Docs: Update for release FEX-2601

Merge pull request #5205 from Sonicadvance1/43
FEXCore: Cleanup pointers structure
2026-10-07 20:00:16 +02:00 · 2026-01-07 14:20:19 -08:00 · 2026-01-07 14:17:30 -08:00 · 2026-01-07 12:55:34 -08:00 · 2026-01-07 12:55:34 -08:00 · 2026-01-07 15:47:58 -05:00
883 changed files with 85555 additions and 60769 deletions
@@ -7,3 +7,6 @@ FEXCore/Source/Interface/Core/X86Tables/*
 # Inline headers with list-like content that can't be processed individually
 Source/Tools/LinuxEmulation/LinuxSyscalls/x*/SyscallsNames.inl
 Source/Tools/LinuxEmulation/LinuxSyscalls/x*/Ioctl/*.inl
+
+# Include files in unittests
+unittests/*ASM/Includes/*.inc
@@ -20,3 +20,5 @@
 # Whole-tree reformat with clang-format-19
 5267cde60e7642852d18f20ae8568643bb5293d5

+# Minor reformat with clang-format-19
+9fdd96af61c969cb5732471223f00eda64b7a069
@@ -34,7 +34,6 @@ jobs:
        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
@@ -41,7 +41,6 @@ jobs:
        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
@@ -34,7 +34,6 @@ jobs:
        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
@@ -33,7 +33,6 @@ jobs:
        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
@@ -48,7 +48,6 @@ jobs:
        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
@@ -78,7 +77,7 @@ 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 -DCMAKE_TOOLCHAIN_FILE=$GITHUB_WORKSPACE/Data/CMake/toolchain_mingw.cmake -DMINGW_TRIPLE=$MINGW_TRIPLE -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True -DBUILD_TESTS=False -DCMAKE_INSTALL_PREFIX=${{runner.workspace}}/build/install
+      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -DCMAKE_TOOLCHAIN_FILE=$GITHUB_WORKSPACE/Data/CMake/toolchain_mingw.cmake -DMINGW_TRIPLE=$MINGW_TRIPLE -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True -DBUILD_TESTING=False -DCMAKE_INSTALL_PREFIX=${{runner.workspace}}/build/install

    - name: Build
      working-directory: ${{runner.workspace}}/build
@@ -0,0 +1,79 @@
+name: steamrt4 build
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - main
+
+env:
+  BUILD_TYPE: Release
+  CC: clang
+  CXX: clang++
+
+jobs:
+  steamrt4_build:
+    runs-on: ${{ matrix.arch }}
+    strategy:
+      matrix:
+        arch: [[self-hosted, ARM64, distrobox]]
+      fail-fast: false
+
+    steps:
+    - uses: actions/checkout@v3
+
+    - name: Set runner label
+      run: echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
+
+    - name : submodule checkout
+      run: |
+        git submodule sync --recursive
+        git submodule update --init --depth 1
+
+    - name: Clean Build Environment
+      run: |
+        rm -Rf ${{runner.workspace}}/build
+        cmake -E make_directory ${{runner.workspace}}/build
+
+    # Setup everything required.
+    - name : distrobox setup
+      run: |
+        distrobox create -Y -i registry.gitlab.steamos.cloud/steamrt/steamrt4/sdk/arm64:4.0.20251117.183306 steamrt4 || true
+        distrobox upgrade steamrt4
+        distrobox enter --name steamrt4 -- sudo apt-get install -y \
+          git cmake ninja-build ccache \
+          lld clang \
+          libclang-dev llvm-dev \
+          libstdc++-14-dev-i386-cross libgcc-14-dev-i386-cross \
+          libstdc++-14-dev-amd64-cross libgcc-14-dev-amd64-cross
+
+    - name: Create Build Environment
+      run: distrobox enter --name steamrt4 -- cmake -E make_directory ${{runner.workspace}}/build
+
+    - name: Configure CMake
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: distrobox enter --name steamrt4 -- cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DBUILD_STEAM_SUPPORT=True -DENABLE_LTO=True -DENABLE_ASSERTIONS=False -DBUILD_THUNKS=True -DBUILD_FEXCONFIG=False -DBUILD_TESTING=False -DENABLE_CLANG_THUNKS=True -DUSE_LINKER=lld -DCMAKE_INSTALL_PREFIX=/usr
+
+    - name: Build
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      run: distrobox enter --name steamrt4 -- cmake --build . --config $BUILD_TYPE
+
+    - name: install
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      env:
+        DESTDIR: ${{runner.workspace}}/install
+      run: distrobox enter --name steamrt4 -- cmake --build . --config $BUILD_TYPE -t install
+    - name: Upload libraries
+      uses: 'actions/upload-artifact@v4'
+      timeout-minutes: 1
+      with:
+        overwrite: true
+        name: steamrt4_steampipe_depot
+        path: ${{runner.workspace}}/install/*
+        retention-days: 60
+        compression-level: 9
@@ -35,7 +35,6 @@ jobs:
        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
@@ -46,12 +46,12 @@ jobs:
    - name: Configure CMake arm64ec
      shell: bash
      working-directory: ${{runner.workspace}}/build_arm64ec
-      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -DCMAKE_TOOLCHAIN_FILE=$GITHUB_WORKSPACE/Data/CMake/toolchain_mingw.cmake -DMINGW_TRIPLE=arm64ec-w64-mingw32 -DCMAKE_INSTALL_LIBDIR=/usr/lib/wine/aarch64-windows -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=False -DENABLE_JEMALLOC_GLIBC_ALLOC=False -DCMAKE_INSTALL_PREFIX=/usr -DBUILD_TESTS=False -DCMAKE_INSTALL_PREFIX=/usr
+      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -DCMAKE_TOOLCHAIN_FILE=$GITHUB_WORKSPACE/Data/CMake/toolchain_mingw.cmake -DMINGW_TRIPLE=arm64ec-w64-mingw32 -DCMAKE_INSTALL_LIBDIR=/usr/lib/wine/aarch64-windows -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=False -DENABLE_JEMALLOC_GLIBC_ALLOC=False -DCMAKE_INSTALL_PREFIX=/usr -DBUILD_TESTING=False -DCMAKE_INSTALL_PREFIX=/usr

    - name: Configure CMake wow64
      shell: bash
      working-directory: ${{runner.workspace}}/build_wow64
-      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -DCMAKE_TOOLCHAIN_FILE=$GITHUB_WORKSPACE/Data/CMake/toolchain_mingw.cmake -DMINGW_TRIPLE=aarch64-w64-mingw32 -DCMAKE_INSTALL_LIBDIR=/usr/lib/wine/aarch64-windows -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=False -DENABLE_JEMALLOC_GLIBC_ALLOC=False -DCMAKE_INSTALL_PREFIX=/usr -DBUILD_TESTS=False -DCMAKE_INSTALL_PREFIX=/usr
+      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -DCMAKE_TOOLCHAIN_FILE=$GITHUB_WORKSPACE/Data/CMake/toolchain_mingw.cmake -DMINGW_TRIPLE=aarch64-w64-mingw32 -DCMAKE_INSTALL_LIBDIR=/usr/lib/wine/aarch64-windows -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=False -DENABLE_JEMALLOC_GLIBC_ALLOC=False -DCMAKE_INSTALL_PREFIX=/usr -DBUILD_TESTING=False -DCMAKE_INSTALL_PREFIX=/usr

    - name: Build arm64ec
      working-directory: ${{runner.workspace}}/build_arm64ec
@@ -1,47 +1,49 @@
 cmake_minimum_required(VERSION 3.14)
 project(FEX C CXX ASM)

-INCLUDE (CheckIncludeFiles)
-CHECK_INCLUDE_FILES ("gdb/jit-reader.h" HAVE_GDB_JIT_READER_H)
+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_FEX_LINUX_TESTS "Build FEXLinuxTests (requires x86 compiler)" FALSE)
 option(BUILD_THUNKS "Build thunks" FALSE)
 option(BUILD_FEXCONFIG "Build FEXConfig" TRUE)
 option(ENABLE_CLANG_THUNKS "Build thunks with clang" TRUE)
-option(ENABLE_IWYU "Enables include what you use program" FALSE)
+option(ENABLE_IWYU "Enable the Include What You Use sanitizer" FALSE)
 option(ENABLE_LTO "Enable LTO with compilation" TRUE)
 option(ENABLE_XRAY "Enable building with LLVM X-Ray" FALSE)
-set(USE_LINKER "" CACHE STRING "Allow overriding the linker path directly")
-option(ENABLE_UBSAN "Enables Clang UBSAN" FALSE)
-option(ENABLE_ASAN "Enables Clang ASAN" FALSE)
-option(ENABLE_TSAN "Enables Clang TSAN" FALSE)
-option(ENABLE_COVERAGE "Enables Coverage" FALSE)
-option(ENABLE_ASSERTIONS "Enables assertions in build" FALSE)
-option(ENABLE_GDB_SYMBOLS "Enables GDBSymbols integration support" ${HAVE_GDB_JIT_READER_H})
-option(ENABLE_STRICT_WERROR "Enables stricter -Werror for CI" FALSE)
-option(ENABLE_WERROR "Enables -Werror" FALSE)
-option(ENABLE_JEMALLOC "Enables jemalloc allocator" TRUE)
-option(ENABLE_JEMALLOC_GLIBC_ALLOC "Enables jemalloc glibc 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_CCACHE "Enables ccache for compile caching" TRUE)
-option(ENABLE_VIXL_SIMULATOR "Enable use of VIXL simulator for emulation (only useful for CI testing)" FALSE)
-option(ENABLE_VIXL_DISASSEMBLER "Enables debug disassembler output with VIXL" FALSE)
-option(USE_LEGACY_BINFMTMISC "Uses legacy method of setting up binfmt_misc" FALSE)
-option(ENABLE_FEXCORE_PROFILER "Enables use of the FEXCore timeline profiling capabilities" FALSE)
-set (FEXCORE_PROFILER_BACKEND "gpuvis" CACHE STRING "Set which backend to use for the FEXCore profiler (gpuvis, tracy)")
+set(USE_LINKER "" CACHE STRING "Path to a custom linker program")
+option(ENABLE_UBSAN "Enable the Clang Undefined Behavior Sanitizer" FALSE)
+option(ENABLE_ASAN "Enable the Clang Address Sanitizer" FALSE)
+option(ENABLE_TSAN "Enable the Clang Thread Sanitizer" FALSE)
+option(ENABLE_COVERAGE "Enable Code Coverage" FALSE)
+option(ENABLE_ASSERTIONS "Enable debug assertions" FALSE)
+option(ENABLE_GDB_SYMBOLS "Enable GDBSymbols integration support" ${HAVE_GDB_JIT_READER_H})
+option(ENABLE_STRICT_WERROR "Enable stricter -Werror" FALSE)
+option(ENABLE_WERROR "Enable -Werror" FALSE)
+option(ENABLE_JEMALLOC "Enable jemalloc allocator" TRUE)
+option(ENABLE_JEMALLOC_GLIBC_ALLOC "Enable jemalloc glibc allocator" TRUE)
+option(ENABLE_OFFLINE_TELEMETRY "Enable FEX offline telemetry" TRUE)
+option(ENABLE_COMPILE_TIME_TRACE "Enable time trace compile option" FALSE)
+option(ENABLE_LIBCXX "Use LLVM's libc++ instead of the GNU libstdc++" FALSE)
+option(ENABLE_CCACHE "Enable ccache for build caching" TRUE)
+option(ENABLE_VIXL_SIMULATOR "Use the VIXL simulator for emulation (only useful for CI testing)" FALSE)
+option(ENABLE_VIXL_DISASSEMBLER "Enable debug disassembler output with VIXL" FALSE)
+option(USE_LEGACY_BINFMTMISC "Use legacy method of setting up binfmt_misc" FALSE)
+option(ENABLE_FEXCORE_PROFILER "Enable FEXCore's timeline profiling capabilities" FALSE)
+set(FEXCORE_PROFILER_BACKEND "gpuvis" CACHE STRING "Set which backend to use for FEXCore's profiler")
+set_property(CACHE FEXCORE_PROFILER_BACKEND PROPERTY STRINGS gpuvis tracy)
 option(ENABLE_GLIBC_ALLOCATOR_HOOK_FAULT "Enables glibc memory allocation hooking with fault for CI testing")
-option(USE_PDB_DEBUGINFO "Builds debug info in PDB format" FALSE)
+option(USE_PDB_DEBUGINFO "Build debug info in PDB format" FALSE)
+option(BUILD_STEAM_SUPPORT "Enable Steam integration" FALSE)
+
+set(X86_32_TOOLCHAIN_FILE "${CMAKE_CURRENT_SOURCE_DIR}/Data/CMake/toolchain_x86_32.cmake" CACHE FILEPATH "Toolchain file for the (cross-)compiler targeting i686")
+set(X86_64_TOOLCHAIN_FILE "${CMAKE_CURRENT_SOURCE_DIR}/Data/CMake/toolchain_x86_64.cmake" CACHE FILEPATH "Toolchain file for the (cross-)compiler targeting x86_64")
+set(X86_DEV_ROOTFS "/" CACHE FILEPATH "Path to the sysroot used for cross-compiling for i686 and x86_64")
+set(DATA_DIRECTORY "" CACHE PATH "Global data directory (override)")
+set(HOSTLIBS_DATA_DIRECTORY "" CACHE PATH "Global data directory (override)")

-set (X86_32_TOOLCHAIN_FILE "${CMAKE_CURRENT_SOURCE_DIR}/Data/CMake/toolchain_x86_32.cmake" CACHE FILEPATH "Toolchain file for the (cross-)compiler targeting i686")
-set (X86_64_TOOLCHAIN_FILE "${CMAKE_CURRENT_SOURCE_DIR}/Data/CMake/toolchain_x86_64.cmake" CACHE FILEPATH "Toolchain file for the (cross-)compiler targeting x86_64")
-set (X86_DEV_ROOTFS "/" CACHE FILEPATH "Path to the sysroot used for cross-compiling for i686 and x86_64")
-set (DATA_DIRECTORY "" CACHE PATH "Global data directory (override)")
-set (HOSTLIBS_DATA_DIRECTORY "" CACHE PATH "Global data directory (override)")
 if (NOT DATA_DIRECTORY)
-  set (DATA_DIRECTORY "${CMAKE_INSTALL_PREFIX}/share/fex-emu")
+  set(DATA_DIRECTORY "${CMAKE_INSTALL_PREFIX}/share/fex-emu")
 endif()

 include(GNUInstallDirs)
@@ -49,22 +51,74 @@ if (NOT HOSTLIBS_DATA_DIRECTORY)
  set(HOSTLIBS_DATA_DIRECTORY "${CMAKE_INSTALL_FULL_LIBDIR}/fex-emu")
 endif()

-string(FIND ${CMAKE_BASE_NAME} mingw CONTAINS_MINGW)
-if (NOT CONTAINS_MINGW EQUAL -1)
-  message (STATUS "Mingw build")
-  set (MINGW_BUILD TRUE)
-  set (ENABLE_JEMALLOC TRUE)
-  set (ENABLE_JEMALLOC_GLIBC_ALLOC FALSE)
+## Platform Checks ##
+# Only 64-bit Linux and Windows are supported
+
+# NB: SIZEOF_VOID_P is in bytes, not bits
+# On 32-bit systems this is set to 4
+if (NOT CMAKE_SIZEOF_VOID_P EQUAL 8)
+  message(FATAL_ERROR "Unsupported pointer size ${CMAKE_SIZEOF_VOID_P}."
+    " FEX only supports 64-bit (8-byte pointer) systems."
+    " If you believe this is in error, file an issue.")
+elseif (NOT (WIN32 OR CMAKE_SYSTEM_NAME STREQUAL "Linux"))
+  message(FATAL_ERROR "Unsupported system type ${CMAKE_SYSTEM_NAME}."
+    " FEX only supports Linux and Windows."
+    " If you believe this is in error, file an issue.")
 endif()

-if (NOT MINGW_BUILD)
-  message (STATUS "Clang version ${CMAKE_CXX_COMPILER_VERSION}")
-  set (CLANG_MINIMUM_VERSION 13.0)
+## Compiler Checks ##
+# GCC and MSVC are unsupported
+if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
+  message(FATAL_ERROR "FEX doesn't support GCC! Use Clang instead.")
+elseif (MSVC)
+  message(FATAL_ERROR "FEX doesn't support MSVC! Use Clang on MinGW instead.")
+elseif (MINGW)
+  message(STATUS "Building for MinGW")
+  set(ENABLE_JEMALLOC TRUE)
+  set(ENABLE_JEMALLOC_GLIBC_ALLOC FALSE)
+else ()
+  message(STATUS "Clang version ${CMAKE_CXX_COMPILER_VERSION}")
+  set(CLANG_MINIMUM_VERSION 13.0)
  if (CMAKE_CXX_COMPILER_VERSION VERSION_LESS ${CLANG_MINIMUM_VERSION})
-    message (FATAL_ERROR "Clang version too old for FEX. Need at least ${CLANG_MINIMUM_VERSION} but has ${CMAKE_CXX_COMPILER_VERSION}")
+    message(FATAL_ERROR "Clang version too old for FEX. Need at least ${CLANG_MINIMUM_VERSION} but has ${CMAKE_CXX_COMPILER_VERSION}")
  endif()
 endif()

+## Architecture Handling ##
+string(TOLOWER ${CMAKE_SYSTEM_PROCESSOR} processor)
+if (processor MATCHES "x86|amd64")
+  option(ENABLE_X86_HOST_DEBUG "Enables compiling on x86_64 host" FALSE)
+  if (NOT ENABLE_X86_HOST_DEBUG)
+    message(FATAL_ERROR
+      " FEX doesn't support compiling for x86-64 hosts!"
+      " This is /only/ a supported configuration for FEX CI and nothing else!")
+  else()
+    message(STATUS "x86_64 debug build")
+  endif()
+
+  set(ARCHITECTURE_x86_64 1)
+  add_definitions(-DARCHITECTURE_x86_64=1)
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mcx16")
+elseif (processor MATCHES "^aarch64|^arm64|^armv8\.*")
+  set(ARCHITECTURE_arm64 1)
+  add_definitions(-DARCHITECTURE_arm64=1)
+
+  # arm64ec needs to define both arm64 and arm64ec
+  if (processor MATCHES "^arm64ec")
+    set(ARCHITECTURE_arm64ec 1)
+    add_definitions(-DARCHITECTURE_arm64ec=1)
+  endif()
+endif()
+
+if (NOT (ARCHITECTURE_arm64 OR ARCHITECTURE_arm64ec OR ARCHITECTURE_x86_64))
+  message(FATAL_ERROR "Unsupported processor type ${processor}."
+    " If you believe this is in error, file an issue.")
+endif()
+
+if (BUILD_STEAM_SUPPORT)
+  add_definitions(-DFEX_STEAM_SUPPORT=1)
+endif()
+
 if (ENABLE_FEXCORE_PROFILER)
  add_definitions(-DENABLE_FEXCORE_PROFILER=1)
  string(TOUPPER "${FEXCORE_PROFILER_BACKEND}" FEXCORE_PROFILER_BACKEND)
@@ -84,8 +138,8 @@ if (ENABLE_FEXCORE_PROFILER)
    add_definitions(-DTRACY_NO_SAMPLING=1)
    # This pulls in libbacktrace which allocators in global constructors (before FEX can set up its allocator hooks)
    add_definitions(-DTRACY_NO_CALLSTACK=1)
-    if (MINGW_BUILD)
-      message(FATAL_ERROR "Tracy profiler not supported")
+    if (MINGW)
+      message(FATAL_ERROR "Tracy profiler not supported on MinGW")
    endif()
  else()
    message(FATAL_ERROR "Unknown FEXCore profiler backend ${FEXCORE_PROFILER_BACKEND}")
@@ -112,9 +166,10 @@ if(NOT TARGET uninstall)
 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>}")
+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")
+set(OVERRIDE_HASH "detect" CACHE STRING "Override the FEX git hash")

 string(TOUPPER "${CMAKE_BUILD_TYPE}" CMAKE_BUILD_TYPE)
 if (CMAKE_BUILD_TYPE MATCHES "DEBUG")
@@ -131,7 +186,6 @@ if (ENABLE_GDB_SYMBOLS)
  add_definitions(-DGDB_SYMBOLS_ENABLED=1)
 endif()

-
 set(CMAKE_CXX_STANDARD 20)
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/Bin)
@@ -141,33 +195,7 @@ cmake_policy(SET CMP0083 NEW) # Follow new PIE policy
 include(CheckPIESupported)
 check_pie_supported()

-if (ENABLE_LTO)
-  set(CMAKE_INTERPROCEDURAL_OPTIMIZATION TRUE)
-else()
-  set(CMAKE_INTERPROCEDURAL_OPTIMIZATION FALSE)
-endif()
-
-if (CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64")
-  option(ENABLE_X86_HOST_DEBUG "Enables compiling on x86_64 host" FALSE)
-  if (NOT ENABLE_X86_HOST_DEBUG)
-    message(FATAL_ERROR
-    " FEX-Emu doesn't support compiling for x86-64 hosts!"
-    " This is /only/ a supported configuration for FEX CI and nothing else!")
-  endif()
-  set(_M_X86_64 1)
-  add_definitions(-D_M_X86_64=1)
-  set (CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mcx16")
-endif()
-
-if (CMAKE_SYSTEM_PROCESSOR MATCHES "^aarch64|^arm64|^armv8\.*")
-  set(_M_ARM_64 1)
-  add_definitions(-D_M_ARM_64=1)
-endif()
-
-if (CMAKE_SYSTEM_PROCESSOR MATCHES "^arm64ec")
-  set(_M_ARM_64EC 1)
-  add_definitions(-D_M_ARM_64EC=1)
-endif()
+set(CMAKE_INTERPROCEDURAL_OPTIMIZATION ${ENABLE_LTO})

 include(CheckCXXSourceCompiles)
 set(CMAKE_REQUIRED_FLAGS "-std=c++11 -Wattributes -Werror=attributes")
@@ -183,15 +211,15 @@ check_cxx_source_compiles(
  HAS_CLANG_PRESERVE_ALL)
 unset(CMAKE_REQUIRED_FLAGS)
 if (HAS_CLANG_PRESERVE_ALL)
-  if (MINGW_BUILD)
+  if (MINGW)
    message(STATUS "Ignoring broken clang::preserve_all support")
    set(HAS_CLANG_PRESERVE_ALL FALSE)
  else()
    message(STATUS "Has clang::preserve_all")
  endif()
-endif ()
+endif()

-if (_M_ARM_64 AND HAS_CLANG_PRESERVE_ALL)
+if (ARCHITECTURE_arm64 AND HAS_CLANG_PRESERVE_ALL)
  add_definitions("-DFEX_PRESERVE_ALL_ATTR=__attribute__((preserve_all))" "-DFEX_HAS_PRESERVE_ALL_ATTR=1")
 else()
  add_definitions("-DFEX_PRESERVE_ALL_ATTR=" "-DFEX_HAS_PRESERVE_ALL_ATTR=0")
@@ -220,7 +248,7 @@ if (ENABLE_COMPILE_TIME_TRACE)
  link_libraries(-ftime-trace)
 endif()

-set (PTHREAD_LIB pthread)
+set(PTHREAD_LIB pthread)

 if (USE_LINKER)
  message(STATUS "Overriding linker to: ${USE_LINKER}")
@@ -272,8 +300,8 @@ if (ENABLE_JEMALLOC_GLIBC_ALLOC)
  # Required for thunks to work.
  # All host native libraries will use this allocator, while *most* other FEX internal allocations will use the other jemalloc allocator.
  add_subdirectory(External/jemalloc_glibc/)
-elseif (NOT MINGW_BUILD)
-  message (STATUS
+elseif (NOT MINGW)
+  message(STATUS
    " jemalloc glibc allocator disabled!\n"
    " This is not a recommended configuration!\n"
    " This will very explicitly break thunk execution!\n"
@@ -283,8 +311,8 @@ endif()
 if (ENABLE_JEMALLOC)
  # The jemalloc subproject that all FEXCore fextl objects allocate through.
  add_subdirectory(External/jemalloc/)
-elseif (NOT MINGW_BUILD)
-  message (STATUS
+elseif (NOT MINGW)
+  message(STATUS
    " jemalloc disabled!\n"
    " This is not a recommended configuration!\n"
    " This will very explicitly break 32-bit application execution!\n"
@@ -296,15 +324,22 @@ if (USE_PDB_DEBUGINFO)
  add_link_options(-g -Wl,--pdb=)
 endif()

-set (CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} -fno-omit-frame-pointer")
-set (CMAKE_LINKER_FLAGS_RELWITHDEBINFO "${CMAKE_LINKER_FLAGS_RELWITHDEBINFO} -fno-omit-frame-pointer")
+set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} -fno-omit-frame-pointer")
+set(CMAKE_LINKER_FLAGS_RELWITHDEBINFO "${CMAKE_LINKER_FLAGS_RELWITHDEBINFO} -fno-omit-frame-pointer")

-set (CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -fomit-frame-pointer")
-set (CMAKE_LINKER_FLAGS_RELEASE "${CMAKE_LINKER_FLAGS_RELEASE} -fomit-frame-pointer")
+set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -fomit-frame-pointer")
+set(CMAKE_LINKER_FLAGS_RELEASE "${CMAKE_LINKER_FLAGS_RELEASE} -fomit-frame-pointer")

+## Modules ##
+list(APPEND CMAKE_MODULE_PATH ${CMAKE_SOURCE_DIR}/Data/CMake/)
+
+include(LinkerGC)
+
+## Externals ##
 include_directories(External/robin-map/include/)

-if (BUILD_TESTS OR ENABLE_VIXL_DISASSEMBLER OR ENABLE_VIXL_SIMULATOR)
+include(CTest)
+if (BUILD_TESTING OR ENABLE_VIXL_DISASSEMBLER OR ENABLE_VIXL_SIMULATOR)
  add_subdirectory(External/vixl/)
  include_directories(SYSTEM External/vixl/src/)
 endif()
@@ -313,20 +348,16 @@ if (ENABLE_FEXCORE_PROFILER AND FEXCORE_PROFILER_BACKEND STREQUAL "TRACY")
  add_subdirectory(External/tracy)
 endif()

-if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
-  # This means we were attempted to get compiled with GCC
-  message(FATAL_ERROR "FEX doesn't support getting compiled with GCC!")
-endif()
-
 find_package(PkgConfig REQUIRED)
-find_package(Python 3.0 REQUIRED COMPONENTS Interpreter)
+find_package(Python 3.9 REQUIRED COMPONENTS Interpreter)

 set(BUILD_SHARED_LIBS OFF)

-pkg_search_module(xxhash IMPORTED_TARGET xxhash libxxhash)
-if (TARGET PkgConfig::xxhash AND NOT CMAKE_CROSSCOMPILING)
-  add_library(xxHash::xxhash ALIAS PkgConfig::xxhash)
-else()
+if (NOT CMAKE_CROSSCOMPILING)
+  find_package(xxhash MODULE QUIET)
+endif()
+
+if (NOT TARGET xxHash::xxhash)
  set(XXHASH_BUNDLED_MODE TRUE)
  set(XXHASH_BUILD_XXHSUM FALSE)
  add_subdirectory(External/xxhash/cmake_unofficial/)
@@ -335,7 +366,7 @@ endif()
 add_definitions(-Wno-trigraphs)
 add_definitions(-DGLOBAL_DATA_DIRECTORY="${DATA_DIRECTORY}/")

-if (BUILD_TESTS)
+if (BUILD_TESTING)
  find_package(Catch2 3 QUIET)
  if (NOT Catch2_FOUND)
    add_subdirectory(External/Catch2/)
@@ -345,6 +376,9 @@ if (BUILD_TESTS)
  endif()

  include(Catch)
+else ()
+  # Override any previously generated test list to avoid running stale test binaries
+  file(GENERATE OUTPUT CTestTestfile.cmake CONTENT "# No tests since BUILD_TESTING is disabled")
 endif()

 find_package(fmt QUIET)
@@ -404,7 +438,7 @@ if (NOT TUNE_ARCH STREQUAL "generic")
 endif()

 if (TUNE_CPU STREQUAL "native")
-  if(_M_ARM_64)
+  if(ARCHITECTURE_arm64)
    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
@@ -445,6 +479,40 @@ elseif (NOT TUNE_CPU STREQUAL "none")
  endif()
 endif()

+set(GIT_DESCRIBE_STRING "FEX-Unknown")
+
+if (OVERRIDE_VERSION STREQUAL "detect")
+  find_package(Git)
+
+  if (GIT_FOUND)
+    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_DESCRIBE_STRING "${OVERRIDE_VERSION}")
+endif()
+
+set(GIT_SHORT_HASH "Unknown")
+
+if (OVERRIDE_HASH STREQUAL "detect")
+  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)
+  endif()
+else()
+  set(GIT_SHORT_HASH "${OVERRIDE_HASH}")
+endif()
+
 if (ENABLE_IWYU)
  find_program(IWYU_EXE "iwyu")
  if (IWYU_EXE)
@@ -455,15 +523,10 @@ endif()

 add_compile_options(-Wall)

-include(CTest)
-if (BUILD_TESTS)
+if (BUILD_TESTING)
  message(STATUS "Unit tests are enabled")
-  if (NOT BUILD_TESTING)
-    # CMake checks this variable before generating CTestTestfile.cmake
-    message(SEND_ERROR "Unit tests require BUILD_TESTING to be enabled")
-  endif()

-  set (TEST_JOB_COUNT "" CACHE STRING "Override number of parallel jobs to use while running tests")
+  set(TEST_JOB_COUNT "" CACHE STRING "Override number of parallel jobs to use while running tests")
  if (TEST_JOB_COUNT)
    message(STATUS "Running tests with ${TEST_JOB_COUNT} jobs")
  elseif(CMAKE_VERSION VERSION_LESS "3.29")
@@ -478,13 +541,16 @@ add_subdirectory(FEXHeaderUtils/)
 add_subdirectory(CodeEmitter/)
 add_subdirectory(FEXCore/)

-if (_M_ARM_64 AND NOT MINGW_BUILD)
+if (ARCHITECTURE_arm64 AND NOT MINGW AND NOT BUILD_STEAM_SUPPORT)
  # Binfmt_misc files must be installed prior to Source/ installs
  add_subdirectory(Data/binfmts/)
 endif()

 add_subdirectory(Source/)
-add_subdirectory(Data/AppConfig/)
+
+if (NOT BUILD_STEAM_SUPPORT)
+  add_subdirectory(Data/AppConfig/)
+endif()

 # Install the ThunksDB file
 file(GLOB CONFIG_SOURCES CONFIGURE_DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/Data/*.json)
@@ -492,15 +558,16 @@ file(GLOB CONFIG_SOURCES CONFIGURE_DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/Data/*.js
 # Any application configuration json file gets installed
 foreach(CONFIG_SRC ${CONFIG_SOURCES})
  install(FILES ${CONFIG_SRC}
-    DESTINATION ${DATA_DIRECTORY}/)
+    DESTINATION ${DATA_DIRECTORY}/
+    COMPONENT Runtime)
 endforeach()

-if (BUILD_TESTS)
+if (BUILD_TESTING)
  add_subdirectory(unittests/)
 endif()

 if (BUILD_THUNKS)
-  set (FEX_PROJECT_SOURCE_DIR ${PROJECT_SOURCE_DIR})
+  set(FEX_PROJECT_SOURCE_DIR ${PROJECT_SOURCE_DIR})
  add_subdirectory(ThunkLibs/Generator)

  # Thunk targets for both host libraries and IDE integration
@@ -527,8 +594,7 @@ if (BUILD_THUNKS)
      "-DX86_DEV_ROOTFS=${X86_DEV_ROOTFS}"
    INSTALL_COMMAND ""
    BUILD_ALWAYS ON
-    DEPENDS thunkgen
-  )
+    DEPENDS thunkgen)

  ExternalProject_Add(guest-libs-32
    PREFIX guest-libs-32
@@ -546,106 +612,36 @@ if (BUILD_THUNKS)
      "-DX86_DEV_ROOTFS=${X86_DEV_ROOTFS}"
    INSTALL_COMMAND ""
    BUILD_ALWAYS ON
-    DEPENDS thunkgen
-  )
+    DEPENDS thunkgen)

  install(
-    CODE "MESSAGE(\"-- Installing: guest-libs\")"
+    CODE "message(\"-- Installing: guest-libs\")"
    CODE "
-    EXECUTE_PROCESS(COMMAND ${CMAKE_COMMAND} --build . --target install
-    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest
-    )"
+      execute_process(COMMAND ${CMAKE_COMMAND} --build . --target install
+        WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest)"
    DEPENDS guest-libs
-  )
+    COMPONENT Runtime)

  install(
-    CODE "MESSAGE(\"-- Installing: guest-libs-32\")"
+    CODE "message(\"-- Installing: guest-libs-32\")"
    CODE "
-    EXECUTE_PROCESS(COMMAND ${CMAKE_COMMAND} --build . --target install
-    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest_32
-    )"
+      execute_process(COMMAND ${CMAKE_COMMAND} --build . --target install
+        WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest_32)"
    DEPENDS guest-libs-32
-  )
+    COMPONENT Runtime)

  add_custom_target(uninstall_guest-libs
    COMMAND ${CMAKE_COMMAND} "--build" "." "--target" "uninstall"
-    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest
-  )
+    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
-  )
+    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")
-
-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 ${GIT_ERROR} EQUAL 0)
-      # Likely built in a way that doesn't have tags
-      # Setup a version tag that is unknown
-      set(GIT_DESCRIBE_STRING "FEX-0000")
-    endif()
-  endif()
-else()
-  set(GIT_DESCRIBE_STRING "FEX-${OVERRIDE_VERSION}")
+if (BUILD_STEAM_SUPPORT)
+  add_subdirectory(Source/Steam/)
 endif()
-
-# 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)
-
-# Change `2106.1` in to a list
-string(REPLACE "." ";" DESCRIBE_LIST ${DESCRIBE_LIST})
-
-# Calculate list size
-list(LENGTH DESCRIBE_LIST LIST_SIZE)
-
-# Pull out the major version
-list(GET DESCRIBE_LIST 0 FEX_VERSION_MAJOR)
-
-# Minor version only exists if there is a .1 at the end
-# eg: 2106 versus 2106.1
-if (LIST_SIZE GREATER 1)
-  list(GET DESCRIBE_LIST 1 FEX_VERSION_MINOR)
-endif()
-
-# Package creation
-set (CPACK_GENERATOR "DEB")
-set (CPACK_PACKAGE_NAME fex-emu)
-set (CPACK_PACKAGE_FILE_NAME "${CPACK_PACKAGE_NAME}-${GIT_DESCRIBE_STRING}_${CMAKE_SYSTEM_PROCESSOR}")
-set (CPACK_PACKAGE_CONTACT "FEX-Emu Maintainers <team@fex-emu.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}/Data/CMake/CPack/Description.txt")
-
-# Debian defines
-set (CPACK_DEBIAN_PACKAGE_DEPENDS "libc6, libstdc++6, libepoxy0, libsdl2-2.0-0, libegl1, libx11-6, squashfuse")
-set (CPACK_DEBIAN_PACKAGE_CONTROL_EXTRA
-  "${CMAKE_CURRENT_SOURCE_DIR}/Data/CMake/CPack/postinst;${CMAKE_CURRENT_SOURCE_DIR}/Data/CMake/CPack/prerm;${CMAKE_CURRENT_SOURCE_DIR}/Data/CMake/CPack/triggers")
-if (CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64")
-  # binfmt_misc conflicts with qemu-user-static
-  # We also only install binfmt_misc on aarch64 hosts
-  set (CPACK_DEBIAN_PACKAGE_CONFLICTS "${CPACK_DEBIAN_PACKAGE_CONFLICTS}, qemu-user-static")
-endif()
-include (CPack)
@@ -129,4 +129,4 @@
 			"variables": []
 		}
 	]
-}
+}
@@ -36,24 +36,31 @@ public:
    DataProcessing_PCRel_Imm(Op, rd, Imm);
  }

-  void adr(ARMEmitter::Register rd, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded adr(ARMEmitter::Register rd, const BackwardLabel* Label) {
    int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    LOGMAN_THROW_A_FMT(IsADRRange(Imm), "Unscaled offset too large");
+    if (IsADRRange(Imm)) {
+      constexpr uint32_t Op = 0b0001'0000 << 24;
+      DataProcessing_PCRel_Imm(Op, rd, Imm);
+      return BranchEncodeSucceeded::Success;
+    }

-    constexpr uint32_t Op = 0b0001'0000 << 24;
-    DataProcessing_PCRel_Imm(Op, rd, Imm);
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
  }
-  void adr(ARMEmitter::Register rd, ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded adr(ARMEmitter::Register rd, ForwardLabel* Label) {
    AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::ADR});
    constexpr uint32_t Op = 0b0001'0000 << 24;
    DataProcessing_PCRel_Imm(Op, rd, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
  }

-  void adr(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded adr(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
    if (Label->Backward.Location) {
-      adr(rd, &Label->Backward);
+      return adr(rd, &Label->Backward);
    } else {
-      adr(rd, &Label->Forward);
+      return adr(rd, &Label->Forward);
    }
  }

@@ -62,38 +69,53 @@ public:
    DataProcessing_PCRel_Imm(Op, rd, Imm);
  }

-  void adrp(ARMEmitter::Register rd, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded adrp(ARMEmitter::Register rd, const BackwardLabel* Label) {
    int64_t Imm = reinterpret_cast<int64_t>(Label->Location) - (GetCursorAddress<int64_t>() & ~0xFFFLL);
-    LOGMAN_THROW_A_FMT(IsADRPRange(Imm) && IsADRPAligned(Imm), "Unscaled offset too large");

-    constexpr uint32_t Op = 0b1001'0000 << 24;
-    DataProcessing_PCRel_Imm(Op, rd, Imm);
+    if (IsADRPRange(Imm) && IsADRPAligned(Imm)) {
+      constexpr uint32_t Op = 0b1001'0000 << 24;
+      DataProcessing_PCRel_Imm(Op, rd, Imm);
+      return BranchEncodeSucceeded::Success;
+    }
+
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
  }
-  void adrp(ARMEmitter::Register rd, ForwardLabel* Label) {
+
+  [[nodiscard]] BranchEncodeSucceeded adrp(ARMEmitter::Register rd, ForwardLabel* Label) {
    AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::ADRP});
    constexpr uint32_t Op = 0b1001'0000 << 24;
    DataProcessing_PCRel_Imm(Op, rd, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
  }

-  void adrp(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded adrp(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
    if (Label->Backward.Location) {
-      adrp(rd, &Label->Backward);
+      return adrp(rd, &Label->Backward);
    } else {
-      adrp(rd, &Label->Forward);
+      return adrp(rd, &Label->Forward);
    }
  }

-  void LongAddressGen(ARMEmitter::Register rd, const BackwardLabel* Label) {
-    int64_t Imm = reinterpret_cast<int64_t>(Label->Location) - (GetCursorAddress<int64_t>());
+  [[nodiscard]] BranchEncodeSucceeded LongAddressGen(ARMEmitter::Register rd, const BackwardLabel* Label) {
+    const auto SLocation = reinterpret_cast<int64_t>(Label->Location);
+    const auto ULocation = std::bit_cast<uint64_t>(SLocation);
+
+    const int64_t Imm = SLocation - (GetCursorAddress<int64_t>());
+    const auto UImm = std::bit_cast<uint64_t>(Imm);
+
    if (IsADRRange(Imm)) {
      // If the range is in ADR range then we can just use ADR.
-      adr(rd, Label);
-    } else if (IsADRPRange(Imm)) {
-      int64_t ADRPImm = (reinterpret_cast<int64_t>(Label->Location) & ~0xFFFLL) - (GetCursorAddress<int64_t>() & ~0xFFFLL);
+      return adr(rd, Label);
+    }
+    if (IsADRPRange(Imm)) {
+      const int64_t ADRPImm = (SLocation & ~0xFFFLL) - (GetCursorAddress<int64_t>() & ~0xFFFLL);

      // If the range is in the ADRP range then we can use ADRP.
-      bool NeedsOffset = !IsADRPAligned(reinterpret_cast<uint64_t>(Label->Location));
-      uint64_t AlignedOffset = reinterpret_cast<uint64_t>(Label->Location) & 0xFFFULL;
+      const bool NeedsOffset = !IsADRPAligned(ULocation);
+      const uint64_t AlignedOffset = ULocation & 0xFFFULL;

      // First emit ADRP
      adrp(rd, ADRPImm >> 12);
@@ -102,23 +124,33 @@ public:
        // Now even an add
        add(ARMEmitter::Size::i64Bit, rd, rd, AlignedOffset);
      }
-    } else {
-      LOGMAN_MSG_A_FMT("Unscaled offset too large");
-      FEX_UNREACHABLE;
+
+      return BranchEncodeSucceeded::Success;
    }
+
+    // Stinky path, we need to load the address as a sequence of movz+movk+movk
+    movz(ARMEmitter::Size::i64Bit, rd, (UImm >> 32) & 0xFFFF, 32);
+    movk(ARMEmitter::Size::i64Bit, rd, (UImm >> 16) & 0xFFFF, 16);
+    movk(ARMEmitter::Size::i64Bit, rd, UImm & 0xFFFF);
+
+    return BranchEncodeSucceeded::Success;
  }
-  void LongAddressGen(ARMEmitter::Register rd, ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded LongAddressGen(ARMEmitter::Register rd, ForwardLabel* Label) {
    AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::LONG_ADDRESS_GEN});
-    // Emit a register index and a nop. These will be backpatched.
+    // Emit a register index and two nops. These will be backpatched.
    dc32(rd.Idx());
    nop();
+    nop();
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
  }

-  void LongAddressGen(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded LongAddressGen(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
    if (Label->Backward.Location) {
-      LongAddressGen(rd, &Label->Backward);
+      return LongAddressGen(rd, &Label->Backward);
    } else {
-      LongAddressGen(rd, &Label->Forward);
+      return LongAddressGen(rd, &Label->Forward);
    }
  }

@@ -862,12 +894,6 @@ public:
  }

 private:
-  static constexpr Condition InvertCondition(Condition cond) {
-    // These behave as always, so it makes no sense to allow inverting these.
-    LOGMAN_THROW_A_FMT(cond != Condition::CC_AL && cond != Condition::CC_NV, "Cannot invert CC_AL or CC_NV");
-    return static_cast<Condition>(FEXCore::ToUnderlying(cond) ^ 1);
-  }
-
  void and_(ARMEmitter::Size s, ARMEmitter::Register rd, 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);
@@ -2244,8 +2244,7 @@ public:

  template<IsQOrDRegister T>
  void movi(SubRegSize size, T rd, uint64_t Imm, uint16_t Shift = 0) {
-    LOGMAN_THROW_A_FMT(size == SubRegSize::i8Bit || size == SubRegSize::i16Bit || size == SubRegSize::i32Bit ||
-                        size == SubRegSize::i64Bit,
+    LOGMAN_THROW_A_FMT(size == SubRegSize::i8Bit || size == SubRegSize::i16Bit || size == SubRegSize::i32Bit || size == SubRegSize::i64Bit,
                       "Unsupported movi size");

    uint32_t cmode;
@@ -20,23 +20,31 @@ public:
    constexpr uint32_t Op = 0b0101'010 << 25;
    Branch_Conditional(Op, 0, 0, Cond, Imm);
  }
-  void b(ARMEmitter::Condition Cond, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded b(ARMEmitter::Condition Cond, const BackwardLabel* 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);
+    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) {
+      constexpr uint32_t Op = 0b0101'010 << 25;
+      Branch_Conditional(Op, 0, 0, Cond, Imm >> 2);
+      return BranchEncodeSucceeded::Success;
+    }
+
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
  }
-  void b(ARMEmitter::Condition Cond, ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded b(ARMEmitter::Condition Cond, ForwardLabel* Label) {
    AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::BC});
    constexpr uint32_t Op = 0b0101'010 << 25;
    Branch_Conditional(Op, 0, 0, Cond, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
  }

-  void b(ARMEmitter::Condition Cond, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded b(ARMEmitter::Condition Cond, BiDirectionalLabel* Label) {
    if (Label->Backward.Location) {
-      b(Cond, &Label->Backward);
+      return b(Cond, &Label->Backward);
    } else {
-      b(Cond, &Label->Forward);
+      return b(Cond, &Label->Forward);
    }
  }

@@ -45,24 +53,32 @@ public:
    constexpr uint32_t Op = 0b0101'010 << 25;
    Branch_Conditional(Op, 0, 1, Cond, Imm);
  }
-  void bc(ARMEmitter::Condition Cond, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded bc(ARMEmitter::Condition Cond, const BackwardLabel* 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);
+    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) {
+      constexpr uint32_t Op = 0b0101'010 << 25;
+      Branch_Conditional(Op, 0, 1, Cond, Imm >> 2);
+      return BranchEncodeSucceeded::Success;
+    }
+
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
  }

-  void bc(ARMEmitter::Condition Cond, ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded bc(ARMEmitter::Condition Cond, ForwardLabel* Label) {
    AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::BC});
    constexpr uint32_t Op = 0b0101'010 << 25;
    Branch_Conditional(Op, 0, 1, Cond, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
  }

-  void bc(ARMEmitter::Condition Cond, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded bc(ARMEmitter::Condition Cond, BiDirectionalLabel* Label) {
    if (Label->Backward.Location) {
-      bc(Cond, &Label->Backward);
+      return bc(Cond, &Label->Backward);
    } else {
-      bc(Cond, &Label->Forward);
+      return bc(Cond, &Label->Forward);
    }
  }

@@ -98,25 +114,32 @@ public:

    UnconditionalBranch(Op, Imm);
  }
-  void b(const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded b(const BackwardLabel* 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;
+    if (Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0)) {
+      constexpr uint32_t Op = 0b0001'01 << 26;
+      UnconditionalBranch(Op, Imm >> 2);
+      return BranchEncodeSucceeded::Success;
+    }

-    UnconditionalBranch(Op, Imm >> 2);
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
  }
-  void b(ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded b(ForwardLabel* Label) {
    AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::B});
    constexpr uint32_t Op = 0b0001'01 << 26;

    UnconditionalBranch(Op, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
  }

-  void b(BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded b(BiDirectionalLabel* Label) {
    if (Label->Backward.Location) {
-      b(&Label->Backward);
+      return b(&Label->Backward);
    } else {
-      b(&Label->Forward);
+      return b(&Label->Forward);
    }
  }

@@ -126,25 +149,33 @@ public:
    UnconditionalBranch(Op, Imm);
  }

-  void bl(const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded bl(const BackwardLabel* 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;
+    if (Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0)) {
+      constexpr uint32_t Op = 0b1001'01 << 26;
+      UnconditionalBranch(Op, Imm >> 2);

-    UnconditionalBranch(Op, Imm >> 2);
+      return BranchEncodeSucceeded::Success;
+    }
+
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
  }
-  void bl(ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded bl(ForwardLabel* Label) {
    AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::B});
    constexpr uint32_t Op = 0b1001'01 << 26;

    UnconditionalBranch(Op, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
  }

-  void bl(BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded bl(BiDirectionalLabel* Label) {
    if (Label->Backward.Location) {
-      bl(&Label->Backward);
+      return bl(&Label->Backward);
    } else {
-      bl(&Label->Forward);
+      return bl(&Label->Forward);
    }
  }

@@ -155,28 +186,35 @@ public:
    CompareAndBranch(Op, s, rt, Imm);
  }

-  void cbz(ARMEmitter::Size s, ARMEmitter::Register rt, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded cbz(ARMEmitter::Size s, ARMEmitter::Register rt, const BackwardLabel* 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;
+    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) {
+      constexpr uint32_t Op = 0b0011'0100 << 24;
+      CompareAndBranch(Op, s, rt, Imm >> 2);
+      return BranchEncodeSucceeded::Success;
+    }

-    CompareAndBranch(Op, s, rt, Imm >> 2);
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
  }

-  void cbz(ARMEmitter::Size s, ARMEmitter::Register rt, ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded cbz(ARMEmitter::Size s, ARMEmitter::Register rt, ForwardLabel* Label) {
    AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::BC});

    constexpr uint32_t Op = 0b0011'0100 << 24;

    CompareAndBranch(Op, s, rt, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
  }

-  void cbz(ARMEmitter::Size s, ARMEmitter::Register rt, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded cbz(ARMEmitter::Size s, ARMEmitter::Register rt, BiDirectionalLabel* Label) {
    if (Label->Backward.Location) {
-      cbz(s, rt, &Label->Backward);
+      return cbz(s, rt, &Label->Backward);
    } else {
-      cbz(s, rt, &Label->Forward);
+      return cbz(s, rt, &Label->Forward);
    }
  }

@@ -186,28 +224,35 @@ public:
    CompareAndBranch(Op, s, rt, Imm);
  }

-  void cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, const BackwardLabel* 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;
+    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) {
+      constexpr uint32_t Op = 0b0011'0101 << 24;
+      CompareAndBranch(Op, s, rt, Imm >> 2);
+      return BranchEncodeSucceeded::Success;
+    }

-    CompareAndBranch(Op, s, rt, Imm >> 2);
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
  }

-  void cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, ForwardLabel* Label) {
    AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::BC});

    constexpr uint32_t Op = 0b0011'0101 << 24;

    CompareAndBranch(Op, s, rt, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
  }

-  void cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, BiDirectionalLabel* Label) {
    if (Label->Backward.Location) {
-      cbnz(s, rt, &Label->Backward);
+      return cbnz(s, rt, &Label->Backward);
    } else {
-      cbnz(s, rt, &Label->Forward);
+      return cbnz(s, rt, &Label->Forward);
    }
  }

@@ -217,28 +262,35 @@ public:

    TestAndBranch(Op, rt, Bit, Imm);
  }
-  void tbz(ARMEmitter::Register rt, uint32_t Bit, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded tbz(ARMEmitter::Register rt, uint32_t Bit, const BackwardLabel* 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;
+    if (Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0)) {
+      constexpr uint32_t Op = 0b0011'0110 << 24;
+      TestAndBranch(Op, rt, Bit, Imm >> 2);
+      return BranchEncodeSucceeded::Success;
+    }

-    TestAndBranch(Op, rt, Bit, Imm >> 2);
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
  }

-  void tbz(ARMEmitter::Register rt, uint32_t Bit, ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded tbz(ARMEmitter::Register rt, uint32_t Bit, ForwardLabel* Label) {
    AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::TEST_BRANCH});

    constexpr uint32_t Op = 0b0011'0110 << 24;

    TestAndBranch(Op, rt, Bit, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
  }

-  void tbz(ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded tbz(ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel* Label) {
    if (Label->Backward.Location) {
-      tbz(rt, Bit, &Label->Backward);
+      return tbz(rt, Bit, &Label->Backward);
    } else {
-      tbz(rt, Bit, &Label->Forward);
+      return tbz(rt, Bit, &Label->Forward);
    }
  }

@@ -247,27 +299,34 @@ public:

    TestAndBranch(Op, rt, Bit, Imm);
  }
-  void tbnz(ARMEmitter::Register rt, uint32_t Bit, const BackwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded tbnz(ARMEmitter::Register rt, uint32_t Bit, const BackwardLabel* 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;
+    if (Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0)) {
+      constexpr uint32_t Op = 0b0011'0111 << 24;
+      TestAndBranch(Op, rt, Bit, Imm >> 2);
+      return BranchEncodeSucceeded::Success;
+    }

-    TestAndBranch(Op, rt, Bit, Imm >> 2);
+    // Can't encode.
+    return BranchEncodeSucceeded::Failure;
  }

-  void tbnz(ARMEmitter::Register rt, uint32_t Bit, ForwardLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded tbnz(ARMEmitter::Register rt, uint32_t Bit, ForwardLabel* Label) {
    AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::TEST_BRANCH});
    constexpr uint32_t Op = 0b0011'0111 << 24;

    TestAndBranch(Op, rt, Bit, 0);
+
+    // Forward label doesn't know if it can encode until Bind.
+    return BranchEncodeSucceeded::Success;
  }

-  void tbnz(ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel* Label) {
+  [[nodiscard]] BranchEncodeSucceeded tbnz(ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel* Label) {
    if (Label->Backward.Location) {
-      tbnz(rt, Bit, &Label->Backward);
+      return tbnz(rt, Bit, &Label->Backward);
    } else {
-      tbnz(rt, Bit, &Label->Forward);
+      return tbnz(rt, Bit, &Label->Forward);
    }
  }

@@ -586,6 +586,15 @@ concept IsXOrWRegister = std::is_same_v<T, XRegister> || std::is_same_v<T, WRegi
 template<typename T>
 concept IsQOrDRegister = std::is_same_v<T, QRegister> || std::is_same_v<T, DRegister>;

+template<typename T>
+concept IsLabel = std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
+                  std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>;
+
+enum class BranchEncodeSucceeded {
+  Success,
+  Failure,
+};
+
 // Whether or not a given set of vector registers are sequential
 // in increasing order as far as the register file is concerned (modulo its size)
 //
@@ -638,19 +647,25 @@ public:

  // Bind a backward label to an address.
  // Address that is bound is the current emitter location.
-  void Bind(BackwardLabel* Label) {
+  [[nodiscard]] bool Bind(BackwardLabel* Label) {
    LOGMAN_THROW_A_FMT(Label->Location == nullptr, "Trying to bind a label twice");
    Label->Location = GetCursorAddress<uint8_t*>();
+
+    // Always binds because it is only storing a location.
+    return true;
  }

-  void Bind(const ForwardLabel::Reference* Label) {
+  [[nodiscard]] bool Bind(const ForwardLabel::Reference* Label) {
    uint8_t* CurrentAddress = GetCursorAddress<uint8_t*>();
    // Patch up the instructions
    switch (Label->Type) {
    case ForwardLabel::InstType::ADR: {
      uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
      int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
-      LOGMAN_THROW_A_FMT(IsADRRange(Imm), "Unscaled offset too large");
+      if (!IsADRRange(Imm)) {
+        // Can't bind.
+        return false;
+      }
      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;
@@ -662,7 +677,12 @@ public:
    case ForwardLabel::InstType::ADRP: {
      uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
      int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
-      LOGMAN_THROW_A_FMT(IsADRPRange(Imm) && IsADRPAligned(Imm), "Unscaled offset too large");
+
+      if (!(IsADRPRange(Imm) && IsADRPAligned(Imm))) {
+        // Can't bind.
+        return false;
+      }
+
      Imm >>= 12;
      uint32_t InstMask = 0b11 << 29 | 0b1111'1111'1111'1111'111 << 5;
      uint32_t Offset = static_cast<uint32_t>(Imm) & 0x3F'FFFF;
@@ -672,11 +692,13 @@ public:
      *Instruction = Inst;
      break;
    }
-
    case ForwardLabel::InstType::B: {
      uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->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");
+      if (!(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0))) {
+        // Can't bind.
+        return false;
+      }
      Imm >>= 2;
      uint32_t InstMask = 0x3FF'FFFF;
      uint32_t Offset = static_cast<uint32_t>(Imm) & InstMask;
@@ -686,11 +708,13 @@ public:

      break;
    }
-
    case ForwardLabel::InstType::TEST_BRANCH: {
      uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->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");
+      if (!(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0))) {
+        // Can't bind.
+        return false;
+      }
      Imm >>= 2;
      uint32_t InstMask = 0x3FFF;
      uint32_t Offset = static_cast<uint32_t>(Imm) & InstMask;
@@ -704,7 +728,10 @@ public:
    case ForwardLabel::InstType::RELATIVE_LOAD: {
      uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->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");
+      if (!(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0))) {
+        // Can't bind.
+        return false;
+      }
      Imm >>= 2;
      uint32_t InstMask = 0x7'FFFF;
      uint32_t Offset = static_cast<uint32_t>(Imm) & InstMask;
@@ -714,38 +741,44 @@ public:
      break;
    }
    case ForwardLabel::InstType::LONG_ADDRESS_GEN: {
-      uint32_t* Instructions = reinterpret_cast<uint32_t*>(Label->Location);
-      int64_t ImmInstOne = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[0]);
-      int64_t ImmInstTwo = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[1]);
-      auto OriginalOffset = GetCursorOffset();
+      const auto* Instructions = reinterpret_cast<uint32_t*>(Label->Location);
+      const auto ImmInstOne = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[0]);
+      const auto ImmInstTwo = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[1]);
+      const auto ImmInstThree = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[2]);
+      const auto OriginalOffset = GetCursorOffset();

-      auto InstOffset = GetCursorOffsetFromAddress(Instructions);
+      const auto InstOffset = GetCursorOffsetFromAddress(Instructions);
      SetCursorOffset(InstOffset);

      // We encoded the destination register in to the first instruction space.
      // Read it back.
      ARMEmitter::Register DestReg(Instructions[0]);

-      if (IsADRRange(ImmInstTwo)) {
-        // If within ADR range from the second instruction, then we can emit NOP+ADR
+      if (IsADRRange(ImmInstThree)) {
+        // If within ADR range from the third instruction, then we can emit NOP+NOP+ADR
        nop();
-        adr(DestReg, static_cast<uint32_t>(ImmInstTwo) & 0x7FFF);
-      } else if (IsADRPRange(ImmInstOne)) {
+        nop();
+        adr(DestReg, static_cast<uint32_t>(ImmInstThree) & 0x7FFF);
+      } else if (IsADRPRange(ImmInstTwo)) {

        // If within ADRP range from the first instruction, then we are /definitely/ in range for the second instruction.
        // First check if we are in non-offset range for second instruction.
        if (IsADRPAligned(reinterpret_cast<uint64_t>(CurrentAddress))) {
-          // We can emit nop + adrp
+          // We can emit nop + nop + adrp
+          nop();
+          nop();
+          adrp(DestReg, static_cast<uint32_t>(ImmInstThree >> 12) & 0x7FFF);
+        } else {
+          // Not aligned, need nop + adrp + add
          nop();
          adrp(DestReg, static_cast<uint32_t>(ImmInstTwo >> 12) & 0x7FFF);
-        } else {
-          // Not aligned, need adrp + add
-          adrp(DestReg, static_cast<uint32_t>(ImmInstOne >> 12) & 0x7FFF);
-          add(ARMEmitter::Size::i64Bit, DestReg, DestReg, ImmInstOne & 0xFFF);
+          add(ARMEmitter::Size::i64Bit, DestReg, DestReg, ImmInstTwo & 0xFFF);
        }
      } else {
-        LOGMAN_MSG_A_FMT("Unscaled offset is too large");
-        FEX_UNREACHABLE;
+        // Stinky path, we need to emit a movz+movk+movk sequence.
+        movz(ARMEmitter::Size::i64Bit, DestReg, uint32_t(ImmInstOne >> 32) & 0x7FFF, 32);
+        movk(ARMEmitter::Size::i64Bit, DestReg, uint32_t(ImmInstOne >> 16) & 0xFFFF, 16);
+        movk(ARMEmitter::Size::i64Bit, DestReg, uint32_t(ImmInstOne) & 0xFFFF);
      }

      SetCursorOffset(OriginalOffset);
@@ -753,27 +786,41 @@ public:
    }
    default: LOGMAN_MSG_A_FMT("Unexpected inst type in label fixup");
    }
+
+    return true;
  }

  // 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.
-  void Bind(ForwardLabel* Label) {
+  [[nodiscard]] bool Bind(ForwardLabel* Label) {
+    bool Bound = true;
    if (Label->FirstInst.Location) {
-      Bind(&Label->FirstInst);
+      Bound &= Bind(&Label->FirstInst);
    }
    for (auto& Inst : Label->Insts) {
-      Bind(&Inst);
+      Bound &= Bind(&Inst);
    }
+
+    return Bound;
  }

  // Bind a bidirectional location to a location.
  // Binds both forwards and backwards depending on how the label was used.
-  void Bind(BiDirectionalLabel* Label) {
+  [[nodiscard]] bool Bind(BiDirectionalLabel* Label) {
+    bool Bound = true;
    if (!Label->Backward.Location) {
-      Bind(&Label->Backward);
+      Bound &= Bind(&Label->Backward);
    }
-    Bind(&Label->Forward);
+    Bound &= Bind(&Label->Forward);
+
+    return Bound;
+  }
+
+  static constexpr Condition InvertCondition(Condition cond) {
+    // These behave as always, so it makes no sense to allow inverting these.
+    LOGMAN_THROW_A_FMT(cond != Condition::CC_AL && cond != Condition::CC_NV, "Cannot invert CC_AL or CC_NV");
+    return static_cast<Condition>(FEXCore::ToUnderlying(cond) ^ 1);
  }

 #include <CodeEmitter/VixlUtils.inl>
@@ -5125,7 +5125,7 @@ private:
  requires (std::is_same_v<T, float> || std::is_same_v<T, double>)
  [[nodiscard]]
  static bool IsValidFPValueForImm8(T value) {
-    const uint64_t bits = FEXCore::BitCast<FloatToEquivalentUInt<T>>(value);
+    const uint64_t bits = std::bit_cast<FloatToEquivalentUInt<T>>(value);
    const uint64_t datasize_idx = FEXCore::ilog2(sizeof(T)) - 1;

    static constexpr std::array mantissa_masks {
@@ -5171,7 +5171,7 @@ protected:
    LOGMAN_THROW_A_FMT(IsValidFPValueForImm8(value), "Value ({}) cannot be encoded into an 8-bit immediate", value);
 #endif

-    const auto bits = FEXCore::BitCast<uint32_t>(value);
+    const auto bits = std::bit_cast<uint32_t>(value);
    const auto sign = (bits & 0x80000000) >> 24;
    const auto expb2 = (bits & 0x20000000) >> 23;
    const auto b5_to_0 = (bits >> 19) & 0x3F;
@@ -5184,7 +5184,7 @@ protected:
    LOGMAN_THROW_A_FMT(IsValidFPValueForImm8(value), "Value ({}) cannot be encoded into an 8-bit immediate", value);
 #endif

-    const auto bits = FEXCore::BitCast<uint64_t>(value);
+    const auto bits = std::bit_cast<uint64_t>(value);
    const auto sign = (bits & 0x80000000'00000000) >> 56;
    const auto expb2 = (bits & 0x20000000'00000000) >> 55;
    const auto b5_to_0 = (bits >> 48) & 0x3F;
@@ -4,7 +4,8 @@ file(GLOB GEN_CONFIG_SOURCES CONFIGURE_DEPENDS *.json.in)
 # Any application configuration json file gets installed
 foreach(CONFIG_SRC ${CONFIG_SOURCES})
  install(FILES ${CONFIG_SRC}
-    DESTINATION ${DATA_DIRECTORY}/AppConfig/)
+    DESTINATION ${DATA_DIRECTORY}/AppConfig/
+    COMPONENT Runtime)
 endforeach()

 # Any configuration file json file that needs to be generated
@@ -14,12 +15,10 @@ foreach(GEN_CONFIG_SRC ${GEN_CONFIG_SOURCES})
  get_filename_component(CONFIG_NAME ${GEN_CONFIG_SRC} NAME_WLE)

  # Configure it
-  configure_file(
-    ${GEN_CONFIG_SRC}
-    ${CMAKE_BINARY_DIR}/Data/AppConfig/${CONFIG_NAME})
+  configure_file(${GEN_CONFIG_SRC} ${CMAKE_BINARY_DIR}/Data/AppConfig/${CONFIG_NAME})

  # Then install the configured json
-  install(
-    FILES ${CMAKE_BINARY_DIR}/Data/AppConfig/${CONFIG_NAME}
-    DESTINATION ${DATA_DIRECTORY}/AppConfig/)
+  install(FILES ${CMAKE_BINARY_DIR}/Data/AppConfig/${CONFIG_NAME}
+    DESTINATION ${DATA_DIRECTORY}/AppConfig/
+    COMPONENT Runtime)
 endforeach()
@@ -1,3 +0,0 @@
-x86 and x86-64 Linux emulator
-
-FEX allows you to run x86 applications on ARM64 Linux devices. It offers broad compatibility with both 32-bit and 64-bit binaries, and it can be used alongside Wine/Proton to play Windows games.
@@ -1,18 +0,0 @@
-#!/bin/sh
-set -e
-update_binfmt() {
-  # Check for update-binfmts
-  command -v update-binfmts >/dev/null || return 0
-
-  # Setup binfmt_misc
-  update-binfmts --import FEX-x86
-  update-binfmts --import FEX-x86_64
-}
-
-# Install FEXInterpreter hardlink
-# Needs to be done before setting up binfmt_misc
-ln -f /usr/bin/FEXLoader /usr/bin/FEXInterpreter
-
-if [ $(uname -m) = 'aarch64' ]; then
-  update_binfmt
-fi
@@ -1,17 +0,0 @@
-#!/bin/sh
-set -e
-update_binfmt() {
-  # Check for update-binfmts
-  command -v update-binfmts >/dev/null || return 0
-
-  # Uninstall
-  update-binfmts --unimport FEX-x86
-  update-binfmts --unimport FEX-x86_64
-}
-
-if [ $(uname -m) = 'aarch64' ]; then
-  update_binfmt
-fi
-
-# Remove FEXInterpreter hardlink
-unlink /usr/bin/FEXInterpreter
@@ -1 +0,0 @@
-activate-noawait ldconfig
@@ -0,0 +1,18 @@
+# SPDX-License-Identifier: MIT
+
+include(FindPackageHandleStandardArgs)
+
+find_package(PkgConfig QUIET)
+pkg_search_module(xxhash QUIET IMPORTED_TARGET xxhash libxxhash)
+find_package_handle_standard_args(xxhash
+    REQUIRED_VARS xxhash_LINK_LIBRARIES
+    VERSION_VAR xxhash_VERSION
+)
+
+if (xxhash_FOUND AND NOT TARGET xxHash::xxhash)
+    if (TARGET PkgConfig::xxhash)
+        add_library(xxHash::xxhash ALIAS PkgConfig::xxhash)
+    else()
+        add_library(xxHash::xxhash ALIAS xxhash)
+    endif()
+endif()
@@ -0,0 +1,15 @@
+# SPDX-License-Identifier: MIT
+
+# This applies some common linker options that reduce code size and linking time in Release mode. Namely:
+# --gc-sections: Linktime garbage collection, discards unused sections from the final output
+# --strip-all  : Similar to running `strip`, discards the symbol table from the final output
+# --as-needed  : Only includes libraries that are actually needed in the final output.
+
+macro(LinkerGC target)
+  if (CMAKE_BUILD_TYPE MATCHES "RELEASE")
+    target_link_options(${target} PRIVATE
+      "LINKER:--gc-sections"
+      "LINKER:--strip-all"
+      "LINKER:--as-needed")
+  endif()
+endmacro()
@@ -14,7 +14,7 @@ RUN mkdir build

 ARG CC=clang-13
 ARG CXX=clang++-13
-RUN cmake -DCMAKE_INSTALL_PREFIX=/usr -DCMAKE_BUILD_TYPE=Release -DUSE_LINKER=lld -DENABLE_LTO=True -DBUILD_TESTS=False -DENABLE_ASSERTIONS=False -G Ninja .
+RUN cmake -DCMAKE_INSTALL_PREFIX=/usr -DCMAKE_BUILD_TYPE=Release -DUSE_LINKER=lld -DENABLE_LTO=True -DBUILD_TESTING=False -DENABLE_ASSERTIONS=False -G Ninja .
 RUN ninja

 WORKDIR /FEX/build
@@ -3,23 +3,21 @@ function(GenBinFmt Name)
  get_filename_component(FMT_NAME ${Name} NAME_WE)

  # Configure it
-  configure_file(
-    ${Name}
-    ${CMAKE_BINARY_DIR}/Data/binfmts/${FMT_NAME})
+  configure_file(${Name} ${CMAKE_BINARY_DIR}/Data/binfmts/${FMT_NAME})

  # Then install the configured binfmt
-  install(
-    FILES ${CMAKE_BINARY_DIR}/Data/binfmts/${FMT_NAME}
-    DESTINATION ${CMAKE_INSTALL_PREFIX}/share/binfmts/)
+  install(FILES ${CMAKE_BINARY_DIR}/Data/binfmts/${FMT_NAME}
+    DESTINATION ${CMAKE_INSTALL_PREFIX}/share/binfmts/
+    COMPONENT Runtime)
 endfunction()

 if (NOT USE_LEGACY_BINFMTMISC)
  configure_file(FEX-x86.conf.in ${CMAKE_BINARY_DIR}/Data/binfmts/FEX-x86.conf)
  configure_file(FEX-x86_64.conf.in ${CMAKE_BINARY_DIR}/Data/binfmts/FEX-x86_64.conf)

-  install(
-    FILES ${CMAKE_BINARY_DIR}/Data/binfmts/FEX-x86.conf ${CMAKE_BINARY_DIR}/Data/binfmts/FEX-x86_64.conf
-    DESTINATION ${CMAKE_INSTALL_PREFIX}/lib/binfmt.d/)
+  install(FILES ${CMAKE_BINARY_DIR}/Data/binfmts/FEX-x86.conf ${CMAKE_BINARY_DIR}/Data/binfmts/FEX-x86_64.conf
+    DESTINATION ${CMAKE_INSTALL_PREFIX}/lib/binfmt.d/
+    COMPONENT Runtime)
 else()
  GenBinFmt(FEX-x86.in)
  GenBinFmt(FEX-x86_64.in)
@@ -1 +1 @@
-:FEX-x86:M:0:\x7fELF\x01\x01\x01\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02\x00\x03\x00:\xff\xff\xff\xff\xff\xfe\xfe\x00\x00\x00\x00\xff\xff\xff\xff\xff\xfe\xff\xff\xff:@CMAKE_INSTALL_PREFIX@/bin/FEXInterpreter:POCF
+:FEX-x86:M:0:\x7fELF\x01\x01\x01\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02\x00\x03\x00:\xff\xff\xff\xff\xff\xfe\xfe\x00\x00\x00\x00\xff\xff\xff\xff\xff\xfe\xff\xff\xff:@CMAKE_INSTALL_PREFIX@/bin/FEX:POCF
@@ -1,5 +1,5 @@
 package fex
-interpreter @CMAKE_INSTALL_PREFIX@/bin/FEXInterpreter
+interpreter @CMAKE_INSTALL_PREFIX@/bin/FEX
 magic \x7fELF\x01\x01\x01\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02\x00\x03\x00
 offset 0
 mask \xff\xff\xff\xff\xff\xfe\xfe\x00\x00\x00\x00\xff\xff\xff\xff\xff\xfe\xff\xff\xff
@@ -1 +1 @@
-:FEX-x86_64:M:0:\x7fELF\x02\x01\x01\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02\x00\x3e\x00:\xff\xff\xff\xff\xff\xfe\xfe\x00\x00\x00\x00\xff\xff\xff\xff\xff\xfe\xff\xff\xff:@CMAKE_INSTALL_PREFIX@/bin/FEXInterpreter:POCF
+:FEX-x86_64:M:0:\x7fELF\x02\x01\x01\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02\x00\x3e\x00:\xff\xff\xff\xff\xff\xfe\xfe\x00\x00\x00\x00\xff\xff\xff\xff\xff\xfe\xff\xff\xff:@CMAKE_INSTALL_PREFIX@/bin/FEX:POCF
@@ -1,5 +1,5 @@
 package fex
-interpreter @CMAKE_INSTALL_PREFIX@/bin/FEXInterpreter
+interpreter @CMAKE_INSTALL_PREFIX@/bin/FEX
 magic \x7fELF\x02\x01\x01\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02\x00\x3e\x00
 offset 0
 mask \xff\xff\xff\xff\xff\xfe\xfe\x00\x00\x00\x00\xff\xff\xff\xff\xff\xfe\xff\xff\xff
@@ -2,8 +2,8 @@

 let
  toolchain = pkgs.fetchzip {
-    url = "https://github.com/bylaws/llvm-mingw/releases/download/20250305/llvm-mingw-20250305-ucrt-ubuntu-20.04-aarch64.tar.xz";
-    sha256 = "sha256-cA03/ab9O61eO9+S2JzIXD4V0HzTXK5/AYyxW2d73Po=";
+    url = "https://github.com/bylaws/llvm-mingw/releases/download/20250920/llvm-mingw-20250920-ucrt-ubuntu-22.04-aarch64.tar.xz";
+    sha256 = "sha256-LaojKjC8KzY+soW5u6eoDoXE3qtYk9Ejr7M3enTqRAE=";
  };

  cmakeToolchainFile = pkgs.substitute {
@@ -45,7 +45,7 @@ pkgs.mkShell {
    fi
  '';

-  # E.g. cmake $FEX_CMAKE_TOOLCHAIN_ARM64EC -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr -DENABLE_LTO=False -DBUILD_TESTS=False
+  # E.g. cmake $FEX_CMAKE_TOOLCHAIN_ARM64EC -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr -DENABLE_LTO=False -DBUILD_TESTING=False
  FEX_CMAKE_TOOLCHAIN_ARM64EC = "--toolchain ${cmakeToolchainFile} -DMINGW_TRIPLE=arm64ec-w64-mingw32 -DCMAKE_INSTALL_LIBDIR=/usr/lib/wine/aarch64-windows";
  FEX_CMAKE_TOOLCHAIN_WOW64 = "--toolchain ${cmakeToolchainFile} -DMINGW_TRIPLE=aarch64-w64-mingw32 -DCMAKE_INSTALL_LIBDIR=/usr/lib/wine/aarch64-windows";
  FEX_MESON_CROSSFILE = "--cross-file ${mesonCrossFile}";
@@ -18,4 +18,4 @@ then
 fi

 set -o xtrace
-cmake $FEX_CMAKE_TOOLCHAIN_WOW64 -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr -DENABLE_LTO=False -DBUILD_TESTS=False $@
+cmake $FEX_CMAKE_TOOLCHAIN_WOW64 -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr -DENABLE_LTO=False -DBUILD_TESTING=False $@
@@ -18,4 +18,4 @@ then
 fi

 set -o xtrace
-cmake $FEX_CMAKE_TOOLCHAIN_ARM64EC -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr -DENABLE_LTO=False -DBUILD_TESTS=False $@
+cmake $FEX_CMAKE_TOOLCHAIN_ARM64EC -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr -DENABLE_LTO=False -DBUILD_TESTING=False $@
@@ -14,4 +14,4 @@ fi
 rm -rf unittests/FEXLinuxTests

 set -o xtrace
-cmake . $FEX_CMAKE_TOOLCHAINS -DBUILD_TESTS=ON -DBUILD_FEX_LINUX_TESTS=ON
+cmake . $FEX_CMAKE_TOOLCHAINS -DBUILD_TESTING=ON -DBUILD_FEX_LINUX_TESTS=ON
@@ -1,5 +1,5 @@

-set (SRCS
+add_library(softfloat_3e STATIC
  # F80 support
  src/extF80_add.c
  src/extF80_div.c
@@ -84,7 +84,7 @@ set (SRCS
  src/s_normSubnormalF32Sig.c
  src/s_f32UIToCommonNaN.c)

-if (_M_ARM_64 AND HAS_CLANG_PRESERVE_ALL)
+if (ARCHITECTURE_arm64 AND HAS_CLANG_PRESERVE_ALL)
  list(APPEND DEFINES "-DFEXCORE_PRESERVE_ALL_ATTR=__attribute__((preserve_all));-DFEXCORE_HAS_PRESERVE_ALL_ATTR=1")
 else()
  list(APPEND DEFINES "-DFEXCORE_PRESERVE_ALL_ATTR=;-DFEXCORE_HAS_PRESERVE_ALL_ATTR=0")
@@ -92,7 +92,6 @@ endif()

 list(APPEND DEFINES "-DSOFTFLOAT_BUILTIN_CLZ=1;-DINLINE=static inline;-DINLINE_LEVEL=4;-DSOFTFLOAT_FAST_INT64=1;-DSOFTFLOAT_FAST_DIV32TO16=1;-DSOFTFLOAT_FAST_DIV64TO32=1")

-add_library(softfloat_3e STATIC ${SRCS})
 target_include_directories(softfloat_3e PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/include/)
 target_include_directories(softfloat_3e PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/include/SoftFloat-3e/)
 target_compile_definitions(softfloat_3e PUBLIC ${DEFINES})
@@ -1,4 +1,4 @@
-set(SRCS_128BIT
+add_library(cephes_128bit STATIC
  src/128bit/Impl.cpp
  src/128bit/atanll.c
  src/128bit/constll.c
@@ -11,7 +11,6 @@ set(SRCS_128BIT
  src/128bit/tanll.c)

 # 128-bit library
-add_library(cephes_128bit STATIC ${SRCS_128BIT})
 target_link_libraries(cephes_128bit softfloat_3e)
 target_include_directories(cephes_128bit PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/include/)
 target_compile_options(cephes_128bit PRIVATE -fno-builtin)
@@ -306,9 +306,9 @@ typing-extensions==4.14.1 \
    --hash=sha256:38b39f4aeeab64884ce9f74c94263ef78f3c22467c8724005483154c26648d36 \
    --hash=sha256:d1e1e3b58374dc93031d6eda2420a48ea44a36c2b4766a4fdeb3710755731d76
    # via pygithub
-urllib3==2.5.0 \
-    --hash=sha256:3fc47733c7e419d4bc3f6b3dc2b4f890bb743906a30d56ba4a5bfa4bbff92760 \
-    --hash=sha256:e6b01673c0fa6a13e374b50871808eb3bf7046c4b125b216f6bf1cc604cff0dc
+urllib3==2.6.0 \
+    --hash=sha256:c90f7a39f716c572c4e3e58509581ebd83f9b59cced005b7db7ad2d22b0db99f \
+    --hash=sha256:cb9bcef5a4b345d5da5d145dc3e30834f58e8018828cbc724d30b4cb7d4d49f1
    # via
    #   -r requirements_formatting.txt.in
    #   pygithub
@@ -2,7 +2,7 @@ black~=25.1
 darker==2.1.1
 PyGithub==2.6.1
 cryptography>=43.0.1
-urllib3>=2.5.0
+urllib3>=2.6.0
 requests>=2.32.4
 idna>=3.7
 certifi>=2024.7.4
@@ -1,16 +1,16 @@
 cmake_minimum_required(VERSION 3.14)
-set (PROJECT_NAME FEXCore)
+set(PROJECT_NAME FEXCore)
 project(${PROJECT_NAME}
  VERSION 0.01
  LANGUAGES CXX)

 if (CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64")
-  set(_M_X86_64 1)
+  set(ARCHITECTURE_x86_64 1)
  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mcx16")
 endif()

 if (CMAKE_SYSTEM_PROCESSOR MATCHES "^aarch64|^arm64|^armv8\.*")
-  set(_M_ARM_64 1)
+  set(ARCHITECTURE_arm64 1)
 endif()

