Docs: Update for release FEX-2507.1

JIT: fix divisor masking
oversight. should fix Steam. Fixes: de4becc26 ("OpcodeDispatcher: mask certain divisors") Closes: #4652 Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
2026-10-17 00:00:39 +02:00 · 2025-07-11 11:37:44 -07:00 · 2025-07-11 11:34:45 -07:00 · 2025-07-11 11:34:07 -07:00 · 2025-07-07 23:49:56 -07:00 · 2025-07-07 21:16:54 -07:00
1868 changed files with 167412 additions and 299813 deletions
@@ -3,10 +3,11 @@
 # Ignore all files in the External directory
 External/*

-# SoftFloat-3e code doesn't belong to us 
-FEXCore/Source/Common/SoftFloat-3e/*
 Source/Common/cpp-optparse/*

 # Files with human-indented tables for readability - don't mess with these
 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
@@ -13,3 +13,6 @@

 # Second reformat to find fixed point PR#3577
 905aa935f5ce344a48ef4d5edab3c31efa8d793e
+
+# Reformat of CodeEmitter inl files
+8760c593ece92d7e9fa94c40da0368fd367c9cad
@@ -250,7 +250,7 @@ jobs:

    - name: Upload results
      if: ${{ always() }}
-      uses: 'actions/upload-artifact@v3'
+      uses: 'actions/upload-artifact@v4'
      timeout-minutes: 1
      with:
        name: Results-${{ env.runner_name }}
@@ -184,7 +184,7 @@ jobs:

    - name: Upload results
      if: ${{ always() }}
-      uses: 'actions/upload-artifact@v3'
+      uses: 'actions/upload-artifact@v4'
      timeout-minutes: 1
      with:
        name: Results-${{ env.runner_name }}
@@ -97,7 +97,7 @@ jobs:

    - name: Upload results
      if: ${{ always() }}
-      uses: 'actions/upload-artifact@v3'
+      uses: 'actions/upload-artifact@v4'
      timeout-minutes: 1
      with:
        name: Results-${{ env.runner_name }}
@@ -128,7 +128,7 @@ jobs:

    - name: Upload results
      if: ${{ always() }}
-      uses: 'actions/upload-artifact@v3'
+      uses: 'actions/upload-artifact@v4'
      timeout-minutes: 1
      with:
        name: Results-${{ env.runner_name }}
@@ -137,7 +137,7 @@ jobs:

    - name: Upload results InstCountCI
      if: ${{ always() }}
-      uses: 'actions/upload-artifact@v3'
+      uses: 'actions/upload-artifact@v4'
      timeout-minutes: 1
      with:
        name: Results-${{ env.runner_name }}-instcountci
@@ -16,7 +16,7 @@ jobs:
    runs-on: ${{ matrix.arch }}
    strategy:
      matrix:
-        arch: [[self-hosted, ARM64, mingw]]
+        arch: [[self-hosted, ARM64, mingw], [self-hosted, ARM64EC, mingw, ARM64]]
      fail-fast: false

    steps:
@@ -38,6 +38,11 @@ jobs:
      run: |
        echo "MINGW_TRIPLE=aarch64-w64-mingw32" >> $GITHUB_ENV

+    - name: Set CC Arm64EC
+      if: matrix.arch[1] == 'ARM64EC'
+      run: |
+        echo "MINGW_TRIPLE=arm64ec-w64-mingw32" >> $GITHUB_ENV
+
    - name: Set rootfs paths
      run: |
        echo "FEX_ROOTFS_MOUNT=/mnt/AutoNFS/rootfs/" >> $GITHUB_ENV
@@ -73,7 +78,7 @@ 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=$GITHUB_WORKSPACE/toolchain_mingw.cmake -DMINGW_TRIPLE=$MINGW_TRIPLE -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True -DBUILD_TESTS=False -DENABLE_JEMALLOC=False -DENABLE_JEMALLOC_GLIBC_ALLOC=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_TESTS=False -DCMAKE_INSTALL_PREFIX=${{runner.workspace}}/build/install

    - name: Build
      working-directory: ${{runner.workspace}}/build
@@ -87,7 +92,7 @@ jobs:

    - name: Upload results
      if: ${{ always() }}
-      uses: 'actions/upload-artifact@v3'
+      uses: 'actions/upload-artifact@v4'
      timeout-minutes: 1
      with:
        name: Results-${{ env.runner_name }}
@@ -28,7 +28,7 @@ jobs:

      - name: Get changed files
        id: changed-files
-        uses: tj-actions/changed-files@v39
+        uses: step-security/changed-files@3dbe17c78367e7d60f00d78ae6781a35be47b4a1 # v45.0.1
        with:
          separator: ","
          skip_initial_fetch: true
@@ -72,23 +72,22 @@ jobs:
      # Execute the build.  You can specify a specific target with "--target <NAME>"
      run: cmake --build . --config $BUILD_TYPE

-    - name: ASM Tests
+    - name: ASM Tests - SVE256
      working-directory: ${{runner.workspace}}/build
      shell: bash
      # Execute the unit tests
      run: cmake --build . --config $BUILD_TYPE --target asm_tests

-    - name: ASM Test Results move
+    - name: ASM Test SVE256 Results move
      if: ${{ always() }}
      shell: bash
      working-directory: ${{runner.workspace}}/build
-      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_ASM.log || true
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_ASM_SVE256Bit.log || true

-    - name: ASM Tests 128-bit
+    - name: ASM Tests - SVE128
      working-directory: ${{runner.workspace}}/build
      shell: bash
      env:
-        FEX_HOSTFEATURES: "disableavx"
        FEX_FORCESVEWIDTH: "128"
      # Execute the unit tests
      run: cmake --build . --config $BUILD_TYPE --target asm_tests
@@ -97,7 +96,21 @@ jobs:
      if: ${{ always() }}
      shell: bash
      working-directory: ${{runner.workspace}}/build
-      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_ASM128bit.log || true
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_ASM_SVE128Bit.log || true
+
+    - name: ASM Tests - ASIMD
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      env:
+        FEX_HOSTFEATURES: "disablesve"
+      # Execute the unit tests
+      run: cmake --build . --config $BUILD_TYPE --target asm_tests
+
+    - name: ASM Test ASIMD Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_ASM_ASIMD.log || true

    - name: Truncate test results
      if: ${{ always() }}
@@ -113,7 +126,7 @@ jobs:

    - name: Upload results
      if: ${{ always() }}
-      uses: 'actions/upload-artifact@v3'
+      uses: 'actions/upload-artifact@v4'
      timeout-minutes: 1
      with:
        name: Results-${{ env.runner_name }}
@@ -4,7 +4,7 @@ compile_commands.json
 vim_rc
 Config.json

-[Bb]uild*/
+[Bb]uild*
 [Bb]in/
 out/
 .vscode/
@@ -5,19 +5,6 @@
 [submodule "External/cpp-optparse"]
 	path = Source/Common/cpp-optparse
 	url = https://github.com/Sonicadvance1/cpp-optparse
-[submodule "External/imgui"]
-	path = External/imgui
-	url = https://github.com/Sonicadvance1/imgui.git
-[submodule "External/json-maker"]
-	path = External/json-maker
-	url = https://github.com/Sonicadvance1/json-maker.git
-[submodule "External/tiny-json"]
-	path = External/tiny-json
-	url = https://github.com/Sonicadvance1/tiny-json.git
-[submodule "External/xbyak"]
-  shallow = true
-	path = External/xbyak
-	url = https://github.com/herumi/xbyak.git
 [submodule "External/fex-posixtest-bins"]
  shallow = true
 	path = External/fex-posixtest-bins
@@ -56,3 +43,6 @@
 [submodule "External/jemalloc_glibc"]
 	path = External/jemalloc_glibc
 	url = https://github.com/FEX-Emu/jemalloc.git
+[submodule "External/tracy"]
+	path = External/tracy
+	url = https://github.com/wolfpld/tracy
@@ -1,5 +1,5 @@
 cmake_minimum_required(VERSION 3.14)
-project(FEX)
+project(FEX C CXX ASM)

 INCLUDE (CheckIncludeFiles)
 CHECK_INCLUDE_FILES ("gdb/jit-reader.h" HAVE_GDB_JIT_READER_H)
@@ -7,14 +7,16 @@ CHECK_INCLUDE_FILES ("gdb/jit-reader.h" HAVE_GDB_JIT_READER_H)
 option(BUILD_TESTS "Build unit tests to ensure sanity" TRUE)
 option(BUILD_FEX_LINUX_TESTS "Build FEXLinuxTests, requires x86 compiler" FALSE)
 option(BUILD_THUNKS "Build thunks" FALSE)
-option(BUILD_FEXCONFIG "Build FEXConfig, requires SDL2 and X11" TRUE)
-option(ENABLE_CLANG_THUNKS "Build thunks with clang" 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_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)
@@ -25,27 +27,33 @@ 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 "Forces the FEX JIT to use the VIXL simulator" FALSE)
+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(COMPILE_VIXL_DISASSEMBLER "Compiles the vixl disassembler in to 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 you want to use for the FEXCore profiler")
+set (FEXCORE_PROFILER_BACKEND "gpuvis" CACHE STRING "Set which backend to use for the FEXCore profiler (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)

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

 if (NOT MINGW_BUILD)
  message (STATUS "Clang version ${CMAKE_CXX_COMPILER_VERSION}")
-  set (CLANG_MINIMUM_VERSION 12.0)
+  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}")
  endif()
@@ -57,6 +65,22 @@ if (ENABLE_FEXCORE_PROFILER)

  if (FEXCORE_PROFILER_BACKEND STREQUAL "GPUVIS")
    add_definitions(-DFEXCORE_PROFILER_BACKEND=1)
+  elseif (FEXCORE_PROFILER_BACKEND STREQUAL "TRACY")
+    add_definitions(-DFEXCORE_PROFILER_BACKEND=2)
+    add_definitions(-DTRACY_ENABLE=1)
+    # Required so that Tracy will only start in the selected guest application
+    add_definitions(-DTRACY_MANUAL_LIFETIME=1)
+    add_definitions(-DTRACY_DELAYED_INIT=1)
+    # This interferes with FEX's signal handling
+    add_definitions(-DTRACY_NO_CRASH_HANDLER=1)
+    # Tracy can gather call stack samples in regular intervals, but this
+    # isn't useful for us since it would usually sample opaque JIT code
+    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")
+    endif()
  else()
    message(FATAL_ERROR "Unknown FEXCore profiler backend ${FEXCORE_PROFILER_BACKEND}")
  endif()
@@ -73,7 +97,7 @@ endif()
 # uninstall target
 if(NOT TARGET uninstall)
  configure_file(
-    "${CMAKE_CURRENT_SOURCE_DIR}/CMakeFiles/cmake_uninstall.cmake.in"
+    "${CMAKE_CURRENT_SOURCE_DIR}/Data/CMake/cmake_uninstall.cmake.in"
    "${CMAKE_CURRENT_BINARY_DIR}/CMakeFiles/cmake_uninstall.cmake"
    IMMEDIATE @ONLY)

@@ -137,9 +161,39 @@ endif()
 if (CMAKE_SYSTEM_PROCESSOR MATCHES "^arm64ec")
  set(_M_ARM_64EC 1)
  add_definitions(-D_M_ARM_64EC=1)
+endif()