 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
@@ -25,44 +25,15 @@ include(CheckIncludeFileCXX)
 include(CheckCXXSourceCompiles)

 if (EXISTS ${CMAKE_CURRENT_DIR}/External/vixl/)
-    # Useful to have for freestanding libFEXCore
-    add_subdirectory(External/vixl/)
-    include_directories(External/vixl/src/)
+  # Useful to have for freestanding libFEXCore
+  add_subdirectory(External/vixl/)
+  include_directories(External/vixl/src/)
 endif()

 set(CMAKE_CXX_STANDARD 20)
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)

-set(GIT_SHORT_HASH "Unknown")
-set(GIT_DESCRIBE_STRING "FEX-Unknown")
-
-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(
-  ${CMAKE_CURRENT_SOURCE_DIR}/include/git_version.h.in
+configure_file(${CMAKE_CURRENT_SOURCE_DIR}/include/git_version.h.in
  ${CMAKE_BINARY_DIR}/generated/git_version.h)

 include_directories(${CMAKE_BINARY_DIR}/generated)
@@ -74,10 +45,12 @@ add_compile_options($<$<COMPILE_LANGUAGE:CXX>:-fno-strict-aliasing> $<$<COMPILE_

 add_subdirectory(Source/)

-install (DIRECTORY include/FEXCore ${CMAKE_BINARY_DIR}/include/FEXCore
-  DESTINATION include
-  COMPONENT Development)
+if (NOT BUILD_STEAM_SUPPORT)
+  install (DIRECTORY include/FEXCore ${CMAKE_BINARY_DIR}/include/FEXCore
+    DESTINATION include
+    COMPONENT Development)
+endif()

-if (BUILD_TESTS)
+if (BUILD_TESTING)
  add_subdirectory(unittests/)
 endif()
@@ -118,41 +118,6 @@ def print_man_env_option(name, desc, default, no_json_key):
    output_man.write("\\fBdefault:\\fR {0}\n".format(default))
    output_man.write(".Pp\n\n")

-def print_man_options(options):
-    output_man.write(".Sh OPTIONS\n")
-    output_man.write(".Bl -tag -width -indent\n")
-    for op_group, group_vals in options.items():
-        for op_key, op_vals in group_vals.items():
-            short = None
-            long = op_key.lower()
-
-            if ("ShortArg" in op_vals):
-                short = op_vals["ShortArg"]
-
-            default = op_vals["Default"]
-            value_type = op_vals["Type"]
-
-            # Textual default rather than enum based
-            if ("TextDefault" in op_vals):
-                default = op_vals["TextDefault"]
-
-            if (value_type == "str" or value_type == "strarray" or value_type == "strenum"):
-                # Wrap the string argument in quotes
-                default = "'" + default + "'"
-            print_man_option(
-                short,
-                long,
-                op_vals["Desc"],
-                default
-            )
-            if (value_type == "strenum"):
-                Enums = op_vals["Enums"]
-                output_man.write("\\fBAvailable Options:\\fR\n")
-                output_man.write(", ".join(f"{enum_op_val}" for [_, enum_op_val] in Enums.items()))
-                output_man.write("\n.sp\n")
-
-    output_man.write(".El\n")
-
 def print_man_environment(options):
    output_man.write(".Sh ENVIRONMENT\n")
    output_man.write(".Bl -tag -width -indent\n")
@@ -194,7 +159,7 @@ def print_man_environment_tail():
    "By default FEX will look in {$HOME, $XDG_CONFIG_HOME}/.fex-emu/",
    "This will override the full path",
    "If FEX_PORTABLE is declared then relative paths are also supported",
-    "For FEXInterpreter: Relative to the FEXInterpreter binary",
+    "For FEX: Relative to the FEX binary",
    "For WINE: Relative to %LOCALAPPDATA%"
    ],
    "''", True)
@@ -208,7 +173,7 @@ def print_man_environment_tail():
    "One must be careful with this option as it will override any applications that load with execve as well"
    "If you need to support applications that execve then use FEX_APP_CONFIG_LOCATION instead"
    "If FEX_PORTABLE is declared then relative paths are also supported",
-    "For FEXInterpreter: Relative to the FEXInterpreter binary",
+    "For FEX: Relative to the FEX binary",
    "For WINE: Relative to %LOCALAPPDATA%"
    ],
    "''", True)
@@ -227,33 +192,34 @@ def print_man_environment_tail():
    "PORTABLE",
    [
    "Allows FEX to run without installation. Global locations for configuration and binfmt_misc are ignored.",
-    "For FEXInterpreter on Linux:",
-    "These files are instead read from <FEXInterpreterPath>/fex-emu/ by default.",
+    "For FEX on Linux:",
+    "These files are instead read from <FEXPath>/fex-emu/ by default.",
    "For Arm64ec/Wow64 WINE builds:",
    "These files are instead read from $LOCALAPPDATA/fex-emu/ by default.",
    "For further customization, see FEX_APP_CONFIG_LOCATION and FEX_APP_DATA_LOCATION."
    ],
    "''", True)

+    print_man_env_option(
+    "APP_CACHE_LOCATION",
+    [
+    "Allows the user to override where FEX stores and loads cache files",
+    "By default FEX will look in $XDG_CACHE_HOME/fex-emu/ or $HOME/.cache/fex-emu/",
+    "This will override the full path, trailing forward-slash is expected to exist",
+    ],
+    "''", True)
+
 def print_man_header():
    header ='''.Dd {0}
 .Dt FEX
 .Os Linux
 .Sh NAME
-.Nm FEXLoader
-.Nm FEXInterpreter
+.Nm FEX
 .Nm FEXBash
 .Nd Fast x86-64 and x86 emulation.
 .Sh SYNOPSIS
 .Nm
-.Op options
-.Op Ar --
-.Ar Application
-<args> ...
-.Pp
-.Nm FEXInterpreter
-.Ar Application
-<args> ...
+.Ar <args> ...
 .Pp
 .Nm FEXBash
 .Ar <args> ...
@@ -361,82 +327,6 @@ def print_config_option(type, group_name, json_name, default_value, short, choic

    output_argloader.write("\n");

-def print_argloader_options(options):
-    output_argloader.write("#ifdef BEFORE_PARSE\n")
-    output_argloader.write("#undef BEFORE_PARSE\n")
-    for op_group, group_vals in options.items():
-        for op_key, op_vals in group_vals.items():
-            default = op_vals["Default"]
-
-            if (op_vals["Type"] == "str" or op_vals["Type"] == "strarray" or op_vals["Type"] == "strenum"):
-                # Wrap the string argument in quotes
-                default = "\"" + default + "\""
-
-            # Textual default rather than enum based
-            if ("TextDefault" in op_vals):
-                default = "\"" + op_vals["TextDefault"] + "\""
-
-            short = None
-            choices = None
-
-            if ("ShortArg" in op_vals):
-                short = op_vals["ShortArg"]
-            if ("Choices" in op_vals):
-                choices = op_vals["Choices"]
-
-            print_config_option(
-                op_vals["Type"],
-                op_group,
-                op_key,
-                default,
-                short,
-                choices,
-                op_vals["Desc"])
-
-        output_argloader.write("\n")
-    output_argloader.write("#endif\n")
-
-def print_parse_argloader_options(options):
-    output_argloader.write("#ifdef AFTER_PARSE\n")
-    output_argloader.write("#undef AFTER_PARSE\n")
-    for op_group, group_vals in options.items():
-        for op_key, op_vals in group_vals.items():
-            output_argloader.write("if (Options.is_set_by_user(\"{0}\")) {{\n".format(op_key))
-
-            value_type = op_vals["Type"]
-            NeedsString = False
-            conversion_func = "fextl::fmt::format(\"{}\", "
-            if ("ArgumentHandler" in op_vals):
-                NeedsString = True
-                conversion_func = "FEXCore::Config::Handler::{0}(".format(op_vals["ArgumentHandler"])
-            if (value_type == "str"):
-                NeedsString = True
-                conversion_func = "std::move("
-            if (value_type == "bool"):
-                # boolean values need a decimal specifier. Otherwise fmt prints strings.
-                conversion_func = "fextl::fmt::format(\"{:d}\", "
-
-            if (value_type == "strenum"):
-                output_argloader.write("\tfextl::string UserValue = Options[\"{0}\"];\n".format(op_key))
-                output_argloader.write("\tSet(FEXCore::Config::ConfigOption::CONFIG_{}, FEXCore::Config::EnumParser<FEXCore::Config::{}ConfigPair>(FEXCore::Config::{}_EnumPairs, UserValue));\n".format(op_key.upper(), op_key, op_key))
-            elif (value_type == "strarray"):
-                # these need a bit more help
-                output_argloader.write("\tauto Array = Options.all(\"{0}\");\n".format(op_key))
-                output_argloader.write("\tfor (auto iter = Array.begin(); iter != Array.end(); ++iter) {\n")
-                output_argloader.write("\t\tAppendStrArrayValue(FEXCore::Config::ConfigOption::CONFIG_{0}, *iter);\n".format(op_key.upper()))
-                output_argloader.write("\t}\n")
-            else:
-                if (NeedsString):
-                    output_argloader.write("\tfextl::string UserValue = Options[\"{0}\"];\n".format(op_key))
-                else:
-                    output_argloader.write("\t{0} UserValue = Options.get(\"{1}\");\n".format(value_type, op_key))
-
-                output_argloader.write("\tSet(FEXCore::Config::ConfigOption::CONFIG_{0}, {1}UserValue));\n".format(op_key.upper(), conversion_func))
-            output_argloader.write("}\n")
-
-    output_argloader.write("#endif\n")
-
-
 def print_parse_envloader_options(options):
    output_argloader.write("#ifdef ENVLOADER\n")
    output_argloader.write("#undef ENVLOADER\n")
@@ -517,41 +407,6 @@ def print_parse_enum_options(options):

    output_argloader.write("#endif\n")

-def check_for_duplicate_options(options):
-    short_map = []
-    long_map = []
-
-    # Spin through all the items and see if we have a duplicate option
-    for op_group, group_vals in options.items():
-        for op_key, op_vals in group_vals.items():
-            short = None
-            long = op_key.lower()
-            long_invert = None
-            if ("ShortArg" in op_vals):
-                short = op_vals["ShortArg"]
-            if (op_vals["Type"] == "bool"):
-                long_invert = "no-" + long
-
-            # Check for short key duplication
-            if (short != None):
-                if (short in short_map):
-                    raise Exception("Short config '{0}' for option '{1}' has duplicate entry!".format(short, op_key))
-                else:
-                    short_map.append(short)
-
-            # Check for long key duplication
-            if (long in long_map):
-                raise Exception("Long config '{0}' has duplicate entry!".format(long))
-            else:
-                long_map.append(long)
-
-            # Check for long key duplication
-            if (long_invert != None):
-                if (long_invert in long_map):
-                    raise Exception("Long config '{0}' has duplicate entry!".format(long_invert))
-                else:
-                    long_map.append(long_invert)
-
 if (len(sys.argv) < 5):
    sys.exit()

@@ -568,8 +423,6 @@ json_object = json.loads(json_text)
 options = json_object["Options"]
 unnamed_options = json_object["UnnamedOptions"]

-check_for_duplicate_options(options)
-
 # Generate config include file
 output_file = open(output_filename, "w")
 print_header()
@@ -581,7 +434,6 @@ output_file.close()
 # Generate man file
 output_man = open(output_man_page, "w")
 print_man_header()
-print_man_options(options)
 print_man_environment(options)
 print_man_tail()

@@ -589,8 +441,6 @@ output_man.close()

 # Generate argument loader code
 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);
@@ -58,10 +58,10 @@ class OpDefinition:
    JITDispatch: bool
    JITDispatchOverride: str
    TiedSource: int
-    Inline: list
-    Arguments: list
-    EmitValidation: list
-    Desc: list
+    Inline: list[str]
+    Arguments: list[OpArgument]
+    EmitValidation: list[str]
+    Desc: list[str]

    def __init__(self):
        self.Name = None
@@ -92,19 +92,14 @@ class OpDefinition:
        attrs = vars(self)
        print(", ".join("%s: %s" % item for item in attrs.items()))

-IRTypesToCXX = {}
-CXXTypeToIR = {}
-IROps = []
+IRTypesToCXX: dict[str, IRType] = {}
+CXXTypeToIR: dict[str, IRType] = {}
+IROps: list[OpDefinition] = []

-IROpNameMap = {}
+IROpNameSet: set[str] = set()

-def is_ssa_type(type):
-    if (type == "SSA" or
-       type == "GPR" or
-       type == "GPRPair" or
-       type == "FPR"):
-       return True
-    return False
+def is_ssa_type(op_type: str):
+    return op_type in {"SSA", "GPR", "GPRPair", "FPR"}

 def parse_irtypes(irtypes):
    for op_key, op_val in irtypes.items():
@@ -219,11 +214,8 @@ def parse_ops(ops):
                        OpArg.DefaultInitializer = DefaultInit[1][:-1]

                    # If SSA type then we can generate validation for this op
-                    if (OpArg.IsSSA and
-                        (OpArg.Type == "GPR" or
-                        OpArg.Type == "GPRPair" or
-                        OpArg.Type == "FPR")):
-                        OpDef.EmitValidation.append(f"GetOpRegClass({ArgName}) == InvalidClass || WalkFindRegClass({ArgName}) == {OpArg.Type}Class")
+                    if OpArg.IsSSA and OpArg.Type in {"GPR", "GPRPair", "FPR"}:
+                        OpDef.EmitValidation.append(f"GetOpRegClass({ArgName}) == RegClass::Invalid || WalkFindRegClass({ArgName}) == RegClass::{OpArg.Type}")

                    OpArg.Name = ArgName
                    OpArg.NameWithPrefix = NameWithPrefix
@@ -296,22 +288,29 @@ def parse_ops(ops):
            #OpDef.print()

            # Error on duplicate op
-            if OpDef.Name in IROpNameMap:
+            if OpDef.Name in IROpNameSet:
                ExitError("Duplicate Op defined! {}".format(OpDef.Name))

            IROps.append(OpDef)
-            IROpNameMap[OpDef.Name] = 1
+            IROpNameSet.add(OpDef.Name)

 # Print out enum values
-def print_enums():
+def print_enums(enums):
    output_file.write("#ifdef IROP_ENUM\n")
    output_file.write("enum IROps : uint16_t {\n")
-
    for op in IROps:
        output_file.write("\tOP_{},\n" .format(op.Name.upper()))
-
    output_file.write("};\n")

+    for name, members in enums.items():
+        output_file.write(f"enum {name} {{\n")
+        for member in members:
+            if member:
+                output_file.write(f"\t{member}\n")
+            else:
+                output_file.write("\n")
+        output_file.write("};\n\n")
+
    output_file.write("#undef IROP_ENUM\n")
    output_file.write("#endif\n\n")

@@ -408,7 +407,7 @@ def print_ir_sizes():
    [[nodiscard, gnu::const]] std::string_view const& GetName(IROps Op);
    [[nodiscard, gnu::const]] uint8_t GetArgs(IROps Op);
    [[nodiscard, gnu::const]] uint8_t GetRAArgs(IROps Op);
-    [[nodiscard, gnu::const]] FEXCore::IR::RegisterClassType GetRegClass(IROps Op);
+    [[nodiscard, gnu::const]] FEXCore::IR::RegClass GetRegClass(IROps Op);
    [[nodiscard, gnu::const]] bool HasSideEffects(IROps Op);
    [[nodiscard, gnu::const]] bool ImplicitFlagClobber(IROps Op);
    [[nodiscard, gnu::const]] bool GetHasDest(IROps Op);
@@ -422,30 +421,29 @@ def print_ir_sizes():
 def print_ir_reg_classes():
    output_file.write("#ifdef IROP_REG_CLASSES_IMPL\n")

-    output_file.write("constexpr std::array<FEXCore::IR::RegisterClassType, IROps::OP_LAST + 1> IRRegClasses = {\n")
+    output_file.write("constexpr std::array<FEXCore::IR::RegClass, IROps::OP_LAST + 1> IRRegClasses = {\n")
    for op in IROps:
        if op.Name == "Last":
-            output_file.write("\tFEXCore::IR::InvalidClass,\n")
+            output_file.write("\tRegClass::Invalid,\n")
        else:
-            Class = "Invalid"
-            if op.HasDest and op.DestType == None:
+            if op.HasDest and op.DestType is None:
                ExitError("IR op {} has destination with no destination class".format(op.Name))

            if op.HasDest and op.DestType == "SSA": # Special case SSA type
-                output_file.write("\tFEXCore::IR::ComplexClass,\n")
+                output_file.write("\tRegClass::Complex,\n")
            elif op.HasDest:
-                output_file.write("\tFEXCore::IR::{}Class,\n".format(op.DestType))
+                output_file.write("\tRegClass::{},\n".format(op.DestType))
            else:
                # No destination so it has an invalid destination class
-                output_file.write("\tFEXCore::IR::InvalidClass, // No destination\n")
+                output_file.write("\tRegClass::Invalid, // No destination\n")


    output_file.write("};\n\n")

    output_file.write("// Make sure our array maps directly to the IROps enum\n")
-    output_file.write("static_assert(IRRegClasses[IROps::OP_LAST] == FEXCore::IR::InvalidClass);\n\n")
+    output_file.write("static_assert(IRRegClasses[IROps::OP_LAST] == RegClass::Invalid);\n\n")

-    output_file.write("FEXCore::IR::RegisterClassType GetRegClass(IROps Op) { return IRRegClasses[Op]; }\n\n")
+    output_file.write("FEXCore::IR::RegClass GetRegClass(IROps Op) { return IRRegClasses[Op]; }\n\n")

    output_file.write("#undef IROP_REG_CLASSES_IMPL\n")
    output_file.write("#endif\n\n")
@@ -568,9 +566,7 @@ def print_ir_arg_printer():

            SSAArgNum = 0
            FirstArg = True
-            for i in range(0, len(op.Arguments)):
-                arg = op.Arguments[i]
-
+            for arg in op.Arguments:
                # No point printing temporaries that we can't recover
                if arg.Temporary:
                    continue
@@ -671,7 +667,7 @@ def print_ir_allocator_helpers():
    output_file.write("\t\treturn HeaderOp->Op;\n")
    output_file.write("\t}\n\n")

-    output_file.write("\tFEXCore::IR::RegisterClassType GetOpRegClass(const OrderedNode *Op) const {\n")
+    output_file.write("\tFEXCore::IR::RegClass GetOpRegClass(const OrderedNode *Op) const {\n")
    output_file.write("\t\treturn GetRegClass(GetOpType(Op));\n")
    output_file.write("\t}\n\n")

@@ -685,22 +681,21 @@ def print_ir_allocator_helpers():
            output_file.write("\tIRPair<IROp_{}> _{}(" .format(op.Name, op.Name))

            # Output SSA args first
-            for i in range(0, len(op.Arguments)):
-                arg = op.Arguments[i]
-                LastArg = len(op.Arguments) - i - 1 == 0
+            for i, arg in enumerate(op.Arguments):
+                LastArg = i == len(op.Arguments) - 1

                if arg.Temporary:
                    CType = IRTypesToCXX[arg.Type].CXXName
-                    output_file.write("{} {}".format(CType, arg.Name));
+                    output_file.write("{} {}".format(CType, arg.Name))
                elif arg.IsSSA:
                    # SSA value
                    output_file.write("OrderedNodeWrapper {}".format(arg.Name))
                else:
                    # User defined op that is stored
                    CType = IRTypesToCXX[arg.Type].CXXName
-                    output_file.write("{} {}".format(CType, arg.Name));
+                    output_file.write("{} {}".format(CType, arg.Name))

-                if arg.DefaultInitializer != None:
+                if arg.DefaultInitializer:
                    output_file.write(" = {}".format(arg.DefaultInitializer))

                if not LastArg:
@@ -758,20 +753,19 @@ def print_ir_allocator_helpers():
            if op.SSAArgNum:
                output_file.write("\tIRPair<IROp_{}> _{}(" .format(op.Name, op.Name))

-                for i in range(0, len(op.Arguments)):
-                    arg = op.Arguments[i]
-                    LastArg = len(op.Arguments) - i - 1 == 0
+                for i, arg in enumerate(op.Arguments):
+                    LastArg = i == len(op.Arguments) - 1

                    if arg.Temporary:
                        CType = IRTypesToCXX[arg.Type].CXXName
-                        output_file.write("{} {}".format(CType, arg.Name));
+                        output_file.write("{} {}".format(CType, arg.Name))
                    elif arg.IsSSA:
                        output_file.write("OrderedNode *{}".format(arg.Name))
                    else:
                        CType = IRTypesToCXX[arg.Type].CXXName
-                        output_file.write("{} {}".format(CType, arg.Name));
+                        output_file.write("{} {}".format(CType, arg.Name))

-                    if arg.DefaultInitializer != None:
+                    if arg.DefaultInitializer:
                        output_file.write(" = {}".format(arg.DefaultInitializer))

                    if not LastArg:
@@ -812,16 +806,15 @@ def print_ir_allocator_helpers():
                print_validation(op)

                output_file.write(f"\t\treturn _{op.Name}(")
-                for i in range(0, len(op.Arguments)):
-                    arg = op.Arguments[i]
-                    LastArg = len(op.Arguments) - i - 1 == 0
+                for i, arg in enumerate(op.Arguments):
+                    LastArg = i == len(op.Arguments) - 1
                    output_file.write(arg.Name)
                    if arg.IsSSA:
                        output_file.write("->Wrapped(ListDataBegin)")
                    if not LastArg:
                        output_file.write(", ")
-                output_file.write(");\n");
-                output_file.write("\t}\n\n");
+                output_file.write(");\n")
+                output_file.write("\t}\n\n")

    output_file.write("#undef IROP_ALLOCATE_HELPERS\n")
    output_file.write("#endif\n")
@@ -852,8 +845,8 @@ def print_ir_dispatcher_dispatch():
    output_dispatch_file.write("#endif\n")


-if (len(sys.argv) < 4):
-    ExitError()
+if len(sys.argv) < 4:
+    ExitError("Insufficient parameters passed to script")

 output_filename = sys.argv[2]
 output_dispatcher_filename = sys.argv[3]
@@ -865,6 +858,7 @@ json_file.close()
 json_object = json.loads(json_text)
 json_object = {k.upper(): v for k, v in json_object.items()}

+enums = json_object["ENUMS"]
 ops = json_object["OPS"]
 irtypes = json_object["IRTYPES"]
 defines = json_object["DEFINES"]
@@ -874,7 +868,7 @@ parse_ops(ops)

 output_file = open(output_filename, "w")

-print_enums()
+print_enums(enums)
 print_ir_structs(defines)
 print_ir_sizes()
 print_ir_reg_classes()
@@ -1,23 +1,23 @@
-set (MAN_DIR share/man CACHE PATH "MAN_DIR")
+set(MAN_DIR share/man CACHE PATH "MAN_DIR")

-set (FEXCORE_BASE_SRCS
+set(FEXCORE_BASE_SRCS
  Interface/Config/Config.cpp
  Utils/Allocator.cpp
  Utils/FileLoading.cpp
  Utils/ForcedAssert.cpp
  Utils/LogManager.cpp
-  Utils/SpinWaitLock.cpp
-  )
+  Utils/SpinWaitLock.cpp)

-if (NOT MINGW_BUILD)
+if (NOT MINGW)
  list(APPEND FEXCORE_BASE_SRCS
    Utils/Allocator/64BitAllocator.cpp)
 endif()

-set (SRCS
+set(SRCS
  Common/JitSymbols.cpp
  Interface/Context/Context.cpp
  Interface/Core/LookupCache.cpp
+  Interface/Core/CodeCache.cpp
  Interface/Core/Core.cpp
  Interface/Core/CPUBackend.cpp
  Interface/Core/Addressing.cpp
@@ -30,7 +30,6 @@ set (SRCS
  Interface/Core/OpcodeDispatcher/X87.cpp
  Interface/Core/OpcodeDispatcher/X87F64.cpp
  Interface/Core/OpcodeDispatcher.cpp
-  Interface/Core/X86HelperGen.cpp
  Interface/Core/ArchHelpers/Arm64Emitter.cpp
  Interface/Core/Dispatcher/Dispatcher.cpp
  Interface/Core/Interpreter/Fallbacks/InterpreterFallbacks.cpp
@@ -57,7 +56,6 @@ set (SRCS
  Interface/Core/X86Tables/VEXTables.cpp
  Interface/Core/X86Tables/X87Tables.cpp
  Interface/GDBJIT/GDBJIT.cpp
-  Interface/IR/AOTIR.cpp
  Interface/IR/IRDumper.cpp
  Interface/IR/IREmitter.cpp
  Interface/IR/PassManager.cpp
@@ -66,12 +64,12 @@ set (SRCS
  Interface/IR/Passes/RedundantFlagCalculationElimination.cpp
  Interface/IR/Passes/RegisterAllocationPass.cpp
  Interface/IR/Passes/x87StackOptimizationPass.cpp
+  Utils/LongJump.cpp
  Utils/Telemetry.cpp
  Utils/Threads.cpp
-  Utils/Profiler.cpp
-  )
+  Utils/Profiler.cpp)

-if (_M_ARM_64)
+if (ARCHITECTURE_arm64)
  list(APPEND SRCS Utils/ArchHelpers/Arm64.cpp)
 else()
  list(APPEND SRCS Utils/ArchHelpers/Arm64_stubs.cpp)
@@ -84,42 +82,41 @@ endif()

 set(DEFINES -DJIT_ARM64)

-if (_M_X86_64)
-  list(APPEND DEFINES -D_M_X86_64=1)
+if (ARCHITECTURE_x86_64)
+  list(APPEND DEFINES -DARCHITECTURE_x86_64=1)
 endif()

-if (_M_ARM_64)
-  list(APPEND DEFINES -D_M_ARM_64=1)
+if (ARCHITECTURE_arm64)
+  list(APPEND DEFINES -DARCHITECTURE_arm64=1)
 endif()

 if (ENABLE_VIXL_DISASSEMBLER)
  list(APPEND DEFINES -DVIXL_DISASSEMBLER=1)
 endif()

-if (_M_ARM_64 AND HAS_CLANG_PRESERVE_ALL)
+if (ARCHITECTURE_arm64 AND HAS_CLANG_PRESERVE_ALL)
  list(APPEND DEFINES "-DFEXCORE_PRESERVE_ALL_ATTR=__attribute__((preserve_all));-DFEXCORE_HAS_PRESERVE_ALL_ATTR=1")
 else()
  list(APPEND DEFINES "-DFEXCORE_PRESERVE_ALL_ATTR=;-DFEXCORE_HAS_PRESERVE_ALL_ATTR=0")
 endif()

-set (LIBS fmt::fmt xxHash::xxhash FEXHeaderUtils CodeEmitter cephes_128bit)
+set(LIBS fmt::fmt xxHash::xxhash FEXHeaderUtils CodeEmitter cephes_128bit)

 if (ENABLE_VIXL_DISASSEMBLER OR ENABLE_VIXL_SIMULATOR)
-  list (APPEND LIBS vixl)
+  list(APPEND LIBS vixl)
 endif()

-if (NOT MINGW_BUILD)
-  list (APPEND LIBS dl)
+if (NOT MINGW)
+  list(APPEND LIBS dl)
 else()
-  list (APPEND LIBS synchronization)
-  if (_M_ARM_64EC)
-    list (APPEND LIBS mincore)
+  list(APPEND LIBS synchronization)
+  if (ARCHITECTURE_arm64ec)
+    list(APPEND LIBS mincore)
  endif()
 endif()

 # Generate config
-configure_file(
-  ${CMAKE_CURRENT_SOURCE_DIR}/Interface/Config/Config.json.in
+configure_file(${CMAKE_CURRENT_SOURCE_DIR}/Interface/Config/Config.json.in
  ${CMAKE_BINARY_DIR}/generated/Config/Config.json)

 # Generate IR include file
@@ -134,11 +131,10 @@ add_custom_command(
  OUTPUT "${OUTPUT_NAME}" "${OUTPUT_DISPATCHER_NAME}"
  DEPENDS "${INPUT_NAME}"
  DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_generator.py"
-  COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_generator.py" "${INPUT_NAME}" "${OUTPUT_NAME}" "${OUTPUT_DISPATCHER_NAME}"
-  )
+  COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_generator.py"
+    "${INPUT_NAME}" "${OUTPUT_NAME}" "${OUTPUT_DISPATCHER_NAME}")

-set_source_files_properties(${OUTPUT_NAME} PROPERTIES
-  GENERATED TRUE)
+set_source_files_properties(${OUTPUT_NAME} PROPERTIES GENERATED TRUE)

 # Generate IR documentation
 set(OUTPUT_IR_DOC "${CMAKE_BINARY_DIR}/IR.md")
@@ -147,11 +143,10 @@ add_custom_command(
  OUTPUT "${OUTPUT_IR_DOC}"
  DEPENDS "${INPUT_NAME}"
  DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_doc_generator.py"
-  COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_doc_generator.py" "${INPUT_NAME}" "${OUTPUT_IR_DOC}"
-  )
+  COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_doc_generator.py"
+    "${INPUT_NAME}" "${OUTPUT_IR_DOC}")

-set_source_files_properties(${OUTPUT_IR_NAME} PROPERTIES
-  GENERATED TRUE)
+set_source_files_properties(${OUTPUT_IR_NAME} PROPERTIES GENERATED TRUE)

 # Create the target
 add_custom_target(IR_INC
@@ -175,14 +170,12 @@ add_custom_command(
  DEPENDS "${INPUT_CONFIG_NAME}"
  DEPENDS "${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}"
-  "${OUTPUT_CONFIG_OPTION_NAME}"
-  )
+  "${OUTPUT_CONFIG_OPTION_NAME}")

 add_custom_command(
  OUTPUT "${OUTPUT_MAN_NAME_COMPRESS}"
  DEPENDS "${OUTPUT_MAN_NAME}"
-  COMMAND "gzip" "-kf9n" "${OUTPUT_MAN_NAME}"
-  )
+  COMMAND "gzip" "-kf9n" "${OUTPUT_MAN_NAME}")

 set_source_files_properties(${OUTPUT_CONFIG_NAME} PROPERTIES
  GENERATED TRUE)
@@ -201,8 +194,10 @@ add_custom_target(CONFIG_INC
  DEPENDS "${OUTPUT_MAN_NAME}"
  DEPENDS "${OUTPUT_MAN_NAME_COMPRESS}")

-# Install the compressed man page
-install(FILES ${OUTPUT_MAN_NAME_COMPRESS} DESTINATION ${MAN_DIR}/man1)
+if (NOT BUILD_STEAM_SUPPORT)
+  # Install the compressed man page
+  install(FILES ${OUTPUT_MAN_NAME_COMPRESS} COMPONENT Runtime DESTINATION ${MAN_DIR}/man1)
+endif()

 # Add in diagnostic colours if the option is available.
 # Ninja code generator will kill colours if this isn't here
@@ -224,8 +219,7 @@ function(AddDefaultOptionsToTarget Name)
  target_compile_definitions(${Name} PRIVATE ${DEFINES})
  add_dependencies(${Name} CONFIG_INC IR_INC)

-  target_compile_options(${Name}
-    PRIVATE
+  target_compile_options(${Name} PRIVATE
    -Wall
    -Werror=cast-qual
    -Werror=ignored-qualifiers
@@ -233,31 +227,20 @@ function(AddDefaultOptionsToTarget Name)

    -Wno-trigraphs
    -ffunction-sections
-    -fwrapv
-  )
+    -fwrapv)

  if (GCC_COLOR)
-    target_compile_options(${Name}
-      PRIVATE
-      "-fdiagnostics-color=always")
-  endif()
-  if (CLANG_COLOR)
-    target_compile_options(${Name}
-      PRIVATE
-      "-fcolor-diagnostics")
+    target_compile_options(${Name} PRIVATE "-fdiagnostics-color=always")
  endif()

-  if (CMAKE_BUILD_TYPE MATCHES "RELEASE")
-    target_link_options(${Name}
-      PRIVATE
-      "LINKER:--gc-sections"
-      "LINKER:--strip-all"
-      "LINKER:--as-needed"
-    )
+  if (CLANG_COLOR)
+    target_compile_options(${Name} PRIVATE "-fcolor-diagnostics")
  endif()
+
+  LinkerGC(${Name})
 endfunction()

-# Build FEXCore_Config static library
+# Build FEXCore_Base static library
 add_library(FEXCore_Base STATIC ${FEXCORE_BASE_SRCS})
 target_link_libraries(FEXCore_Base ${LIBS})
 AddDefaultOptionsToTarget(FEXCore_Base)
@@ -266,34 +249,34 @@ if (ENABLE_FEXCORE_PROFILER AND FEXCORE_PROFILER_BACKEND STREQUAL "TRACY")
  target_link_libraries(FEXCore_Base TracyClient)
 endif()

-function(AddObject Name Type)
-  add_library(${Name} ${Type} ${SRCS})
+function(AddObject Name)
+  add_library(${Name} OBJECT ${SRCS})

-  target_link_libraries(${Name} FEXCore_Base)
+  target_link_libraries(${Name} PRIVATE FEXCore_Base)
  target_compile_options(${Name} PRIVATE ${FEX_TUNE_COMPILE_FLAGS})
  AddDefaultOptionsToTarget(${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(${Name} FEXCore_Base)
-  target_compile_options(${Name} PRIVATE ${FEX_TUNE_COMPILE_FLAGS})
  set_target_properties(${Name} PROPERTIES OUTPUT_NAME FEXCore)

+  # During generation of the import library (dll.a), MinGW needs some extra symbols from libraries
+  # such as fmt, which are propagated by FEXCore_Base. Wonderful.
+  if (MINGW)
+    target_link_libraries(${Name} FEXCore_Base)
+  endif()
  AddDefaultOptionsToTarget(${Name})
 endfunction()

-AddObject(${PROJECT_NAME}_object OBJECT)
+AddObject(${PROJECT_NAME}_object)
 AddLibrary(${PROJECT_NAME} STATIC)
 AddLibrary(${PROJECT_NAME}_shared SHARED)

-if (NOT MINGW_BUILD)
-  install(TARGETS ${PROJECT_NAME}_shared
-    LIBRARY
-      DESTINATION ${CMAKE_INSTALL_LIBDIR}
-      COMPONENT Libraries)
+if (NOT MINGW AND NOT BUILD_STEAM_SUPPORT)
+  install(TARGETS ${PROJECT_NAME}_shared LIBRARY
+    DESTINATION ${CMAKE_INSTALL_LIBDIR}
+    COMPONENT Libraries)
 endif()

 # Meta-library to link jemalloc libraries enabled in the build configuration.
@@ -308,7 +291,7 @@ if (ENABLE_JEMALLOC_GLIBC_ALLOC)
  target_link_libraries(JemallocLibs INTERFACE FEX_jemalloc_glibc)
 endif()

-if (NOT MINGW_BUILD)
+if (NOT MINGW)
  # Dummy project to use for host tools.
  # This overrides use of jemalloc in FEXCore with the normal glibc allocator.
  add_library(JemallocDummy STATIC Utils/AllocatorHooks.cpp)
@@ -1,5 +1,6 @@
 // SPDX-License-Identifier: MIT
 #pragma once
+#include <FEXCore/Utils/TypeDefines.h>

 #include <FEXCore/fextl/memory.h>

@@ -12,7 +13,7 @@ namespace FEXCore {
 // Buffered JIT symbol tracking.
 struct JITSymbolBuffer {
  // Maximum buffer size to ensure we are a page in size.
-  constexpr static size_t BUFFER_SIZE = 4096 - (8 * 2);
+  constexpr static size_t BUFFER_SIZE = FEXCore::Utils::FEX_PAGE_SIZE - (8 * 2);
  // Maximum distance until the end of the buffer to do a write.
  constexpr static size_t NEEDS_WRITE_DISTANCE = BUFFER_SIZE - 64;
  // Maximum time threshhold to wait before a buffer write occurs.
@@ -27,7 +28,7 @@ struct JITSymbolBuffer {
  size_t Offset {};
  char Buffer[BUFFER_SIZE] {};
 };
-static_assert(sizeof(JITSymbolBuffer) == 4096, "Ensure this is one page in size");
+static_assert(sizeof(JITSymbolBuffer) == FEXCore::Utils::FEX_PAGE_SIZE, "Ensure this is one page in size");

 class JITSymbols final {
 public:
@@ -4,9 +4,9 @@
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/fextl/sstream.h>
 #include <FEXCore/fextl/string.h>
-#include <FEXHeaderUtils/BitUtils.h>
 #include "cephes_128bit.h"

+#include <bit>
 #include <cmath>
 #include <cstring>
 #include <stdint.h>
@@ -19,7 +19,7 @@ extern "C" {
 }

 struct FEX_PACKED X80SoftFloat {
-#ifdef _M_X86_64
+#ifdef ARCHITECTURE_x86_64
 // Define this to push some operations to x87
 // Only useful to see if precision loss is killing something
 // #define DEBUG_X86_FLOAT
@@ -30,7 +30,7 @@ struct FEX_PACKED X80SoftFloat {
 #define BIGFLOAT float128_t
 #define BIGFLOATSIZE 16
 #endif
-#elif defined(_M_ARM_64)
+#elif defined(ARCHITECTURE_arm64)
 #define BIGFLOAT float128_t
 #define BIGFLOATSIZE 16
 #else
@@ -501,12 +501,12 @@ struct FEX_PACKED X80SoftFloat {

  float ToF32(softfloat_state* state) const {
    const float32_t Result = extF80_to_f32(state, *this);
-    return FEXCore::BitCast<float>(Result);
+    return std::bit_cast<float>(Result);
  }

  double ToF64(softfloat_state* state) const {
    const float64_t Result = extF80_to_f64(state, *this);
-    return FEXCore::BitCast<double>(Result);
+    return std::bit_cast<double>(Result);
  }

  FEXCore::VectorRegType ToVector() const {
@@ -518,7 +518,7 @@ struct FEX_PACKED X80SoftFloat {
  BIGFLOAT ToFMax(softfloat_state* state) const {
 #if BIGFLOATSIZE == 16
    const float128_t Result = extF80_to_f128(state, *this);
-    return FEXCore::BitCast<BIGFLOAT>(Result);
+    return std::bit_cast<BIGFLOAT>(Result);
 #else
    BIGFLOAT result {};
    memcpy(&result, this, sizeof(result));
@@ -577,18 +577,18 @@ struct FEX_PACKED X80SoftFloat {
  }

  X80SoftFloat(softfloat_state* state, const float rhs) {
-    *this = f32_to_extF80(state, FEXCore::BitCast<float32_t>(rhs));
+    *this = f32_to_extF80(state, std::bit_cast<float32_t>(rhs));
  }

  X80SoftFloat(softfloat_state* state, const double rhs) {
-    *this = f64_to_extF80(state, FEXCore::BitCast<float64_t>(rhs));
+    *this = f64_to_extF80(state, std::bit_cast<float64_t>(rhs));
  }

  X80SoftFloat(softfloat_state* state, BIGFLOAT rhs) {
 #if BIGFLOATSIZE == 16
-    *this = f128_to_extF80(state, FEXCore::BitCast<float128_t>(rhs));
+    *this = f128_to_extF80(state, std::bit_cast<float128_t>(rhs));
 #else
-    *this = FEXCore::BitCast<long double>(rhs);
+    *this = std::bit_cast<long double>(rhs);
 #endif
  }

@@ -2,59 +2,23 @@
 #pragma once
 #include <FEXCore/fextl/string.h>

-#include <cstdint>
+#include <concepts>
 #include <string_view>
-#include <optional>

 namespace FEXCore::StrConv {
-inline bool Conv(std::string_view Value, bool* Result) {
-  *Result = std::strtoull(Value.data(), nullptr, 0);
+template<std::integral T>
+bool Conv(std::string_view Value, T* Result) {
+  if constexpr (std::is_signed_v<T>) {
+    *Result = static_cast<T>(std::strtoll(Value.data(), nullptr, 0));
+  } else {
+    *Result = static_cast<T>(std::strtoull(Value.data(), nullptr, 0));
+  }
  return true;
 }

-inline bool Conv(std::string_view Value, uint8_t* Result) {
-  *Result = std::strtoul(Value.data(), nullptr, 0);
-  return true;
-}
-
-inline bool Conv(std::string_view Value, int8_t* Result) {
-  *Result = std::strtol(Value.data(), nullptr, 0);
-  return true;
-}
-
-inline bool Conv(std::string_view Value, uint16_t* Result) {
-  *Result = std::strtoul(Value.data(), nullptr, 0);
-  return true;
-}
-
-inline bool Conv(std::string_view Value, int16_t* Result) {
-  *Result = std::strtol(Value.data(), nullptr, 0);
-  return true;
-}
-
-inline bool Conv(std::string_view Value, uint32_t* Result) {
-  *Result = std::strtoul(Value.data(), nullptr, 0);
-  return true;
-}
-
-inline bool Conv(std::string_view Value, int32_t* Result) {
-  *Result = std::strtol(Value.data(), nullptr, 0);
-  return true;
-}
-
-inline bool Conv(std::string_view Value, uint64_t* Result) {
-  *Result = std::strtoull(Value.data(), nullptr, 0);
-  return true;
-}
-
-inline bool Conv(std::string_view Value, int64_t* Result) {
-  *Result = std::strtoll(Value.data(), nullptr, 0);
-  return true;
-}
-
-template<typename T, typename = std::enable_if<std::is_enum<T>::value, T>>
-inline bool Conv(std::string_view Value, T* Result) {
-  *Result = static_cast<T>(std::stoull(Value.data(), nullptr, 0));
+template<typename T, typename = std::enable_if_t<std::is_enum_v<T>, T>>
+bool Conv(std::string_view Value, T* Result) {
+  *Result = static_cast<T>(std::strtoull(Value.data(), nullptr, 0));
  return true;
 }

@@ -1,9 +1,11 @@
 // SPDX-License-Identifier: MIT
 #pragma once

-#ifdef _M_X86_64
+#ifdef ARCHITECTURE_x86_64
 #include <xmmintrin.h>
 #include <immintrin.h>
+#else
+#include <cstdint>
 #endif

 namespace FEXCore {
@@ -11,7 +13,7 @@ struct VectorScalarF64Pair {
  double val[2];
 };

-#ifdef _M_ARM_64
+#ifdef ARCHITECTURE_arm64
 // Can't use uint8x16_t directly from arm_neon.h here.
 // Overrides softfloat-3e's defines which causes problems.
 using VectorRegType = __attribute__((neon_vector_type(16))) uint8_t;
@@ -23,7 +25,7 @@ static inline VectorRegPairType MakeVectorRegPair(VectorRegType low, VectorRegTy
  return VectorRegPairType {low, high};
 }

-#elif defined(_M_X86_64)
+#elif defined(ARCHITECTURE_x86_64)
 using VectorRegType = __m128i;
 using VectorRegPairType = __m256i;

@@ -30,14 +30,14 @@ class Context;
 }

 namespace FEXCore::Config {
-namespace DefaultValues {
+namespace detail {
 #define P(x) x
 #define OPT_BASE(type, group, enum, json, default) const P(type) P(enum) = P(default);
 #define OPT_STR(group, enum, json, default) const std::string_view P(enum) = P(default);
 #define OPT_STRARRAY(group, enum, json, default) OPT_STR(group, enum, json, default)
 #define OPT_STRENUM(group, enum, json, default) const uint64_t P(enum) = FEXCore::ToUnderlying(P(default));
 #include <FEXCore/Config/ConfigValues.inl>
-} // namespace DefaultValues
+} // namespace detail

 enum Paths {
  PATH_DATA_DIR_LOCAL = 0,
@@ -134,7 +134,7 @@ public:
  void Load();

  template<typename T>
-  requires (!std::is_same_v<fextl::string, T> && !std::is_same_v<DefaultValues::Type::StringArrayType, T>)
+  requires (!std::is_same_v<fextl::string, T> && !std::is_same_v<StringArrayType, T>)
  std::optional<T> GetConv(ConfigOption Option) {
    const auto it = OptionMap.find(Option);
    if (it == OptionMap.end()) {
@@ -142,7 +142,7 @@ public:
    }

    const auto& Value = it->second;
-    LOGMAN_THROW_A_FMT(!std::holds_alternative<DefaultValues::Type::StringArrayType>(Value), "Tried to get config of invalid type!");
+    LOGMAN_THROW_A_FMT(!std::holds_alternative<StringArrayType>(Value), "Tried to get config of invalid type!");

    if (std::holds_alternative<T>(Value)) [[likely]] {
      return std::get<T>(Value);
@@ -165,7 +165,7 @@ public:

 private:
  void MergeConfigMap(const LayerOptions& Options);
-  void MergeEnvironmentVariables(const ConfigOption& Option, const DefaultValues::Type::StringArrayType& Value);
+  void MergeEnvironmentVariables(const ConfigOption& Option, const StringArrayType& Value);
 };

 void MetaLayer::Load() {
@@ -181,7 +181,7 @@ void MetaLayer::Load() {
 }


-void MetaLayer::MergeEnvironmentVariables(const ConfigOption& Option, const DefaultValues::Type::StringArrayType& Value) {
+void MetaLayer::MergeEnvironmentVariables(const ConfigOption& Option, const StringArrayType& Value) {
  // Environment variables need a bit of additional work
  // We want to merge the arrays rather than overwrite entirely
  auto MetaEnvironment = OptionMap.find(Option);
@@ -193,7 +193,7 @@ void MetaLayer::MergeEnvironmentVariables(const ConfigOption& Option, const Defa

  // If an environment variable exists in both current meta and in the incoming layer then the meta layer value is overwritten
  fextl::unordered_map<fextl::string, fextl::string> LookupMap;
-  const auto AddToMap = [&LookupMap](const DefaultValues::Type::StringArrayType& Value) {
+  const auto AddToMap = [&LookupMap](const StringArrayType& Value) {
    for (const auto& EnvVar : Value) {
      const auto ItEq = EnvVar.find_first_of('=');
      if (ItEq == fextl::string::npos) {
@@ -209,7 +209,7 @@ void MetaLayer::MergeEnvironmentVariables(const ConfigOption& Option, const Defa
    }
  };

-  AddToMap(std::get<DefaultValues::Type::StringArrayType>(MetaEnvironment->second));
+  AddToMap(std::get<StringArrayType>(MetaEnvironment->second));
  AddToMap(Value);

  // Now with the two layers merged in the map
@@ -225,8 +225,8 @@ void MetaLayer::MergeConfigMap(const LayerOptions& Options) {
  // Insert this layer's options, overlaying previous options that exist here
  for (auto& it : Options) {
    if (it.first == FEXCore::Config::ConfigOption::CONFIG_ENV || it.first == FEXCore::Config::ConfigOption::CONFIG_HOSTENV) {
-      LOGMAN_THROW_A_FMT(std::holds_alternative<DefaultValues::Type::StringArrayType>(it.second), "Tried to get config of invalid type!");
-      MergeEnvironmentVariables(it.first, std::get<DefaultValues::Type::StringArrayType>(it.second));
+      LOGMAN_THROW_A_FMT(std::holds_alternative<StringArrayType>(it.second), "Tried to get config of invalid type!");
+      MergeEnvironmentVariables(it.first, std::get<StringArrayType>(it.second));
    } else {
      OptionMap.insert_or_assign(it.first, it.second);
    }
@@ -423,7 +423,7 @@ bool Exists(ConfigOption Option) {
  return Meta->OptionExists(Option);
 }

-std::optional<DefaultValues::Type::StringArrayType*> All(ConfigOption Option) {
+std::optional<StringArrayType*> All(ConfigOption Option) {
  return Meta->All(Option);
 }

@@ -436,6 +436,12 @@ std::optional<T> GetConv(ConfigOption Option) {
  return Meta->GetConv<T>(Option);
 }

+template std::optional<bool> GetConv(ConfigOption Option);
+template std::optional<uint8_t> GetConv(ConfigOption Option);
+template std::optional<int32_t> GetConv(ConfigOption Option);
+template std::optional<uint32_t> GetConv(ConfigOption Option);
+template std::optional<uint64_t> GetConv(ConfigOption Option);
+
 void Set(ConfigOption Option, std::string_view Data) {
  Meta->Set(Option, Data);
 }
@@ -491,13 +497,12 @@ template Value<uint8_t>::Value(FEXCore::Config::ConfigOption _Option, uint8_t De
 template Value<uint64_t>::Value(FEXCore::Config::ConfigOption _Option, uint64_t Default);

 template<typename T>
-void Value<T>::GetListIfExists(FEXCore::Config::ConfigOption Option, DefaultValues::Type::StringArrayType* List) {
+void Value<T>::GetListIfExists(FEXCore::Config::ConfigOption Option, StringArrayType* List) {
  auto Value = FEXCore::Config::All(Option);
  List->clear();
  if (Value) {
    *List = **Value;
  }
 }
-template void Value<DefaultValues::Type::StringArrayType>::GetListIfExists(FEXCore::Config::ConfigOption Option,
-                                                                           DefaultValues::Type::StringArrayType* List);
+template void Value<StringArrayType>::GetListIfExists(FEXCore::Config::ConfigOption Option, StringArrayType* List);
 } // namespace FEXCore::Config
@@ -4,7 +4,6 @@
      "Multiblock": {
        "Type": "bool",
        "Default": "true",
-        "ShortArg": "m",
        "Desc": [
          "Controls multiblock code compilation",
          "Can cause long JIT compilation times and stutter"
@@ -13,11 +12,24 @@
      "MaxInst": {
        "Type": "int32",
        "Default": "5000",
-        "ShortArg": "n",
        "Desc": [
          "Maximum number of instruction to store in a block"
        ]
      },
+      "EnableCodeCachingWIP": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Enable the code caching subsystem"
+        ]
+      },
+      "EnableCodeCacheValidation": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Enable expensive validation when loading code caches"
+        ]
+      },
      "HostFeatures": {
        "Type": "strenum",
        "Default": "FEXCore::Config::HostFeatures::OFF",
@@ -61,7 +73,9 @@
          "ENABLEWFXT": "enablewfxt",
          "DISABLEWFXT": "disablewfxt",
          "ENABLE3DNOW": "enable3dnow",
-          "DISABLE3DNOW": "disable3dnow"
+          "DISABLE3DNOW": "disable3dnow",
+          "ENABLESSE4A": "enablesse4a",
+          "DISABLESSE4A": "disablesse4a"
        },
        "Desc": [
          "Allows controlling of the CPU features in the JIT.",
@@ -84,7 +98,8 @@
          "\t{enable,disable}svebitperm: Will force enable or disable svebitperm even if the host doesn't support it",
          "\t{enable,disable}preserveallabi: Will force enable or disable preserve_all abi even if the host doesn't support it",
          "\t{enable,disable}wfxt: Will force enable or disable wfxt even if the host doesn't support it",
-          "\t{enable,disable}3dnow: Will force enable or disable 3DNow even if the host doesn't support it"
+          "\t{enable,disable}3dnow: Will force enable or disable 3DNow! even if the host doesn't support it",
+          "\t{enable,disable}sse4a: Will force enable or disable SSE4a even if the host doesn't support it"
        ]
      },
      "SmallTSCScale": {
@@ -93,13 +108,19 @@
        "Desc": [
          "Scales the cycle counter on systems that have low frequencies."
        ]
+      },
+      "CPUFeatureRegisters": {
+        "Type": "str",
+        "Default": "",
+        "Desc": [
+          "Allows overriding cpu feature flags for manual testing"
+        ]
      }
    },
    "Emulation": {
      "RootFS": {
        "Type": "str",
        "Default": "",
-        "ShortArg": "R",
        "Desc": [
          "Which Root filesystem prefix to use",
          "This can be a filesystem path",
@@ -114,7 +135,6 @@
      "ThunkHostLibs": {
        "Type": "str",
        "Default": "@CMAKE_INSTALL_FULL_LIBDIR@/fex-emu/HostThunks",
-        "ShortArg": "t",
        "Desc": [
          "Folder to find the host-side thunking libraries."
        ]
@@ -122,7 +142,6 @@
      "ThunkGuestLibs": {
        "Type": "str",
        "Default": "@CMAKE_INSTALL_PREFIX@/share/fex-emu/GuestThunks",
-        "ShortArg": "j",
        "Desc": [
          "Folder to find the guest-side thunking libraries."
        ]
@@ -130,7 +149,6 @@
      "ThunkConfig": {
        "Type": "str",
        "Default": "",
-        "ShortArg": "k",
        "Desc": [
          "A json file specifying where to overlay the thunks.",
          "This can be a filesystem path",
@@ -145,7 +163,6 @@
      "Env": {
        "Type": "strarray",
        "Default": "",
-        "ShortArg": "E",
        "Desc": [
          "Adds an environment variable to the emulated environment."
        ]
@@ -153,7 +170,6 @@
      "HostEnv": {
        "Type": "strarray",
        "Default": "",
-        "ShortArg": "H",
        "Desc": [
          "Adds an environment variable to the host environment.",
          "This can be useful for setting environment variables that thunks can pick up.",
@@ -166,13 +182,50 @@
        "Desc": [
          "Allows the user to pass additional arguments to the application"
        ]
+      },
+      "DisableL2Cache": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Disables FEXCore's JIT L2 cache lookup. Saving memory.",
+          "Can potentially introduce more stutters."
+        ]
+      },
+      "DynamicL1Cache": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Switches FEXCore's JIT L1 cache to be dynamically sized. Saving memory.",
+          "Can potentially introduce more stutters."
+        ]
+      },
+      "DynamicL1CacheIncreaseCountHeuristic": {
+        "Type": "uint64",
+        "Default": "250",
+        "Desc": [
+          "Threshold of lookups per second that the L1 dynamic cache should increase its size.",
+          "Lower numbers means more aggressive scaling upward to the maximum size.",
+          "Higher numbers means more conservative scaling, using less memory.",
+          "Can potentially introduce stutters, more likely the higher the number.",
+          "Don't have this number smaller than the decrease count!"
+        ]
+      },
+      "DynamicL1CacheDecreaseCountHeuristic": {
+        "Type": "uint64",
+        "Default": "50",
+        "Desc": [
+          "Threshold of lookups per second that the L1 dynamic cache should decrease its size.",
+          "The higher the number, the more aggressively it reduces the L1 cache size.",
+          "Lower numbers means more conservative memory savings.",
+          "Can potentially introduce more stutters, more likely the higher the number.",
+          "Don't have this number larger than the increase count!"
+        ]
      }
    },
    "Debug": {
      "SingleStep": {
        "Type": "bool",
        "Default": "false",
-        "ShortArg": "S",
        "Desc": [
          "Single stepping configuration."
        ]
@@ -180,7 +233,6 @@
      "GdbServer": {
        "Type": "bool",
        "Default": "false",
-        "ShortArg": "G",
        "Desc": [
          "Enables the GDB server."
        ]
@@ -214,7 +266,6 @@
      "DumpGPRs": {
        "Type": "bool",
        "Default": "false",
-        "ShortArg": "g",
        "Desc": [
          "When the test harness ends, print the GPR state."
        ]
@@ -222,7 +273,6 @@
      "O0": {
        "Type": "bool",
        "Default": "false",
-        "ShortArg": "O0",
        "Desc": [
          "Disables optimizations passes for debugging."
        ]
@@ -312,7 +362,6 @@
      "SilentLog": {
        "Type": "bool",
        "Default": "true",
-        "ShortArg": "s",
        "Desc": [
          "Disables logging"
        ]
@@ -320,10 +369,9 @@
      "OutputLog": {
        "Type": "str",
        "Default": "server",
-        "ShortArg": "o",
        "Desc": [
          "File to write FEX output to.",
-          "[stdout, stderr, server, <Filename>]"
+          "[stderr, server, <Filename>]"
        ]
      },
      "TelemetryDirectory": {
@@ -341,6 +389,13 @@
          "Enables FEX's low-overhead sampling profile statistics.",
          "Requires a supported version of Mangohud to see the results"
        ]
+      },
+      "EnableGpuvisProfiling": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Enables profiling when FEX was built with the gpuvis profiler backend."
+        ]
      }
    },
    "Hacks": {
@@ -395,12 +450,11 @@
          "This is required to ensure a split-lock doesn't tear inside the process"
        ]
      },
-      "TSOAutoMigration": {
+      "KernelUnalignedAtomicBackpatching": {
        "Type": "bool",
        "Default": "true",
        "Desc": [
-          "Automatically enables TSO when shared memory is used.",
-          "Should work without issues in most cases."
+          "When the kernel unaligned atomic handler is enabled, use backpatching to reduce kernel context switches."
        ]
      },
      "VolatileMetadata": {
@@ -408,7 +462,7 @@
        "Default": "true",
        "Desc": [
          "Use volatile metadata in PE files to inform TSO instructions when available.",
-	  "When metadata is unavailable falls back to the currently enabled TSO options."
+          "When metadata is unavailable falls back to the currently enabled TSO options."
        ]
      },
      "X87ReducedPrecision": {
@@ -418,23 +472,6 @@
          "Emulates X87 floating point using 64-bit precision. This reduces emulation accuracy and may result in rendering bugs."
        ]
      },
-      "ABILocalFlags": {
-        "Type": "bool",
-        "Default": "false",
-        "Desc": [
-          "When enabled enables an optimization around flags.",
-          "Assumes flags are not used across cals.",
-          "Hand-written assembly can violate this assumption."
-        ]
-      },
-      "ParanoidTSO": {
-        "Type": "bool",
-        "Default": "false",
-        "Desc": [
-          "Makes TSO operations even more strict.",
-          "Forces vector loadstores to also become atomic."
-        ]
-      },
      "StallProcess": {
        "Type": "bool",
        "Default": "false",
@@ -514,10 +551,6 @@
  },
  "UnnamedOptions": {
    "Misc": {
-      "IS_INTERPRETER": {
-        "Type": "bool",
-        "Default": "false"
-      },
      "INTERPRETER_INSTALLED": {
        "Type": "bool",
        "Default": "false"
@@ -1,6 +1,7 @@
 // SPDX-License-Identifier: MIT
 #include "Interface/Context/Context.h"
 #include "Interface/Core/OpcodeDispatcher.h"
+#include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Core/X86Tables/X86Tables.h"

 #include <FEXCore/Core/CoreState.h>
@@ -4,54 +4,47 @@
 #include "Common/JitSymbols.h"
 #include "Interface/Core/CPUBackend.h"
 #include "Interface/Core/CPUID.h"
-#include "Interface/Core/X86HelperGen.h"
-#include "Interface/Core/Dispatcher/Dispatcher.h"
-#include "Interface/IR/AOTIR.h"
 #include <Interface/IR/IntrusiveIRList.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/IR/IR.h>
 #include <FEXCore/Utils/CompilerDefs.h>
-#include <FEXCore/Utils/Event.h>
 #include <FEXCore/Utils/SignalScopeGuards.h>
 #include <FEXCore/fextl/memory.h>
 #include <FEXCore/fextl/set.h>
 #include <FEXCore/fextl/string.h>
 #include <FEXCore/fextl/unordered_map.h>
 #include <FEXCore/fextl/vector.h>
-#include <FEXHeaderUtils/Syscalls.h>
-#include <stdint.h>

 #include <atomic>
+#include <cstddef>
+#include <cstdint>
 #include <mutex>
+#include <optional>
 #include <shared_mutex>

 namespace FEXCore {
-class CodeLoader;
+class SignalDelegator;
 class ThunkHandler;
+struct LookupCacheWriteLockToken;
+
+namespace Core {
+  struct DebugData;
+  struct InternalThreadState;
+} // namespace Core

 namespace CPU {
-  class Arm64JITCore;
  class Dispatcher;
 } // namespace CPU
+
 namespace HLE {
-  struct SyscallArguments;
-  class SyscallHandler;
  class SourcecodeResolver;
-  struct SourcecodeMap;
+  class SyscallHandler;
 } // namespace HLE
 } // namespace FEXCore

-namespace FEXCore::IR {
-namespace Validation {
-  class IRValidation;
-}
-} // namespace FEXCore::IR
-
 namespace FEXCore::Context {
 struct FEX_PACKED ExitFunctionLinkData {
  uint64_t HostCode;
@@ -68,10 +61,66 @@ struct CustomIRResult {
    , Data(Data) {}
 };

-using BlockDelinkerFunc = void (*)(FEXCore::Core::CpuStateFrame* Frame, FEXCore::Context::ExitFunctionLinkData* Record);
+using BlockDelinkerFunc = void (*)(FEXCore::Context::ExitFunctionLinkData* Record);
 constexpr uint32_t TSC_SCALE_MAXIMUM = 1'000'000'000; ///< 1Ghz

-class ContextImpl final : public FEXCore::Context::Context, CPU::CodeBufferManager {
+class CodeCache : public AbstractCodeCache {
+public:
+  CodeCache(ContextImpl&);
+  ~CodeCache();
+
+  ContextImpl& CTX;
+  fextl::unique_ptr<ContextImpl> ValidationCTX;
+  fextl::unique_ptr<Core::InternalThreadState> ValidationThread;
+  FEXCore::Core::CPUState::gdt_segment ValidationGDT[32] {};
+  bool IsGeneratingCache = false;
+
+  FEX_CONFIG_OPT(EnableCodeCaching, ENABLECODECACHINGWIP);
+  FEX_CONFIG_OPT(EnableCodeCacheValidation, ENABLECODECACHEVALIDATION);
+
+  uint64_t ComputeCodeMapId(std::string_view Filename, int FD) override;
+  bool SaveData(Core::InternalThreadState&, int TargetFD, const ExecutableFileSectionInfo&, uint64_t SerializedBaseAddress) override;
+  bool LoadData(Core::InternalThreadState*, std::byte* MappedCacheFile, const ExecutableFileSectionInfo&) override;
+
+  /**
+   * Performs expensive extra validation on the loaded code cache data.
+   *
+   * This kicks off an in-process recompile of all cached blocks and compares
+   * them with the cached data. Differences will be reported as fatal errors,
+   * which can uncover bugs like for example:
+   * - mismatches of the JIT configuration used during cache generation
+   * - hidden position dependencies due to missing FEX relocations
+   * - incorrect instruction padding
+   */
+  void Validate(const ExecutableFileSectionInfo&, fextl::set<uint64_t> GuestBlocks, const fextl::set<uint64_t>& HostBlocks,
+                std::span<std::byte> CachedCode);
+
+  void InitiateCacheGeneration() override {
+    IsGeneratingCache = true;
+  }
+
+  /**
+   * Applies a set of FEX relocations to the given code section.
+   *
+   * FEX relocations describe runtime-dependencies of FEX-generated code.
+   * When loading a code cache, they are used to move cached code to the
+   * dynamically chosen base address of the guest binary.
+   *
+   * Conversely, relocations are applied in reverse when writing code caches
+   * to ensure consistency across generation runs.
+   *
+   * Note that FEX relocations are unrelated to ELF/PE relocations.
+   *
+   * @param GuestDelta Guest address offset to apply to RIP-relative data
+   * @param ForStorage True for serializing data (producing deterministic output); false for de-serializing it (resolving dynamic symbols)
+   *
+   * @return Returns true on success
+   */
+  [[nodiscard]]
+  bool ApplyCodeRelocations(uint64_t GuestDelta, std::span<std::byte> Code, std::span<const CPU::Relocation> Relocations, bool ForStorage);
+};
+
+class ContextImpl final : public FEXCore::Context::Context, public CPU::CodeBufferManager {
 public:
  // Context base class implementation.
  bool InitCore() override;
@@ -141,21 +190,28 @@ public:
  FEXCore::CPUID::XCRResults RunXCRFunction(uint32_t Function) override;
  FEXCore::CPUID::FunctionResults RunCPUIDFunctionName(uint32_t Function, uint32_t Leaf, uint32_t CPU) override;

-  FEXCore::IR::AOTIRCacheEntry* LoadAOTIRCacheEntry(const fextl::string& Name) override;
-  void UnloadAOTIRCacheEntry(FEXCore::IR::AOTIRCacheEntry* Entry) override;
+  CodeCache& GetCodeCache() override {
+    return CodeCache;
+  }

-  void FinalizeAOTIRCache() override {}
+  void SetCodeMapWriter(fextl::unique_ptr<CodeMapWriter> Writer) override {
+    CodeMapWriter = std::move(Writer);
+  }

-  void OnCodeBufferAllocated(CPU::CodeBuffer&) override;
+  void FlushAndCloseCodeMap() override {
+    if (CodeMapWriter) {
+      CodeMapWriter.reset();
+    }
+  }
+
+  void OnCodeBufferAllocated(const std::shared_ptr<CPU::CodeBuffer>&) override;
  void ClearCodeCache(FEXCore::Core::InternalThreadState* Thread, bool NewCodeBuffer = true) override;
-  void InvalidateGuestCodeRange(FEXCore::Core::InternalThreadState* Thread, InvalidatedEntryAccumulator& Accumulator, uint64_t Start,
-                                uint64_t Length) override;
+  void InvalidateCodeBuffersCodeRange(uint64_t Start, uint64_t Length) override;
+  void InvalidateThreadCachedCodeRange(FEXCore::Core::InternalThreadState* Thread, uint64_t Start, uint64_t Length) override;
  FEXCore::ForkableSharedMutex& GetCodeInvalidationMutex() override {
    return CodeInvalidationMutex;
  }

-  void MarkMemoryShared(FEXCore::Core::InternalThreadState* Thread) override;
-
  void ConfigureAOTGen(FEXCore::Core::InternalThreadState* Thread, fextl::set<uint64_t>* ExternalBranches, uint64_t SectionMaxAddress) override;

  bool IsAddressInCodeBuffer(FEXCore::Core::InternalThreadState* Thread, uintptr_t Address) const override;
@@ -177,13 +233,6 @@ public:
  void MarkMonoBackpatcherBlock(uint64_t BlockEntry) override;

 public:
-  friend class FEXCore::HLE::SyscallHandler;
-#ifdef JIT_ARM64
-  friend class FEXCore::CPU::Arm64JITCore;
-#endif
-
-  friend class FEXCore::IR::Validation::IRValidation;
-
  struct {
    uint64_t VirtualMemSize {1ULL << 36};
    uint64_t TSCScale = 0;
@@ -196,10 +245,8 @@ public:
    FEX_CONFIG_OPT(GdbServer, GDBSERVER);
    FEX_CONFIG_OPT(Is64BitMode, IS64BIT_MODE);
    FEX_CONFIG_OPT(TSOEnabled, TSOENABLED);
-    FEX_CONFIG_OPT(TSOAutoMigration, TSOAUTOMIGRATION);
    FEX_CONFIG_OPT(VectorTSOEnabled, VECTORTSOENABLED);
    FEX_CONFIG_OPT(MemcpySetTSOEnabled, MEMCPYSETTSOENABLED);
-    FEX_CONFIG_OPT(ABILocalFlags, ABILOCALFLAGS);
    FEX_CONFIG_OPT(SMCChecks, SMCCHECKS);
    FEX_CONFIG_OPT(MaxInstPerBlock, MAXINST);
    FEX_CONFIG_OPT(RootFSPath, ROOTFS);
@@ -207,7 +254,6 @@ public:
    FEX_CONFIG_OPT(LibraryJITNaming, LIBRARYJITNAMING);
    FEX_CONFIG_OPT(BlockJITNaming, BLOCKJITNAMING);
    FEX_CONFIG_OPT(GDBSymbols, GDBSYMBOLS);
-    FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
    FEX_CONFIG_OPT(x87ReducedPrecision, X87REDUCEDPRECISION);
    FEX_CONFIG_OPT(DisableTelemetry, DISABLETELEMETRY);
    FEX_CONFIG_OPT(DisableVixlIndirectCalls, DISABLE_VIXL_INDIRECT_RUNTIME_CALLS);
@@ -227,14 +273,13 @@ public:
  FEXCore::HLE::SourcecodeResolver* SourcecodeResolver {};
  FEXCore::ThunkHandler* ThunkHandler {};
  fextl::unique_ptr<FEXCore::CPU::Dispatcher> Dispatcher;
+  CodeCache CodeCache;
+  fextl::unique_ptr<CodeMapWriter> CodeMapWriter;

  SignalDelegator* SignalDelegation {};
-  X86GeneratedCode X86CodeGen;

  ContextImpl(const FEXCore::HostFeatures& Features);

-  static bool ThreadRemoveCodeEntry(FEXCore::Core::InternalThreadState* Thread, uint64_t GuestRIP);
-
  static void ThreadRemoveCodeEntryFromJit(FEXCore::Core::CpuStateFrame* Frame, uint64_t GuestRIP);

  // This is used as a replacement for the SMC writes in the mono callsite backpatcher that avoids atomic operations
@@ -269,9 +314,9 @@ public:

  FEXCore::JITSymbols Symbols;

-  FEXCore::Utils::PooledAllocatorVirtual OpDispatcherAllocator;
-  FEXCore::Utils::PooledAllocatorVirtual FrontendAllocator;
-  FEXCore::Utils::PooledAllocatorVirtual CPUBackendAllocator;
+  FEXCore::Utils::PooledAllocatorVirtual OpDispatcherAllocator {"FEXMem_OpDispatcher"};
+  FEXCore::Utils::PooledAllocatorVirtual FrontendAllocator {"FEXMem_Frontend"};
+  FEXCore::Utils::PooledAllocatorVirtualWithGuard CPUBackendAllocator {"FEXMem_CPUBackend"};

  // If Atomic-based TSO emulation is enabled or not.
  bool IsAtomicTSOEnabled() const {
@@ -312,17 +357,10 @@ protected:
      AtomicTSOEmulationEnabled = false;
      VectorAtomicTSOEmulationEnabled = false;
      MemcpyAtomicTSOEmulationEnabled = false;
-    } else if (Config.ParanoidTSO) {
-      AtomicTSOEmulationEnabled = true;
-      VectorAtomicTSOEmulationEnabled = true;
-      MemcpyAtomicTSOEmulationEnabled = true;
    } else {
-      // Atomic TSO emulation only enabled if the config option is enabled.
-      AtomicTSOEmulationEnabled = (IsMemoryShared || !Config.TSOAutoMigration) && Config.TSOEnabled;
-      // Atomic vector TSO emulation only enabled if TSO emulation is enabled and also vector TSO is enabled.
-      VectorAtomicTSOEmulationEnabled = (IsMemoryShared || !Config.TSOAutoMigration) && Config.TSOEnabled && Config.VectorTSOEnabled;
-      // Atomic memcpy TSO emulation only enabled if TSO emulation is enabled and also memcpy TSO is enabled.
-      MemcpyAtomicTSOEmulationEnabled = (IsMemoryShared || !Config.TSOAutoMigration) && Config.TSOEnabled && Config.MemcpySetTSOEnabled;
+      AtomicTSOEmulationEnabled = Config.TSOEnabled;
+      VectorAtomicTSOEmulationEnabled = Config.TSOEnabled && Config.VectorTSOEnabled;
+      MemcpyAtomicTSOEmulationEnabled = Config.TSOEnabled && Config.MemcpySetTSOEnabled;
    }
  }

@@ -336,9 +374,6 @@ private:
   */
  void InitializeCompiler(FEXCore::Core::InternalThreadState* Thread);

-  IR::AOTIRCaptureCache IRCaptureCache;
-
-  bool IsMemoryShared = false;
  bool SupportsHardwareTSO = false;
  bool AtomicTSOEmulationEnabled = true;
  bool VectorAtomicTSOEmulationEnabled = false;
@@ -351,8 +386,8 @@ private:
  std::atomic<bool> HasCustomIRHandlers {};
  struct CustomIRHandlerEntry final {
    CustomIREntrypointHandler Handler;
-    void *Creator;
-    void *Data;
+    void* Creator;
+    void* Data;
  };
  fextl::unordered_map<uint64_t, CustomIRHandlerEntry> CustomIRHandlers;
  IntervalList<uint64_t> ForceTSOValidRanges; // The ranges for which ForceTSOInstructions has populated data
@@ -360,5 +395,8 @@ private:

  bool MonoDetected = false;
  std::atomic<uint64_t> MonoBackpatcherBlock;
+
+  std::mutex CodeBufferListLock;
+  fextl::vector<std::weak_ptr<CPU::CodeBuffer>> CodeBufferList;
 };
 } // namespace FEXCore::Context
@@ -7,7 +7,7 @@

 namespace FEXCore::IR {

-Ref LoadEffectiveAddress(IREmitter* IREmit, AddressMode A, IR::OpSize GPRSize, bool AddSegmentBase, bool AllowUpperGarbage) {
+Ref LoadEffectiveAddress(IREmitter* IREmit, const AddressMode& A, IR::OpSize GPRSize, bool AddSegmentBase, bool AllowUpperGarbage) {
  Ref Tmp = A.Base;

  if (A.Offset) {
@@ -51,8 +51,8 @@ Ref LoadEffectiveAddress(IREmitter* IREmit, AddressMode A, IR::OpSize GPRSize, b
  return Tmp ?: IREmit->Constant(0);
 }

-AddressMode SelectAddressMode(IREmitter* IREmit, AddressMode A, IR::OpSize GPRSize, bool HostSupportsTSOImm9, bool AtomicTSO, bool Vector,
-                              IR::OpSize AccessSize) {
+AddressMode SelectAddressMode(IREmitter* IREmit, const AddressMode& A, IR::OpSize GPRSize, bool HostSupportsTSOImm9, bool AtomicTSO,
+                              bool Vector, IR::OpSize AccessSize) {
  const auto Is32Bit = GPRSize == OpSize::i32Bit;
  const auto GPRSizeMatchesAddrSize = A.AddrSize == GPRSize;
  const auto OffsetIndexToLargeFor32Bit = Is32Bit && (A.Offset <= -16384 || A.Offset >= 16384);
@@ -103,7 +103,7 @@ AddressMode SelectAddressMode(IREmitter* IREmit, AddressMode A, IR::OpSize GPRSi
    return {
      .Base = LoadEffectiveAddress(IREmit, B, GPRSize, true /* AddSegmentBase */, false),
      .Index = IREmit->Constant(A.Offset),
-      .IndexType = MEM_OFFSET_SXTX,
+      .IndexType = MemOffsetType::SXTX,
      .IndexScale = 1,
    };
  }
@@ -111,15 +111,17 @@ AddressMode SelectAddressMode(IREmitter* IREmit, AddressMode A, IR::OpSize GPRSi
  if (AtomicTSO) {
    // TODO: LRCPC3 support for vector Imm9.
  } else if (!Is32Bit && A.Base && (A.Index || A.Segment) && !A.Offset && (A.IndexScale == 1 || A.IndexScale == AccessSizeAsImm)) {
+    AddressMode B = A;
+
    // ScaledRegisterLoadstore
-    if (A.Index && A.Segment) {
-      A.Base = IREmit->Add(GPRSize, A.Base, A.Segment);
-    } else if (A.Segment) {
-      A.Index = A.Segment;
-      A.IndexScale = 1;
+    if (B.Index && B.Segment) {
+      B.Base = IREmit->Add(GPRSize, B.Base, B.Segment);
+    } else if (B.Segment) {
+      B.Index = B.Segment;
+      B.IndexScale = 1;
    }

-    return A;
+    return B;
  }

  if (Vector || !AtomicTSO) {
@@ -134,7 +136,7 @@ AddressMode SelectAddressMode(IREmitter* IREmit, AddressMode A, IR::OpSize GPRSi
        return {
          .Base = LoadEffectiveAddress(IREmit, B, GPRSize, true /* AddSegmentBase */, false),
          .Index = IREmit->Constant(A.Offset),
-          .IndexType = MEM_OFFSET_SXTX,
+          .IndexType = MemOffsetType::SXTX,
          .IndexScale = 1,
        };
      }
@@ -11,17 +11,18 @@ struct AddressMode {
  Ref Segment {nullptr};
  Ref Base {nullptr};
  Ref Index {nullptr};
-  MemOffsetType IndexType = MEM_OFFSET_SXTX;
-  uint8_t IndexScale = 1;
  int64_t Offset = 0;

+  MemOffsetType IndexType = MemOffsetType::SXTX;
+  uint8_t IndexScale = 1;
+
  // Size in bytes for the address calculation. 8 for an arm64 hardware mode.
  IR::OpSize AddrSize;
  bool NonTSO;
 };

-Ref LoadEffectiveAddress(IREmitter* IREmit, AddressMode A, IR::OpSize GPRSize, bool AddSegmentBase, bool AllowUpperGarbage = false);
-AddressMode SelectAddressMode(IREmitter* IREmit, AddressMode A, IR::OpSize GPRSize, bool HostSupportsTSOImm9, bool AtomicTSO, bool Vector,
-                              IR::OpSize AccessSize);
+Ref LoadEffectiveAddress(IREmitter* IREmit, const AddressMode& A, IR::OpSize GPRSize, bool AddSegmentBase, bool AllowUpperGarbage = false);
+AddressMode SelectAddressMode(IREmitter* IREmit, const AddressMode& A, IR::OpSize GPRSize, bool HostSupportsTSOImm9, bool AtomicTSO,
+                              bool Vector, IR::OpSize AccessSize);

-}; // namespace FEXCore::IR
+} // namespace FEXCore::IR
@@ -1,10 +1,10 @@
 // SPDX-License-Identifier: MIT
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
-#include "FEXCore/Core/X86Enums.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Context/Context.h"

 #include <FEXCore/Core/CoreState.h>
+#include <FEXCore/Core/X86Enums.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/MathUtils.h>

@@ -41,7 +41,7 @@ namespace FEXCore::CPU {
 //    r19-r29 and SP.

 namespace x64 {
-#ifndef _M_ARM_64EC
+#ifndef ARCHITECTURE_arm64ec
  // All but x19 and x29 are caller saved
  // Note that rax/rdx are rearranged here so we can coalesce cmpxchg.
  constexpr std::array<ARMEmitter::Register, 18> SRA = {
@@ -417,18 +417,34 @@ FEXCore::X86State::X86Reg Arm64Emitter::GetX86RegRelationToARMReg(ARMEmitter::Re
  return FEXCore::X86State::X86Reg::REG_INVALID;
 }