-  # Required as FEX is not allowed to lock the CRT heap lock during compilation or callbacks
-  set(ENABLE_JEMALLOC TRUE)
+include(CheckCXXSourceCompiles)
+set(CMAKE_REQUIRED_FLAGS "-std=c++11 -Wattributes -Werror=attributes")
+check_cxx_source_compiles(
+  "
+  __attribute__((preserve_all))
+  int Testy(int a, int b, int c, int d, int e, int f) {
+  return a + b + c + d + e + f;
+  }
+  int main() {
+  return Testy(0, 1, 2, 3, 4, 5);
+  }"
+  HAS_CLANG_PRESERVE_ALL)
+unset(CMAKE_REQUIRED_FLAGS)
+if (HAS_CLANG_PRESERVE_ALL)
+  if (MINGW_BUILD)
+    message(STATUS "Ignoring broken clang::preserve_all support")
+    set(HAS_CLANG_PRESERVE_ALL FALSE)
+  else()
+    message(STATUS "Has clang::preserve_all")
+  endif()
+endif ()
+
+if (_M_ARM_64 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")
+endif()
+
+if (ENABLE_VIXL_SIMULATOR)
+  # We can run the simulator on both x86-64 or AArch64 hosts
+  add_definitions(-DVIXL_SIMULATOR=1 -DVIXL_INCLUDE_SIMULATOR_AARCH64=1)
 endif()

 if (ENABLE_CCACHE)
@@ -178,6 +232,19 @@ if (NOT ENABLE_OFFLINE_TELEMETRY)
  add_definitions(-DFEX_DISABLE_TELEMETRY=1)
 endif()

+if (ENABLE_UBSAN)
+  # See https://github.com/FEX-Emu/FEX/pull/4494#issuecomment-2800608944
+  # and related discussion for the use of -fno-sanitize=alignment -fno-sanitize=function
+  # with UBSAN.
+  # alignment: we don't follow a strict alignment policy, for example IR uses packed structs
+  # that are regularly access unaligned.
+  # function: syscalls cast function pointers to void (*)(unsigned long...), causing warnings
+  # related to this access.
+  add_definitions(-DENABLE_UBSAN=1)
+  add_compile_options(-fno-omit-frame-pointer -fsanitize=undefined -fno-sanitize=alignment -fno-sanitize=function -fno-sanitize-recover=undefined)
+  link_libraries(-fno-omit-frame-pointer -fsanitize=undefined -fno-sanitize=alignment -fno-sanitize=function -fno-sanitize-recover=undefined)
+endif()
+
 if (ENABLE_ASAN)
  add_definitions(-DENABLE_ASAN=1)
  add_compile_options(-fno-omit-frame-pointer -fsanitize=address -fsanitize-address-use-after-scope)
@@ -189,13 +256,17 @@ if (ENABLE_TSAN)
  link_libraries(-fno-omit-frame-pointer -fsanitize=thread)
 endif()

+if (ENABLE_COVERAGE)
+  add_compile_options(-fprofile-instr-generate -fcoverage-mapping)
+  link_libraries(-fprofile-instr-generate -fcoverage-mapping)
+endif()
+
 if (ENABLE_JEMALLOC_GLIBC_ALLOC)
  # The glibc jemalloc subproject which hooks the glibc allocator.
  # 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_definitions(-DENABLE_JEMALLOC_GLIBC=1)
  add_subdirectory(External/jemalloc_glibc/)
-else()
+elseif (NOT MINGW_BUILD)
  message (STATUS
    " jemalloc glibc allocator disabled!\n"
    " This is not a recommended configuration!\n"
@@ -205,10 +276,8 @@ endif()

 if (ENABLE_JEMALLOC)
  # The jemalloc subproject that all FEXCore fextl objects allocate through.
-  add_definitions(-DENABLE_JEMALLOC=1)
  add_subdirectory(External/jemalloc/)
-  include_directories(External/jemalloc/pregen/include/)
-else()
+elseif (NOT MINGW_BUILD)
  message (STATUS
    " jemalloc disabled!\n"
    " This is not a recommended configuration!\n"
@@ -216,6 +285,11 @@ else()
    " Use at your own risk!")
 endif()

+if (USE_PDB_DEBUGINFO)
+  add_compile_options(-g -gcodeview)
+  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")

@@ -224,13 +298,14 @@ set (CMAKE_LINKER_FLAGS_RELEASE "${CMAKE_LINKER_FLAGS_RELEASE} -fomit-frame-poin

 include_directories(External/robin-map/include/)

-if (BUILD_TESTS)
-  # Enable vixl disassembler if tests are enabled.
-  set(COMPILE_VIXL_DISASSEMBLER TRUE)
+if (BUILD_TESTS OR ENABLE_VIXL_DISASSEMBLER OR ENABLE_VIXL_SIMULATOR)
+  add_subdirectory(External/vixl/)
+  include_directories(SYSTEM External/vixl/src/)
 endif()

-add_subdirectory(External/vixl/)
-include_directories(SYSTEM External/vixl/src/)
+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
@@ -240,37 +315,42 @@ endif()
 find_package(PkgConfig REQUIRED)
 find_package(Python 3.0 REQUIRED COMPONENTS Interpreter)

-set(XXHASH_BUNDLED_MODE TRUE)
-set(XXHASH_BUILD_XXHSUM FALSE)
 set(BUILD_SHARED_LIBS OFF)
-add_subdirectory(External/xxhash/cmake_unofficial/)
+
+pkg_search_module(xxhash IMPORTED_TARGET xxhash libxxhash)
+if (TARGET PkgConfig::xxhash AND NOT CMAKE_CROSSCOMPILING)
+  add_library(xxHash::xxhash ALIAS PkgConfig::xxhash)
+else()
+  set(XXHASH_BUNDLED_MODE TRUE)
+  set(XXHASH_BUILD_XXHSUM FALSE)
+  add_subdirectory(External/xxhash/cmake_unofficial/)
+endif()

 add_definitions(-Wno-trigraphs)
 add_definitions(-DGLOBAL_DATA_DIRECTORY="${DATA_DIRECTORY}/")

 if (BUILD_TESTS)
-  add_subdirectory(External/Catch2/)
+  find_package(Catch2 3 QUIET)
+  if (NOT Catch2_FOUND)
+    add_subdirectory(External/Catch2/)
+
+    # Pull in catch_discover_tests definition
+    list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/External/Catch2/contrib/")
+  endif()

-  # Pull in catch_discover_tests definition
-  list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/External/Catch2/contrib/")
  include(Catch)
 endif()

-# Disable fmt install
-set(FMT_INSTALL OFF)
-add_subdirectory(External/fmt/)
-
-add_subdirectory(External/imgui/)
-include_directories(External/imgui/)
-
-add_subdirectory(External/json-maker/)
-include_directories(External/json-maker/)
+find_package(fmt QUIET)
+if (NOT fmt_FOUND)
+  # Disable fmt install
+  set(FMT_INSTALL OFF)
+  add_subdirectory(External/fmt/)
+endif()

 add_subdirectory(External/tiny-json/)
 include_directories(External/tiny-json/)

-include_directories(External/xbyak/)
-
 include_directories(Source/)
 include_directories("${CMAKE_BINARY_DIR}/Source/")

@@ -321,10 +401,6 @@ if (TUNE_CPU STREQUAL "native")
        list(APPEND FEX_TUNE_COMPILE_FLAGS "-mcpu=native")
      endif()
    else()
-      # Due to an oversight in llvm, it declares any reasonably new Kryo CPU to only be ARMv8.0
-      # Manually detect newer CPU revisions until clang and llvm fixes their bug
-      # This script will either provide a supported CPU or 'native'
-      # Additionally -march doesn't work under AArch64+Clang, so you have to use -mcpu or -mtune
      execute_process(COMMAND python3 "${PROJECT_SOURCE_DIR}/Scripts/aarch64_fit_native.py" "/proc/cpuinfo" "${CMAKE_CXX_COMPILER_VERSION}"
        OUTPUT_VARIABLE AARCH64_CPU)

@@ -341,7 +417,7 @@ if (TUNE_CPU STREQUAL "native")
      list(APPEND FEX_TUNE_COMPILE_FLAGS "-march=native")
    endif()
  endif()
-else()
+elseif (NOT TUNE_CPU STREQUAL "none")
  check_cxx_compiler_flag("-mcpu=${TUNE_CPU}" COMPILER_SUPPORTS_CPU_TYPE)
  if(COMPILER_SUPPORTS_CPU_TYPE)
    list(APPEND FEX_TUNE_COMPILE_FLAGS "-mcpu=${TUNE_CPU}")
@@ -360,31 +436,33 @@ endif()

 add_compile_options(-Wall)

-configure_file(
-  ${CMAKE_CURRENT_SOURCE_DIR}/include/Config.h.in
-  ${CMAKE_BINARY_DIR}/generated/ConfigDefines.h)
-
+include(CTest)
 if (BUILD_TESTS)
-  include(CTest)
-  enable_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")
  if (TEST_JOB_COUNT)
    message(STATUS "Running tests with ${TEST_JOB_COUNT} jobs")
-  endif()
-  if (CMAKE_VERSION VERSION_LESS "3.29")
+  elseif(CMAKE_VERSION VERSION_LESS "3.29")
    execute_process(COMMAND "nproc" OUTPUT_STRIP_TRAILING_WHITESPACE OUTPUT_VARIABLE TEST_JOB_COUNT)
  endif()
  set(TEST_JOB_FLAG "-j${TEST_JOB_COUNT}")
 endif()

+add_subdirectory(External/SoftFloat-3e/)
+add_subdirectory(External/cephes/)
 add_subdirectory(FEXHeaderUtils/)
 add_subdirectory(CodeEmitter/)
 add_subdirectory(FEXCore/)

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

 add_subdirectory(Source/)
 add_subdirectory(Data/AppConfig/)
@@ -427,6 +505,7 @@ if (BUILD_THUNKS)
      "-DCMAKE_INSTALL_PREFIX=${CMAKE_INSTALL_PREFIX}"
      "-DFEX_PROJECT_SOURCE_DIR=${FEX_PROJECT_SOURCE_DIR}"
      "-DGENERATOR_EXE=$<TARGET_FILE:thunkgen>"
+      "-DX86_DEV_ROOTFS=${X86_DEV_ROOTFS}"
    INSTALL_COMMAND ""
    BUILD_ALWAYS ON
    DEPENDS thunkgen
@@ -445,6 +524,7 @@ if (BUILD_THUNKS)
      "-DCMAKE_INSTALL_PREFIX=${CMAKE_INSTALL_PREFIX}"
      "-DFEX_PROJECT_SOURCE_DIR=${FEX_PROJECT_SOURCE_DIR}"
      "-DGENERATOR_EXE=$<TARGET_FILE:thunkgen>"
+      "-DX86_DEV_ROOTFS=${X86_DEV_ROOTFS}"
    INSTALL_COMMAND ""
    BUILD_ALWAYS ON
    DEPENDS thunkgen
@@ -538,12 +618,12 @@ set (CPACK_PACKAGE_CONTACT "FEX-Emu Maintainers <team@fex-emu.com>")
 set (CPACK_PACKAGE_VERSION_MAJOR "${FEX_VERSION_MAJOR}")
 set (CPACK_PACKAGE_VERSION_MINOR "${FEX_VERSION_MINOR}")
 set (CPACK_PACKAGE_VERSION_PATCH "${FEX_VERSION_PATCH}")
-set (CPACK_PACKAGE_DESCRIPTION_FILE "${CMAKE_CURRENT_SOURCE_DIR}/CPack/Description.txt")
+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
-  "${CMAKE_CURRENT_SOURCE_DIR}/CPack/postinst;${CMAKE_CURRENT_SOURCE_DIR}/CPack/prerm;${CMAKE_CURRENT_SOURCE_DIR}/CPack/triggers")
+  "${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
@@ -1,3 +0,0 @@
-x86 and x86-64 Linux emulator
-
-FEX is very much work in progress, so expect things to change.
@@ -11,6 +11,14 @@
 * FEX-Emu ALU operations usually have a 32-bit or 64-bit operating size encoded in the IR operation,
 * This allows FEX to use a single helper function which decodes to both handlers.
 */
+
+#pragma once
+#ifndef INCLUDED_BY_EMITTER
+#include <CodeEmitter/Emitter.h>
+namespace ARMEmitter {
+struct EmitterOps : Emitter {
+#endif
+
 private:
  static bool IsADRRange(int64_t Imm) {
    return Imm >= -1048576 && Imm <= 1048575;
@@ -28,26 +36,23 @@ public:
    DataProcessing_PCRel_Imm(Op, rd, Imm);
  }