-void Arm64Emitter::LoadConstant(ARMEmitter::Size s, ARMEmitter::Register Reg, uint64_t Constant, bool NOPPad) {
+void Arm64Emitter::LoadConstant(ARMEmitter::Size s, ARMEmitter::Register Reg, uint64_t Constant, PadType Pad, int MaxBytes) {
+  bool NOPPad = false;
+  if (Pad == PadType::DOPAD) {
+    NOPPad = true;
+  } else if (Pad == PadType::NOPAD) {
+    NOPPad = false;
+  } else if (Pad == PadType::AUTOPAD) {
+    // Force NOP padding to ensure relocated constants always have enough encoding space available
+    NOPPad = EnableCodeCaching;
+  }
+
  bool Is64Bit = s == ARMEmitter::Size::i64Bit;
-  int Segments = Is64Bit ? 4 : 2;
+  const auto UpperBound = Is64Bit ? 4 : 2;
+  int Segments = MaxBytes ? (MaxBytes / 2) : UpperBound;
+
+  LOGMAN_THROW_A_FMT(MaxBytes >= 0 && MaxBytes <= (UpperBound * 2) && (MaxBytes & 1) == 0,
+                     "MaxBytes must be bounded in the range of [0, {}] and 16-bit aligned", UpperBound);
+  // If MaxBytes specified then make sure to sanity check incoming data.
+  LOGMAN_THROW_A_FMT(MaxBytes == 0 || (Constant >> (MaxBytes * 8)) == 0, "MaxBytes provided but data can't fit within provided range.");

  if (Is64Bit && ((~Constant) >> 16) == 0) {
-    movn(s, Reg, (~Constant) & 0xFFFF);
-
    if (NOPPad) {
      nop();
      nop();
      nop();
    }
+
+    movn(s, Reg, (~Constant) & 0xFFFF);
    return;
  }

@@ -436,17 +452,17 @@ void Arm64Emitter::LoadConstant(ARMEmitter::Size s, ARMEmitter::Register Reg, ui
    // If the upper 32-bits is all zero, we can now switch to a 32-bit move.
    s = ARMEmitter::Size::i32Bit;
    Is64Bit = false;
-    Segments = 2;
+    Segments = std::min(Segments, 2);
  }

  if (!Is64Bit && ((~Constant) & 0xFFFF0000) == 0) {
-    movn(s, Reg.W(), (~Constant) & 0xFFFF);
-
    if (NOPPad) {
      nop();
      nop();
      nop();
    }
+
+    movn(s, Reg.W(), (~Constant) & 0xFFFF);
    return;
  }

@@ -467,24 +483,24 @@ void Arm64Emitter::LoadConstant(ARMEmitter::Size s, ARMEmitter::Register Reg, ui
    // `movz` is better than `orr` since hardware will rename or merge if possible when `movz` is used.
    const auto IsImm = ARMEmitter::Emitter::IsImmLogical(Constant, RegSizeInBits(s));
    if (IsImm) {
-      orr(s, Reg, ARMEmitter::Reg::zr, Constant);
      if (NOPPad) {
        nop();
        nop();
        nop();
      }
+      orr(s, Reg, ARMEmitter::Reg::zr, Constant);
      return;
    }
  }

  // If we can't handle negatives with the orr, try with movn+movk
  if (Is64Bit && ((~Constant) >> 32) == 0) {
-    movn(s, Reg, (~Constant) & 0xFFFF);
-    movk(s, Reg, (Constant >> 16) & 0xFFFF, 16);
    if (NOPPad) {
      nop();
      nop();
    }
+    movn(s, Reg, (~Constant) & 0xFFFF);
+    movk(s, Reg, (Constant >> 16) & 0xFFFF, 16);
    return;
  }

@@ -786,7 +802,7 @@ void Arm64Emitter::FillStaticRegs(bool FPRs, uint32_t GPRFillMask, uint32_t FPRF
  auto TmpReg = *OptionalReg;
  auto TmpReg2 = *OptionalReg2;

-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
  // Load STATE in from the CPU area as x28 is not callee saved in the ARM64EC ABI.
  ldr(TmpReg.X(), ARMEmitter::Reg::r18, TEB_CPU_AREA_OFFSET);
  ldr(STATE, TmpReg, CPU_AREA_EMULATOR_DATA_OFFSET);
@@ -1,36 +1,38 @@
 // SPDX-License-Identifier: MIT
 #pragma once

-#include "FEXCore/Utils/EnumUtils.h"
-#include "Interface/Core/JIT/Relocations.h"
+#include <FEXCore/Config/Config.h>

 #ifdef VIXL_DISASSEMBLER
 #include <aarch64/disasm-aarch64.h>
+#include <FEXCore/fextl/memory.h>
+#include <FEXCore/fextl/vector.h>
 #endif
 #ifdef VIXL_SIMULATOR
 #include <aarch64/simulator-aarch64.h>
 #include <aarch64/simulator-constants-aarch64.h>
 #endif

-#include <FEXCore/Core/X86Enums.h>
-#include <FEXCore/Config/Config.h>
-#include <FEXCore/fextl/vector.h>
 #include <CodeEmitter/Emitter.h>
 #include <CodeEmitter/Registers.h>

 #include <cstddef>
 #include <cstdint>
+#include <optional>
 #include <span>

 namespace FEXCore::Context {
 class ContextImpl;
 }
+namespace FEXCore::X86State {
+enum X86Reg : uint32_t;
+}

 namespace FEXCore::CPU {
 // Contains the address to the currently available CPU state
 constexpr auto STATE = ARMEmitter::XReg::x28;

-#ifndef _M_ARM_64EC
+#ifndef ARCHITECTURE_arm64ec
 // 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 = ARMEmitter::XReg::x0;
@@ -104,9 +106,20 @@ constexpr ARMEmitter::PRegister PRED_TMP_32B = ARMEmitter::PReg::p7;
 // This class contains common emitter utility functions that can
 // be used by both Arm64 JIT and ARM64 Dispatcher
 class Arm64Emitter : public ARMEmitter::Emitter {
-protected:
+public:
  Arm64Emitter(FEXCore::Context::ContextImpl* ctx, void* EmissionPtr = nullptr, size_t size = 0);

+  enum class PadType {
+    // Explicitly does not need padding, even if code-caching is enabled.
+    NOPAD,
+    // Explicitly needs padding, even if code-caching is disabled.
+    DOPAD,
+    // Choose to pad or not depending on if code-caching is enabled.
+    AUTOPAD,
+  };
+  void LoadConstant(ARMEmitter::Size s, ARMEmitter::Register Reg, uint64_t Constant, PadType Pad = PadType::NOPAD, int MaxBytes = 0);
+
+protected:
  FEXCore::Context::ContextImpl* EmitterCTX;

  std::span<const ARMEmitter::Register> StaticRegisters {};
@@ -116,8 +129,6 @@ protected:
  std::span<const ARMEmitter::VRegister> GeneralFPRegisters {};
  uint32_t PairRegisters = 0;

-  void LoadConstant(ARMEmitter::Size s, ARMEmitter::Register Reg, uint64_t Constant, bool NOPPad = false);
-
  void FillSpecialRegs(ARMEmitter::Register TmpReg, ARMEmitter::Register TmpReg2, bool SetFIZ, bool SetPredRegs);

  // Correlate an ARM register back to an x86 register index.
@@ -270,6 +281,8 @@ protected:

  FEX_CONFIG_OPT(Disassemble, DISASSEMBLE);
 #endif
+
+  FEX_CONFIG_OPT(EnableCodeCaching, ENABLECODECACHINGWIP);
 };

 } // namespace FEXCore::CPU
@@ -1,14 +1,17 @@
 // SPDX-License-Identifier: MIT
-#include "FEXCore/IR/IR.h"
-#include "FEXCore/Utils/AllocatorHooks.h"
 #include "Interface/Context/Context.h"
 #include "Interface/Core/CPUBackend.h"
+#include "Interface/Core/LookupCache.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
+
+#include <FEXCore/IR/IR.h>
+#include <FEXCore/Utils/AllocatorHooks.h>
+#include <FEXCore/Utils/PrctlUtils.h>
+
 #include <cstdint>

-#include "LookupCache.h"
-
 #ifndef _WIN32
+#include <linux/prctl.h>
 #include <sys/prctl.h>
 #endif

@@ -274,37 +277,37 @@ namespace CPU {
    : ThreadState(ThreadState)
    , CodeBuffers(CodeBuffers) {

-    auto& Common = ThreadState->CurrentFrame->Pointers.Common;
+    auto& Ptrs = ThreadState->CurrentFrame->Pointers;

    // Initialize named vector constants.
    for (size_t i = 0; i < FEXCore::IR::NamedVectorConstant::NAMED_VECTOR_CONST_POOL_MAX; ++i) {
-      Common.NamedVectorConstantPointers[i] = reinterpret_cast<uint64_t>(NamedVectorConstants[i]);
+      Ptrs.NamedVectorConstantPointers[i] = reinterpret_cast<uint64_t>(NamedVectorConstants[i]);
    }

    // Copy named vector constants.
-    memcpy(Common.NamedVectorConstants, NamedVectorConstants, sizeof(NamedVectorConstants));
+    memcpy(Ptrs.NamedVectorConstants, NamedVectorConstants, sizeof(NamedVectorConstants));

    // Initialize Indexed named vector constants.
-    Common.IndexedNamedVectorConstantPointers[FEXCore::IR::IndexNamedVectorConstant::INDEXED_NAMED_VECTOR_PSHUFLW] =
+    Ptrs.IndexedNamedVectorConstantPointers[FEXCore::IR::IndexNamedVectorConstant::INDEXED_NAMED_VECTOR_PSHUFLW] =
      reinterpret_cast<uint64_t>(PSHUFLW_LUT.data());
-    Common.IndexedNamedVectorConstantPointers[FEXCore::IR::IndexNamedVectorConstant::INDEXED_NAMED_VECTOR_PSHUFHW] =
+    Ptrs.IndexedNamedVectorConstantPointers[FEXCore::IR::IndexNamedVectorConstant::INDEXED_NAMED_VECTOR_PSHUFHW] =
      reinterpret_cast<uint64_t>(PSHUFHW_LUT.data());
-    Common.IndexedNamedVectorConstantPointers[FEXCore::IR::IndexNamedVectorConstant::INDEXED_NAMED_VECTOR_PSHUFD] =
+    Ptrs.IndexedNamedVectorConstantPointers[FEXCore::IR::IndexNamedVectorConstant::INDEXED_NAMED_VECTOR_PSHUFD] =
      reinterpret_cast<uint64_t>(PSHUFD_LUT.data());
-    Common.IndexedNamedVectorConstantPointers[FEXCore::IR::IndexNamedVectorConstant::INDEXED_NAMED_VECTOR_SHUFPS] =
+    Ptrs.IndexedNamedVectorConstantPointers[FEXCore::IR::IndexNamedVectorConstant::INDEXED_NAMED_VECTOR_SHUFPS] =
      reinterpret_cast<uint64_t>(SHUFPS_LUT.data());
-    Common.IndexedNamedVectorConstantPointers[FEXCore::IR::IndexNamedVectorConstant::INDEXED_NAMED_VECTOR_DPPS_MASK] =
+    Ptrs.IndexedNamedVectorConstantPointers[FEXCore::IR::IndexNamedVectorConstant::INDEXED_NAMED_VECTOR_DPPS_MASK] =
      reinterpret_cast<uint64_t>(DPPS_MASK.data());
-    Common.IndexedNamedVectorConstantPointers[FEXCore::IR::IndexNamedVectorConstant::INDEXED_NAMED_VECTOR_DPPD_MASK] =
+    Ptrs.IndexedNamedVectorConstantPointers[FEXCore::IR::IndexNamedVectorConstant::INDEXED_NAMED_VECTOR_DPPD_MASK] =
      reinterpret_cast<uint64_t>(DPPD_MASK.data());
-    Common.IndexedNamedVectorConstantPointers[FEXCore::IR::IndexNamedVectorConstant::INDEXED_NAMED_VECTOR_PBLENDW] =
+    Ptrs.IndexedNamedVectorConstantPointers[FEXCore::IR::IndexNamedVectorConstant::INDEXED_NAMED_VECTOR_PBLENDW] =
      reinterpret_cast<uint64_t>(PBLENDW_LUT.data());

 #ifndef FEX_DISABLE_TELEMETRY
    // Fill in telemetry values
    for (size_t i = 0; i < FEXCore::Telemetry::TYPE_LAST; ++i) {
      auto& Telem = FEXCore::Telemetry::GetTelemetryValue(static_cast<FEXCore::Telemetry::TelemetryType>(i));
-      Common.TelemetryValueAddresses[i] = reinterpret_cast<uint64_t>(&Telem);
+      Ptrs.TelemetryValueAddresses[i] = reinterpret_cast<uint64_t>(&Telem);
    }
 #endif
  }
@@ -357,6 +360,8 @@ namespace CPU {
      LogMan::Msg::EFmt("Failed to mprotect last page of code buffer.");
    }

+    FEXCore::Allocator::VirtualName("FEXMemJIT", reinterpret_cast<void*>(Ptr), Size);
+
    LookupCache = fextl::make_unique<GuestToHostMap>();
  }

@@ -395,7 +400,7 @@ namespace CPU {
    Latest = Buffer;
    LatestOffset = 0;

-    OnCodeBufferAllocated(*Buffer);
+    OnCodeBufferAllocated(Buffer);

    return Buffer;
  }
@@ -17,6 +17,10 @@ $end_info$

 #include <cstdint>

+namespace FEXCore::CPU {
+union Relocation;
+}
+
 namespace FEXCore {

 namespace IR {
@@ -77,7 +81,7 @@ namespace CPU {
    // Protects writes to the latest CodeBuffer and changes to LatestOffset
    FEXCore::ForkableUniqueMutex CodeBufferWriteMutex;

-    virtual void OnCodeBufferAllocated(CodeBuffer&) {};
+    virtual void OnCodeBufferAllocated(const std::shared_ptr<CodeBuffer>&) {};

  private:
    fextl::shared_ptr<CodeBuffer> Latest;
@@ -157,18 +161,7 @@ namespace CPU {
    virtual CompiledCode CompileCode(uint64_t Entry, uint64_t Size, bool SingleInst, const FEXCore::IR::IRListView* IR,
                                     FEXCore::Core::DebugData* DebugData, bool CheckTF) = 0;

-    /**
-     * @brief Relocates a block of code from the JIT code object cache
-     *
-     * @param Entry - RIP of the entry
-     * @param SerializationData - Serialization data referring to the object cache for `Entry`
-     *
-     * @return An executable function pointer relocated from the cache object
-     */
-    [[nodiscard]]
-    virtual void* RelocateJITObjectCode(uint64_t /* Entry */, const CodeSerialize::CodeObjectFileSection* /* SerializationData */) {
-      return nullptr;
-    }
+    virtual fextl::vector<FEXCore::CPU::Relocation> TakeRelocations(uint64_t GuestBaseAddress) = 0;

    virtual void ClearCache() {}

@@ -14,6 +14,7 @@ $end_info$
 #include <FEXCore/Core/CPUID.h>
 #include <FEXCore/Core/HostFeatures.h>
 #include <FEXCore/Utils/FileLoading.h>
+#include <FEXCore/Utils/MathUtils.h>
 #include <FEXCore/fextl/string.h>
 #include <FEXHeaderUtils/Syscalls.h>

@@ -23,7 +24,7 @@ $end_info$

 namespace FEXCore {
 namespace ProductNames {
-#ifdef _M_ARM_64
+#ifdef ARCHITECTURE_arm64
  static const char ARM_UNKNOWN[] = "Unknown ARM CPU";
  static const char ARM_A57[] = "Cortex-A57";
  static const char ARM_A72[] = "Cortex-A72";
@@ -43,12 +44,15 @@ namespace ProductNames {
  static const char ARM_A715[] = "Cortex-A715";
  static const char ARM_A720[] = "Cortex-A720";
  static const char ARM_A725[] = "Cortex-A725";
+  static const char ARM_C1Pro[] = "C1-Pro";
+  static const char ARM_C1Premium[] = "C1-Premium";
  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_X3[] = "Cortex-X3";
  static const char ARM_X4[] = "Cortex-X4";
  static const char ARM_X925[] = "Cortex-X925";
+  static const char ARM_C1Ultra[] = "C1-Ultra";
  static const char ARM_N1[] = "Neoverse N1";
  static const char ARM_N2[] = "Neoverse N2";
  static const char ARM_N3[] = "Neoverse N3";
@@ -59,6 +63,7 @@ namespace ProductNames {
  static const char ARM_A65[] = "Cortex-A65";
  static const char ARM_A510[] = "Cortex-A510";
  static const char ARM_A520[] = "Cortex-A520";
+  static const char ARM_C1Nano[] = "C1-Nano";

  static const char ARM_Kryo200[] = "Kryo 2xx";
  static const char ARM_Kryo300[] = "Kryo 3xx";
@@ -70,6 +75,7 @@ namespace ProductNames {

  static const char ARM_Denver[] = "Nvidia Denver";
  static const char ARM_Carmel[] = "Nvidia Carmel";
+  static const char ARM_Olympus[] = "Nvidia Olympus";

  static const char ARM_Firestorm_M1[] = "Apple Firestorm (M1)";
  static const char ARM_Icestorm_M1[] = "Apple Icestorm (M1)";
@@ -83,8 +89,12 @@ namespace ProductNames {
  static const char ARM_Blizzard_M2Pro[] = "Apple Blizzard (M2 Pro)";
  static const char ARM_Avalanche_M2Max[] = "Apple Avalanche (M2 Max)";
  static const char ARM_Blizzard_M2Max[] = "Apple Blizzard (M2 Max)";
+  static const char ARM_AppleSilicon[] = "Apple Silicon";

  static const char ARM_ORYON_1[] = "Oryon-1";
+  static const char ARM_Ampere_1[] = "AmpereOne";
+  static const char ARM_Ampere_1A[] = "AmpereOneA";
+  static const char ARM_Ampere_1B[] = "AmpereOneB";
 #else
 #endif
 } // namespace ProductNames
@@ -130,7 +140,7 @@ constexpr uint32_t FAMILY_IDENTIFIER = GenerateFamily(CPUFamily {
 });
 #endif

-#ifdef _M_ARM_64
+#ifdef ARCHITECTURE_arm64
 uint32_t GetCycleCounterFrequency() {
  uint64_t Result {};
  __asm("mrs %[Res], CNTFRQ_EL0" : [Res] "=r"(Result));
@@ -170,7 +180,7 @@ void CPUIDEmu::SetupHostHybridFlag() {
  // CPU priority order
  // This is mostly arbitrary but will sort by some sort of CPU priority by performance
  // Relative list so things they will commonly end up in big.little configurations sort of relate
-  static constexpr std::array<CPUMIDR, 58> CPUMIDRs = {{
+  static constexpr std::array<CPUMIDR, 66> CPUMIDRs = {{
    // Typically big CPU cores
    {0x51, 0x001, 1, ProductNames::ARM_ORYON_1}, // Qualcomm Oryon-1

@@ -180,39 +190,48 @@ void CPUIDEmu::SetupHostHybridFlag() {
    {0x61, 0x029, 1, ProductNames::ARM_Firestorm_M1Max}, // Apple Firestorm (M1 Max)
    {0x61, 0x025, 1, ProductNames::ARM_Firestorm_M1Pro}, // Apple Firestorm (M1 Pro)
    {0x61, 0x023, 1, ProductNames::ARM_Firestorm_M1},    // Apple Firestorm (M1)
+    {0x61, 0, 1, ProductNames::ARM_AppleSilicon},        // QEmu Apple Silicon

-    {0x41, 0xd85, 1, ProductNames::ARM_X925},    // X925
-    {0x41, 0xd87, 1, ProductNames::ARM_A725},    // A725
-    {0x41, 0xd84, 1, ProductNames::ARM_V3},      // V3
-    {0x41, 0xd83, 1, ProductNames::ARM_V3AE},    // V3AE
-    {0x41, 0xd8e, 1, ProductNames::ARM_N3},      // N3
-    {0x41, 0xd82, 1, ProductNames::ARM_X4},      // X4
-    {0x41, 0xd81, 1, ProductNames::ARM_A720},    // A720
-    {0x41, 0xd4e, 1, ProductNames::ARM_X3},      // X3
-    {0x41, 0xd4d, 1, ProductNames::ARM_A715},    // A715
-    {0x41, 0xd4f, 1, ProductNames::ARM_V2},      // V2
-    {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
+    {0x41, 0xd8c, 1, ProductNames::ARM_C1Ultra},   // C1-Ultra
+    {0x41, 0xd90, 1, ProductNames::ARM_C1Premium}, // C1-Premium
+    {0x41, 0xd8b, 1, ProductNames::ARM_C1Pro},     // C1-Pro
+    {0x41, 0xd85, 1, ProductNames::ARM_X925},      // X925
+    {0x41, 0xd87, 1, ProductNames::ARM_A725},      // A725
+    {0x41, 0xd84, 1, ProductNames::ARM_V3},        // V3
+    {0x41, 0xd83, 1, ProductNames::ARM_V3AE},      // V3AE
+    {0x41, 0xd8e, 1, ProductNames::ARM_N3},        // N3
+    {0x41, 0xd82, 1, ProductNames::ARM_X4},        // X4
+    {0x41, 0xd81, 1, ProductNames::ARM_A720},      // A720
+    {0x41, 0xd4e, 1, ProductNames::ARM_X3},        // X3
+    {0x41, 0xd4d, 1, ProductNames::ARM_A715},      // A715
+    {0x41, 0xd4f, 1, ProductNames::ARM_V2},        // V2
+    {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
+    {0xc0, 0xac3, 1, ProductNames::ARM_Ampere_1},  // AmpereOne
+    {0xc0, 0xac4, 1, ProductNames::ARM_Ampere_1A}, // AmpereOneA
+    {0xc0, 0xac5, 1, ProductNames::ARM_Ampere_1B}, // AmpereOneB
+
+    {0x4e, 0x010, 1, ProductNames::ARM_Olympus}, // Olympus
+    {0x4e, 0x004, 1, ProductNames::ARM_Carmel},  // Carmel

    // Denver rated above A57 to match TX2 weirdness
    {0x4e, 0x003, 1, ProductNames::ARM_Denver}, // Denver
@@ -227,6 +246,7 @@ void CPUIDEmu::SetupHostHybridFlag() {
    {0x61, 0x024, 0, ProductNames::ARM_Icestorm_M1Pro}, // Apple Icestorm (M1 Pro)
    {0x61, 0x022, 0, ProductNames::ARM_Icestorm_M1},    // Apple Icestorm (M1)

+    {0x41, 0xd8a, 1, ProductNames::ARM_C1Nano},   // C1-Nano
    {0x41, 0xd80, 0, ProductNames::ARM_A520},     // A520
    {0x41, 0xd46, 0, ProductNames::ARM_A510},     // A510
    {0x41, 0xd06, 0, ProductNames::ARM_A65},      // A65
@@ -424,10 +444,10 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) const {

  Res.eax = FAMILY_IDENTIFIER;

-  Res.ebx = 0 |             // Brand index
-            (8 << 8) |      // Cache line size in bytes
-            (Cores << 16) | // Number of addressable IDs for the logical cores in the physical CPU
-            (0 << 24);      // Local APIC ID
+  Res.ebx = 0 |                 // Brand index
+            (8 << 8) |          // Cache line size in bytes
+            (Cores << 16) |     // Number of addressable IDs for the logical cores in the physical CPU
+            (GetCPUID() << 24); // Local APIC ID

  Res.ecx = (1 << 0) |                                      // SSE3
            (CTX->HostFeatures.SupportsPMULL_128Bit << 1) | // PCLMULQDQ
@@ -490,7 +510,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) const {
            (1 << 25) | // SSE
            (1 << 26) | // SSE2
            (0 << 27) | // Self Snoop
-            (1 << 28) | // Max APIC IDs reserved field is valid
+            (0 << 28) | // (HTT) Max APIC IDs reserved field is valid
            (1 << 29) | // Thermal monitor
            (0 << 30) | // Reserved
            (0 << 31);  // Pending break enable
@@ -837,10 +857,10 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_4000_0001h(uint32_t Leaf) con
  constexpr uint32_t MaximumSubLeafNumber = 2;
  if (Leaf == 0) {
    // EAX[3:0] Is the host architecture that FEX is running under
-#ifdef _M_X86_64
+#ifdef ARCHITECTURE_x86_64
    // EAX[3:0] = 1 = x86_64 host architecture
    Res.eax |= 0b0001;
-#elif defined(_M_ARM_64)
+#elif defined(ARCHITECTURE_arm64)
    // EAX[3:0] = 2 = AArch64 host architecture
    Res.eax |= 0b0010;
 #else
@@ -892,38 +912,38 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0001h(uint32_t Leaf) con

  Res.eax = FAMILY_IDENTIFIER;

-  Res.ecx = (1 << 0) |  // LAHF/SAHF
-            (1 << 1) |  // 0 = Single core product, 1 = multi core product
-            (0 << 2) |  // SVM
-            (1 << 3) |  // Extended APIC register space
-            (0 << 4) |  // LOCK MOV CR0 means MOV CR8
-            (1 << 5) |  // ABM instructions
-            (0 << 6) |  // SSE4a
-            (0 << 7) |  // Misaligned SSE mode
-            (1 << 8) |  // PREFETCHW
-            (0 << 9) |  // OS visible workaround support
-            (0 << 10) | // Instruction based sampling support
-            (0 << 11) | // XOP
-            (0 << 12) | // SKINIT
-            (0 << 13) | // Watchdog timer support
-            (0 << 14) | // Reserved
-            (0 << 15) | // Lightweight profiling support
-            (0 << 16) | // FMA4
-            (1 << 17) | // Translation cache extension
-            (0 << 18) | // Reserved
-            (0 << 19) | // Reserved
-            (0 << 20) | // Reserved
-            (0 << 21) | // XOP-TBM
-            (0 << 22) | // Topology extensions support
-            (0 << 23) | // Core performance counter extensions
-            (0 << 24) | // NB performance counter extensions
-            (0 << 25) | // Reserved
-            (0 << 26) | // Data breakpoints extensions
-            (0 << 27) | // Performance TSC
-            (0 << 28) | // L2 perf counter extensions
-            (0 << 29) | // MONITORX
-            (0 << 30) | // Reserved
-            (0 << 31);  // Reserved
+  Res.ecx = (1 << 0) |                               // LAHF/SAHF
+            (1 << 1) |                               // 0 = Single core product, 1 = multi core product
+            (0 << 2) |                               // SVM
+            (1 << 3) |                               // Extended APIC register space
+            (0 << 4) |                               // LOCK MOV CR0 means MOV CR8
+            (1 << 5) |                               // ABM instructions
+            (CTX->HostFeatures.SupportsSSE4a << 6) | // SSE4a
+            (0 << 7) |                               // Misaligned SSE mode
+            (1 << 8) |                               // PREFETCHW
+            (0 << 9) |                               // OS visible workaround support
+            (0 << 10) |                              // Instruction based sampling support
+            (0 << 11) |                              // XOP
+            (0 << 12) |                              // SKINIT
+            (0 << 13) |                              // Watchdog timer support
+            (0 << 14) |                              // Reserved
+            (0 << 15) |                              // Lightweight profiling support
+            (0 << 16) |                              // FMA4
+            (1 << 17) |                              // Translation cache extension
+            (0 << 18) |                              // Reserved
+            (0 << 19) |                              // Reserved
+            (0 << 20) |                              // Reserved
+            (0 << 21) |                              // XOP-TBM
+            (0 << 22) |                              // Topology extensions support
+            (0 << 23) |                              // Core performance counter extensions
+            (0 << 24) |                              // NB performance counter extensions
+            (0 << 25) |                              // Reserved
+            (0 << 26) |                              // Data breakpoints extensions
+            (0 << 27) |                              // Performance TSC
+            (0 << 28) |                              // L2 perf counter extensions
+            (0 << 29) |                              // MONITORX
+            (0 << 30) |                              // Reserved
+            (0 << 31);                               // Reserved

  Res.edx = (1 << 0) |                                // FPU
            (1 << 1) |                                // Virtual mode extensions
@@ -1077,9 +1097,9 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0008h(uint32_t Leaf) con
            (CTX->HostFeatures.SupportsCLZERO << 0); // CLZERO support

  uint32_t CoreCount = Cores - 1;
-  Res.ecx = (0 << 16) |                                  // PerfTscSize: Performance timestamp count size
-            ((uint32_t)std::log2(CoreCount + 1) << 12) | // ApicIdSize: Number of bits in ApicID
-            (CoreCount << 0);                            // Count count subtract one
+  Res.ecx = (0 << 16) |                    // PerfTscSize: Performance timestamp count size
+            (std::bit_ceil(Cores) << 12) | // ApicIdSize: Number of bits in ApicID
+            (CoreCount << 0);              // Count count subtract one

  return Res;
 }
@@ -1210,7 +1230,7 @@ CPUIDEmu::CPUIDEmu(const FEXCore::Context::ContextImpl* ctx)

  SetupFeatures();

-#ifdef _M_ARM_64
+#ifdef ARCHITECTURE_arm64
  if (SupportsCPUIndexInTPIDRRO) {
    GetCPUID = GetCPUID_TPIDRRO;
  }
@@ -159,7 +159,7 @@ private:

  struct CPUData {
    const char* ProductName {};
-#ifdef _M_ARM_64
+#ifdef ARCHITECTURE_arm64
    uint32_t MIDR {};
 #endif
    bool IsBig {};
@@ -277,7 +277,7 @@ private:
    // 0: Highest function parameter and ID
    {SupportsConstant::CONSTANT, NeedsLeafConstant::NOLEAFCONSTANT},
    // 1: Processor info
-    {SupportsConstant::CONSTANT, NeedsLeafConstant::NOLEAFCONSTANT},
+    {SupportsConstant::NONCONSTANT, NeedsLeafConstant::NOLEAFCONSTANT},
    // 2: Cache and TLB info
    {SupportsConstant::CONSTANT, NeedsLeafConstant::NOLEAFCONSTANT},
    // 3: Serial Number(previously), now reserved
@@ -0,0 +1,660 @@
+// SPDX-License-Identifier: MIT
+#include "Utils/SpinWaitLock.h"
+
+#include <Interface/Context/Context.h>
+#include <Interface/Core/ArchHelpers/Arm64Emitter.h>
+#include <Interface/Core/Dispatcher/Dispatcher.h>
+#include <Interface/Core/JIT/DebugData.h>
+#include <Interface/Core/JIT/Relocations.h>
+#include <Interface/Core/LookupCache.h>
+#include <Interface/Core/OpcodeDispatcher.h>
+#include <Interface/IR/PassManager.h>
+
+#include <FEXCore/Core/Thunks.h>
+#include <FEXCore/HLE/SyscallHandler.h>
+
+#include <FEXHeaderUtils/Filesystem.h>
+
+#include <git_version.h>
+
+#include <xxhash.h>
+
+#include <fstream>
+
+namespace FEXCore {
+
+#if __clang_major__ < 16
+ExecutableFileInfo::ExecutableFileInfo(fextl::unique_ptr<HLE::SourcecodeMap> Map, uint64_t FileId, fextl::string Filename)
+  : SourcecodeMap(std::move(Map))
+  , FileId(FileId)
+  , Filename(Filename) {}
+#endif
+
+fextl::string CodeMap::GetBaseFilename(const ExecutableFileInfo& MainExecutable, bool AddNombSuffix) {
+  auto FileId = MainExecutable.FileId;
+
+  std::string_view base_filename = FHU::Filesystem::GetFilename(std::string_view {MainExecutable.Filename});
+  if (FileId != 0xffff'ffff'ffff'ffff) {
+    return fextl::fmt::format("{}-{:016x}{}", base_filename, MainExecutable.FileId, AddNombSuffix ? "-nomb" : "");
+  }
+
+  return "";
+}
+
+fextl::map<CodeMapFileId, CodeMap::ParsedContents> CodeMap::ParseCodeMap(std::ifstream& File) {
+  fextl::map<CodeMapFileId, CodeMap::ParsedContents> Ret;
+  while (true) {
+    Entry Entry;
+    File.read(reinterpret_cast<char*>(&Entry), sizeof(Entry));
+    if (!File) {
+      break;
+    }
+
+    if (Entry.FileId == LoadExternalLibrary.FileId && Entry.BlockOffset == LoadExternalLibrary.BlockOffset) {
+      ExternalLibraryInfo Info;
+      File.read(reinterpret_cast<char*>(&Info), sizeof(Info));
+
+      fextl::string Filename;
+      std::getline(File, Filename, '\0');
+
+      // Align to 4-byte boundary
+      char Null[4];
+      File.read(Null, AlignUp(Filename.size() + 1, 4) - Filename.size() - 1);
+      if (!File) {
+        break;
+      }
+      Ret[Info.ExternalFileId].Filename = std::move(Filename);
+    } else if (Entry.FileId == SetExecutableFileId {}.Marker.FileId && Entry.BlockOffset == SetExecutableFileId {}.Marker.BlockOffset) {
+      CodeMapFileId ExecutableFileId;
+      File.read(reinterpret_cast<char*>(&ExecutableFileId), sizeof(ExecutableFileId));
+      if (!File) {
+        break;
+      }
+      Ret[ExecutableFileId].IsExecutable = true;
+    } else {
+      if (!Ret.contains(Entry.FileId)) {
+        LogMan::Msg::EFmt("Code map referenced unknown file id {:016x}", Entry.FileId);
+      } else {
+        Ret[Entry.FileId].Blocks.insert(Entry.BlockOffset);
+      }
+    }
+
+    if (!File) {
+      break;
+    }
+  }
+  return Ret;
+}
+
+CodeMapWriter::CodeMapWriter(CodeMapOpener& Opener, bool OpenEagerly)
+  : Buffer(4096)
+  , FileOpener(Opener) {
+  if (OpenEagerly) {
+    CodeMapFD = FileOpener.OpenCodeMapFile();
+  }
+}
+
+CodeMapWriter::~CodeMapWriter() {
+  if (CodeMapFD.value_or(-1) != -1) {
+    Flush(BufferOffset);
+    close(*CodeMapFD);
+  }
+}
+
+bool CodeMapWriter::IsWriteEnabled(const ExecutableFileSectionInfo& Section) {
+  if (CodeMapFD == -1) {
+    return false;
+  }
+
+  // PV libraries can't yet be read by FEXServer, so skip dumping them
+  if (Section.FileInfo.Filename.starts_with("/run/pressure-vessel")) {
+    return false;
+  }
+
+  if (CodeMapFD) {
+    return true;
+  }
+
+  // Acquire mutex and re-check CodeMapFD to avoid race conditions
+  auto lk = std::unique_lock {Mutex};
+  if (!CodeMapFD) {
+    CodeMapFD = FileOpener.OpenCodeMapFile();
+  }
+
+  return CodeMapFD != -1;
+}
+
+void CodeMapWriter::Flush(size_t Offset) {
+  // Acquire exclusive lock and flush circular buffer
+  std::unique_lock Lock {Mutex};
+  Flush(Offset, Lock);
+}
+
+void CodeMapWriter::Flush(size_t Offset, std::unique_lock<std::shared_mutex>&) {
+  write(*CodeMapFD, Buffer.data(), Offset);
+  BufferOffset = 0;
+}
+
+void CodeMapWriter::AppendBlock(const FEXCore::ExecutableFileSectionInfo& SectionInfo, uint64_t BlockEntry) {
+  if (!IsWriteEnabled(SectionInfo)) {
+    return;
+  }
+
+  BlockEntry -= SectionInfo.FileStartVA;
+  if (BlockEntry > std::numeric_limits<uint32_t>::max()) {
+    ERROR_AND_DIE_FMT("Cannot write code map");
+  }
+
+  // Register new library if not already known
+  bool NewLibraryLoad = false;
+  {
+    // Check prior registration with shared lock
+    std::shared_lock Lock {Mutex};
+    NewLibraryLoad = !KnownFileIds.contains(SectionInfo.FileInfo.FileId);
+  }
+  if (NewLibraryLoad) {
+    // Register to map with exclusive lock
+    std::unique_lock Lock {Mutex};
+    NewLibraryLoad &= KnownFileIds.insert(SectionInfo.FileInfo.FileId).second;
+  }
+  if (NewLibraryLoad) {
+    // Add entry to code map
+    AppendLibraryLoad(SectionInfo.FileInfo);
+  }
+
+  // Register the actual code block
+  CodeMap::Entry DataEntry {SectionInfo.FileInfo.FileId, static_cast<uint32_t>(BlockEntry)};
+  AppendData(std::as_bytes(std::span {&DataEntry, 1}));
+}
+
+void CodeMapWriter::AppendLibraryLoad(const FEXCore::ExecutableFileInfo& FileInfo) {
+  // See CodeMap::ExternalLibraryInfo
+  auto ExternalFileId = FileInfo.FileId;
+  auto TotalSize = AlignUp(sizeof(CodeMap::LoadExternalLibrary) + sizeof(ExternalFileId) + FileInfo.Filename.size() + 1, 4);
+  const auto Data = reinterpret_cast<char*>(alloca(TotalSize));
+  auto WritePtr = std::copy_n(reinterpret_cast<const char*>(&CodeMap::LoadExternalLibrary), sizeof(CodeMap::LoadExternalLibrary), Data);
+  WritePtr = std::copy_n(reinterpret_cast<const char*>(&ExternalFileId), sizeof(ExternalFileId), WritePtr);
+  WritePtr = std::copy(FileInfo.Filename.begin(), FileInfo.Filename.end(), WritePtr);
+  std::fill(WritePtr, Data + TotalSize, 0);
+  AppendData(std::as_bytes(std::span {Data, TotalSize}));
+}
+
+void CodeMapWriter::AppendSetMainExecutable(const FEXCore::ExecutableFileInfo& FileInfo) {
+  CodeMap::SetExecutableFileId Data {.ExecutableFileId = FileInfo.FileId};
+  AppendData(std::span {reinterpret_cast<const std::byte*>(&Data), sizeof(Data)});
+}
+
+void CodeMapWriter::AppendData(std::span<const std::byte> Data) {
+  std::shared_lock Lock {Mutex};
+  auto Offset = BufferOffset.fetch_add(Data.size_bytes());
+  if (Offset + Data.size_bytes() > Buffer.size()) {
+    // Acquire exclusive lock and flush the buffer.
+    // Under heavy pressure, multiple threads may observe an exhausted buffer simultaneously.
+    // The thread with the last in-bounds Offset is responsible for flushing the buffer.
+    Lock.unlock();
+    bool IsResponsibleForFlush = false;
+    {
+      std::unique_lock ExclusiveLock {Mutex};
+      IsResponsibleForFlush = (Offset <= Buffer.size());
+      if (IsResponsibleForFlush) {
+        Flush(Offset, ExclusiveLock);
+      }
+    }
+    if (!IsResponsibleForFlush) {
+      // Wait for the buffer to be flushed on the responsible thread
+      Utils::SpinWaitLock::WaitPred<std::less_equal<>, size_t>(reinterpret_cast<size_t*>(&BufferOffset), Buffer.size());
+    }
+    AppendData(Data);
+    return;
+  }
+
+  memcpy(&Buffer.at(Offset), Data.data(), Data.size_bytes());
+}
+
+} // namespace FEXCore
+
+namespace FEXCore::Context {
+
+CodeCache::CodeCache(ContextImpl& CTX_)
+  : CTX(CTX_) {}
+CodeCache::~CodeCache() = default;
+
+uint64_t CodeCache::ComputeCodeMapId(std::string_view Filename, int FD) {
+  if (Filename.empty()) {
+    return 0xffff'ffff'ffff'ffff;
+  }
+
+  // For now, we just use the file path as an identifier.
+  // TODO: Ensure the hash is unique enough to distinguish executables while remaining independent of the installation location
+  return XXH3_64bits(Filename.data(), Filename.size());
+}
+
+struct CodeCacheHeader {
+  std::array<char, 4> Magic = ExpectedMagic;
+  uint32_t FormatVersion = 1;
+  char FEXVersion[8] = {};
+  uint32_t NumBlocks;
+  uint32_t NumCodePages;
+  uint32_t CodeBufferSize;
+  uint32_t NumRelocations;
+  uint64_t SerializedBaseAddress;
+  // TODO: Consider including information from LookupCache.BlockLinks
+
+  static constexpr std::array<char, 4> ExpectedMagic = {'F', 'X', 'C', 'C'};
+};
+
+template<typename T>
+concept OrderedContainer = requires { typename T::key_compare; };
+
+bool CodeCache::SaveData(Core::InternalThreadState& Thread, int fd, const ExecutableFileSectionInfo& SourceBinary, uint64_t SerializedBaseAddress) {
+  auto CodeBuffer = CTX.GetLatest();
+  auto& LookupCache = *Thread.LookupCache->Shared;
+  auto Relocations = Thread.CPUBackend->TakeRelocations(SourceBinary.FileStartVA);
+
+  // Write file header
+  CodeCacheHeader header {};
+  constexpr std::string_view git_hash = GIT_SHORT_HASH;
+  static_assert(git_hash.size() <= sizeof(header.FEXVersion));
+  std::ranges::copy(git_hash, header.FEXVersion);
+  header.NumBlocks = LookupCache.BlockList.size();
+  header.NumCodePages = LookupCache.CodePages.size();
+  header.CodeBufferSize = CTX.LatestOffset;
+  header.NumRelocations = Relocations.size();
+  header.SerializedBaseAddress = SerializedBaseAddress;
+  ::write(fd, &header, sizeof(header));
+
+  // Dump guest<->host block mappings
+  {
+    // Cache contents must be deterministic, so copy the unordered block list and then sort by key
+    static_assert(!OrderedContainer<decltype(LookupCache.BlockList)>, "Already deterministic; drop temporary container");
+    fextl::vector<std::pair<uint64_t, const GuestToHostMap::BlockEntry*>> BlockList;
+    BlockList.reserve(LookupCache.BlockList.size());
+    for (auto& [Guest, BlockEntry] : LookupCache.BlockList) {
+      static_assert(sizeof(Guest) == 8, "Breaking change in code cache data layout");
+      BlockList.emplace_back(Guest, &BlockEntry);
+    }
+    std::ranges::sort(BlockList);
+
+    for (auto [Guest, Host] : BlockList) {
+      static_assert(sizeof(Host->HostCode) == 8, "Breaking change in code cache data layout");
+      static_assert(sizeof(Host->CodePages[0]) == 8, "Breaking change in code cache data layout");
+
+      Guest -= SourceBinary.FileStartVA;
+      ::write(fd, &Guest, sizeof(Guest));
+      uint64_t HostCode = Host->HostCode - reinterpret_cast<uintptr_t>(CodeBuffer->Ptr);
+      ::write(fd, &HostCode, sizeof(HostCode));
+      uint64_t NumCodePages = Host->CodePages.size();
+      ::write(fd, &NumCodePages, sizeof(NumCodePages));
+      LOGMAN_THROW_A_FMT(std::ranges::is_sorted(Host->CodePages), "Code pages aren't sorted");
+      for (auto CodePage : Host->CodePages) {
+        CodePage -= SourceBinary.FileStartVA;
+        ::write(fd, &CodePage, sizeof(CodePage));
+      }
+    }
+  }
+
+  // Dump relocations
+  static_assert(sizeof(Relocations[0]) == 48, "Breaking change in code cache data layout");
+  ::write(fd, Relocations.data(), Relocations.size() * sizeof(Relocations[0]));
+
+  // Pad to next page in file so that the CodeBuffer can be mmap'ed into process on load
+  char Zero[64] {};
+  auto Off = lseek(fd, 0, SEEK_CUR);
+  while (Off != AlignUp(Off, Utils::FEX_PAGE_SIZE)) {
+    auto BytesToWrite = std::min(AlignUp(Off, Utils::FEX_PAGE_SIZE) - Off, sizeof(Zero));
+    ::write(fd, Zero, BytesToWrite);
+    Off += BytesToWrite;
+  }
+
+  // Dump the host code (relocated for position-independent serialization)
+  std::vector CodeBufferData(reinterpret_cast<std::byte*>(CodeBuffer->Ptr), reinterpret_cast<std::byte*>(CodeBuffer->Ptr) + CTX.LatestOffset);
+  if (!ApplyCodeRelocations(SerializedBaseAddress, CodeBufferData, Relocations, true)) {
+    LOGMAN_THROW_A_FMT(false, "Failed to apply code relocations");
+    return false;
+  }
+  ::write(fd, CodeBufferData.data(), CodeBufferData.size());
+
+  // Dump code pages
+  static_assert(OrderedContainer<decltype(LookupCache.CodePages)>, "Non-deterministic data source");
+  for (const auto& [PageIndex, Entrypoints] : LookupCache.CodePages) {
+    uint64_t PageAddr = (PageIndex << 12) - SourceBinary.FileStartVA;
+    ::write(fd, &PageAddr, sizeof(PageAddr));
+    uint64_t NumEntrypoints = Entrypoints.size();
+    ::write(fd, &NumEntrypoints, sizeof(NumEntrypoints));
+    for (uint64_t Entrypoint : Entrypoints) {
+      Entrypoint -= SourceBinary.FileStartVA;
+      ::write(fd, &Entrypoint, sizeof(Entrypoint));
+    }
+  }
+
+  return true;
+}
+
+bool CodeCache::LoadData(Core::InternalThreadState* Thread, std::byte* MappedCacheFile, const ExecutableFileSectionInfo& BinarySection) {
+  if (!EnableCodeCaching) {
+    return true;
+  }
+
+  namespace ranges = std::ranges;
+
+  // Read file header
+  CodeCacheHeader header {};
+  ::memcpy(&header, MappedCacheFile, sizeof(header));
+  MappedCacheFile += sizeof(header);
+
+  LogMan::Msg::IFmt("Cache load: {:5} blocks; base={:#14x}; off={:#9x}-{:#09x}; {:016x} {}", header.NumBlocks, BinarySection.FileStartVA,
+                    BinarySection.BeginVA - BinarySection.FileStartVA, BinarySection.EndVA - BinarySection.FileStartVA,
+                    BinarySection.FileInfo.FileId, BinarySection.FileInfo.Filename);
+
+  if (!ranges::equal(header.Magic, header.ExpectedMagic)) {
+    LogMan::Msg::EFmt("Invalid cache file header");
+    return false;
+  }
+
+  char ExpectedVersion[8] = GIT_SHORT_HASH;
+  ranges::fill(ranges::find(ExpectedVersion, 0), std::end(ExpectedVersion), 0);
+  if (!ranges::equal(header.FEXVersion, ExpectedVersion)) {
+    LogMan::Msg::IFmt("Cache generated from old FEX version {}, current is {}; skipping", fmt::join(header.FEXVersion, ""),
+                      fmt::join(ExpectedVersion, ""));
+    return false;
+  }
+
+  if (header.NumBlocks == 0) {
+    // Valid caches are never empty
+    LogMan::Msg::IFmt("Code cache empty, aborting");
+    return false;
+  }
+
+  // Read guest<->host block mappings
+  using BlockListEntry = decltype(GuestToHostMap::BlockList)::value_type;
+  fextl::vector<BlockListEntry> BlockList(header.NumBlocks);
+  {
+    for (auto& BlockPtr : BlockList) {
+      ::memcpy(&BlockPtr.first, MappedCacheFile, sizeof(BlockPtr.first));
+      MappedCacheFile += sizeof(BlockPtr.first);
+      ::memcpy(&BlockPtr.second.HostCode, MappedCacheFile, sizeof(BlockPtr.second.HostCode));
+      MappedCacheFile += sizeof(BlockPtr.second.HostCode);
+      uint64_t NumGuestPages;
+      ::memcpy(&NumGuestPages, MappedCacheFile, sizeof(NumGuestPages));
+      MappedCacheFile += sizeof(NumGuestPages);
+
+      BlockPtr.second.CodePages.resize(NumGuestPages);
+      ::memcpy(BlockPtr.second.CodePages.data(), MappedCacheFile, std::span {BlockPtr.second.CodePages}.size_bytes());
+      MappedCacheFile += std::span {BlockPtr.second.CodePages}.size_bytes();
+    }
+
+    // Consistency check: VMA regions at the top and end should belong to the same file
+    auto [min_val, max_val] = ranges::minmax_element(BlockList, std::less {}, &decltype(BlockList)::value_type::first);
+    auto MinBound = CTX.SyscallHandler->LookupExecutableFileSection(Thread, min_val->first + BinarySection.FileStartVA);
+    auto MaxBound = CTX.SyscallHandler->LookupExecutableFileSection(Thread, max_val->first + BinarySection.FileStartVA);
+    if (&MinBound->FileInfo != &BinarySection.FileInfo || &MaxBound->FileInfo != &BinarySection.FileInfo) {
+      ERROR_AND_DIE_FMT("Cached blocks offsets {:#x}-{:#x} out of bounds for guest library {} ({:016x} @ {:#x}) while trying to load "
+                        "section {:#x}-{:#x}!",
+                        min_val->first, max_val->first, BinarySection.FileInfo.Filename, BinarySection.FileInfo.FileId,
+                        BinarySection.FileStartVA, BinarySection.BeginVA, BinarySection.EndVA);
+    }
+
+    // Constrain BlockList to the given ExecutableFileSectionInfo
+    LOGMAN_THROW_A_FMT(ranges::is_sorted(BlockList, [](auto& a, auto& b) { return a.first < b.first; }), "Expected sorted block list");
+    auto begin = ranges::lower_bound(BlockList, BinarySection.BeginVA - BinarySection.FileStartVA, std::less {}, &BlockListEntry::first);
+    auto end =
+      ranges::upper_bound(begin, BlockList.end(), BinarySection.EndVA - BinarySection.FileStartVA - 1, std::less {}, &BlockListEntry::first);
+    BlockList.erase(end, BlockList.end());
+    BlockList.erase(BlockList.begin(), begin);
+    if (BlockList.empty()) {
+      // Not an error since there is just no data to load
+      LogMan::Msg::IFmt("No blocks cached in this range, aborting");
+      return true;
+    }
+  }
+
+  // Read relocations
+  fextl::vector<FEXCore::CPU::Relocation> Relocations(header.NumRelocations, FEXCore::CPU::Relocation::Default());
+  ::memcpy(Relocations.data(), MappedCacheFile, Relocations.size() * sizeof(Relocations[0]));
+  MappedCacheFile += Relocations.size() * sizeof(Relocations[0]);
+
+  // Pad to next page in file, which contains CodeBuffer data
+  MappedCacheFile = reinterpret_cast<std::byte*>(AlignUp(reinterpret_cast<uintptr_t>(MappedCacheFile), Utils::FEX_PAGE_SIZE));
+
+  // Prepare CodeBuffer: Page aligned and big enough to hold all cached data
+  auto Lock = std::unique_lock {CTX.CodeBufferWriteMutex};
+  if (Thread) {
+    if (auto Prev = Thread->CPUBackend->CheckCodeBufferUpdate()) {
+      Allocator::VirtualDontNeed(Thread->CallRetStackBase, FEXCore::Core::InternalThreadState::CALLRET_STACK_SIZE);
+      auto lk = Thread->LookupCache->AcquireWriteLock();
+      Thread->LookupCache->ChangeGuestToHostMapping(*Prev, *CTX.GetLatest()->LookupCache, lk);
+    }
+  }
+
+  auto CodeBuffer = CTX.GetLatest();
+  LOGMAN_THROW_A_FMT(header.CodeBufferSize <= CodeBuffer->Size, "CodeBuffer too small to load code cache");
+  LOGMAN_THROW_A_FMT(reinterpret_cast<uintptr_t>(CodeBuffer->Ptr) % 0x1000 == 0, "Expected CodeBuffer base to be page-aligned");
+  const auto Delta = AlignUp(CTX.LatestOffset, 0x1000) - CTX.LatestOffset;
+  CTX.LatestOffset += Delta;
+
+  while (CTX.LatestOffset + header.CodeBufferSize > CodeBuffer->Size - Utils::FEX_PAGE_SIZE) {
+    if (Thread) {
+      CTX.ClearCodeCache(Thread);
+      CodeBuffer = CTX.GetLatest();
+      LogMan::Msg::IFmt("Increased code buffer size to {} MiB for cache load", CodeBuffer->Size / 1024 / 1024);
+    } else {
+      ERROR_AND_DIE_FMT("Cannot extend codebuffer without thread!");
+    }
+  }
+
+  // Read CodeBuffer data from file. Make sure the destination is page-aligned.
+  // TODO: Only load the data needed for the selected section
+  auto CodeBufferRange = std::as_writable_bytes(std::span {CodeBuffer->Ptr, CodeBuffer->Size}).subspan(CTX.LatestOffset, header.CodeBufferSize);
+  ::memcpy(CodeBufferRange.data(), MappedCacheFile, header.CodeBufferSize);
+  MappedCacheFile += header.CodeBufferSize;
+  CTX.LatestOffset += header.CodeBufferSize;
+
+  // Apply FEX relocations
+  auto Ret = ApplyCodeRelocations(BinarySection.FileStartVA, CodeBufferRange, Relocations, false);
+  LOGMAN_THROW_A_FMT(Ret == true, "Failed to apply code cache relocations");
+
+  {
+    auto& LookupCache = *CodeBuffer->LookupCache;
+    auto WriteLock = LookupCache.AcquireWriteLock();
+
+    // Register blocks to LookupCache
+    for (auto& [Guest, Host] : BlockList) {
+      for (auto& CodePage : Host.CodePages) {
+        CodePage += BinarySection.FileStartVA;
+      }
+      auto HostCode = reinterpret_cast<void*>(Host.HostCode + reinterpret_cast<uintptr_t>(CodeBufferRange.data()));
+      LookupCache.AddBlockMapping(Guest + BinarySection.FileStartVA, std::move(Host.CodePages), HostCode, WriteLock);
+    }
+
+    // Register loaded code ranges
+    fextl::vector<uint64_t> Entrypoints;
+    for (uint32_t i = 0; i < header.NumCodePages; ++i) {
+      uint64_t CodePage;
+      memcpy(&CodePage, MappedCacheFile, sizeof(CodePage));
+      CodePage += BinarySection.FileStartVA;
+      MappedCacheFile += sizeof(CodePage);
+
+      uint64_t NumEntrypoints;
+      memcpy(&NumEntrypoints, MappedCacheFile, sizeof(NumEntrypoints));
+      MappedCacheFile += sizeof(NumEntrypoints);
+
+      Entrypoints.resize(NumEntrypoints);
+      memcpy(Entrypoints.data(), MappedCacheFile, NumEntrypoints * sizeof(Entrypoints[0]));
+      MappedCacheFile += NumEntrypoints * sizeof(Entrypoints[0]);
+      for (auto& Entrypoint : Entrypoints) {
+        Entrypoint += BinarySection.FileStartVA;
+      }
+
+      if (LookupCache.AddBlockExecutableRange(Entrypoints, CodePage, FEXCore::Utils::FEX_PAGE_SIZE, WriteLock)) {
+        CTX.SyscallHandler->MarkGuestExecutableRange(Thread, CodePage, FEXCore::Utils::FEX_PAGE_SIZE);
+      }
+    }
+  }
+
+  if (EnableCodeCacheValidation) {
+    fextl::set<uint64_t> GuestBlocks, HostBlocks;
+    for (auto& [Guest, Host] : BlockList) {
+      GuestBlocks.insert(Guest + BinarySection.FileStartVA);
+      HostBlocks.insert(Host.HostCode);
+    }
+
+    Validate(BinarySection, std::move(GuestBlocks), HostBlocks, CodeBufferRange);
+  }
+
+  return true;
+}
+
+void CodeCache::Validate(const ExecutableFileSectionInfo& Section, fextl::set<uint64_t> GuestBlocks, const fextl::set<uint64_t>& HostBlocks,
+                         std::span<std::byte> CachedCode) {
+  LOGMAN_THROW_A_FMT(!HostBlocks.empty(), "Tried to validate without any host blocks");
+  // Skip any cached data before the first host block
+  CachedCode = CachedCode.subspan(*HostBlocks.begin() - sizeof(CPU::CPUBackend::JITCodeHeader));
+
+  if (!ValidationCTX) {
+    ValidationCTX.reset(static_cast<ContextImpl*>(FEXCore::Context::Context::CreateNewContext(CTX.HostFeatures).release()));
+    ValidationCTX->SetSignalDelegator(CTX.SignalDelegation);
+    ValidationCTX->SetSyscallHandler(CTX.SyscallHandler);
+    ValidationCTX->SetThunkHandler(CTX.ThunkHandler);
+    if (!ValidationCTX->InitCore()) {
+      ERROR_AND_DIE_FMT("Failed to create cache load validation context");
+    }
+
+    ValidationThread.reset(ValidationCTX->CreateThread(0, 0, nullptr));
+
+    auto Frame = ValidationThread->CurrentFrame;
+    Frame->State.segment_arrays[FEXCore::Core::CPUState::SEGMENT_ARRAY_INDEX_GDT] = &ValidationGDT[0];
+    Frame->State.segment_arrays[FEXCore::Core::CPUState::SEGMENT_ARRAY_INDEX_LDT] = &ValidationGDT[0];
+    Frame->State.cs_idx = 0;
+    Frame->State.cs_cached = 0;
+
+    if (ValidationCTX->Config.Is64BitMode()) {
+      ValidationGDT[0].L = 1; // L = Long Mode = 64-bit
+      ValidationGDT[0].D = 0; // D = Default Operand Size = Reserved
+    } else {
+      ValidationGDT[0].L = 0; // L = Long Mode = 32-bit
+      ValidationGDT[0].D = 1; // D = Default Operand Size = 32-bit
+    }
+  }
+
+  auto NewCodeBuffer = ValidationCTX->GetLatest();
+
+  std::span<std::byte> CodeBufferRangeRef =
+    std::as_writable_bytes(std::span {NewCodeBuffer->Ptr, NewCodeBuffer->Ptr + NewCodeBuffer->Size}).subspan(0, CachedCode.size_bytes());
+
+  while (!GuestBlocks.empty()) {
+    auto [CompiledBlocks, _, _2, _3, _4] = ValidationCTX->CompileCode(ValidationThread.get(), *GuestBlocks.begin(), 0 /* TODO: Set MaxInst? */);
+    for (auto& Entry : CompiledBlocks.EntryPoints) {
+      GuestBlocks.erase(Entry.first);
+    }
+  }
+
+  // Patch FEX-internal function addresses with values from the main Context to ensure the code blocks are comparable
+  auto NewRelocations = ValidationThread->CPUBackend->TakeRelocations(Section.FileStartVA);
+  NewRelocations.erase(std::remove_if(NewRelocations.begin(), NewRelocations.end(), [](const CPU::Relocation& Reloc) {
+    return Reloc.Header.Type != CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL && Reloc.Header.Type != CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE;
+  }));
+  (void)ApplyCodeRelocations(Section.FileStartVA, CodeBufferRangeRef, NewRelocations, false);
+
+  if (ValidationCTX->LatestOffset <= CodeBufferRangeRef.size()) {
+    // Reference compilation produced fewer bytes than our cache, so validation is going to fail.
+    // Make sure we don't output any garbage bytes though.
+    CodeBufferRangeRef = CodeBufferRangeRef.subspan(0, ValidationCTX->LatestOffset);
+  }
+
+  auto [Mismatch, _] = std::mismatch(CodeBufferRangeRef.begin(), CodeBufferRangeRef.end(), CachedCode.begin());
+  if (Mismatch != CodeBufferRangeRef.end()) {
+    // Align down to instruction size
+    auto Idx = AlignDown(std::distance(CodeBufferRangeRef.begin(), Mismatch), 4);
+
+    auto BlockIt = std::prev(HostBlocks.lower_bound(*HostBlocks.begin() + Idx + 1));
+    std::optional<uint64_t> GuestBlockAddr;
+    std::optional<uint64_t> GuestBlockAddrRef;
+    if (BlockIt != HostBlocks.end()) {
+      for (int i : {0, 1}) {
+        std::span Buffer = (i == 0 ? CachedCode : CodeBufferRangeRef);
+
+        // Second instruction is always a constant load for relative offset to the (multi)block start
+        int32_t addr = (*reinterpret_cast<uint32_t*>(&Buffer[*BlockIt - *HostBlocks.begin() + 4]) & 0x3ff'ffe0) << 11;
+        addr >>= 14;
+        auto header = reinterpret_cast<CPU::CPUBackend::JITCodeHeader*>(&Buffer[*BlockIt - *HostBlocks.begin() + 4 + addr]);
+        auto tail = reinterpret_cast<CPU::CPUBackend::JITCodeTail*>(reinterpret_cast<uintptr_t>(header) + header->OffsetToBlockTail);
+        (i == 0 ? GuestBlockAddr : GuestBlockAddrRef) = tail->RIP - Section.FileStartVA;
+        LogMan::Msg::EFmt("Recorded rip {}: {:#x} (offset {:#x})", i, tail->RIP, tail->RIP - Section.FileStartVA);
+
+        if (i == 1) {
+          if (tail->RIP >= Section.BeginVA && tail->RIP < Section.EndVA) {
+            auto [IRView, TotalInstructions, TotalInstructionsLength, StartAddr, Length, _] =
+              ValidationCTX->GenerateIR(ValidationThread.get(), tail->RIP, false, FEXCore::Config::Get_MAXINST());
+            fextl::stringstream ss;
+            FEXCore::IR::Dump(&ss, &*IRView);
+            LogMan::Msg::EFmt("IR:\n{}", ss.str());
+          } else {
+            LogMan::Msg::EFmt("Can't dump IR for out-of-range RIP {:#x}", tail->RIP);
+          }
+        }
+      }
+    }
+
+    fextl::string GuestBlockInfo = "UNKNOWN";
+    if (GuestBlockAddr) {
+      GuestBlockInfo = fextl::fmt::format("{:#x}", GuestBlockAddr.value());
+    }
+    if (GuestBlockAddr != GuestBlockAddrRef) {
+      GuestBlockInfo += " (MISMATCH)";
+    }
+    ERROR_AND_DIE_FMT("Cache validation failed at offset {:#x}: {:02x} <-> {:02x} (at {} <-> {}, guest block {})", Idx,
+                      fmt::join(CachedCode.subspan(Idx, 4), ""), fmt::join(CodeBufferRangeRef.subspan(Idx, 4), ""),
+                      fmt::ptr(CachedCode.data()), fmt::ptr(CodeBufferRangeRef.data()), GuestBlockInfo);
+  }
+
+  // Reset Context state for next validation
+  ValidationThread->LookupCache->ClearCache(ValidationThread->LookupCache->AcquireWriteLock());
+  ValidationCTX->LatestOffset = 0;
+
+  LogMan::Msg::IFmt("\tSuccessfully validated cache");
+}
+
+bool CodeCache::ApplyCodeRelocations(uint64_t GuestEntry, std::span<std::byte> Code,
+                                     std::span<const FEXCore::CPU::Relocation> EntryRelocations, bool ForStorage) {
+  CPU::Arm64Emitter Emitter(&CTX, Code.data(), Code.size_bytes());
+  for (size_t j = 0; j < EntryRelocations.size(); ++j) {
+    const FEXCore::CPU::Relocation& Reloc = EntryRelocations[j];
+    Emitter.SetCursorOffset(Reloc.Header.Offset);
+
+    switch (Reloc.Header.Type) {
+    case FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL: {
+      // Generate a literal so we can place it
+      uint64_t Pointer = ForStorage ? 0 : GetNamedSymbolLiteral(CTX, Reloc.NamedSymbolLiteral.Symbol);
+      Emitter.dc64(Pointer);
+      break;
+    }
+    case FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE: {
+      uint64_t Pointer = ForStorage ? 0 : reinterpret_cast<uint64_t>(CTX.ThunkHandler->LookupThunk(Reloc.NamedThunkMove.Symbol));
+      if (Pointer == ~0ULL) {
+        return false;
+      }
+      // Pointers are required to fit within 48-bit VA space.
+      Emitter.LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Register(Reloc.NamedThunkMove.RegisterIndex), Pointer,
+                           CPU::Arm64Emitter::PadType::DOPAD, 6);
+      break;
+    }
+    case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_LITERAL: {
+      Emitter.dc64(GuestEntry + Reloc.GuestRIP.GuestRIP);
+      break;
+    }
+    case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE: {
+      uint64_t Pointer = Reloc.GuestRIP.GuestRIP + GuestEntry;
+      // Pointers are required to fit within 48-bit VA space.
+      Emitter.LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Register(Reloc.GuestRIP.RegisterIndex), Pointer,
+                           CPU::Arm64Emitter::PadType::DOPAD, 6);
+      break;
+    }
+
+    default: ERROR_AND_DIE_FMT("Unknown relocation type {}", ToUnderlying(Reloc.Header.Type));
+    }
+  }
+
+  return true;
+}
+
+} // namespace FEXCore::Context
@@ -18,6 +18,7 @@ $end_info$
 #include "Interface/Core/JIT/JITClass.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Core/X86Tables/X86Tables.h"
+#include <Interface/GDBJIT/GDBJIT.h>
 #include "Interface/IR/IR.h"
 #include "Interface/IR/IREmitter.h"
 #include "Interface/IR/Passes/RegisterAllocationPass.h"
@@ -77,7 +78,7 @@ namespace FEXCore::Context {
 ContextImpl::ContextImpl(const FEXCore::HostFeatures& Features)
  : HostFeatures {Features}
  , CPUID {this}
-  , IRCaptureCache {this} {
+  , CodeCache {*this} {
  if (!Config.Is64BitMode()) {
    // When operating in 32-bit mode, the virtual memory we care about is only the lower 32-bits.
    Config.VirtualMemSize = 1ULL << 32;
@@ -344,7 +345,7 @@ bool ContextImpl::InitCore() {
  // Set up the SignalDelegator config since core is initialized.
  SignalDelegation->SetConfig(Dispatcher->MakeSignalDelegatorConfig());

-#if defined(_WIN32) && !defined(_M_ARM_64EC)
+#if defined(_WIN32) && !defined(ARCHITECTURE_arm64ec)
  // WOW64 always needs the interrupt fault check to be enabled.
  Config.NeedsPendingInterruptFaultCheck = true;
 #endif
@@ -362,6 +363,9 @@ void ContextImpl::HandleCallback(FEXCore::Core::InternalThreadState* Thread, uin
 }

 void ContextImpl::ExecuteThread(FEXCore::Core::InternalThreadState* Thread) {
+  // Update the thread pointer for Thunk return to the latest.
+  Thread->CurrentFrame->Pointers.ThunkCallbackRet = SignalDelegation->GetThunkCallbackRET();
+
  Dispatcher->ExecuteDispatch(Thread->CurrentFrame);

  // If it is the parent thread that died then just leave
@@ -375,8 +379,10 @@ void ContextImpl::InitializeCompiler(FEXCore::Core::InternalThreadState* Thread)
  Thread->FrontendDecoder = fextl::make_unique<FEXCore::Frontend::Decoder>(Thread);
  Thread->PassManager = fextl::make_unique<FEXCore::IR::PassManager>();

-  Thread->CurrentFrame->Pointers.Common.L1Pointer = Thread->LookupCache->GetL1Pointer();
-  Thread->CurrentFrame->Pointers.Common.L2Pointer = Thread->LookupCache->GetPagePointer();
+  Thread->CurrentFrame->State.L1Pointer = Thread->LookupCache->GetL1Pointer();
+  Thread->CurrentFrame->State.L1Mask = Thread->LookupCache->GetScaledL1PointerMask();
+
+  Thread->CurrentFrame->Pointers.L2Pointer = Thread->LookupCache->GetPagePointer();

  Dispatcher->InitThreadPointers(Thread);

@@ -397,6 +403,7 @@ ContextImpl::CreateThread(uint64_t InitialRIP, uint64_t StackPointer, const FEXC
  FEXCore::Core::InternalThreadState* Thread = new FEXCore::Core::InternalThreadState {
    .CTX = this,
  };
+  FEXCore::Allocator::VirtualName("FEXMem_ThreadState", Thread, sizeof(*Thread));

  Thread->CurrentFrame->State.gregs[X86State::REG_RSP] = StackPointer;
  Thread->CurrentFrame->State.rip = InitialRIP;
@@ -433,6 +440,10 @@ void ContextImpl::UnlockAfterFork(FEXCore::Core::InternalThreadState* LiveThread

  Profiler::PostForkAction(Child);
  if (Child) {
+    if (CodeMapWriter) {
+      CodeMapWriter->ResetAfterFork();
+    }
+
    CodeInvalidationMutex.StealAndDropActiveLocks();
    if (Config.StrictInProcessSplitLocks) {
      StrictSplitLockMutex = 0;
@@ -455,9 +466,14 @@ void ContextImpl::LockBeforeFork(FEXCore::Core::InternalThreadState* Thread) {
 }
 #endif

-void ContextImpl::OnCodeBufferAllocated(CPU::CodeBuffer& Buffer) {
+void ContextImpl::OnCodeBufferAllocated(const fextl::shared_ptr<CPU::CodeBuffer>& Buffer) {
  if (Config.GlobalJITNaming()) {
-    Symbols.RegisterJITSpace(Buffer.Ptr, Buffer.Size);
+    Symbols.RegisterJITSpace(Buffer->Ptr, Buffer->Size);
+  }
+
+  {
+    std::scoped_lock lk {CodeBufferListLock};
+    CodeBufferList.emplace_back(Buffer);
  }
 }

@@ -469,7 +485,8 @@ void ContextImpl::ClearCodeCache(FEXCore::Core::InternalThreadState* Thread, boo
    Thread->CPUBackend->ClearCache();
  } else {
    // Clear L1+L2 cache of this thread, and clear L3 cache across any threads using it
-    Thread->LookupCache->ClearCache();
+    auto lk = Thread->LookupCache->AcquireWriteLock();
+    Thread->LookupCache->ClearCache(lk);
  }
  Allocator::VirtualDontNeed(Thread->CallRetStackBase, FEXCore::Core::InternalThreadState::CALLRET_STACK_SIZE);
 }
@@ -610,7 +627,7 @@ ContextImpl::GenerateIR(FEXCore::Core::InternalThreadState* Thread, uint64_t Gue
            } else {
              ForceTSO = IR::ForceTSOMode::ForceDisabled;
            }
-          } else if (DecodedInfo->ForceTSO) {
+          } else if (DecodedInfo->Flags & X86Tables::DecodeFlags::FLAG_FORCE_TSO) {
            ForceTSO = IR::ForceTSOMode::ForceEnabled;
          }

@@ -641,10 +658,11 @@ ContextImpl::GenerateIR(FEXCore::Core::InternalThreadState* Thread, uint64_t Gue
              LogMan::Msg::EFmt("Invalid or Unknown instruction: {} 0x{:x}", TableInfo->Name ?: "UND", Block.Entry - GuestRIP);
            }

-            if (Block.BlockStatus == Frontend::Decoder::DecodedBlockStatus::NOEXEC_INST) {
-              Thread->OpDispatcher->NoExecOp(DecodedInfo);
-            } else {
+            if (Block.BlockStatus == Frontend::Decoder::DecodedBlockStatus::INVALID_INST ||
+                Block.BlockStatus == Frontend::Decoder::DecodedBlockStatus::BAD_RELOCATION) {
              Thread->OpDispatcher->InvalidOp(DecodedInfo);
+            } else {
+              Thread->OpDispatcher->NoExecOp(DecodedInfo);
            }
          }

@@ -708,9 +726,10 @@ ContextImpl::GenerateIR(FEXCore::Core::InternalThreadState* Thread, uint64_t Gue

 ContextImpl::CompileCodeResult ContextImpl::CompileCode(FEXCore::Core::InternalThreadState* Thread, uint64_t GuestRIP, uint64_t MaxInst) {
  if (SourcecodeResolver && Config.GDBSymbols()) {
-    auto AOTIRCacheEntry = SyscallHandler->LookupAOTIRCacheEntry(Thread, GuestRIP);
-    if (AOTIRCacheEntry.Entry) {
-      AOTIRCacheEntry.Entry->SourcecodeMap = SourcecodeResolver->GenerateMap(AOTIRCacheEntry.Entry->Filename, AOTIRCacheEntry.Entry->FileId);
+    auto MappedSection = SyscallHandler->LookupExecutableFileSection(Thread, GuestRIP);
+    if (MappedSection) {
+      MappedSection->FileInfo.SourcecodeMap =
+        SourcecodeResolver->GenerateMap(MappedSection->FileInfo.Filename, CodeMap::GetBaseFilename(MappedSection->FileInfo, false));
    }
  }

@@ -727,7 +746,7 @@ ContextImpl::CompileCodeResult ContextImpl::CompileCode(FEXCore::Core::InternalT
  // but this would increase lock contention. Redundant frontend runs aren't
  // as expensive and are easily reverted.
  if (MaxInst != 1) {
-    if (auto Block = Thread->LookupCache->FindBlock(GuestRIP)) {
+    if (auto Block = Thread->LookupCache->FindBlock(Thread, GuestRIP)) {
      Thread->OpDispatcher->DelayedDisownBuffer();
      return {.CompiledCode = {.BlockBegin = reinterpret_cast<uint8_t*>(Block), .EntryPoints = {{GuestRIP, reinterpret_cast<uint8_t*>(Block)}}},
              .DebugData = nullptr,
@@ -768,10 +787,13 @@ uintptr_t ContextImpl::CompileBlock(FEXCore::Core::CpuStateFrame* Frame, uint64_

  // Is the code in the cache?
  // The backends only check L1 and L2, not L3
-  if (auto HostCode = Thread->LookupCache->FindBlock(GuestRIP)) {
+  if (auto HostCode = Thread->LookupCache->FindBlock(Thread, GuestRIP)) {
    return HostCode;
  }

+  // Accumulate a JIT count now, as even if another thread raced us, it should count as a compile.
+  FEXCORE_PROFILE_INSTANT_INCREMENT(Thread, AccumulatedJITCount, 1);
+
  auto [CompiledCode, DebugData, StartAddr, Length, NeedsAddGuestCodeRanges] = CompileCode(Thread, GuestRIP, MaxInst);
  auto CodePtr = CompiledCode.EntryPoints[GuestRIP];
  if (CodePtr == nullptr) {
@@ -785,52 +807,72 @@ uintptr_t ContextImpl::CompileBlock(FEXCore::Core::CpuStateFrame* Frame, uint64_
  if (Config.BlockJITNaming()) {
    auto FragmentBasePtr = CompiledCode.BlockBegin;

-    if (DebugData) {
-      auto GuestRIPLookup = SyscallHandler->LookupAOTIRCacheEntry(Thread, GuestRIP);
+    auto GuestRIPLookup = SyscallHandler->LookupExecutableFileSection(Thread, GuestRIP);

-      if (DebugData->Subblocks.size()) {
-        for (auto& Subblock : DebugData->Subblocks) {
-          auto BlockBasePtr = FragmentBasePtr + Subblock.HostCodeOffset;
-          if (GuestRIPLookup.Entry) {
-            Symbols.Register(Thread->SymbolBuffer.get(), BlockBasePtr, CompiledCode.Size, GuestRIPLookup.Entry->Filename,
-                             GuestRIP - GuestRIPLookup.VAFileStart);
-          } else {
-            Symbols.Register(Thread->SymbolBuffer.get(), BlockBasePtr, GuestRIP, Subblock.HostCodeSize);
-          }
-        }
-      } else {
-        if (GuestRIPLookup.Entry) {
-          Symbols.Register(Thread->SymbolBuffer.get(), FragmentBasePtr, CompiledCode.Size, GuestRIPLookup.Entry->Filename,
-                           GuestRIP - GuestRIPLookup.VAFileStart);
+    if (DebugData->Subblocks.size()) {
+      for (auto& Subblock : DebugData->Subblocks) {
+        auto BlockBasePtr = FragmentBasePtr + Subblock.HostCodeOffset;
+        if (GuestRIPLookup) {
+          Symbols.Register(Thread->SymbolBuffer.get(), BlockBasePtr, CompiledCode.Size, GuestRIPLookup->FileInfo.Filename,
+                           GuestRIP - GuestRIPLookup->FileStartVA);
        } else {
-          Symbols.Register(Thread->SymbolBuffer.get(), FragmentBasePtr, GuestRIP, CompiledCode.Size);
+          Symbols.Register(Thread->SymbolBuffer.get(), BlockBasePtr, GuestRIP, Subblock.HostCodeSize);
        }
      }
+    } else {
+      if (GuestRIPLookup) {
+        Symbols.Register(Thread->SymbolBuffer.get(), FragmentBasePtr, CompiledCode.Size, GuestRIPLookup->FileInfo.Filename,
+                         GuestRIP - GuestRIPLookup->FileStartVA);
+      } else {
+        Symbols.Register(Thread->SymbolBuffer.get(), FragmentBasePtr, GuestRIP, CompiledCode.Size);
+      }
+    }
+  }
+
+  if (Config.LibraryJITNaming() || Config.GDBSymbols()) {
+    auto MappedSection = SyscallHandler->LookupExecutableFileSection(Thread, GuestRIP);
+    if (MappedSection) {
+      if (Config.LibraryJITNaming()) {
+        Symbols.RegisterNamedRegion(Thread->SymbolBuffer.get(), CodePtr, DebugData->HostCodeSize, MappedSection->FileInfo.Filename);
+      }
+
+      if (Config.GDBSymbols()) {
+        GDBJITRegister(MappedSection->FileInfo, MappedSection->FileStartVA, GuestRIP, (uintptr_t)CodePtr, *DebugData);
+      }
    }
  }

  // Clear any relocations that might have been generated
-  Thread->CPUBackend->ClearRelocations();
-
-  if (IRCaptureCache.PostCompileCode(Thread, CompiledCode.BlockBegin, GuestRIP, StartAddr, Length, DebugData.get())) {
-    // Early exit
-    return (uintptr_t)CodePtr;
+  if (!CodeCache.IsGeneratingCache) {
+    Thread->CPUBackend->ClearRelocations();
  }