-  void adr(ARMEmitter::Register rd, BackwardLabel const* Label) {
+  void 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");

    constexpr uint32_t Op = 0b0001'0000 << 24;
    DataProcessing_PCRel_Imm(Op, rd, Imm);
  }
-  template<typename LabelType>
-  requires (std::is_same_v<LabelType, ForwardLabel> || std::is_same_v<LabelType, SingleUseForwardLabel>)
-  void adr(ARMEmitter::Register rd, LabelType *Label) {
-    AddLocationToLabel(Label, SingleUseForwardLabel{ .Location = GetCursorAddress<uint8_t*>(), .Type = SingleUseForwardLabel::InstType::ADR });
+  void 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);
  }

-  void adr(ARMEmitter::Register rd, BiDirectionalLabel *Label) {
+  void adr(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
    if (Label->Backward.Location) {
      adr(rd, &Label->Backward);
-    }
-    else {
+    } else {
      adr(rd, &Label->Forward);
    }
  }
@@ -57,39 +62,34 @@ public:
    DataProcessing_PCRel_Imm(Op, rd, Imm);
  }

-  void adrp(ARMEmitter::Register rd, BackwardLabel const* Label) {
+  void 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);
  }
-  template<typename LabelType>
-  requires (std::is_same_v<LabelType, ForwardLabel> || std::is_same_v<LabelType, SingleUseForwardLabel>)
-  void adrp(ARMEmitter::Register rd, LabelType *Label) {
-    AddLocationToLabel(Label, SingleUseForwardLabel{ .Location = GetCursorAddress<uint8_t*>(), .Type = SingleUseForwardLabel::InstType::ADRP });
+  void 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);
  }

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

-  void LongAddressGen(ARMEmitter::Register rd, BackwardLabel const* Label) {
+  void LongAddressGen(ARMEmitter::Register rd, const BackwardLabel* Label) {
    int64_t Imm = reinterpret_cast<int64_t>(Label->Location) - (GetCursorAddress<int64_t>());
    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);
+    } else if (IsADRPRange(Imm)) {
+      int64_t ADRPImm = (reinterpret_cast<int64_t>(Label->Location) & ~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));
@@ -102,24 +102,22 @@ public:
        // Now even an add
        add(ARMEmitter::Size::i64Bit, rd, rd, AlignedOffset);
      }
-    }
-    else {
+    } else {
      LOGMAN_MSG_A_FMT("Unscaled offset too large");
      FEX_UNREACHABLE;
    }
  }
  void LongAddressGen(ARMEmitter::Register rd, ForwardLabel* Label) {
-    Label->Insts.emplace_back(SingleUseForwardLabel{ .Location = GetCursorAddress<uint8_t*>(), .Type = SingleUseForwardLabel::InstType::LONG_ADDRESS_GEN });
+    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.
    dc32(rd.Idx());
    nop();
  }

-  void LongAddressGen(ARMEmitter::Register rd, BiDirectionalLabel *Label) {
+  void LongAddressGen(ARMEmitter::Register rd, BiDirectionalLabel* Label) {
    if (Label->Backward.Location) {
      LongAddressGen(rd, &Label->Backward);
-    }
-    else {
+    } else {
      LongAddressGen(rd, &Label->Forward);
    }
  }
@@ -176,11 +174,7 @@ public:
  // Logical immediate
  void and_(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, uint64_t Imm) {
    uint32_t n, immr, imms;
-    [[maybe_unused]] const auto IsImm = IsImmLogical(Imm,
-                           RegSizeInBits(s),
-                           &n,
-                           &imms,
-                           &immr);
+    [[maybe_unused]] const auto IsImm = IsImmLogical(Imm, RegSizeInBits(s), &n, &imms, &immr);
    LOGMAN_THROW_A_FMT(IsImm, "Couldn't encode immediate to logical op");
    and_(s, rd, rn, n, immr, imms);
  }
@@ -191,11 +185,7 @@ public:

  void ands(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, uint64_t Imm) {
    uint32_t n, immr, imms;
-    [[maybe_unused]] const auto IsImm = IsImmLogical(Imm,
-                           RegSizeInBits(s),
-                           &n,
-                           &imms,
-                           &immr);
+    [[maybe_unused]] const auto IsImm = IsImmLogical(Imm, RegSizeInBits(s), &n, &imms, &immr);
    LOGMAN_THROW_A_FMT(IsImm, "Couldn't encode immediate to logical op");
    ands(s, rd, rn, n, immr, imms);
  }
@@ -206,22 +196,14 @@ public:

  void orr(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, uint64_t Imm) {
    uint32_t n, immr, imms;
-    [[maybe_unused]] const auto IsImm = IsImmLogical(Imm,
-                           RegSizeInBits(s),
-                           &n,
-                           &imms,
-                           &immr);
+    [[maybe_unused]] const auto IsImm = IsImmLogical(Imm, RegSizeInBits(s), &n, &imms, &immr);
    LOGMAN_THROW_A_FMT(IsImm, "Couldn't encode immediate to logical op");
    orr(s, rd, rn, n, immr, imms);
  }

  void eor(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, uint64_t Imm) {
    uint32_t n, immr, imms;
-    [[maybe_unused]] const auto IsImm = IsImmLogical(Imm,
-                           RegSizeInBits(s),
-                           &n,
-                           &imms,
-                           &immr);
+    [[maybe_unused]] const auto IsImm = IsImmLogical(Imm, RegSizeInBits(s), &n, &imms, &immr);
    LOGMAN_THROW_A_FMT(IsImm, "Couldn't encode immediate to logical op");
    eor(s, rd, rn, n, immr, imms);
  }
@@ -301,8 +283,9 @@ public:
    xbfiz_helper(true, s, rd, rn, lsb, width);
  }
  void asr(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, uint32_t shift) {
-    LOGMAN_THROW_A_FMT(shift <= RegSizeInBits(s), "Tried to asr a region larger than the register");
-    sbfm(s, rd, rn, shift, RegSizeInBits(s) - 1);
+    const auto RegSize_m1 = RegSizeInBits(s) - 1;
+    shift &= RegSize_m1;
+    sbfm(s, rd, rn, shift, RegSize_m1);
  }

  void uxtb(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn) {
@@ -325,14 +308,14 @@ public:
  }

  void lsl(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, uint32_t shift) {
-    const auto RegSize = RegSizeInBits(s);
-    LOGMAN_THROW_A_FMT(shift < RegSize, "Tried to lsl a region larger than the register");
-    ubfm(s, rd, rn, (RegSize - shift) % RegSize, RegSize - shift - 1);
+    const auto RegSize_m1 = RegSizeInBits(s) - 1;
+    shift &= RegSize_m1;
+    ubfm(s, rd, rn, (RegSizeInBits(s) - shift) & RegSize_m1, RegSize_m1 - shift);
  }
  void lsr(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, uint32_t shift) {
-    const auto RegSize = RegSizeInBits(s);
-    LOGMAN_THROW_A_FMT(shift < RegSize, "Tried to lsr a region larger than the register");
-    ubfm(s, rd, rn, shift, RegSize - 1);
+    const auto RegSize_m1 = RegSizeInBits(s) - 1;
+    shift &= RegSize_m1;
+    ubfm(s, rd, rn, shift, RegSize_m1);
  }
  void ubfx(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, uint32_t lsb, uint32_t width) {
    LOGMAN_THROW_A_FMT(width > 0, "ubfx needs width > 0");
@@ -354,8 +337,8 @@ public:
    const auto lsb_p_width = lsb + width;

    LOGMAN_THROW_A_FMT(width >= 1, "bfxil needs width >= 1");
-    LOGMAN_THROW_A_FMT(lsb_p_width <= reg_size_bits, "bfxil lsb + width ({}) must be <= {}. lsb={}, width={}",
-                       lsb_p_width, reg_size_bits, lsb, width);
+    LOGMAN_THROW_A_FMT(lsb_p_width <= reg_size_bits, "bfxil lsb + width ({}) must be <= {}. lsb={}, width={}", lsb_p_width, reg_size_bits,
+                       lsb, width);

    bfm(s, rd, rn, lsb, lsb_p_width - 1);
  }
@@ -368,193 +351,148 @@ public:
  }

  void ror(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, uint32_t Imm) {
+    Imm &= RegSizeInBits(s) - 1;
    extr(s, rd, rn, rn, Imm);
  }

  // Data processing - 2 source
  void udiv(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0000'10U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0000'10U << 10);
    DataProcessing_2Source(Op, s, rd, rn, rm);
  }
  void sdiv(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0000'11U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0000'11U << 10);
    DataProcessing_2Source(Op, s, rd, rn, rm);
  }