+  fextl::vector<uint64_t> CodePages;
+
  if (NeedsAddGuestCodeRanges) {
    // Track in the guest to host map all entrypoints for all pages the compiled block touches, if any page didn't previously
    // contain code, inform the frontend so it can setup SMC detection.
    auto BlockInfo = Thread->FrontendDecoder->GetDecodedBlockInfo();
+    CodePages.reserve(BlockInfo->CodePages.size());
+    CodePages.insert(CodePages.end(), BlockInfo->CodePages.begin(), BlockInfo->CodePages.end());
    for (auto CodePage : BlockInfo->CodePages) {
-      if (Thread->LookupCache->AddBlockExecutableRange(BlockInfo->EntryPoints, CodePage, FEXCore::Utils::FEX_PAGE_SIZE)) {
+      if (Thread->LookupCache->AddBlockExecutableRange(Thread, BlockInfo->EntryPoints, CodePage, FEXCore::Utils::FEX_PAGE_SIZE)) {
        SyscallHandler->MarkGuestExecutableRange(Thread, CodePage, FEXCore::Utils::FEX_PAGE_SIZE);
      }
    }
  }

  // Insert to lookup cache
+
  for (auto [GuestAddr, HostAddr] : CompiledCode.EntryPoints) {
-    Thread->LookupCache->AddBlockMapping(GuestAddr, HostAddr);
+    Thread->LookupCache->AddBlockMapping(Thread, GuestAddr, CodePages, HostAddr);
+  }
+
+  if (CodeMapWriter) {
+    auto Region = SyscallHandler->LookupExecutableFileSection(Thread, GuestRIP);
+    if (Region && Region->FileStartVA != 0) {
+      CodeMapWriter->AppendBlock(*Region, GuestRIP);
+    }
  }

  return (uintptr_t)CodePtr;
@@ -857,66 +899,37 @@ uintptr_t ContextImpl::CompileSingleStep(FEXCore::Core::CpuStateFrame* Frame, ui
  return (uintptr_t)CodePtr;
 }

-static void InvalidateGuestThreadCodeRange(FEXCore::Core::InternalThreadState* Thread, InvalidatedEntryAccumulator& Accumulator,
-                                           uint64_t Start, uint64_t Length) {
+void ContextImpl::InvalidateCodeBuffersCodeRange(uint64_t Start, uint64_t Length) {
+  FEXCORE_PROFILE_SCOPED("InvalidateCodeBuffersCodeRange");
+
+  LOGMAN_THROW_A_FMT(CodeInvalidationMutex.try_lock() == false, "CodeInvalidationMutex needs to be unique_locked here");
+  std::scoped_lock lk {CodeBufferListLock};
+  auto it = CodeBufferList.begin();
+  while (it != CodeBufferList.end()) {
+    if (auto Strong = it->lock()) {
+      Strong->LookupCache->InvalidateRange(Start, Length);
+      it++;
+    } else {
+      it = CodeBufferList.erase(it);
+    }
+  }
+}
+
+void ContextImpl::InvalidateThreadCachedCodeRange(FEXCore::Core::InternalThreadState* Thread, uint64_t Start, uint64_t Length) {
+  LOGMAN_THROW_A_FMT(CodeInvalidationMutex.try_lock() == false, "CodeInvalidationMutex needs to be unique_locked here");
+
  // Ensures now-modified mappings aren't cached as being in their previous non-executable state.
  // Accessing FrontendDecoder is safe as the thread's code invalidation mutex must be locked here.
  Thread->FrontendDecoder->ResetExecutableRangeCache();

-  auto lk = Thread->LookupCache->AcquireLock();
-  auto& CodePages = Thread->LookupCache->Shared->CodePages;
+  if (Thread->LookupCache->InvalidateCacheRange(Start, Length)) {
+    FEXCORE_PROFILE_SCOPED("InvalidateCallRet");

-  auto lower = CodePages.lower_bound(Start >> 12);
-  auto upper = CodePages.upper_bound((Start + Length - 1) >> 12);
-
-  for (auto it = lower; it != upper; it++) {
-    Accumulator.emplace_back(std::move(it->second));
-  }
-
-  bool InvalidatedAnyEntries = false;
-  for (const auto& PageEntries : Accumulator) {
-    for (const auto& Entry : PageEntries) {
-      if (ContextImpl::ThreadRemoveCodeEntry(Thread, Entry)) {
-        InvalidatedAnyEntries = true;
-      }
-    }
-  }
-
-  if (InvalidatedAnyEntries) {
    // This may cause access violations in the thread on Windows as zeroing is not atomic, this is handled by the frontend
    Allocator::VirtualDontNeed(Thread->CallRetStackBase, FEXCore::Core::InternalThreadState::CALLRET_STACK_SIZE);
  }
 }

-void ContextImpl::InvalidateGuestCodeRange(FEXCore::Core::InternalThreadState* Thread, InvalidatedEntryAccumulator& Accumulator,
-                                           uint64_t Start, uint64_t Length) {
-  InvalidateGuestThreadCodeRange(Thread, Accumulator, Start, Length);
-}
-
-void ContextImpl::MarkMemoryShared(FEXCore::Core::InternalThreadState* Thread) {
-  if (!Thread) {
-    return;
-  }
-
-  if (!IsMemoryShared) {
-    IsMemoryShared = true;
-    UpdateAtomicTSOEmulationConfig();
-
-    if (Config.TSOAutoMigration) {
-      // Only the lookup cache is cleared here, so that old code can keep running until next compilation.
-      // This will leak previously compiled blocks until the CodeBuffer is cleared for some other reason.
-      Thread->LookupCache->ClearCache();
-    }
-  }
-}
-
-bool ContextImpl::ThreadRemoveCodeEntry(FEXCore::Core::InternalThreadState* Thread, uint64_t GuestRIP) {
-  LogMan::Throw::AFmt(static_cast<ContextImpl*>(Thread->CTX)->CodeInvalidationMutex.try_lock() == false, "CodeInvalidationMutex needs to "
-                                                                                                         "be unique_locked here");
-
-  return Thread->LookupCache->Erase(Thread->CurrentFrame, GuestRIP);
-}
-
 void ContextImpl::ThreadRemoveCodeEntryFromJit(FEXCore::Core::CpuStateFrame* Frame, uint64_t GuestRIP) {
  static_cast<ContextImpl*>(Frame->Thread->CTX)->SyscallHandler->InvalidateGuestCodeRange(Frame->Thread, GuestRIP, 1);
 }
@@ -958,12 +971,13 @@ void ContextImpl::AddThunkTrampolineIRHandler(uintptr_t Entrypoint, uintptr_t Gu

      const auto GPRSize = this->Config.Is64BitMode ? IR::OpSize::i64Bit : IR::OpSize::i32Bit;

+      // Thunk entry-points don't get cached, don't need to be padded.
      if (GPRSize == IR::OpSize::i64Bit) {
        IR::Ref R = emit->_StoreRegister(emit->Constant(Entrypoint), GPRSize);
-        R->Reg = IR::PhysicalRegister(IR::GPRFixedClass, X86State::REG_R11).Raw;
+        R->Reg = IR::PhysicalRegister(IR::RegClass::GPRFixed, X86State::REG_R11).Raw;
      } else {
-        emit->_StoreContext(GPRSize, IR::FPRClass, emit->_VCastFromGPR(IR::OpSize::i64Bit, IR::OpSize::i64Bit, emit->Constant(Entrypoint)),
-                            offsetof(Core::CPUState, mm[0][0]));
+        emit->_StoreContextFPR(GPRSize, emit->_VCastFromGPR(IR::OpSize::i64Bit, IR::OpSize::i64Bit, emit->Constant(Entrypoint)),
+                               offsetof(Core::CPUState, mm[0][0]));
      }
      emit->_ExitFunction(IR::OpSize::i64Bit, emit->Constant(GuestThunkEntrypoint), IR::BranchHint::None, emit->Invalid(), emit->Invalid());
    },
@@ -984,7 +998,7 @@ void ContextImpl::AddThunkTrampolineIRHandler(uintptr_t Entrypoint, uintptr_t Gu
 void ContextImpl::AddForceTSOInformation(const IntervalList<uint64_t>& ValidRanges, fextl::set<uint64_t>&& Instructions) {
  LogMan::Throw::AFmt(CodeInvalidationMutex.try_lock() == false, "CodeInvalidationMutex needs to be unique_locked here");
  ForceTSOValidRanges.Insert(ValidRanges);
-  ForceTSOInstructions.merge(Instructions);
+  ForceTSOInstructions.merge(std::move(Instructions));
 }

 void ContextImpl::RemoveForceTSOInformation(uint64_t Address, uint64_t Size) {
@@ -1015,9 +1029,9 @@ void ContextImpl::MonoBackpatcherWrite(FEXCore::Core::CpuStateFrame* Frame, uint
    auto lk = GuardSignalDeferringSection(CTX->CodeInvalidationMutex, Thread);

    if (Size == 8) {
-      *reinterpret_cast<uint64_t *>(Address) = Value;
+      *reinterpret_cast<uint64_t*>(Address) = Value;
    } else if (Size == 4) {
-      *reinterpret_cast<uint32_t *>(Address) = Value;
+      *reinterpret_cast<uint32_t*>(Address) = Value;
    } else {
      ERROR_AND_DIE_FMT("Unexpected write size for backpatcher: {}", Size);
    }
@@ -1026,15 +1040,7 @@ void ContextImpl::MonoBackpatcherWrite(FEXCore::Core::CpuStateFrame* Frame, uint
  CTX->SyscallHandler->InvalidateGuestCodeRange(Thread, Address, Size);
 }

-IR::AOTIRCacheEntry* ContextImpl::LoadAOTIRCacheEntry(const fextl::string& filename) {
-  auto rv = IRCaptureCache.LoadAOTIRCacheEntry(filename);
-  return rv;
-}
-
-void ContextImpl::UnloadAOTIRCacheEntry(IR::AOTIRCacheEntry* Entry) {}
-
 void ContextImpl::ConfigureAOTGen(FEXCore::Core::InternalThreadState* Thread, fextl::set<uint64_t>* ExternalBranches, uint64_t SectionMaxAddress) {
  Thread->FrontendDecoder->SetExternalBranches(ExternalBranches);
-  Thread->FrontendDecoder->SetSectionMaxAddress(SectionMaxAddress);
 }
 } // namespace FEXCore::Context
@@ -1,11 +1,10 @@
 // SPDX-License-Identifier: MIT

-#include "Common/SoftFloat.h"
+#include "Common/VectorRegType.h"
 #include "Interface/Context/Context.h"
 #include "Interface/Core/CPUBackend.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Core/LookupCache.h"
-#include "Interface/Core/X86HelperGen.h"
 #include "Utils/MemberFunctionToPointer.h"

 #include <FEXCore/Config/Config.h>
@@ -17,6 +16,7 @@
 #include <FEXCore/Utils/Event.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/MathUtils.h>
+#include <FEXHeaderUtils/Syscalls.h>

 #include <CodeEmitter/Emitter.h>

@@ -25,9 +25,7 @@
 #endif

 #include <array>
-#include <atomic>
 #include <bit>
-#include <condition_variable>
 #include <csignal>
 #include <cstring>

@@ -37,12 +35,14 @@ static void SleepThread(FEXCore::Context::ContextImpl* CTX, FEXCore::Core::CpuSt
  CTX->SyscallHandler->SleepThread(CTX, Frame);
 }

-constexpr size_t MAX_DISPATCHER_CODE_SIZE = 4096 * 4;
+constexpr size_t MAX_DISPATCHER_CODE_SIZE = FEXCore::Utils::FEX_PAGE_SIZE * 4;

 Dispatcher::Dispatcher(FEXCore::Context::ContextImpl* ctx)
  : Arm64Emitter(ctx, FEXCore::Allocator::VirtualAlloc(MAX_DISPATCHER_CODE_SIZE, true), MAX_DISPATCHER_CODE_SIZE)
  , CTX {ctx} {
  EmitDispatcher();
+
+  FEXCore::Allocator::VirtualName("FEXMem_Misc", reinterpret_cast<void*>(GetBufferBase()), MAX_DISPATCHER_CODE_SIZE);
 }

 Dispatcher::~Dispatcher() {
@@ -92,12 +92,12 @@ void Dispatcher::EmitDispatcher() {

  FillStaticRegs();
  ldr(RipReg, STATE_PTR(CpuStateFrame, State.rip));
-  cbnz(ARMEmitter::Size::i32Bit, ENTRY_FILL_SRA_SINGLE_INST_REG, &CompileSingleStep);
+  (void)cbnz(ARMEmitter::Size::i32Bit, ENTRY_FILL_SRA_SINGLE_INST_REG, &CompileSingleStep);

  ARMEmitter::BiDirectionalLabel LoopTop {};

-#ifdef _M_ARM_64EC
-  b(&LoopTop);
+#ifdef ARCHITECTURE_arm64ec
+  (void)b(&LoopTop);

  AbsoluteLoopTopAddressEnterECFillSRA = GetCursorAddress<uint64_t>();
  ldr(STATE, EC_ENTRY_CPUAREA_REG, CPU_AREA_EMULATOR_DATA_OFFSET);
@@ -105,10 +105,10 @@ void Dispatcher::EmitDispatcher() {

  ldr(RipReg, STATE_PTR(CpuStateFrame, State.rip));
  // Force a single instruction block if ENTRY_FILL_SRA_SINGLE_INST_REG is nonzero entering the JIT, used for inline SMC handling.
-  cbnz(ARMEmitter::Size::i32Bit, ENTRY_FILL_SRA_SINGLE_INST_REG, &CompileSingleStep);
+  (void)cbnz(ARMEmitter::Size::i32Bit, ENTRY_FILL_SRA_SINGLE_INST_REG, &CompileSingleStep);

  // Enter JIT
-  b(&LoopTop);
+  (void)b(&LoopTop);

  AbsoluteLoopTopAddressEnterEC = GetCursorAddress<uint64_t>();
  // Load ThreadState and write the target PC there
@@ -129,7 +129,7 @@ void Dispatcher::EmitDispatcher() {
  ldp<ARMEmitter::IndexType::OFFSET>(TMP1, TMP2, REG_CALLRET_SP);
  // EC_CALL_CHECKER_PC_REG is REG_PF which isn't touched by any of the above
  sub(ARMEmitter::Size::i64Bit, TMP1, EC_CALL_CHECKER_PC_REG, TMP1);
-  cbnz(ARMEmitter::Size::i64Bit, TMP1, &LoopTop);
+  (void)cbnz(ARMEmitter::Size::i64Bit, TMP1, &LoopTop);

  // If the entry at the TOS is for the target address, pop it and return to the JIT code
  add(ARMEmitter::Size::i64Bit, REG_CALLRET_SP, REG_CALLRET_SP, 0x10);
@@ -141,13 +141,13 @@ void Dispatcher::EmitDispatcher() {
  // We want to ensure that we are 16 byte aligned at the top of this loop
  Align16B();

-  Bind(&LoopTop);
+  (void)Bind(&LoopTop);
  AbsoluteLoopTopAddress = GetCursorAddress<uint64_t>();

  // Load in our RIP
  ldr(RipReg, STATE_PTR(CpuStateFrame, State.rip));

-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
  // Clobbers TMP1/2
  // Check the EC code bitmap incase we need to exit the JIT to call into native code.
  ARMEmitter::ForwardLabel l_NotECCode;
@@ -159,75 +159,82 @@ void Dispatcher::EmitDispatcher() {
  ldr(TMP1, TMP1, TMP2, ARMEmitter::ExtendedType::LSL_64, 0);
  lsr(ARMEmitter::Size::i64Bit, TMP2, RipReg, 12);
  lsrv(ARMEmitter::Size::i64Bit, TMP1, TMP1, TMP2);
-  tbz(TMP1, 0, &l_NotECCode);
+  (void)tbz(TMP1, 0, &l_NotECCode);

  str(REG_CALLRET_SP, STATE_PTR(CpuStateFrame, State.callret_sp));

  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, StaticRegisters[X86State::REG_RSP], 0);
  mov(EC_CALL_CHECKER_PC_REG, RipReg);
-  ldr(TMP2, STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionEC));
+  ldr(TMP2, STATE_PTR(CpuStateFrame, Pointers.ExitFunctionEC));
  br(TMP2);

-  Bind(&l_NotECCode);
+  (void)Bind(&l_NotECCode);
 #endif

  ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
-  cbnz(ARMEmitter::Size::i32Bit, TMP1, &CompileSingleStep);
-
-  // This is the block cache lookup routine
-  // It matches what is going on it LookupCache.h::FindBlock
-  ldr(TMP1, 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_(ARMEmitter::Size::i64Bit, TMP4, RipReg.R(), VirtualMemorySize - 1);
-  } else {
-    LoadConstant(ARMEmitter::Size::i64Bit, TMP4, VirtualMemorySize);
-    and_(ARMEmitter::Size::i64Bit, TMP4, RipReg.R(), TMP4);
-  }
+  (void)cbnz(ARMEmitter::Size::i32Bit, TMP1, &CompileSingleStep);

  ARMEmitter::ForwardLabel NoBlock;

-  {
-    // Offset the address and add to our page pointer
-    lsr(ARMEmitter::Size::i64Bit, TMP2, TMP4, 12);
+  if (DisableL2Cache()) {
+    (void)b(&NoBlock);
+  } else {
+    // This is the block cache lookup routine
+    // It matches what is going on it LookupCache.h::FindBlock
+    ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.L2Pointer));

-    // Load the pointer from the offset
-    ldr(TMP1, TMP1, TMP2, ARMEmitter::ExtendedType::LSL_64, 3);
+    // 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_(ARMEmitter::Size::i64Bit, TMP4, RipReg.R(), VirtualMemorySize - 1);
+    } else {
+      LoadConstant(ARMEmitter::Size::i64Bit, TMP4, VirtualMemorySize);
+      and_(ARMEmitter::Size::i64Bit, TMP4, RipReg.R(), TMP4);
+    }

-    // If page pointer is zero then we have no block
-    cbz(ARMEmitter::Size::i64Bit, TMP1, &NoBlock);
-
-    // Steal the page offset
-    and_(ARMEmitter::Size::i64Bit, TMP2, TMP4, 0x0FFF);
-
-    // Shift the offset by the size of the block cache entry
-    add(TMP1, TMP1, TMP2, ARMEmitter::ShiftType::LSL, (int)log2(sizeof(FEXCore::LookupCache::LookupCacheEntry)));
-
-    // The the full LookupCacheEntry with a single LDP.
-    // Check the guest address first to ensure it maps to the address we are currently at.
-    // This fixes aliasing problems
-    ldp<ARMEmitter::IndexType::OFFSET>(TMP4, TMP2, TMP1, 0);
-
-    // If the guest address doesn't match, Compile the block.
-    sub(TMP2, TMP2, RipReg);
-    cbnz(ARMEmitter::Size::i64Bit, TMP2, &NoBlock);
-
-    // Check the host address to see if it matches, else compile the block.
-    cbz(ARMEmitter::Size::i64Bit, TMP4, &NoBlock);
-
-    // If we've made it here then we have a real compiled block
    {
-      // update L1 cache
-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));
+      // Offset the address and add to our page pointer
+      lsr(ARMEmitter::Size::i64Bit, TMP2, TMP4, 12);

-      and_(ARMEmitter::Size::i64Bit, TMP2, RipReg.R(), LookupCache::L1_ENTRIES_MASK);
-      add(TMP1, TMP1, TMP2, ARMEmitter::ShiftType::LSL, 4);
-      stp<ARMEmitter::IndexType::OFFSET>(TMP4, RipReg, TMP1);
+      // Load the pointer from the offset
+      ldr(TMP1, TMP1, TMP2, ARMEmitter::ExtendedType::LSL_64, 3);

-      // Jump to the block
-      br(TMP4);
+      // If page pointer is zero then we have no block
+      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &NoBlock);
+
+      // Steal the page offset
+      and_(ARMEmitter::Size::i64Bit, TMP2, TMP4, 0x0FFF);
+
+      // Shift the offset by the size of the block cache entry
+      add(TMP1, TMP1, TMP2, ARMEmitter::ShiftType::LSL, FEXCore::ilog2(sizeof(LookupCache::LookupCacheEntry)));
+
+      // The the full LookupCacheEntry with a single LDP.
+      // Check the guest address first to ensure it maps to the address we are currently at.
+      // This fixes aliasing problems
+      ldp<ARMEmitter::IndexType::OFFSET>(TMP4, TMP2, TMP1, 0);
+
+      // If the guest address doesn't match, Compile the block.
+      sub(TMP2, TMP2, RipReg);
+      (void)cbnz(ARMEmitter::Size::i64Bit, TMP2, &NoBlock);
+
+      // Check the host address to see if it matches, else compile the block.
+      (void)cbz(ARMEmitter::Size::i64Bit, TMP4, &NoBlock);
+
+      // If we've made it here then we have a real compiled block
+      {
+        // update L1 cache
+        ldp<ARMEmitter::IndexType::OFFSET>(TMP1, TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, State.L1Pointer));
+
+        // Calculate (tmp1 + ((ripreg & L1_ENTRIES_MASK) << 4)) for the address
+        // L1Mask is pre-shifted.
+        and_(ARMEmitter::Size::i64Bit, TMP2, TMP2, RipReg.R(), ARMEmitter::ShiftType::LSL, FEXCore::ilog2(sizeof(LookupCache::LookupCacheEntry)));
+        add(TMP1, TMP1, TMP2);
+
+        stp<ARMEmitter::IndexType::OFFSET>(TMP4, RipReg, TMP1);
+
+        // Jump to the block
+        br(TMP4);
+      }
    }
  }

@@ -252,7 +259,7 @@ void Dispatcher::EmitDispatcher() {
    str(TMP2, STATE, offsetof(FEXCore::Core::CPUState, DeferredSignalRefCount));
 #endif

-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
    ldr(TMP2, ARMEmitter::XReg::x18, TEB_CPU_AREA_OFFSET);
    LoadConstant(ARMEmitter::Size::i32Bit, TMP1, 1);
    strb(TMP1.W(), TMP2, CPU_AREA_IN_SYSCALL_CALLBACK_OFFSET);
@@ -260,7 +267,7 @@ void Dispatcher::EmitDispatcher() {

    Body();

-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
    ldr(TMP2, ARMEmitter::XReg::x18, TEB_CPU_AREA_OFFSET);
    strb(ARMEmitter::WReg::zr, TMP2, CPU_AREA_IN_SYSCALL_CALLBACK_OFFSET);
 #endif
@@ -284,7 +291,7 @@ void Dispatcher::EmitDispatcher() {
      mov(ARMEmitter::XReg::x0, STATE);
      mov(ARMEmitter::XReg::x1, ARMEmitter::XReg::lr);

-      ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionLink));
+      ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, Pointers.ExitFunctionLink));
      if (!CTX->Config.DisableVixlIndirectCalls) [[unlikely]] {
        GenerateIndirectRuntimeCall<uintptr_t, void*, void*>(ARMEmitter::Reg::r2);
      } else {
@@ -303,7 +310,7 @@ void Dispatcher::EmitDispatcher() {

  // Need to create the block
  {
-    Bind(&NoBlock);
+    (void)Bind(&NoBlock);

    EmitSignalGuardedRegion([&]() {
      SpillStaticRegs(TMP1);
@@ -337,7 +344,7 @@ void Dispatcher::EmitDispatcher() {
  }

  {
-    Bind(&CompileSingleStep);
+    (void)Bind(&CompileSingleStep);

    EmitSignalGuardedRegion([&]() {
      SpillStaticRegs(TMP1);
@@ -481,7 +488,7 @@ void Dispatcher::EmitDispatcher() {

    // 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(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, CTX->X86CodeGen.CallbackReturn);
+    ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Pointers.ThunkCallbackRet));

    ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]));
    sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r2, CTX->Config.Is64BitMode ? 16 : 12);
@@ -499,7 +506,7 @@ void Dispatcher::EmitDispatcher() {
    stp<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::zr, ARMEmitter::XReg::zr, REG_CALLRET_SP, -0x10);

    // Now go back to the regular dispatcher loop
-    b(&LoopTop);
+    (void)b(&LoopTop);
  }

  auto EmitLongALUOpHandler = [&](auto R, auto Offset) {
@@ -537,8 +544,8 @@ void Dispatcher::EmitDispatcher() {
    return Address;
  };

-  LUDIVHandlerAddress = EmitLongALUOpHandler(STATE_PTR(CpuStateFrame, Pointers.AArch64.LUDIV));
-  LDIVHandlerAddress = EmitLongALUOpHandler(STATE_PTR(CpuStateFrame, Pointers.AArch64.LDIV));
+  LUDIVHandlerAddress = EmitLongALUOpHandler(STATE_PTR(CpuStateFrame, Pointers.LUDIV));
+  LDIVHandlerAddress = EmitLongALUOpHandler(STATE_PTR(CpuStateFrame, Pointers.LDIV));

  // Interpreter fallbacks
  {
@@ -568,14 +575,15 @@ void Dispatcher::EmitDispatcher() {
    }
  }

-  Bind(&l_CTX);
+  (void)Bind(&l_CTX);
  dc64(reinterpret_cast<uintptr_t>(CTX));
-  Bind(&l_Sleep);
+  (void)Bind(&l_Sleep);
  dc64(reinterpret_cast<uint64_t>(SleepThread));
-  Bind(&l_CompileBlock);
+  (void)Bind(&l_CompileBlock);
  FEXCore::Utils::MemberFunctionToPointerCast PMFCompileBlock(&FEXCore::Context::ContextImpl::CompileBlock);
  dc64(PMFCompileBlock.GetConvertedPointer());
-  Bind(&l_CompileSingleStep);
+  (void)Bind(&l_CompileSingleStep);
+
  FEXCore::Utils::MemberFunctionToPointerCast PMFCompileSingleStep(&FEXCore::Context::ContextImpl::CompileSingleStep);
  dc64(PMFCompileSingleStep.GetConvertedPointer());

@@ -757,7 +765,7 @@ uint64_t Dispatcher::GenerateABICall(FallbackABI ABI) {

    ldrh(ARMEmitter::WReg::w0, STATE, offsetof(FEXCore::Core::CPUState, FCW));
    if (!TMP_ABIARGS) {
-      fmov(VABI1.D(), VTMP1.D());
+      mov(VABI1.Q(), VTMP1.Q());
    }

    mov(ARMEmitter::XReg::x1, STATE);
@@ -1078,27 +1086,25 @@ uint64_t Dispatcher::GenerateABICall(FallbackABI ABI) {
 void Dispatcher::InitThreadPointers(FEXCore::Core::InternalThreadState* Thread) {
  // Setup dispatcher specific pointers that need to be accessed from JIT code
  {
-    auto& Common = Thread->CurrentFrame->Pointers.Common;
+    auto& Ptrs = Thread->CurrentFrame->Pointers;

-    Common.DispatcherLoopTop = AbsoluteLoopTopAddress;
-    Common.DispatcherLoopTopFillSRA = AbsoluteLoopTopAddressFillSRA;
-    Common.DispatcherLoopTopEnterEC = AbsoluteLoopTopAddressEnterEC;
-    Common.DispatcherLoopTopEnterECFillSRA = AbsoluteLoopTopAddressEnterECFillSRA;
-    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;
-    Common.SignalReturnHandlerRT = SignalHandlerReturnAddressRT;
-
-    auto& AArch64 = Thread->CurrentFrame->Pointers.AArch64;
-    AArch64.LUDIVHandler = LUDIVHandlerAddress;
-    AArch64.LDIVHandler = LDIVHandlerAddress;
+    Ptrs.DispatcherLoopTop = AbsoluteLoopTopAddress;
+    Ptrs.DispatcherLoopTopFillSRA = AbsoluteLoopTopAddressFillSRA;
+    Ptrs.DispatcherLoopTopEnterEC = AbsoluteLoopTopAddressEnterEC;
+    Ptrs.DispatcherLoopTopEnterECFillSRA = AbsoluteLoopTopAddressEnterECFillSRA;
+    Ptrs.ExitFunctionLinker = ExitFunctionLinkerAddress;
+    Ptrs.ThreadStopHandlerSpillSRA = ThreadStopHandlerAddressSpillSRA;
+    Ptrs.ThreadPauseHandlerSpillSRA = ThreadPauseHandlerAddressSpillSRA;
+    Ptrs.GuestSignal_SIGILL = GuestSignal_SIGILL;
+    Ptrs.GuestSignal_SIGTRAP = GuestSignal_SIGTRAP;
+    Ptrs.GuestSignal_SIGSEGV = GuestSignal_SIGSEGV;
+    Ptrs.SignalReturnHandler = SignalHandlerReturnAddress;
+    Ptrs.SignalReturnHandlerRT = SignalHandlerReturnAddressRT;
+    Ptrs.LUDIVHandler = LUDIVHandlerAddress;
+    Ptrs.LDIVHandler = LDIVHandlerAddress;

    // Fill in the fallback handlers
-    InterpreterOps::FillFallbackIndexPointers(Common.FallbackHandlerPointers, &ABIPointers[0]);
+    InterpreterOps::FillFallbackIndexPointers(Ptrs.FallbackHandlerPointers, &ABIPointers[0]);
  }
 }

@@ -4,6 +4,7 @@
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
 #include "Interface/Core/Interpreter/InterpreterOps.h"

+#include <FEXCore/Config/Config.h>
 #include <FEXCore/fextl/memory.h>

 #include <array>
@@ -50,6 +51,10 @@ public:
  }
 #endif

+  uint64_t GetExitFunctionLinkerAddress() const {
+    return ExitFunctionLinkerAddress;
+  }
+
  SignalDelegatorConfig MakeSignalDelegatorConfig() const;

 protected:
@@ -92,6 +97,8 @@ private:

  void EmitDispatcher();
  uint64_t GenerateABICall(FallbackABI ABI);
+
+  FEX_CONFIG_OPT(DisableL2Cache, DISABLEL2CACHE);
 };

 } // namespace FEXCore::CPU
@@ -9,7 +9,6 @@ $end_info$
 #include "Interface/Context/Context.h"
 #include "Interface/Core/Frontend.h"
 #include "Interface/Core/X86Tables/X86Tables.h"
-#include "Interface/Core/X86HelperGen.h"
 #include "Interface/Core/LookupCache.h"

 #include <array>
@@ -90,11 +89,6 @@ Decoder::Decoder(FEXCore::Core::InternalThreadState* Thread)
 }

 bool Decoder::CheckRangeExecutable(uint64_t Address, uint64_t Size) {
-  // Treat FEX-internal X86 callbacks as always executable
-  if (EntryPoint == CTX->X86CodeGen.CallbackReturn) {
-    return true;
-  }
-
  while (Address < ExecutableRangeBase || Address + Size > ExecutableRangeEnd) {
    auto RangeInfo = CTX->SyscallHandler->QueryGuestExecutableRange(Thread, Address);
    ExecutableRangeBase = RangeInfo.Base;
@@ -138,7 +132,7 @@ std::optional<uint8_t> Decoder::PeekByte(uint8_t Offset) {
  }
 }

-uint64_t Decoder::ReadData(uint8_t Size) {
+std::pair<uint64_t, bool> Decoder::ReadData(uint8_t Size) {
  LOGMAN_THROW_A_FMT(Size != 0 && Size <= sizeof(uint64_t), "Unknown data size to read");

  uint64_t Res = 0;
@@ -160,7 +154,21 @@ uint64_t Decoder::ReadData(uint8_t Size) {
  SkipBytes(Size);
 #endif

-  return Res;
+  if (Relocations) {
+    uint32_t SectionOffset = static_cast<uint32_t>(Address - SectionMinAddress);
+    if (auto It = Relocations->find(SectionOffset); It != Relocations->end()) {
+      if (It->second == GuestRelocationType::Rel32 && Size == 4) {
+        return {static_cast<int64_t>(static_cast<int32_t>(Res) - static_cast<int32_t>(EntryPoint)), true};
+      } else if (It->second == GuestRelocationType::Rel64 && Size == 8) {
+        return {static_cast<int64_t>(Res) - static_cast<int64_t>(EntryPoint), true};
+      } else {
+        HitBadRelocation = true;
+        Res = 0;
+      }
+    }
+  }
+
+  return {Res, false};
 }

 void Decoder::DecodeModRM_16(X86Tables::DecodedOperand* Operand, X86Tables::ModRMDecoded ModRM) {
@@ -192,7 +200,9 @@ void Decoder::DecodeModRM_16(X86Tables::DecodedOperand* Operand, X86Tables::ModR
    DisplacementSize = 1;
  }
  if (DisplacementSize) {
-    Literal = ReadData(DisplacementSize);
+    bool IsRelocation = false;
+    std::tie(Literal, IsRelocation) = ReadData(DisplacementSize);
+    LOGMAN_THROW_A_FMT(!IsRelocation, "1/2 byte relocations unsupported");
    if (DisplacementSize == 1) {
      Literal = static_cast<int8_t>(Literal);
    }
@@ -259,13 +269,13 @@ void Decoder::DecodeModRM_64(X86Tables::DecodedOperand* Operand, X86Tables::ModR

  if (HasSIB) {
    FEXCore::X86Tables::SIBDecoded SIB;
-    if (DecodeInst->DecodedSIB) {
+    if (DecodeInst->Flags & DecodeFlags::FLAG_DECODED_SIB) {
      SIB.Hex = DecodeInst->SIB;
    } else {
      // Haven't yet grabbed SIB, pull it now
      DecodeInst->SIB = ReadByte();
      SIB.Hex = DecodeInst->SIB;
-      DecodeInst->DecodedSIB = true;
+      DecodeInst->Flags |= DecodeFlags::FLAG_DECODED_SIB;
    }

    // If the SIB base is 0b101, aka BP or R13 then we have a 32bit displacement
@@ -298,7 +308,10 @@ void Decoder::DecodeModRM_64(X86Tables::DecodedOperand* Operand, X86Tables::ModR
    LOGMAN_THROW_A_FMT(Displacement <= 4, "Number of bytes should be <= 4 for literal src");

    if (Displacement) {
-      uint64_t Literal = ReadData(Displacement);
+      auto [Literal, IsRelocation] = ReadData(Displacement);
+      if (IsRelocation) {
+        Operand->Type = DecodedOperand::OpType::SIBRelocation;
+      }
      if (Displacement == 1) {
        Literal = static_cast<int8_t>(Literal);
      }
@@ -308,10 +321,9 @@ void Decoder::DecodeModRM_64(X86Tables::DecodedOperand* Operand, X86Tables::ModR
    // Explained in Table 1-14. "Operand Addressing Using ModRM and SIB Bytes"
    if (ModRM.rm == 0b101) {
      // 32bit Displacement
-      const uint32_t Literal = ReadData(4);
-
-      Operand->Type = DecodedOperand::OpType::RIPRelative;
-      Operand->Data.RIPLiteral.Value.u = Literal;
+      auto [Literal, IsRelocation] = ReadData(4);
+      Operand->Type = IsRelocation ? DecodedOperand::OpType::RIPRelativeRelocation : DecodedOperand::OpType::RIPRelative;
+      Operand->Data.RIPLiteral.Value = Literal;
    } else {
      // Register-direct addressing
      Operand->Type = DecodedOperand::OpType::GPRDirect;
@@ -319,12 +331,12 @@ void Decoder::DecodeModRM_64(X86Tables::DecodedOperand* Operand, X86Tables::ModR
    }
  } else {
    uint8_t DisplacementSize = ModRM.mod == 1 ? 1 : 4;
-    uint32_t Literal = ReadData(DisplacementSize);
+    auto [Literal, IsRelocation] = ReadData(DisplacementSize);
    if (DisplacementSize == 1) {
      Literal = static_cast<int8_t>(Literal);
    }

-    Operand->Type = DecodedOperand::OpType::GPRIndirect;
+    Operand->Type = IsRelocation ? DecodedOperand::OpType::GPRIndirectRelocation : DecodedOperand::OpType::GPRIndirect;
    Operand->Data.GPRIndirect.GPR = MapModRMToReg(DecodeInst->Flags & DecodeFlags::FLAG_REX_XGPR_B ? 1 : 0, ModRM.rm, false, false, false, false);
    Operand->Data.GPRIndirect.Displacement = Literal;
  }
@@ -401,9 +413,9 @@ bool Decoder::NormalOp(const FEXCore::X86Tables::X86InstInfo* Info, uint16_t Op,

  // If we require ModRM and haven't decoded it yet, do it now
  // Some instructions have to read modrm upfront, others do it later
-  if (HasMODRM && !DecodeInst->DecodedModRM) {
+  if (HasMODRM && !(DecodeInst->Flags & DecodeFlags::FLAG_DECODED_MODRM)) {
    DecodeInst->ModRM = ReadByte();
-    DecodeInst->DecodedModRM = true;
+    DecodeInst->Flags |= DecodeFlags::FLAG_DECODED_MODRM;
  }

  // New instruction size decoding
@@ -436,9 +448,8 @@ bool Decoder::NormalOp(const FEXCore::X86Tables::X86InstInfo* Info, uint16_t Op,
      // If the default operating mode is 32bit and we have the operand size flag then the operating size drops to 16bit
      DecodeInst->Flags |= DecodeFlags::GenSizeDstSize(DecodeFlags::SIZE_16BIT);
      DestSize = 2;
-    } else if ((HasXMMDst || HasMMDst || BlockInfo.Is64BitMode) &&
-               (HasWideningDisplacement || DstSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_64BIT ||
-                DstSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_64BITDEF)) {
+    } else if ((HasXMMDst || HasMMDst || BlockInfo.Is64BitMode) && (HasWideningDisplacement || DstSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_64BIT ||
+                                                                    DstSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_64BITDEF)) {
      DecodeInst->Flags |= DecodeFlags::GenSizeDstSize(DecodeFlags::SIZE_64BIT);
      DestSize = 8;
    } else {
@@ -465,9 +476,8 @@ bool Decoder::NormalOp(const FEXCore::X86Tables::X86InstInfo* Info, uint16_t Op,
      // See table 1-2. Operand-Size Overrides for this decoding
      // If the default operating mode is 32bit and we have the operand size flag then the operating size drops to 16bit
      DecodeInst->Flags |= DecodeFlags::GenSizeSrcSize(DecodeFlags::SIZE_16BIT);
-    } else if ((HasXMMSrc || HasMMSrc || BlockInfo.Is64BitMode) &&
-               (HasWideningDisplacement || SrcSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_64BIT ||
-                SrcSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_64BITDEF)) {
+    } else if ((HasXMMSrc || HasMMSrc || BlockInfo.Is64BitMode) && (HasWideningDisplacement || SrcSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_64BIT ||
+                                                                    SrcSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_64BITDEF)) {
      DecodeInst->Flags |= DecodeFlags::GenSizeSrcSize(DecodeFlags::SIZE_64BIT);
    } else {
      DecodeInst->Flags |= DecodeFlags::GenSizeSrcSize(DecodeFlags::SIZE_32BIT);
@@ -622,25 +632,32 @@ bool Decoder::NormalOp(const FEXCore::X86Tables::X86InstInfo* Info, uint16_t Op,
  if (Bytes != 0) {
    LOGMAN_THROW_A_FMT(Bytes <= 8, "Number of bytes should be <= 8 for literal src");

-    DecodeInst->Src[CurrentSrc].Data.Literal.Size = Bytes;

-    uint64_t Literal = ReadData(Bytes);
+    auto [Literal, IsRelocation] = ReadData(Bytes);
+    if (IsRelocation) {
+      DecodeInst->Src[CurrentSrc].Type = DecodedOperand::OpType::LiteralRelocation;
+      DecodeInst->Src[CurrentSrc].Data.LiteralRelocation.EntrypointOffset = Literal;
+    } else {
+      DecodeInst->Src[CurrentSrc].Data.Literal.Size = Bytes;

-    if ((Info->Flags & FEXCore::X86Tables::InstFlags::FLAGS_SRC_SEXT) || (DecodeFlags::GetSizeDstFlags(DecodeInst->Flags) == DecodeFlags::SIZE_64BIT &&
-                                                                          Info->Flags & FEXCore::X86Tables::InstFlags::FLAGS_SRC_SEXT64BIT)) {
-      if (Bytes == 1) {
-        Literal = static_cast<int8_t>(Literal);
-      } else if (Bytes == 2) {
-        Literal = static_cast<int16_t>(Literal);
-      } else {
-        Literal = static_cast<int32_t>(Literal);
+      if ((Info->Flags & FEXCore::X86Tables::InstFlags::FLAGS_SRC_SEXT) ||
+          (DecodeFlags::GetSizeDstFlags(DecodeInst->Flags) == DecodeFlags::SIZE_64BIT &&
+           Info->Flags & FEXCore::X86Tables::InstFlags::FLAGS_SRC_SEXT64BIT)) {
+        if (Bytes == 1) {
+          Literal = static_cast<int8_t>(Literal);
+        } else if (Bytes == 2) {
+          Literal = static_cast<int16_t>(Literal);
+        } else {
+          Literal = static_cast<int32_t>(Literal);
+        }
+        DecodeInst->Src[CurrentSrc].Data.Literal.Size = DestSize;
      }
-      DecodeInst->Src[CurrentSrc].Data.Literal.Size = DestSize;
+
+      DecodeInst->Src[CurrentSrc].Type = DecodedOperand::OpType::Literal;
+      DecodeInst->Src[CurrentSrc].Data.Literal.Value = Literal;
    }

    Bytes = 0;
-    DecodeInst->Src[CurrentSrc].Type = DecodedOperand::OpType::Literal;
-    DecodeInst->Src[CurrentSrc].Data.Literal.Value = Literal;
  }

  LOGMAN_THROW_A_FMT(Bytes == 0, "Inst at 0x{:x}: 0x{:04x} '{}' Had an instruction of size {} with {} remaining", DecodeInst->PC,
@@ -672,7 +689,7 @@ bool Decoder::NormalOpHeader(const FEXCore::X86Tables::X86InstInfo* Info, uint16
  } else if (Info->Type >= FEXCore::X86Tables::TYPE_GROUP_1 && Info->Type <= FEXCore::X86Tables::TYPE_GROUP_11) {
    uint8_t ModRMByte = ReadByte();
    DecodeInst->ModRM = ModRMByte;
-    DecodeInst->DecodedModRM = true;
+    DecodeInst->Flags |= DecodeFlags::FLAG_DECODED_MODRM;

    FEXCore::X86Tables::ModRMDecoded ModRM;
    ModRM.Hex = DecodeInst->ModRM;
@@ -689,18 +706,18 @@ bool Decoder::NormalOpHeader(const FEXCore::X86Tables::X86InstInfo* Info, uint16
    constexpr uint16_t PF_F2 = 3;

    uint16_t PrefixType = PF_NONE;
-    if (DecodeInst->LastEscapePrefix == 0xF3) {
+    if (LastEscapePrefix == 0xF3) {
      PrefixType = PF_F3;
-    } else if (DecodeInst->LastEscapePrefix == 0xF2) {
+    } else if (LastEscapePrefix == 0xF2) {
      PrefixType = PF_F2;
-    } else if (DecodeInst->LastEscapePrefix == 0x66) {
+    } else if (LastEscapePrefix == 0x66) {
      PrefixType = PF_66;
    }

    // We have ModRM
    uint8_t ModRMByte = ReadByte();
    DecodeInst->ModRM = ModRMByte;
-    DecodeInst->DecodedModRM = true;
+    DecodeInst->Flags |= DecodeFlags::FLAG_DECODED_MODRM;

    FEXCore::X86Tables::ModRMDecoded ModRM;
    ModRM.Hex = DecodeInst->ModRM;
@@ -727,7 +744,7 @@ bool Decoder::NormalOpHeader(const FEXCore::X86Tables::X86InstInfo* Info, uint16
    // We have ModRM
    uint8_t ModRMByte = ReadByte();
    DecodeInst->ModRM = ModRMByte;
-    DecodeInst->DecodedModRM = true;
+    DecodeInst->Flags |= DecodeFlags::FLAG_DECODED_MODRM;

    uint16_t X87Op = ((Op - 0xD8) << 8) | ModRMByte;
    return NormalOp(&(*X87Table)[X87Op], X87Op);
@@ -789,7 +806,7 @@ bool Decoder::NormalOpHeader(const FEXCore::X86Tables::X86InstInfo* Info, uint16
      // We have ModRM
      uint8_t ModRMByte = ReadByte();
      DecodeInst->ModRM = ModRMByte;
-      DecodeInst->DecodedModRM = true;
+      DecodeInst->Flags |= DecodeFlags::FLAG_DECODED_MODRM;

      FEXCore::X86Tables::ModRMDecoded ModRM;
      ModRM.Hex = DecodeInst->ModRM;
@@ -813,6 +830,7 @@ bool Decoder::NormalOpHeader(const FEXCore::X86Tables::X86InstInfo* Info, uint16

 bool Decoder::DecodeInstructionImpl(uint64_t PC) {
  InstructionSize = 0;
+  LastEscapePrefix = 0;
  Instruction.fill(0);

  DecodeInst = &DecodedBuffer[DecodedSize];
@@ -830,11 +848,12 @@ bool Decoder::DecodeInstructionImpl(uint64_t PC) {
      switch (EscapeOp) {
      case 0x0F:
        [[unlikely]] { // 3DNow!
+          DecodeREXIfValid(-2);
          // 3DNow! Instruction Encoding: 0F 0F [ModRM] [SIB] [Displacement] [Opcode]
          // Decode ModRM
          uint8_t ModRMByte = ReadByte();
          DecodeInst->ModRM = ModRMByte;
-          DecodeInst->DecodedModRM = true;
+          DecodeInst->Flags |= DecodeFlags::FLAG_DECODED_MODRM;

          FEXCore::X86Tables::ModRMDecoded ModRM;
          ModRM.Hex = DecodeInst->ModRM;
@@ -854,6 +873,7 @@ bool Decoder::DecodeInstructionImpl(uint64_t PC) {
          break;
        }
      case 0x38: { // F38 Table!
+        DecodeREXIfValid(-2);
        constexpr uint16_t PF_38_NONE = 0;
        constexpr uint16_t PF_38_66 = (1U << 0);
        constexpr uint16_t PF_38_F2 = (1U << 1);
@@ -873,23 +893,23 @@ bool Decoder::DecodeInstructionImpl(uint64_t PC) {
        uint16_t LocalOp = (Prefix << 8) | ReadByte();

        bool NoOverlay66 = (FEXCore::X86Tables::H0F38TableOps[LocalOp].Flags & InstFlags::FLAGS_NO_OVERLAY66) != 0;
-        if (DecodeInst->LastEscapePrefix == 0x66 && NoOverlay66) { // Operand Size
+        if (LastEscapePrefix == 0x66 && NoOverlay66) { // Operand Size
          // Remove prefix so it doesn't effect calculations.
          // This is only an escape prefix rather than modifier now
          DecodeInst->Flags &= ~DecodeFlags::FLAG_OPERAND_SIZE;
          DecodeFlags::PopOpAddrIf(&DecodeInst->Flags, DecodeFlags::FLAG_OPERAND_SIZE_LAST);
        }
-
        return NormalOpHeader(&FEXCore::X86Tables::H0F38TableOps[LocalOp], LocalOp);
        break;
      }
      case 0x3A: { // F3A Table!
+        DecodeREXIfValid(-2);
        constexpr uint16_t PF_3A_NONE = 0;
        constexpr uint16_t PF_3A_66 = (1 << 0);
        constexpr uint16_t PF_3A_REX = (1 << 1);

        uint16_t Prefix = PF_3A_NONE;
-        if (DecodeInst->LastEscapePrefix == 0x66) { // Operand Size
+        if (LastEscapePrefix == 0x66) { // Operand Size
          Prefix = PF_3A_66;
        }

@@ -913,19 +933,20 @@ bool Decoder::DecodeInstructionImpl(uint64_t PC) {
          bool NoOverlay = (FEXCore::X86Tables::SecondBaseOps[EscapeOp].Flags & InstFlags::FLAGS_NO_OVERLAY) != 0;
          bool NoOverlay66 = (FEXCore::X86Tables::SecondBaseOps[EscapeOp].Flags & InstFlags::FLAGS_NO_OVERLAY66) != 0;

+          DecodeREXIfValid(-2);
          if (NoOverlay) { // This section of the table ignores prefix extention
            return NormalOpHeader(&FEXCore::X86Tables::SecondBaseOps[EscapeOp], EscapeOp);
-          } else if (DecodeInst->LastEscapePrefix == 0xF3) { // REP
+          } else if (LastEscapePrefix == 0xF3) { // REP
            // Remove prefix so it doesn't effect calculations.
            // This is only an escape prefix rather tan modifier now
            DecodeInst->Flags &= ~DecodeFlags::FLAG_REP_PREFIX;
            return NormalOpHeader(&FEXCore::X86Tables::RepModOps[EscapeOp], EscapeOp);
-          } else if (DecodeInst->LastEscapePrefix == 0xF2) { // REPNE
+          } else if (LastEscapePrefix == 0xF2) { // REPNE
            // Remove prefix so it doesn't effect calculations.
            // This is only an escape prefix rather tan modifier now
            DecodeInst->Flags &= ~DecodeFlags::FLAG_REPNE_PREFIX;
            return NormalOpHeader(&FEXCore::X86Tables::RepNEModOps[EscapeOp], EscapeOp);
-          } else if (DecodeInst->LastEscapePrefix == 0x66 && !NoOverlay66) { // Operand Size
+          } else if (LastEscapePrefix == 0x66 && !NoOverlay66) { // Operand Size
            // Remove prefix so it doesn't effect calculations.
            // This is only an escape prefix rather tan modifier now
            DecodeInst->Flags &= ~DecodeFlags::FLAG_OPERAND_SIZE;
@@ -941,7 +962,7 @@ bool Decoder::DecodeInstructionImpl(uint64_t PC) {
    }
    case 0x66: // Operand Size prefix
      DecodeInst->Flags |= DecodeFlags::FLAG_OPERAND_SIZE;
-      DecodeInst->LastEscapePrefix = Op;
+      LastEscapePrefix = Op;
      DecodeFlags::PushOpAddr(&DecodeInst->Flags, DecodeFlags::FLAG_OPERAND_SIZE_LAST);
      break;
    case 0x67: // Address Size override prefix
@@ -972,11 +993,11 @@ bool Decoder::DecodeInstructionImpl(uint64_t PC) {
      break;
    case 0xF2: // REPNE prefix
      DecodeInst->Flags |= DecodeFlags::FLAG_REPNE_PREFIX;
-      DecodeInst->LastEscapePrefix = Op;
+      LastEscapePrefix = Op;
      break;
    case 0xF3: // REP prefix
      DecodeInst->Flags |= DecodeFlags::FLAG_REP_PREFIX;
-      DecodeInst->LastEscapePrefix = Op;
+      LastEscapePrefix = Op;
      break;
    case 0x64: // FS prefix
      DecodeInst->Flags = (DecodeInst->Flags & ~FEXCore::X86Tables::DecodeFlags::FLAG_SEGMENTS) | DecodeFlags::FLAG_FS_PREFIX;
@@ -992,29 +1013,9 @@ bool Decoder::DecodeInstructionImpl(uint64_t PC) {
        }

        if (Info->Type == FEXCore::X86Tables::TYPE_REX_PREFIX) {
-          DecodeInst->Flags |= DecodeFlags::FLAG_REX_PREFIX;
-
-          // Widening displacement
-          if (Op & 0b1000) {
-            DecodeInst->Flags |= DecodeFlags::FLAG_REX_WIDENING;
-            DecodeFlags::PushOpAddr(&DecodeInst->Flags, DecodeFlags::FLAG_WIDENING_SIZE_LAST);
-          }
-
-          // XGPR_B bit set
-          if (Op & 0b0001) {
-            DecodeInst->Flags |= DecodeFlags::FLAG_REX_XGPR_B;
-          }
-
-          // XGPR_X bit set
-          if (Op & 0b0010) {
-            DecodeInst->Flags |= DecodeFlags::FLAG_REX_XGPR_X;
-          }
-
-          // XGPR_R bit set
-          if (Op & 0b0100) {
-            DecodeInst->Flags |= DecodeFlags::FLAG_REX_XGPR_R;
-          }
+          DecodeInst->REXIndex = InstructionSize;
        } else {
+          DecodeREXIfValid();
          return NormalOpHeader(Info, Op);
        }

@@ -1030,17 +1031,53 @@ bool Decoder::DecodeInstructionImpl(uint64_t PC) {
  return true;
 }

+void Decoder::DecodeREXIfValid(int8_t ExpectedOffset) {
+  LOGMAN_THROW_A_FMT(ExpectedOffset < 0, "Expecting an negative offset for the REX offset!");
+  const int8_t REXIndex = InstructionSize + ExpectedOffset;
+
+  if (DecodeInst->REXIndex != 0 && DecodeInst->REXIndex == REXIndex) {
+    const uint8_t Op = Instruction[REXIndex - 1];
+    DecodeInst->Flags |= DecodeFlags::FLAG_REX_PREFIX;
+
+    // Widening displacement
+    if (Op & 0b1000) {
+      DecodeInst->Flags |= DecodeFlags::FLAG_REX_WIDENING;
+      DecodeFlags::PushOpAddr(&DecodeInst->Flags, DecodeFlags::FLAG_WIDENING_SIZE_LAST);
+    }
+
+    // XGPR_B bit set
+    if (Op & 0b0001) {
+      DecodeInst->Flags |= DecodeFlags::FLAG_REX_XGPR_B;
+    }
+
+    // XGPR_X bit set
+    if (Op & 0b0010) {
+      DecodeInst->Flags |= DecodeFlags::FLAG_REX_XGPR_X;
+    }
+
+    // XGPR_R bit set
+    if (Op & 0b0100) {
+      DecodeInst->Flags |= DecodeFlags::FLAG_REX_XGPR_R;
+    }
+  }
+}
+
 Decoder::DecodedBlockStatus Decoder::DecodeInstruction(uint64_t PC) {
  // Will be set if DecodeInstructionImpl tries to read non-executable memory
  HitNonExecutableRange = false;
+  HitBadRelocation = false;
  bool ErrorDuringDecoding = !DecodeInstructionImpl(PC);

-  if (ErrorDuringDecoding || HitNonExecutableRange) [[unlikely]] {
+  if (ErrorDuringDecoding || HitNonExecutableRange || HitBadRelocation) [[unlikely]] {
    // Put an invalid instruction in the stream so the core can raise SIGILL if hit
    // Error while decoding instruction. We don't know the table or instruction size
    DecodeInst->TableInfo = nullptr;
+    auto Result = ErrorDuringDecoding   ? DecodedBlockStatus::INVALID_INST :
+                  DecodeInst->InstSize  ? DecodedBlockStatus::PARTIAL_DECODE_INST :
+                  HitNonExecutableRange ? DecodedBlockStatus::NOEXEC_INST :
+                                          DecodedBlockStatus::BAD_RELOCATION;
    DecodeInst->InstSize = 0;
-    return ErrorDuringDecoding ? DecodedBlockStatus::INVALID_INST : DecodedBlockStatus::NOEXEC_INST;
+    return Result;
  } else if (!DecodeInst->TableInfo || (DecodeInst->TableInfo->Type == TYPE_INST && !DecodeInst->TableInfo->OpcodeDispatcher.OpDispatch)) {
    // If there wasn't an error during decoding but we have no dispatcher for the instruction then claim invalid instruction.
    return DecodedBlockStatus::INVALID_INST;
@@ -1055,10 +1092,10 @@ Decoder::DecodedBlockStatus Decoder::DecodeInstruction(uint64_t PC) {

    if (DecodeInst->OP == 0x8b && DecodeInst->Src[0].IsGPRIndirect() &&
        IsKnownAtomicDisplacement(DecodeInst->Src[0].Data.GPRIndirect.Displacement)) {
-      DecodeInst->ForceTSO = true;
+      DecodeInst->Flags |= X86Tables::DecodeFlags::FLAG_FORCE_TSO;
    }
    if (DecodeInst->OP == 0x89 && DecodeInst->Dest.IsGPRIndirect() && IsKnownAtomicDisplacement(DecodeInst->Dest.Data.GPRIndirect.Displacement)) {
-      DecodeInst->ForceTSO = true;
+      DecodeInst->Flags |= X86Tables::DecodeFlags::FLAG_FORCE_TSO;
    }
  }

@@ -1124,9 +1161,9 @@ void Decoder::BranchTargetInMultiblockRange() {
  // Forbid distant branches to have the cost code better match the guest code layout, avoiding massive (range-wise) code
  // blocks in highly fragmented guest code. Such branches are often not-taken branches to garbage in obfuscated code.
  constexpr uint64_t MAX_FORWARD_BRANCH_DIST = FEXCore::Utils::FEX_PAGE_SIZE * 4;
-  bool ValidMultiblockMember = TargetRIP >= SymbolMinAddress && TargetRIP < std::min(InstEnd + MAX_FORWARD_BRANCH_DIST, SymbolMaxAddress);
+  bool ValidMultiblockMember = TargetRIP >= EntryPoint && TargetRIP < std::min(InstEnd + MAX_FORWARD_BRANCH_DIST, SectionMaxAddress);

-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
  ValidMultiblockMember = ValidMultiblockMember && !RtlIsEcCode(TargetRIP);
 #endif

@@ -1301,7 +1338,7 @@ const uint8_t* Decoder::AdjustAddrForSpecialRegion(const uint8_t* _InstStream, u
  return _InstStream - EntryPoint + RIP;
 }

-void Decoder::DecodeInstructionsAtEntry(FEXCore::Core::InternalThreadState *Thread, const uint8_t* _InstStream, uint64_t PC, uint64_t MaxInst) {
+void Decoder::DecodeInstructionsAtEntry(FEXCore::Core::InternalThreadState* Thread, const uint8_t* _InstStream, uint64_t PC, uint64_t MaxInst) {
  FEXCORE_PROFILE_SCOPED("DecodeInstructions");
  BlockInfo.TotalInstructionCount = 0;
  BlockInfo.Blocks.clear();
@@ -1317,19 +1354,23 @@ void Decoder::DecodeInstructionsAtEntry(FEXCore::Core::InternalThreadState *Thre
  BlockInfo.Is64BitMode = CSSegment->L == 1;
  LOGMAN_THROW_A_FMT(BlockInfo.Is64BitMode == CTX->Config.Is64BitMode, "Expected operating mode to not change at runtime!");

-  // XXX: Load symbol data
-  SymbolAvailable = false;
  EntryPoint = PC;
  BlockInfo.EntryPoints = {PC};
  InstStream = _InstStream;

  uint64_t TotalInstructions {};

-  // If we don't have symbols available then we become a bit optimistic about multiblock ranges
-  if (!SymbolAvailable) {
-    // If we don't have a symbol available then assume all branches are valid for multiblock
-    SymbolMaxAddress = SectionMaxAddress;
-    SymbolMinAddress = EntryPoint;
+  SectionMinAddress = 0;
+  SectionMaxAddress = ~0ULL;
+  Relocations = nullptr;
+
+  if (CTX->GetCodeCache().IsGeneratingCache || EnableCodeCacheValidation) {
+    // If generating cache, attempt to load section bounds and relocations
+    if (auto SectionInfo = CTX->SyscallHandler->LookupExecutableFileSection(Thread, EntryPoint)) {
+      SectionMinAddress = SectionInfo->FileStartVA;
+      SectionMaxAddress = SectionInfo->EndVA;
+      Relocations = &SectionInfo->FileInfo.Relocations;
+    }
  }

  DecodedMinAddress = EntryPoint;
@@ -1442,7 +1483,11 @@ void Decoder::DecodeInstructionsAtEntry(FEXCore::Core::InternalThreadState *Thre
          EraseBlock = true;
        } else {
          LogMan::Msg::EFmt("{} instruction in entry block: {:X}",
-                            BlockIt->BlockStatus == DecodedBlockStatus::INVALID_INST ? "Invalid" : "NoExec", OpAddress);
+                            BlockIt->BlockStatus == DecodedBlockStatus::INVALID_INST   ? "Invalid" :
+                            BlockIt->BlockStatus == DecodedBlockStatus::NOEXEC_INST    ? "NoExec" :
+                            BlockIt->BlockStatus == DecodedBlockStatus::BAD_RELOCATION ? "BadRelocation" :
+                                                                                         "PartialDecode",
+                            OpAddress);
        }
        break;
      }
@@ -4,9 +4,12 @@
 #include "Interface/Core/X86Tables/X86Tables.h"
 #include "Interface/IR/IR.h"

+#include <FEXCore/Config/Config.h>
+#include <FEXCore/Core/CodeCache.h>
 #include <FEXCore/Utils/ThreadPoolAllocator.h>
 #include <FEXCore/fextl/set.h>
 #include <FEXCore/fextl/vector.h>
+#include <FEXCore/fextl/robin_map.h>

 #include <array>
 #include <cstddef>
@@ -27,6 +30,8 @@ public:
    SUCCESS,
    INVALID_INST,
    NOEXEC_INST,
+    PARTIAL_DECODE_INST,
+    BAD_RELOCATION,
  };

  // New Frontend decoding
@@ -49,7 +54,7 @@ public:
  };

  Decoder(FEXCore::Core::InternalThreadState* Thread);
-  void DecodeInstructionsAtEntry(FEXCore::Core::InternalThreadState *Thread, const uint8_t* InstStream, uint64_t PC, uint64_t MaxInst);
+  void DecodeInstructionsAtEntry(FEXCore::Core::InternalThreadState* Thread, const uint8_t* InstStream, uint64_t PC, uint64_t MaxInst);

  const DecodedBlockInformation* GetDecodedBlockInfo() const {
    return &BlockInfo;
@@ -58,9 +63,6 @@ public:
  uint64_t DecodedMinAddress {};
  uint64_t DecodedMaxAddress {~0ULL};

-  void SetSectionMaxAddress(uint64_t v) {
-    SectionMaxAddress = v;
-  }
  void SetExternalBranches(fextl::set<uint64_t>* v) {
    ExternalBranches = v;
  }
@@ -86,6 +88,8 @@ private:
  FEXCore::Context::ContextImpl* CTX;
  const FEXCore::HLE::SyscallOSABI OSABI {};

+  FEX_CONFIG_OPT(EnableCodeCacheValidation, ENABLECODECACHEVALIDATION);
+
  bool DecodeInstructionImpl(uint64_t PC);
  DecodedBlockStatus DecodeInstruction(uint64_t PC);

@@ -99,7 +103,8 @@ private:

  uint8_t ReadByte();
  std::optional<uint8_t> PeekByte(uint8_t Offset);
-  uint64_t ReadData(uint8_t Size);
+  std::pair<uint64_t, bool> ReadData(uint8_t Size);
+
  void SkipBytes(uint8_t Size) {
    InstructionSize += Size;
  }
@@ -107,6 +112,8 @@ private:
  bool NormalOp(const FEXCore::X86Tables::X86InstInfo* Info, uint16_t Op, DecodedHeader Options = {});
  bool NormalOpHeader(const FEXCore::X86Tables::X86InstInfo* Info, uint16_t Op);

+  void DecodeREXIfValid(int8_t ExpectedOffset = -1);
+
  static constexpr size_t DefaultDecodedBufferSize = 0x10000;
  FEXCore::X86Tables::DecodedInst* DecodedBuffer {};
  Utils::PoolBufferWithTimedRetirement<FEXCore::X86Tables::DecodedInst*, 5000, 500> PoolObject;
@@ -116,6 +123,7 @@ private:
  uint64_t ExecutableRangeEnd {};
  bool ExecutableRangeWritable {};
  bool HitNonExecutableRange {};
+  bool HitBadRelocation {};

  const uint8_t* InstStream {};
  IR::OpSize GetGPROpSize() const {
@@ -125,16 +133,15 @@ private:
  static constexpr size_t MAX_INST_SIZE = 15;
  uint8_t InstructionSize {};
  std::array<uint8_t, MAX_INST_SIZE> Instruction;
+  uint8_t LastEscapePrefix {};
  FEXCore::X86Tables::DecodedInst* DecodeInst;

  // This is for multiblock data tracking
-  bool SymbolAvailable {false};
  uint64_t EntryPoint {};
  uint64_t MaxCondBranchForward {};
  uint64_t MaxCondBranchBackwards {~0ULL};
-  uint64_t SymbolMaxAddress {};
-  uint64_t SymbolMinAddress {~0ULL};
  uint64_t SectionMaxAddress {~0ULL};
+  uint64_t SectionMinAddress {};
  uint64_t NextBlockStartAddress {~0ULL};

  DecodedBlockInformation BlockInfo;
@@ -143,6 +150,8 @@ private:
  fextl::set<uint64_t> VisitedBlocks;
  fextl::set<uint64_t>* ExternalBranches {nullptr};

+  const fextl::robin_map<uint32_t, GuestRelocationType>* Relocations {nullptr};
+
  // ModRM rm decoding
  using DecodeModRMPtr = void (FEXCore::Frontend::Decoder::*)(X86Tables::DecodedOperand* Operand, X86Tables::ModRMDecoded ModRM);
  void DecodeModRM_16(X86Tables::DecodedOperand* Operand, X86Tables::ModRMDecoded ModRM);
@@ -87,12 +87,10 @@ void InterpreterOps::FillFallbackIndexPointers(Core::FallbackABIInfo* Info, uint
  Info[Core::OPINDEX_F64SINCOS] = {ABIHandlers[FABI_F64x2_F64_PTR],
                                   reinterpret_cast<uint64_t>(&FEXCore::CPU::OpHandlers<IR::OP_F64SINCOS>::handle)};
  Info[Core::OPINDEX_F64TAN] = {ABIHandlers[FABI_F64_F64_PTR], reinterpret_cast<uint64_t>(&FEXCore::CPU::OpHandlers<IR::OP_F64TAN>::handle)};
-  Info[Core::OPINDEX_F64F2XM1] = {ABIHandlers[FABI_F64_F64_PTR],
-                                  reinterpret_cast<uint64_t>(&FEXCore::CPU::OpHandlers<IR::OP_F64F2XM1>::handle)};
+  Info[Core::OPINDEX_F64F2XM1] = {ABIHandlers[FABI_F64_F64_PTR], reinterpret_cast<uint64_t>(&FEXCore::CPU::OpHandlers<IR::OP_F64F2XM1>::handle)};

  // Double Precision Binary
-  Info[Core::OPINDEX_F64ATAN] = {ABIHandlers[FABI_F64_F64_F64_PTR],
-                                 reinterpret_cast<uint64_t>(&FEXCore::CPU::OpHandlers<IR::OP_F64ATAN>::handle)};
+  Info[Core::OPINDEX_F64ATAN] = {ABIHandlers[FABI_F64_F64_F64_PTR], reinterpret_cast<uint64_t>(&FEXCore::CPU::OpHandlers<IR::OP_F64ATAN>::handle)};
  Info[Core::OPINDEX_F64FPREM] = {ABIHandlers[FABI_F64_F64_F64_PTR],
                                  reinterpret_cast<uint64_t>(&FEXCore::CPU::OpHandlers<IR::OP_F64FPREM>::handle)};
  Info[Core::OPINDEX_F64FPREM1] = {ABIHandlers[FABI_F64_F64_F64_PTR],
@@ -220,21 +218,21 @@ bool InterpreterOps::GetFallbackHandler(const IR::IROp_Header* IROp, FallbackInf
    return true;                                                                                       \
  }

-#define COMMON_UNARY_F64_OP(OP)                            \
-  case IR::OP_F64##OP: {                                   \
+#define COMMON_UNARY_F64_OP(OP)                        \
+  case IR::OP_F64##OP: {                               \
    *Info = {FABI_F64_F64_PTR, Core::OPINDEX_F64##OP}; \
-    return true;                                           \
+    return true;                                       \
  }
-#define COMMON_UNARYPAIR_F64_OP(OP)                          \
-  case IR::OP_F64##OP: {                                     \
+#define COMMON_UNARYPAIR_F64_OP(OP)                      \
+  case IR::OP_F64##OP: {                                 \
    *Info = {FABI_F64x2_F64_PTR, Core::OPINDEX_F64##OP}; \
-    return true;                                             \
+    return true;                                         \
  }

-#define COMMON_BINARY_F64_OP(OP)                               \
-  case IR::OP_F64##OP: {                                       \
+#define COMMON_BINARY_F64_OP(OP)                           \
+  case IR::OP_F64##OP: {                                   \
    *Info = {FABI_F64_F64_F64_PTR, Core::OPINDEX_F64##OP}; \
-    return true;                                               \
+    return true;                                           \
  }

    // Unary
@@ -2,14 +2,14 @@
 #include "Interface/Core/Interpreter/Fallbacks/VectorFallbacks.h"
 #include "Interface/IR/IR.h"

-#ifdef _M_ARM_64
+#ifdef ARCHITECTURE_arm64
 #include <arm_neon.h>
 #endif

 #include <cstring>

 namespace FEXCore::CPU {
-#ifdef _M_ARM_64
+#ifdef ARCHITECTURE_arm64
 FEXCORE_PRESERVE_ALL_ATTR static int32_t GetImplicitLength(FEXCore::VectorRegType data, uint16_t control) {
  const auto is_using_words = (control & 1) != 0;

@@ -43,21 +43,28 @@ DEF_BINOP_WITH_CONSTANT(Ror, rorv, ror)
 DEF_OP(Constant) {
  auto Op = IROp->C<IR::IROp_Constant>();
  auto Dst = GetReg(Node);
-  LoadConstant(ARMEmitter::Size::i64Bit, Dst, Op->Constant);
+
+  const auto PadType = [Pad = Op->Pad]() {
+    switch (Pad) {
+    case IR::ConstPad::NoPad: return CPU::Arm64Emitter::PadType::NOPAD;
+    case IR::ConstPad::DoPad: return CPU::Arm64Emitter::PadType::DOPAD;
+    default: return CPU::Arm64Emitter::PadType::AUTOPAD;
+    }
+  }();
+  LoadConstant(ARMEmitter::Size::i64Bit, Dst, Op->Constant, PadType, Op->MaxBytes);
 }

 DEF_OP(EntrypointOffset) {
  auto Op = IROp->C<IR::IROp_EntrypointOffset>();

  auto Constant = Entry + Op->Offset;
-  auto Dst = GetReg(Node);
  uint64_t Mask = ~0ULL;
  const auto OpSize = IROp->Size;
  if (OpSize == IR::OpSize::i32Bit) {
    Mask = 0xFFFF'FFFFULL;
  }

-  LoadConstant(ARMEmitter::Size::i64Bit, Dst, Constant & Mask);
+  InsertGuestRIPMove(GetReg(Node), Constant & Mask);
 }

 DEF_OP(InlineConstant) {
@@ -372,7 +379,7 @@ DEF_OP(CondSubNZCV) {
 DEF_OP(Neg) {
  auto Op = IROp->C<IR::IROp_Neg>();

-  if (Op->Cond == FEXCore::IR::COND_AL) {
+  if (Op->Cond == IR::CondClass::AL) {
    neg(ConvertSize48(IROp), GetReg(Node), GetReg(Op->Src));
  } else {
    cneg(ConvertSize48(IROp), GetReg(Node), GetReg(Op->Src), MapCC(Op->Cond));
@@ -588,7 +595,7 @@ DEF_OP(ShiftFlags) {
  and_(ARMEmitter::Size::i32Bit, TMP1, Src2, OpSize == IR::OpSize::i64Bit ? 0x3f : 0x1f);

  ARMEmitter::ForwardLabel Done;
-  cbz(EmitSize, TMP1, &Done);
+  (void)cbz(EmitSize, TMP1, &Done);
  {
    // PF/SF/ZF/OF
    if (OpSize >= IR::OpSize::i32Bit) {
@@ -652,7 +659,7 @@ DEF_OP(ShiftFlags) {
      msr(ARMEmitter::SystemRegister::NZCV, TMP2);
    }
  }
-  Bind(&Done);
+  (void)Bind(&Done);

  // TODO: Make RA less dumb so this can't happen (e.g. with late-kill).
  if (PFOutput != PFTemp) {
@@ -669,7 +676,7 @@ DEF_OP(RotateFlags) {

  // If shift=0, flags are unaffected. Wrap the whole implementation in a cbz.
  ARMEmitter::ForwardLabel Done;
-  cbz(EmitSize, Shift, &Done);
+  (void)cbz(EmitSize, Shift, &Done);
  {
    // Extract the last bit shifted in to CF
    const auto BitSize = IR::OpSizeToSize(Op->Size) * 8;
@@ -701,7 +708,7 @@ DEF_OP(RotateFlags) {
      msr(ARMEmitter::SystemRegister::NZCV, TMP3);
    }
  }
-  Bind(&Done);
+  (void)Bind(&Done);
 }

 DEF_OP(Extr) {
@@ -767,14 +774,14 @@ DEF_OP(PDep) {
    // Now, they're copied, so we can start setting Dest (even if it overlaps with
    // one of them).  Handle early exit case
    mov(EmitSize, Dest, 0);
-    cbz(EmitSize, OrigMask, &Done);
+    (void)cbz(EmitSize, OrigMask, &Done);

    // Setup for first iteration
    neg(EmitSize, T0, Mask);
    and_(EmitSize, T0, T0, Mask);

    // Main loop
-    Bind(&NextBit);
+    (void)Bind(&NextBit);
    sbfx(EmitSize, T1, Input, 0, 1);
    eor(EmitSize, Mask, Mask, T0);
    and_(EmitSize, T0, T1, T0);
@@ -782,10 +789,10 @@ DEF_OP(PDep) {
    orr(EmitSize, Dest, Dest, T0);
    lsr(EmitSize, Input, Input, 1);
    and_(EmitSize, T0, Mask, T1);
-    cbnz(EmitSize, T0, &NextBit);
+    (void)cbnz(EmitSize, T0, &NextBit);

    // All done with nothing to do.
-    Bind(&Done);
+    (void)Bind(&Done);
  }
 }

@@ -821,27 +828,27 @@ DEF_OP(PExt) {
    ARMEmitter::BackwardLabel NextBit;
    ARMEmitter::ForwardLabel Done;

-    cbz(EmitSize, Mask, &EarlyExit);
+    (void)cbz(EmitSize, Mask, &EarlyExit);
    mov(EmitSize, MaskReg, Mask);
    mov(EmitSize, ValueReg, Input);
    mov(EmitSize, Dest, ARMEmitter::Reg::zr);

    // Main loop
-    Bind(&NextBit);
-    cbz(EmitSize, MaskReg, &Done);
+    (void)Bind(&NextBit);
+    (void)cbz(EmitSize, MaskReg, &Done);
    clz(EmitSize, BitReg, MaskReg);
    lslv(EmitSize, ValueReg, ValueReg, BitReg);
    lslv(EmitSize, MaskReg, MaskReg, BitReg);
    extr(EmitSize, Dest, Dest, ValueReg, OpSizeBitsM1);
    bfc(EmitSize, MaskReg, OpSizeBitsM1, 1);
-    b(&NextBit);
+    (void)b(&NextBit);

    // Early exit
-    Bind(&EarlyExit);
+    (void)Bind(&EarlyExit);
    mov(EmitSize, Dest, ARMEmitter::Reg::zr);

    // All done with nothing to do.
-    Bind(&Done);
+    (void)Bind(&Done);
  }
 }

@@ -909,7 +916,7 @@ DEF_OP(Div) {
    eor(EmitSize, TMP1, TMP1, Upper);

    // If the sign bit matches then the result is zero
-    cbz(EmitSize, TMP1, &Only64Bit);
+    (void)cbz(EmitSize, TMP1, &Only64Bit);

    // Long divide
    {
@@ -917,7 +924,7 @@ DEF_OP(Div) {
      mov(EmitSize, TMP2, Lower);
      mov(EmitSize, TMP3, Divisor);

-      ldr(TMP4, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.LDIVHandler));
+      ldr(TMP4, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.LDIVHandler));

      str<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, -16);
      blr(TMP4);
@@ -928,17 +935,17 @@ DEF_OP(Div) {
      mov(EmitSize, Remainder, TMP2);

      // Skip 64-bit path
-      b(&LongDIVRet);
+      (void)b(&LongDIVRet);
    }

-    Bind(&Only64Bit);
+    (void)Bind(&Only64Bit);
    // 64-Bit only
    {
      sdiv(EmitSize, Quotient, Lower, Divisor);
      msub(EmitSize, Remainder, Quotient, Divisor, Lower);
    }

-    Bind(&LongDIVRet);
+    (void)Bind(&LongDIVRet);
    break;
  }
  default: LOGMAN_MSG_A_FMT("Unknown DIV Size: {}", OpSize); break;
@@ -992,7 +999,7 @@ DEF_OP(UDiv) {

    // Check the upper bits for zero
    // If the upper bits are zero then we can do a 64-bit divide
-    cbz(EmitSize, Upper, &Only64Bit);
+    (void)cbz(EmitSize, Upper, &Only64Bit);

    // Long divide
    {
@@ -1000,7 +1007,7 @@ DEF_OP(UDiv) {
      mov(EmitSize, TMP2, Lower);
      mov(EmitSize, TMP3, Divisor);

-      ldr(TMP4, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.LUDIVHandler));
+      ldr(TMP4, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.LUDIVHandler));

      str<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, -16);
      blr(TMP4);
@@ -1011,17 +1018,17 @@ DEF_OP(UDiv) {
      mov(EmitSize, Remainder, TMP2);

      // Skip 64-bit path
-      b(&LongDIVRet);
+      (void)b(&LongDIVRet);
    }

-    Bind(&Only64Bit);
+    (void)Bind(&Only64Bit);
    // 64-Bit only
    {
      udiv(EmitSize, Quotient, Lower, Divisor);
      msub(EmitSize, Remainder, Quotient, Divisor, Lower);
    }