  void lslv(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0010'00U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0010'00U << 10);
    DataProcessing_2Source(Op, s, rd, rn, rm);
  }
  void lsrv(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0010'01U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0010'01U << 10);
    DataProcessing_2Source(Op, s, rd, rn, rm);
  }
  void asrv(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0010'10U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0010'10U << 10);
    DataProcessing_2Source(Op, s, rd, rn, rm);
  }
  void rorv(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0010'11U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0010'11U << 10);
    DataProcessing_2Source(Op, s, rd, rn, rm);
  }
  void crc32b(ARMEmitter::WRegister rd, ARMEmitter::WRegister rn, ARMEmitter::WRegister rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0100'00U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0100'00U << 10);
    DataProcessing_2Source(Op, ARMEmitter::Size::i32Bit, rd, rn, rm);
  }
  void crc32h(ARMEmitter::WRegister rd, ARMEmitter::WRegister rn, ARMEmitter::WRegister rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0100'01U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0100'01U << 10);
    DataProcessing_2Source(Op, ARMEmitter::Size::i32Bit, rd, rn, rm);
  }
  void crc32w(ARMEmitter::WRegister rd, ARMEmitter::WRegister rn, ARMEmitter::WRegister rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0100'10U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0100'10U << 10);
    DataProcessing_2Source(Op, ARMEmitter::Size::i32Bit, rd, rn, rm);
  }
  void crc32cb(ARMEmitter::WRegister rd, ARMEmitter::WRegister rn, ARMEmitter::WRegister rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0101'00U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0101'00U << 10);
    DataProcessing_2Source(Op, ARMEmitter::Size::i32Bit, rd, rn, rm);
  }
  void crc32ch(ARMEmitter::WRegister rd, ARMEmitter::WRegister rn, ARMEmitter::WRegister rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0101'01U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0101'01U << 10);
    DataProcessing_2Source(Op, ARMEmitter::Size::i32Bit, rd, rn, rm);
  }
  void crc32cw(ARMEmitter::WRegister rd, ARMEmitter::WRegister rn, ARMEmitter::WRegister rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0101'10U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0101'10U << 10);
    DataProcessing_2Source(Op, ARMEmitter::Size::i32Bit, rd, rn, rm);
  }
  void smax(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0110'00U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0110'00U << 10);
    DataProcessing_2Source(Op, s, rd, rn, rm);
  }
  void umax(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0110'01U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0110'01U << 10);
    DataProcessing_2Source(Op, s, rd, rn, rm);
  }
  void smin(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0110'10U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0110'10U << 10);
    DataProcessing_2Source(Op, s, rd, rn, rm);
  }
  void umin(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0110'11U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0110'11U << 10);
    DataProcessing_2Source(Op, s, rd, rn, rm);
  }
  void subp(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn, ARMEmitter::XRegister rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0000'00U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0000'00U << 10);
    DataProcessing_2Source(Op, ARMEmitter::Size::i64Bit, rd, rn, rm);
  }
  void irg(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn, ARMEmitter::XRegister rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0001'00U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0001'00U << 10);
    DataProcessing_2Source(Op, ARMEmitter::Size::i64Bit, rd, rn, rm);
  }
  void gmi(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn, ARMEmitter::XRegister rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0001'01U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0001'01U << 10);
    DataProcessing_2Source(Op, ARMEmitter::Size::i64Bit, rd, rn, rm);
  }
  void pacga(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn, ARMEmitter::XRegister rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0011'00U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0011'00U << 10);
    DataProcessing_2Source(Op, ARMEmitter::Size::i64Bit, rd, rn, rm);
  }
  void crc32x(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn, ARMEmitter::XRegister rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0100'11U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0100'11U << 10);
    DataProcessing_2Source(Op, ARMEmitter::Size::i64Bit, rd, rn, rm);
  }
  void crc32cx(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn, ARMEmitter::XRegister rm) {
-    constexpr uint32_t Op = (0b001'1010'110U << 21) |
-                        (0b0101'11U << 10);
+    constexpr uint32_t Op = (0b001'1010'110U << 21) | (0b0101'11U << 10);
    DataProcessing_2Source(Op, ARMEmitter::Size::i64Bit, rd, rn, rm);
  }
  void subps(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn, ARMEmitter::XRegister rm) {
-    constexpr uint32_t Op = (0b011'1010'110U << 21) |
-                        (0b0000'00U << 10);
+    constexpr uint32_t Op = (0b011'1010'110U << 21) | (0b0000'00U << 10);
    DataProcessing_2Source(Op, ARMEmitter::Size::i64Bit, rd, rn, rm);
  }

  // Data processing - 1 source
  void rbit(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn) {
-    constexpr uint32_t Op = (0b101'1010'110U << 21) |
-                        (0b0'0000U << 16) |
-                        (0b0000'00U << 10);
+    constexpr uint32_t Op = (0b101'1010'110U << 21) | (0b0'0000U << 16) | (0b0000'00U << 10);
    DataProcessing_1Source(Op, s, rd, rn);
  }
  void rev16(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn) {
-    constexpr uint32_t Op = (0b101'1010'110U << 21) |
-                        (0b0'0000U << 16) |
-                        (0b0000'01U << 10);
+    constexpr uint32_t Op = (0b101'1010'110U << 21) | (0b0'0000U << 16) | (0b0000'01U << 10);
    DataProcessing_1Source(Op, s, rd, rn);
  }
  void rev(ARMEmitter::WRegister rd, ARMEmitter::WRegister rn) {
-    constexpr uint32_t Op = (0b101'1010'110U << 21) |
-                        (0b0'0000U << 16) |
-                        (0b0000'10U << 10);
+    constexpr uint32_t Op = (0b101'1010'110U << 21) | (0b0'0000U << 16) | (0b0000'10U << 10);
    DataProcessing_1Source(Op, ARMEmitter::Size::i32Bit, rd, rn);
  }
  void rev32(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn) {
-    constexpr uint32_t Op = (0b101'1010'110U << 21) |
-                        (0b0'0000U << 16) |
-                        (0b0000'10U << 10);
+    constexpr uint32_t Op = (0b101'1010'110U << 21) | (0b0'0000U << 16) | (0b0000'10U << 10);
    DataProcessing_1Source(Op, ARMEmitter::Size::i64Bit, rd, rn);
  }
  void clz(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn) {
-    constexpr uint32_t Op = (0b101'1010'110U << 21) |
-                        (0b0'0000U << 16) |
-                        (0b0001'00U << 10);
+    constexpr uint32_t Op = (0b101'1010'110U << 21) | (0b0'0000U << 16) | (0b0001'00U << 10);
    DataProcessing_1Source(Op, s, rd, rn);
  }
  void cls(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn) {
-    constexpr uint32_t Op = (0b101'1010'110U << 21) |
-                        (0b0'0000U << 16) |
-                        (0b0001'01U << 10);
+    constexpr uint32_t Op = (0b101'1010'110U << 21) | (0b0'0000U << 16) | (0b0001'01U << 10);
    DataProcessing_1Source(Op, s, rd, rn);
  }
  void rev(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn) {
-    constexpr uint32_t Op = (0b101'1010'110U << 21) |
-                        (0b0'0000U << 16) |
-                        (0b0000'11U << 10);
+    constexpr uint32_t Op = (0b101'1010'110U << 21) | (0b0'0000U << 16) | (0b0000'11U << 10);
    DataProcessing_1Source(Op, ARMEmitter::Size::i64Bit, rd, rn);
  }
  void rev(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn) {
-    uint32_t Op = (0b101'1010'110U << 21) |
-                        (0b0'0000U << 16) |
-                        (0b0000'10U << 10) |
-                        (s == ARMEmitter::Size::i64Bit ? (1U << 10) : 0);
+    uint32_t Op = (0b101'1010'110U << 21) | (0b0'0000U << 16) | (0b0000'10U << 10) | (s == ARMEmitter::Size::i64Bit ? (1U << 10) : 0);
    DataProcessing_1Source(Op, s, rd, rn);
  }
  void ctz(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn) {
-    constexpr uint32_t Op = (0b101'1010'110U << 21) |
-                        (0b0'0000U << 16) |
-                        (0b0001'10U << 10);
+    constexpr uint32_t Op = (0b101'1010'110U << 21) | (0b0'0000U << 16) | (0b0001'10U << 10);
    DataProcessing_1Source(Op, s, rd, rn);
  }
  void cnt(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn) {
-    constexpr uint32_t Op = (0b101'1010'110U << 21) |
-                        (0b0'0000U << 16) |
-                        (0b0001'11U << 10);
+    constexpr uint32_t Op = (0b101'1010'110U << 21) | (0b0'0000U << 16) | (0b0001'11U << 10);
    DataProcessing_1Source(Op, s, rd, rn);
  }
  void abs(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn) {
-    constexpr uint32_t Op = (0b101'1010'110U << 21) |
-                        (0b0'0000U << 16) |
-                        (0b0010'00U << 10);
+    constexpr uint32_t Op = (0b101'1010'110U << 21) | (0b0'0000U << 16) | (0b0010'00U << 10);
    DataProcessing_1Source(Op, s, rd, rn);
  }

@@ -571,27 +509,33 @@ public:
    orr(ARMEmitter::Size::i32Bit, rd.R(), ARMEmitter::Reg::zr, rn.R(), ARMEmitter::ShiftType::LSL, 0);
  }

-  void mvn(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void mvn(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL,
+           uint32_t amt = 0) {
    orn(s, rd, ARMEmitter::Reg::zr, rn, Shift, amt);
  }

-  void and_(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void and_(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm,
+            ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    constexpr uint32_t Op = 0b000'1010'000U << 21;
    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
  }
-  void ands(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void ands(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm,
+            ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    constexpr uint32_t Op = 0b110'1010'000U << 21;
    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
  }
-  void bic(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void bic(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm,
+           ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    constexpr uint32_t Op = 0b000'1010'001U << 21;
    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
  }
-  void bics(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void bics(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm,
+            ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    constexpr uint32_t Op = 0b110'1010'001U << 21;
    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
  }
-  void orr(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void orr(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm,
+           ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    constexpr uint32_t Op = 0b010'1010'000U << 21;
    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
  }
@@ -599,30 +543,36 @@ public:
    ands(s, Reg::zr, rn, rm, shift, amt);
  }

-  void orn(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void orn(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm,
+           ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    constexpr uint32_t Op = 0b010'1010'001U << 21;
    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
  }
-  void eor(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void eor(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm,
+           ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    constexpr uint32_t Op = 0b100'1010'000U << 21;
    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
  }
-  void eon(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void eon(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm,
+           ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    constexpr uint32_t Op = 0b100'1010'001U << 21;
    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
  }

  // AddSub - shifted register
-  void add(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn, ARMEmitter::XRegister rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void add(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn, ARMEmitter::XRegister rm,
+           ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    add(ARMEmitter::Size::i64Bit, rd.R(), rn.R(), rm.R(), Shift, amt);
  }
-  void adds(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn, ARMEmitter::XRegister rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void adds(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn, ARMEmitter::XRegister rm,
+            ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    adds(ARMEmitter::Size::i64Bit, rd.R(), rn.R(), rm.R(), Shift, amt);
  }
  void cmn(ARMEmitter::XRegister rn, ARMEmitter::XRegister rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    adds(ARMEmitter::Size::i64Bit, ARMEmitter::XReg::zr, rn.R(), rm.R(), Shift, amt);
  }
-  void sub(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn, ARMEmitter::XRegister rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void sub(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn, ARMEmitter::XRegister rm,
+           ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    sub(ARMEmitter::Size::i64Bit, rd.R(), rn.R(), rm.R(), Shift, amt);
  }
  void neg(ARMEmitter::XRegister rd, ARMEmitter::XRegister rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
@@ -631,23 +581,27 @@ public:
  void cmp(ARMEmitter::XRegister rn, ARMEmitter::XRegister rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    subs(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, rn.R(), rm.R(), Shift, amt);
  }
-  void subs(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn, ARMEmitter::XRegister rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void subs(ARMEmitter::XRegister rd, ARMEmitter::XRegister rn, ARMEmitter::XRegister rm,
+            ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    subs(ARMEmitter::Size::i64Bit, rd.R(), rn.R(), rm.R(), Shift, amt);
  }
  void negs(ARMEmitter::XRegister rd, ARMEmitter::XRegister rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    subs(rd, ARMEmitter::XReg::zr, rm, Shift, amt);
  }