-    Bind(&LongDIVRet);
+    (void)Bind(&LongDIVRet);
    break;
  }
  default: LOGMAN_MSG_A_FMT("Unknown LUDIV Size: {}", OpSize); break;
@@ -1046,24 +1053,19 @@ DEF_OP(Popcount) {

  if (CTX->HostFeatures.SupportsCSSC) {
    switch (OpSize) {
-      case IR::OpSize::i8Bit:
-        uxtb(ARMEmitter::Size::i32Bit, Dst, Src);
-        cnt(ARMEmitter::Size::i32Bit, Dst, Dst);
-        break;
-      case IR::OpSize::i16Bit:
-        uxth(ARMEmitter::Size::i32Bit, Dst, Src);
-        cnt(ARMEmitter::Size::i32Bit, Dst, Dst);
-        break;
-      case IR::OpSize::i32Bit:
-        cnt(ARMEmitter::Size::i32Bit, Dst, Src);
-        break;
-      case IR::OpSize::i64Bit:
-        cnt(ARMEmitter::Size::i64Bit, Dst, Src);
-        break;
-      default: LOGMAN_MSG_A_FMT("Unsupported Popcount size: {}", OpSize);
+    case IR::OpSize::i8Bit:
+      uxtb(ARMEmitter::Size::i32Bit, Dst, Src);
+      cnt(ARMEmitter::Size::i32Bit, Dst, Dst);
+      break;
+    case IR::OpSize::i16Bit:
+      uxth(ARMEmitter::Size::i32Bit, Dst, Src);
+      cnt(ARMEmitter::Size::i32Bit, Dst, Dst);
+      break;
+    case IR::OpSize::i32Bit: cnt(ARMEmitter::Size::i32Bit, Dst, Src); break;
+    case IR::OpSize::i64Bit: cnt(ARMEmitter::Size::i64Bit, Dst, Src); break;
+    default: LOGMAN_MSG_A_FMT("Unsupported Popcount size: {}", OpSize);
    }
-  }
-  else {
+  } else {
    switch (OpSize) {
    case IR::OpSize::i8Bit:
      fmov(ARMEmitter::Size::i32Bit, VTMP1.S(), Src);
@@ -1195,6 +1197,19 @@ DEF_OP(Rev) {
  }
 }

+DEF_OP(Rbit) {
+  auto Op = IROp->C<IR::IROp_Rbit>();
+  const auto OpSize = IROp->Size;
+
+  LOGMAN_THROW_A_FMT(OpSize == IR::OpSize::i32Bit || OpSize == IR::OpSize::i64Bit, "Unsupported {} size: {}", __func__, OpSize);
+  const auto EmitSize = ConvertSize48(IROp);
+
+  const auto Dst = GetReg(Node);
+  const auto Src = GetReg(Op->Src);
+
+  rbit(EmitSize, Dst, Src);
+}
+
 DEF_OP(Bfi) {
  auto Op = IROp->C<IR::IROp_Bfi>();
  const auto EmitSize = ConvertSize(IROp);
@@ -1278,6 +1293,16 @@ DEF_OP(Sbfe) {
  sbfx(ConvertSize(IROp), Dst, Src, Op->lsb, Op->Width);
 }

+DEF_OP(MaskGenerateFromBitWidth) {
+  auto Op = IROp->C<IR::IROp_MaskGenerateFromBitWidth>();
+  auto BitWidth = GetReg(Op->BitWidth);
+
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP1, -1);
+  cmp(ARMEmitter::Size::i64Bit, BitWidth, 0);
+  lslv(ARMEmitter::Size::i64Bit, TMP2, TMP1, BitWidth);
+  csinv(ARMEmitter::Size::i64Bit, GetReg(Node), TMP1, TMP2, ARMEmitter::Condition::CC_EQ);
+}
+
 DEF_OP(Select) {
  auto Op = IROp->C<IR::IROp_Select>();
  const auto OpSize = IROp->Size;
@@ -11,36 +11,32 @@ $end_info$
 #include <FEXCore/Core/Thunks.h>

 namespace FEXCore::CPU {
-
-uint64_t Arm64JITCore::GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
+uint64_t GetNamedSymbolLiteral(FEXCore::Context::ContextImpl& CTX, 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 CTX.Dispatcher->GetExitFunctionLinkerAddress();
+
+  default: ERROR_AND_DIE_FMT("Unknown named symbol literal: {}", static_cast<uint32_t>(Op));
  }
-  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 - CodeData.BlockBegin;
+  MoveABI.NamedThunkMove.Header = {.Offset = GetCursorOffset(), .Type = FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE};
  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, false);
+  // Pointers are required to fit within 48-bit VA space.
+  LoadConstant(ARMEmitter::Size::i64Bit, Reg, Pointer, FEXCore::CPU::Arm64Emitter::PadType::AUTOPAD, 6);
  Relocations.emplace_back(MoveABI);
 }

 Arm64JITCore::NamedSymbolLiteralPair Arm64JITCore::InsertNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
-  uint64_t Pointer = GetNamedSymbolLiteral(Op);
+  uint64_t Pointer = GetNamedSymbolLiteral(*CTX, Op);

-  Arm64JITCore::NamedSymbolLiteralPair Lit {
+  NamedSymbolLiteralPair Lit {
    .Lit = Pointer,
    .MoveABI =
      {
@@ -48,87 +44,77 @@ Arm64JITCore::NamedSymbolLiteralPair Arm64JITCore::InsertNamedSymbolLiteral(FEXC
          {
            .Header =
              {
+                .Offset = 0, // Set by PlaceNamedSymbolLiteral
                .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 - CodeData.BlockBegin;
+void Arm64JITCore::PlaceNamedSymbolLiteral(NamedSymbolLiteralPair Lit) {
+  switch (Lit.MoveABI.Header.Type) {
+  case RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL:
+  case RelocationTypes::RELOC_GUEST_RIP_LITERAL: {
+    Lit.MoveABI.Header.Offset = GetCursorOffset();
+    break;
+  }

-  Bind(&Lit.Loc);
+  default: ERROR_AND_DIE_FMT("Unknown relocation type for {}", __FUNCTION__);
+  }
+
+  BindOrRestart(&Lit.Loc);
  dc64(Lit.Lit);
  Relocations.emplace_back(Lit.MoveABI);
 }

+auto Arm64JITCore::InsertGuestRIPLiteral(uint64_t GuestRIP) -> NamedSymbolLiteralPair {
+  return {
+    .Lit = GuestRIP,
+    .MoveABI =
+      {
+        .GuestRIP = {.Header =
+                       {
+                         .Offset = 0, // Set by PlaceNamedSymbolLiteral
+                         .Type = FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_LITERAL,
+                       },
+                     // NOTE: Cache serialization will subtract the guest binary base address later to produce consistency results
+                     .GuestRIP = GuestRIP},
+      },
+  };
+}
+
 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 - CodeData.BlockBegin;
-  MoveABI.GuestRIPMove.GuestRIP = Constant;
-  MoveABI.GuestRIPMove.RegisterIndex = Reg.Idx();
+  MoveABI.GuestRIP.Header = {.Offset = GetCursorOffset(), .Type = FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE};
+  // NOTE: Cache serialization will subtract the guest binary base address later to produce consistency results
+  MoveABI.GuestRIP.GuestRIP = Constant;
+  MoveABI.GuestRIP.RegisterIndex = Reg.Idx();

-  LoadConstant(ARMEmitter::Size::i64Bit, Reg, Constant, false);
+  // Pointers are required to fit within 48-bit VA space.
+  // TODO: Force 6-byte `MaxSize`, with sign extension to 64-bit. Current code not smart enough to handle negatives.
+  // 48-bit sign extension works because x86-64 guests only receive 47-bit VA space, with 48-bit being reserved for kernel.
+  // Additional quirk, "canonical" 48-bit pointers on x86-64, sign extend the 48-bit as well (Which is why kernel pointers are negative).
+  LoadConstant(ARMEmitter::Size::i64Bit, Reg, Constant, FEXCore::CPU::Arm64Emitter::PadType::AUTOPAD);
  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_A_FMT((DataIndex % alignof(Relocation)) == 0, "Alignment of relocation wasn't adhered to");
-
-    switch (Reloc->Header.Type) {
-    case FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL: {
-      uint64_t Pointer = GetNamedSymbolLiteral(Reloc->NamedSymbolLiteral.Symbol);
-      // Relocation occurs at the cursorEntry + offset relative to that cursor
-      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);
+fextl::vector<FEXCore::CPU::Relocation> Arm64JITCore::TakeRelocations(uint64_t GuestBaseAddress) {
+  // Rebase relocations to library base address
+  for (auto& Relocation : Relocations) {
+    switch (Relocation.Header.Type) {
+    case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE:
+    case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_LITERAL: {
+      Relocation.GuestRIP.GuestRIP -= GuestBaseAddress;
      break;
    }
+    default:;
    }
  }

-  return true;
+  return std::move(Relocations);
 }
+
 } // namespace FEXCore::CPU
@@ -62,27 +62,27 @@ DEF_OP(CASPair) {
    ARMEmitter::BackwardLabel LoopTop;
    ARMEmitter::ForwardLabel LoopNotExpected;
    ARMEmitter::ForwardLabel LoopExpected;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);

    // This instruction sequence must be synced with HandleCASPAL_Armv8.
    ldaxp(EmitSize, TMP2, TMP3, MemSrc);
    cmp(EmitSize, TMP2, Expected0);
    ccmp(EmitSize, TMP3, Expected1, ARMEmitter::StatusFlags::None, ARMEmitter::Condition::CC_EQ);
-    b(ARMEmitter::Condition::CC_NE, &LoopNotExpected);
+    (void)b(ARMEmitter::Condition::CC_NE, &LoopNotExpected);
    stlxp(EmitSize, TMP2, Desired0, Desired1, MemSrc);
-    cbnz(EmitSize, TMP2, &LoopTop);
+    (void)cbnz(EmitSize, TMP2, &LoopTop);
    mov(EmitSize, Dst0, Expected0);
    mov(EmitSize, Dst1, Expected1);

-    b(&LoopExpected);
+    (void)b(&LoopExpected);

-    Bind(&LoopNotExpected);
+    (void)Bind(&LoopNotExpected);
    mov(EmitSize, Dst0, TMP2.R());
    mov(EmitSize, Dst1, TMP3.R());
    // exclusive monitor needs to be cleared here
    // Might have hit the case where ldaxr was hit but stlxr wasn't
    clrex();
-    Bind(&LoopExpected);
+    (void)Bind(&LoopExpected);

    // Restore
    msr(ARMEmitter::SystemRegister::NZCV, TMP1);
@@ -114,7 +114,7 @@ DEF_OP(CAS) {
    ARMEmitter::BackwardLabel LoopTop;
    ARMEmitter::ForwardLabel LoopNotExpected;
    ARMEmitter::ForwardLabel LoopExpected;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
    ldaxr(SubEmitSize, TMP2, MemSrc);
    if (IROp->Size == IR::OpSize::i8Bit) {
      cmp(EmitSize, TMP2, Expected, ARMEmitter::ExtendedType::UXTB, 0);
@@ -123,38 +123,18 @@ DEF_OP(CAS) {
    } else {
      cmp(EmitSize, TMP2, Expected);
    }
-    b(ARMEmitter::Condition::CC_NE, &LoopNotExpected);
+    (void)b(ARMEmitter::Condition::CC_NE, &LoopNotExpected);
    stlxr(SubEmitSize, TMP3, Desired, MemSrc);
-    cbnz(EmitSize, TMP3, &LoopTop);
+    (void)cbnz(EmitSize, TMP3, &LoopTop);
    mov(EmitSize, Dst, Expected);
-    b(&LoopExpected);
+    (void)b(&LoopExpected);

-    Bind(&LoopNotExpected);
+    (void)Bind(&LoopNotExpected);
    mov(EmitSize, Dst, TMP2.R());
    // exclusive monitor needs to be cleared here
    // Might have hit the case where ldaxr was hit but stlxr wasn't
    clrex();
-    Bind(&LoopExpected);
-  }
-}
-
-DEF_OP(AtomicXor) {
-  auto Op = IROp->C<IR::IROp_AtomicXor>();
-  const auto EmitSize = ConvertSize(IROp);
-  const auto SubEmitSize = ConvertSubRegSize8(IROp->Size);
-
-  auto MemSrc = GetReg(Op->Addr);
-  auto Src = GetReg(Op->Value);
-
-  if (CTX->HostFeatures.SupportsAtomics) {
-    steorl(SubEmitSize, Src, MemSrc);
-  } else {
-    ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
-    ldaxr(SubEmitSize, TMP2, MemSrc);
-    eor(EmitSize, TMP2, TMP2, Src);
-    stlxr(SubEmitSize, TMP2, TMP2, MemSrc);
-    cbnz(EmitSize, TMP2, &LoopTop);
+    (void)Bind(&LoopExpected);
  }
 }

@@ -179,10 +159,10 @@ DEF_OP(AtomicSwap) {
    ldswpal(SubEmitSize, Src, GetReg(Node), MemSrc);
  } else {
    ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
    ldaxr(SubEmitSize, TMP2, MemSrc);
    stlxr(SubEmitSize, TMP4, Src, MemSrc);
-    cbnz(EmitSize, TMP4, &LoopTop);
+    (void)cbnz(EmitSize, TMP4, &LoopTop);
    ubfm(EmitSize, GetReg(Node), TMP2, 0, IR::OpSizeAsBits(OpSize) - 1);
  }
 }
@@ -199,11 +179,11 @@ DEF_OP(AtomicFetchAdd) {
    ldaddal(SubEmitSize, Src, GetReg(Node), MemSrc);
  } else {
    ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
    ldaxr(SubEmitSize, TMP2, MemSrc);
    add(EmitSize, TMP3, TMP2, Src);
    stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    cbnz(EmitSize, TMP4, &LoopTop);
+    (void)cbnz(EmitSize, TMP4, &LoopTop);
    mov(EmitSize, GetReg(Node), TMP2.R());
  }
 }
@@ -221,11 +201,11 @@ DEF_OP(AtomicFetchSub) {
    ldaddal(SubEmitSize, TMP2, GetReg(Node), MemSrc);
  } else {
    ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
    ldaxr(SubEmitSize, TMP2, MemSrc);
    sub(EmitSize, TMP3, TMP2, Src);
    stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    cbnz(EmitSize, TMP4, &LoopTop);
+    (void)cbnz(EmitSize, TMP4, &LoopTop);
    mov(EmitSize, GetReg(Node), TMP2.R());
  }
 }
@@ -243,11 +223,11 @@ DEF_OP(AtomicFetchAnd) {
    ldclral(SubEmitSize, TMP2, GetReg(Node), MemSrc);
  } else {
    ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
    ldaxr(SubEmitSize, TMP2, MemSrc);
    and_(EmitSize, TMP3, TMP2, Src);
    stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    cbnz(EmitSize, TMP4, &LoopTop);
+    (void)cbnz(EmitSize, TMP4, &LoopTop);
    mov(EmitSize, GetReg(Node), TMP2.R());
  }
 }
@@ -264,11 +244,11 @@ DEF_OP(AtomicFetchCLR) {
    ldclral(SubEmitSize, Src, GetReg(Node), MemSrc);
  } else {
    ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
    ldaxr(SubEmitSize, TMP2, MemSrc);
    bic(EmitSize, TMP3, TMP2, Src);
    stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    cbnz(EmitSize, TMP4, &LoopTop);
+    (void)cbnz(EmitSize, TMP4, &LoopTop);
    mov(EmitSize, GetReg(Node), TMP2.R());
  }
 }
@@ -285,11 +265,11 @@ DEF_OP(AtomicFetchOr) {
    ldsetal(SubEmitSize, Src, GetReg(Node), MemSrc);
  } else {
    ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
    ldaxr(SubEmitSize, TMP2, MemSrc);
    orr(EmitSize, TMP3, TMP2, Src);
    stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    cbnz(EmitSize, TMP4, &LoopTop);
+    (void)cbnz(EmitSize, TMP4, &LoopTop);
    mov(EmitSize, GetReg(Node), TMP2.R());
  }
 }
@@ -306,11 +286,11 @@ DEF_OP(AtomicFetchXor) {
    ldeoral(SubEmitSize, Src, GetReg(Node), MemSrc);
  } else {
    ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
    ldaxr(SubEmitSize, TMP2, MemSrc);
    eor(EmitSize, TMP3, TMP2, Src);
    stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    cbnz(EmitSize, TMP4, &LoopTop);
+    (void)cbnz(EmitSize, TMP4, &LoopTop);
    mov(EmitSize, GetReg(Node), TMP2.R());
  }
 }
@@ -326,20 +306,20 @@ DEF_OP(AtomicFetchNeg) {
    // Use a CAS loop to avoid needing to emulate unaligned LLSC atomics
    ldr(SubEmitSize, TMP2, MemSrc);
    ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
    mov(EmitSize, TMP4, TMP2);
    neg(EmitSize, TMP3, TMP2);
    casal(SubEmitSize, TMP2, TMP3, MemSrc);
    sub(EmitSize, TMP3, TMP2, TMP4);
-    cbnz(EmitSize, TMP3, &LoopTop);
+    (void)cbnz(EmitSize, TMP3, &LoopTop);
    mov(EmitSize, GetReg(Node), TMP2.R());
  } else {
    ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
    ldaxr(SubEmitSize, TMP2, MemSrc);
    neg(EmitSize, TMP3, TMP2);
    stlxr(SubEmitSize, TMP4, TMP3, MemSrc);
-    cbnz(EmitSize, TMP4, &LoopTop);
+    (void)cbnz(EmitSize, TMP4, &LoopTop);
    mov(EmitSize, GetReg(Node), TMP2.R());
  }
 }
@@ -349,7 +329,7 @@ DEF_OP(TelemetrySetValue) {
  auto Op = IROp->C<IR::IROp_TelemetrySetValue>();
  auto Src = GetReg(Op->Value);

-  ldr(TMP2, STATE_PTR(CpuStateFrame, Pointers.Common.TelemetryValueAddresses[Op->TelemetryValueIndex]));
+  ldr(TMP2, STATE_PTR(CpuStateFrame, Pointers.TelemetryValueAddresses[Op->TelemetryValueIndex]));

  // Cortex fuses cmp+cset.
  cmp(ARMEmitter::Size::i32Bit, Src, 0);
@@ -359,11 +339,11 @@ DEF_OP(TelemetrySetValue) {
    stsetl(ARMEmitter::SubRegSize::i64Bit, TMP1, TMP2);
  } else {
    ARMEmitter::BackwardLabel LoopTop;
-    Bind(&LoopTop);
+    (void)Bind(&LoopTop);
    ldaxr(ARMEmitter::SubRegSize::i64Bit, TMP3, TMP2);
    orr(ARMEmitter::Size::i32Bit, TMP3, TMP3, Src);
    stlxr(ARMEmitter::SubRegSize::i64Bit, TMP3, TMP3, TMP2);
-    cbnz(ARMEmitter::Size::i32Bit, TMP3, &LoopTop);
+    (void)cbnz(ARMEmitter::Size::i32Bit, TMP3, &LoopTop);
  }
 #endif
 }
@@ -56,12 +56,12 @@ DEF_OP(ExitFunction) {
  uint64_t NewRIP;

  if (IsInlineConstant(Op->NewRIP, &NewRIP) || IsInlineEntrypointOffset(Op->NewRIP, &NewRIP)) {
-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
    if (NewRIP < EC_CODE_BITMAP_MAX_ADDRESS && RtlIsEcCode(NewRIP)) {
      str(REG_CALLRET_SP, STATE_PTR(CpuStateFrame, State.callret_sp));
      add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, StaticRegisters[X86State::REG_RSP], 0);
-      LoadConstant(ARMEmitter::Size::i64Bit, EC_CALL_CHECKER_PC_REG, NewRIP);
-      ldr(TMP2, STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionEC));
+      InsertGuestRIPMove(EC_CALL_CHECKER_PC_REG, NewRIP);
+      ldr(TMP2, STATE_PTR(CpuStateFrame, Pointers.ExitFunctionEC));
      br(TMP2);
    } else {
 #endif
@@ -141,7 +141,7 @@ DEF_OP(ExitFunction) {
        if (!Op->CallReturnBlock.IsInvalid()) {
          auto CallReturnAddressReg = GetReg(Op->CallReturnAddress).X();
          PendingCallReturnTargetLabel = &CallReturnTargets.try_emplace(Op->CallReturnBlock.ID()).first->second;
-          adr(TMP1, &l_CallReturn);
+          (void)adr(TMP1, &l_CallReturn);
          stp<ARMEmitter::IndexType::PRE>(CallReturnAddressReg, TMP1, REG_CALLRET_SP, -0x10);
        } else {
          stp<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::zr, ARMEmitter::XReg::zr, REG_CALLRET_SP, -0x10);
@@ -149,17 +149,17 @@ DEF_OP(ExitFunction) {
      } else if (Op->Hint == IR::BranchHint::CheckTF) {
        ARMEmitter::ForwardLabel TFUnset;
        ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
-        cbz(ARMEmitter::Size::i32Bit, TMP1, &TFUnset);
-        LoadConstant(ARMEmitter::Size::i64Bit, TMP1, NewRIP);
+        (void)cbz(ARMEmitter::Size::i32Bit, TMP1, &TFUnset);
+        InsertGuestRIPMove(TMP1, NewRIP);
        str(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip));
-        ldr(TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.DispatcherLoopTop));
+        ldr(TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.DispatcherLoopTop));
        blr(TMP2);
-        Bind(&TFUnset);
+        (void)Bind(&TFUnset);
      }

      EmitLinkedBranch(NewRIP, Op->Hint == IR::BranchHint::Call);
-      Bind(&l_CallReturn);
-#ifdef _M_ARM_64EC
+      (void)Bind(&l_CallReturn);
+#ifdef ARCHITECTURE_arm64ec
    }
 #endif
  } else {
@@ -170,40 +170,38 @@ DEF_OP(ExitFunction) {
      // First try to pop from the call-ret stack, otherwise follow the normal path (but ending in a ret)
      ldp<ARMEmitter::IndexType::POST>(TMP1, TMP2, REG_CALLRET_SP, 0x10);
      sub(TMP1, TMP1, RipReg.X());
-      cbz(ARMEmitter::Size::i64Bit, TMP1, &SkipFullLookup);
+      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &SkipFullLookup);
    }

    // L1 Cache
-    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.L1Pointer));
+    ldp<ARMEmitter::IndexType::OFFSET>(TMP1, TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, State.L1Pointer));

    // Calculate (tmp1 + ((ripreg & L1_ENTRIES_MASK) << 4)) for the address
-    // arithmetic. ubfiz+add is marginally faster on Firestorm than
-    // and+add(shift). Same performance on Cortex.
-    static_assert(LookupCache::L1_ENTRIES_MASK == ((1u << 20) - 1));
-    ubfiz(ARMEmitter::Size::i64Bit, TMP4, RipReg, 4, 20);
-    add(TMP1, TMP1, TMP4);
+    // L1Mask is pre-shifted.
+    and_(ARMEmitter::Size::i64Bit, TMP2, TMP2, RipReg, ARMEmitter::ShiftType::LSL, FEXCore::ilog2(sizeof(LookupCache::LookupCacheEntry)));
+    add(TMP1, TMP1, TMP2);

    ldp<ARMEmitter::IndexType::OFFSET>(TMP2, TMP1, TMP1, 0);

    // Note: sub+cbnz used over cmp+br to preserve flags.
    sub(TMP1, TMP1, RipReg.X());
-    cbz(ARMEmitter::Size::i64Bit, TMP1, &SkipFullLookup);
-    ldr(TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.DispatcherLoopTop));
+    (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &SkipFullLookup);
+    ldr(TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.DispatcherLoopTop));
    str(RipReg.X(), STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip));

-    Bind(&SkipFullLookup);
+    (void)Bind(&SkipFullLookup);
    if (Op->Hint == IR::BranchHint::Call) {
      ARMEmitter::ForwardLabel l_CallReturn;
      if (!Op->CallReturnBlock.IsInvalid()) {
        auto CallReturnAddressReg = GetReg(Op->CallReturnAddress).X();
        PendingCallReturnTargetLabel = &CallReturnTargets.try_emplace(Op->CallReturnBlock.ID()).first->second;
-        adr(TMP1, &l_CallReturn);
+        (void)adr(TMP1, &l_CallReturn);
        stp<ARMEmitter::IndexType::PRE>(CallReturnAddressReg, TMP1, REG_CALLRET_SP, -0x10);
      } else {
        stp<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::zr, ARMEmitter::XReg::zr, REG_CALLRET_SP, -0x10);
      }
      blr(TMP2);
-      Bind(&l_CallReturn);
+      (void)Bind(&l_CallReturn);
    } else if (Op->Hint == IR::BranchHint::Return) {
      ret(TMP2);
    } else {
@@ -224,7 +222,7 @@ DEF_OP(CondJump) {
  auto TrueTargetLabel = JumpTarget(Op->TrueBlock);

  if (Op->FromNZCV) {
-    b(MapCC(Op->Cond), TrueTargetLabel);
+    b_OrRestart(MapCC(Op->Cond), TrueTargetLabel);
  } else {
    uint64_t Const;
    const bool isConst = IsInlineConstant(Op->Cmp2, &Const);
@@ -235,18 +233,18 @@ DEF_OP(CondJump) {
    LOGMAN_THROW_A_FMT(IsGPR(Op->Cmp1), "CondJump: Expected GPR");
    LOGMAN_THROW_A_FMT(isConst, "CondJump: Expected constant source");

-    if (Op->Cond.Val == FEXCore::IR::COND_EQ) {
+    if (Op->Cond == IR::CondClass::EQ) {
      LOGMAN_THROW_A_FMT(Const == 0, "CondJump: Expected 0 source");
-      cbz(Size, Reg, TrueTargetLabel);
-    } else if (Op->Cond.Val == FEXCore::IR::COND_NEQ) {
+      cbz_OrRestart(Size, Reg, TrueTargetLabel);
+    } else if (Op->Cond == IR::CondClass::NEQ) {
      LOGMAN_THROW_A_FMT(Const == 0, "CondJump: Expected 0 source");
-      cbnz(Size, Reg, TrueTargetLabel);
-    } else if (Op->Cond.Val == FEXCore::IR::COND_TSTZ) {
+      cbnz_OrRestart(Size, Reg, TrueTargetLabel);
+    } else if (Op->Cond == IR::CondClass::TSTZ) {
      LOGMAN_THROW_A_FMT(Const < 64, "CondJump: Expected valid bit source");
-      tbz(Reg, Const, TrueTargetLabel);
-    } else if (Op->Cond.Val == FEXCore::IR::COND_TSTNZ) {
+      tbz_OrRestart(Reg, Const, TrueTargetLabel);
+    } else if (Op->Cond == IR::CondClass::TSTNZ) {
      LOGMAN_THROW_A_FMT(Const < 64, "CondJump: Expected valid bit source");
-      tbnz(Reg, Const, TrueTargetLabel);
+      tbnz_OrRestart(Reg, Const, TrueTargetLabel);
    } else {
      LOGMAN_THROW_A_FMT(false, "CondJump expected simple condition");
    }
@@ -262,16 +260,10 @@ DEF_OP(Syscall) {
  // X1: ThreadState
  // X2: Pointer to SyscallArguments

-  FEXCore::IR::SyscallFlags Flags = Op->Flags;
  PushDynamicRegs(TMP1);

  uint32_t GPRSpillMask = ~0U;
  uint32_t FPRSpillMask = ~0U;
-  if ((Flags & FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY) == FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY) {
-    // Need to spill all caller saved registers still
-    GPRSpillMask = CALLER_GPR_MASK;
-    FPRSpillMask = CALLER_FPR_MASK;
-  }

  SpillStaticRegs(TMP1, true, GPRSpillMask, FPRSpillMask);

@@ -291,8 +283,8 @@ DEF_OP(Syscall) {
    str(GetReg(Op->Header.Args[i]).X(), ARMEmitter::Reg::rsp, i * 8);
  }

-  ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SyscallHandlerObj));
-  ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SyscallHandlerFunc));
+  ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.SyscallHandlerObj));
+  ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.SyscallHandlerFunc));
  mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, STATE.R());

  // SP supporting move
@@ -305,117 +297,22 @@ DEF_OP(Syscall) {

  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, SPOffset);

-  if ((Flags & FEXCore::IR::SyscallFlags::NORETURN) != FEXCore::IR::SyscallFlags::NORETURN) {
-    // Result is now in x0
-    // Fix the stack and any values that were stepped on
-    FillStaticRegs(true, GPRSpillMask, FPRSpillMask, ARMEmitter::Reg::r1, ARMEmitter::Reg::r2);
+  // Result is now in x0
+  // Fix the stack and any values that were stepped on
+  FillStaticRegs(true, GPRSpillMask, FPRSpillMask, ARMEmitter::Reg::r1, ARMEmitter::Reg::r2);

-    // 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));
+  // 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));

-    PopDynamicRegs();
+  PopDynamicRegs();

-    if ((Flags & FEXCore::IR::SyscallFlags::NORETURNEDRESULT) != FEXCore::IR::SyscallFlags::NORETURNEDRESULT) {
-      // Move result to its destination register.
-      // Only if `NORETURNEDRESULT` wasn't set, otherwise we might overwrite the CPUState refilled with `FillStaticRegs`
-      mov(ARMEmitter::Size::i64Bit, GetReg(Node), ARMEmitter::Reg::r0);
-    }
-  }
-}
+  const auto OSABI = CTX->SyscallHandler->GetOSABI();

-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]);
-    if (Reg == ARMEmitter::Reg::r8 || Reg == ARMEmitter::Reg::r4 || Reg == ARMEmitter::Reg::r5) {
-
-      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(TMP1, 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]);
-      if (SpillMask & (1U << Reg.Idx())) {
-        // In the case of intersection with x4, x5, or x8 then these are currently SRA
-        // for registers RAX, RDX, and RSP. Which have just been spilled
-        // Just load back from the context.
-        auto Correlation = GetX86RegRelationToARMReg(Reg);
-        LOGMAN_THROW_A_FMT(Correlation != X86State::REG_INVALID, "Invalid register mapping");
-        ldr(EmitSubSize, RegArgs[i].R(), STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[Correlation]));
-      } 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]));
-    }
-  }
-
-  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, ~0U, ARMEmitter::Reg::r8, ARMEmitter::Reg::r1);
-
-    // 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);
+  if (OSABI != FEXCore::HLE::SyscallOSABI::OS_GENERIC) {
+    // Move result to its destination register.
+    // Only if `NORETURNEDRESULT` wasn't set, otherwise we might overwrite the CPUState refilled with `FillStaticRegs`
+    mov(ARMEmitter::Size::i64Bit, GetReg(Node), ARMEmitter::Reg::r0);
  }
 }

@@ -431,8 +328,7 @@ DEF_OP(Thunk) {

  mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, GetReg(Op->ArgPtr));

-  auto thunkFn = static_cast<Context::ContextImpl*>(ThreadState->CTX)->ThunkHandler->LookupThunk(Op->ThunkNameHash);
-  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, (uintptr_t)thunkFn);
+  InsertNamedThunkRelocation(ARMEmitter::Reg::r2, Op->ThunkNameHash);
  if (!CTX->Config.DisableVixlIndirectCalls) [[unlikely]] {
    GenerateIndirectRuntimeCall<void, void*, void*>(ARMEmitter::Reg::r2);
  } else {
@@ -458,7 +354,7 @@ DEF_OP(ValidateCode) {
    while (len >= Size) {
      LoadData();
      sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, TMP2);
-      cbnz(ARMEmitter::Size::i64Bit, TMP1, &Fail);
+      cbnz_OrRestart(ARMEmitter::Size::i64Bit, TMP1, &Fail);
      len -= Size;
      Offset += Size;
    }
@@ -486,10 +382,10 @@ DEF_OP(ValidateCode) {

  ARMEmitter::ForwardLabel End;
  LoadConstant(ARMEmitter::Size::i32Bit, Dst, 0);
-  b(&End);
-  Bind(&Fail);
+  b_OrRestart(&End);
+  BindOrRestart(&Fail);
  LoadConstant(ARMEmitter::Size::i32Bit, Dst, 1);
-  Bind(&End);
+  BindOrRestart(&End);
 }

 DEF_OP(ThreadRemoveCodeEntry) {
@@ -501,9 +397,10 @@ DEF_OP(ThreadRemoveCodeEntry) {
  // X1: RIP
  mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, STATE.R());

-  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, Entry);
+  // TODO: Relocations don't seem to be wired up to this...?
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, Entry, CPU::Arm64Emitter::PadType::AUTOPAD);

-  ldr(ARMEmitter::XReg::x2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.ThreadRemoveCodeEntryFromJIT));
+  ldr(ARMEmitter::XReg::x2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.ThreadRemoveCodeEntryFromJIT));
  if (!CTX->Config.DisableVixlIndirectCalls) [[unlikely]] {
    GenerateIndirectRuntimeCall<void, void*, void*>(ARMEmitter::Reg::r2);
  } else {
@@ -528,8 +425,8 @@ DEF_OP(CPUID) {
  // x0 = CPUID Handler
  // x1 = CPUID Function
  // x2 = CPUID Leaf
-  ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDObj));
-  ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDFunction));
+  ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.CPUIDObj));
+  ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.CPUIDFunction));

  if (!TMP_ABIARGS) {
    mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, TMP2);
@@ -569,8 +466,8 @@ DEF_OP(XGetBV) {

  // x0 = CPUID Handler
  // x1 = XCR Function
-  ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDObj));
-  ldr(ARMEmitter::XReg::x2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.XCRFunction));
+  ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.CPUIDObj));
+  ldr(ARMEmitter::XReg::x2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.XCRFunction));
  if (!CTX->Config.DisableVixlIndirectCalls) [[unlikely]] {
    GenerateIndirectRuntimeCall<uint64_t, void*, uint32_t>(ARMEmitter::Reg::r2);
  } else {
@@ -423,11 +423,11 @@ DEF_OP(Vector_FToI) {
    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 FEXCore::IR::Round_Negative_Infinity.Val: frintm(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
-    case FEXCore::IR::Round_Positive_Infinity.Val: frintp(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
-    case FEXCore::IR::Round_Towards_Zero.Val: frintz(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
-    case FEXCore::IR::Round_Host.Val: frinti(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::Nearest: frintn(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::NegInfinity: frintm(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::PosInfinity: frintp(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::TowardsZero: frintz(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::Host: frinti(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
    }
  } else {
    const auto IsScalar = ElementSize == OpSize;
@@ -449,21 +449,21 @@ DEF_OP(Vector_FToI) {
  }

      switch (Op->Round) {
-      case IR::Round_Nearest.Val: ROUNDING_FN(frintn); break;
-      case IR::Round_Negative_Infinity.Val: ROUNDING_FN(frintm); break;
-      case IR::Round_Positive_Infinity.Val: ROUNDING_FN(frintp); break;
-      case IR::Round_Towards_Zero.Val: ROUNDING_FN(frintz); break;
-      case IR::Round_Host.Val: ROUNDING_FN(frinti); break;
+      case IR::RoundMode::Nearest: ROUNDING_FN(frintn); break;
+      case IR::RoundMode::NegInfinity: ROUNDING_FN(frintm); break;
+      case IR::RoundMode::PosInfinity: ROUNDING_FN(frintp); break;
+      case IR::RoundMode::TowardsZero: ROUNDING_FN(frintz); break;
+      case IR::RoundMode::Host: ROUNDING_FN(frinti); break;
      }

 #undef ROUNDING_FN
    } else {
      switch (Op->Round) {
-      case FEXCore::IR::Round_Nearest.Val: frintn(SubEmitSize, Dst.Q(), Vector.Q()); break;
-      case FEXCore::IR::Round_Negative_Infinity.Val: frintm(SubEmitSize, Dst.Q(), Vector.Q()); break;
-      case FEXCore::IR::Round_Positive_Infinity.Val: frintp(SubEmitSize, Dst.Q(), Vector.Q()); break;
-      case FEXCore::IR::Round_Towards_Zero.Val: frintz(SubEmitSize, Dst.Q(), Vector.Q()); break;
-      case FEXCore::IR::Round_Host.Val: frinti(SubEmitSize, Dst.Q(), Vector.Q()); break;
+      case IR::RoundMode::Nearest: frintn(SubEmitSize, Dst.Q(), Vector.Q()); break;
+      case IR::RoundMode::NegInfinity: frintm(SubEmitSize, Dst.Q(), Vector.Q()); break;
+      case IR::RoundMode::PosInfinity: frintp(SubEmitSize, Dst.Q(), Vector.Q()); break;
+      case IR::RoundMode::TowardsZero: frintz(SubEmitSize, Dst.Q(), Vector.Q()); break;
+      case IR::RoundMode::Host: frinti(SubEmitSize, Dst.Q(), Vector.Q()); break;
      }
    }
  }
@@ -539,11 +539,11 @@ DEF_OP(Vector_F64ToI32) {
    // Then convert to integers using fcvtzs.
    auto CVTReg = Dst.Z();
    switch (Round) {
-    case IR::Round_Nearest.Val: frintn(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Vector.Z()); break;
-    case IR::Round_Negative_Infinity.Val: frintm(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Vector.Z()); break;
-    case IR::Round_Positive_Infinity.Val: frintp(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Vector.Z()); break;
-    case IR::Round_Towards_Zero.Val: CVTReg = Vector.Z(); break;
-    case IR::Round_Host.Val: frinti(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::Nearest: frintn(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::NegInfinity: frintm(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::PosInfinity: frintp(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::TowardsZero: CVTReg = Vector.Z(); break;
+    case IR::RoundMode::Host: frinti(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Vector.Z()); break;
    }

    fcvtzs(Dst.Z(), ARMEmitter::SubRegSize::i32Bit, Mask, CVTReg, ARMEmitter::SubRegSize::i64Bit);
@@ -567,11 +567,11 @@ DEF_OP(Vector_F64ToI32) {

    ///< Round float to integral depending on rounding mode.
    switch (Round) {
-    case FEXCore::IR::Round_Nearest.Val: frintn(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
-    case FEXCore::IR::Round_Negative_Infinity.Val: frintm(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
-    case FEXCore::IR::Round_Positive_Infinity.Val: frintp(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
-    case FEXCore::IR::Round_Towards_Zero.Val: frintz(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
-    case FEXCore::IR::Round_Host.Val: frinti(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
+    case IR::RoundMode::Nearest: frintn(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
+    case IR::RoundMode::NegInfinity: frintm(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
+    case IR::RoundMode::PosInfinity: frintp(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
+    case IR::RoundMode::TowardsZero: frintz(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
+    case IR::RoundMode::Host: frinti(ARMEmitter::SubRegSize::i64Bit, Dst.Q(), Vector.Q()); break;
    }

    // Now narrow from f64 to f32.
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MIT
+#pragma once
+
+#include <FEXCore/Utils/AllocatorHooks.h>
+#include <FEXCore/fextl/vector.h>
+
+#include <cstdint>
+
+namespace FEXCore::CPU {
+union Relocation;
+} // namespace FEXCore::CPU
+
+namespace FEXCore::Core {
+struct DebugDataSubblock {
+  uint32_t HostCodeOffset;
+  uint32_t HostCodeSize;
+};
+
+struct DebugDataGuestOpcode {
+  uint64_t GuestEntryOffset;
+  ptrdiff_t HostEntryOffset;
+};
+
+/**
+ * @brief Contains debug data for a block of code for later debugger analysis
+ *
+ * Needs to remain around for as long as the code could be executed at least
+ */
+struct DebugData : public FEXCore::Allocator::FEXAllocOperators {
+  uint64_t HostCodeSize; ///< The size of the code generated in the host JIT
+  fextl::vector<DebugDataSubblock> Subblocks;
+  fextl::vector<DebugDataGuestOpcode> GuestOpcodes;
+  fextl::vector<FEXCore::CPU::Relocation>* Relocations;
+};
+} // namespace FEXCore::Core
@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: MIT
 /*
 $info$
-glossary: Splatter ~ a code generator backend that concaternates configurable macros instead of doing isel
+glossary: Splatter ~ a code generator backend that concatenates configurable macros instead of doing isel
 glossary: IR ~ Intermediate Representation, our high-level opcode representation, loosely modeling arm64
 glossary: SSA ~ Single Static Assignment, a form of representing IR in memory
 glossary: Basic Block ~ A block of instructions with no control flow, terminated by control flow
@@ -11,15 +11,12 @@ desc: Main glue logic of the arm64 splatter backend
 $end_info$
 */

-#include "Common/SoftFloat.h"
-#include "FEXCore/Utils/Telemetry.h"
-#include "FEXCore/Utils/TypeDefines.h"
 #include "Interface/Context/Context.h"
 #include "Interface/Core/LookupCache.h"
-
 #include "Interface/Core/Dispatcher/Dispatcher.h"
+#include "Interface/Core/Interpreter/InterpreterOps.h"
+#include "Interface/Core/JIT/DebugData.h"
 #include "Interface/Core/JIT/JITClass.h"
-
 #include "Interface/IR/Passes/RegisterAllocationPass.h"

 #include "Utils/MemberFunctionToPointer.h"
@@ -30,15 +27,16 @@ $end_info$
 #include <FEXCore/Utils/Allocator.h>
 #include <FEXCore/Utils/CompilerDefs.h>
 #include <FEXCore/Utils/EnumUtils.h>
+#include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/LongJump.h>
 #include <FEXCore/Utils/Profiler.h>
+#include <FEXCore/Utils/Telemetry.h>
+#include <FEXCore/Utils/TypeDefines.h>
 #include <FEXCore/HLE/SyscallHandler.h>

-#include "Interface/Core/Interpreter/InterpreterOps.h"
-
-#include <stdio.h>
+#include <cstdio>
+#include <cstring>
 #include <unistd.h>
-#include <string.h>
-#include <limits>

 namespace {
 struct DivRem {
@@ -135,8 +133,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
      const auto Src1 = GetVReg(IROp->Args[0]);
      fmov(VTMP1.S(), Src1.S());

-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));
      blr(TMP1);

      ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
@@ -153,8 +151,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
      const auto Src1 = GetVReg(IROp->Args[0]);
      fmov(VTMP1.D(), Src1.D());

-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));
      blr(TMP1);

      ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
@@ -178,8 +176,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
        mov(ARMEmitter::Size::i32Bit, TMP2, Src1);
      }

-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));
      blr(TMP1);

      ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
@@ -196,8 +194,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
      const auto Src1 = GetVReg(IROp->Args[0]);
      mov(VTMP1.Q(), Src1.Q());

-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));
      blr(TMP1);

      ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
@@ -214,8 +212,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
      const auto Src1 = GetVReg(IROp->Args[0]);
      mov(VTMP1.Q(), Src1.Q());

-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));
      blr(TMP1);

      ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
@@ -232,8 +230,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
      const auto Src1 = GetVReg(IROp->Args[0]);
      fmov(VTMP1.D(), Src1.D());

-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));
      blr(TMP1);

      ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
@@ -256,8 +254,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {

      fmov(VTMP1.D(), Src1.D());

-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));
      blr(TMP1);

      ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
@@ -278,8 +276,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
      fmov(VTMP1.D(), Src1.D());
      fmov(VTMP2.D(), Src2.D());

-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));
      blr(TMP1);

      ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
@@ -296,8 +294,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
      const auto Src1 = GetVReg(IROp->Args[0]);
      mov(VTMP1.Q(), Src1.Q());

-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));
      blr(TMP1);

      ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
@@ -314,8 +312,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
      const auto Src1 = GetVReg(IROp->Args[0]);
      mov(VTMP1.Q(), Src1.Q());

-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));
      blr(TMP1);

      ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
@@ -332,8 +330,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
      const auto Src1 = GetVReg(IROp->Args[0]);
      mov(VTMP1.Q(), Src1.Q());

-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));
      blr(TMP1);

      ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
@@ -353,8 +351,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
      mov(VTMP1.Q(), Src1.Q());
      mov(VTMP2.Q(), Src2.Q());

-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));
      blr(TMP1);

      ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
@@ -371,8 +369,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
      const auto Src1 = GetVReg(IROp->Args[0]);
      mov(VTMP1.Q(), Src1.Q());

-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));
      blr(TMP1);

      ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
@@ -396,8 +394,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {

      mov(VTMP1.Q(), Src1.Q());

-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));
      blr(TMP1);

      ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
@@ -418,8 +416,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
      mov(VTMP1.Q(), Src1.Q());
      mov(VTMP2.Q(), Src2.Q());

-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));
      blr(TMP1);

      ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
@@ -436,8 +434,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
      // tmp2 (x1/x11): source 2
      // tmp3 (x2/x12): source 3
      const auto Op = IROp->C<IR::IROp_VPCMPESTRX>();
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));

      stp<ARMEmitter::IndexType::PRE>(TMP1, ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, -16);

@@ -478,8 +476,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
      mov(VTMP2.Q(), Src2.Q());
      movz(ARMEmitter::Size::i32Bit, TMP1, Control);

-      ldr(TMP2, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
-      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex].Func));
+      ldr(TMP2, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].ABIHandler));
+      ldr(TMP4, STATE_PTR(CpuStateFrame, Pointers.FallbackHandlerPointers[Info.HandlerIndex].Func));
      blr(TMP2);

      ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
@@ -495,7 +493,7 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
  }
 }

-static void DirectBlockDelinker(FEXCore::Core::CpuStateFrame* Frame, FEXCore::Context::ExitFunctionLinkData* Record, bool Call) {
+static void DirectBlockDelinker(FEXCore::Context::ExitFunctionLinkData* Record, bool Call) {
  uintptr_t JumpThunkStartAddress = reinterpret_cast<uintptr_t>(Record) - 0x10;
  uintptr_t CallerAddress = JumpThunkStartAddress + Record->CallerOffset;
  auto BranchOffset = JumpThunkStartAddress / 4 - CallerAddress / 4;
@@ -513,11 +511,12 @@ static void DirectBlockDelinker(FEXCore::Core::CpuStateFrame* Frame, FEXCore::Co
  ARMEmitter::Emitter::ClearICache(reinterpret_cast<void*>(CallerAddress), 4);
 }

-static void IndirectBlockDelinker(FEXCore::Core::CpuStateFrame* Frame, FEXCore::Context::ExitFunctionLinkData* Record) {
+static void IndirectBlockDelinker(FEXCore::Context::ExitFunctionLinkData* Record) {
  uintptr_t JumpThunkStartAddress = reinterpret_cast<uintptr_t>(Record) - 0x10;
  uint32_t BranchInst = 0;
  ARMEmitter::Emitter BranchEmit(reinterpret_cast<uint8_t*>(&BranchInst), 4);
-  BranchEmit.b(0x8);
+  // Restore branch +2 instructions to jump to the linker block
+  BranchEmit.b(0x2);

  std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(JumpThunkStartAddress)).store(BranchInst, std::memory_order::relaxed);
  ARMEmitter::Emitter::ClearICache(reinterpret_cast<void*>(JumpThunkStartAddress), 4);
@@ -534,12 +533,13 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX
  if (TFSet) {
    // If TF is set, the cache must be skipped as different code needs to be generated.
    Frame->State.rip = GuestRip;
-    return Frame->Pointers.Common.DispatcherLoopTop;
+    return Frame->Pointers.DispatcherLoopTop;
  } else {
    {
      // Guard the LookupCache lock with the code invalidation mutex, to avoid issues with forking
-      auto lk_inval = GuardSignalDeferringSection<std::shared_lock>(static_cast<Context::ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);
-      HostCode = Thread->LookupCache->FindBlock(GuestRip);
+      auto lk_inval =
+        GuardSignalDeferringSection<std::shared_lock>(static_cast<Context::ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);
+      HostCode = Thread->LookupCache->FindBlock(Thread, GuestRip);
    }
    if (!HostCode) {
      // Hold a reference to the code buffer, to avoid linking unmapped code if compilation triggers a recreation.
@@ -564,7 +564,7 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX
  auto lk_inval = GuardSignalDeferringSection<std::shared_lock>(static_cast<Context::ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);

  // Lock here is necessary to prevent simultaneous linking and delinking
-  auto lk = Thread->LookupCache->AcquireLock();
+  auto lk = Thread->LookupCache->AcquireWriteLock();

  // For non-calls, this would extend into the block's code, however that's fine as an out-of-range adr would never
  // be generated avoiding any false positives.
@@ -577,14 +577,12 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX

    if (KnownCallMarkerInst == ExpectedKnownCallMarkerInst) {
      BranchEmit.bl(BranchOffset);
-      Thread->LookupCache->AddBlockLink(GuestRip, Record, [](FEXCore::Core::CpuStateFrame* Frame, FEXCore::Context::ExitFunctionLinkData* Record) {
-        DirectBlockDelinker(Frame, Record, true);
-      });
+      Thread->LookupCache->AddBlockLink(
+        GuestRip, Record, [](FEXCore::Context::ExitFunctionLinkData* Record) { DirectBlockDelinker(Record, true); }, lk);
    } else {
      BranchEmit.b(BranchOffset);
-      Thread->LookupCache->AddBlockLink(GuestRip, Record, [](FEXCore::Core::CpuStateFrame* Frame, FEXCore::Context::ExitFunctionLinkData* Record) {
-        DirectBlockDelinker(Frame, Record, false);
-      });
+      Thread->LookupCache->AddBlockLink(
+        GuestRip, Record, [](FEXCore::Context::ExitFunctionLinkData* Record) { DirectBlockDelinker(Record, false); }, lk);
    }

    std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(CallerAddress)).store(BranchInst, std::memory_order::relaxed);
@@ -592,7 +590,7 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX
  } else {
    // This case is common between calls and jumps as the thunk callsite can be left untouched.
    std::atomic_ref<uint64_t>(Record->HostCode).store(HostCode, std::memory_order::seq_cst);
-#ifdef _M_ARM_64
+#ifdef ARCHITECTURE_arm64
    // Make memory write visible to other threads reading the same location
    asm volatile("dc cvau, %0; dsb ish" : : "r"(Record->HostCode) :);
 #endif
@@ -603,7 +601,7 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX
    std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(JumpThunkStartAddress)).store(LdrInst, std::memory_order::relaxed);
    ARMEmitter::Emitter::ClearICache(reinterpret_cast<void*>(JumpThunkStartAddress), 4);

-    Thread->LookupCache->AddBlockLink(GuestRip, Record, IndirectBlockDelinker);
+    Thread->LookupCache->AddBlockLink(GuestRip, Record, IndirectBlockDelinker, lk);
  }

  return HostCode;
@@ -624,59 +622,46 @@ Arm64JITCore::Arm64JITCore(FEXCore::Context::ContextImpl* ctx, FEXCore::Core::In

  RAPass = Thread->PassManager->GetPass<IR::RegisterAllocationPass>("RA");

-  RAPass->AddRegisters(FEXCore::IR::GPRClass, GeneralRegisters.size());
-  RAPass->AddRegisters(FEXCore::IR::GPRFixedClass, StaticRegisters.size());
-  RAPass->AddRegisters(FEXCore::IR::FPRClass, GeneralFPRegisters.size());
-  RAPass->AddRegisters(FEXCore::IR::FPRFixedClass, StaticFPRegisters.size());
+  RAPass->AddRegisters(IR::RegClass::GPR, GeneralRegisters.size());
+  RAPass->AddRegisters(IR::RegClass::GPRFixed, StaticRegisters.size());
+  RAPass->AddRegisters(IR::RegClass::FPR, GeneralFPRegisters.size());
+  RAPass->AddRegisters(IR::RegClass::FPRFixed, StaticFPRegisters.size());
  RAPass->PairRegs = PairRegisters;

  {
    // Set up pointers that the JIT needs to load

    // Common
-    auto& Common = ThreadState->CurrentFrame->Pointers.Common;
+    auto& Ptrs = ThreadState->CurrentFrame->Pointers;

-    Common.PrintValue = reinterpret_cast<uint64_t>(PrintValue);
-    Common.PrintVectorValue = reinterpret_cast<uint64_t>(PrintVectorValue);
-    Common.ThreadRemoveCodeEntryFromJIT = reinterpret_cast<uintptr_t>(&Context::ContextImpl::ThreadRemoveCodeEntryFromJit);
-    Common.MonoBackpatcherWrite = reinterpret_cast<uint64_t>(&Context::ContextImpl::MonoBackpatcherWrite);
-    Common.CPUIDObj = reinterpret_cast<uint64_t>(&CTX->CPUID);
+    Ptrs.PrintValue = reinterpret_cast<uint64_t>(PrintValue);
+    Ptrs.PrintVectorValue = reinterpret_cast<uint64_t>(PrintVectorValue);
+    Ptrs.ThreadRemoveCodeEntryFromJIT = reinterpret_cast<uintptr_t>(&Context::ContextImpl::ThreadRemoveCodeEntryFromJit);
+    Ptrs.MonoBackpatcherWrite = reinterpret_cast<uint64_t>(&Context::ContextImpl::MonoBackpatcherWrite);
+    Ptrs.CPUIDObj = reinterpret_cast<uint64_t>(&CTX->CPUID);

    {
      FEXCore::Utils::MemberFunctionToPointerCast PMF(&FEXCore::CPUIDEmu::RunFunction);
-      Common.CPUIDFunction = PMF.GetConvertedPointer();
+      Ptrs.CPUIDFunction = PMF.GetConvertedPointer();
    }

    {
      FEXCore::Utils::MemberFunctionToPointerCast PMF(&FEXCore::CPUIDEmu::RunXCRFunction);
-      Common.XCRFunction = PMF.GetConvertedPointer();
+      Ptrs.XCRFunction = PMF.GetConvertedPointer();
    }

    {
      FEXCore::Utils::MemberFunctionToPointerCast PMF(&FEXCore::HLE::SyscallHandler::HandleSyscall);
-      Common.SyscallHandlerObj = reinterpret_cast<uint64_t>(CTX->SyscallHandler);
-      Common.SyscallHandlerFunc = PMF.GetVTableEntry(CTX->SyscallHandler);
+      Ptrs.SyscallHandlerObj = reinterpret_cast<uint64_t>(CTX->SyscallHandler);
+      Ptrs.SyscallHandlerFunc = PMF.GetVTableEntry(CTX->SyscallHandler);
    }
-    Common.ExitFunctionLink = reinterpret_cast<uintptr_t>(&Arm64JITCore::ExitFunctionLink);
-
-    // Platform Specific
-    auto& AArch64 = ThreadState->CurrentFrame->Pointers.AArch64;
-
-    AArch64.LUDIV = reinterpret_cast<uint64_t>(LUDIV);
-    AArch64.LDIV = reinterpret_cast<uint64_t>(LDIV);
+    Ptrs.ExitFunctionLink = reinterpret_cast<uintptr_t>(&Arm64JITCore::ExitFunctionLink);
+    Ptrs.LUDIV = reinterpret_cast<uint64_t>(LUDIV);
+    Ptrs.LDIV = reinterpret_cast<uint64_t>(LDIV);
  }

  CurrentCodeBuffer = CodeBuffers.GetLatest();
  ThreadState->LookupCache->Shared = CurrentCodeBuffer->LookupCache.get();
-
-  // Setup dynamic dispatch.
-  if (ParanoidTSO()) {
-    RT_LoadMemTSO = &Arm64JITCore::Op_ParanoidLoadMemTSO;
-    RT_StoreMemTSO = &Arm64JITCore::Op_ParanoidStoreMemTSO;
-  } else {
-    RT_LoadMemTSO = &Arm64JITCore::Op_LoadMemTSO;
-    RT_StoreMemTSO = &Arm64JITCore::Op_StoreMemTSO;
-  }
 }

 void Arm64JITCore::EmitDetectionString() {
@@ -688,13 +673,13 @@ void Arm64JITCore::EmitDetectionString() {
 void Arm64JITCore::ClearCache() {
  // NOTE: Holding on to the reference here is required to ensure validity of the WriteLock mutex
  auto PrevCodeBuffer = CurrentCodeBuffer;
-  std::lock_guard lk(PrevCodeBuffer->LookupCache->WriteLock);
+  auto lk = PrevCodeBuffer->LookupCache->AcquireWriteLock();

  auto CodeBuffer = GetEmptyCodeBuffer();
  SetBuffer(CodeBuffer->Ptr, CodeBuffer->Size);
  EmitDetectionString();

-  ThreadState->LookupCache->ChangeGuestToHostMapping(*PrevCodeBuffer, *CurrentCodeBuffer->LookupCache);
+  ThreadState->LookupCache->ChangeGuestToHostMapping(*PrevCodeBuffer, *CurrentCodeBuffer->LookupCache, lk);
 }

 Arm64JITCore::~Arm64JITCore() {}
@@ -740,48 +725,48 @@ bool Arm64JITCore::IsInlineEntrypointOffset(const IR::OrderedNodeWrapper& WNode,
  }
 }

-void Arm64JITCore::EmitInterruptChecks(bool CheckTF) {
-  if (CheckTF) {
-    ARMEmitter::ForwardLabel l_TFUnset;
-    ARMEmitter::ForwardLabel l_TFBlocked;
+void Arm64JITCore::EmitTFCheck() {
+  ARMEmitter::ForwardLabel l_TFUnset;
+  ARMEmitter::ForwardLabel l_TFBlocked;

-    // Note that this needs to be before the below suspend checks, as X86 checks this flag immediately after executing an instruction.
-    ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
+  // Note that this needs to be before the below suspend checks, as X86 checks this flag immediately after executing an instruction.
+  ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));

-    cbz(ARMEmitter::Size::i32Bit, TMP1, &l_TFUnset);
+  (void)cbz(ARMEmitter::Size::i32Bit, TMP1, &l_TFUnset);

-    // X86 semantically checks TF after executing each instruction, so e.g. setting a context with TF set will execute a single instruction
-    // and then raise an exception. However on the FEX side this is simpler to implement by checking at the start of each instruction, handle this by having bit 1 being unset in the flag state indicate that TF is blocked for a single instruction.
-    tbz(TMP1, 1, &l_TFBlocked);
+  // X86 semantically checks TF after executing each instruction, so e.g. setting a context with TF set will execute a single instruction
+  // and then raise an exception. However on the FEX side this is simpler to implement by checking at the start of each instruction, handle this by having bit 1 being unset in the flag state indicate that TF is blocked for a single instruction.
+  (void)tbz(TMP1, 1, &l_TFBlocked);

-    // Block TF for a single instruction when the frontend jumps to a new context by unsetting bit 1.
-    ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
-    and_(ARMEmitter::Size::i32Bit, TMP1, TMP1, ~(1 << 1));
-    strb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
+  // Block TF for a single instruction when the frontend jumps to a new context by unsetting bit 1.
+  ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
+  and_(ARMEmitter::Size::i32Bit, TMP1, TMP1, ~(1 << 1));
+  strb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));

-    Core::CpuStateFrame::SynchronousFaultDataStruct State = {
-      .FaultToTopAndGeneratedException = 1,
-      .Signal = Core::FAULT_SIGTRAP,
-      .TrapNo = X86State::X86_TRAPNO_DB,
-      .si_code = 2,
-      .err_code = 0,
-    };
+  Core::CpuStateFrame::SynchronousFaultDataStruct State = {
+    .FaultToTopAndGeneratedException = 1,
+    .Signal = Core::FAULT_SIGTRAP,
+    .TrapNo = X86State::X86_TRAPNO_DB,
+    .si_code = 2,
+    .err_code = 0,
+  };

-    uint64_t Constant {};
-    memcpy(&Constant, &State, sizeof(State));
+  uint64_t Constant {};
+  memcpy(&Constant, &State, sizeof(State));

-    LoadConstant(ARMEmitter::Size::i64Bit, TMP1, Constant);
-    str(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, SynchronousFaultData));
-    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP));
-    br(TMP1);
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP1, Constant);
+  str(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, SynchronousFaultData));
+  ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.GuestSignal_SIGTRAP));
+  br(TMP1);

-    Bind(&l_TFBlocked);
-    // If TF was blocked for this instruction, unblock it for the next.
-    LoadConstant(ARMEmitter::Size::i32Bit, TMP1, 0b11);
-    strb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
-    Bind(&l_TFUnset);
-  }
+  (void)Bind(&l_TFBlocked);
+  // If TF was blocked for this instruction, unblock it for the next.
+  LoadConstant(ARMEmitter::Size::i32Bit, TMP1, 0b11);
+  strb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
+  (void)Bind(&l_TFUnset);
+}

+void Arm64JITCore::EmitSuspendInterruptCheck() {
  if (CTX->Config.NeedsPendingInterruptFaultCheck) {
    // Trigger a fault if there are any pending interrupts
    // Used only for suspend on WIN32 at the moment
@@ -789,24 +774,26 @@ void Arm64JITCore::EmitInterruptChecks(bool CheckTF) {
         offsetof(FEXCore::Core::InternalThreadState, InterruptFaultPage) - offsetof(FEXCore::Core::InternalThreadState, BaseFrameState));
  }

-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
  static constexpr uint16_t SuspendMagic {0xCAFE};

  ldr(TMP2.W(), STATE_PTR(CpuStateFrame, SuspendDoorbell));
  ARMEmitter::ForwardLabel l_NoSuspend;
-  cbz(ARMEmitter::Size::i32Bit, TMP2, &l_NoSuspend);
+  (void)cbz(ARMEmitter::Size::i32Bit, TMP2, &l_NoSuspend);
  brk(SuspendMagic);
-  Bind(&l_NoSuspend);
+  (void)Bind(&l_NoSuspend);
 #endif
 }

 void Arm64JITCore::EmitEntryPoint(ARMEmitter::BackwardLabel& HeaderLabel, bool CheckTF) {
  // Get the address of the JITCodeHeader and store in to the core state.
  // Two instruction cost, each 1 cycle.
-  adr(TMP1, &HeaderLabel);
+  adr_OrRestart(TMP1, &HeaderLabel);
  str(TMP1, STATE, offsetof(FEXCore::Core::CPUState, InlineJITBlockHeader));

-  EmitInterruptChecks(CheckTF);
+  if (CheckTF) {
+    EmitTFCheck();
+  }

  if (SpillSlots) {
    const auto TotalSpillSlotsSize = SpillSlots * MaxSpillSlotSize;
@@ -818,36 +805,65 @@ void Arm64JITCore::EmitEntryPoint(ARMEmitter::BackwardLabel& HeaderLabel, bool C
      sub(ARMEmitter::Size::i64Bit, ARMEmitter::XReg::rsp, ARMEmitter::XReg::rsp, TMP1, ARMEmitter::ExtendedType::LSL_64, 0);
    }
  }
+
+  EmitSuspendInterruptCheck();
 }

 CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size, bool SingleInst, const FEXCore::IR::IRListView* IR,
                                                   FEXCore::Core::DebugData* DebugData, bool CheckTF) {
  FEXCORE_PROFILE_SCOPED("Arm64::CompileCode");

-  JumpTargets.clear();
-  CallReturnTargets.clear();
-  PendingJumpThunks.clear();
-  uint32_t SSACount = IR->GetSSACount();
-  JumpTargets.resize(IR->GetHeader()->BlockCount, {});
+  const auto PrevNumAllocations = Relocations.size();

  this->Entry = Entry;
  this->DebugData = DebugData;
  this->IR = IR;
+  RequiresFarARM64Jumps = false;
+  SSANodeMultiplier = 24;
+
+  // Prepare restart via long jump in case branch encoding fails.
+  // This uses UncheckedLongJump since we don't implement std::longjmp in WoA setups
+  switch (static_cast<RestartOptions::Control>(FEXCore::UncheckedLongJump::SetJump(ThreadState->RestartJump))) {
+  case RestartOptions::Control::Incoming:
+    // Nothing
+    break;
+  case RestartOptions::Control::EnableFarARM64Jumps: RequiresFarARM64Jumps = true; break;
+  case RestartOptions::Control::NeedsLargerJITSpace:
+    // Get rid of the claimed buffer immediately, we can't fit in it at all.
+    TempAllocator.UnclaimBuffer();
+    SSANodeMultiplier *= 2;
+    break;
+  default: LOGMAN_MSG_A_FMT("Unhandled Arm64 restart condition!");
+  }
+
+  uint32_t SSACount = IR->GetSSACount();
+  JumpTargets.clear();
+  CallReturnTargets.clear();
+  PendingJumpThunks.clear();
+  JumpTargets.resize(IR->GetHeader()->BlockCount, {});
+
  CodeData.EntryPoints.clear();

  // Fairly excessive buffer range to make sure we don't overflow
-  uint32_t BufferRange = 0x1000 + SSACount * 24;
+  // One page baseline, plus SSANodeMultipler bytes, plus another page for guard page.
+  const uint32_t DesiredBufferRange = AlignUp(FEXCore::Utils::FEX_PAGE_SIZE * 2 + SSACount * SSANodeMultiplier, FEXCore::Utils::FEX_PAGE_SIZE);

  // JIT output is first written to a temporary buffer and later relocated to the CodeBuffer.
  // This minimizes lock contention of CodeBufferWriteMutex.
-  auto TempCodeBuffer = TempAllocator.ReownOrClaimBuffer(BufferRange);
-  SetBuffer(TempCodeBuffer, BufferRange);
+  auto TempCodeBufferInfo = TempAllocator.ReownOrClaimBufferWithSize(DesiredBufferRange);
+  auto TempCodeBuffer = TempCodeBufferInfo.Ptr;
+  const uint32_t UsableBufferRange = TempCodeBufferInfo.Size - FEXCore::Utils::FEX_PAGE_SIZE;
+
+  SetBuffer(TempCodeBuffer, UsableBufferRange);
+
+  ThreadState->JITGuardPage = reinterpret_cast<uintptr_t>(TempCodeBuffer) + UsableBufferRange;
+  ThreadState->JITGuardOverflowArgument = FEXCore::ToUnderlying(RestartOptions::Control::NeedsLargerJITSpace);

  CodeData.BlockBegin = GetCursorAddress<uint8_t*>();

  // Put the code header at the start of the data block.
  ARMEmitter::BackwardLabel JITCodeHeaderLabel {};
-  Bind(&JITCodeHeaderLabel);
+  (void)Bind(&JITCodeHeaderLabel);
  JITCodeHeader* CodeHeader = GetCursorAddress<JITCodeHeader*>();
  CursorIncrement(sizeof(JITCodeHeader));

@@ -892,7 +908,10 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size

      // if there's a pending branch, and it is not fall-through
      if (PendingTargetLabel && PendingTargetLabel != Target) {
-        b(PendingTargetLabel);
+        if (PendingTargetLabel->Backward.Location) {
+          EmitSuspendInterruptCheck();
+        }
+        b_OrRestart(PendingTargetLabel);
        PendingTargetLabel = nullptr;
      }

@@ -902,14 +921,14 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
        const auto IsReturnTarget = CallReturnTargets.try_emplace(Node).first;
        if (PendingTargetLabel) {
          // If there is a fallthrough branch to this block, skip over the entrypoint code.
-          b(Target);
+          b_OrRestart(Target);
        } else if (PendingCallReturnTargetLabel && PendingCallReturnTargetLabel != &IsReturnTarget->second) {
          // If we just emitted a call, but the block we're now emitting is not the return block so don't fallthrough.
-          b(PendingCallReturnTargetLabel);
+          b_OrRestart(PendingCallReturnTargetLabel);
        }
        PendingCallReturnTargetLabel = nullptr;

-        Bind(&IsReturnTarget->second);
+        BindOrRestart(&IsReturnTarget->second);
        CodeData.EntryPoints.emplace(BlockStartRIP, GetCursorAddress<uint8_t*>());
        DebugData->GuestOpcodes.push_back({BlockIROp->GuestEntryOffset, GetCursorAddress<uint8_t*>() - CodeData.BlockBegin});

@@ -918,18 +937,16 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size

      if (PendingCallReturnTargetLabel) {
        // If there is still a pending call return target, then the block we're emitting is not the return block so don't fallthrough.
-        b(PendingCallReturnTargetLabel);
+        b_OrRestart(PendingCallReturnTargetLabel);
        PendingCallReturnTargetLabel = nullptr;
      }
      PendingTargetLabel = nullptr;

-      Bind(Target);
+      BindOrRestart(Target);
    }

    for (auto [CodeNode, IROp] : IR->GetCode(BlockNode)) {
      switch (IROp->Op) {
-#define REGISTER_OP_RT(op, x) \
-  case FEXCore::IR::IROps::OP_##op: std::invoke(RT_##x, this, IROp, CodeNode); break
 #define REGISTER_OP(op, x) \
  case FEXCore::IR::IROps::OP_##op: Op_##x(IROp, CodeNode); break

@@ -947,7 +964,10 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size

  // Make sure last branch is generated. It certainly can't be eliminated here.
  if (PendingTargetLabel) {
-    b(PendingTargetLabel);
+    if (PendingTargetLabel->Backward.Location) {
+      EmitSuspendInterruptCheck();
+    }
+    b_OrRestart(PendingTargetLabel);
  }
  PendingTargetLabel = nullptr;

@@ -958,31 +978,37 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size

    ARMEmitter::ForwardLabel l_DoLink;
    uint64_t ThunkAddress = GetCursorAddress<uint64_t>();
-    Bind(&PendingJumpThunk.Label);
-    b(&l_DoLink);
+    BindOrRestart(&PendingJumpThunk.Label);
+    b_OrRestart(&l_DoLink);
    br(TMP1);
-    Bind(&l_DoLink);
+    BindOrRestart(&l_DoLink);
    ldr(TMP1, &l_ExitLink);
    blr(TMP1);

    // This is a ExitFunctionLinkData struct
-    Bind(&l_ExitLink);
-    dc64(0);                                             // HostCode
-    dc64(PendingJumpThunk.GuestRIP);                     // GuestRIP
-    dc64(PendingJumpThunk.CallerAddress - ThunkAddress); // CallerOffset
+    BindOrRestart(&l_ExitLink);
+    dc64(0);                                                                   // HostCode
+    PlaceNamedSymbolLiteral(InsertGuestRIPLiteral(PendingJumpThunk.GuestRIP)); // GuestRIP
+    dc64(PendingJumpThunk.CallerAddress - ThunkAddress);                       // CallerOffset
  }

-  Bind(&l_ExitLink);
-  dc64(ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker);
+  BindOrRestart(&l_ExitLink);
+  PlaceNamedSymbolLiteral(InsertNamedSymbolLiteral(RelocNamedSymbolLiteral::NamedSymbol::SYMBOL_LITERAL_EXITFUNCTION_LINKER));

  // CodeSize not including the header or tail data.
  const uint64_t CodeOnlySize = GetCursorAddress<uint8_t*>() - CodeBegin;

-  // Add the JitCodeTail
+  // Add the JitCodeTail (written later)
  Align(alignof(JITCodeTail));
-  auto JITBlockTailLocation = GetCursorAddress<uint8_t*>();
-  auto JITBlockTail = GetCursorAddress<JITCodeTail*>();
-  CursorIncrement(sizeof(JITCodeTail));
+  const auto JITBlockTailLocation = GetCursorAddress<uint8_t*>();
+  CodeHeader->OffsetToBlockTail = JITBlockTailLocation - CodeData.BlockBegin;
+
+  JITCodeTail JITBlockTail {
+    .RIP = Entry,
+    .GuestSize = Size,
+    .SpinLockFutex = 0,
+    .SingleInst = SingleInst,
+  };

  // Entries that live after the JITCodeTail.
  // These entries correlate JIT code regions with guest RIP regions.
@@ -1000,23 +1026,13 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
  //   FEXCore::Utils::vl64 GuestRIPOffset;
  // };

-  auto JITRIPEntriesBegin = GetCursorAddress<uint8_t*>();
-
-  // Put the block's RIP entry in the tail.
-  // This will be used for RIP reconstruction in the future.
-  // TODO: This needs to be a data RIP relocation once code caching works.
-  //   Current relocation code doesn't support this feature yet.
-  JITBlockTail->RIP = Entry;
-  JITBlockTail->GuestSize = Size;
-  JITBlockTail->SingleInst = SingleInst;
-  JITBlockTail->SpinLockFutex = 0;
-
+  const auto JITRIPEntriesBegin = JITBlockTailLocation + sizeof(JITBlockTail);
  auto JITRIPEntriesLocation = JITRIPEntriesBegin;

  {
    // Store the RIP entries.
-    JITBlockTail->NumberOfRIPEntries = DebugData->GuestOpcodes.size();
-    JITBlockTail->OffsetToRIPEntries = JITRIPEntriesBegin - JITBlockTailLocation;
+    JITBlockTail.NumberOfRIPEntries = DebugData->GuestOpcodes.size();
+    JITBlockTail.OffsetToRIPEntries = JITRIPEntriesBegin - JITBlockTailLocation;
    uintptr_t CurrentRIPOffset = 0;
    uint64_t CurrentPCOffset = 0;

@@ -1032,14 +1048,20 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
    }
  }

-  CursorIncrement(JITRIPEntriesLocation - JITRIPEntriesBegin);
+  SetCursorOffset(JITRIPEntriesLocation - CodeData.BlockBegin);
  Align();

-  CodeHeader->OffsetToBlockTail = JITBlockTailLocation - CodeData.BlockBegin;
-
  CodeData.Size = GetCursorAddress<uint8_t*>() - CodeData.BlockBegin;

-  JITBlockTail->Size = CodeData.Size;
+  // Finalize and write block tail data
+  JITBlockTail.Size = CodeData.Size;
+  {
+    auto PrevCur = GetCursorOffset();
+    memcpy(JITBlockTailLocation, &JITBlockTail, sizeof(JITBlockTail));
+    SetCursorOffset(JITBlockTailLocation - CodeData.BlockBegin + offsetof(JITCodeTail, RIP));
+    PlaceNamedSymbolLiteral(InsertGuestRIPLiteral(JITBlockTail.RIP));
+    SetCursorOffset(PrevCur);
+  }

  // Migrate the compile output from temporary storage to the actual CodeBuffer.
  // This can block progress in other compiling threads, so the duration of the lock should be as small as possible.
@@ -1048,7 +1070,6 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size

    // Query size of generated code
    const auto TempSize = GetCursorOffset();
-    LOGMAN_THROW_A_FMT(TempSize <= BufferRange, "Exceeded bounds of temporary buffer ({:#x} vs {:#x})", TempSize, BufferRange);

    // Bring CodeBuffer up to date
    {
@@ -1056,7 +1077,8 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
                                                                                                   "doesn't match up!\n");
      if (auto Prev = CheckCodeBufferUpdate()) {
        Allocator::VirtualDontNeed(ThreadState->CallRetStackBase, FEXCore::Core::InternalThreadState::CALLRET_STACK_SIZE);
-        ThreadState->LookupCache->ChangeGuestToHostMapping(*Prev, *CurrentCodeBuffer->LookupCache);
+        auto lk = ThreadState->LookupCache->AcquireWriteLock();
+        ThreadState->LookupCache->ChangeGuestToHostMapping(*Prev, *CurrentCodeBuffer->LookupCache, lk);
      }

      // NOTE: 16-byte alignment of the new cursor offset must be preserved for block linking records
@@ -1079,6 +1101,10 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
    }
    CodeBegin += Delta;

+    for (std::size_t Idx = PrevNumAllocations; Idx != Relocations.size(); ++Idx) {
+      Relocations[Idx].Header.Offset += CodeBuffers.LatestOffset;
+    }
+
    // Copy over CodeBuffer contents
    memcpy(GetCursorAddress<uint8_t*>(), TempCodeBuffer, TempSize);
    SetCursorOffset(CodeBuffers.LatestOffset + TempSize);
@@ -10,31 +10,39 @@ $end_info$
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
 #include "Interface/Core/CPUBackend.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
+#include "Interface/Core/JIT/Relocations.h"
 #include "Interface/IR/IR.h"
 #include "Interface/IR/IntrusiveIRList.h"
 #include "Interface/IR/RegisterAllocationData.h"

+#include <FEXCore/Config/Config.h>
 #include <FEXCore/Core/CoreState.h>
 #include <FEXCore/IR/IR.h>
+#include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/fextl/map.h>
+#include <FEXCore/fextl/memory.h>
 #include <FEXCore/fextl/string.h>
 #include <FEXCore/fextl/vector.h>
+#include <FEXCore/Utils/LongJump.h>

 #include <CodeEmitter/Emitter.h>

 #include <array>
 #include <cstdint>
 #include <functional>
+#include <optional>
 #include <utility>
 #include <variant>

 namespace FEXCore::Core {
 struct InternalThreadState;
 }
-
 namespace FEXCore::Context {
 struct ExitFunctionLinkData;
 }
+namespace FEXCore::IR {
+class RegisterAllocationPass;
+}

 namespace FEXCore::CPU {
 class Arm64JITCore final : public CPUBackend, public Arm64Emitter {
@@ -53,14 +61,27 @@ public:
  }

 private:
-  FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
-
  const bool HostSupportsSVE128 {};
  const bool HostSupportsSVE256 {};
  const bool HostSupportsAVX256 {};
  const bool HostSupportsRPRES {};
  const bool HostSupportsAFP {};

+  struct RestartOptions {
+    enum class Control : uint64_t {
+      Incoming = 0,
+      EnableFarARM64Jumps = 1,
+      NeedsLargerJITSpace = 2,
+    };
+  };
+
+  // FEXCore makes assumptions in the JIT about certain conditions being true.
+  // In the rare case when those assumptions are broken, FEX needs to safely restart the JIT.
+  RestartOptions RestartControl {};
+  bool RequiresFarARM64Jumps {};
+  // Default to 6 instructions per SSA node.
+  uint32_t SSANodeMultiplier {24};
+
  ARMEmitter::BiDirectionalLabel* PendingTargetLabel {};
  ARMEmitter::BiDirectionalLabel* PendingCallReturnTargetLabel {};
  FEXCore::Context::ContextImpl* CTX {};
@@ -90,11 +111,13 @@ private:

  [[nodiscard]]
  ARMEmitter::Register GetReg(IR::PhysicalRegister Reg) const {
-    LOGMAN_THROW_A_FMT(Reg.Class == IR::GPRFixedClass.Val || Reg.Class == IR::GPRClass.Val, "Unexpected Class: {}", Reg.Class);
+    const auto RegClass = Reg.AsRegClass();

-    if (Reg.Class == IR::GPRFixedClass.Val) {
+    LOGMAN_THROW_A_FMT(RegClass == IR::RegClass::GPRFixed || RegClass == IR::RegClass::GPR, "Unexpected Class: {}", Reg.Class);
+
+    if (RegClass == IR::RegClass::GPRFixed) {
      return StaticRegisters[Reg.Reg];
-    } else if (Reg.Class == IR::GPRClass.Val) {
+    } else if (RegClass == IR::RegClass::GPR) {
      return GeneralRegisters[Reg.Reg];
    }

@@ -113,11 +136,13 @@ private:

  [[nodiscard]]
  ARMEmitter::VRegister GetVReg(IR::PhysicalRegister Reg) const {
-    LOGMAN_THROW_A_FMT(Reg.Class == IR::FPRFixedClass.Val || Reg.Class == IR::FPRClass.Val, "Unexpected Class: {}", Reg.Class);
+    const auto RegClass = Reg.AsRegClass();

-    if (Reg.Class == IR::FPRFixedClass.Val) {
+    LOGMAN_THROW_A_FMT(RegClass == IR::RegClass::FPRFixed || RegClass == IR::RegClass::FPR, "Unexpected Class: {}", Reg.Class);
+
+    if (RegClass == IR::RegClass::FPRFixed) {
      return StaticFPRegisters[Reg.Reg];
-    } else if (Reg.Class == IR::FPRClass.Val) {
+    } else if (RegClass == IR::RegClass::FPR) {
      return GeneralFPRegisters[Reg.Reg];
    }

@@ -135,8 +160,8 @@ private:
  }

  [[nodiscard]]
-  FEXCore::IR::RegisterClassType GetRegClass(IR::Ref Node) const {
-    return FEXCore::IR::RegisterClassType {IR::PhysicalRegister(Node).Class};
+  static IR::RegClass GetRegClass(IR::Ref Node) {
+    return IR::PhysicalRegister(Node).AsRegClass();
  }

  [[nodiscard]]
@@ -153,7 +178,7 @@ private:
  // Converts IR-base shift type to ARMEmitter shift type.
  // Will be a no-op, only a type conversion since the two definitions match.
  [[nodiscard]]
-  ARMEmitter::ShiftType ConvertIRShiftType(IR::ShiftType Shift) const {
+  static ARMEmitter::ShiftType ConvertIRShiftType(IR::ShiftType Shift) {
    return Shift == IR::ShiftType::LSL ? ARMEmitter::ShiftType::LSL :
           Shift == IR::ShiftType::LSR ? ARMEmitter::ShiftType::LSR :
           Shift == IR::ShiftType::ASR ? ARMEmitter::ShiftType::ASR :
@@ -161,18 +186,23 @@ private:
  }

  [[nodiscard]]
-  ARMEmitter::Size ConvertSize(const IR::IROp_Header* Op) {
+  static ARMEmitter::Size ConvertSize(const IR::IROp_Header* Op) {
    return Op->Size == IR::OpSize::i64Bit ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
  }

  [[nodiscard]]
-  ARMEmitter::Size ConvertSize48(const IR::IROp_Header* Op) {
+  static ARMEmitter::Size ConvertSize48(const IR::IROp_Header* Op) {
    LOGMAN_THROW_A_FMT(Op->Size == IR::OpSize::i32Bit || Op->Size == IR::OpSize::i64Bit, "Invalid size");
    return ConvertSize(Op);
  }

  [[nodiscard]]
-  ARMEmitter::SubRegSize ConvertSubRegSize16(IR::OpSize ElementSize) {
+  static ARMEmitter::Size ConvertSize(IR::OpSize Size) {
+    return Size == IR::OpSize::i64Bit ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
+  }
+
+  [[nodiscard]]
+  static ARMEmitter::SubRegSize ConvertSubRegSize16(IR::OpSize ElementSize) {
    LOGMAN_THROW_A_FMT(ElementSize == IR::OpSize::i8Bit || ElementSize == IR::OpSize::i16Bit || ElementSize == IR::OpSize::i32Bit ||
                         ElementSize == IR::OpSize::i64Bit || ElementSize == IR::OpSize::i128Bit,
                       "Invalid size");
@@ -184,105 +214,105 @@ private:
  }

  [[nodiscard]]
-  ARMEmitter::SubRegSize ConvertSubRegSize16(const IR::IROp_Header* Op) {
+  static ARMEmitter::SubRegSize ConvertSubRegSize16(const IR::IROp_Header* Op) {
    return ConvertSubRegSize16(Op->ElementSize);
  }

  [[nodiscard]]
-  ARMEmitter::SubRegSize ConvertSubRegSize8(IR::OpSize ElementSize) {
+  static ARMEmitter::SubRegSize ConvertSubRegSize8(IR::OpSize ElementSize) {
    LOGMAN_THROW_A_FMT(ElementSize != IR::OpSize::i128Bit, "Invalid size");
    return ConvertSubRegSize16(ElementSize);
  }

  [[nodiscard]]
-  ARMEmitter::SubRegSize ConvertSubRegSize8(const IR::IROp_Header* Op) {
+  static ARMEmitter::SubRegSize ConvertSubRegSize8(const IR::IROp_Header* Op) {
    return ConvertSubRegSize8(Op->ElementSize);
  }

  [[nodiscard]]
-  ARMEmitter::SubRegSize ConvertSubRegSize4(const IR::IROp_Header* Op) {
+  static ARMEmitter::SubRegSize ConvertSubRegSize4(const IR::IROp_Header* Op) {
    LOGMAN_THROW_A_FMT(Op->ElementSize != IR::OpSize::i64Bit, "Invalid size");
    return ConvertSubRegSize8(Op);
  }

  [[nodiscard]]
-  ARMEmitter::SubRegSize ConvertSubRegSize248(const IR::IROp_Header* Op) {
+  static ARMEmitter::SubRegSize ConvertSubRegSize248(const IR::IROp_Header* Op) {
    LOGMAN_THROW_A_FMT(Op->ElementSize != IR::OpSize::i8Bit, "Invalid size");
    return ConvertSubRegSize8(Op);
  }

  [[nodiscard]]
-  ARMEmitter::VectorRegSizePair ConvertSubRegSizePair16(const IR::IROp_Header* Op) {
+  static ARMEmitter::VectorRegSizePair ConvertSubRegSizePair16(const IR::IROp_Header* Op) {
    return ARMEmitter::ToVectorSizePair(ConvertSubRegSize16(Op));
  }

  [[nodiscard]]
-  ARMEmitter::VectorRegSizePair ConvertSubRegSizePair8(const IR::IROp_Header* Op) {
+  static ARMEmitter::VectorRegSizePair ConvertSubRegSizePair8(const IR::IROp_Header* Op) {
    LOGMAN_THROW_A_FMT(Op->ElementSize != IR::OpSize::i128Bit, "Invalid size");
    return ConvertSubRegSizePair16(Op);
  }

  [[nodiscard]]
-  ARMEmitter::VectorRegSizePair ConvertSubRegSizePair248(const IR::IROp_Header* Op) {
+  static ARMEmitter::VectorRegSizePair ConvertSubRegSizePair248(const IR::IROp_Header* Op) {
    LOGMAN_THROW_A_FMT(Op->ElementSize != IR::OpSize::i8Bit, "Invalid size");
    return ConvertSubRegSizePair8(Op);
  }

  [[nodiscard]]
-  ARMEmitter::Condition MapCC(IR::CondClassType Cond) {
-    switch (Cond.Val) {
-    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:
-    case FEXCore::IR::COND_VS: return ARMEmitter::Condition::CC_VS;
-    case FEXCore::IR::COND_FNU:
-    case FEXCore::IR::COND_VC: return ARMEmitter::Condition::CC_VC;
-    case FEXCore::IR::COND_MI: return ARMEmitter::Condition::CC_MI;
-    case FEXCore::IR::COND_PL: return ARMEmitter::Condition::CC_PL;
+  static ARMEmitter::Condition MapCC(IR::CondClass Cond) {
+    switch (Cond) {
+    case IR::CondClass::EQ: return ARMEmitter::Condition::CC_EQ;
+    case IR::CondClass::NEQ: return ARMEmitter::Condition::CC_NE;
+    case IR::CondClass::SGE: return ARMEmitter::Condition::CC_GE;
+    case IR::CondClass::SLT: return ARMEmitter::Condition::CC_LT;
+    case IR::CondClass::SGT: return ARMEmitter::Condition::CC_GT;
+    case IR::CondClass::SLE: return ARMEmitter::Condition::CC_LE;
+    case IR::CondClass::UGE: return ARMEmitter::Condition::CC_CS;
+    case IR::CondClass::ULT: return ARMEmitter::Condition::CC_CC;
+    case IR::CondClass::UGT: return ARMEmitter::Condition::CC_HI;
+    case IR::CondClass::ULE: return ARMEmitter::Condition::CC_LS;
+    case IR::CondClass::FLU: return ARMEmitter::Condition::CC_LT;
+    case IR::CondClass::FGE: return ARMEmitter::Condition::CC_GE;
+    case IR::CondClass::FLEU: return ARMEmitter::Condition::CC_LE;
+    case IR::CondClass::FGT: return ARMEmitter::Condition::CC_GT;
+    case IR::CondClass::FU:
+    case IR::CondClass::VS: return ARMEmitter::Condition::CC_VS;
+    case IR::CondClass::FNU:
+    case IR::CondClass::VC: return ARMEmitter::Condition::CC_VC;
+    case IR::CondClass::MI: return ARMEmitter::Condition::CC_MI;
+    case IR::CondClass::PL: return ARMEmitter::Condition::CC_PL;
    default: LOGMAN_MSG_A_FMT("Unsupported compare type"); return ARMEmitter::Condition::CC_NV;
    }
  }

  [[nodiscard]]
-  bool IsFPR(IR::RegisterClassType Class) const {
-    return Class == IR::FPRClass || Class == IR::FPRFixedClass;
+  static bool IsFPR(IR::RegClass Class) {
+    return Class == IR::RegClass::FPR || Class == IR::RegClass::FPRFixed;
  }

  [[nodiscard]]
-  bool IsGPR(IR::RegisterClassType Class) const {
-    return Class == IR::GPRClass || Class == IR::GPRFixedClass;
+  static bool IsGPR(IR::RegClass Class) {
+    return Class == IR::RegClass::GPR || Class == IR::RegClass::GPRFixed;
  }

  [[nodiscard]]
-  bool IsGPR(IR::Ref Node) {
+  static bool IsGPR(IR::Ref Node) {
    return IsGPR(GetRegClass(Node));
  }

  [[nodiscard]]
-  bool IsFPR(IR::Ref Node) {
+  static bool IsFPR(IR::Ref Node) {
    return IsFPR(GetRegClass(Node));
  }

  [[nodiscard]]
-  bool IsGPR(IR::OrderedNodeWrapper Wrap) {
-    return IsGPR(IR::RegisterClassType {IR::PhysicalRegister(Wrap).Class});
+  static bool IsGPR(IR::OrderedNodeWrapper Wrap) {
+    return IsGPR(IR::PhysicalRegister(Wrap).AsRegClass());
  }

  [[nodiscard]]
-  bool IsFPR(IR::OrderedNodeWrapper Wrap) {
-    return IsFPR(IR::RegisterClassType {IR::PhysicalRegister(Wrap).Class});
+  static bool IsFPR(IR::OrderedNodeWrapper Wrap) {
+    return IsFPR(IR::PhysicalRegister(Wrap).AsRegClass());
  }

  [[nodiscard]]
@@ -315,14 +345,187 @@ private:
  void EmitLinkedBranch(uint64_t GuestRIP, bool Call) {
    PendingJumpThunks.push_back({GetCursorAddress<uint64_t>(), GuestRIP, {}});
    auto& Thunk = PendingJumpThunks.back();
-    Bind(&Thunk.Label);
+    BindOrRestart(&Thunk.Label);
    if (Call) {
-      bl(&Thunk.Label);
+      bl_OrRestart(&Thunk.Label);
    } else {
-      b(&Thunk.Label);
+      b_OrRestart(&Thunk.Label);
    }
  }

+  // Restart helpers
+  template<ARMEmitter::IsLabel T>
+  void bl_OrRestart(T* Label) {
+    if (bl(Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    // We can support this but currently unnecessary.
+    ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void b_OrRestart(T* Label) {
+    if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    // We can support this but currently unnecessary.
+    ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void b_OrRestart(ARMEmitter::Condition Cond, T* Label) {
+    if (RequiresFarARM64Jumps) {
+      ARMEmitter::ForwardLabel Skip {};
+      // Wrap a manual Cond check around an unconditional branch; this can encode larger offsets
+      (void)b(InvertCondition(Cond), &Skip);
+      if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
+        ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
+      }
+
+      (void)Bind(&Skip);
+      return;
+    }
+
+    if (b(Cond, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void cbz_OrRestart(ARMEmitter::Size s, ARMEmitter::Register rt, T* Label) {
+    if (RequiresFarARM64Jumps) {
+      ARMEmitter::ForwardLabel Skip {};
+      // Wrap a manual Cond check around an unconditional branch; this can encode larger offsets
+      (void)cbnz(s, rt, &Skip);
+      if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
+        ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
+      }
+
+      (void)Bind(&Skip);
+      return;
+    }
+
+    if (cbz(s, rt, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void cbnz_OrRestart(ARMEmitter::Size s, ARMEmitter::Register rt, T* Label) {
+    if (RequiresFarARM64Jumps) {
+      ARMEmitter::ForwardLabel Skip {};
+      // Wrap a manual Cond check around an unconditional branch; this can encode larger offsets
+      (void)cbz(s, rt, &Skip);
+      if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
+        ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
+      }
+
+      (void)Bind(&Skip);
+      return;
+    }
+
+    if (cbnz(s, rt, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void tbz_OrRestart(ARMEmitter::Register rt, uint32_t Bit, T* Label) {
+    if (RequiresFarARM64Jumps) {
+      ARMEmitter::ForwardLabel Skip {};
+      // Wrap a manual Cond check around an unconditional branch; this can encode larger offsets
+      (void)tbnz(rt, Bit, &Skip);
+      if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
+        ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
+      }
+
+      (void)Bind(&Skip);
+      return;
+    }
+
+    if (tbz(rt, Bit, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void tbnz_OrRestart(ARMEmitter::Register rt, uint32_t Bit, T* Label) {
+    if (RequiresFarARM64Jumps) {
+      ARMEmitter::ForwardLabel Skip {};
+      // Wrap a manual Cond check around an unconditional branch; this can encode larger offsets
+      (void)tbz(rt, Bit, &Skip);
+      if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
+        ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
+      }
+
+      (void)Bind(&Skip);
+      return;
+    }
+
+    if (tbnz(rt, Bit, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void adr_OrRestart(ARMEmitter::Register rd, T* Label) {
+    if (RequiresFarARM64Jumps) {
+      if (LongAddressGen(rd, Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
+        ERROR_AND_DIE_FMT("Unable to encode long ADR.");
+      }
+      return;
+    }
+    if (adr(rd, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void adrp_OrRestart(ARMEmitter::Register rd, T* Label) {
+    if (RequiresFarARM64Jumps) {
+      if (LongAddressGen(rd, Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
+        ERROR_AND_DIE_FMT("Unable to encode long ADRP.");
+      }
+      return;
+    }
+    if (adrp(rd, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
+      return;
+    }
+
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
+  template<ARMEmitter::IsLabel T>
+  void BindOrRestart(T* Label) {
+    if (Bind(Label)) {
+      return;
+    }
+
+    if (RequiresFarARM64Jumps) {
+      // This should have been caught before this point.
+      ERROR_AND_DIE_FMT("Unhandled long bind");
+      return;
+    }
+
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+  }
+
  // This is purely a debugging aid for developers to see if they are in JIT code space when inspecting raw memory
  void EmitDetectionString();
  IR::RegisterAllocationPass* RAPass {};
@@ -333,8 +536,6 @@ private:
   * @name Relocations
   * @{ */

-  uint64_t GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op);
-
  /**
   * @brief A literal pair relocation object for named symbol literals
   */
@@ -371,17 +572,30 @@ private:
   */
  NamedSymbolLiteralPair InsertNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op);

+  /**
+   * @brief Inserts a relocation for a constant value relative to the guest entrypoint
+   *
+   * @param Reg - The GPR to move the guest RIP in to
+   * @param Constant - The guest RIP that will be relocated
+   */
+  NamedSymbolLiteralPair InsertGuestRIPLiteral(uint64_t GuestRIP);
+
  /**
   * @brief Place the named symbol literal relocation in memory
   *
   * @param Lit - Which literal to place
   */
-  void PlaceNamedSymbolLiteral(NamedSymbolLiteralPair& Lit);
+  void PlaceNamedSymbolLiteral(NamedSymbolLiteralPair Lit);

  fextl::vector<FEXCore::CPU::Relocation> Relocations;

-  ///< Relocation code loading
-  bool ApplyRelocations(uint64_t GuestEntry, uint64_t CodeEntry, uint64_t CursorEntry, size_t NumRelocations, const char* EntryRelocations);
+  /**
+   * Returns any relocations generated since the last call to TakeRelocations.
+   *
+   * GuestBaseAddress must match the base virtual address to which the
+   * input x86 binary is mapped.
+   */
+  fextl::vector<FEXCore::CPU::Relocation> TakeRelocations(uint64_t GuestBaseAddress) override;

  /**  @} */

@@ -404,23 +618,16 @@ private:
  void Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize, ARMEmitter::VRegister Dst, ARMEmitter::VRegister IncomingDst,
                           std::optional<ARMEmitter::Register> BaseAddr, ARMEmitter::VRegister VectorIndexLow,
                           std::optional<ARMEmitter::VRegister> VectorIndexHigh, ARMEmitter::VRegister MaskReg, IR::OpSize VectorIndexSize,
-                           size_t DataElementOffsetStart, size_t IndexElementOffsetStart, uint8_t OffsetScale);
+                           size_t DataElementOffsetStart, size_t IndexElementOffsetStart, uint8_t OffsetScale, IR::OpSize AddrSize);

-  void EmitInterruptChecks(bool CheckTF);
+  void EmitTFCheck();
+
+  void EmitSuspendInterruptCheck();

  void EmitEntryPoint(ARMEmitter::BackwardLabel& HeaderLabel, bool CheckTF);

-  // Runtime selection;
-  // Load and store TSO memory style
-  OpType RT_LoadMemTSO;
-  OpType RT_StoreMemTSO;
-
 #define DEF_OP(x) void Op_##x(IR::IROp_Header const* IROp, IR::Ref Node)

-  // Dynamic Dispatcher supporting operations
-  DEF_OP(ParanoidLoadMemTSO);
-  DEF_OP(ParanoidStoreMemTSO);
-
  ///< Unhandled handler
  DEF_OP(Unhandled);

@@ -21,7 +21,7 @@ DEF_OP(LoadContext) {
  const auto Op = IROp->C<IR::IROp_LoadContext>();
  const auto OpSize = IROp->Size;

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    auto Dst = GetReg(Node);

    switch (OpSize) {
@@ -52,7 +52,7 @@ DEF_OP(LoadContext) {
 DEF_OP(LoadContextPair) {
  const auto Op = IROp->C<IR::IROp_LoadContextPair>();

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    const auto Dst1 = GetReg(Op->OutValue1);
    const auto Dst2 = GetReg(Op->OutValue2);

@@ -78,7 +78,7 @@ DEF_OP(StoreContext) {
  const auto Op = IROp->C<IR::IROp_StoreContext>();
  const auto OpSize = IROp->Size;

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    auto Src = GetZeroableReg(Op->Value);

    switch (OpSize) {
@@ -110,7 +110,7 @@ DEF_OP(StoreContextPair) {
  const auto Op = IROp->C<IR::IROp_StoreContextPair>();
  const auto OpSize = IROp->Size;

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    auto Src1 = GetZeroableReg(Op->Value1);
    auto Src2 = GetZeroableReg(Op->Value2);

@@ -135,11 +135,11 @@ DEF_OP(StoreContextPair) {
 DEF_OP(LoadRegister) {
  const auto Op = IROp->C<IR::IROp_LoadRegister>();

-  if (Op->Class == IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    LOGMAN_THROW_A_FMT(Op->Reg < StaticRegisters.size(), "out of range reg");

    mov(GetReg(Node).X(), StaticRegisters[Op->Reg].X());
-  } else if (Op->Class == IR::FPRClass) {
+  } else if (Op->Class == IR::RegClass::FPR) {
    const auto regSize = HostSupportsAVX256 ? IR::OpSize::i256Bit : IR::OpSize::i128Bit;
    LOGMAN_THROW_A_FMT(Op->Reg < StaticFPRegisters.size(), "out of range reg");
    LOGMAN_THROW_A_FMT(IROp->Size == regSize, "expected sized");
@@ -175,12 +175,13 @@ DEF_OP(LoadAF) {

 DEF_OP(StoreRegister) {
  const auto Op = IROp->C<IR::IROp_StoreRegister>();
-  auto Reg = IR::PhysicalRegister(Node);
+  const auto Reg = IR::PhysicalRegister(Node);
+  const auto RegClass = Reg.AsRegClass();

-  if (Reg.Class == IR::GPRFixedClass) {
+  if (RegClass == IR::RegClass::GPRFixed) {
    // Always use 64-bit, it's faster. Upper bits ignored for 32-bit mode.
    mov(ARMEmitter::Size::i64Bit, GetReg(Reg), GetReg(Op->Value));
-  } else if (Reg.Class == IR::FPRFixedClass) {
+  } else if (RegClass == IR::RegClass::FPRFixed) {
    const auto regSize = HostSupportsAVX256 ? IR::OpSize::i256Bit : IR::OpSize::i128Bit;
    LOGMAN_THROW_A_FMT(IROp->Size == regSize, "expected sized");

@@ -193,7 +194,7 @@ DEF_OP(StoreRegister) {
      mov(guest.Q(), host.Q());
    }
  } else {
-    LOGMAN_THROW_A_FMT(false, "Unhandled Op->Class {}", Reg.Class);
+    LOGMAN_THROW_A_FMT(false, "Unhandled Op->Class {}", RegClass);
  }
 }

@@ -225,7 +226,7 @@ DEF_OP(LoadContextIndexed) {

  const auto Index = GetReg(Op->Index);

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    switch (Op->Stride) {
    case 1:
    case 2:
@@ -288,7 +289,7 @@ DEF_OP(StoreContextIndexed) {

  const auto Index = GetReg(Op->Index);

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    const auto Value = GetReg(Op->Value);

    switch (Op->Stride) {
@@ -348,12 +349,31 @@ DEF_OP(StoreContextIndexed) {
  }
 }

+DEF_OP(FormContextAddress) {
+  const auto Op = IROp->C<IR::IROp_FormContextAddress>();
+  const auto Index = GetReg(Op->Index);
+  const auto Dst = GetReg(Node);
+
+  switch (Op->Stride) {
+  case 1:
+  case 2:
+  case 4:
+  case 8:
+  case 16:
+  case 32: {
+    add(ARMEmitter::Size::i64Bit, Dst, STATE, Index, ARMEmitter::ShiftType::LSL, FEXCore::ilog2(Op->Stride));
+    break;
+  }
+  default: LOGMAN_MSG_A_FMT("Unhandled FormContextAddress stride: {}", Op->Stride); break;
+  }
+}
+
 DEF_OP(SpillRegister) {
  const auto Op = IROp->C<IR::IROp_SpillRegister>();
  const auto OpSize = IROp->Size;
  const uint32_t SlotOffset = Op->Slot * MaxSpillSlotSize;

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    const auto Src = GetReg(Op->Value);
    switch (OpSize) {
    case IR::OpSize::i8Bit: {
@@ -394,7 +414,7 @@ DEF_OP(SpillRegister) {
    }
    default: LOGMAN_MSG_A_FMT("Unhandled SpillRegister size: {}", OpSize); break;
    }
-  } else if (Op->Class == FEXCore::IR::FPRClass) {
+  } else if (Op->Class == FEXCore::IR::RegClass::FPR) {
    const auto Src = GetVReg(Op->Value);

    switch (OpSize) {
@@ -433,7 +453,7 @@ DEF_OP(SpillRegister) {
    default: LOGMAN_MSG_A_FMT("Unhandled SpillRegister size: {}", OpSize); break;
    }
  } else {
-    LOGMAN_MSG_A_FMT("Unhandled SpillRegister class: {}", Op->Class.Val);
+    LOGMAN_MSG_A_FMT("Unhandled SpillRegister class: {}", Op->Class);
  }
 }

@@ -442,7 +462,7 @@ DEF_OP(FillRegister) {
  const auto OpSize = IROp->Size;
  const uint32_t SlotOffset = Op->Slot * MaxSpillSlotSize;

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    const auto Dst = GetReg(Node);
    switch (OpSize) {
    case IR::OpSize::i8Bit: {
@@ -483,7 +503,7 @@ DEF_OP(FillRegister) {
    }
    default: LOGMAN_MSG_A_FMT("Unhandled FillRegister size: {}", OpSize); break;
    }
-  } else if (Op->Class == FEXCore::IR::FPRClass) {
+  } else if (Op->Class == FEXCore::IR::RegClass::FPR) {
    const auto Dst = GetVReg(Node);

    switch (OpSize) {
@@ -522,7 +542,7 @@ DEF_OP(FillRegister) {
    default: LOGMAN_MSG_A_FMT("Unhandled FillRegister size: {}", OpSize); break;
    }
  } else {
-    LOGMAN_MSG_A_FMT("Unhandled FillRegister class: {}", Op->Class.Val);
+    LOGMAN_MSG_A_FMT("Unhandled FillRegister class: {}", Op->Class);
  }
 }

@@ -559,14 +579,14 @@ ARMEmitter::ExtendedMemOperand Arm64JITCore::GenerateMemOperand(
      return ARMEmitter::ExtendedMemOperand(Base.X(), ARMEmitter::IndexType::OFFSET, Const);
    } else {
      auto RegOffset = GetReg(Offset);
-      switch (OffsetType.Val) {
-      case IR::MEM_OFFSET_SXTX.Val:
+      switch (OffsetType) {
+      case IR::MemOffsetType::SXTX:
        return ARMEmitter::ExtendedMemOperand(Base.X(), RegOffset.X(), ARMEmitter::ExtendedType::SXTX, FEXCore::ilog2(OffsetScale));
-      case IR::MEM_OFFSET_UXTW.Val:
+      case IR::MemOffsetType::UXTW:
        return ARMEmitter::ExtendedMemOperand(Base.X(), RegOffset.X(), ARMEmitter::ExtendedType::UXTW, FEXCore::ilog2(OffsetScale));
-      case IR::MEM_OFFSET_SXTW.Val:
+      case IR::MemOffsetType::SXTW:
        return ARMEmitter::ExtendedMemOperand(Base.X(), RegOffset.X(), ARMEmitter::ExtendedType::SXTW, FEXCore::ilog2(OffsetScale));
-      default: LOGMAN_MSG_A_FMT("Unhandled GenerateMemOperand OffsetType: {}", OffsetType.Val); break;
+      default: LOGMAN_MSG_A_FMT("Unhandled GenerateMemOperand OffsetType: {}", OffsetType); break;
      }
    }
  }
@@ -593,20 +613,20 @@ ARMEmitter::Register Arm64JITCore::ApplyMemOperand(IR::OpSize AccessSize, ARMEmi
    add(ARMEmitter::Size::i64Bit, Tmp, Base, Tmp, ARMEmitter::ShiftType::LSL, FEXCore::ilog2(OffsetScale));
  } else {
    auto RegOffset = GetReg(Offset);
-    switch (OffsetType.Val) {
-    case IR::MEM_OFFSET_SXTX.Val:
+    switch (OffsetType) {
+    case IR::MemOffsetType::SXTX:
      add(ARMEmitter::Size::i64Bit, Tmp, Base, RegOffset, ARMEmitter::ExtendedType::SXTX, FEXCore::ilog2(OffsetScale));
      break;

-    case IR::MEM_OFFSET_UXTW.Val:
+    case IR::MemOffsetType::UXTW:
      add(ARMEmitter::Size::i64Bit, Tmp, Base, RegOffset, ARMEmitter::ExtendedType::UXTW, FEXCore::ilog2(OffsetScale));
      break;

-    case IR::MEM_OFFSET_SXTW.Val:
+    case IR::MemOffsetType::SXTW:
      add(ARMEmitter::Size::i64Bit, Tmp, Base, RegOffset, ARMEmitter::ExtendedType::SXTW, FEXCore::ilog2(OffsetScale));
      break;

-    default: LOGMAN_MSG_A_FMT("Unhandled OffsetType: {}", OffsetType.Val); break;
+    default: LOGMAN_MSG_A_FMT("Unhandled OffsetType: {}", OffsetType); break;
    }
  }
  return Tmp;
@@ -657,7 +677,7 @@ ARMEmitter::SVEMemOperand Arm64JITCore::GenerateSVEMemOperand(IR::OpSize AccessS
  // Note that we do nothing with the offset type and offset scale,
  // since SVE loads and stores don't have the ability to perform an
  // optional extension or shift as part of their behavior.
-  LOGMAN_THROW_A_FMT(OffsetType.Val == IR::MEM_OFFSET_SXTX.Val, "Currently only the default offset type (SXTX) is supported.");
+  LOGMAN_THROW_A_FMT(OffsetType == IR::MemOffsetType::SXTX, "Currently only the default offset type (SXTX) is supported.");

  const auto RegOffset = GetReg(Offset);
  return ARMEmitter::SVEMemOperand(Base.X(), RegOffset.X());
@@ -670,7 +690,7 @@ DEF_OP(LoadMem) {
  const auto MemReg = GetReg(Op->Addr);
  const auto MemSrc = GenerateMemOperand(OpSize, MemReg, Op->Offset, Op->OffsetType, Op->OffsetScale);

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    const auto Dst = GetReg(Node);

    switch (OpSize) {
@@ -704,7 +724,7 @@ DEF_OP(LoadMemPair) {
  const auto Op = IROp->C<IR::IROp_LoadMemPair>();
  const auto Addr = GetReg(Op->Addr);

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    const auto Dst1 = GetReg(Op->OutValue1);
    const auto Dst2 = GetReg(Op->OutValue2);

@@ -732,13 +752,13 @@ DEF_OP(LoadMemTSO) {

  const auto MemReg = GetReg(Op->Addr);

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    LOGMAN_THROW_A_FMT(Op->Offset.IsInvalid() || CTX->HostFeatures.SupportsTSOImm9, "unexpected offset");
    LOGMAN_THROW_A_FMT(Op->OffsetScale == 1, "unexpected offset scale");
-    LOGMAN_THROW_A_FMT(Op->OffsetType == IR::MEM_OFFSET_SXTX, "unexpected offset type");
+    LOGMAN_THROW_A_FMT(Op->OffsetType == IR::MemOffsetType::SXTX, "unexpected offset type");
  }

-  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == FEXCore::IR::GPRClass) {
+  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == IR::RegClass::GPR) {
    const auto Dst = GetReg(Node);
    uint64_t Offset = 0;
    if (!Op->Offset.IsInvalid()) {
@@ -760,7 +780,7 @@ DEF_OP(LoadMemTSO) {
      // Half-barrier once back-patched.
      nop();
    }
-  } else if (CTX->HostFeatures.SupportsRCPC && Op->Class == FEXCore::IR::GPRClass) {
+  } else if (CTX->HostFeatures.SupportsRCPC && Op->Class == IR::RegClass::GPR) {
    const auto Dst = GetReg(Node);
    if (OpSize == IR::OpSize::i8Bit) {
      // 8bit load is always aligned to natural alignment
@@ -775,7 +795,7 @@ DEF_OP(LoadMemTSO) {
      // Half-barrier once back-patched.
      nop();
    }
-  } else if (Op->Class == FEXCore::IR::GPRClass) {
+  } else if (Op->Class == IR::RegClass::GPR) {
    const auto Dst = GetReg(Node);
    if (OpSize == IR::OpSize::i8Bit) {
      // 8bit load is always aligned to natural alignment
@@ -892,7 +912,7 @@ DEF_OP(VLoadVectorMasked) {

      // If the sign bit is zero then skip the load
      ARMEmitter::ForwardLabel Skip {};
-      tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
+      (void)tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
      // Do the gather load for this element into the destination
      switch (IROp->ElementSize) {
      case IR::OpSize::i8Bit: ld1<ARMEmitter::SubRegSize::i8Bit>(TempDst.Q(), i, TempMemReg); break;
@@ -903,7 +923,7 @@ DEF_OP(VLoadVectorMasked) {
      default: LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, IROp->ElementSize); return;
      }

-      Bind(&Skip);
+      (void)Bind(&Skip);

      if ((i + 1) != NumElements) {
        // Handle register rename to save a move.
@@ -993,7 +1013,7 @@ DEF_OP(VStoreVectorMasked) {

      // If the sign bit is zero then skip the load
      ARMEmitter::ForwardLabel Skip {};
-      tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
+      (void)tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
      // Do the gather load for this element into the destination
      switch (IROp->ElementSize) {
      case IR::OpSize::i8Bit: st1<ARMEmitter::SubRegSize::i8Bit>(RegData.Q(), i, TempMemReg); break;
@@ -1004,7 +1024,7 @@ DEF_OP(VStoreVectorMasked) {
      default: LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, IROp->ElementSize); return;
      }

-      Bind(&Skip);
+      (void)Bind(&Skip);

      if ((i + 1) != NumElements) {
        // Handle register rename to save a move.
@@ -1020,7 +1040,7 @@ void Arm64JITCore::Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize,
                                       ARMEmitter::VRegister IncomingDst, std::optional<ARMEmitter::Register> BaseAddr,
                                       ARMEmitter::VRegister VectorIndexLow, std::optional<ARMEmitter::VRegister> VectorIndexHigh,
                                       ARMEmitter::VRegister MaskReg, IR::OpSize VectorIndexSize, size_t DataElementOffsetStart,
-                                       size_t IndexElementOffsetStart, uint8_t OffsetScale) {
+                                       size_t IndexElementOffsetStart, uint8_t OffsetScale, IR::OpSize AddrSize) {
  LOGMAN_THROW_A_FMT(ElementSize >= IR::OpSize::i8Bit && ElementSize <= IR::OpSize::i64Bit, "Invalid element size");

  const auto PerformSMove = [this](IR::OpSize ElementSize, const ARMEmitter::Register Dst, const ARMEmitter::VRegister Vector, int index) {
@@ -1082,7 +1102,7 @@ void Arm64JITCore::Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize,
    PerformMove(ElementSize, WorkingReg, MaskReg, i);

    // Skip if the mask's sign bit isn't set
-    tbz(WorkingReg, ElementSizeInBits - 1, &Skip);
+    (void)tbz(WorkingReg, ElementSizeInBits - 1, &Skip);

    // Extract Index Element
    if ((IndexElement * IR::OpSizeToSize(VectorIndexSize)) >= 16) {
@@ -1096,17 +1116,17 @@ void Arm64JITCore::Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize,
    // Calculate memory position for this gather load
    if (BaseAddr.has_value()) {
      if (VectorIndexSize == IR::OpSize::i32Bit) {
-        add(ARMEmitter::Size::i64Bit, TempMemReg, *BaseAddr, WorkingReg, ARMEmitter::ExtendedType::SXTW, FEXCore::ilog2(OffsetScale));
+        add(ConvertSize(AddrSize), TempMemReg, *BaseAddr, WorkingReg, ARMEmitter::ExtendedType::SXTW, FEXCore::ilog2(OffsetScale));
      } else {
-        add(ARMEmitter::Size::i64Bit, TempMemReg, *BaseAddr, WorkingReg, ARMEmitter::ShiftType::LSL, FEXCore::ilog2(OffsetScale));
+        add(ConvertSize(AddrSize), TempMemReg, *BaseAddr, WorkingReg, ARMEmitter::ShiftType::LSL, FEXCore::ilog2(OffsetScale));
      }
    } else {
      ///< In this case we have no base address, All addresses come from the vector register itself
      if (VectorIndexSize == IR::OpSize::i32Bit) {
        // Sign extend and shift in to the 64-bit register
-        sbfiz(ARMEmitter::Size::i64Bit, TempMemReg, WorkingReg, FEXCore::ilog2(OffsetScale), 32);
+        sbfiz(ConvertSize(AddrSize), TempMemReg, WorkingReg, FEXCore::ilog2(OffsetScale), 32);
      } else {
-        lsl(ARMEmitter::Size::i64Bit, TempMemReg, WorkingReg, FEXCore::ilog2(OffsetScale));
+        lsl(ConvertSize(AddrSize), TempMemReg, WorkingReg, FEXCore::ilog2(OffsetScale));
      }
    }

@@ -1120,7 +1140,7 @@ void Arm64JITCore::Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize,
    default: LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, ElementSize); FEX_UNREACHABLE;
    }

-    Bind(&Skip);
+    (void)Bind(&Skip);
  }

  if (NeedsDestTmp) {
@@ -1164,7 +1184,8 @@ DEF_OP(VLoadVectorGatherMasked) {

  ///< If the host supports SVE and the offset scale matches SVE limitations then it can do an SVE style load.
  const bool SupportsSVELoad = (HostSupportsSVE128 || HostSupportsSVE256) &&
-                               (OffsetScale == 1 || OffsetScale == IR::OpSizeToSize(VectorIndexSize)) && VectorIndexSize == IROp->ElementSize;
+                               (OffsetScale == 1 || OffsetScale == IR::OpSizeToSize(VectorIndexSize)) &&
+                               VectorIndexSize == IROp->ElementSize && Op->AddrSize == IR::OpSize::i64Bit;

  if (SupportsSVELoad) {
    uint8_t SVEScale = FEXCore::ilog2(OffsetScale);
@@ -1222,7 +1243,7 @@ DEF_OP(VLoadVectorGatherMasked) {
  } else {
    LOGMAN_THROW_A_FMT(!Is256Bit, "Can't emulate this gather load in the backend! Programming error!");
    Emulate128BitGather(IROp->Size, IROp->ElementSize, Dst, IncomingDst, BaseAddr, VectorIndexLow, VectorIndexHigh, MaskReg,
-                        VectorIndexSize, DataElementOffsetStart, IndexElementOffsetStart, OffsetScale);
+                        VectorIndexSize, DataElementOffsetStart, IndexElementOffsetStart, OffsetScale, Op->AddrSize);
  }
 }

@@ -1247,7 +1268,9 @@ DEF_OP(VLoadVectorGatherMaskedQPS) {
    !Op->VectorIndexHigh.IsInvalid() ? std::make_optional(GetVReg(Op->VectorIndexHigh)) : std::nullopt;

  ///< If the host supports SVE and the offset scale matches SVE limitations then it can do an SVE style load.
-  if (HostSupportsSVE128 && (OffsetScale == 1 || OffsetScale == 4)) {
+  const bool SupportsSVELoad = HostSupportsSVE128 && (OffsetScale == 1 || OffsetScale == 4) && Op->AddrSize == IR::OpSize::i64Bit;
+
+  if (SupportsSVELoad) {
    ARMEmitter::SVEModType ModType = ARMEmitter::SVEModType::MOD_NONE;
    if (OffsetScale != 1) {
      ModType = ARMEmitter::SVEModType::MOD_LSL;
@@ -1301,7 +1324,7 @@ DEF_OP(VLoadVectorGatherMaskedQPS) {
    }
  } else {
    Emulate128BitGather(IR::OpSize::i128Bit, IR::OpSize::i32Bit, Dst, IncomingDst, BaseAddr, VectorIndexLow, VectorIndexHigh, MaskReg,
-                        IR::OpSize::i64Bit, 0, 0, OffsetScale);
+                        IR::OpSize::i64Bit, 0, 0, OffsetScale, Op->AddrSize);
  }
 }

@@ -1602,7 +1625,7 @@ DEF_OP(StoreMem) {
  const auto MemReg = GetReg(Op->Addr);
  const auto MemSrc = GenerateMemOperand(OpSize, MemReg, Op->Offset, Op->OffsetType, Op->OffsetScale);

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    const auto Src = GetZeroableReg(Op->Value);
    switch (OpSize) {
    case IR::OpSize::i8Bit: strb(Src, MemSrc); break;
@@ -1713,7 +1736,7 @@ DEF_OP(StoreMemPair) {
  const auto OpSize = IROp->Size;
  const auto Addr = GetReg(Op->Addr);

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    const auto Src1 = GetZeroableReg(Op->Value1);
    const auto Src2 = GetZeroableReg(Op->Value2);
    switch (OpSize) {
@@ -1740,13 +1763,13 @@ DEF_OP(StoreMemTSO) {

  const auto MemReg = GetReg(Op->Addr);

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    LOGMAN_THROW_A_FMT(Op->Offset.IsInvalid() || CTX->HostFeatures.SupportsTSOImm9, "unexpected offset");
    LOGMAN_THROW_A_FMT(Op->OffsetScale == 1, "unexpected offset scale");
-    LOGMAN_THROW_A_FMT(Op->OffsetType == IR::MEM_OFFSET_SXTX, "unexpected offset type");
+    LOGMAN_THROW_A_FMT(Op->OffsetType == IR::MemOffsetType::SXTX, "unexpected offset type");
  }

-  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == FEXCore::IR::GPRClass) {
+  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == IR::RegClass::GPR) {
    const auto Src = GetZeroableReg(Op->Value);
    uint64_t Offset = 0;
    if (!Op->Offset.IsInvalid()) {
@@ -1767,7 +1790,7 @@ DEF_OP(StoreMemTSO) {
      default: LOGMAN_MSG_A_FMT("Unhandled StoreMemTSO size: {}", OpSize); break;
      }
    }
-  } else if (Op->Class == FEXCore::IR::GPRClass) {
+  } else if (Op->Class == IR::RegClass::GPR) {
    const auto Src = GetZeroableReg(Op->Value);

    if (OpSize == IR::OpSize::i8Bit) {
@@ -1851,7 +1874,7 @@ DEF_OP(MemSet) {

  if (!DirectionIsInline) {
    // Backward or forwards implementation depends on flag
-    tbnz(DirectionReg, 1, &BackwardImpl);
+    (void)tbnz(DirectionReg, 1, &BackwardImpl);
  }

  auto MemStore = [this](auto Value, uint32_t OpSize, int32_t Size) {
@@ -1899,7 +1922,7 @@ DEF_OP(MemSet) {
    ARMEmitter::ForwardLabel DoneInternal {};

    // Early exit if zero count.
-    cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+    (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);

    if (!IsAtomic) {
      ARMEmitter::ForwardLabel AgainInternal256Exit {};
@@ -1916,50 +1939,50 @@ DEF_OP(MemSet) {
      // Do this in two parts, to fallback to the byte by byte loop if size < 32, and to the
      // single copy loop if size < 64.
      sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      tbnz(TMP1, 63, &AgainInternal128Exit);
+      (void)tbnz(TMP1, 63, &AgainInternal128Exit);

      // Fill VTMP2 with the set pattern
      dup(SubRegSize, VTMP2.Q(), Value);

      sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      tbnz(TMP1, 63, &AgainInternal256Exit);
+      (void)tbnz(TMP1, 63, &AgainInternal256Exit);

-      Bind(&AgainInternal256);
+      (void)Bind(&AgainInternal256);
      stp<ARMEmitter::IndexType::POST>(VTMP2.Q(), VTMP2.Q(), TMP2, 32 * Direction);
      stp<ARMEmitter::IndexType::POST>(VTMP2.Q(), VTMP2.Q(), TMP2, 32 * Direction);
      sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 64 / Size);
-      tbz(TMP1, 63, &AgainInternal256);
+      (void)tbz(TMP1, 63, &AgainInternal256);

-      Bind(&AgainInternal256Exit);
+      (void)Bind(&AgainInternal256Exit);
      add(ARMEmitter::Size::i64Bit, TMP1, TMP1, 64 / Size);
-      cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);

      sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      tbnz(TMP1, 63, &AgainInternal128Exit);
-      Bind(&AgainInternal128);
+      (void)tbnz(TMP1, 63, &AgainInternal128Exit);
+      (void)Bind(&AgainInternal128);
      stp<ARMEmitter::IndexType::POST>(VTMP2.Q(), VTMP2.Q(), TMP2, 32 * Direction);
      sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      tbz(TMP1, 63, &AgainInternal128);
+      (void)tbz(TMP1, 63, &AgainInternal128);

-      Bind(&AgainInternal128Exit);
+      (void)Bind(&AgainInternal128Exit);
      add(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);

      if (Direction == -1) {
        add(ARMEmitter::Size::i64Bit, TMP2, TMP2, 32 - Size);
      }
    }

-    Bind(&AgainInternal);
+    (void)Bind(&AgainInternal);
    if (IsAtomic) {
      MemStoreTSO(Value, OpSize, SizeDirection);
    } else {
      MemStore(Value, OpSize, SizeDirection);
    }
    sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 1);
-    cbnz(ARMEmitter::Size::i64Bit, TMP1, &AgainInternal);
+    (void)cbnz(ARMEmitter::Size::i64Bit, TMP1, &AgainInternal);

-    Bind(&DoneInternal);
+    (void)Bind(&DoneInternal);

    if (SizeDirection >= 0) {
      switch (OpSize) {
@@ -1989,12 +2012,12 @@ DEF_OP(MemSet) {
      EmitMemset(Direction);

      if (Direction == 1) {
-        b(&Done);
-        Bind(&BackwardImpl);
+        (void)b(&Done);
+        (void)Bind(&BackwardImpl);
      }
    }

-    Bind(&Done);
+    (void)Bind(&Done);
    // Destination already set to the final pointer.
  }
 }
@@ -2044,7 +2067,7 @@ DEF_OP(MemCpy) {

  if (!DirectionIsInline) {
    // Backward or forwards implementation depends on flag
-    tbnz(DirectionReg, 1, &BackwardImpl);
+    (void)tbnz(DirectionReg, 1, &BackwardImpl);
  }

  auto MemCpy = [this](uint32_t OpSize, int32_t Size) {
@@ -2141,7 +2164,7 @@ DEF_OP(MemCpy) {
    ARMEmitter::ForwardLabel DoneInternal {};

    // Early exit if zero count.
-    cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+    (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);

    if (!IsAtomic) {
      ARMEmitter::ForwardLabel AbsPos {};
@@ -2151,11 +2174,11 @@ DEF_OP(MemCpy) {
      ARMEmitter::BackwardLabel AgainInternal256 {};

      sub(ARMEmitter::Size::i64Bit, TMP4, TMP2, TMP3);
-      tbz(TMP4, 63, &AbsPos);
+      (void)tbz(TMP4, 63, &AbsPos);
      neg(ARMEmitter::Size::i64Bit, TMP4, TMP4);
-      Bind(&AbsPos);
+      (void)Bind(&AbsPos);
      sub(ARMEmitter::Size::i64Bit, TMP4, TMP4, 32);
-      tbnz(TMP4, 63, &AgainInternal);
+      (void)tbnz(TMP4, 63, &AgainInternal);

      if (Direction == -1) {
        sub(ARMEmitter::Size::i64Bit, TMP2, TMP2, 32 - Size);
@@ -2167,30 +2190,30 @@ DEF_OP(MemCpy) {
      // Do this in two parts, to fallback to the byte by byte loop if size < 32, and to the
      // single copy loop if size < 64.
      sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      tbnz(TMP1, 63, &AgainInternal128Exit);
+      (void)tbnz(TMP1, 63, &AgainInternal128Exit);
      sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      tbnz(TMP1, 63, &AgainInternal256Exit);
+      (void)tbnz(TMP1, 63, &AgainInternal256Exit);

-      Bind(&AgainInternal256);
+      (void)Bind(&AgainInternal256);
      MemCpy(32, 32 * Direction);
      MemCpy(32, 32 * Direction);
      sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 64 / Size);
-      tbz(TMP1, 63, &AgainInternal256);
+      (void)tbz(TMP1, 63, &AgainInternal256);

-      Bind(&AgainInternal256Exit);
+      (void)Bind(&AgainInternal256Exit);
      add(ARMEmitter::Size::i64Bit, TMP1, TMP1, 64 / Size);
-      cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);

      sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      tbnz(TMP1, 63, &AgainInternal128Exit);
-      Bind(&AgainInternal128);
+      (void)tbnz(TMP1, 63, &AgainInternal128Exit);
+      (void)Bind(&AgainInternal128);
      MemCpy(32, 32 * Direction);
      sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      tbz(TMP1, 63, &AgainInternal128);
+      (void)tbz(TMP1, 63, &AgainInternal128);

-      Bind(&AgainInternal128Exit);
+      (void)Bind(&AgainInternal128Exit);
      add(ARMEmitter::Size::i64Bit, TMP1, TMP1, 32 / Size);
-      cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);
+      (void)cbz(ARMEmitter::Size::i64Bit, TMP1, &DoneInternal);

      if (Direction == -1) {
        add(ARMEmitter::Size::i64Bit, TMP2, TMP2, 32 - Size);
@@ -2198,16 +2221,16 @@ DEF_OP(MemCpy) {
      }
    }

-    Bind(&AgainInternal);
+    (void)Bind(&AgainInternal);
    if (IsAtomic) {
      MemCpyTSO(OpSize, SizeDirection);
    } else {
      MemCpy(OpSize, SizeDirection);
    }
    sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 1);
-    cbnz(ARMEmitter::Size::i64Bit, TMP1, &AgainInternal);
+    (void)cbnz(ARMEmitter::Size::i64Bit, TMP1, &AgainInternal);

-    Bind(&DoneInternal);
+    (void)Bind(&DoneInternal);

    // Needs to use temporaries just in case of overwrite
    mov(TMP1, MemRegDest.X());
@@ -2265,186 +2288,15 @@ DEF_OP(MemCpy) {
    for (int32_t Direction : {1, -1}) {
      EmitMemcpy(Direction);
      if (Direction == 1) {
-        b(&Done);
-        Bind(&BackwardImpl);
+        (void)b(&Done);
+        (void)Bind(&BackwardImpl);
      }
    }
-    Bind(&Done);
+    (void)Bind(&Done);
    // Destination already set to the final pointer.
  }
 }

-DEF_OP(ParanoidLoadMemTSO) {
-  const auto Op = IROp->C<IR::IROp_LoadMemTSO>();
-  const auto OpSize = IROp->Size;
-
-  auto MemReg = GetReg(Op->Addr);
-
-  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == FEXCore::IR::GPRClass) {
-    const auto Dst = GetReg(Node);
-    uint64_t Offset = 0;
-    if (!Op->Offset.IsInvalid()) {
-      if (!IsInlineConstant(Op->Offset, &Offset)) {
-        MemReg = ApplyMemOperand(OpSize, MemReg, TMP4, Op->Offset, Op->OffsetType, Op->OffsetScale);
-      }
-    }
-
-    if (OpSize == IR::OpSize::i8Bit) {
-      // 8bit load is always aligned to natural alignment
-      const auto Dst = GetReg(Node);
-      ldapurb(Dst, MemReg, Offset);
-    } else {
-      switch (OpSize) {
-      case IR::OpSize::i16Bit: ldapurh(Dst, MemReg, Offset); break;
-      case IR::OpSize::i32Bit: ldapur(Dst.W(), MemReg, Offset); break;
-      case IR::OpSize::i64Bit: ldapur(Dst.X(), MemReg, Offset); break;
-      default: LOGMAN_MSG_A_FMT("Unhandled ParanoidLoadMemTSO size: {}", OpSize); break;
-      }
-    }
-  } else if (CTX->HostFeatures.SupportsRCPC && Op->Class == FEXCore::IR::GPRClass) {
-    const auto Dst = GetReg(Node);
-    MemReg = ApplyMemOperand(OpSize, MemReg, TMP4, Op->Offset, Op->OffsetType, Op->OffsetScale);
-    if (OpSize == IR::OpSize::i8Bit) {
-      // 8bit load is always aligned to natural alignment
-      ldaprb(Dst.W(), MemReg);
-    } else {
-      switch (OpSize) {
-      case IR::OpSize::i16Bit: ldaprh(Dst.W(), MemReg); break;
-      case IR::OpSize::i32Bit: ldapr(Dst.W(), MemReg); break;
-      case IR::OpSize::i64Bit: ldapr(Dst.X(), MemReg); break;
-      default: LOGMAN_MSG_A_FMT("Unhandled ParanoidLoadMemTSO size: {}", OpSize); break;
-      }
-    }
-  } else if (Op->Class == FEXCore::IR::GPRClass) {
-    const auto Dst = GetReg(Node);
-    MemReg = ApplyMemOperand(OpSize, MemReg, TMP4, Op->Offset, Op->OffsetType, Op->OffsetScale);
-    switch (OpSize) {
-    case IR::OpSize::i8Bit: ldarb(Dst, MemReg); break;
-    case IR::OpSize::i16Bit: ldarh(Dst, MemReg); break;
-    case IR::OpSize::i32Bit: ldar(Dst.W(), MemReg); break;
-    case IR::OpSize::i64Bit: ldar(Dst.X(), MemReg); break;
-    default: LOGMAN_MSG_A_FMT("Unhandled ParanoidLoadMemTSO size: {}", OpSize); break;
-    }
-  } else {
-    const auto Dst = GetVReg(Node);
-    MemReg = ApplyMemOperand(OpSize, MemReg, TMP4, Op->Offset, Op->OffsetType, Op->OffsetScale);
-    switch (OpSize) {
-    case IR::OpSize::i8Bit:
-      ldarb(TMP1, MemReg);
-      fmov(ARMEmitter::Size::i32Bit, Dst.S(), TMP1.W());
-      break;
-    case IR::OpSize::i16Bit:
-      ldarh(TMP1, MemReg);
-      fmov(ARMEmitter::Size::i32Bit, Dst.S(), TMP1.W());
-      break;
-    case IR::OpSize::i32Bit:
-      ldar(TMP1.W(), MemReg);
-      fmov(ARMEmitter::Size::i32Bit, Dst.S(), TMP1.W());
-      break;
-    case IR::OpSize::i64Bit:
-      ldar(TMP1, MemReg);
-      fmov(ARMEmitter::Size::i64Bit, Dst.D(), TMP1);
-      break;
-    case IR::OpSize::i128Bit:
-      ldaxp(ARMEmitter::Size::i64Bit, TMP1, TMP2, MemReg);
-      clrex();
-      ins(ARMEmitter::SubRegSize::i64Bit, Dst, 0, TMP1);
-      ins(ARMEmitter::SubRegSize::i64Bit, Dst, 1, TMP2);
-      break;
-    case IR::OpSize::i256Bit:
-      LOGMAN_THROW_A_FMT(HostSupportsSVE256, "Need SVE256 support in order to use {} with 256-bit operation", __func__);
-      dmb(ARMEmitter::BarrierScope::ISH);
-      ld1b<ARMEmitter::SubRegSize::i8Bit>(Dst.Z(), PRED_TMP_32B.Zeroing(), MemReg);
-      dmb(ARMEmitter::BarrierScope::ISH);
-      break;
-    default: LOGMAN_MSG_A_FMT("Unhandled ParanoidLoadMemTSO size: {}", OpSize); break;
-    }
-  }
-}
-
-DEF_OP(ParanoidStoreMemTSO) {
-  const auto Op = IROp->C<IR::IROp_StoreMemTSO>();
-  const auto OpSize = IROp->Size;
-
-  auto MemReg = GetReg(Op->Addr);
-
-  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == FEXCore::IR::GPRClass) {
-    const auto Src = GetZeroableReg(Op->Value);
-    uint64_t Offset = 0;
-    if (!Op->Offset.IsInvalid()) {
-      if (!IsInlineConstant(Op->Offset, &Offset)) {
-        MemReg = ApplyMemOperand(OpSize, MemReg, TMP1, Op->Offset, Op->OffsetType, Op->OffsetScale);
-      }
-    }
-
-    if (OpSize == IR::OpSize::i8Bit) {
-      // 8bit load is always aligned to natural alignment
-      stlurb(Src, MemReg, Offset);
-    } else {
-      switch (OpSize) {
-      case IR::OpSize::i16Bit: stlurh(Src, MemReg, Offset); break;
-      case IR::OpSize::i32Bit: stlur(Src.W(), MemReg, Offset); break;
-      case IR::OpSize::i64Bit: stlur(Src.X(), MemReg, Offset); break;
-      default: LOGMAN_MSG_A_FMT("Unhandled ParanoidStoreMemTSO size: {}", OpSize); break;
-      }
-    }
-  } else if (Op->Class == FEXCore::IR::GPRClass) {
-    const auto Src = GetZeroableReg(Op->Value);
-    MemReg = ApplyMemOperand(OpSize, MemReg, TMP1, Op->Offset, Op->OffsetType, Op->OffsetScale);
-    switch (OpSize) {
-    case IR::OpSize::i8Bit: stlrb(Src, MemReg); break;
-    case IR::OpSize::i16Bit: stlrh(Src, MemReg); break;
-    case IR::OpSize::i32Bit: stlr(Src.W(), MemReg); break;
-    case IR::OpSize::i64Bit: stlr(Src.X(), MemReg); break;
-    default: LOGMAN_MSG_A_FMT("Unhandled ParanoidStoreMemTSO size: {}", OpSize); break;
-    }
-  } else {
-    const auto Src = GetVReg(Op->Value);
-
-    MemReg = ApplyMemOperand(OpSize, MemReg, TMP4, Op->Offset, Op->OffsetType, Op->OffsetScale);
-
-    switch (OpSize) {
-    case IR::OpSize::i8Bit:
-      umov<ARMEmitter::SubRegSize::i8Bit>(TMP1, Src, 0);
-      stlrb(TMP1, MemReg);
-      break;
-    case IR::OpSize::i16Bit:
-      umov<ARMEmitter::SubRegSize::i16Bit>(TMP1, Src, 0);
-      stlrh(TMP1, MemReg);
-      break;
-    case IR::OpSize::i32Bit:
-      umov<ARMEmitter::SubRegSize::i32Bit>(TMP1, Src, 0);
-      stlr(TMP1.W(), MemReg);
-      break;
-    case IR::OpSize::i64Bit:
-      umov<ARMEmitter::SubRegSize::i64Bit>(TMP1, Src, 0);
-      stlr(TMP1, MemReg);
-      break;
-    case IR::OpSize::i128Bit: {
-      // Move vector to GPRs
-      umov<ARMEmitter::SubRegSize::i64Bit>(TMP1, Src, 0);
-      umov<ARMEmitter::SubRegSize::i64Bit>(TMP2, Src, 1);
-      ARMEmitter::BackwardLabel B;
-      Bind(&B);
-
-      // ldaxp must not have both the destination registers be the same
-      ldaxp(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::zr, TMP3, MemReg); // <- Can hit SIGBUS. Overwritten with DMB
-      stlxp(ARMEmitter::Size::i64Bit, TMP3, TMP1, TMP2, MemReg);          // <- Can also hit SIGBUS
-      cbnz(ARMEmitter::Size::i64Bit, TMP3, &B);                           // < Overwritten with DMB
-      break;
-    }
-    case IR::OpSize::i256Bit: {
-      LOGMAN_THROW_A_FMT(HostSupportsSVE256, "Need SVE256 support in order to use {} with 256-bit operation", __func__);
-      dmb(ARMEmitter::BarrierScope::ISH);
-      st1b<ARMEmitter::SubRegSize::i8Bit>(Src.Z(), PRED_TMP_32B, MemReg, 0);
-      dmb(ARMEmitter::BarrierScope::ISH);
-      break;
-    }
-    default: LOGMAN_MSG_A_FMT("Unhandled ParanoidStoreMemTSO size: {}", OpSize); break;
-    }
-  }
-}
-
 DEF_OP(CacheLineClear) {
  if (!CTX->HostFeatures.SupportsCacheMaintenanceOps) {
    dmb(ARMEmitter::BarrierScope::SY);
@@ -10,10 +10,12 @@ $end_info$
 #endif

 #include "Interface/Context/Context.h"
+#include "Interface/Core/JIT/DebugData.h"
 #include "Interface/Core/JIT/JITClass.h"
-#include "FEXCore/Debug/InternalThreadState.h"

 #include <FEXCore/Core/SignalDelegator.h>
+#include <FEXCore/Debug/InternalThreadState.h>
+#include <FEXCore/Utils/EnumUtils.h>

 namespace FEXCore::CPU {

@@ -46,10 +48,10 @@ DEF_OP(GuestOpcode) {
 DEF_OP(Fence) {
  auto Op = IROp->C<IR::IROp_Fence>();
  switch (Op->Fence) {
-  case IR::Fence_Load.Val: dmb(ARMEmitter::BarrierScope::LD); break;
-  case IR::Fence_LoadStore.Val: dmb(ARMEmitter::BarrierScope::SY); break;
-  case IR::Fence_Store.Val: dmb(ARMEmitter::BarrierScope::ST); break;
-  case IR::Fence_Inst.Val: isb(); break;
+  case IR::FenceType::Load: dmb(ARMEmitter::BarrierScope::LD); break;
+  case IR::FenceType::LoadStore: dmb(ARMEmitter::BarrierScope::SY); break;
+  case IR::FenceType::Store: dmb(ARMEmitter::BarrierScope::ST); break;
+  case IR::FenceType::Inst: isb(); break;
  default: LOGMAN_MSG_A_FMT("Unknown Fence: {}", Op->Fence); break;
  }
 }
@@ -76,19 +78,19 @@ DEF_OP(Break) {

  switch (Op->Reason.Signal) {
  case Core::FAULT_SIGILL:
-    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGILL));
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.GuestSignal_SIGILL));
    br(TMP1);
    break;
  case Core::FAULT_SIGTRAP:
-    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP));
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.GuestSignal_SIGTRAP));
    br(TMP1);
    break;
  case Core::FAULT_SIGSEGV:
-    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGSEGV));
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.GuestSignal_SIGSEGV));
    br(TMP1);
    break;
  default:
-    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP));
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.GuestSignal_SIGTRAP));
    br(TMP1);
    break;
  }
@@ -106,10 +108,10 @@ DEF_OP(GetRoundingMode) {
  // zero. Just swapping 01 and 10. That's a bitfield reverse. Round mode is in
  // bottom two bits. After reversing as a 32-bit operation, it'll be in [31:30]
  // and ripe for reinsertion back at 0.
-  static_assert(IR::ROUND_MODE_NEAREST == 0);
-  static_assert(IR::ROUND_MODE_NEGATIVE_INFINITY == 1);
-  static_assert(IR::ROUND_MODE_POSITIVE_INFINITY == 2);
-  static_assert(IR::ROUND_MODE_TOWARDS_ZERO == 3);
+  static_assert(FEXCore::ToUnderlying(IR::RoundMode::Nearest) == 0);
+  static_assert(FEXCore::ToUnderlying(IR::RoundMode::NegInfinity) == 1);
+  static_assert(FEXCore::ToUnderlying(IR::RoundMode::PosInfinity) == 2);
+  static_assert(FEXCore::ToUnderlying(IR::RoundMode::TowardsZero) == 3);

  rbit(ARMEmitter::Size::i32Bit, TMP1, Dst);
  bfi(ARMEmitter::Size::i64Bit, Dst, TMP1, 30, 2);
@@ -187,11 +189,11 @@ DEF_OP(Print) {

  if (IsGPR(Op->Value)) {
    mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, GetReg(Op->Value));
-    ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.PrintValue));
+    ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.PrintValue));
  } else {
    fmov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, GetVReg(Op->Value), false);
    fmov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, GetVReg(Op->Value), true);
-    ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.PrintVectorValue));
+    ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.PrintVectorValue));
  }

  if (!CTX->Config.DisableVixlIndirectCalls) [[unlikely]] {
@@ -239,7 +241,7 @@ DEF_OP(ProcessorID) {
  sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, 16);

  // Load the getcpu syscall number
-#if defined(_M_X86_64)
+#if defined(ARCHITECTURE_x86_64)
  // Just to ensure the syscall number doesn't change if compiled for an x86_64 host.
  constexpr auto GetCPUSyscallNum = 0xa8;
 #else
@@ -303,20 +305,20 @@ DEF_OP(MonoBackpatcherWrite) {
    mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, TMP4);
  }

-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
  ldr(TMP2, ARMEmitter::XReg::x18, TEB_CPU_AREA_OFFSET);
  LoadConstant(ARMEmitter::Size::i32Bit, TMP1, 1);
  strb(TMP1.W(), TMP2, CPU_AREA_IN_SYSCALL_CALLBACK_OFFSET);
 #endif

-  ldr(ARMEmitter::XReg::x4, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.MonoBackpatcherWrite));
+  ldr(ARMEmitter::XReg::x4, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.MonoBackpatcherWrite));
  if (!CTX->Config.DisableVixlIndirectCalls) [[unlikely]] {
    GenerateIndirectRuntimeCall<void, void*, uint8_t, uint64_t, uint64_t>(ARMEmitter::Reg::r4);
  } else {
    blr(ARMEmitter::Reg::r4);
  }

-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
  ldr(TMP2, ARMEmitter::XReg::x18, TEB_CPU_AREA_OFFSET);
  strb(ARMEmitter::WReg::zr, TMP2, CPU_AREA_IN_SYSCALL_CALLBACK_OFFSET);
 #endif
@@ -18,18 +18,4 @@ DEF_OP(RMWHandle) {
  mov(ARMEmitter::Size::i64Bit, GetReg(Node), GetReg(IROp->Args[0]));
 }

-DEF_OP(Swap1) {
-  auto Op = IROp->C<IR::IROp_Swap1>();
-  auto A = GetReg(Op->A), B = GetReg(Op->B);
-  LOGMAN_THROW_A_FMT(B == GetReg(Node), "Invariant");
-
-  mov(ARMEmitter::Size::i64Bit, TMP1, A);
-  mov(ARMEmitter::Size::i64Bit, A, B);
-  mov(ARMEmitter::Size::i64Bit, B, TMP1);
-}
-
-DEF_OP(Swap2) {
-  // Implemented above
-}
-
 } // namespace FEXCore::CPU
@@ -1,79 +1,100 @@
 // SPDX-License-Identifier: MIT
 #pragma once
 #include <FEXCore/IR/IR.h>
+#include <FEXCore/Utils/CompilerDefs.h>
+
+namespace FEXCore::Context {
+class ContextImpl;
+}

 namespace FEXCore::CPU {
-enum class RelocationTypes : uint8_t {
+enum class RelocationTypes : uint32_t {
  // 8 byte literal in memory for symbol
  // Aligned to struct RelocNamedSymbolLiteral
  RELOC_NAMED_SYMBOL_LITERAL,

  // Fixed size named thunk move
-  // 4 instruction constant generation on AArch64
-  // 64-bit mov on x86-64
+  // 4 instruction constant generation
  // Aligned to struct RelocNamedThunkMove
  RELOC_NAMED_THUNK_MOVE,

+  // 8 byte literal (relative to binary base address)
+  RELOC_GUEST_RIP_LITERAL,
+
  // Fixed size guest RIP move
-  // 4 instruction constant generation on AArch64
-  // 64-bit mov on x86-64
-  // Aligned to struct RelocGuestRIPMove
+  // 4 instruction constant generation
+  // Aligned to struct RelocGuestRIP
  RELOC_GUEST_RIP_MOVE,
 };

-struct RelocationTypeHeader final {
+struct FEX_PACKED RelocationHeader final {
+  // Offset to the relocated host code data
+  uint64_t Offset {};
+
  RelocationTypes Type;
 };

 struct RelocNamedSymbolLiteral final {
-  enum class NamedSymbol : uint8_t {
+  enum class NamedSymbol : uint32_t {
    ///< Thread specific relocations
    // JIT Literal pointers
    SYMBOL_LITERAL_EXITFUNCTION_LINKER,
  };

-  RelocationTypeHeader Header {};
+  RelocationHeader Header {};

  NamedSymbol Symbol;

-  // Offset in to the code section to begin the relocation
-  uint64_t Offset {};
+  uint32_t Pad[8];
 };

 struct RelocNamedThunkMove final {
-  RelocationTypeHeader Header {};
+  RelocationHeader Header {};

  // GPR index the constant is being moved to
-  uint8_t RegisterIndex;
+  uint32_t RegisterIndex;

  // The thunk SHA256 hash
  IR::SHA256Sum Symbol;
-
-  // Offset in to the code section to begin the relocation
-  uint64_t Offset {};
 };

-struct RelocGuestRIPMove final {
-  RelocationTypeHeader Header {};
+struct RelocGuestRIP final {
+  RelocationHeader Header {};

-  // GPR index the constant is being moved to
+  // GPR index the constant is being moved to (for non-literal relocations)
  uint8_t RegisterIndex;

-  // Offset in to the code section to begin the relocation
-  uint64_t Offset {};
+  char Pad[3];

-  // The unrelocated RIP that is being moved
+  // The base RIP (to be moved by the register for non-literal relocations).
+  // In a serialized code cache, this is relative to the binary base address.
  uint64_t GuestRIP;
+
+  uint32_t pad2[6] {};
 };

 union Relocation {
-  RelocationTypeHeader Header {};
+  // Clang 16 Can't default-initialize this union
+  static Relocation Default() {
+#if __clang_major__ < 17
+    Relocation Ret {.Header {}};
+    memset(&Ret, 0, sizeof(Ret));
+    return Ret;
+#else
+    return {};
+#endif
+  }
+
+  RelocationHeader Header {};

  RelocNamedSymbolLiteral NamedSymbolLiteral;
  // This makes our union of relocations at least 48 bytes
  // It might be more efficient to not use a union
  RelocNamedThunkMove NamedThunkMove;

-  RelocGuestRIPMove GuestRIPMove;
+  RelocGuestRIP GuestRIP;
 };
+
+uint64_t GetNamedSymbolLiteral(FEXCore::Context::ContextImpl&, RelocNamedSymbolLiteral::NamedSymbol);
+
 } // namespace FEXCore::CPU
@@ -41,6 +41,7 @@ namespace FEXCore::CPU {
    const auto Op = IROp->C<IR::IROp_##FEXOp>();                                                                                    \
    const auto OpSize = IROp->Size;                                                                                                 \
    const auto Is256Bit = OpSize == IR::OpSize::i256Bit;                                                                            \
+    const auto Is128Bit = OpSize == IR::OpSize::i128Bit;                                                                            \
    LOGMAN_THROW_A_FMT(!Is256Bit || HostSupportsSVE256, "Need SVE256 support in order to use {} with 256-bit operation", __func__); \
                                                                                                                                    \
    const auto Dst = GetVReg(Node);                                                                                                 \
@@ -49,8 +50,10 @@ namespace FEXCore::CPU {
                                                                                                                                    \
    if (HostSupportsSVE256 && Is256Bit) {                                                                                           \
      ARMOp(Dst.Z(), Vector1.Z(), Vector2.Z());                                                                                     \
-    } else {                                                                                                                        \
+    } else if (Is128Bit) {                                                                                                          \
      ARMOp(Dst.Q(), Vector1.Q(), Vector2.Q());                                                                                     \
+    } else {                                                                                                                        \
+      ARMOp(Dst.D(), Vector1.D(), Vector2.D());                                                                                     \
    }                                                                                                                               \
  }

@@ -744,11 +747,11 @@ DEF_OP(VFToIScalarInsert) {
    auto Src = *std::get_if<ARMEmitter::VRegister>(&SrcVar);

    switch (RoundMode) {
-    case IR::Round_Nearest: frintn(SubRegSize.Scalar, Dst, Src); break;
-    case IR::Round_Negative_Infinity: frintm(SubRegSize.Scalar, Dst, Src); break;
-    case IR::Round_Positive_Infinity: frintp(SubRegSize.Scalar, Dst, Src); break;
-    case IR::Round_Towards_Zero: frintz(SubRegSize.Scalar, Dst, Src); break;
-    case IR::Round_Host: frinti(SubRegSize.Scalar, Dst, Src); break;
+    case IR::RoundMode::Nearest: frintn(SubRegSize.Scalar, Dst, Src); break;
+    case IR::RoundMode::NegInfinity: frintm(SubRegSize.Scalar, Dst, Src); break;
+    case IR::RoundMode::PosInfinity: frintp(SubRegSize.Scalar, Dst, Src); break;
+    case IR::RoundMode::TowardsZero: frintz(SubRegSize.Scalar, Dst, Src); break;
+    case IR::RoundMode::Host: frinti(SubRegSize.Scalar, Dst, Src); break;
    }
  };

@@ -974,7 +977,7 @@ DEF_OP(LoadNamedVectorConstant) {
  }
  // Load the pointer.
  auto GenerateMemOperand = [this](IR::OpSize OpSize, uint32_t NamedConstant, ARMEmitter::Register Base) {
-    const auto ConstantOffset = offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.NamedVectorConstants[NamedConstant]);
+    const auto ConstantOffset = offsetof(FEXCore::Core::CpuStateFrame, Pointers.NamedVectorConstants[NamedConstant]);

    if (ConstantOffset <= 255 || // Unscaled 9-bit signed
        ((ConstantOffset & (IR::OpSizeToSize(OpSize) - 1)) == 0 &&
@@ -982,13 +985,13 @@ DEF_OP(LoadNamedVectorConstant) {
      return ARMEmitter::ExtendedMemOperand(Base.X(), ARMEmitter::IndexType::OFFSET, ConstantOffset);
    }

-    ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.NamedVectorConstantPointers[NamedConstant]));
+    ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.NamedVectorConstantPointers[NamedConstant]));
    return ARMEmitter::ExtendedMemOperand(TMP1, ARMEmitter::IndexType::OFFSET, 0);
  };

  if (OpSize == IR::OpSize::i256Bit) {
    // Handle SVE 32-byte variant upfront.
-    ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.NamedVectorConstantPointers[Op->Constant]));
+    ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.NamedVectorConstantPointers[Op->Constant]));
    ld1b<ARMEmitter::SubRegSize::i8Bit>(Dst.Z(), PRED_TMP_32B.Zeroing(), TMP1, 0);
    return;
  }
@@ -1010,7 +1013,7 @@ DEF_OP(LoadNamedVectorIndexedConstant) {
  const auto Dst = GetVReg(Node);

  // Load the pointer.
-  ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.IndexedNamedVectorConstantPointers[Op->Constant]));
+  ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.IndexedNamedVectorConstantPointers[Op->Constant]));

  switch (OpSize) {
  case IR::OpSize::i8Bit: ldrb(Dst, TMP1, Op->Index); break;
@@ -15,7 +15,7 @@ $end_info$

 namespace FEXCore {
 GuestToHostMap::GuestToHostMap()
-  : BlockLinks_mbr {fextl::pmr::get_default_resource()} {
+  : BlockLinks_mbr {"FEXMem_BlockLinks"} {
  BlockLinks_pma = fextl::make_unique<std::pmr::polymorphic_allocator<std::byte>>(&BlockLinks_mbr);
  // Setup our PMR map.
  BlockLinks = BlockLinks_pma->new_object<BlockLinksMapType>();
@@ -24,7 +24,7 @@ GuestToHostMap::GuestToHostMap()
 LookupCache::LookupCache(FEXCore::Context::ContextImpl* CTX)
  : ctx {CTX} {

-  TotalCacheSize = ctx->Config.VirtualMemSize / 4096 * 8 + CODE_SIZE + L1_SIZE;
+  TotalCacheSize = ctx->Config.VirtualMemSize / FEXCore::Utils::FEX_PAGE_SIZE * 8 + CODE_SIZE + MAX_L1_SIZE;

  // Block cache ends up looking like this
  // PageMemoryMap[VirtualMemoryRegion >> 12]
@@ -39,6 +39,10 @@ LookupCache::LookupCache(FEXCore::Context::ContextImpl* CTX)
  // We need one pointer per page of virtual memory
  // At 64GB of virtual memory this will allocate 128MB of virtual memory space
  PagePointer = reinterpret_cast<uintptr_t>(FEXCore::Allocator::VirtualAlloc(TotalCacheSize, false, false));
+  LOGMAN_THROW_A_FMT(PagePointer != -1ULL, "Failed to allocate PagePointer");
+
+  FEXCore::Allocator::VirtualName("FEXMem_Lookup", reinterpret_cast<void*>(PagePointer),
+                                  ctx->Config.VirtualMemSize / FEXCore::Utils::FEX_PAGE_SIZE * 8 + CODE_SIZE);
  CTX->SyscallHandler->MarkOvercommitRange(PagePointer, TotalCacheSize);

  // Allocate our memory backing our pages
@@ -46,14 +50,21 @@ LookupCache::LookupCache(FEXCore::Context::ContextImpl* CTX)
  // XXX: We can drop down to 16KB if we store 4byte offsets from the code base
  // We currently limit to 128MB of real memory for caching for the total cache size.
  // Can end up being inefficient if we compile a small number of blocks per page
-  PageMemory = PagePointer + ctx->Config.VirtualMemSize / 4096 * 8;
-  LOGMAN_THROW_A_FMT(PageMemory != -1ULL, "Failed to allocate page memory");
+  PageMemory = PagePointer + ctx->Config.VirtualMemSize / FEXCore::Utils::FEX_PAGE_SIZE * 8;

  // L1 Cache
  L1Pointer = PageMemory + CODE_SIZE;
-  LOGMAN_THROW_A_FMT(L1Pointer != -1ULL, "Failed to allocate L1Pointer");
+  FEXCore::Allocator::VirtualName("FEXMem_Lookup_L1", reinterpret_cast<void*>(L1Pointer), MAX_L1_SIZE);

  VirtualMemSize = ctx->Config.VirtualMemSize;
+
+  if (DynamicL1Cache()) {
+    // Start at minimum size when dynamic.
+    L1PointerMask = MIN_L1_ENTRIES - 1;
+  } else {
+    // Start at maximum instead.
+    L1PointerMask = MAX_L1_ENTRIES - 1;
+  }
 }

 LookupCache::~LookupCache() {
@@ -64,31 +75,27 @@ LookupCache::~LookupCache() {
  // These will get freed when their memory allocators are deallocated.
 }

-void LookupCache::ClearL2Cache() {
-  auto lk = Shared->AcquireLock();
+void LookupCache::ClearL2Cache(const FEXCore::LookupCacheBaseLockToken& lk) {
  // Clear out the page memory
  // PagePointer and PageMemory are sequential with each other. Clear both at once.
-  FEXCore::Allocator::VirtualDontNeed(reinterpret_cast<void*>(PagePointer), ctx->Config.VirtualMemSize / 4096 * 8 + CODE_SIZE, false);
+  FEXCore::Allocator::VirtualDontNeed(reinterpret_cast<void*>(PagePointer),
+                                      ctx->Config.VirtualMemSize / FEXCore::Utils::FEX_PAGE_SIZE * 8 + CODE_SIZE, false);
  AllocateOffset = 0;
 }

-void LookupCache::ClearThreadLocalCaches() {
-  auto lk = Shared->AcquireLock();
-
+void LookupCache::ClearThreadLocalCaches(const LookupCacheWriteLockToken&) {
  // Clear L1 and L2 by clearing the full cache.
  FEXCore::Allocator::VirtualDontNeed(reinterpret_cast<void*>(PagePointer), TotalCacheSize, false);
+  CachedCodePages.clear();
 }

-void LookupCache::ClearCache() {
-  auto lk = Shared->AcquireLock();
-
+void LookupCache::ClearCache(const LookupCacheWriteLockToken& lk) {
  // Clear L1 and L2 by clearing the full cache.
-  FEXCore::Allocator::VirtualDontNeed(reinterpret_cast<void*>(PagePointer), TotalCacheSize, false);
-
+  ClearThreadLocalCaches(lk);
  Shared->ClearCache(lk);
 }