-  void add(ARMEmitter::WRegister rd, ARMEmitter::WRegister rn, ARMEmitter::WRegister rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void add(ARMEmitter::WRegister rd, ARMEmitter::WRegister rn, ARMEmitter::WRegister rm,
+           ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    add(ARMEmitter::Size::i32Bit, rd.R(), rn.R(), rm.R(), Shift, amt);
  }
-  void adds(ARMEmitter::WRegister rd, ARMEmitter::WRegister rn, ARMEmitter::WRegister rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void adds(ARMEmitter::WRegister rd, ARMEmitter::WRegister rn, ARMEmitter::WRegister rm,
+            ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    adds(ARMEmitter::Size::i32Bit, rd.R(), rn.R(), rm.R(), Shift, amt);
  }
  void cmn(ARMEmitter::WRegister rn, ARMEmitter::WRegister rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    adds(ARMEmitter::Size::i32Bit, ARMEmitter::WReg::zr, rn.R(), rm.R(), Shift, amt);
  }
-  void sub(ARMEmitter::WRegister rd, ARMEmitter::WRegister rn, ARMEmitter::WRegister rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void sub(ARMEmitter::WRegister rd, ARMEmitter::WRegister rn, ARMEmitter::WRegister rm,
+           ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    sub(ARMEmitter::Size::i32Bit, rd.R(), rn.R(), rm.R(), Shift, amt);
  }
  void neg(ARMEmitter::WRegister rd, ARMEmitter::WRegister rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
@@ -656,65 +610,78 @@ public:
  void cmp(ARMEmitter::WRegister rn, ARMEmitter::WRegister rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    subs(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::rsp, rn.R(), rm.R(), Shift, amt);
  }
-  void subs(ARMEmitter::WRegister rd, ARMEmitter::WRegister rn, ARMEmitter::WRegister rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void subs(ARMEmitter::WRegister rd, ARMEmitter::WRegister rn, ARMEmitter::WRegister rm,
+            ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    subs(ARMEmitter::Size::i32Bit, rd.R(), rn.R(), rm.R(), Shift, amt);
  }
  void negs(ARMEmitter::WRegister rd, ARMEmitter::WRegister rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
    subs(rd, ARMEmitter::WReg::zr, rm, Shift, amt);
  }

-  void add(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
-    LOGMAN_THROW_AA_FMT(Shift != ARMEmitter::ShiftType::ROR, "Doesn't support ROR");
+  void add(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm,
+           ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    LOGMAN_THROW_A_FMT(Shift != ARMEmitter::ShiftType::ROR, "Doesn't support ROR");
    constexpr uint32_t Op = 0b000'1011'000U << 21;
    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
  }
-  void adds(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
-    LOGMAN_THROW_AA_FMT(Shift != ARMEmitter::ShiftType::ROR, "Doesn't support ROR");
+  void adds(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm,
+            ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    LOGMAN_THROW_A_FMT(Shift != ARMEmitter::ShiftType::ROR, "Doesn't support ROR");
    constexpr uint32_t Op = 0b010'1011'000U << 21;
    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
  }
-  void cmn(ARMEmitter::Size s, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void cmn(ARMEmitter::Size s, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL,
+           uint32_t amt = 0) {
    adds(s, ARMEmitter::Reg::zr, rn, rm, Shift, amt);
  }
-  void sub(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
-    LOGMAN_THROW_AA_FMT(Shift != ARMEmitter::ShiftType::ROR, "Doesn't support ROR");
+  void sub(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm,
+           ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    LOGMAN_THROW_A_FMT(Shift != ARMEmitter::ShiftType::ROR, "Doesn't support ROR");
    constexpr uint32_t Op = 0b100'1011'000U << 21;
    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
  }
-  void neg(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void neg(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL,
+           uint32_t amt = 0) {
    sub(s, rd, ARMEmitter::Reg::zr, rm, Shift, amt);
  }
-  void cmp(ARMEmitter::Size s, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void cmp(ARMEmitter::Size s, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL,
+           uint32_t amt = 0) {
    subs(s, ARMEmitter::Reg::zr, rn, rm, Shift, amt);
  }

-  void subs(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
-    LOGMAN_THROW_AA_FMT(Shift != ARMEmitter::ShiftType::ROR, "Doesn't support ROR");
+  void subs(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm,
+            ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    LOGMAN_THROW_A_FMT(Shift != ARMEmitter::ShiftType::ROR, "Doesn't support ROR");
    constexpr uint32_t Op = 0b110'1011'000U << 21;
    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
  }
-  void negs(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+  void negs(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift = ARMEmitter::ShiftType::LSL,
+            uint32_t amt = 0) {
    subs(s, rd, ARMEmitter::Reg::zr, rm, Shift, amt);
  }

  // AddSub - extended register
-  void add(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ExtendedType Option, uint32_t Shift = 0) {
-    LOGMAN_THROW_AA_FMT(Shift <= 4, "Shift amount is too large");
+  void add(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ExtendedType Option,
+           uint32_t Shift = 0) {
+    LOGMAN_THROW_A_FMT(Shift <= 4, "Shift amount is too large");
    constexpr uint32_t Op = 0b000'1011'001U << 21;
    DataProcessing_Extended_Reg(Op, s, rd, rn, rm, Option, Shift);
  }
-  void adds(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ExtendedType Option, uint32_t Shift = 0) {
+  void adds(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ExtendedType Option,
+            uint32_t Shift = 0) {
    constexpr uint32_t Op = 0b010'1011'001U << 21;
    DataProcessing_Extended_Reg(Op, s, rd, rn, rm, Option, Shift);
  }
  void cmn(ARMEmitter::Size s, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ExtendedType Option, uint32_t Shift = 0) {
    adds(s, ARMEmitter::Reg::zr, rn, rm, Option, Shift);
  }
-  void sub(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ExtendedType Option, uint32_t Shift = 0) {
+  void sub(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ExtendedType Option,
+           uint32_t Shift = 0) {
    constexpr uint32_t Op = 0b100'1011'001U << 21;
    DataProcessing_Extended_Reg(Op, s, rd, rn, rm, Option, Shift);
  }
-  void subs(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ExtendedType Option, uint32_t Shift = 0) {
+  void subs(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ExtendedType Option,
+            uint32_t Shift = 0) {
    constexpr uint32_t Op = 0b110'1011'001U << 21;
    DataProcessing_Extended_Reg(Op, s, rd, rn, rm, Option, Shift);
  }
@@ -749,8 +716,8 @@ public:

  // Rotate right into flags
  void rmif(XRegister rn, uint32_t shift, uint32_t mask) {
-    LOGMAN_THROW_AA_FMT(shift <= 63, "Shift must be within 0-63. Shift: {}", shift);
-    LOGMAN_THROW_AA_FMT(mask <= 15, "Mask must be within 0-15. Mask: {}", mask);
+    LOGMAN_THROW_A_FMT(shift <= 63, "Shift must be within 0-63. Shift: {}", shift);
+    LOGMAN_THROW_A_FMT(mask <= 15, "Mask must be within 0-15. Mask: {}", mask);

    uint32_t Op = 0b1011'1010'0000'0000'0000'0100'0000'0000;
    Op |= rn.Idx() << 5;
@@ -814,7 +781,8 @@ public:
  }
  void cset(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Condition Cond) {
    constexpr uint32_t Op = 0b0001'1010'100 << 21;
-    ConditionalCompare(Op, 0, 0b01, s, rd, ARMEmitter::Reg::zr, ARMEmitter::Reg::zr, static_cast<ARMEmitter::Condition>(FEXCore::ToUnderlying(Cond) ^ FEXCore::ToUnderlying(ARMEmitter::Condition::CC_NE)));
+    ConditionalCompare(Op, 0, 0b01, s, rd, ARMEmitter::Reg::zr, ARMEmitter::Reg::zr,
+                       static_cast<ARMEmitter::Condition>(FEXCore::ToUnderlying(Cond) ^ FEXCore::ToUnderlying(ARMEmitter::Condition::CC_NE)));
  }
  void csinc(ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::Condition Cond) {
    constexpr uint32_t Op = 0b0001'1010'100 << 21;
@@ -896,8 +864,7 @@ public:
 private:
  static constexpr Condition InvertCondition(Condition cond) {
    // These behave as always, so it makes no sense to allow inverting these.
-    LOGMAN_THROW_AA_FMT(cond != Condition::CC_AL && cond != Condition::CC_NV,
-                        "Cannot invert CC_AL or CC_NV");
+    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);
  }

@@ -948,7 +915,7 @@ private:
      LSL12 = true;
      Imm >>= 12;
    }
-    LOGMAN_THROW_AA_FMT(TooLarge == false, "Imm amount too large: 0x{:x}", Imm);
+    LOGMAN_THROW_A_FMT(TooLarge == false, "Imm amount too large: 0x{:x}", Imm);

    const uint32_t SF = s == ARMEmitter::Size::i64Bit ? (1U << 31) : 0;

@@ -993,7 +960,8 @@ private:
  }

  // Logical immediate
-  void DataProcessing_Logical_Imm(uint32_t Op, ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, uint32_t n, uint32_t immr, uint32_t imms) {
+  void DataProcessing_Logical_Imm(uint32_t Op, ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, uint32_t n,
+                                  uint32_t immr, uint32_t imms) {
    const uint32_t SF = s == ARMEmitter::Size::i64Bit ? (1U << 31) : 0;

    uint32_t Instr = Op;
@@ -1012,9 +980,8 @@ private:
    [[maybe_unused]] const auto lsb_p_width = lsb + width;
    const auto reg_size_bits = RegSizeInBits(s);

-    LOGMAN_THROW_AA_FMT(lsb_p_width <= reg_size_bits, "lsb + width ({}) must be <= {}. lsb={}, width={}",
-                        lsb_p_width, reg_size_bits, lsb, width);
-    LOGMAN_THROW_AA_FMT(width >= 1, "xbfiz width must be >= 1");
+    LOGMAN_THROW_A_FMT(lsb_p_width <= reg_size_bits, "lsb + width ({}) must be <= {}. lsb={}, width={}", lsb_p_width, reg_size_bits, lsb, width);
+    LOGMAN_THROW_A_FMT(width >= 1, "xbfiz width must be >= 1");

    const auto immr = (reg_size_bits - lsb) & (reg_size_bits - 1);
    const auto imms = width - 1;
@@ -1026,12 +993,13 @@ private:
    }
  }

-  void DataProcessing_Extract(uint32_t Op, ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, uint32_t Imm) {
+  void DataProcessing_Extract(uint32_t Op, ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm,
+                              uint32_t Imm) {
    const uint32_t SF = s == ARMEmitter::Size::i64Bit ? (1U << 31) : 0;

    // Current ARMv8 spec hardcodes SF == N for this class of instructions.
    // Anythign else is undefined behaviour.
-    const uint32_t N  = s == ARMEmitter::Size::i64Bit ? (1U << 22) : 0;
+    const uint32_t N = s == ARMEmitter::Size::i64Bit ? (1U << 22) : 0;

    uint32_t Instr = Op;

@@ -1074,10 +1042,11 @@ private:
  }

  // AddSub - shifted register
-  void DataProcessing_Shifted_Reg(uint32_t Op, ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ShiftType Shift, uint32_t amt) {
-    LOGMAN_THROW_AA_FMT((amt & ~0b11'1111U) == 0, "Shift amount too large");
+  void DataProcessing_Shifted_Reg(uint32_t Op, ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn,
+                                  ARMEmitter::Register rm, ARMEmitter::ShiftType Shift, uint32_t amt) {
+    LOGMAN_THROW_A_FMT((amt & ~0b11'1111U) == 0, "Shift amount too large");
    if (s == ARMEmitter::Size::i32Bit) {
-      LOGMAN_THROW_AA_FMT(amt < 32, "Shift amount for 32-bit must be below 32");
+      LOGMAN_THROW_A_FMT(amt < 32, "Shift amount for 32-bit must be below 32");
    }

    const uint32_t SF = s == ARMEmitter::Size::i64Bit ? (1U << 31) : 0;
@@ -1095,7 +1064,8 @@ private:
  }