-void GuestToHostMap::ClearCache(const LockToken&) {
+void GuestToHostMap::ClearCache(const LookupCacheWriteLockToken&) {
  // Allocate a new pointer from the BlockLinks pma again.
  BlockLinks = BlockLinks_pma->new_object<BlockLinksMapType>();
  // All code is gone, clear the block list
@@ -2,30 +2,57 @@
 #pragma once
 #include "Interface/Context/Context.h"
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/SHMStats.h>
+#include "Utils/WritePriorityMutex.h"
+
 #include <FEXCore/fextl/map.h>
 #include <FEXCore/fextl/memory_resource.h>
 #include <FEXCore/fextl/robin_map.h>
+#include <FEXCore/fextl/robin_set.h>
 #include <FEXCore/fextl/vector.h>
 #include <FEXCore/fextl/memory_resource.h>

 #include <cstdint>
-#include <functional>
 #include <stddef.h>
 #include <utility>
 #include <mutex>

 namespace FEXCore {
+struct LookupCacheBaseLockToken {
+protected:
+  // Protected constructor - only derived classes can construct
+  LookupCacheBaseLockToken() = default;
+};
+
+struct LookupCacheWriteLockToken : public LookupCacheBaseLockToken {
+private:
+  // Only constructible by GuestToHostMap
+  friend struct GuestToHostMap;
+  LookupCacheWriteLockToken(FEXCore::Utils::WritePriorityMutex::Mutex& Mutex)
+    : Lock {Mutex} {}
+  std::lock_guard<FEXCore::Utils::WritePriorityMutex::Mutex> Lock;
+};
+
+struct LookupCacheReadLockToken : public LookupCacheBaseLockToken {
+private:
+  // Only constructible by GuestToHostMap
+  friend struct GuestToHostMap;
+  LookupCacheReadLockToken(FEXCore::Utils::WritePriorityMutex::Mutex& Mutex)
+    : Lock {Mutex} {}
+  std::shared_lock<FEXCore::Utils::WritePriorityMutex::Mutex> Lock;
+};

 struct GuestToHostMap {
-  std::recursive_mutex WriteLock;
-
-  struct LockToken {
-    std::lock_guard<std::recursive_mutex> Lock;
-  };
+  FEXCore::Utils::WritePriorityMutex::Mutex Lock {};

  [[nodiscard]]
-  LockToken AcquireLock() {
-    return LockToken {std::lock_guard {WriteLock}};
+  LookupCacheWriteLockToken AcquireWriteLock() {
+    return LookupCacheWriteLockToken {Lock};
+  }
+
+  [[nodiscard]]
+  LookupCacheReadLockToken AcquireReadLock() {
+    return LookupCacheReadLockToken {Lock};
  }

  struct BlockLinkTag {
@@ -49,53 +76,72 @@ struct GuestToHostMap {
  // walking each block member and destructing objects.
  //
  // This makes `BlockLinks` look like a raw pointer that could memory leak, but since it is backed by the MBR, it won't.
-  std::pmr::monotonic_buffer_resource BlockLinks_mbr;
+  fextl::pmr::named_monotonic_page_buffer_resource BlockLinks_mbr;
  using BlockLinksMapType = std::pmr::map<BlockLinkTag, FEXCore::Context::BlockDelinkerFunc>;
  fextl::unique_ptr<std::pmr::polymorphic_allocator<std::byte>> BlockLinks_pma;
  BlockLinksMapType* BlockLinks;

-  fextl::robin_map<uint64_t, uint64_t> BlockList;
+  struct BlockEntry {
+    uint64_t HostCode;
+    fextl::vector<uint64_t> CodePages;
+  };
+
+  fextl::robin_map<uint64_t, BlockEntry> BlockList;

  fextl::map<uint64_t, fextl::vector<uint64_t>> CodePages;

  GuestToHostMap();

  // Adds to Guest -> Host code mapping
-  void AddBlockMapping(uint64_t Address, void* HostCode, const LockToken&) {
+  const BlockEntry& AddBlockMapping(uint64_t Address, const fextl::vector<uint64_t>& CodePages, void* HostCode, const LookupCacheWriteLockToken&) {
    // This may replace an existing mapping
    // NOTE: Generally no previous entry should exist, however there is one exception:
    //       If the backend updates the active thread's CodeBuffer, the new associated LookupCache
    //       may already contain the block address. Since is comparatively rare, we'll just leak
    //       one of the two blocks in this case.
-    BlockList[Address] = (uintptr_t)HostCode;
+    return BlockList.insert_or_assign(Address, BlockEntry {(uintptr_t)HostCode, CodePages}).first->second;
  }

-  std::optional<uintptr_t> FindBlock(uint64_t Address, const LockToken&) {
+  const BlockEntry* FindBlock(uint64_t Address, const LookupCacheReadLockToken&) {
    auto HostCode = BlockList.find(Address);
    if (HostCode == BlockList.end()) {
-      return std::nullopt;
+      return nullptr;
    }
-    return HostCode->second;
+    return &HostCode->second;
  }

-  bool Erase(FEXCore::Core::CpuStateFrame* Frame, uint64_t Address, const LockToken&) {
+  bool Erase(uint64_t Address, const LookupCacheWriteLockToken&) {
    // Sever any links to this block
    auto lower = BlockLinks->lower_bound({Address, nullptr});
    auto upper = BlockLinks->upper_bound({Address, reinterpret_cast<FEXCore::Context::ExitFunctionLinkData*>(UINTPTR_MAX)});
    for (auto it = lower; it != upper; it = BlockLinks->erase(it)) {
-      it->second(Frame, it->first.HostLink);
+      it->second(it->first.HostLink);
    }

    // Remove from BlockList
    return BlockList.erase(Address) != 0;
  }

+  void InvalidateRange(uint64_t Start, uint64_t Length) {
+    auto lk = AcquireWriteLock();
+
+    auto lower = CodePages.lower_bound(Start >> 12);
+    auto upper = CodePages.upper_bound((Start + Length - 1) >> 12);
+
+    for (auto it = lower; it != upper; it++) {
+      for (const auto& Entry : it->second) {
+        Erase(Entry, lk);
+      }
+    }
+    CodePages.erase(lower, upper);
+  }
+
  void AddBlockLink(uint64_t GuestDestination, FEXCore::Context::ExitFunctionLinkData* HostLink,
-                    const FEXCore::Context::BlockDelinkerFunc& delinker, const LockToken&) {
+                    const FEXCore::Context::BlockDelinkerFunc& delinker, const LookupCacheWriteLockToken&) {
    BlockLinks->insert({{GuestDestination, HostLink}, delinker});
  }

-  bool AddBlockExecutableRange(const fextl::set<uint64_t>& Addresses, uint64_t Start, uint64_t Length, const LockToken&) {
+  bool AddBlockExecutableRange(const std::ranges::input_range auto& Addresses, uint64_t Start, uint64_t Length, const LookupCacheWriteLockToken&) {
    bool rv = false;

    for (auto CurrentPage = Start >> 12, EndPage = (Start + Length - 1) >> 12; CurrentPage <= EndPage; CurrentPage++) {
@@ -107,7 +153,7 @@ struct GuestToHostMap {
    return rv;
  }

-  void ClearCache(const LockToken&);
+  void ClearCache(const LookupCacheWriteLockToken&);
 };

 class LookupCache {
@@ -122,122 +168,199 @@ public:

  // Swaps out the underlying GuestToHostMap and clears all associated caches.
  // This interface requires the previous CodeBuffer to be provided despite not using it. This ensures the shared write lock is still valid.
-  void ChangeGuestToHostMapping([[maybe_unused]] CPU::CodeBuffer& Prev, GuestToHostMap& NewMap) {
-    ClearThreadLocalCaches();
+  void ChangeGuestToHostMapping([[maybe_unused]] CPU::CodeBuffer& Prev, GuestToHostMap& NewMap, const LookupCacheWriteLockToken& lk) {
+    ClearThreadLocalCaches(lk);
    Shared = &NewMap;
  }

-  uintptr_t FindBlock(uint64_t Address) {
+  uintptr_t FindBlock(FEXCore::Core::InternalThreadState* Thread, uint64_t Address) {
    // Try L1, no lock needed
-    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
+    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1PointerMask];
    if (L1Entry.GuestCode == Address) {
      return L1Entry.HostCode;
    }

    // L2 and L3 need to be locked
-    auto lk = Shared->AcquireLock();
+    uintptr_t HostPtr {};
+    {
+      std::optional<FEXCore::SHMStats::AccumulationBlock<uint64_t>> LockTime(
+        Thread->ThreadStats ? &Thread->ThreadStats->AccumulatedCacheReadLockTime : nullptr);
+      auto lk = Shared->AcquireReadLock();
+      LockTime.reset();

-    // Try L2
-    const auto PageIndex = (Address & (VirtualMemSize - 1)) >> 12;
-    const auto PageOffset = Address & (0x0FFF);
+      if (!DisableL2Cache()) {
+        // Try L2
+        const auto PageIndex = (Address & (VirtualMemSize - 1)) >> 12;
+        const auto PageOffset = Address & (0x0FFF);

-    const auto Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
-    auto LocalPagePointer = Pointers[PageIndex];
+        const auto Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
+        auto LocalPagePointer = Pointers[PageIndex];

-    // Do we a page pointer for this address?
-    if (LocalPagePointer) {
-      // Find there pointer for the address in the blocks
-      auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
+        // Do we a page pointer for this address?
+        if (LocalPagePointer) {
+          // Find there pointer for the address in the blocks
+          auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);

-      if (BlockPointers[PageOffset].GuestCode == Address) {
-        L1Entry.GuestCode = Address;
-        L1Entry.HostCode = BlockPointers[PageOffset].HostCode;
-        return L1Entry.HostCode;
+          if (BlockPointers[PageOffset].GuestCode == Address) {
+            L1Entry.GuestCode = Address;
+            L1Entry.HostCode = BlockPointers[PageOffset].HostCode;
+            HostPtr = L1Entry.HostCode;
+          }
+        }
+      }
+
+      if (!HostPtr) {
+        // Try L3
+        auto Entry = Shared->FindBlock(Address, lk);
+        if (Entry) {
+          CacheBlockMapping(Address, *Entry, false, lk);
+          HostPtr = Entry->HostCode;
+        }
      }
    }

-    // Try L3
-    auto HostCode = Shared->FindBlock(Address, lk);
-    if (HostCode) {
-      CacheBlockMapping(Address, HostCode.value());
-      return HostCode.value();
+    if (HostPtr && DynamicL1Cache()) {
+      UpdateDynamicL1Stats(Thread);
    }

-    // Failed to find
-    return 0;
+    FEXCORE_PROFILE_INSTANT_INCREMENT(Thread, AccumulatedCacheMissCount, 1);
+
+    return HostPtr;
+  }
+
+  void UpdateDynamicL1Stats(FEXCore::Core::InternalThreadState* Thread) {
+    // If host pointer was found in L2 or L3, then add it to the counter.
+    // Keeping track not L1 misses, but specifically L2/L3 hits.
+    ++L2L3CacheHits;
+
+    const auto CurrentTime = std::chrono::system_clock::now();
+    const auto Period = CurrentTime - LastPeriod;
+    if (Period >= SamplePeriod) {
+      // If larger than the sample period then check if we need to increase L1 cache size.
+      const double AveragePerSecond = static_cast<double>(L2L3CacheHits) /
+                                      static_cast<double>(std::chrono::duration_cast<std::chrono::milliseconds>(Period).count()) * 1000.0;
+
+      if (AveragePerSecond >= DynamicL1CacheIncreaseCountHeuristic()) {
+        if (CurrentL1Entries < MAX_L1_ENTRIES) {
+          CurrentL1Entries <<= 1;
+          L1PointerMask = CurrentL1Entries - 1;
+
+          // Update the thread's L1 pointer mask to increase how much cache it uses.
+          // Since we're in C-code, this is safe to update here.
+          Thread->CurrentFrame->State.L1Mask = GetScaledL1PointerMask();
+        }
+      } else if (AveragePerSecond < DynamicL1CacheDecreaseCountHeuristic()) {
+        if (CurrentL1Entries > MIN_L1_ENTRIES) {
+          CurrentL1Entries >>= 1;
+          L1PointerMask = CurrentL1Entries - 1;
+
+          // Madvise the entries that we are dropping. Gives the memory back to the OS.
+          LookupCacheEntry* FirstZeroL1Entry = &reinterpret_cast<LookupCacheEntry*>(L1Pointer)[CurrentL1Entries];
+          size_t ZeroMemorySize = (MAX_L1_ENTRIES - CurrentL1Entries) * sizeof(LookupCacheEntry);
+          FEXCore::Allocator::VirtualDontNeed(FirstZeroL1Entry, ZeroMemorySize, false);
+
+          // Update the thread's L1 pointer mask to increase how much cache it uses.
+          // Since we're in C-code, this is safe to update here.
+          Thread->CurrentFrame->State.L1Mask = GetScaledL1PointerMask();
+        }
+      }
+
+      // Update Last period to start again.
+      LastPeriod = CurrentTime;
+      L2L3CacheHits = 0;
+    }
  }

  GuestToHostMap* Shared = nullptr;

  // Appends a list of Block {Address} to CodePages [Start, Start + Length)
  // Returns true if new pages are marked as containing code
-  bool AddBlockExecutableRange(const fextl::set<uint64_t>& Addresses, uint64_t Start, uint64_t Length) {
-    auto lk = Shared->AcquireLock();
+  bool AddBlockExecutableRange(FEXCore::Core::InternalThreadState* Thread, const fextl::set<uint64_t>& Addresses, uint64_t Start, uint64_t Length) {
+    std::optional<FEXCore::SHMStats::AccumulationBlock<uint64_t>> LockTime(
+      Thread->ThreadStats ? &Thread->ThreadStats->AccumulatedCacheWriteLockTime : nullptr);
+    auto lk = Shared->AcquireWriteLock();
+    LockTime.reset();
+
    return Shared->AddBlockExecutableRange(Addresses, Start, Length, lk);
  }

  // Adds to Guest -> Host code mapping
-  void AddBlockMapping(uint64_t Address, void* HostCode) {
-    auto lk = Shared->AcquireLock();
+  void AddBlockMapping(FEXCore::Core::InternalThreadState* Thread, uint64_t Address, const fextl::vector<uint64_t>& CodePages, void* HostCode) {
+    std::optional<FEXCore::SHMStats::AccumulationBlock<uint64_t>> LockTime(
+      Thread->ThreadStats ? &Thread->ThreadStats->AccumulatedCacheWriteLockTime : nullptr);
+    auto lk = Shared->AcquireWriteLock();
+    LockTime.reset();

-    Shared->AddBlockMapping(Address, HostCode, lk);
+    const auto& Entry = Shared->AddBlockMapping(Address, CodePages, HostCode, lk);

    // There is no need to update L1 or L2, they will get updated on first lookup
    // However, adding to L1 here increases performance
-    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
-    L1Entry.GuestCode = Address;
-    L1Entry.HostCode = (uintptr_t)HostCode;
+    CacheBlockMapping(Address, Entry, true, lk);
  }

-  // NOTE: It's the caller's responsibility to call Erase() for all other
-  //       GuestToHostMaps that share the same LookupCache. Otherwise, the
-  //       L1/L2 caches will contain stale references to deallocated memory.
-  bool Erase(FEXCore::Core::CpuStateFrame* Frame, uint64_t Address) {
-    auto lk = Shared->AcquireLock();
-
-    bool ErasedAny = Shared->Erase(Frame, Address, lk);
-
+  // Invalidates L1/L2 for a given guest block
+  void InvalidateCache(uint64_t Address, const LookupCacheWriteLockToken& lk) {
    // Do L1
-    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
+    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1PointerMask];
    if (L1Entry.GuestCode == Address) {
      L1Entry.GuestCode = 0;
-      ErasedAny = true;
      // Leave L1Entry.HostCode as is, so that concurrent lookups won't read a null pointer
      // This is a soft guarantee for cross thread invalidation, as atomics are not used
      // and it hasn't been thoroughly tested
    }

-    // Do full map
-    Address = Address & (VirtualMemSize - 1);
-    uint64_t PageOffset = Address & (0x0FFF);
-    Address >>= 12;
+    if (!DisableL2Cache()) {
+      // Do full map
+      Address = Address & (VirtualMemSize - 1);
+      uint64_t PageOffset = Address & (0x0FFF);
+      Address >>= 12;

-    uintptr_t* Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
-    uint64_t LocalPagePointer = Pointers[Address];
-    if (!LocalPagePointer) {
-      // Page for this code didn't even exist, nothing to do
-      return ErasedAny;
+      uintptr_t* Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
+      uint64_t LocalPagePointer = Pointers[Address];
+      if (!LocalPagePointer) {
+        // Page for this code didn't even exist, nothing to do
+        return;
+      }
+
+      // Page exists, just set the offset to zero
+      auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
+      BlockPointers[PageOffset].GuestCode = 0;
+      BlockPointers[PageOffset].HostCode = 0;
    }
-
-    // Page exists, just set the offset to zero
-    auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
-    BlockPointers[PageOffset].GuestCode = 0;
-    BlockPointers[PageOffset].HostCode = 0;
-    return true;
  }

-  void AddBlockLink(uint64_t GuestDestination, FEXCore::Context::ExitFunctionLinkData* HostLink, const FEXCore::Context::BlockDelinkerFunc& delinker) {
-    auto lk = Shared->AcquireLock();
+  // Invalidates all L1/L2 entries for all guest block that intersect the given range
+  bool InvalidateCacheRange(uint64_t Start, uint64_t Length) {
+    auto lk = Shared->AcquireWriteLock();
+
+    auto lower = CachedCodePages.lower_bound(Start >> 12);
+    auto upper = CachedCodePages.upper_bound((Start + Length - 1) >> 12);
+
+    for (auto it = lower; it != upper; it++) {
+      for (const auto& Entry : it->second) {
+        InvalidateCache(Entry, lk);
+      }
+    }
+    bool ret = upper != lower;
+    CachedCodePages.erase(lower, upper);
+    return ret;
+  }
+
+  void AddBlockLink(uint64_t GuestDestination, FEXCore::Context::ExitFunctionLinkData* HostLink,
+                    const FEXCore::Context::BlockDelinkerFunc& delinker, const LookupCacheWriteLockToken& lk) {
    Shared->AddBlockLink(GuestDestination, HostLink, delinker, lk);
  }

-  void ClearCache();
-  void ClearL2Cache();
-  void ClearThreadLocalCaches();
+  void ClearCache(const LookupCacheWriteLockToken&);
+  void ClearL2Cache(const LookupCacheBaseLockToken&);
+  void ClearThreadLocalCaches(const LookupCacheWriteLockToken&);

  uintptr_t GetL1Pointer() const {
    return L1Pointer;
  }
+  uintptr_t GetScaledL1PointerMask() const {
+    return L1PointerMask << FEXCore::ilog2(sizeof(LookupCache::LookupCacheEntry));
+  }
  uintptr_t GetPagePointer() const {
    return PagePointer;
  }
@@ -245,9 +368,6 @@ public:
    return VirtualMemSize;
  }

-  constexpr static size_t L1_ENTRIES = 1 * 1024 * 1024; // Must be a power of 2
-  constexpr static size_t L1_ENTRIES_MASK = L1_ENTRIES - 1;
-
  // This needs to be taken before reads or writes to L2, L3, CodePages,
  // and before writes to L1. Concurrent access from a thread that this LookupCache doesn't belong to
  // may only happen during cross thread invalidation (::Erase).
@@ -255,45 +375,52 @@ public:
  // Some care is taken so that L1 lookups can be done without locks, and even tearing is unlikely to lead to a crash.
  // This approach has not been fully vetted yet.
  // Also note that L1 lookups might be inlined in the JIT Dispatcher and/or block ends.
-  auto AcquireLock() {
-    return Shared->AcquireLock();
+  auto AcquireWriteLock() {
+    return Shared->AcquireWriteLock();
  }

 private:
-  void CacheBlockMapping(uint64_t Address, uintptr_t HostCode) {
-    // Do L1
-    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
-    L1Entry.GuestCode = Address;
-    L1Entry.HostCode = HostCode;
-
-    // Do ful map
-    auto FullAddress = Address;
-    Address = Address & (VirtualMemSize - 1);
-
-    uint64_t PageOffset = Address & (0x0FFF);
-    Address >>= 12;
-    uintptr_t* Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
-    uint64_t LocalPagePointer = Pointers[Address];
-    if (!LocalPagePointer) {
-      // We don't have a page pointer for this address
-      // Allocate one now if we can
-      uintptr_t NewPageBacking = AllocateBackingForPage();
-      if (!NewPageBacking) {
-        // Couldn't allocate, clear L2 and retry
-        ClearL2Cache();
-        CacheBlockMapping(Address, HostCode);
-        return;
-      }
-      Pointers[Address] = NewPageBacking;
-      LocalPagePointer = NewPageBacking;
+  void CacheBlockMapping(uint64_t Address, const GuestToHostMap::BlockEntry& Entry, bool L1Only, const LookupCacheBaseLockToken& lk) {
+    for (const auto& CodePage : Entry.CodePages) {
+      CachedCodePages[CodePage >> 12].insert(Address);
    }

-    // Add the new pointer to the page block
-    auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
+    // Do L1
+    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1PointerMask];
+    L1Entry.GuestCode = Address;
+    L1Entry.HostCode = Entry.HostCode;

-    // This silently replaces existing mappings
-    BlockPointers[PageOffset].GuestCode = FullAddress;
-    BlockPointers[PageOffset].HostCode = HostCode;
+    if (!DisableL2Cache() && !L1Only) {
+      // Do ful map
+      auto FullAddress = Address;
+      Address = Address & (VirtualMemSize - 1);
+
+      uint64_t PageOffset = Address & (0x0FFF);
+      Address >>= 12;
+
+      uintptr_t* Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
+      uint64_t LocalPagePointer = Pointers[Address];
+      if (!LocalPagePointer) {
+        // We don't have a page pointer for this address
+        // Allocate one now if we can
+        uintptr_t NewPageBacking = AllocateBackingForPage();
+        if (!NewPageBacking) {
+          // Couldn't allocate, clear L2 and retry
+          ClearL2Cache(lk);
+          CacheBlockMapping(FullAddress, Entry, false, lk);
+          return;
+        }
+        Pointers[Address] = NewPageBacking;
+        LocalPagePointer = NewPageBacking;
+      }
+
+      // Add the new pointer to the page block
+      auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
+
+      // This silently replaces existing mappings
+      BlockPointers[PageOffset].GuestCode = FullAddress;
+      BlockPointers[PageOffset].HostCode = Entry.HostCode;
+    }
  }

  uintptr_t AllocateBackingForPage() {
@@ -310,19 +437,38 @@ private:
    return PageMemory + NewBase;
  }

+  // Maps from a page index to all blocks in the page that have at some point been fetched into L1/L2
+  fextl::map<uint64_t, fextl::robin_set<uint64_t>> CachedCodePages;
+
  uintptr_t PagePointer;
  uintptr_t PageMemory;
  uintptr_t L1Pointer;
+  uintptr_t L1PointerMask;

  size_t TotalCacheSize;

+  // Start with 8k entries in L1 to give 128KB of L1 cache to each thread.
+  // Max out at 1 million entries to give each thread 16MB of L1 cache maximum.
+  constexpr static size_t MIN_L1_ENTRIES = 8 * 1024;        // Must be a power of 2
+  constexpr static size_t MAX_L1_ENTRIES = 1 * 1024 * 1024; // Must be a power of 2
+
  constexpr static size_t CODE_SIZE = 128 * 1024 * 1024;
-  constexpr static size_t SIZE_PER_PAGE = 4096 * sizeof(LookupCacheEntry);
-  constexpr static size_t L1_SIZE = L1_ENTRIES * sizeof(LookupCacheEntry);
+  constexpr static size_t SIZE_PER_PAGE = FEXCore::Utils::FEX_PAGE_SIZE * sizeof(LookupCacheEntry);
+  constexpr static size_t MAX_L1_SIZE = MAX_L1_ENTRIES * sizeof(LookupCacheEntry);

  size_t AllocateOffset {};

  FEXCore::Context::ContextImpl* ctx;
  uint64_t VirtualMemSize {};
+
+  size_t CurrentL1Entries = MIN_L1_ENTRIES;
+  uint64_t L2L3CacheHits {};
+  std::chrono::time_point<std::chrono::system_clock> LastPeriod {};
+  constexpr static std::chrono::seconds SamplePeriod {1};
+  FEX_CONFIG_OPT(DynamicL1CacheIncreaseCountHeuristic, DYNAMICL1CACHEINCREASECOUNTHEURISTIC);
+  FEX_CONFIG_OPT(DynamicL1CacheDecreaseCountHeuristic, DYNAMICL1CACHEDECREASECOUNTHEURISTIC);
+
+  FEX_CONFIG_OPT(DynamicL1Cache, DYNAMICL1CACHE);
+  FEX_CONFIG_OPT(DisableL2Cache, DISABLEL2CACHE);
 };
 } // namespace FEXCore
@@ -139,27 +139,28 @@ public:
    FlushRegisterCache();
    return _Jump(_TargetBlock);
  }
-  IRPair<IROp_CondJump> CondJump(Ref _Cmp1, Ref _Cmp2, Ref _TrueBlock, Ref _FalseBlock, CondClassType _Cond = {COND_NEQ},
+  IRPair<IROp_CondJump> CondJump(Ref _Cmp1, Ref _Cmp2, Ref _TrueBlock, Ref _FalseBlock, CondClass _Cond = CondClass::NEQ,
                                 IR::OpSize _CompareSize = OpSize::iInvalid) {
    FlushRegisterCache();
    return _CondJump(_Cmp1, _Cmp2, _TrueBlock, _FalseBlock, _Cond, _CompareSize);
  }
-  IRPair<IROp_CondJump> CondJump(Ref ssa0, CondClassType cond = {COND_NEQ}) {
+  IRPair<IROp_CondJump> CondJump(Ref ssa0, CondClass cond = CondClass::NEQ) {
    FlushRegisterCache();
    return _CondJump(ssa0, cond);
  }
-  IRPair<IROp_CondJump> CondJump(Ref ssa0, Ref ssa1, Ref ssa2, CondClassType cond = {COND_NEQ}) {
+  IRPair<IROp_CondJump> CondJump(Ref ssa0, Ref ssa1, Ref ssa2, CondClass cond = CondClass::NEQ) {
    FlushRegisterCache();
    return _CondJump(ssa0, ssa1, ssa2, cond);
  }
-  IRPair<IROp_CondJump> CondJumpNZCV(CondClassType Cond) {
+  IRPair<IROp_CondJump> CondJumpNZCV(CondClass Cond) {
    FlushRegisterCache();
    return _CondJump(InvalidNode, InvalidNode, InvalidNode, InvalidNode, Cond, OpSize::iInvalid, true);
  }
  IRPair<IROp_CondJump> CondJumpBit(Ref Src, unsigned Bit, bool Set) {
    FlushRegisterCache();
    auto InlineConst = _InlineConstant(Bit);
-    return _CondJump(Src, InlineConst, InvalidNode, InvalidNode, {Set ? COND_TSTNZ : COND_TSTZ}, OpSize::iInvalid, false);
+    auto Cond = Set ? CondClass::TSTNZ : CondClass::TSTZ;
+    return _CondJump(Src, InlineConst, InvalidNode, InvalidNode, Cond, OpSize::iInvalid, false);
  }
  IRPair<IROp_ExitFunction> ExitFunction(Ref NewRIP, BranchHint Hint = BranchHint::None) {
    FlushRegisterCache();
@@ -209,10 +210,17 @@ public:
  }

  static bool CanHaveSideEffects(const FEXCore::X86Tables::X86InstInfo* TableInfo, FEXCore::X86Tables::DecodedOp Op) {
-    if (TableInfo && TableInfo->Flags & X86Tables::InstFlags::FLAGS_DEBUG_MEM_ACCESS) {
-      // If it is marked as having memory access then always say it has a side-effect.
-      // Not always true but better to be safe.
-      return true;
+    if (TableInfo) {
+      if (TableInfo->Flags & X86Tables::InstFlags::FLAGS_DEBUG_MEM_ACCESS) {
+        // If it is marked as having memory access then always say it has a side-effect.
+        // Not always true but better to be safe.
+        return true;
+      }
+
+      if (TableInfo->Flags & (X86Tables::InstFlags::FLAGS_SETS_RIP | X86Tables::InstFlags::FLAGS_BLOCK_END)) {
+        // Cooperative suspend interrupts can be triggered at any back-edge, the RIP must be reconstructed correctly in such cases
+        return true;
+      }
    }

    auto CanHaveSideEffects = false;
@@ -244,7 +252,7 @@ public:
    auto ExitBlock = CreateNewCodeBlockAfter(BackwardBlock);

    auto DF = GetRFLAG(X86State::RFLAG_DF_RAW_LOC);
-    CondJump(DF, Zero, ForwardBlock, BackwardBlock, {COND_EQ});
+    CondJump(DF, Zero, ForwardBlock, BackwardBlock, CondClass::EQ);

    for (auto D = 0; D < 2; ++D) {
      SetCurrentCodeBlock(D ? BackwardBlock : ForwardBlock);
@@ -293,7 +301,8 @@ public:
    return ShouldDump;
  }

-  void BeginFunction(uint64_t RIP, const fextl::vector<FEXCore::Frontend::Decoder::DecodedBlocks>* Blocks, uint32_t NumInstructions, bool Is64BitMode, bool MonoBackpatcherBlock);
+  void BeginFunction(uint64_t RIP, const fextl::vector<FEXCore::Frontend::Decoder::DecodedBlocks>* Blocks, uint32_t NumInstructions,
+                     bool Is64BitMode, bool MonoBackpatcherBlock);
  void Finalize();

  // Dispatch builder functions
@@ -363,7 +372,7 @@ public:
  void CMOVOp(OpcodeArgs);
  void CPUIDOp(OpcodeArgs);
  void XGetBVOp(OpcodeArgs);
-  uint32_t LoadConstantShift(X86Tables::DecodedOp Op, bool Is1Bit);
+  uint32_t GetConstantShift(X86Tables::DecodedOp Op, bool Is1Bit);
  void SHLOp(OpcodeArgs);
  void SHLImmediateOp(OpcodeArgs, bool SHL1Bit);
  void SHROp(OpcodeArgs);
@@ -555,7 +564,7 @@ public:
  template<IR::OpSize DstElementSize, IR::OpSize SrcElementSize>
  void AVXInsertScalar_CVT_Float_To_Float(OpcodeArgs);

-  RoundType TranslateRoundType(uint8_t Mode);
+  RoundMode TranslateRoundType(uint8_t Mode);

  template<IR::OpSize ElementSize>
  void InsertScalarRound(OpcodeArgs);
@@ -749,7 +758,6 @@ public:
  void X87FXTRACT(OpcodeArgs);
  void X87FYL2X(OpcodeArgs, bool IsFYL2XP1);
  void X87LDENV(OpcodeArgs);
-  void X87LDSW(OpcodeArgs);
  void X87ModifySTP(OpcodeArgs, bool Inc);
  void X87OpHelper(OpcodeArgs, FEXCore::IR::IROps IROp, bool ZeroC2);

@@ -898,6 +906,10 @@ public:
  void VPCLMULQDQOp(OpcodeArgs);

  void CRC32(OpcodeArgs);
+  void Extrq_imm(OpcodeArgs);
+  void Insertq_imm(OpcodeArgs);
+  void Extrq(OpcodeArgs);
+  void Insertq(OpcodeArgs);

  void BreakOp(OpcodeArgs, FEXCore::IR::BreakDefinition BreakDefinition);
  void UnimplementedOp(OpcodeArgs);
@@ -976,7 +988,6 @@ public:
  void AVX128_VPSIGN(OpcodeArgs, IR::OpSize ElementSize);
  void AVX128_UCOMISx(OpcodeArgs, IR::OpSize ElementSize);
  void AVX128_VectorScalarInsertALU(OpcodeArgs, FEXCore::IR::IROps IROp, IR::OpSize ElementSize);
-  Ref AVX128_VFCMPImpl(IR::OpSize ElementSize, Ref Src1, Ref Src2, uint8_t CompType);
  void AVX128_VFCMP(OpcodeArgs, IR::OpSize ElementSize);
  void AVX128_InsertScalarFCMP(OpcodeArgs, IR::OpSize ElementSize);
  void AVX128_MOVBetweenGPR_FPR(OpcodeArgs);
@@ -995,9 +1006,7 @@ public:
  void AVX128_VINSERT(OpcodeArgs);
  void AVX128_VINSERTPS(OpcodeArgs);

-  Ref AVX128_PHSUBImpl(Ref Src1, Ref Src2, size_t ElementSize);
  void AVX128_VPHSUB(OpcodeArgs, IR::OpSize ElementSize);
-
  void AVX128_VPHSUBSW(OpcodeArgs);

  void AVX128_VADDSUBP(OpcodeArgs, IR::OpSize ElementSize);
@@ -1089,8 +1098,8 @@ public:
  void AVX128_VFMAScalarImpl(OpcodeArgs, IROps IROp, uint8_t Src1Idx, uint8_t Src2Idx, uint8_t AddendIdx);
  void AVX128_VFMAddSubImpl(OpcodeArgs, bool AddSub, uint8_t Src1Idx, uint8_t Src2Idx, uint8_t AddendIdx);

-  RefPair AVX128_VPGatherQPSImpl(Ref Dest, Ref Mask, RefVSIB VSIB);
-  RefPair AVX128_VPGatherImpl(OpSize Size, OpSize ElementLoadSize, OpSize AddrElementSize, RefPair Dest, RefPair Mask, RefVSIB VSIB);
+  RefPair AVX128_VPGatherQPSImpl(OpcodeArgs, Ref Dest, Ref Mask, RefVSIB VSIB);
+  RefPair AVX128_VPGatherImpl(OpcodeArgs, OpSize Size, OpSize ElementLoadSize, OpSize AddrElementSize, RefPair Dest, RefPair Mask, RefVSIB VSIB);

  void AVX128_VPGATHER(OpcodeArgs, OpSize AddrElementSize);

@@ -1100,8 +1109,8 @@ public:
  // End of AVX 128-bit implementation

  // AVX 256-bit operations
-  void StoreResult_WithAVXInsert(VectorOpType Type, FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, Ref Value,
-                                 IR::OpSize Align, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+  void StoreResult_WithAVXInsert(VectorOpType Type, RegClass Class, FEXCore::X86Tables::DecodedOp Op, Ref Value,
+                                 IR::OpSize Align = IR::OpSize::iInvalid, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
    if (Op->Dest.IsGPR() && Op->Dest.Data.GPR.GPR >= X86State::REG_XMM_0 && Op->Dest.Data.GPR.GPR <= X86State::REG_XMM_15 &&
        GetGuestVectorLength() == OpSize::i256Bit && Type == VectorOpType::SSE) {
      const auto gpr = Op->Dest.Data.GPR.GPR;
@@ -1152,7 +1161,7 @@ public:
    }
  }

-  void StoreContextHelper(IR::OpSize Size, RegisterClassType Class, Ref Value, uint32_t Offset) {
+  void StoreContextHelper(IR::OpSize Size, RegClass Class, Ref Value, uint32_t Offset) {
    // For i128Bit, we won't see a normal Constant to inline, but as a special
    // case we can replace with a 2x64-bit store which can use inline zeroes.
    if (Size == OpSize::i128Bit) {
@@ -1164,7 +1173,7 @@ public:

        if (Const->Constant == IR::NamedVectorConstant::NAMED_VECTOR_ZERO) {
          Ref Zero = _Constant(0);
-          Ref STP = _StoreContextPair(IR::OpSize::i64Bit, GPRClass, Zero, Zero, Offset);
+          Ref STP = _StoreContextPair(IR::OpSize::i64Bit, RegClass::GPR, Zero, Zero, Offset);

          // XXX: This works around InlineConstant not having an associated
          // register class, else we'd just do InlineConstant above.
@@ -1179,13 +1188,15 @@ public:
    _StoreContext(Size, Class, Value, Offset);
  }

-  void FlushRegisterCache(bool SRAOnly = false) {
+  void FlushRegisterCache(bool SRAOnly = false, bool MMXOnly = false) {
    // At block boundaries, fix up the carry flag.
    if (!SRAOnly) {
      RectifyCarryInvert(CFInvertedABI);
    }

-    CalculateDeferredFlags();
+    if (!MMXOnly) {
+      CalculateDeferredFlags();
+    }

    const auto GPRSize = GetGPROpSize();
    const auto VectorSize = GetGuestVectorLength();
@@ -1207,22 +1218,27 @@ public:
      Bits &= Mask;
    }

+    if (MMXOnly) {
+      Mask &= ((1ull << (MM7Index - MM0Index + 1)) - 1) << MM0Index;
+      Bits &= Mask;
+    }
+
    while (Bits != 0) {
      uint32_t Index = 63 - std::countl_zero(Bits);
      Ref Value = RegCache.Value[Index];

      if (Index >= GPR0Index && Index <= GPR15Index) {
        Ref R = _StoreRegister(Value, GPRSize);
-        R->Reg = PhysicalRegister(GPRFixedClass, Index - GPR0Index).Raw;
+        R->Reg = PhysicalRegister(RegClass::GPRFixed, Index - GPR0Index).Raw;
      } else if (Index == PFIndex) {
        _StorePF(Value, GPRSize);
      } else if (Index == AFIndex) {
        _StoreAF(Value, GPRSize);
      } else if (Index >= FPR0Index && Index <= FPR15Index) {
        Ref R = _StoreRegister(Value, VectorSize);
-        R->Reg = PhysicalRegister(FPRFixedClass, Index - FPR0Index).Raw;
+        R->Reg = PhysicalRegister(RegClass::FPRFixed, Index - FPR0Index).Raw;
      } else if (Index == DFIndex) {
-        _StoreContext(OpSize::i8Bit, GPRClass, Value, offsetof(Core::CPUState, flags[X86State::RFLAG_DF_RAW_LOC]));
+        _StoreContextGPR(OpSize::i8Bit, Value, offsetof(Core::CPUState, flags[X86State::RFLAG_DF_RAW_LOC]));
      } else {
        bool Partial = RegCache.Partial & (1ull << Index);
        auto Size = Partial ? OpSize::i64Bit : CacheIndexToOpSize(Index);
@@ -1247,7 +1263,7 @@ public:
          StoreContextHelper(Size, Class, Value, Offset);
          // If Partial and MMX register, then we need to store all 1s in bits 64-80
          if (Partial && Index >= MM0Index && Index <= MM7Index) {
-            _StoreContext(OpSize::i16Bit, IR::GPRClass, Constant(0xFFFF), Offset + 8);
+            _StoreContextGPR(OpSize::i16Bit, Constant(0xFFFF), Offset + 8);
          }
        }
      }
@@ -1527,7 +1543,7 @@ private:
  [[nodiscard]]
  static bool IsOperandMem(const X86Tables::DecodedOperand& Operand, bool Load) {
    // Literals are immediates as sources but memory addresses as destinations.
-    return !(Load && Operand.IsLiteral()) && !Operand.IsGPR();
+    return !(Load && (Operand.IsLiteral() || Operand.IsLiteralRelocation())) && !Operand.IsGPR();
  }

  [[nodiscard]]
@@ -1537,23 +1553,63 @@ private:

  AddressMode DecodeAddress(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, MemoryAccessType AccessType, bool IsLoad);

-  Ref LoadSource(RegisterClassType Class, const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, uint32_t Flags,
+  Ref LoadSource(RegClass Class, const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, uint32_t Flags,
                 const LoadSourceOptions& Options = {});
-  Ref LoadSource_WithOpSize(RegisterClassType Class, const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand,
-                            IR::OpSize OpSize, uint32_t Flags, const LoadSourceOptions& Options = {});
-  void StoreResult_WithOpSize(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op,
-                              const FEXCore::X86Tables::DecodedOperand& Operand, const Ref Src, IR::OpSize OpSize, IR::OpSize Align,
-                              MemoryAccessType AccessType = MemoryAccessType::DEFAULT);
-  void StoreResult(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, const FEXCore::X86Tables::DecodedOperand& Operand,
-                   const Ref Src, IR::OpSize Align, MemoryAccessType AccessType = MemoryAccessType::DEFAULT);
-  void StoreResult(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, const Ref Src, IR::OpSize Align,
+  Ref LoadSourceGPR(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, uint32_t Flags,
+                    const LoadSourceOptions& Options = {}) {
+    return LoadSource(RegClass::GPR, Op, Operand, Flags, Options);
+  }
+  Ref LoadSourceFPR(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, uint32_t Flags,
+                    const LoadSourceOptions& Options = {}) {
+    return LoadSource(RegClass::FPR, Op, Operand, Flags, Options);
+  }
+
+  Ref LoadSource_WithOpSize(RegClass Class, const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, IR::OpSize OpSize,
+                            uint32_t Flags, const LoadSourceOptions& Options = {});
+  Ref LoadSourceGPR_WithOpSize(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, IR::OpSize OpSize, uint32_t Flags,
+                               const LoadSourceOptions& Options = {}) {
+    return LoadSource_WithOpSize(RegClass::GPR, Op, Operand, OpSize, Flags, Options);
+  }
+  Ref LoadSourceFPR_WithOpSize(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, IR::OpSize OpSize, uint32_t Flags,
+                               const LoadSourceOptions& Options = {}) {
+    return LoadSource_WithOpSize(RegClass::FPR, Op, Operand, OpSize, Flags, Options);
+  }
+
+  void StoreResult_WithOpSize(RegClass Class, X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, IR::OpSize OpSize,
+                              IR::OpSize Align, MemoryAccessType AccessType = MemoryAccessType::DEFAULT);
+  void StoreResultGPR_WithOpSize(X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, IR::OpSize OpSize,
+                                 IR::OpSize Align = IR::OpSize::iInvalid, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult_WithOpSize(RegClass::GPR, Op, Operand, Src, OpSize, Align, AccessType);
+  }
+  void StoreResultFPR_WithOpSize(X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, IR::OpSize OpSize,
+                                 IR::OpSize Align = IR::OpSize::iInvalid, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult_WithOpSize(RegClass::FPR, Op, Operand, Src, OpSize, Align, AccessType);
+  }
+
+  void StoreResult(RegClass Class, X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, OpSize Align,
                   MemoryAccessType AccessType = MemoryAccessType::DEFAULT);
+  void StoreResultGPR(X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, OpSize Align = OpSize::iInvalid,
+                      MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult(RegClass::GPR, Op, Operand, Src, Align, AccessType);
+  }
+  void StoreResultFPR(X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, OpSize Align = OpSize::iInvalid,
+                      MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult(RegClass::FPR, Op, Operand, Src, Align, AccessType);
+  }
+
+  void StoreResult(RegClass Class, X86Tables::DecodedOp Op, Ref Src, OpSize Align, MemoryAccessType AccessType = MemoryAccessType::DEFAULT);
+  void StoreResultGPR(X86Tables::DecodedOp Op, Ref Src, OpSize Align = OpSize::iInvalid, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult(RegClass::GPR, Op, Src, Align, AccessType);
+  }
+  void StoreResultFPR(X86Tables::DecodedOp Op, Ref Src, OpSize Align = OpSize::iInvalid, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult(RegClass::FPR, Op, Src, Align, AccessType);
+  }

  // In several instances, it's desirable to get a base address with the segment offset
  // applied to it. This pulls all the common-case appending into a single set of functions.
  [[nodiscard]]
  Ref MakeSegmentAddress(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, IR::OpSize OpSize) {
-    Ref Mem = LoadSource_WithOpSize(GPRClass, Op, Operand, OpSize, Op->Flags, {.LoadData = false});
+    Ref Mem = LoadSourceGPR_WithOpSize(Op, Operand, OpSize, Op->Flags, {.LoadData = false});
    return AppendSegmentOffset(Mem, Op->Flags);
  }
  [[nodiscard]]
@@ -1598,6 +1654,9 @@ private:
    return IR::SizeToOpSize(GetSrcSize(Op));
  }

+  [[nodiscard]]
+  IR::OpSize GetStringOpSize(X86Tables::DecodedOp Op) const;
+
  // Set flag tracking to prepare for an operation that directly writes NZCV.
  void HandleNZCVWrite() {
    CachedNZCV = nullptr;
@@ -1789,14 +1848,15 @@ private:
      // For DF, we need to transform 0/1 into 1/-1
      StoreDF(_SubShift(OpSize::i64Bit, Constant(1), Value, ShiftType::LSL, 1));
    } else if (BitOffset == FEXCore::X86State::RFLAG_TF_RAW_LOC) {
-      auto PackedTF = _LoadContext(OpSize::i8Bit, GPRClass, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
+      auto PackedTF = _LoadContextGPR(OpSize::i8Bit, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
      // An exception should still be raised after an instruction that unsets TF, leave the unblocked bit set but unset
      // the TF bit to cause such behaviour. The handling code at the start of the next block will then unset the
      // unblocked bit before raising the exception.
-      auto NewPackedTF = _Select(FEXCore::IR::COND_EQ, Value, Constant(0), _And(OpSize::i32Bit, PackedTF, Constant(~1)), Constant(1));
-      _StoreContext(OpSize::i8Bit, GPRClass, NewPackedTF, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
+      auto NewPackedTF =
+        _Select(OpSize::i64Bit, OpSize::i64Bit, CondClass::EQ, Value, Constant(0), _And(OpSize::i32Bit, PackedTF, Constant(~1)), Constant(1));
+      _StoreContextGPR(OpSize::i8Bit, NewPackedTF, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
    } else {
-      _StoreContext(OpSize::i8Bit, GPRClass, Value, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
+      _StoreContextGPR(OpSize::i8Bit, Value, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
    }
  }

@@ -1822,12 +1882,12 @@ private:
  }

  [[nodiscard]]
-  static CondClassType CondForNZCVBit(unsigned BitOffset, bool Invert) {
+  static CondClass CondForNZCVBit(unsigned BitOffset, bool Invert) {
    switch (BitOffset) {
-    case X86State::RFLAG_SF_RAW_LOC: return {Invert ? COND_PL : COND_MI};
-    case X86State::RFLAG_ZF_RAW_LOC: return {Invert ? COND_NEQ : COND_EQ};
-    case X86State::RFLAG_CF_RAW_LOC: return {Invert ? COND_ULT : COND_UGE};
-    case X86State::RFLAG_OF_RAW_LOC: return {Invert ? COND_FNU : COND_FU};
+    case X86State::RFLAG_SF_RAW_LOC: return Invert ? CondClass::PL : CondClass::MI;
+    case X86State::RFLAG_ZF_RAW_LOC: return Invert ? CondClass::NEQ : CondClass::EQ;
+    case X86State::RFLAG_CF_RAW_LOC: return Invert ? CondClass::ULT : CondClass::UGE;
+    case X86State::RFLAG_OF_RAW_LOC: return Invert ? CondClass::FNU : CondClass::FU;
    default: FEX_UNREACHABLE;
    }
  }
@@ -1838,10 +1898,10 @@ private:
  static const int PFIndex = 16;
  static const int AFIndex = 17;
  /* Gap 18..19 */
+  /* Note this range is only valid if MMXState = MMXState_MMX */
  static const int MM0Index = 20;
  static const int MM7Index = 27;
-  static const int AbridgedFTWIndex = 28;
-  /* Gap 29..30 */
+  /* Gap 28..30 */
  static const int DFIndex = 31;
  static const int FPR0Index = 32;
  static const int FPR15Index = 47;
@@ -1853,17 +1913,16 @@ private:
    switch (Index) {
    case MM0Index ... MM7Index: return offsetof(FEXCore::Core::CPUState, mm[Index - MM0Index]);
    case AVXHigh0Index ... AVXHigh15Index: return offsetof(FEXCore::Core::CPUState, avx_high[Index - AVXHigh0Index][0]);
-    case AbridgedFTWIndex: return offsetof(FEXCore::Core::CPUState, AbridgedFTW);
    default: return ~0U;
    }
  }

  [[nodiscard]]
-  static RegisterClassType CacheIndexClass(int Index) {
+  static RegClass CacheIndexClass(int Index) {
    if ((Index >= MM0Index && Index <= MM7Index) || Index >= FPR0Index) {
-      return FPRClass;
+      return RegClass::FPR;
    } else {
-      return GPRClass;
+      return RegClass::GPR;
    }
  }

@@ -1895,14 +1954,14 @@ private:
    RegCache.Written &= ~Bit;
  }

-  Ref LoadRegCache(uint64_t Offset, uint8_t Index, RegisterClassType RegClass, IR::OpSize Size) {
+  Ref LoadRegCache(uint64_t Offset, uint8_t Index, RegClass Class, IR::OpSize Size) {
    LOGMAN_THROW_A_FMT(Index < 64, "valid index");
    uint64_t Bit = (1ull << (uint64_t)Index);

    if (Size == OpSize::i128Bit && (RegCache.Partial & Bit)) {
      // We need to load the full register extend if we previously did a partial access.
      Ref Value = RegCache.Value[Index];
-      Ref Full = _LoadContext(Size, RegClass, Offset);
+      Ref Full = _LoadContext(Size, Class, Offset);

      // If we did a partial store, we're inserting into the full register
      if (RegCache.Written & Bit) {
@@ -1915,8 +1974,8 @@ private:
    if (!(RegCache.Cached & Bit)) {
      if (Index == DFIndex) {
        RegCache.Value[Index] = _LoadDF();
-      } else if ((Index >= MM0Index && Index <= AbridgedFTWIndex) || Index >= AVXHigh0Index) {
-        RegCache.Value[Index] = _LoadContext(Size, RegClass, Offset);
+      } else if ((Index >= MM0Index && Index <= MM7Index) || Index >= AVXHigh0Index) {
+        RegCache.Value[Index] = _LoadContext(Size, Class, Offset);

        // We may have done a partial load, this requires special handling.
        if (Size == OpSize::i64Bit) {
@@ -1927,7 +1986,7 @@ private:
      } else if (Index == AFIndex) {
        RegCache.Value[Index] = _LoadAF(Size);
      } else {
-        RegCache.Value[Index] = _LoadRegister(Offset, RegClass, Size);
+        RegCache.Value[Index] = _LoadRegister(Offset, Class, Size);
      }

      RegCache.Cached |= Bit;
@@ -1936,21 +1995,21 @@ private:
    return RegCache.Value[Index];
  }

-  RefPair AllocatePair(FEXCore::IR::RegisterClassType Class, IR::OpSize Size) {
-    if (Class == FPRClass) {
+  RefPair AllocatePair(RegClass Class, IR::OpSize Size) {
+    if (Class == RegClass::FPR) {
      return {_AllocateFPR(Size, Size), _AllocateFPR(Size, Size)};
    } else {
      return {_AllocateGPR(false), _AllocateGPR(false)};
    }
  }

-  RefPair LoadContextPair_Uncached(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, unsigned Offset) {
+  RefPair LoadContextPair_Uncached(RegClass Class, IR::OpSize Size, unsigned Offset) {
    RefPair Values = AllocatePair(Class, Size);
    _LoadContextPair(Size, Class, Offset, Values.Low, Values.High);
    return Values;
  }

-  RefPair LoadRegCachePair(uint64_t Offset, uint8_t Index, RegisterClassType RegClass, IR::OpSize Size) {
+  RefPair LoadRegCachePair(uint64_t Offset, uint8_t Index, RegClass Class, IR::OpSize Size) {
    LOGMAN_THROW_A_FMT(Index != DFIndex, "must be pairable");
    LOGMAN_THROW_A_FMT(Size != IR::OpSize::iUnsized, "Invalid size!");

@@ -1958,7 +2017,7 @@ private:
    uint64_t Bits = (3ull << (uint64_t)Index);
    const auto SizeInt = IR::OpSizeToSize(Size);
    if (((RegCache.Partial | RegCache.Cached) & Bits) == 0 && ((Offset / SizeInt) < 64)) {
-      auto Values = LoadContextPair_Uncached(RegClass, Size, Offset);
+      auto Values = LoadContextPair_Uncached(Class, Size, Offset);
      RegCache.Value[Index] = Values.Low;
      RegCache.Value[Index + 1] = Values.High;
      RegCache.Cached |= Bits;
@@ -1970,13 +2029,13 @@ private:

    // Fallback on a pair of loads
    return {
-      .Low = LoadRegCache(Offset, Index, RegClass, Size),
-      .High = LoadRegCache(Offset + SizeInt, Index + 1, RegClass, Size),
+      .Low = LoadRegCache(Offset, Index, Class, Size),
+      .High = LoadRegCache(Offset + SizeInt, Index + 1, Class, Size),
    };
  }

  Ref LoadGPR(uint8_t Reg) {
-    return LoadRegCache(Reg, GPR0Index + Reg, GPRClass, GetGPROpSize());
+    return LoadRegCache(Reg, GPR0Index + Reg, RegClass::GPR, GetGPROpSize());
  }

  Ref LoadContext(IR::OpSize Size, uint8_t Index) {
@@ -1992,7 +2051,7 @@ private:
  }

  Ref LoadXMMRegister(uint8_t Reg) {
-    return LoadRegCache(Reg, FPR0Index + Reg, FPRClass, GetGuestVectorLength());
+    return LoadRegCache(Reg, FPR0Index + Reg, RegClass::FPR, GetGuestVectorLength());
  }

  Ref LoadDF() {
@@ -2047,7 +2106,7 @@ private:
      // Recover the sign bit, it is the logical DF value
      return _Lshr(OpSize::i64Bit, LoadDF(), Constant(63));
    } else {
-      return _LoadContext(OpSize::i8Bit, GPRClass, offsetof(Core::CPUState, flags[BitOffset]));
+      return _LoadContextGPR(OpSize::i8Bit, offsetof(Core::CPUState, flags[BitOffset]));
    }
  }

@@ -2064,18 +2123,18 @@ private:
  }

  // Safe version of NZCVSelect that handles inverted carries automatically.
-  Ref NZCVSelect(OpSize OpSize, CondClassType Cond, Ref TrueV, Ref FalseV, bool CarryIsInverted = false) {
+  Ref NZCVSelect(OpSize OpSize, CondClass Cond, Ref TrueV, Ref FalseV, bool CarryIsInverted = false) {
    switch (Cond) {
-    case IR::COND_UGE: /* cs */
-    case IR::COND_ULT: /* cc */
+    case CondClass::UGE: /* cs */
+    case CondClass::ULT: /* cc */
      // Invert the condition to match our expectations.
      if (CarryIsInverted != CFInverted) {
-        Cond = {Cond == COND_UGE ? COND_ULT : COND_UGE};
+        Cond = (Cond == CondClass::UGE) ? CondClass::ULT : CondClass::UGE;
      }
      break;

-    case IR::COND_UGT: /* hi */
-    case IR::COND_ULE: /* ls */
+    case CondClass::UGT: /* hi */
+    case CondClass::ULE: /* ls */
      // No clever optimization we can do here, rectify carry itself.
      RectifyCarryInvert(CarryIsInverted);
      break;
@@ -2164,7 +2223,7 @@ private:

    HandleNZCV_RMW();
    CalculatePF(_ShiftFlags(OpSizeFromSrc(Op), Result, Dest, Shift, Src, OldPF, CFInverted));
-    StoreResult(GPRClass, Op, Result, OpSize::iInvalid);
+    StoreResultGPR(Op, Result);
  }

  // Helper to derive Dest by a given builder-using Expression with the opcode
@@ -2237,8 +2296,7 @@ private:
    CachedIndexedNamedVectorConstants.clear();
  }

-  std::optional<CondClassType> DecodeNZCVCondition(uint8_t OP);
-  Ref SelectBit(Ref Cmp, IR::OpSize ResultSize, Ref TrueValue, Ref FalseValue);
+  std::optional<CondClass> DecodeNZCVCondition(uint8_t OP);
  Ref SelectCC0All1(uint8_t OP);

  /**
@@ -2252,8 +2310,8 @@ private:
    if (Size != OpSize::i32Bit) {
      return;
    }
-    auto Dest = LoadSource(GPRClass, Op, Op->Dest, Op->Flags);
-    StoreResult(GPRClass, Op, Dest, OpSize::iInvalid);
+    auto Dest = LoadSourceGPR(Op, Op->Dest, Op->Flags);
+    StoreResultGPR(Op, Dest);
  }

  using ZeroShiftFunctionPtr = void (OpDispatchBuilder::*)(FEXCore::X86Tables::DecodedOp Op);
@@ -2288,7 +2346,7 @@ private:

    ///< Jump to zeroshift block or end block depending on if it was provided.
    IRPair<IROp_CodeBlock> TailHandling = ZeroShiftResult ? ZeroShiftBlock : EndBlock;
-    CondJump(Shift, Zero, TailHandling, SetBlock, {COND_EQ});
+    CondJump(Shift, Zero, TailHandling, SetBlock, CondClass::EQ);

    SetCurrentCodeBlock(SetBlock);
    StartNewBlock();
@@ -2331,9 +2389,7 @@ private:
  void CalculateFlags_MUL(IR::OpSize SrcSize, Ref Res, Ref High);
  void CalculateFlags_UMUL(Ref High);
  void CalculateFlags_Logical(IR::OpSize SrcSize, Ref Res);
-  void CalculateFlags_ShiftLeft(IR::OpSize SrcSize, Ref Res, Ref Src1, Ref Src2);
  void CalculateFlags_ShiftLeftImmediate(IR::OpSize SrcSize, Ref Res, Ref Src1, uint64_t Shift);
-  void CalculateFlags_ShiftRight(IR::OpSize SrcSize, Ref Res, Ref Src1, Ref Src2);
  void CalculateFlags_ShiftRightImmediate(IR::OpSize SrcSize, Ref Res, Ref Src1, uint64_t Shift);
  void CalculateFlags_ShiftRightDoubleImmediate(IR::OpSize SrcSize, Ref Res, Ref Src1, uint64_t Shift);
  void CalculateFlags_ShiftRightImmediateCommon(IR::OpSize SrcSize, Ref Res, Ref Src1, uint64_t Shift);
@@ -2349,17 +2405,19 @@ private:
  void ChgStateX87_MMX() override {
    LOGMAN_THROW_A_FMT(MMXState == MMXState_X87, "Expected state to be x87");
    _StackForceSlow();
-    SetX87Top(Constant(0));                             // top reset to zero
-    StoreContext(AbridgedFTWIndex, Constant(0xFFFFUL)); // all valid
+    SetX87Top(Constant(0)); // top reset to zero
+    _StoreContextGPR(OpSize::i8Bit, Constant(0xFFFFUL), offsetof(FEXCore::Core::CPUState, AbridgedFTW));
    MMXState = MMXState_MMX;
  }

  void ChgStateMMX_X87() override {
    LOGMAN_THROW_A_FMT(MMXState == MMXState_MMX, "Expected state to be MMX");
+    // The opcode dispatcher register cache is used for MMX, but the x87 pass register cache is used for x87, spill to
+    // context to ensure coherence.
+    FlushRegisterCache(false, true);
    // We explicitly initialize to x87 state in StartNewBlock.
    // So if we ever change this to do something else, we need to
    // make sure that we consider if we need to explicitly set it there.
-    FlushRegisterCache();
    MMXState = MMXState_X87;
  }

@@ -2385,44 +2443,62 @@ private:
  IROp_IRHeader* CurrentHeader {};

  [[nodiscard]]
-  bool IsTSOEnabled(FEXCore::IR::RegisterClassType Class) const {
+  bool IsTSOEnabled(RegClass Class) const {
    if (ForceTSO == ForceTSOMode::ForceEnabled) {
      return true;
    } else if (ForceTSO == ForceTSOMode::ForceDisabled) {
      return false;
-    } else if (Class == FPRClass) {
+    } else if (Class == RegClass::FPR) {
      return CTX->IsVectorAtomicTSOEnabled();
    } else {
      return CTX->IsAtomicTSOEnabled();
    }
  }

-  Ref _StoreMemAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, Ref Addr, Ref Value, IR::OpSize Align = IR::OpSize::i8Bit) {
+  Ref _StoreMemAutoTSO(RegClass Class, OpSize Size, Ref Addr, Ref Value, OpSize Align = OpSize::i8Bit) {
    if (IsTSOEnabled(Class)) {
-      return _StoreMemTSO(Class, Size, Value, Addr, Invalid(), Align, MEM_OFFSET_SXTX, 1);
+      return _StoreMemTSO(Class, Size, Value, Addr, Invalid(), Align, MemOffsetType::SXTX, 1);
    } else {
-      return _StoreMem(Class, Size, Value, Addr, Invalid(), Align, MEM_OFFSET_SXTX, 1);
+      return _StoreMem(Class, Size, Value, Addr, Invalid(), Align, MemOffsetType::SXTX, 1);
    }
  }
-
-  Ref _LoadMemAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, Ref ssa0, IR::OpSize Align = IR::OpSize::i8Bit) {
-    if (IsTSOEnabled(Class)) {
-      return _LoadMemTSO(Class, Size, ssa0, Invalid(), Align, MEM_OFFSET_SXTX, 1);
-    } else {
-      return _LoadMem(Class, Size, ssa0, Invalid(), Align, MEM_OFFSET_SXTX, 1);
-    }
+  Ref _StoreMemGPRAutoTSO(OpSize Size, Ref Addr, Ref Value, OpSize Align = OpSize::i8Bit) {
+    return _StoreMemAutoTSO(RegClass::GPR, Size, Addr, Value, Align);
+  }
+  Ref _StoreMemFPRAutoTSO(OpSize Size, Ref Addr, Ref Value, OpSize Align = OpSize::i8Bit) {
+    return _StoreMemAutoTSO(RegClass::FPR, Size, Addr, Value, Align);
  }

-  Ref _LoadMemAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, AddressMode A, IR::OpSize Align = IR::OpSize::i8Bit) {
-    bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
-    A = SelectAddressMode(this, A, GetGPROpSize(), CTX->HostFeatures.SupportsTSOImm9, AtomicTSO, Class != GPRClass, Size);
+  Ref _LoadMemAutoTSO(RegClass Class, OpSize Size, Ref ssa0, OpSize Align = OpSize::i8Bit) {
+    if (IsTSOEnabled(Class)) {
+      return _LoadMemTSO(Class, Size, ssa0, Invalid(), Align, MemOffsetType::SXTX, 1);
+    } else {
+      return _LoadMem(Class, Size, ssa0, Invalid(), Align, MemOffsetType::SXTX, 1);
+    }
+  }
+  Ref _LoadMemGPRAutoTSO(OpSize Size, Ref ssa0, OpSize Align = OpSize::i8Bit) {
+    return _LoadMemAutoTSO(RegClass::GPR, Size, ssa0, Align);
+  }
+  Ref _LoadMemFPRAutoTSO(OpSize Size, Ref ssa0, OpSize Align = OpSize::i8Bit) {
+    return _LoadMemAutoTSO(RegClass::FPR, Size, ssa0, Align);
+  }
+
+  Ref _LoadMemAutoTSO(RegClass Class, OpSize Size, const AddressMode& A, OpSize Align = OpSize::i8Bit) {
+    const bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
+    const auto B = SelectAddressMode(this, A, GetGPROpSize(), CTX->HostFeatures.SupportsTSOImm9, AtomicTSO, Class != RegClass::GPR, Size);

    if (AtomicTSO) {
-      return _LoadMemTSO(Class, Size, A.Base, A.Index, Align, A.IndexType, A.IndexScale);
+      return _LoadMemTSO(Class, Size, B.Base, B.Index, Align, B.IndexType, B.IndexScale);
    } else {
-      return _LoadMem(Class, Size, A.Base, A.Index, Align, A.IndexType, A.IndexScale);
+      return _LoadMem(Class, Size, B.Base, B.Index, Align, B.IndexType, B.IndexScale);
    }
  }
+  Ref _LoadMemGPRAutoTSO(OpSize Size, const AddressMode& A, OpSize Align = OpSize::i8Bit) {
+    return _LoadMemAutoTSO(RegClass::GPR, Size, A, Align);
+  }
+  Ref _LoadMemFPRAutoTSO(OpSize Size, const AddressMode& A, OpSize Align = OpSize::i8Bit) {
+    return _LoadMemAutoTSO(RegClass::FPR, Size, A, Align);
+  }

  AddressMode SelectPairAddressMode(AddressMode A, IR::OpSize Size) {
    LOGMAN_THROW_A_FMT(Size != IR::OpSize::iUnsized, "Invalid size!");
@@ -2440,56 +2516,72 @@ private:
  }


-  RefPair LoadMemPair(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, Ref Base, unsigned Offset) {
+  RefPair LoadMemPair(RegClass Class, OpSize Size, Ref Base, uint32_t Offset) {
    RefPair Values = AllocatePair(Class, Size);
    _LoadMemPair(Class, Size, Base, Offset, Values.Low, Values.High);
    return Values;
  }
+  RefPair LoadMemPairFPR(OpSize Size, Ref Base, uint32_t Offset) {
+    return LoadMemPair(RegClass::FPR, Size, Base, Offset);
+  }

-  RefPair _LoadMemPairAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, AddressMode A, IR::OpSize Align = IR::OpSize::i8Bit) {
-    bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
+  RefPair _LoadMemPairAutoTSO(RegClass Class, OpSize Size, const AddressMode& A, OpSize Align = OpSize::i8Bit) {
+    const bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;

    // Use ldp if possible, otherwise fallback on two loads.
-    if (!AtomicTSO && !A.Segment && Size >= OpSize::i32Bit & Size <= OpSize::i128Bit) {
-      A = SelectPairAddressMode(A, Size);
-      return LoadMemPair(Class, Size, A.Base, A.Offset);
-    } else {
-      AddressMode HighA = A;
-      HighA.Offset += 16;
-
-      return {
-        .Low = _LoadMemAutoTSO(Class, Size, A, Align),
-        .High = _LoadMemAutoTSO(Class, Size, HighA, Align),
-      };
+    if (!AtomicTSO && !A.Segment && Size >= OpSize::i32Bit && Size <= OpSize::i128Bit) {
+      const auto B = SelectPairAddressMode(A, Size);
+      return LoadMemPair(Class, Size, B.Base, B.Offset);
    }
+
+    AddressMode HighA = A;
+    HighA.Offset += 16;
+
+    return {
+      .Low = _LoadMemAutoTSO(Class, Size, A, Align),
+      .High = _LoadMemAutoTSO(Class, Size, HighA, Align),
+    };
+  }
+  RefPair _LoadMemPairFPRAutoTSO(OpSize Size, const AddressMode& A, OpSize Align = OpSize::i8Bit) {
+    return _LoadMemPairAutoTSO(RegClass::FPR, Size, A, Align);
  }

-  Ref _StoreMemAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, AddressMode A, Ref Value, IR::OpSize Align = IR::OpSize::i8Bit) {
-    bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
-    A = SelectAddressMode(this, A, GetGPROpSize(), CTX->HostFeatures.SupportsTSOImm9, AtomicTSO, Class != GPRClass, Size);
+  Ref _StoreMemAutoTSO(RegClass Class, OpSize Size, const AddressMode& A, Ref Value, OpSize Align = OpSize::i8Bit) {
+    const bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
+    const auto B = SelectAddressMode(this, A, GetGPROpSize(), CTX->HostFeatures.SupportsTSOImm9, AtomicTSO, Class != RegClass::GPR, Size);

    if (AtomicTSO) {
-      return _StoreMemTSO(Class, Size, Value, A.Base, A.Index, Align, A.IndexType, A.IndexScale);
+      return _StoreMemTSO(Class, Size, Value, B.Base, B.Index, Align, B.IndexType, B.IndexScale);
    } else {
-      return _StoreMem(Class, Size, Value, A.Base, A.Index, Align, A.IndexType, A.IndexScale);
+      return _StoreMem(Class, Size, Value, B.Base, B.Index, Align, B.IndexType, B.IndexScale);
    }
  }
+  Ref _StoreMemGPRAutoTSO(OpSize Size, const AddressMode& A, Ref Value, OpSize Align = OpSize::i8Bit) {
+    return _StoreMemAutoTSO(RegClass::GPR, Size, A, Value, Align);
+  }
+  Ref _StoreMemFPRAutoTSO(OpSize Size, const AddressMode& A, Ref Value, OpSize Align = OpSize::i8Bit) {
+    return _StoreMemAutoTSO(RegClass::FPR, Size, A, Value, Align);
+  }

-  void _StoreMemPairAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, AddressMode A, Ref Value1, Ref Value2,
-                            IR::OpSize Align = IR::OpSize::i8Bit) {
+  void _StoreMemPairAutoTSO(RegClass Class, OpSize Size, const AddressMode& A, Ref Value1, Ref Value2, OpSize Align = OpSize::i8Bit) {
    const auto SizeInt = IR::OpSizeToSize(Size);
-    bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
+    const bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;

    // Use stp if possible, otherwise fallback on two stores.
-    if (!AtomicTSO && !A.Segment && Size >= OpSize::i32Bit & Size <= OpSize::i128Bit) {
-      A = SelectPairAddressMode(A, Size);
-      _StoreMemPair(Class, Size, Value1, Value2, A.Base, A.Offset);
+    if (!AtomicTSO && !A.Segment && Size >= OpSize::i32Bit && Size <= OpSize::i128Bit) {
+      const auto B = SelectPairAddressMode(A, Size);
+      _StoreMemPair(Class, Size, Value1, Value2, B.Base, B.Offset);
    } else {
-      _StoreMemAutoTSO(Class, Size, A, Value1, OpSize::i8Bit);
-      A.Offset += SizeInt;
-      _StoreMemAutoTSO(Class, Size, A, Value2, OpSize::i8Bit);
+      auto B = A;
+
+      _StoreMemAutoTSO(Class, Size, B, Value1, OpSize::i8Bit);
+      B.Offset += SizeInt;
+      _StoreMemAutoTSO(Class, Size, B, Value2, OpSize::i8Bit);
    }
  }
+  void _StoreMemPairFPRAutoTSO(OpSize Size, const AddressMode& A, Ref Value1, Ref Value2, OpSize Align = OpSize::i8Bit) {
+    return _StoreMemPairAutoTSO(RegClass::FPR, Size, A, Value1, Value2, Align);
+  }

  Ref Pop(IR::OpSize Size, Ref SP_RMW) {
    Ref Value = _AllocateGPR(false);
@@ -35,20 +35,16 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_LoadSource_WithOpSize(
  } else {
    LOGMAN_THROW_A_FMT(IsOperandMem(Operand, true), "only memory sources");

-    AddressMode A = DecodeAddress(Op, Operand, AccessType, true /* IsLoad */);
-
-    AddressMode HighA = A;
-    HighA.Offset += 16;
-
    if (Operand.IsSIB()) {
      const bool IsVSIB = (Op->Flags & X86Tables::DecodeFlags::FLAG_VSIB_BYTE) != 0;
      LOGMAN_THROW_A_FMT(!IsVSIB, "VSIB uses LoadVSIB instead");
    }

+    const AddressMode A = DecodeAddress(Op, Operand, AccessType, true /* IsLoad */);
    if (NeedsHigh) {
-      return _LoadMemPairAutoTSO(FPRClass, OpSize::i128Bit, A, OpSize::i8Bit);
+      return _LoadMemPairFPRAutoTSO(OpSize::i128Bit, A, OpSize::i8Bit);
    } else {
-      return {.Low = _LoadMemAutoTSO(FPRClass, OpSize::i128Bit, A, OpSize::i8Bit)};
+      return {.Low = _LoadMemFPRAutoTSO(OpSize::i128Bit, A, OpSize::i8Bit)};
    }
  }
 }
@@ -56,7 +52,8 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_LoadSource_WithOpSize(
 OpDispatchBuilder::RefVSIB
 OpDispatchBuilder::AVX128_LoadVSIB(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, uint32_t Flags, bool NeedsHigh) {
  const bool IsVSIB = (Op->Flags & X86Tables::DecodeFlags::FLAG_VSIB_BYTE) != 0;
-  LOGMAN_THROW_A_FMT(Operand.IsSIB() && IsVSIB, "Trying to load VSIB for something that isn't the correct type!");
+  LOGMAN_THROW_A_FMT((Operand.IsSIB() || Operand.IsSIBRelocation()) && IsVSIB, "Trying to load VSIB for something that isn't the correct "
+                                                                               "type!");

  // VSIB is a very special case which has a ton of encoded data.
  // Get it in a format we can reason about.
@@ -68,13 +65,25 @@ OpDispatchBuilder::AVX128_LoadVSIB(const X86Tables::DecodedOp& Op, const X86Tabl
                     "Base must be a GPR.");
  const auto Index_XMM_gpr = Index_gpr - X86State::REG_XMM_0;

-  return {
+  OpDispatchBuilder::RefVSIB A {
    .Low = AVX128_LoadXMMRegister(Index_XMM_gpr, false),
    .High = NeedsHigh ? AVX128_LoadXMMRegister(Index_XMM_gpr, true) : Invalid(),
    .BaseAddr = Base_gpr != FEXCore::X86State::REG_INVALID ? LoadGPRRegister(Base_gpr, OpSize::i64Bit, 0, false) : nullptr,
-    .Displacement = Operand.Data.SIB.Offset,
    .Scale = Operand.Data.SIB.Scale,
  };
+
+  if (Operand.IsSIBRelocation()) {
+    auto EPOffset = _EntrypointOffset(OpSize::i64Bit, Operand.Data.SIB.Offset);
+    if (A.BaseAddr) {
+      A.BaseAddr = Add(OpSize::i64Bit, EPOffset, A.BaseAddr);
+    } else {
+      A.BaseAddr = EPOffset;
+    }
+  } else {
+    A.Displacement = static_cast<int32_t>(Operand.Data.SIB.Offset);
+  }
+
+  return A;
 }

 void OpDispatchBuilder::AVX128_StoreResult_WithOpSize(FEXCore::X86Tables::DecodedOp Op, const FEXCore::X86Tables::DecodedOperand& Operand,
@@ -95,9 +104,9 @@ void OpDispatchBuilder::AVX128_StoreResult_WithOpSize(FEXCore::X86Tables::Decode
    AddressMode A = DecodeAddress(Op, Operand, AccessType, false /* IsLoad */);

    if (Src.High) {
-      _StoreMemPairAutoTSO(FPRClass, OpSize::i128Bit, A, Src.Low, Src.High, OpSize::i8Bit);
+      _StoreMemPairFPRAutoTSO(OpSize::i128Bit, A, Src.Low, Src.High, OpSize::i8Bit);
    } else {
-      _StoreMemAutoTSO(FPRClass, OpSize::i128Bit, A, Src.Low, OpSize::i8Bit);
+      _StoreMemFPRAutoTSO(OpSize::i128Bit, A, Src.Low, OpSize::i8Bit);
    }
  }
 }
@@ -151,13 +160,13 @@ void OpDispatchBuilder::AVX128_VMOVScalarImpl(OpcodeArgs, IR::OpSize ElementSize
    AVX128_StoreResult_WithOpSize(Op, Op->Dest, RefPair {.Low = Result, .High = High});
  } else if (Op->Dest.IsGPR()) {
    // VMOVSS/SD xmm1, mem32/mem64
-    Ref Src = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], ElementSize, Op->Flags);
+    Ref Src = LoadSourceFPR_WithOpSize(Op, Op->Src[1], ElementSize, Op->Flags);
    auto High = LoadZeroVector(OpSize::i128Bit);
    AVX128_StoreResult_WithOpSize(Op, Op->Dest, RefPair {.Low = Src, .High = High});
  } else {
    // VMOVSS/SD mem32/mem64, xmm1
    auto Src = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, false);
-    StoreResult_WithOpSize(FPRClass, Op, Op->Dest, Src.Low, ElementSize, OpSize::iInvalid);
+    StoreResultFPR_WithOpSize(Op, Op->Dest, Src.Low, ElementSize);
  }
 }

@@ -351,7 +360,7 @@ void OpDispatchBuilder::AVX128_MOVVectorNT(OpcodeArgs) {
  if (Op->Dest.IsGPR()) {
    ///< MOVNTDQA load non-temporal comes from SSE4.1 and is extended by AVX/AVX2.
    RefPair Src {};
-    Ref SrcAddr = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags, {.LoadData = false});
+    Ref SrcAddr = LoadSourceGPR(Op, Op->Src[0], Op->Flags, {.LoadData = false});
    Src.Low = _VLoadNonTemporal(OpSize::i128Bit, SrcAddr, 0);

    if (Is128Bit) {
@@ -362,7 +371,7 @@ void OpDispatchBuilder::AVX128_MOVVectorNT(OpcodeArgs) {
    AVX128_StoreResult_WithOpSize(Op, Op->Dest, Src);
  } else {
    auto Src = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, !Is128Bit, MemoryAccessType::STREAM);
-    Ref Dest = LoadSource(GPRClass, Op, Op->Dest, Op->Flags, {.LoadData = false});
+    Ref Dest = LoadSourceGPR(Op, Op->Dest, Op->Flags, {.LoadData = false});

    if (Is128Bit) {
      // Single store non-temporal for 128-bit operations.
@@ -379,7 +388,7 @@ void OpDispatchBuilder::AVX128_MOVQ(OpcodeArgs) {
  if (Op->Src[0].IsGPR()) {
    Src = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, false);
  } else {
-    Src.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], OpSize::i64Bit, Op->Flags);
+    Src.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], OpSize::i64Bit, Op->Flags);
  }

  // This instruction is a bit special that if the destination is a register then it'll ZEXT the 64bit source to 256bit
@@ -390,7 +399,7 @@ void OpDispatchBuilder::AVX128_MOVQ(OpcodeArgs) {
    Src.High = ZeroVector;
    AVX128_StoreResult_WithOpSize(Op, Op->Dest, Src);
  } else {
-    StoreResult_WithOpSize(FPRClass, Op, Op->Dest, Src.Low, OpSize::i64Bit, OpSize::i64Bit);
+    StoreResultFPR_WithOpSize(Op, Op->Dest, Src.Low, OpSize::i64Bit, OpSize::i64Bit);
  }
 }

@@ -399,7 +408,7 @@ void OpDispatchBuilder::AVX128_VMOVLP(OpcodeArgs) {

  if (!Op->Dest.IsGPR()) {
    ///< VMOVLPS/PD mem64, xmm1
-    StoreResult_WithOpSize(FPRClass, Op, Op->Dest, Src1.Low, OpSize::i64Bit, OpSize::i64Bit);
+    StoreResultFPR_WithOpSize(Op, Op->Dest, Src1.Low, OpSize::i64Bit, OpSize::i64Bit);
  } else if (!Op->Src[1].IsGPR()) {
    ///< VMOVLPS/PD xmm1, xmm2, mem64
    // Bits[63:0] come from Src2[63:0]
@@ -463,7 +472,7 @@ void OpDispatchBuilder::AVX128_VMOVDDUP(OpcodeArgs) {
    // 128-bit operation only loads 8-bytes.
    // 256-bit operation loads a full 32-bytes.
    if (Is128Bit) {
-      Src.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], OpSize::i64Bit, Op->Flags);
+      Src.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], OpSize::i64Bit, Op->Flags);
    } else {
      Src = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, true);
    }
@@ -558,18 +567,18 @@ void OpDispatchBuilder::AVX128_InsertCVTGPR_To_FPR(OpcodeArgs, IR::OpSize DstEle

  if (Op->Src[1].IsGPR()) {
    // If the source is a GPR then convert directly from the GPR.
-    auto Src2 = LoadSource_WithOpSize(GPRClass, Op, Op->Src[1], GetGPROpSize(), Op->Flags);
+    auto Src2 = LoadSourceGPR_WithOpSize(Op, Op->Src[1], GetGPROpSize(), Op->Flags);
    Result.Low = _VSToFGPRInsert(OpSize::i128Bit, DstElementSize, SrcSize, Src1.Low, Src2, false);
  } else if (SrcSize != DstElementSize) {
    // If the source is from memory but the Source size and destination size aren't the same,
    // then it is more optimal to load in to a GPR and convert between GPR->FPR.
    // ARM GPR->FPR conversion supports different size source and destinations while FPR->FPR doesn't.
-    auto Src2 = LoadSource(GPRClass, Op, Op->Src[1], Op->Flags);
+    auto Src2 = LoadSourceGPR(Op, Op->Src[1], Op->Flags);
    Result.Low = _VSToFGPRInsert(DstSize, DstElementSize, SrcSize, Src1.Low, Src2, false);
  } else {
    // In the case of cvtsi2s{s,d} where the source and destination are the same size,
    // then it is more optimal to load in to the FPR register directly and convert there.
-    auto Src2 = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+    auto Src2 = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
    // Always signed
    Result.Low = _VSToFVectorInsert(DstSize, DstElementSize, DstElementSize, Src1.Low, Src2, false, false);
  }
@@ -589,11 +598,11 @@ void OpDispatchBuilder::AVX128_CVTFPR_To_GPR(OpcodeArgs, IR::OpSize SrcElementSi
  if (Op->Src[0].IsGPR()) {
    Src = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, false);
  } else {
-    Src.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], SrcElementSize, Op->Flags);
+    Src.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], SrcElementSize, Op->Flags);
  }

  Ref Result = CVTFPR_To_GPRImpl(Op, Src.Low, SrcElementSize, HostRoundingMode);
-  StoreResult(GPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultGPR(Op, Result);
 }

 void OpDispatchBuilder::AVX128_VANDN(OpcodeArgs) {
@@ -636,7 +645,7 @@ void OpDispatchBuilder::AVX128_UCOMISx(OpcodeArgs, IR::OpSize ElementSize) {
  if (Op->Src[0].IsGPR()) {
    Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, false);
  } else {
-    Src2.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], SrcSize, Op->Flags);
+    Src2.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], SrcSize, Op->Flags);
  }

  Comiss(ElementSize, Src1.Low, Src2.Low);
@@ -653,7 +662,7 @@ void OpDispatchBuilder::AVX128_VectorScalarInsertALU(OpcodeArgs, FEXCore::IR::IR
  if (Op->Src[1].IsGPR()) {
    Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, false);
  } else {
-    Src2.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], SrcSize, Op->Flags);
+    Src2.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[1], SrcSize, Op->Flags);
  }

  // If OpSize == ElementSize then it only does the lower scalar op
@@ -690,7 +699,7 @@ void OpDispatchBuilder::AVX128_InsertScalarFCMP(OpcodeArgs, IR::OpSize ElementSi
  if (Op->Src[1].IsGPR()) {
    Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, false);
  } else {
-    Src2.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], SrcSize, Op->Flags);
+    Src2.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[1], SrcSize, Op->Flags);
  }

  const uint8_t CompType = Op->Src[2].Literal();
@@ -708,12 +717,12 @@ void OpDispatchBuilder::AVX128_MOVBetweenGPR_FPR(OpcodeArgs) {
    RefPair Result {};
    if (Op->Src[0].IsGPR()) {
      // Loading from GPR and moving to Vector.
-      Ref Src = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], GetGPROpSize(), Op->Flags);
+      Ref Src = LoadSourceFPR_WithOpSize(Op, Op->Src[0], GetGPROpSize(), Op->Flags);
      // zext to 128bit
      Result.Low = _VCastFromGPR(OpSize::i128Bit, OpSizeFromSrc(Op), Src);
    } else {
      // Loading from Memory as a scalar. Zero extend
-      Result.Low = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      Result.Low = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    }

    Result.High = LoadZeroVector(OpSize::i128Bit);
@@ -726,11 +735,11 @@ void OpDispatchBuilder::AVX128_MOVBetweenGPR_FPR(OpcodeArgs) {
      auto ElementSize = OpSizeFromDst(Op);
      // Extract element from GPR. Zero extending in the process.
      Src.Low = _VExtractToGPR(OpSizeFromSrc(Op), ElementSize, Src.Low, 0);
-      StoreResult(GPRClass, Op, Op->Dest, Src.Low, OpSize::iInvalid);
+      StoreResultGPR(Op, Op->Dest, Src.Low);
    } else {
      // Storing first element to memory.
-      Ref Dest = LoadSource(GPRClass, Op, Op->Dest, Op->Flags, {.LoadData = false});
-      _StoreMem(FPRClass, OpSizeFromDst(Op), Dest, Src.Low, OpSize::i8Bit);
+      Ref Dest = LoadSourceGPR(Op, Op->Dest, Op->Flags, {.LoadData = false});
+      _StoreMemFPR(OpSizeFromDst(Op), Dest, Src.Low, OpSize::i8Bit);
    }
  }
 }
@@ -758,7 +767,7 @@ void OpDispatchBuilder::AVX128_PExtr(OpcodeArgs, IR::OpSize ElementSize) {
    const auto GPRSize = GetGPROpSize();
    // Extract already zero extends the result.
    Ref Result = _VExtractToGPR(OpSize::i128Bit, OverridenElementSize, Src.Low, Index);
-    StoreResult_WithOpSize(GPRClass, Op, Op->Dest, Result, GPRSize, OpSize::iInvalid);
+    StoreResultGPR_WithOpSize(Op, Op->Dest, Result, GPRSize);
    return;
  }

@@ -779,7 +788,7 @@ void OpDispatchBuilder::AVX128_ExtendVectorElements(OpcodeArgs, IR::OpSize Eleme
      const auto SrcSize = OpSizeFromSrc(Op);
      const auto LoadSize = Is256Bit ? IR::SizeToOpSize(IR::OpSizeToSize(SrcSize) * 2) : SrcSize;

-      return LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], LoadSize, Op->Flags);
+      return LoadSourceFPR_WithOpSize(Op, Op->Src[0], LoadSize, Op->Flags);
    }
  };

@@ -868,7 +877,7 @@ void OpDispatchBuilder::AVX128_MOVMSK(OpcodeArgs, IR::OpSize ElementSize) {
    auto GPRHigh = Mask8Byte(Src.High);
    GPR = _Orlshl(OpSize::i64Bit, GPRLow, GPRHigh, 2);
  }
-  StoreResult_WithOpSize(GPRClass, Op, Op->Dest, GPR, GetGPROpSize(), OpSize::iInvalid);
+  StoreResultGPR_WithOpSize(Op, Op->Dest, GPR, GetGPROpSize());
 }

 void OpDispatchBuilder::AVX128_MOVMSKB(OpcodeArgs) {
@@ -897,7 +906,7 @@ void OpDispatchBuilder::AVX128_MOVMSKB(OpcodeArgs) {
    Result = _Orlshl(OpSize::i64Bit, Result, ResultHigh, 16);
  }

-  StoreResult(GPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultGPR(Op, Result);
 }

 void OpDispatchBuilder::AVX128_PINSRImpl(OpcodeArgs, IR::OpSize ElementSize, const X86Tables::DecodedOperand& Src1Op,
@@ -910,7 +919,7 @@ void OpDispatchBuilder::AVX128_PINSRImpl(OpcodeArgs, IR::OpSize ElementSize, con

  if (Src2Op.IsGPR()) {
    // If the source is a GPR then convert directly from the GPR.
-    auto Src2 = LoadSource_WithOpSize(GPRClass, Op, Src2Op, GetGPROpSize(), Op->Flags);
+    auto Src2 = LoadSourceGPR_WithOpSize(Op, Src2Op, GetGPROpSize(), Op->Flags);
    Result.Low = _VInsGPR(OpSize::i128Bit, ElementSize, Index, Src1.Low, Src2);
  } else {
    // If loading from memory then we only load the element size
@@ -1047,7 +1056,7 @@ void OpDispatchBuilder::AVX128_InsertScalar_CVT_Float_To_Float(OpcodeArgs, IR::O
  // Then zero extends the top 128-bit.
  const auto SrcSize = Op->Src[1].IsGPR() ? OpSize::i128Bit : SrcElementSize;
  auto Src1 = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, false);
-  Ref Src2 = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], SrcSize, Op->Flags, {.AllowUpperGarbage = true});
+  Ref Src2 = LoadSourceFPR_WithOpSize(Op, Op->Src[1], SrcSize, Op->Flags, {.AllowUpperGarbage = true});

  Ref Result = _VFToFScalarInsert(OpSize::i128Bit, DstElementSize, SrcElementSize, Src1.Low, Src2, false);
  AVX128_StoreResult_WithOpSize(Op, Op->Dest, AVX128_Zext(Result));
@@ -1076,7 +1085,7 @@ void OpDispatchBuilder::AVX128_Vector_CVT_Float_To_Float(OpcodeArgs, IR::OpSize
  } else {
    // Handle 64-bit memory source.
    // In the case of cvtps2pd xmm, m64.
-    Src.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], LoadSize, Op->Flags);
+    Src.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], LoadSize, Op->Flags);
  }

  RefPair Result {};
@@ -1154,7 +1163,7 @@ void OpDispatchBuilder::AVX128_Vector_CVT_Int_To_Float(OpcodeArgs, IR::OpSize Sr
      // unnecessarily zero extend the vector. Otherwise, if
      // memory, then we want to load the element size exactly.
      const auto LoadSize = IR::SizeToOpSize(8 * (IR::OpSizeToSize(Size) / 16));
-      return RefPair {.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], LoadSize, Op->Flags)};
+      return RefPair {.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], LoadSize, Op->Flags)};
    } else {
      return AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, !Is128Bit);
    }
@@ -1304,7 +1313,7 @@ void OpDispatchBuilder::AVX128_InsertScalarRound(OpcodeArgs, IR::OpSize ElementS
  if (Op->Src[1].IsGPR()) {
    Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, false);
  } else {
-    Src2.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], SrcSize, Op->Flags);
+    Src2.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[1], SrcSize, Op->Flags);
  }

  // If OpSize == ElementSize then it only does the lower scalar op
@@ -1573,20 +1582,20 @@ void OpDispatchBuilder::AVX128_VMASKMOVImpl(OpcodeArgs, IR::OpSize ElementSize,
    auto Address = MakeAddress(Op->Dest);

    auto Data = AVX128_LoadSource_WithOpSize(Op, DataOp, Op->Flags, !Is128Bit);
-    _VStoreVectorMasked(OpSize::i128Bit, ElementSize, Mask.Low, Data.Low, Address, Invalid(), MEM_OFFSET_SXTX, 1);
+    _VStoreVectorMasked(OpSize::i128Bit, ElementSize, Mask.Low, Data.Low, Address, Invalid(), MemOffsetType::SXTX, 1);
    if (!Is128Bit) {
-      _VStoreVectorMasked(OpSize::i128Bit, ElementSize, Mask.High, Data.High, Address, _InlineConstant(16), MEM_OFFSET_SXTX, 1);
+      _VStoreVectorMasked(OpSize::i128Bit, ElementSize, Mask.High, Data.High, Address, _InlineConstant(16), MemOffsetType::SXTX, 1);
    }
  } else {
    auto Address = MakeAddress(DataOp);

    RefPair Result {};
-    Result.Low = _VLoadVectorMasked(OpSize::i128Bit, ElementSize, Mask.Low, Address, Invalid(), MEM_OFFSET_SXTX, 1);
+    Result.Low = _VLoadVectorMasked(OpSize::i128Bit, ElementSize, Mask.Low, Address, Invalid(), MemOffsetType::SXTX, 1);

    if (Is128Bit) {
      Result.High = LoadZeroVector(OpSize::i128Bit);
    } else {
-      Result.High = _VLoadVectorMasked(OpSize::i128Bit, ElementSize, Mask.High, Address, _InlineConstant(16), MEM_OFFSET_SXTX, 1);
+      Result.High = _VLoadVectorMasked(OpSize::i128Bit, ElementSize, Mask.High, Address, _InlineConstant(16), MemOffsetType::SXTX, 1);
    }
    AVX128_StoreResult_WithOpSize(Op, Op->Dest, Result);
  }
@@ -1616,11 +1625,11 @@ void OpDispatchBuilder::AVX128_MASKMOV(OpcodeArgs) {
  // RDI source (DS prefix by default)
  auto MemDest = MakeSegmentAddress(X86State::REG_RDI, Op->Flags, X86Tables::DecodeFlags::FLAG_DS_PREFIX);

-  Ref XMMReg = _LoadMem(FPRClass, Size, MemDest, OpSize::i8Bit);
+  Ref XMMReg = _LoadMemFPR(Size, MemDest, OpSize::i8Bit);

  // If the Mask element high bit is set then overwrite the element with the source, else keep the memory variant
  XMMReg = _VBSL(Size, MaskSrc.Low, VectorSrc.Low, XMMReg);
-  _StoreMem(FPRClass, Size, MemDest, XMMReg, OpSize::i8Bit);
+  _StoreMemFPR(Size, MemDest, XMMReg, OpSize::i8Bit);
 }

 void OpDispatchBuilder::AVX128_VectorVariableBlend(OpcodeArgs, IR::OpSize ElementSize) {
@@ -1660,7 +1669,7 @@ void OpDispatchBuilder::AVX128_SaveAVXState(Ref MemBase) {

  for (uint32_t i = 0; i < NumRegs; i += 2) {
    RefPair Pair = LoadContextPair(OpSize::i128Bit, AVXHigh0Index + i);
-    _StoreMemPair(FPRClass, OpSize::i128Bit, Pair.Low, Pair.High, MemBase, i * 16 + 576);
+    _StoreMemPairFPR(OpSize::i128Bit, Pair.Low, Pair.High, MemBase, i * 16 + 576);
  }
 }

@@ -1668,7 +1677,7 @@ void OpDispatchBuilder::AVX128_RestoreAVXState(Ref MemBase) {
  const auto NumRegs = Is64BitMode ? 16U : 8U;

  for (uint32_t i = 0; i < NumRegs; i += 2) {
-    auto YMMHRegs = LoadMemPair(FPRClass, OpSize::i128Bit, MemBase, i * 16 + 576);
+    auto YMMHRegs = LoadMemPairFPR(OpSize::i128Bit, MemBase, i * 16 + 576);

    AVX128_StoreXMMRegister(i, YMMHRegs.Low, true);
    AVX128_StoreXMMRegister(i + 1, YMMHRegs.High, true);
@@ -1960,7 +1969,7 @@ void OpDispatchBuilder::AVX128_VFMAImpl(OpcodeArgs, IROps IROp, uint8_t Src1Idx,
 }

 void OpDispatchBuilder::AVX128_VFMAScalarImpl(OpcodeArgs, IROps IROp, uint8_t Src1Idx, uint8_t Src2Idx, uint8_t AddendIdx) {
-  const auto SrcSize = OpSizeFromSrc(Op);
+  const OpSize ElementSize = Op->Flags & X86Tables::DecodeFlags::FLAG_OPTION_AVX_W ? OpSize::i64Bit : OpSize::i32Bit;

  auto Dest = AVX128_LoadSource_WithOpSize(Op, Op->Dest, Op->Flags, false).Low;
  auto Src1 = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, false).Low;
@@ -1968,13 +1977,13 @@ void OpDispatchBuilder::AVX128_VFMAScalarImpl(OpcodeArgs, IROps IROp, uint8_t Sr
  if (Op->Src[1].IsGPR()) {
    Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, false).Low;
  } else {
-    Src2 = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], SrcSize, Op->Flags);
+    Src2 = LoadSourceFPR_WithOpSize(Op, Op->Src[1], ElementSize, Op->Flags);
  }

  Ref Sources[3] = {Dest, Src1, Src2};

  DeriveOp(Result_Low, IROp,
-           _VFMLAScalarInsert(OpSize::i128Bit, SrcSize, Dest, Sources[Src1Idx - 1], Sources[Src2Idx - 1], Sources[AddendIdx - 1]));
+           _VFMLAScalarInsert(OpSize::i128Bit, ElementSize, Dest, Sources[Src1Idx - 1], Sources[Src2Idx - 1], Sources[AddendIdx - 1]));
  AVX128_StoreResult_WithOpSize(Op, Op->Dest, AVX128_Zext(Result_Low));
 }

@@ -2016,8 +2025,8 @@ void OpDispatchBuilder::AVX128_VFMAddSubImpl(OpcodeArgs, bool AddSub, uint8_t Sr
  AVX128_StoreResult_WithOpSize(Op, Op->Dest, Result);
 }

-OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherImpl(OpSize Size, OpSize ElementLoadSize, OpSize AddrElementSize, RefPair Dest,
-                                                                  RefPair Mask, RefVSIB VSIB) {
+OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherImpl(OpcodeArgs, OpSize Size, OpSize ElementLoadSize, OpSize AddrElementSize,
+                                                                  RefPair Dest, RefPair Mask, RefVSIB VSIB) {
  LOGMAN_THROW_A_FMT(AddrElementSize == OpSize::i32Bit || AddrElementSize == OpSize::i64Bit, "Unknown address element size");
  const auto Is128Bit = Size == OpSize::i128Bit;

@@ -2061,10 +2070,13 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherImpl(OpSize Size, O
    }
  }

+  const auto GPRSize = GetGPROpSize();
+  auto AddrSize = (Op->Flags & X86Tables::DecodeFlags::FLAG_ADDRESS_SIZE) != 0 ? (GPRSize >> 1) : GPRSize;
+
  RefPair Result {};
  ///< Calculate the low-half.
  Result.Low = _VLoadVectorGatherMasked(OpSize::i128Bit, ElementLoadSize, Dest.Low, Mask.Low, BaseAddr, VSIB.Low, VSIB.High,
-                                        AddrElementSize, VSIB.Scale, 0, 0);
+                                        AddrElementSize, VSIB.Scale, 0, 0, AddrSize);

  if (Is128Bit) {
    Result.High = LoadZeroVector(OpSize::i128Bit);
@@ -2101,7 +2113,7 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherImpl(OpSize Size, O

    ///< Calculate the high-half.
    auto ResultHigh = _VLoadVectorGatherMasked(OpSize::i128Bit, ElementLoadSize, DestReg, MaskReg, BaseAddr, AddrAddressing.Low,
-                                               AddrAddressing.High, AddrElementSize, VSIB.Scale, DataElementOffset, IndexElementOffset);
+                                               AddrAddressing.High, AddrElementSize, VSIB.Scale, DataElementOffset, IndexElementOffset, AddrSize);

    if (AddrElementSize == OpSize::i64Bit && ElementLoadSize == OpSize::i32Bit) {
      // If we only fetched 128-bits worth of data then the upper-result is all zero.
@@ -2114,7 +2126,7 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherImpl(OpSize Size, O
  return Result;
 }

-OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherQPSImpl(Ref Dest, Ref Mask, RefVSIB VSIB) {
+OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherQPSImpl(OpcodeArgs, Ref Dest, Ref Mask, RefVSIB VSIB) {

  ///< BaseAddr doesn't need to exist, calculate that here.
  Ref BaseAddr = VSIB.BaseAddr;
@@ -2142,8 +2154,11 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherQPSImpl(Ref Dest, R

  RefPair Result {};

+  const auto GPRSize = GetGPROpSize();
+  auto AddrSize = (Op->Flags & X86Tables::DecodeFlags::FLAG_ADDRESS_SIZE) != 0 ? (GPRSize >> 1) : GPRSize;
+
  ///< Calculate the low-half.
-  Result.Low = _VLoadVectorGatherMaskedQPS(OpSize::i128Bit, OpSize::i32Bit, Dest, Mask, BaseAddr, VSIB.Low, VSIB.High, VSIB.Scale);
+  Result.Low = _VLoadVectorGatherMaskedQPS(OpSize::i128Bit, OpSize::i32Bit, Dest, Mask, BaseAddr, VSIB.Low, VSIB.High, VSIB.Scale, AddrSize);
  Result.High = LoadZeroVector(OpSize::i128Bit);
  if (VSIB.High == Invalid()) {
    // Special case for only loading two floats.
@@ -2202,15 +2217,15 @@ void OpDispatchBuilder::AVX128_VPGATHER(OpcodeArgs, OpSize AddrElementSize) {
    }

    ///< AddressElementSize is now OpSize::i64Bit
-    Result = AVX128_VPGatherQPSImpl(Dest.Low, Mask.Low, VSIBLow);
+    Result = AVX128_VPGatherQPSImpl(Op, Dest.Low, Mask.Low, VSIBLow);
    if (NeedsHighAddrBytes) {
-      auto Res = AVX128_VPGatherQPSImpl(Dest.High, Mask.High, VSIBHigh);
+      auto Res = AVX128_VPGatherQPSImpl(Op, Dest.High, Mask.High, VSIBHigh);
      Result.High = Res.Low;
    }
  } else if (AddrElementSize == OpSize::i64Bit && ElementLoadSize == OpSize::i32Bit) {
-    Result = AVX128_VPGatherQPSImpl(Dest.Low, Mask.Low, VSIB);
+    Result = AVX128_VPGatherQPSImpl(Op, Dest.Low, Mask.Low, VSIB);
  } else {
-    Result = AVX128_VPGatherImpl(Size, ElementLoadSize, AddrElementSize, Dest, Mask, VSIB);
+    Result = AVX128_VPGatherImpl(Op, Size, ElementLoadSize, AddrElementSize, Dest, Mask, VSIB);
  }
  AVX128_StoreResult_WithOpSize(Op, Op->Dest, Result);

@@ -2234,7 +2249,7 @@ void OpDispatchBuilder::AVX128_VCVTPH2PS(OpcodeArgs) {
    // In the event that a memory operand is used as the source operand,
    // the access width will always be half the size of the destination vector width
    // (i.e. 128-bit vector -> 64-bit mem, 256-bit vector -> 128-bit mem)
-    Src.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], SrcSize, Op->Flags);
+    Src.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], SrcSize, Op->Flags);
  }

  RefPair Result {};
@@ -2289,7 +2304,7 @@ void OpDispatchBuilder::AVX128_VCVTPS2PH(OpcodeArgs) {
  }

  if (!Op->Dest.IsGPR()) {
-    StoreResult_WithOpSize(FPRClass, Op, Op->Dest, Result.Low, StoreSize, OpSize::iInvalid);
+    StoreResultFPR_WithOpSize(Op, Op->Dest, Result.Low, StoreSize);
  } else {
    AVX128_StoreResult_WithOpSize(Op, Op->Dest, Result);
  }
@@ -23,8 +23,8 @@ void OpDispatchBuilder::SHA1NEXTEOp(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);

  // ARMv8 SHA1 extension provides a `SHA1H` instruction which does a fixed rotate by 30.
  // This only operates on element 0 rather than element 3. We don't have the luxury of rewriting the x86 SHA algorithm to take advantage of this.
@@ -36,7 +36,7 @@ void OpDispatchBuilder::SHA1NEXTEOp(OpcodeArgs) {
  auto Tmp = _VAdd(OpSize::i128Bit, OpSize::i32Bit, Src, RotatedNode);
  auto Result = _VInsElement(OpSize::i128Bit, OpSize::i32Bit, 3, 3, Src, Tmp);

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::SHA1MSG1Op(OpcodeArgs) {
@@ -44,15 +44,15 @@ void OpDispatchBuilder::SHA1MSG1Op(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);

  Ref NewVec = _VExtr(OpSize::i128Bit, OpSize::i64Bit, Dest, Src, 1);

  // [W0, W1, W2, W3] ^ [W2, W3, W4, W5]
  Ref Result = _VXor(OpSize::i128Bit, OpSize::i8Bit, Dest, NewVec);

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::SHA1MSG2Op(OpcodeArgs) {
@@ -60,8 +60,8 @@ void OpDispatchBuilder::SHA1MSG2Op(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);

  // ARM SHA1 mostly matches x86 semantics, except the input and outputs are both flipped from elements 0,1,2,3 to 3,2,1,0.
  auto Src1 = SHADataShuffle(Dest);
@@ -70,7 +70,7 @@ void OpDispatchBuilder::SHA1MSG2Op(OpcodeArgs) {
  // The result is swizzled differently than expected
  auto Result = SHADataShuffle(_VSha1SU1(Src1, Src2));

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::SHA1RNDS4Op(OpcodeArgs) {
@@ -79,8 +79,8 @@ void OpDispatchBuilder::SHA1RNDS4Op(OpcodeArgs) {
    return;
  }
  const uint64_t Imm8 = Op->Src[1].Literal() & 0b11;
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);

  Ref Result {};
  Ref ConstantVector {};
@@ -112,7 +112,7 @@ void OpDispatchBuilder::SHA1RNDS4Op(OpcodeArgs) {
  case 3: Result = SHADataShuffle(_VSha1P(Src1, ZeroRegister, Src2)); break;
  }

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::SHA256MSG1Op(OpcodeArgs) {
@@ -120,12 +120,12 @@ void OpDispatchBuilder::SHA256MSG1Op(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);

  auto Result = _VSha256U0(Dest, Src);

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::SHA256MSG2Op(OpcodeArgs) {
@@ -133,8 +133,8 @@ void OpDispatchBuilder::SHA256MSG2Op(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);

  auto Src1 = _VExtr(OpSize::i128Bit, OpSize::i32Bit, Dest, Dest, 3);
  auto DupDst = _VDupElement(OpSize::i128Bit, OpSize::i32Bit, Dest, 3);
@@ -142,7 +142,7 @@ void OpDispatchBuilder::SHA256MSG2Op(OpcodeArgs) {

  auto Result = _VSha256U1(Src1, Src2);

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::SHA256RNDS2Op(OpcodeArgs) {
@@ -150,8 +150,8 @@ void OpDispatchBuilder::SHA256RNDS2Op(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  // Hardcoded to XMM0
  auto XMM0 = LoadXMMRegister(0);

@@ -177,7 +177,7 @@ void OpDispatchBuilder::SHA256RNDS2Op(OpcodeArgs) {
  auto B = _VSha256H2(EFGH, ABCD, Key);
  auto Result = shuffle_abcd(A, B);

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::AESImcOp(OpcodeArgs) {
@@ -185,9 +185,9 @@ void OpDispatchBuilder::AESImcOp(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  Ref Result = _VAESImc(Src);
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::AESEncOp(OpcodeArgs) {
@@ -195,10 +195,10 @@ void OpDispatchBuilder::AESEncOp(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  Ref Result = _VAESEnc(OpSize::i128Bit, Dest, Src, LoadZeroVector(OpSize::i128Bit));
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::VAESEncOp(OpcodeArgs) {
@@ -208,11 +208,11 @@ void OpDispatchBuilder::VAESEncOp(OpcodeArgs) {
  // TODO: Handle 256-bit VAESENC.
  LOGMAN_THROW_A_FMT(Is128Bit, "256-bit VAESENC unimplemented");

-  Ref State = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
-  Ref Key = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+  Ref State = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
+  Ref Key = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
  Ref Result = _VAESEnc(DstSize, State, Key, LoadZeroVector(DstSize));

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::AESEncLastOp(OpcodeArgs) {
@@ -220,10 +220,10 @@ void OpDispatchBuilder::AESEncLastOp(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  Ref Result = _VAESEncLast(OpSize::i128Bit, Dest, Src, LoadZeroVector(OpSize::i128Bit));
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::VAESEncLastOp(OpcodeArgs) {
@@ -233,11 +233,11 @@ void OpDispatchBuilder::VAESEncLastOp(OpcodeArgs) {
  // TODO: Handle 256-bit VAESENCLAST.
  LOGMAN_THROW_A_FMT(Is128Bit, "256-bit VAESENCLAST unimplemented");

-  Ref State = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
-  Ref Key = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+  Ref State = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
+  Ref Key = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
  Ref Result = _VAESEncLast(DstSize, State, Key, LoadZeroVector(DstSize));

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::AESDecOp(OpcodeArgs) {
@@ -245,10 +245,10 @@ void OpDispatchBuilder::AESDecOp(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  Ref Result = _VAESDec(OpSize::i128Bit, Dest, Src, LoadZeroVector(OpSize::i128Bit));
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::VAESDecOp(OpcodeArgs) {
@@ -258,11 +258,11 @@ void OpDispatchBuilder::VAESDecOp(OpcodeArgs) {
  // TODO: Handle 256-bit VAESDEC.
  LOGMAN_THROW_A_FMT(Is128Bit, "256-bit VAESDEC unimplemented");

-  Ref State = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
-  Ref Key = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+  Ref State = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
+  Ref Key = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
  Ref Result = _VAESDec(DstSize, State, Key, LoadZeroVector(DstSize));

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::AESDecLastOp(OpcodeArgs) {
@@ -270,10 +270,10 @@ void OpDispatchBuilder::AESDecLastOp(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  Ref Result = _VAESDecLast(OpSize::i128Bit, Dest, Src, LoadZeroVector(OpSize::i128Bit));
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::VAESDecLastOp(OpcodeArgs) {
@@ -283,15 +283,15 @@ void OpDispatchBuilder::VAESDecLastOp(OpcodeArgs) {
  // TODO: Handle 256-bit VAESDECLAST.
  LOGMAN_THROW_A_FMT(Is128Bit, "256-bit VAESDECLAST unimplemented");

-  Ref State = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
-  Ref Key = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+  Ref State = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
+  Ref Key = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
  Ref Result = _VAESDecLast(DstSize, State, Key, LoadZeroVector(DstSize));

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 Ref OpDispatchBuilder::AESKeyGenAssistImpl(OpcodeArgs) {
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  const uint64_t RCON = Op->Src[1].Literal();

  auto KeyGenSwizzle = LoadAndCacheNamedVectorConstant(OpSize::i128Bit, NAMED_VECTOR_AESKEYGENASSIST_SWIZZLE);
@@ -305,7 +305,7 @@ void OpDispatchBuilder::AESKeyGenAssist(OpcodeArgs) {
  }

  Ref Result = AESKeyGenAssistImpl(Op);
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::PCLMULQDQOp(OpcodeArgs) {
@@ -313,12 +313,12 @@ void OpDispatchBuilder::PCLMULQDQOp(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  const auto Selector = static_cast<uint8_t>(Op->Src[1].Literal());

  auto Res = _PCLMUL(OpSize::i128Bit, Dest, Src, Selector & 0b1'0001);
-  StoreResult(FPRClass, Op, Res, OpSize::iInvalid);
+  StoreResultFPR(Op, Res);
 }

 void OpDispatchBuilder::VPCLMULQDQOp(OpcodeArgs) {
@@ -328,12 +328,12 @@ void OpDispatchBuilder::VPCLMULQDQOp(OpcodeArgs) {
  }
  const auto DstSize = OpSizeFromDst(Op);

-  Ref Src1 = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
-  Ref Src2 = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+  Ref Src1 = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
+  Ref Src2 = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
  const auto Selector = static_cast<uint8_t>(Op->Src[2].Literal());

  Ref Res = _PCLMUL(DstSize, Src1, Src2, Selector & 0b1'0001);
-  StoreResult(FPRClass, Op, Res, OpSize::iInvalid);
+  StoreResultFPR(Op, Res);
 }

 } // namespace FEXCore::IR
@@ -263,7 +263,7 @@ void OpDispatchBuilder::CalculateDeferredFlags() {
 Ref OpDispatchBuilder::IncrementByCarry(OpSize OpSize, Ref Src) {
  // If CF not inverted, we use .cc since the increment happens when the
  // condition is false. If CF inverted, invert to use .cs. A bit mindbendy.
-  return _NZCVSelectIncrement(OpSize, {CFInverted ? COND_UGE : COND_ULT}, Src, Src);
+  return _NZCVSelectIncrement(OpSize, CFInverted ? CondClass::UGE : CondClass::ULT, Src, Src);
 }

 Ref OpDispatchBuilder::CalculateFlags_ADC(IR::OpSize SrcSize, Ref Src1, Ref Src2) {
@@ -290,7 +290,7 @@ Ref OpDispatchBuilder::CalculateFlags_ADC(IR::OpSize SrcSize, Ref Src1, Ref Src2
    Res = _Bfe(OpSize, IR::OpSizeAsBits(SrcSize), 0, Res);

    // TODO: We can fold that second Bfe in (cmp uxth).
-    auto SelectCFInv = Select01(OpSize, CondClassType {COND_UGE}, Res, Src2PlusCF);
+    auto SelectCFInv = Select01(OpSize, CondClass::UGE, Res, Src2PlusCF);

    SetNZ_ZeroCV(SrcSize, Res);
    SetCFInverted(SelectCFInv);
@@ -324,7 +324,7 @@ Ref OpDispatchBuilder::CalculateFlags_SBB(IR::OpSize SrcSize, Ref Src1, Ref Src2
    Res = Sub(OpSize, Src1, Src2PlusCF);
    Res = _Bfe(OpSize, IR::OpSizeAsBits(SrcSize), 0, Res);

-    auto SelectCFInv = Select01(OpSize, CondClassType {COND_UGE}, Src1, Src2PlusCF);
+    auto SelectCFInv = Select01(OpSize, CondClass::UGE, Src1, Src2PlusCF);

    SetNZ_ZeroCV(SrcSize, Res);
    SetCFInverted(SelectCFInv);
@@ -406,7 +406,7 @@ void OpDispatchBuilder::CalculateFlags_MUL(IR::OpSize SrcSize, Ref Res, Ref High
  // If High = SignBit, then sets to nZCv. Else sets to nzcV. Since SF/ZF
  // undefined, this does what we need after inverting carry.
  auto Zero = _InlineConstant(0);
-  _CondSubNZCV(OpSize::i64Bit, Zero, Zero, CondClassType {COND_EQ}, 0x1 /* nzcV */);
+  _CondSubNZCV(OpSize::i64Bit, Zero, Zero, CondClass::EQ, 0x1 /* nzcV */);
  CFInverted = true;
 }

@@ -423,7 +423,7 @@ void OpDispatchBuilder::CalculateFlags_UMUL(Ref High) {

  // If High = 0, then sets to nZCv. Else sets to nzcV. Since SF/ZF undefined,
  // this does what we need.
-  _CondSubNZCV(Size, Zero, Zero, CondClassType {COND_EQ}, 0x1 /* nzcV */);
+  _CondSubNZCV(Size, Zero, Zero, CondClass::EQ, 0x1 /* nzcV */);
  CFInverted = true;
 }

@@ -151,6 +151,9 @@ constexpr DispatchTableEntry OpDispatch_SecondaryGroupTables[] = {
  {OPD(FEXCore::X86Tables::TYPE_GROUP_16, PF_F2, 3), 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::Prefetch, false, false, 3>},
  {OPD(FEXCore::X86Tables::TYPE_GROUP_16, PF_F2, 4), 4, &OpDispatchBuilder::NOPOp},

+  // GROUP 17
+  {OPD(FEXCore::X86Tables::TYPE_GROUP_17, PF_66, 0), 1, &OpDispatchBuilder::Extrq_imm},
+
  // GROUP P
  {OPD(FEXCore::X86Tables::TYPE_GROUP_P, PF_NONE, 0), 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::Prefetch, false, false, 1>},
  {OPD(FEXCore::X86Tables::TYPE_GROUP_P, PF_NONE, 1), 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::Prefetch, true, false, 1>},
@@ -145,7 +145,7 @@ constexpr DispatchTableEntry OpDispatch_TwoByteOpTable[] = {

 #ifndef _WIN32
  // FEX reserved instructions
-  {0x37, 1, &OpDispatchBuilder::CallbackReturnOp},
+  {0x3E, 1, &OpDispatchBuilder::CallbackReturnOp},
  {0x3F, 1, &OpDispatchBuilder::ThunkOp},
 #endif
 };
@@ -198,6 +198,8 @@ constexpr DispatchTableEntry OpDispatch_SecondaryRepNEModTables[] = {
  {0x5E, 1, &OpDispatchBuilder::VectorScalarInsertALUOp<IR::OP_VFDIVSCALARINSERT, OpSize::i64Bit>},
  {0x5F, 1, &OpDispatchBuilder::VectorScalarInsertALUOp<IR::OP_VFMAXSCALARINSERT, OpSize::i64Bit>},
  {0x70, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::PSHUFWOp, true>},
+  {0x78, 1, &OpDispatchBuilder::Insertq_imm},
+  {0x79, 1, &OpDispatchBuilder::Insertq},
  {0x7C, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::VectorALUOp, IR::OP_VFADDP, OpSize::i32Bit>},
  {0x7D, 1, &OpDispatchBuilder::HSUBP<OpSize::i32Bit>},
  {0xD0, 1, &OpDispatchBuilder::ADDSUBPOp<OpSize::i32Bit>},
@@ -256,6 +258,7 @@ constexpr DispatchTableEntry OpDispatch_SecondaryOpSizeModTables[] = {
  {0x75, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::VectorALUOp, IR::OP_VCMPEQ, OpSize::i16Bit>},
  {0x76, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::VectorALUOp, IR::OP_VCMPEQ, OpSize::i32Bit>},
  {0x78, 1, nullptr}, // GROUP 17
+  {0x79, 1, &OpDispatchBuilder::Extrq},
  {0x7C, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::VectorALUOp, IR::OP_VFADDP, OpSize::i64Bit>},
  {0x7D, 1, &OpDispatchBuilder::HSUBP<OpSize::i64Bit>},
  {0x7E, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::MOVBetweenGPR_FPR, OpDispatchBuilder::VectorOpType::SSE>},
@@ -17,7 +17,6 @@ $end_info$
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/FPState.h>

-#include <cmath>
 #include <stddef.h>
 #include <stdint.h>

@@ -28,10 +27,12 @@ class OrderedNode;
 Ref OpDispatchBuilder::GetX87Top() {
  // Yes, we are storing 3 bits in a single flag register.
  // Deal with it
-  return _LoadContext(OpSize::i8Bit, GPRClass, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
+  return _LoadContextGPR(OpSize::i8Bit, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
 }

 void OpDispatchBuilder::SetX87FTW(Ref FTW) {
+  _StackForceSlow(); // Invalidate x87 FTW register cache
+
  // For the output, we want a 1-bit for each pair not equal to 11 (Empty).
  static_assert(static_cast<uint8_t>(FPState::X87Tag::Empty) == 0b11);

@@ -50,24 +51,22 @@ void OpDispatchBuilder::SetX87FTW(Ref FTW) {
  FTW = _Orlshr(OpSize::i32Bit, FTW, FTW, 4);

  // ...and that's it. StoreContext implicitly does the final masking.
-  StoreContext(AbridgedFTWIndex, FTW);
+  _StoreContextGPR(OpSize::i8Bit, FTW, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
 }

 void OpDispatchBuilder::SetX87Top(Ref Value) {
-  _StoreContext(OpSize::i8Bit, GPRClass, Value, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
+  _StoreContextGPR(OpSize::i8Bit, Value, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
 }

 // Float LoaD operation with memory operand
 void OpDispatchBuilder::FLD(OpcodeArgs, IR::OpSize Width) {
-  const auto ReadWidth = (Width == OpSize::f80Bit) ? OpSize::i128Bit : Width;
-
-  Ref Data = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], Width, Op->Flags);
+  Ref Data = LoadSourceFPR_WithOpSize(Op, Op->Src[0], Width, Op->Flags);
  Ref ConvertedData = Data;
  // Convert to 80bit float
  if (Width == OpSize::i32Bit || Width == OpSize::i64Bit) {
-    ConvertedData = _F80CVTTo(Data, ReadWidth);
+    ConvertedData = _F80CVTTo(Data, Width);
  }
-  _PushStack(ConvertedData, Data, ReadWidth, true);
+  _PushStack(ConvertedData, Data, Width);
 }

 // Float LoaD operation with memory operand
@@ -77,27 +76,27 @@ void OpDispatchBuilder::FLDFromStack(OpcodeArgs) {

 void OpDispatchBuilder::FBLD(OpcodeArgs) {
  // Read from memory
-  Ref Data = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], OpSize::f80Bit, Op->Flags);
+  Ref Data = LoadSourceFPR_WithOpSize(Op, Op->Src[0], OpSize::f80Bit, Op->Flags);
  Ref ConvertedData = _F80BCDLoad(Data);
-  _PushStack(ConvertedData, Data, OpSize::i128Bit, true);
+  _PushStack(ConvertedData, Invalid(), OpSize::iInvalid);
 }

 void OpDispatchBuilder::FBSTP(OpcodeArgs) {
  Ref converted = _F80BCDStore(_ReadStackValue(0));
-  StoreResult_WithOpSize(FPRClass, Op, Op->Dest, converted, OpSize::f80Bit, OpSize::i8Bit);
+  StoreResultFPR_WithOpSize(Op, Op->Dest, converted, OpSize::f80Bit, OpSize::i8Bit);
  _PopStackDestroy();
 }

 void OpDispatchBuilder::FLD_Const(OpcodeArgs, NamedVectorConstant K) {
  // Update TOP
  Ref Data = LoadAndCacheNamedVectorConstant(OpSize::i128Bit, K);
-  _PushStack(Data, Data, OpSize::i128Bit, true);
+  _PushStack(Data, Data, OpSize::f80Bit);
 }

 void OpDispatchBuilder::FILD(OpcodeArgs) {
  const auto ReadWidth = OpSizeFromSrc(Op);
  // Read from memory
-  Ref Data = LoadSource_WithOpSize(GPRClass, Op, Op->Src[0], ReadWidth, Op->Flags);
+  Ref Data = LoadSourceGPR_WithOpSize(Op, Op->Src[0], ReadWidth, Op->Flags);

  // Sign extend to 64bits
  if (ReadWidth != OpSize::i64Bit) {
@@ -110,27 +109,28 @@ void OpDispatchBuilder::FILD(OpcodeArgs) {
  // Extract sign and make integer absolute
  auto zero = Constant(0);
  _SubNZCV(OpSize::i64Bit, Data, zero);
-  auto sign = _NZCVSelect(OpSize::i64Bit, CondClassType {COND_SLT}, Constant(0x8000), zero);
-  auto absolute = _Neg(OpSize::i64Bit, Data, CondClassType {COND_MI});
+  auto sign = _NZCVSelect(OpSize::i64Bit, CondClass::SLT, Constant(0x8000), zero);
+  auto absolute = _Neg(OpSize::i64Bit, Data, CondClass::MI);

  // left justify the absolute integer
  auto shift = Sub(OpSize::i64Bit, Constant(63), _FindMSB(IR::OpSize::i64Bit, absolute));
  auto shifted = _Lshl(OpSize::i64Bit, absolute, shift);

  auto adjusted_exponent = Sub(OpSize::i64Bit, Constant(0x3fff + 63), shift);
-  auto zeroed_exponent = _Select(COND_EQ, absolute, zero, zero, adjusted_exponent);
+  auto zeroed_exponent = _Select(OpSize::i64Bit, OpSize::i64Bit, CondClass::EQ, absolute, zero, zero, adjusted_exponent);
  auto upper = _Or(OpSize::i64Bit, sign, zeroed_exponent);

  Ref ConvertedData = _VLoadTwoGPRs(shifted, upper);
-  _PushStack(ConvertedData, Data, ReadWidth, false);
+  _PushStack(ConvertedData, Invalid(), OpSize::iInvalid);
 }

 void OpDispatchBuilder::FST(OpcodeArgs, IR::OpSize Width) {
-  const auto SourceSize = ReducedPrecisionMode ? OpSize::i64Bit : OpSize::i128Bit;
+  LOGMAN_THROW_A_FMT(Width == OpSize::i32Bit || Width == OpSize::i64Bit || Width == OpSize::f80Bit, "Invalid store width for FST");
+  const auto SourceSize = ReducedPrecisionMode ? OpSize::i64Bit : OpSize::f80Bit;
  AddressMode A = DecodeAddress(Op, Op->Dest, MemoryAccessType::DEFAULT, false);

  A = SelectAddressMode(this, A, GetGPROpSize(), CTX->HostFeatures.SupportsTSOImm9, false, false, Width);
-  _StoreStackMem(SourceSize, Width, A.Base, A.Index, OpSize::iInvalid, A.IndexType, A.IndexScale, /*Float=*/true);
+  _StoreStackMem(SourceSize, Width, A.Base, A.Index, OpSize::iInvalid, A.IndexType, A.IndexScale);

  if (Op->TableInfo->Flags & X86Tables::InstFlags::FLAGS_POP) {
    _PopStackDestroy();
@@ -164,12 +164,12 @@ void OpDispatchBuilder::FIST(OpcodeArgs, bool Truncate) {
    // Check for NaN/Infinity: exponent = 0x7fff
    SaveNZCV();
    _TestNZ(OpSize::i64Bit, Exponent, Constant(0x7fff));
-    Ref IsSpecial = _NZCVSelect01({COND_EQ});
+    Ref IsSpecial = _NZCVSelect01(CondClass::EQ);

    // For overflow detection, check if exponent indicates a value >= 2^15
    // Biased exponent for 2^15 is 0x3fff + 15 = 0x400e
    SubWithFlags(OpSize::i64Bit, Exponent, 0x400e);
-    Ref IsOverflow = _NZCVSelect01({COND_UGE});
+    Ref IsOverflow = _NZCVSelect01(CondClass::UGE);

    // Set Invalid Operation flag if overflow or special value
    Ref InvalidFlag = _Or(OpSize::i64Bit, IsSpecial, IsOverflow);
@@ -178,7 +178,7 @@ void OpDispatchBuilder::FIST(OpcodeArgs, bool Truncate) {

  Data = _F80CVTInt(Size, Data, Truncate);

-  StoreResult_WithOpSize(GPRClass, Op, Op->Dest, Data, Size, OpSize::i8Bit);
+  StoreResultGPR_WithOpSize(Op, Op->Dest, Data, Size, OpSize::i8Bit);

  if ((Op->TableInfo->Flags & X86Tables::InstFlags::FLAGS_POP) != 0) {
    _PopStackDestroy();
@@ -204,10 +204,10 @@ void OpDispatchBuilder::FADD(OpcodeArgs, IR::OpSize Width, bool Integer, OpDispa
  // We have one memory argument
  Ref Arg {};
  if (Integer) {
-    Arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    Arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
    Arg = _F80CVTToInt(Arg, Width);
  } else {
-    Arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    Arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    Arg = _F80CVTTo(Arg, Width);
  }

@@ -234,10 +234,10 @@ void OpDispatchBuilder::FMUL(OpcodeArgs, IR::OpSize Width, bool Integer, OpDispa
  // We have one memory argument
  Ref arg {};
  if (Integer) {
-    arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
    arg = _F80CVTToInt(arg, Width);
  } else {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    arg = _F80CVTTo(arg, Width);
  }

@@ -271,10 +271,10 @@ void OpDispatchBuilder::FDIV(OpcodeArgs, IR::OpSize Width, bool Integer, bool Re
  // We have one memory argument
  Ref arg {};
  if (Integer) {
-    arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
    arg = _F80CVTToInt(arg, Width);
  } else {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    arg = _F80CVTTo(arg, Width);
  }

@@ -312,10 +312,10 @@ void OpDispatchBuilder::FSUB(OpcodeArgs, IR::OpSize Width, bool Integer, bool Re
  // We have one memory argument
  Ref Arg {};
  if (Integer) {
-    Arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    Arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
    Arg = _F80CVTToInt(Arg, Width);
  } else {
-    Arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    Arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    Arg = _F80CVTTo(Arg, Width);
  }

@@ -338,7 +338,7 @@ Ref OpDispatchBuilder::GetX87FTW_Helper() {
  // bytes, we use the well-known bit twiddling algorithm:
  //
  // https://graphics.stanford.edu/~seander/bithacks.html#InterleaveBMN
-  Ref X = LoadContext(AbridgedFTWIndex);
+  Ref X = _LoadContextGPR(OpSize::i8Bit, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
  X = _Orlshl(OpSize::i32Bit, X, X, 4);
  X = _And(OpSize::i32Bit, X, Constant(0x0f0f0f0f));
  X = _Orlshl(OpSize::i32Bit, X, X, 2);
@@ -379,41 +379,41 @@ void OpDispatchBuilder::X87FNSTENV(OpcodeArgs) {
  _SyncStackToSlow();

  const auto Size = OpSizeFromSrc(Op);
-  Ref Mem = LoadSource(GPRClass, Op, Op->Dest, Op->Flags, {.LoadData = false});
+  Ref Mem = LoadSourceGPR(Op, Op->Dest, Op->Flags, {.LoadData = false});
  Mem = AppendSegmentOffset(Mem, Op->Flags);

  {
-    auto FCW = _LoadContext(OpSize::i16Bit, GPRClass, offsetof(FEXCore::Core::CPUState, FCW));
-    _StoreMem(GPRClass, Size, Mem, FCW, Size);
+    auto FCW = _LoadContextGPR(OpSize::i16Bit, offsetof(FEXCore::Core::CPUState, FCW));
+    _StoreMemGPR(Size, Mem, FCW, Size);
  }

-  { _StoreMem(GPRClass, Size, ReconstructFSW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MEM_OFFSET_SXTX, 1); }
+  { _StoreMemGPR(Size, ReconstructFSW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MemOffsetType::SXTX, 1); }

  auto ZeroConst = Constant(0);

  {
    // FTW
-    _StoreMem(GPRClass, Size, GetX87FTW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, GetX87FTW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MemOffsetType::SXTX, 1);
  }

  {
    // Instruction Offset
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 3), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 3), Size, MemOffsetType::SXTX, 1);
  }

  {
    // Instruction CS selector (+ Opcode)
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 4), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 4), Size, MemOffsetType::SXTX, 1);
  }

  {
    // Data pointer offset
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 5), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 5), Size, MemOffsetType::SXTX, 1);
  }

  {
    // Data pointer selector
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 6), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 6), Size, MemOffsetType::SXTX, 1);
  }
 }

@@ -439,20 +439,20 @@ void OpDispatchBuilder::X87LDENV(OpcodeArgs) {
  _StackForceSlow();

  const auto Size = OpSizeFromSrc(Op);
-  Ref Mem = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags, {.LoadData = false});
+  Ref Mem = LoadSourceGPR(Op, Op->Src[0], Op->Flags, {.LoadData = false});
  Mem = AppendSegmentOffset(Mem, Op->Flags);

-  auto NewFCW = _LoadMem(GPRClass, OpSize::i16Bit, Mem, OpSize::i16Bit);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  auto NewFCW = _LoadMemGPR(OpSize::i16Bit, Mem, OpSize::i16Bit);
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));

  Ref MemLocation = Add(OpSize::i64Bit, Mem, IR::OpSizeToSize(Size) * 1);
-  auto NewFSW = _LoadMem(GPRClass, Size, MemLocation, Size);
+  auto NewFSW = _LoadMemGPR(Size, MemLocation, Size);
  ReconstructX87StateFromFSW_Helper(NewFSW);

  {
    // FTW
    Ref MemLocation = Add(OpSize::i64Bit, Mem, IR::OpSizeToSize(Size) * 2);
-    SetX87FTW(_LoadMem(GPRClass, Size, MemLocation, Size));
+    SetX87FTW(_LoadMemGPR(Size, MemLocation, Size));
  }
 }

@@ -481,61 +481,61 @@ void OpDispatchBuilder::X87FNSAVE(OpcodeArgs) {
  Ref Mem = MakeSegmentAddress(Op, Op->Dest);
  Ref Top = GetX87Top();
  {
-    auto FCW = _LoadContext(OpSize::i16Bit, GPRClass, offsetof(FEXCore::Core::CPUState, FCW));
-    _StoreMem(GPRClass, Size, Mem, FCW, Size);
+    auto FCW = _LoadContextGPR(OpSize::i16Bit, offsetof(FEXCore::Core::CPUState, FCW));
+    _StoreMemGPR(Size, Mem, FCW, Size);
  }

-  { _StoreMem(GPRClass, Size, ReconstructFSW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MEM_OFFSET_SXTX, 1); }
+  { _StoreMemGPR(Size, ReconstructFSW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MemOffsetType::SXTX, 1); }

  auto ZeroConst = Constant(0);

  {
    // FTW
-    _StoreMem(GPRClass, Size, GetX87FTW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, GetX87FTW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MemOffsetType::SXTX, 1);
  }

  {
    // Instruction Offset
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 3), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 3), Size, MemOffsetType::SXTX, 1);
  }

  {
    // Instruction CS selector (+ Opcode)
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 4), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 4), Size, MemOffsetType::SXTX, 1);
  }

  {
    // Data pointer offset
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 5), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 5), Size, MemOffsetType::SXTX, 1);
  }

  {
    // Data pointer selector
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 6), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 6), Size, MemOffsetType::SXTX, 1);
  }

  auto SevenConst = Constant(7);
  const auto LoadSize = ReducedPrecisionMode ? OpSize::i64Bit : OpSize::i128Bit;
  for (int i = 0; i < 7; ++i) {
-    Ref data = _LoadContextIndexed(Top, LoadSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit), FPRClass);
+    Ref data = _LoadContextFPRIndexed(Top, LoadSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit));
    if (ReducedPrecisionMode) {
      data = _F80CVTTo(data, OpSize::i64Bit);
    }
-    _StoreMem(FPRClass, OpSize::i128Bit, data, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * i)), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
+    _StoreMemFPR(OpSize::i128Bit, data, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * i)), OpSize::i8Bit, MemOffsetType::SXTX, 1);
    Top = _And(OpSize::i32Bit, Add(OpSize::i32Bit, Top, 1), SevenConst);
  }

  // The final st(7) needs a bit of special handling here
-  Ref data = _LoadContextIndexed(Top, LoadSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit), FPRClass);
+  Ref data = _LoadContextFPRIndexed(Top, LoadSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit));
  if (ReducedPrecisionMode) {
    data = _F80CVTTo(data, OpSize::i64Bit);
  }
  // ST7 broken in to two parts
  // Lower 64bits [63:0]
  // upper 16 bits [79:64]
-  _StoreMem(FPRClass, OpSize::i64Bit, data, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (7 * 10)), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
+  _StoreMemFPR(OpSize::i64Bit, data, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (7 * 10)), OpSize::i8Bit, MemOffsetType::SXTX, 1);
  auto topBytes = _VDupElement(OpSize::i128Bit, OpSize::i16Bit, data, 4);
-  _StoreMem(FPRClass, OpSize::i16Bit, topBytes, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (7 * 10) + 8), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
+  _StoreMemFPR(OpSize::i16Bit, topBytes, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (7 * 10) + 8), OpSize::i8Bit, MemOffsetType::SXTX, 1);

  // reset to default
  FNINIT(Op);
@@ -546,8 +546,8 @@ void OpDispatchBuilder::X87FRSTOR(OpcodeArgs) {
  const auto Size = OpSizeFromSrc(Op);
  Ref Mem = MakeSegmentAddress(Op, Op->Src[0]);

-  auto NewFCW = _LoadMem(GPRClass, OpSize::i16Bit, Mem, OpSize::i16Bit);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  auto NewFCW = _LoadMemGPR(OpSize::i16Bit, Mem, OpSize::i16Bit);
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
  if (ReducedPrecisionMode) {
    // ignore the rounding precision, we're always 64-bit in F64.
    // extract rounding mode
@@ -559,11 +559,11 @@ void OpDispatchBuilder::X87FRSTOR(OpcodeArgs) {
    _SetRoundingMode(roundingMode, false, roundingMode);
  }

-  auto NewFSW = _LoadMem(GPRClass, Size, Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MEM_OFFSET_SXTX, 1);
+  auto NewFSW = _LoadMemGPR(Size, Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MemOffsetType::SXTX, 1);
  Ref Top = ReconstructX87StateFromFSW_Helper(NewFSW);
  {
    // FTW
-    SetX87FTW(_LoadMem(GPRClass, Size, Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MEM_OFFSET_SXTX, 1));
+    SetX87FTW(_LoadMemGPR(Size, Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MemOffsetType::SXTX, 1));
  }

  auto SevenConst = Constant(7);
@@ -572,14 +572,14 @@ void OpDispatchBuilder::X87FRSTOR(OpcodeArgs) {
  Ref Mask = _VLoadTwoGPRs(low, high);
  const auto StoreSize = ReducedPrecisionMode ? OpSize::i64Bit : OpSize::i128Bit;
  for (int i = 0; i < 7; ++i) {
-    Ref Reg = _LoadMem(FPRClass, OpSize::i128Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * i)), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
+    Ref Reg = _LoadMemFPR(OpSize::i128Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * i)), OpSize::i8Bit, MemOffsetType::SXTX, 1);
    // Mask off the top bits
    Reg = _VAnd(OpSize::i128Bit, OpSize::i128Bit, Reg, Mask);
    if (ReducedPrecisionMode) {
      // Convert to double precision
      Reg = _F80CVT(OpSize::i64Bit, Reg);
    }
-    _StoreContextIndexed(Reg, Top, StoreSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit), FPRClass);
+    _StoreContextFPRIndexed(Reg, Top, StoreSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit));

    Top = _And(OpSize::i32Bit, Add(OpSize::i32Bit, Top, 1), SevenConst);
  }
@@ -588,20 +588,19 @@ void OpDispatchBuilder::X87FRSTOR(OpcodeArgs) {
  // ST7 broken in to two parts
  // Lower 64bits [63:0]
  // upper 16 bits [79:64]
-  Ref Reg = _LoadMem(FPRClass, OpSize::i64Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * 7)), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
-  Ref RegHigh =
-    _LoadMem(FPRClass, OpSize::i16Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * 7) + 8), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
+  Ref Reg = _LoadMemFPR(OpSize::i64Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * 7)), OpSize::i8Bit, MemOffsetType::SXTX, 1);
+  Ref RegHigh = _LoadMemFPR(OpSize::i16Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * 7) + 8), OpSize::i8Bit, MemOffsetType::SXTX, 1);
  Reg = _VInsElement(OpSize::i128Bit, OpSize::i16Bit, 4, 0, Reg, RegHigh);
  if (ReducedPrecisionMode) {
    Reg = _F80CVT(OpSize::i64Bit, Reg); // Convert to double precision
  }
-  _StoreContextIndexed(Reg, Top, StoreSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit), FPRClass);
+  _StoreContextFPRIndexed(Reg, Top, StoreSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit));
 }

 // Load / Store Control Word
 void OpDispatchBuilder::X87FSTCW(OpcodeArgs) {
-  auto FCW = _LoadContext(OpSize::i16Bit, GPRClass, offsetof(FEXCore::Core::CPUState, FCW));
-  StoreResult(GPRClass, Op, FCW, OpSize::iInvalid);
+  auto FCW = _LoadContextGPR(OpSize::i16Bit, offsetof(FEXCore::Core::CPUState, FCW));
+  StoreResultGPR(Op, FCW);
 }

 void OpDispatchBuilder::X87FLDCW(OpcodeArgs) {
@@ -609,8 +608,8 @@ void OpDispatchBuilder::X87FLDCW(OpcodeArgs) {
  // to switch for now to slow mode whenever these are manually changed.
  // Remove the next line and try DF_04.asm in fast path.
  _StackForceSlow();
-  Ref NewFCW = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  Ref NewFCW = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
 }

 void OpDispatchBuilder::FXCH(OpcodeArgs) {
@@ -646,10 +645,10 @@ void OpDispatchBuilder::FCOMI(OpcodeArgs, IR::OpSize Width, bool Integer, OpDisp
    if (Width == OpSize::i16Bit || Width == OpSize::i32Bit || Width == OpSize::i64Bit) {
      // Memory arg
      if (Integer) {
-        arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+        arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
        b = _F80CVTToInt(arg, Width);
      } else {
-        arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+        arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
        b = _F80CVTTo(arg, Width);
      }
    } else {
@@ -765,7 +764,7 @@ Ref OpDispatchBuilder::ReconstructFSW_Helper(Ref T) {
 void OpDispatchBuilder::X87FNSTSW(OpcodeArgs) {
  Ref TopValue = _SyncStackToSlow();
  Ref StatusWord = ReconstructFSW_Helper(TopValue);
-  StoreResult(GPRClass, Op, StatusWord, OpSize::iInvalid);
+  StoreResultGPR(Op, StatusWord);
 }

 void OpDispatchBuilder::FNCLEX(OpcodeArgs) {
@@ -774,6 +773,8 @@ void OpDispatchBuilder::FNCLEX(OpcodeArgs) {
 }

 void OpDispatchBuilder::FNINIT(OpcodeArgs) {
+  _SyncStackToSlow(); // Invalidate x87 register caches
+
  auto Zero = Constant(0);

  if (ReducedPrecisionMode) {
@@ -782,12 +783,12 @@ void OpDispatchBuilder::FNINIT(OpcodeArgs) {

  // Init FCW to 0x037F
  auto NewFCW = Constant(0x037F);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));

  // Set top to zero
  SetX87Top(Zero);
  // Tags all get marked as invalid
-  StoreContext(AbridgedFTWIndex, Zero);
+  _StoreContextGPR(OpSize::i8Bit, Zero, offsetof(FEXCore::Core::CPUState, AbridgedFTW));

  // Reinits the simulated stack
  _InitStack();
@@ -860,7 +861,7 @@ void OpDispatchBuilder::X87FXAM(OpcodeArgs) {
  auto TopValid = _StackValidTag(0);

  // In the case of top being invalid then C3:C2:C0 is 0b101
-  auto C3 = Select01(OpSize::i32Bit, CondClassType {COND_NEQ}, TopValid, Constant(1));
+  auto C3 = Select01(OpSize::i32Bit, CondClass::NEQ, TopValid, Constant(1));

  auto C2 = TopValid;
  auto C0 = C3; // Mirror C3 until something other than zero is supported
@@ -875,8 +876,8 @@ void OpDispatchBuilder::X87FXTRACT(OpcodeArgs) {
  _PopStackDestroy();
  auto Exp = _F80XTRACT_EXP(Top);
  auto Sig = _F80XTRACT_SIG(Top);
-  _PushStack(Exp, Exp, OpSize::f80Bit, true);
-  _PushStack(Sig, Sig, OpSize::f80Bit, true);
+  _PushStack(Exp, Invalid(), OpSize::iInvalid);
+  _PushStack(Sig, Invalid(), OpSize::iInvalid);
 }

 } // namespace FEXCore::IR
@@ -29,38 +29,37 @@ void OpDispatchBuilder::X87LDENVF64(OpcodeArgs) {
  const auto Size = OpSizeFromSrc(Op);
  Ref Mem = MakeSegmentAddress(Op, Op->Src[0]);

-  auto NewFCW = _LoadMem(GPRClass, OpSize::i16Bit, Mem, OpSize::i16Bit);
+  auto NewFCW = _LoadMemGPR(OpSize::i16Bit, Mem, OpSize::i16Bit);
  // ignore the rounding precision, we're always 64-bit in F64.
  // extract rounding mode
  Ref roundingMode = _Bfe(OpSize::i32Bit, 3, 10, NewFCW);
  _SetRoundingMode(roundingMode, false, roundingMode);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));

-  auto NewFSW = _LoadMem(GPRClass, Size, Mem, Constant(IR::OpSizeToSize(Size)), Size, MEM_OFFSET_SXTX, 1);
+  auto NewFSW = _LoadMemGPR(Size, Mem, Constant(IR::OpSizeToSize(Size)), Size, MemOffsetType::SXTX, 1);
  ReconstructX87StateFromFSW_Helper(NewFSW);

  {
    // FTW
-    SetX87FTW(_LoadMem(GPRClass, Size, Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MEM_OFFSET_SXTX, 1));
+    SetX87FTW(_LoadMemGPR(Size, Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MemOffsetType::SXTX, 1));
  }
 }

 void OpDispatchBuilder::X87FLDCWF64(OpcodeArgs) {
  _StackForceSlow();

-  Ref NewFCW = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+  Ref NewFCW = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
  // ignore the rounding precision, we're always 64-bit in F64.
  // extract rounding mode
  Ref roundingMode = _Bfe(OpSize::i32Bit, 3, 10, NewFCW);
  _SetRoundingMode(roundingMode, false, roundingMode);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
 }

 // F64 ops
 // Float load op with memory operand
 void OpDispatchBuilder::FLDF64(OpcodeArgs, IR::OpSize Width) {
-  const auto ReadWidth = (Width == OpSize::f80Bit) ? OpSize::i128Bit : Width;
-  Ref Data = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], ReadWidth, Op->Flags);
+  Ref Data = LoadSourceFPR_WithOpSize(Op, Op->Src[0], Width, Op->Flags);
  // Convert to 64bit float
  Ref ConvertedData = Data;
  if (Width == OpSize::i32Bit) {
@@ -68,39 +67,39 @@ void OpDispatchBuilder::FLDF64(OpcodeArgs, IR::OpSize Width) {
  } else if (Width == OpSize::f80Bit) {
    ConvertedData = _F80CVT(OpSize::i64Bit, Data);
  }
-  _PushStack(ConvertedData, Data, ReadWidth, true);
+  _PushStack(ConvertedData, Data, Width);
 }

 void OpDispatchBuilder::FBLDF64(OpcodeArgs) {
  // Read from memory
-  Ref Data = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], OpSize::i128Bit, Op->Flags);
+  Ref Data = LoadSourceFPR_WithOpSize(Op, Op->Src[0], OpSize::f80Bit, Op->Flags);
  Ref ConvertedData = _F80BCDLoad(Data);
  ConvertedData = _F80CVT(OpSize::i64Bit, ConvertedData);
-  _PushStack(ConvertedData, Data, OpSize::i64Bit, true);
+  _PushStack(ConvertedData, Invalid(), OpSize::iInvalid);
 }

 void OpDispatchBuilder::FBSTPF64(OpcodeArgs) {
  Ref converted = _F80CVTTo(_ReadStackValue(0), OpSize::i64Bit);
  converted = _F80BCDStore(converted);
-  StoreResult_WithOpSize(FPRClass, Op, Op->Dest, converted, OpSize::f80Bit, OpSize::i8Bit);
+  StoreResultFPR_WithOpSize(Op, Op->Dest, converted, OpSize::f80Bit, OpSize::i8Bit);
  _PopStackDestroy();
 }

 void OpDispatchBuilder::FLDF64_Const(OpcodeArgs, uint64_t Num) {
  auto Data = _VCastFromGPR(OpSize::i64Bit, OpSize::i64Bit, Constant(Num));
-  _PushStack(Data, Data, OpSize::i64Bit, true);
+  _PushStack(Data, Data, OpSize::i64Bit);
 }

 void OpDispatchBuilder::FILDF64(OpcodeArgs) {
  const auto ReadWidth = OpSizeFromSrc(Op);

  // Read from memory
-  Ref Data = LoadSource_WithOpSize(GPRClass, Op, Op->Src[0], ReadWidth, Op->Flags);
+  Ref Data = LoadSourceGPR_WithOpSize(Op, Op->Src[0], ReadWidth, Op->Flags);
  if (ReadWidth == OpSize::i16Bit) {
    Data = _Sbfe(OpSize::i64Bit, IR::OpSizeAsBits(ReadWidth), 0, Data);
  }
  auto ConvertedData = _Float_FromGPR_S(OpSize::i64Bit, ReadWidth == OpSize::i32Bit ? OpSize::i32Bit : OpSize::i64Bit, Data);
-  _PushStack(ConvertedData, Data, ReadWidth, false);
+  _PushStack(ConvertedData, Invalid(), OpSize::iInvalid);
 }

 void OpDispatchBuilder::FISTF64(OpcodeArgs, bool Truncate) {
@@ -112,7 +111,7 @@ void OpDispatchBuilder::FISTF64(OpcodeArgs, bool Truncate) {
  } else {
    data = _Float_ToGPR_S(Size == OpSize::i32Bit ? OpSize::i32Bit : OpSize::i64Bit, OpSize::i64Bit, data);
  }
-  StoreResult_WithOpSize(GPRClass, Op, Op->Dest, data, Size, OpSize::i8Bit);
+  StoreResultGPR_WithOpSize(Op, Op->Dest, data, Size, OpSize::i8Bit);

  if ((Op->TableInfo->Flags & X86Tables::InstFlags::FLAGS_POP) != 0) {
    _PopStackDestroy();
@@ -138,16 +137,16 @@ void OpDispatchBuilder::FADDF64(OpcodeArgs, IR::OpSize Width, bool Integer, OpDi
  Ref arg {};

  if (Integer) {
-    arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
    if (Width == OpSize::i16Bit) {
      arg = _Sbfe(OpSize::i64Bit, 16, 0, arg);
    }
    arg = _Float_FromGPR_S(OpSize::i64Bit, Width == OpSize::i64Bit ? OpSize::i64Bit : OpSize::i32Bit, arg);
  } else if (Width == OpSize::i32Bit) {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    arg = _Float_FToF(OpSize::i64Bit, OpSize::i32Bit, arg);
  } else if (Width == OpSize::i64Bit) {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  } else {
    FEX_UNREACHABLE;
  }
@@ -176,16 +175,16 @@ void OpDispatchBuilder::FMULF64(OpcodeArgs, IR::OpSize Width, bool Integer, OpDi
  Ref arg {};

  if (Integer) {
-    arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
    if (Width == OpSize::i16Bit) {
      arg = _Sbfe(OpSize::i64Bit, 16, 0, arg);
    }
    arg = _Float_FromGPR_S(OpSize::i64Bit, Width == OpSize::i64Bit ? OpSize::i64Bit : OpSize::i32Bit, arg);
  } else if (Width == OpSize::i32Bit) {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    arg = _Float_FToF(OpSize::i64Bit, OpSize::i32Bit, arg);
  } else if (Width == OpSize::i64Bit) {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  } else {
    FEX_UNREACHABLE;
  }
@@ -228,16 +227,16 @@ void OpDispatchBuilder::FDIVF64(OpcodeArgs, IR::OpSize Width, bool Integer, bool

  if (Width == OpSize::i16Bit || Width == OpSize::i32Bit || Width == OpSize::i64Bit) {
    if (Integer) {
-      Arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+      Arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
      if (Width == OpSize::i16Bit) {
        Arg = _Sbfe(OpSize::i64Bit, 16, 0, Arg);
      }
      Arg = _Float_FromGPR_S(OpSize::i64Bit, Width == OpSize::i64Bit ? OpSize::i64Bit : OpSize::i32Bit, Arg);
    } else if (Width == OpSize::i32Bit) {
-      Arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      Arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
      Arg = _Float_FToF(OpSize::i64Bit, OpSize::i32Bit, Arg);
    } else if (Width == OpSize::i64Bit) {
-      Arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      Arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    }
  } else {
    FEX_UNREACHABLE;
@@ -285,16 +284,16 @@ void OpDispatchBuilder::FSUBF64(OpcodeArgs, IR::OpSize Width, bool Integer, bool

  if (Width == OpSize::i16Bit || Width == OpSize::i32Bit || Width == OpSize::i64Bit) {
    if (Integer) {
-      arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+      arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
      if (Width == OpSize::i16Bit) {
        arg = _Sbfe(OpSize::i64Bit, 16, 0, arg);
      }
      arg = _Float_FromGPR_S(OpSize::i64Bit, Width == OpSize::i64Bit ? OpSize::i64Bit : OpSize::i32Bit, arg);
    } else if (Width == OpSize::i32Bit) {
-      arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
      arg = _Float_FToF(OpSize::i64Bit, OpSize::i32Bit, arg);
    } else if (Width == OpSize::i64Bit) {
-      arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    }
  } else {
    FEX_UNREACHABLE;
@@ -332,16 +331,16 @@ void OpDispatchBuilder::FCOMIF64(OpcodeArgs, IR::OpSize Width, bool Integer, OpD
  } else if (Width == OpSize::i16Bit || Width == OpSize::i32Bit || Width == OpSize::i64Bit) {
    // Memory arg
    if (Integer) {
-      arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+      arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
      if (Width == OpSize::i16Bit) {
        arg = _Sbfe(OpSize::i64Bit, 16, 0, arg);
      }
      b = _Float_FromGPR_S(OpSize::i64Bit, Width == OpSize::i64Bit ? OpSize::i64Bit : OpSize::i32Bit, arg);
    } else if (Width == OpSize::i32Bit) {
-      arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
      b = _Float_FToF(OpSize::i64Bit, OpSize::i32Bit, arg);
    } else if (Width == OpSize::i64Bit) {
-      b = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      b = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    }
  } else {
    FEX_UNREACHABLE;
@@ -393,11 +392,11 @@ void OpDispatchBuilder::X87FXTRACTF64(OpcodeArgs) {
  SaveNZCV();
  _TestNZ(OpSize::i64Bit, Gpr, Constant(0x7fff'ffff'ffff'ffffUL));

-  Ref Sig = _NZCVSelectV(OpSize::i64Bit, {COND_EQ}, SigZV, SigNZV);
-  Ref Exp = _NZCVSelectV(OpSize::i64Bit, {COND_EQ}, ExpZV, ExpNZV);
+  Ref Sig = _NZCVSelectV(OpSize::i64Bit, CondClass::EQ, SigZV, SigNZV);
+  Ref Exp = _NZCVSelectV(OpSize::i64Bit, CondClass::EQ, ExpZV, ExpNZV);

  _PopStackDestroy();
-  _PushStack(Exp, Exp, OpSize::i64Bit, true);
-  _PushStack(Sig, Sig, OpSize::i64Bit, true);
+  _PushStack(Exp, Invalid(), OpSize::iInvalid);
+  _PushStack(Sig, Invalid(), OpSize::iInvalid);
 }
 } // namespace FEXCore::IR
@@ -1,87 +0,0 @@
-// SPDX-License-Identifier: MIT
-/*
-$info$
-tags: glue|x86-guest-code
-desc: Guest-side assembly helpers used by the backends
-$end_info$
-*/
-
-#include "Interface/Core/X86HelperGen.h"
-#include "FEXCore/Utils/AllocatorHooks.h"
-
-#include <FEXCore/Config/Config.h>
-#include <FEXCore/Utils/Allocator.h>
-#include <FEXHeaderUtils/Syscalls.h>
-
-#include <cstdint>
-#include <cstring>
-
-namespace FEXCore {
-constexpr size_t CODE_SIZE = 0x1000;
-
-X86GeneratedCode::X86GeneratedCode() {
-#ifdef _WIN32
-  // No need to allocate anything in this config.
-#else
-
-  // Allocate a page for our emulated guest
-  CodePtr = AllocateGuestCodeSpace(CODE_SIZE);
-
-  constexpr std::array<uint8_t, 2> SignalReturnCode = {
-    0x0F, 0x37, // CALLBACKRET FEX Instruction
-  };
-
-  CallbackReturn = reinterpret_cast<uint64_t>(CodePtr);
-
-  memcpy(reinterpret_cast<void*>(CallbackReturn), SignalReturnCode.data(), SignalReturnCode.size());
-
-  mprotect(CodePtr, CODE_SIZE, PROT_READ);
-#endif
-}
-
-X86GeneratedCode::~X86GeneratedCode() {
-#ifndef _WIN32
-  FEXCore::Allocator::VirtualFree(CodePtr, CODE_SIZE);
-#endif
-}
-
-void* X86GeneratedCode::AllocateGuestCodeSpace(size_t Size) {
-#ifndef _WIN32
-  FEX_CONFIG_OPT(Is64BitMode, IS64BIT_MODE);
-
-  if (Is64BitMode()) {
-    // 64bit mode can have its sigret handler anywhere
-    return FEXCore::Allocator::VirtualAlloc(Size);
-  }
-
-  // First 64bit page
-  constexpr uintptr_t LOCATION_MAX = 0x1'0000'0000;
-
-  // 32bit mode
-  // We need to have the sigret handler in the lower 32bits of memory space
-  // Scan top down and try to allocate a location
-  for (size_t Location = 0xFFFF'E000; Location != 0x0; Location -= 0x1000) {
-    void* Ptr = ::mmap(reinterpret_cast<void*>(Location), Size, PROT_READ | PROT_WRITE, MAP_FIXED_NOREPLACE | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
-
-    if (Ptr != MAP_FAILED && reinterpret_cast<uintptr_t>(Ptr) >= LOCATION_MAX) {
-      // Failed to map in the lower 32bits
-      // Try again
-      // Can happen in the case that host kernel ignores MAP_FIXED_NOREPLACE
-      ::munmap(Ptr, Size);
-      continue;
-    }
-
-    if (Ptr != MAP_FAILED) {
-      return Ptr;
-    }
-  }
-
-  // Can't do anything about this
-  // Here's hoping the application doesn't use signals
-  return MAP_FAILED;
-#else
-  return nullptr;
-#endif
-}
-
-} // namespace FEXCore
@@ -1,25 +0,0 @@
-// SPDX-License-Identifier: MIT
-/*
-$info$
-tags: glue|x86-guest-code
-$end_info$
-*/
-
-#pragma once
-
-#include <stddef.h>
-#include <stdint.h>
-
-namespace FEXCore {
-class X86GeneratedCode final {
-public:
-  X86GeneratedCode();
-  ~X86GeneratedCode();
-
-  uint64_t CallbackReturn {};
-
-private:
-  void* CodePtr {};
-  void* AllocateGuestCodeSpace(size_t Size);
-};
-} // namespace FEXCore