  // AddSub - extended register
-  void DataProcessing_Extended_Reg(uint32_t Op, ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::ExtendedType Option, uint32_t Shift) {
+  void DataProcessing_Extended_Reg(uint32_t Op, ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn,
+                                   ARMEmitter::Register rm, ARMEmitter::ExtendedType Option, uint32_t Shift) {
    const uint32_t SF = s == ARMEmitter::Size::i64Bit ? (1U << 31) : 0;

    uint32_t Instr = Op;
@@ -1111,7 +1081,8 @@ private:
  }
  // Conditional compare - register
  template<typename T>
-  void ConditionalCompare(uint32_t Op, uint32_t o1, uint32_t o2, uint32_t o3, ARMEmitter::Size s, ARMEmitter::Register rn, T rm, ARMEmitter::StatusFlags flags, ARMEmitter::Condition Cond) {
+  void ConditionalCompare(uint32_t Op, uint32_t o1, uint32_t o2, uint32_t o3, ARMEmitter::Size s, ARMEmitter::Register rn, T rm,
+                          ARMEmitter::StatusFlags flags, ARMEmitter::Condition Cond) {
    const uint32_t SF = s == ARMEmitter::Size::i64Bit ? (1U << 31) : 0;

    uint32_t Instr = Op;
@@ -1129,7 +1100,8 @@ private:
  }

  template<typename T>
-  void ConditionalCompare(uint32_t Op, uint32_t o1, uint32_t o2, ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, T rm, ARMEmitter::Condition Cond) {
+  void ConditionalCompare(uint32_t Op, uint32_t o1, uint32_t o2, ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, T rm,
+                          ARMEmitter::Condition Cond) {
    const uint32_t SF = s == ARMEmitter::Size::i64Bit ? (1U << 31) : 0;

    uint32_t Instr = Op;
@@ -1146,7 +1118,8 @@ private:
  }

  // Data-processing - 3 source
-  void DataProcessing_3Source(uint32_t Op, uint32_t Op0, ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn, ARMEmitter::Register rm, ARMEmitter::Register ra) {
+  void DataProcessing_3Source(uint32_t Op, uint32_t Op0, ARMEmitter::Size s, ARMEmitter::Register rd, ARMEmitter::Register rn,
+                              ARMEmitter::Register rm, ARMEmitter::Register ra) {
    const uint32_t SF = s == ARMEmitter::Size::i64Bit ? (1U << 31) : 0;

    uint32_t Instr = Op;
@@ -1168,4 +1141,7 @@ private:
    dc32(Instr);
  }

-
+#ifndef INCLUDED_BY_EMITTER
+}; // struct LoadstoreEmitterOps
+} // namespace ARMEmitter
+#endif
@@ -3,339 +3,325 @@
 *
 * Most of these instructions will use `BackwardLabel`, `ForwardLabel`, or `BiDirectionLabel` to determine where a branch targets.
 */
+
+#pragma once
+#ifndef INCLUDED_BY_EMITTER
+#include <CodeEmitter/Emitter.h>
+namespace ARMEmitter {
+struct EmitterOps : Emitter {
+#endif
+
 public:
  // Branches, Exception Generating and System instructions
-  public:
-    // Conditional branch immediate
-    ///< Branch conditional
-    void b(ARMEmitter::Condition Cond, uint32_t Imm) {
-      constexpr uint32_t Op = 0b0101'010 << 25;
-      Branch_Conditional(Op, 0, 0, Cond, Imm);
+public:
+  // Conditional branch immediate
+  ///< Branch conditional
+  void b(ARMEmitter::Condition Cond, uint32_t Imm) {
+    constexpr uint32_t Op = 0b0101'010 << 25;
+    Branch_Conditional(Op, 0, 0, Cond, Imm);
+  }
+  void 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);
+  }
+  void 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);
+  }
+
+  void b(ARMEmitter::Condition Cond, BiDirectionalLabel* Label) {
+    if (Label->Backward.Location) {
+      b(Cond, &Label->Backward);
+    } else {
+      b(Cond, &Label->Forward);
    }
-    void b(ARMEmitter::Condition Cond, BackwardLabel const* Label) {
-      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-      LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
-      constexpr uint32_t Op = 0b0101'010 << 25;
-      Branch_Conditional(Op, 0, 0, Cond, Imm >> 2);
+  }
+
+  ///< Branch consistent conditional
+  void bc(ARMEmitter::Condition Cond, uint32_t Imm) {
+    constexpr uint32_t Op = 0b0101'010 << 25;
+    Branch_Conditional(Op, 0, 1, Cond, Imm);
+  }
+  void 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);
+  }
+
+  void 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);
+  }
+
+  void bc(ARMEmitter::Condition Cond, BiDirectionalLabel* Label) {
+    if (Label->Backward.Location) {
+      bc(Cond, &Label->Backward);
+    } else {
+      bc(Cond, &Label->Forward);
    }
-    template<typename LabelType>
-    requires (std::is_same_v<LabelType, ForwardLabel> || std::is_same_v<LabelType, SingleUseForwardLabel>)
-    void b(ARMEmitter::Condition Cond, LabelType *Label) {
-      AddLocationToLabel(Label, SingleUseForwardLabel{ .Location = GetCursorAddress<uint8_t*>(), .Type = SingleUseForwardLabel::InstType::BC });
-      constexpr uint32_t Op = 0b0101'010 << 25;
-      Branch_Conditional(Op, 0, 0, Cond, 0);
+  }
+
+  // Unconditional branch register
+  void br(ARMEmitter::Register rn) {
+    constexpr uint32_t Op = 0b1101011 << 25 | 0b0'000 << 21 | // opc
+                            0b1'1111 << 16 |                  // op2
+                            0b0000'00 << 10 |                 // op3
+                            0b0'0000;                         // op4
+
+    UnconditionalBranch(Op, rn);
+  }
+  void blr(ARMEmitter::Register rn) {
+    constexpr uint32_t Op = 0b1101011 << 25 | 0b0'001 << 21 | // opc
+                            0b1'1111 << 16 |                  // op2
+                            0b0000'00 << 10 |                 // op3
+                            0b0'0000;                         // op4
+
+    UnconditionalBranch(Op, rn);
+  }
+  void ret(ARMEmitter::Register rn = ARMEmitter::Reg::r30) {
+    constexpr uint32_t Op = 0b1101011 << 25 | 0b0'010 << 21 | // opc
+                            0b1'1111 << 16 |                  // op2
+                            0b0000'00 << 10 |                 // op3
+                            0b0'0000;                         // op4
+
+    UnconditionalBranch(Op, rn);
+  }
+
+  // Unconditional branch immediate
+  void b(uint32_t Imm) {
+    constexpr uint32_t Op = 0b0001'01 << 26;
+
+    UnconditionalBranch(Op, Imm);
+  }
+  void b(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;
+
+    UnconditionalBranch(Op, Imm >> 2);
+  }
+  void 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);
+  }
+
+  void b(BiDirectionalLabel* Label) {
+    if (Label->Backward.Location) {
+      b(&Label->Backward);
+    } else {
+      b(&Label->Forward);
    }
+  }

-    void b(ARMEmitter::Condition Cond, BiDirectionalLabel *Label) {
-      if (Label->Backward.Location) {
-        b(Cond, &Label->Backward);
-      }
-      else {
-        b(Cond, &Label->Forward);
-      }
+  void bl(uint32_t Imm) {
+    constexpr uint32_t Op = 0b1001'01 << 26;
+
+    UnconditionalBranch(Op, Imm);
+  }
+
+  void 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;
+
+    UnconditionalBranch(Op, Imm >> 2);
+  }
+  void 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);
+  }
+
+  void bl(BiDirectionalLabel* Label) {
+    if (Label->Backward.Location) {
+      bl(&Label->Backward);
+    } else {
+      bl(&Label->Forward);
    }
+  }

-    ///< Branch consistent conditional
-    void bc(ARMEmitter::Condition Cond, uint32_t Imm) {
-      constexpr uint32_t Op = 0b0101'010 << 25;
-      Branch_Conditional(Op, 0, 1, Cond, Imm);
+  // Compare and branch
+  void cbz(ARMEmitter::Size s, ARMEmitter::Register rt, uint32_t Imm) {
+    constexpr uint32_t Op = 0b0011'0100 << 24;
+
+    CompareAndBranch(Op, s, rt, Imm);
+  }
+
+  void 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;
+
+    CompareAndBranch(Op, s, rt, Imm >> 2);
+  }
+
+  void 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);
+  }
+
+  void cbz(ARMEmitter::Size s, ARMEmitter::Register rt, BiDirectionalLabel* Label) {
+    if (Label->Backward.Location) {
+      cbz(s, rt, &Label->Backward);
+    } else {
+      cbz(s, rt, &Label->Forward);
    }
-    void bc(ARMEmitter::Condition Cond, BackwardLabel const* Label) {
-      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-      LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
-      constexpr uint32_t Op = 0b0101'010 << 25;
-      Branch_Conditional(Op, 0, 1, Cond, Imm >> 2);
+  }
+
+  void cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, uint32_t Imm) {
+    constexpr uint32_t Op = 0b0011'0101 << 24;
+
+    CompareAndBranch(Op, s, rt, Imm);
+  }
+
+  void 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;
+
+    CompareAndBranch(Op, s, rt, Imm >> 2);
+  }
+
+  void 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);
+  }
+
+  void cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, BiDirectionalLabel* Label) {
+    if (Label->Backward.Location) {
+      cbnz(s, rt, &Label->Backward);
+    } else {
+      cbnz(s, rt, &Label->Forward);
    }
+  }

-    template<typename LabelType>
-    requires (std::is_same_v<LabelType, ForwardLabel> || std::is_same_v<LabelType, SingleUseForwardLabel>)
-    void bc(ARMEmitter::Condition Cond, LabelType *Label) {
-      AddLocationToLabel(Label, SingleUseForwardLabel{ .Location = GetCursorAddress<uint8_t*>(), .Type = SingleUseForwardLabel::InstType::BC });
-      constexpr uint32_t Op = 0b0101'010 << 25;
-      Branch_Conditional(Op, 0, 1, Cond, 0);
+  // Test and branch immediate
+  void tbz(ARMEmitter::Register rt, uint32_t Bit, uint32_t Imm) {
+    constexpr uint32_t Op = 0b0011'0110 << 24;
+
+    TestAndBranch(Op, rt, Bit, Imm);
+  }
+  void 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;
+
+    TestAndBranch(Op, rt, Bit, Imm >> 2);
+  }
+
+  void 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);
+  }
+
+  void tbz(ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel* Label) {
+    if (Label->Backward.Location) {
+      tbz(rt, Bit, &Label->Backward);
+    } else {
+      tbz(rt, Bit, &Label->Forward);
    }
+  }

-    void bc(ARMEmitter::Condition Cond, BiDirectionalLabel *Label) {
-      if (Label->Backward.Location) {
-        bc(Cond, &Label->Backward);
-      }
-      else {
-        bc(Cond, &Label->Forward);
-      }
-    }
+  void tbnz(ARMEmitter::Register rt, uint32_t Bit, uint32_t Imm) {
+    constexpr uint32_t Op = 0b0011'0111 << 24;

-    // Unconditional branch register
-    void br(ARMEmitter::Register rn) {
-      constexpr uint32_t Op = 0b1101011 << 25 |
-                              0b0'000 << 21 |   // opc
-                              0b1'1111 << 16 |  // op2
-                              0b0000'00 << 10 | // op3
-                              0b0'0000;         // op4
+    TestAndBranch(Op, rt, Bit, Imm);
+  }
+  void 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");

-      UnconditionalBranch(Op, rn);
-    }
-    void blr(ARMEmitter::Register rn) {
-      constexpr uint32_t Op = 0b1101011 << 25 |
-                              0b0'001 << 21 |   // opc
-                              0b1'1111 << 16 |  // op2
-                              0b0000'00 << 10 | // op3
-                              0b0'0000;         // op4
+    constexpr uint32_t Op = 0b0011'0111 << 24;

-      UnconditionalBranch(Op, rn);
-    }
-    void ret(ARMEmitter::Register rn = ARMEmitter::Reg::r30) {
-      constexpr uint32_t Op = 0b1101011 << 25 |
-                              0b0'010 << 21 |   // opc
-                              0b1'1111 << 16 |  // op2
-                              0b0000'00 << 10 | // op3
-                              0b0'0000;         // op4
+    TestAndBranch(Op, rt, Bit, Imm >> 2);
+  }

-      UnconditionalBranch(Op, rn);
-    }
+  void 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;

-    // Unconditional branch immediate
-    void b(uint32_t Imm) {
-      constexpr uint32_t Op = 0b0001'01 << 26;
+    TestAndBranch(Op, rt, Bit, 0);
+  }

-      UnconditionalBranch(Op, Imm);
-    }
-    void b(BackwardLabel const* Label) {
-      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-      LOGMAN_THROW_A_FMT(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0), "Unscaled offset too large");
-      constexpr uint32_t Op = 0b0001'01 << 26;
-
-      UnconditionalBranch(Op, Imm >> 2);
-    }
-    template<typename LabelType>
-    requires (std::is_same_v<LabelType, ForwardLabel> || std::is_same_v<LabelType, SingleUseForwardLabel>)
-    void b(LabelType *Label) {
-      AddLocationToLabel(Label, SingleUseForwardLabel{ .Location = GetCursorAddress<uint8_t*>(), .Type = SingleUseForwardLabel::InstType::B });
-      constexpr uint32_t Op = 0b0001'01 << 26;
-
-      UnconditionalBranch(Op, 0);
-    }
-
-    void b(BiDirectionalLabel *Label) {
-      if (Label->Backward.Location) {
-        b(&Label->Backward);
-      }
-      else {
-        b(&Label->Forward);
-      }
-    }
-
-    void bl(uint32_t Imm) {
-      constexpr uint32_t Op = 0b1001'01 << 26;
-
-      UnconditionalBranch(Op, Imm);
-    }
-
-    void bl(BackwardLabel const* Label) {
-      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-      LOGMAN_THROW_A_FMT(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0), "Unscaled offset too large");
-      constexpr uint32_t Op = 0b1001'01 << 26;
-
-      UnconditionalBranch(Op, Imm >> 2);
-    }
-    template<typename LabelType>
-    requires (std::is_same_v<LabelType, ForwardLabel> || std::is_same_v<LabelType, SingleUseForwardLabel>)
-    void bl(LabelType *Label) {
-      AddLocationToLabel(Label, SingleUseForwardLabel{ .Location = GetCursorAddress<uint8_t*>(), .Type = SingleUseForwardLabel::InstType::B });
-      constexpr uint32_t Op = 0b1001'01 << 26;
-
-      UnconditionalBranch(Op, 0);
-    }
-
-    void bl(BiDirectionalLabel *Label) {
-      if (Label->Backward.Location) {
-        bl(&Label->Backward);
-      }
-      else {
-        bl(&Label->Forward);
-      }
-    }
-
-    // Compare and branch
-    void cbz(ARMEmitter::Size s, ARMEmitter::Register rt, uint32_t Imm) {
-      constexpr uint32_t Op = 0b0011'0100 << 24;
-
-      CompareAndBranch(Op, s, rt, Imm);
-    }
-
-    void cbz(ARMEmitter::Size s, ARMEmitter::Register rt, BackwardLabel const* Label) {
-      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-      LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
-
-      constexpr uint32_t Op = 0b0011'0100 << 24;
-
-      CompareAndBranch(Op, s, rt, Imm >> 2);
-    }
-
-    template<typename LabelType>
-    requires (std::is_same_v<LabelType, ForwardLabel> || std::is_same_v<LabelType, SingleUseForwardLabel>)
-    void cbz(ARMEmitter::Size s, ARMEmitter::Register rt, LabelType *Label) {
-      AddLocationToLabel(Label, SingleUseForwardLabel{ .Location = GetCursorAddress<uint8_t*>(), .Type = SingleUseForwardLabel::InstType::BC });
-
-      constexpr uint32_t Op = 0b0011'0100 << 24;
-
-      CompareAndBranch(Op, s, rt, 0);
-    }
-
-    void cbz(ARMEmitter::Size s, ARMEmitter::Register rt, BiDirectionalLabel *Label) {
-      if (Label->Backward.Location) {
-        cbz(s, rt, &Label->Backward);
-      }
-      else {
-        cbz(s, rt, &Label->Forward);
-      }
-    }
-
-    void cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, uint32_t Imm) {
-      constexpr uint32_t Op = 0b0011'0101 << 24;
-
-      CompareAndBranch(Op, s, rt, Imm);
-    }
-
-    void cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, BackwardLabel const* Label) {
-      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-      LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
-
-      constexpr uint32_t Op = 0b0011'0101 << 24;
-
-      CompareAndBranch(Op, s, rt, Imm >> 2);
-    }
-
-    template<typename LabelType>
-    requires (std::is_same_v<LabelType, ForwardLabel> || std::is_same_v<LabelType, SingleUseForwardLabel>)
-    void cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, LabelType *Label) {
-      AddLocationToLabel(Label, SingleUseForwardLabel{ .Location = GetCursorAddress<uint8_t*>(), .Type = SingleUseForwardLabel::InstType::BC });
-
-      constexpr uint32_t Op = 0b0011'0101 << 24;
-
-      CompareAndBranch(Op, s, rt, 0);
-    }
-
-    void cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, BiDirectionalLabel *Label) {
-      if (Label->Backward.Location) {
-        cbnz(s, rt, &Label->Backward);
-      }
-      else {
-        cbnz(s, rt, &Label->Forward);
-      }
-    }
-
-    // Test and branch immediate
-    void tbz(ARMEmitter::Register rt, uint32_t Bit, uint32_t Imm) {
-      constexpr uint32_t Op = 0b0011'0110 << 24;
-
-      TestAndBranch(Op, rt, Bit, Imm);
-    }
-    void tbz(ARMEmitter::Register rt, uint32_t Bit, BackwardLabel const* Label) {
-      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-      LOGMAN_THROW_A_FMT(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0), "Unscaled offset too large");
-
-      constexpr uint32_t Op = 0b0011'0110 << 24;
-
-      TestAndBranch(Op, rt, Bit, Imm >> 2);
-    }
-
-    template<typename LabelType>
-    requires (std::is_same_v<LabelType, ForwardLabel> || std::is_same_v<LabelType, SingleUseForwardLabel>)
-    void tbz(ARMEmitter::Register rt, uint32_t Bit, LabelType *Label) {
-      AddLocationToLabel(Label, SingleUseForwardLabel{ .Location = GetCursorAddress<uint8_t*>(), .Type = SingleUseForwardLabel::InstType::TEST_BRANCH });
-
-      constexpr uint32_t Op = 0b0011'0110 << 24;
-
-      TestAndBranch(Op, rt, Bit, 0);
-    }
-
-    void tbz(ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel *Label) {
-      if (Label->Backward.Location) {
-        tbz(rt, Bit, &Label->Backward);
-      }
-      else {
-        tbz(rt, Bit, &Label->Forward);
-      }
-    }
-
-    void tbnz(ARMEmitter::Register rt, uint32_t Bit, uint32_t Imm) {
-      constexpr uint32_t Op = 0b0011'0111 << 24;
-
-      TestAndBranch(Op, rt, Bit, Imm);
-    }
-    void tbnz(ARMEmitter::Register rt, uint32_t Bit, BackwardLabel const* Label) {
-      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-      LOGMAN_THROW_A_FMT(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0), "Unscaled offset too large");
-
-      constexpr uint32_t Op = 0b0011'0111 << 24;
-
-      TestAndBranch(Op, rt, Bit, Imm >> 2);
-    }
-
-    template<typename LabelType>
-    requires (std::is_same_v<LabelType, ForwardLabel> || std::is_same_v<LabelType, SingleUseForwardLabel>)
-    void tbnz(ARMEmitter::Register rt, uint32_t Bit, LabelType *Label) {
-      AddLocationToLabel(Label, SingleUseForwardLabel{ .Location = GetCursorAddress<uint8_t*>(), .Type = SingleUseForwardLabel::InstType::TEST_BRANCH });
-      constexpr uint32_t Op = 0b0011'0111 << 24;
-
-      TestAndBranch(Op, rt, Bit, 0);
-    }
-
-    void tbnz(ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel *Label) {
-      if (Label->Backward.Location) {
-        tbnz(rt, Bit, &Label->Backward);
-      }
-      else {
-        tbnz(rt, Bit, &Label->Forward);
-      }
+  void tbnz(ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel* Label) {
+    if (Label->Backward.Location) {
+      tbnz(rt, Bit, &Label->Backward);
+    } else {
+      tbnz(rt, Bit, &Label->Forward);
    }
+  }

 private:
-    // Conditional branch immediate
-    void Branch_Conditional(uint32_t Op, uint32_t Op1, uint32_t Op0, ARMEmitter::Condition Cond, uint32_t Imm) {
-      uint32_t Instr = Op;
+  // Conditional branch immediate
+  void Branch_Conditional(uint32_t Op, uint32_t Op1, uint32_t Op0, ARMEmitter::Condition Cond, uint32_t Imm) {
+    uint32_t Instr = Op;

-      Instr |= Op1 << 24;
-      Instr |= (Imm & 0x7'FFFF) << 5;
-      Instr |= Op0 << 4;
-      Instr |= FEXCore::ToUnderlying(Cond);
+    Instr |= Op1 << 24;
+    Instr |= (Imm & 0x7'FFFF) << 5;
+    Instr |= Op0 << 4;
+    Instr |= FEXCore::ToUnderlying(Cond);

-      dc32(Instr);
-    }
+    dc32(Instr);
+  }

-    // Unconditional branch register
-    void UnconditionalBranch(uint32_t Op, ARMEmitter::Register rn) {
-      uint32_t Instr = Op;
-      Instr |= Encode_rn(rn);
-      dc32(Instr);
-    }
+  // Unconditional branch register
+  void UnconditionalBranch(uint32_t Op, ARMEmitter::Register rn) {
+    uint32_t Instr = Op;
+    Instr |= Encode_rn(rn);
+    dc32(Instr);
+  }

-    // Unconditional branch - immediate
-    void UnconditionalBranch(uint32_t Op, uint32_t Imm) {
-      uint32_t Instr = Op;
-      Instr |= Imm & 0x3FF'FFFF;
-      dc32(Instr);
-    }
+  // Unconditional branch - immediate
+  void UnconditionalBranch(uint32_t Op, uint32_t Imm) {
+    uint32_t Instr = Op;
+    Instr |= Imm & 0x3FF'FFFF;
+    dc32(Instr);
+  }

-    // Compare and branch
-    void CompareAndBranch(uint32_t Op, ARMEmitter::Size s, ARMEmitter::Register rt, uint32_t Imm) {
-      const uint32_t SF = s == ARMEmitter::Size::i64Bit ? (1U << 31) : 0;
+  // Compare and branch
+  void CompareAndBranch(uint32_t Op, ARMEmitter::Size s, ARMEmitter::Register rt, uint32_t Imm) {
+    const uint32_t SF = s == ARMEmitter::Size::i64Bit ? (1U << 31) : 0;

-      uint32_t Instr = Op;
+    uint32_t Instr = Op;

-      Instr |= SF;
-      Instr |= (Imm & 0x7'FFFF) << 5;
-      Instr |= Encode_rt(rt);
-      dc32(Instr);
-    }
+    Instr |= SF;
+    Instr |= (Imm & 0x7'FFFF) << 5;
+    Instr |= Encode_rt(rt);
+    dc32(Instr);
+  }

-    // Test and branch - immediate
-    void TestAndBranch(uint32_t Op, ARMEmitter::Register rt, uint32_t Bit, uint32_t Imm) {
-      uint32_t Instr = Op;
+  // Test and branch - immediate
+  void TestAndBranch(uint32_t Op, ARMEmitter::Register rt, uint32_t Bit, uint32_t Imm) {
+    uint32_t Instr = Op;

-      Instr |= (Bit >> 5) << 31;
-      Instr |= (Bit & 0b1'1111) << 19;
-      Instr |= (Imm & 0x3FFF) << 5;
-      Instr |= Encode_rt(rt);
-      dc32(Instr);
-    }
+    Instr |= (Bit >> 5) << 31;
+    Instr |= (Bit & 0b1'1111) << 19;
+    Instr |= (Imm & 0x3FFF) << 5;
+    Instr |= Encode_rt(rt);
+    dc32(Instr);
+  }
+
+#ifndef INCLUDED_BY_EMITTER
+}; // struct LoadstoreEmitterOps
+} // namespace ARMEmitter
+#endif
@@ -3,6 +3,7 @@
 #include <cstddef>
 #include <cstdint>
 #include <cstring>
+#include <type_traits>

 namespace ARMEmitter {
 class Buffer {
@@ -21,42 +22,38 @@ public:
    Size = BaseSize;
  }

+  template<typename T>
+  requires (std::is_trivially_copyable_v<T>)
+  void dcn(const T& Data) {
+    std::memcpy(CurrentOffset, &Data, sizeof(Data));
+    CurrentOffset += sizeof(Data);
+  }
  void dc8(uint8_t Data) {
-    decltype(Data)* Memory = reinterpret_cast<decltype(Data)*>(CurrentOffset);
-    *Memory = Data;
-    CurrentOffset += sizeof(Data);
+    dcn(Data);
  }
-
  void dc16(uint16_t Data) {
-    decltype(Data)* Memory = reinterpret_cast<decltype(Data)*>(CurrentOffset);
-    *Memory = Data;
-    CurrentOffset += sizeof(Data);
+    dcn(Data);
  }
-
  void dc32(uint32_t Data) {
-    decltype(Data)* Memory = reinterpret_cast<decltype(Data)*>(CurrentOffset);
-    *Memory = Data;
-    CurrentOffset += sizeof(Data);
+    dcn(Data);
+  }
+  void dc64(uint64_t Data) {
+    dcn(Data);
  }

-  void dc64(uint64_t Data) {
-    decltype(Data)* Memory = reinterpret_cast<decltype(Data)*>(CurrentOffset);
-    *Memory = Data;
-    CurrentOffset += sizeof(Data);
-  }
  void EmitString(const char* String) {
    const auto StringLength = strlen(String);
    memcpy(CurrentOffset, String, StringLength);
    CurrentOffset += StringLength;
  }

-  void Align() {
-    // Align the buffer to instruction size
-    auto CurrentAlignment = reinterpret_cast<uint64_t>(CurrentOffset) & 0b11;
+  void Align(size_t Size = 4) {
+    // Align the buffer to provided size.
+    auto CurrentAlignment = reinterpret_cast<uint64_t>(CurrentOffset) & (Size - 1);
    if (!CurrentAlignment) {
      return;
    }
-    CurrentOffset += 4 - CurrentAlignment;
+    CurrentOffset += Size - CurrentAlignment;
  }

  template<typename T>
@@ -95,10 +92,6 @@ public:

 protected:

-  void ResetBuffer() {
-    CurrentOffset = BufferBase;
-  }
-
  uint8_t* BufferBase;
  uint8_t* CurrentOffset;
  uint64_t Size;
@@ -341,93 +341,89 @@ public:
 };

 template<uint32_t op0, uint32_t op1, uint32_t CRn, uint32_t CRm, uint32_t op2>
-constexpr uint32_t GenSystemReg() {
-  return op0 << 19 | op1 << 16 | CRn << 12 | CRm << 8 | op2 << 5;
-};
+inline constexpr uint32_t GenSystemReg = op0 << 19 | op1 << 16 | CRn << 12 | CRm << 8 | op2 << 5;

 // This `SystemRegister` enum is used for the mrs/msr instructions.
 enum class SystemRegister : uint32_t {
-  CTR_EL0 = GenSystemReg<0b11, 0b011, 0b0000, 0b0000, 0b001>(),
-  DCZID_EL0 = GenSystemReg<0b11, 0b011, 0b0000, 0b0000, 0b111>(),
-  TPIDR_EL0 = GenSystemReg<0b11, 0b011, 0b1101, 0b0000, 0b010>(),
-  RNDR = GenSystemReg<0b11, 0b011, 0b0010, 0b0100, 0b000>(),
-  RNDRRS = GenSystemReg<0b11, 0b011, 0b0010, 0b0100, 0b001>(),
-  NZCV = GenSystemReg<0b11, 0b011, 0b0100, 0b0010, 0b000>(),
-  FPCR = GenSystemReg<0b11, 0b011, 0b0100, 0b0100, 0b000>(),
-  CNTFRQ_EL0 = GenSystemReg<0b11, 0b011, 0b1110, 0b0000, 0b000>(),
-  CNTVCT_EL0 = GenSystemReg<0b11, 0b011, 0b1110, 0b0000, 0b010>(),
+  CTR_EL0 = GenSystemReg<0b11, 0b011, 0b0000, 0b0000, 0b001>,
+  DCZID_EL0 = GenSystemReg<0b11, 0b011, 0b0000, 0b0000, 0b111>,
+  TPIDR_EL0 = GenSystemReg<0b11, 0b011, 0b1101, 0b0000, 0b010>,
+  RNDR = GenSystemReg<0b11, 0b011, 0b0010, 0b0100, 0b000>,
+  RNDRRS = GenSystemReg<0b11, 0b011, 0b0010, 0b0100, 0b001>,
+  NZCV = GenSystemReg<0b11, 0b011, 0b0100, 0b0010, 0b000>,
+  FPCR = GenSystemReg<0b11, 0b011, 0b0100, 0b0100, 0b000>,
+  TPIDRRO_EL0 = GenSystemReg<0b11, 0b011, 0b1101, 0b0000, 0b011>,
+  CNTFRQ_EL0 = GenSystemReg<0b11, 0b011, 0b1110, 0b0000, 0b000>,
+  CNTVCT_EL0 = GenSystemReg<0b11, 0b011, 0b1110, 0b0000, 0b010>,
+  CNTVCTSS_EL0 = GenSystemReg<0b11, 0b011, 0b1110, 0b0000, 0b110>,
 };

 template<uint32_t op1, uint32_t CRm, uint32_t op2>
-constexpr uint32_t GenDCReg() {
-  return op1 << 16 | CRm << 8 | op2 << 5;
-};
+inline constexpr uint32_t GenDCReg = op1 << 16 | CRm << 8 | op2 << 5;

 // This `DataCacheOperation` enum is used for the dc instruction.
 enum class DataCacheOperation : uint32_t {
-  IVAC = GenDCReg<0b000, 0b0110, 0b001>(),
-  ISW = GenDCReg<0b000, 0b0110, 0b010>(),
-  CSW = GenDCReg<0b000, 0b1010, 0b010>(),
-  CISW = GenDCReg<0b000, 0b1110, 0b010>(),
-  ZVA = GenDCReg<0b011, 0b0100, 0b001>(),
-  CVAC = GenDCReg<0b011, 0b1010, 0b001>(),
-  CVAU = GenDCReg<0b011, 0b1011, 0b001>(),
-  CIVAC = GenDCReg<0b011, 0b1110, 0b001>(),
+  IVAC = GenDCReg<0b000, 0b0110, 0b001>,
+  ISW = GenDCReg<0b000, 0b0110, 0b010>,
+  CSW = GenDCReg<0b000, 0b1010, 0b010>,
+  CISW = GenDCReg<0b000, 0b1110, 0b010>,
+  ZVA = GenDCReg<0b011, 0b0100, 0b001>,
+  CVAC = GenDCReg<0b011, 0b1010, 0b001>,
+  CVAU = GenDCReg<0b011, 0b1011, 0b001>,
+  CIVAC = GenDCReg<0b011, 0b1110, 0b001>,

  // MTE2
-  IGVAC = GenDCReg<0b000, 0b0110, 0b011>(),
-  IGSW = GenDCReg<0b000, 0b0110, 0b100>(),
-  IGDVAC = GenDCReg<0b000, 0b0110, 0b101>(),
-  IGDSW = GenDCReg<0b000, 0b0110, 0b110>(),
-  CGSW = GenDCReg<0b000, 0b1010, 0b100>(),
-  CGDSW = GenDCReg<0b000, 0b1010, 0b110>(),
-  CIGSW = GenDCReg<0b000, 0b1110, 0b100>(),
-  CIGDSW = GenDCReg<0b000, 0b1110, 0b110>(),
+  IGVAC = GenDCReg<0b000, 0b0110, 0b011>,
+  IGSW = GenDCReg<0b000, 0b0110, 0b100>,
+  IGDVAC = GenDCReg<0b000, 0b0110, 0b101>,
+  IGDSW = GenDCReg<0b000, 0b0110, 0b110>,
+  CGSW = GenDCReg<0b000, 0b1010, 0b100>,
+  CGDSW = GenDCReg<0b000, 0b1010, 0b110>,
+  CIGSW = GenDCReg<0b000, 0b1110, 0b100>,
+  CIGDSW = GenDCReg<0b000, 0b1110, 0b110>,

  // MTE
-  GVA = GenDCReg<0b011, 0b0100, 0b011>(),
-  GZVA = GenDCReg<0b011, 0b0100, 0b100>(),
-  CGVAC = GenDCReg<0b011, 0b1010, 0b011>(),
-  CGDVAC = GenDCReg<0b011, 0b1010, 0b101>(),
-  CGVAP = GenDCReg<0b011, 0b1100, 0b011>(),
-  CGDVAP = GenDCReg<0b011, 0b1100, 0b101>(),
-  CGVADP = GenDCReg<0b011, 0b1101, 0b011>(),
-  CGDVADP = GenDCReg<0b011, 0b1101, 0b101>(),
-  CIGVAC = GenDCReg<0b011, 0b1110, 0b011>(),
-  CIGDVAC = GenDCReg<0b011, 0b1110, 0b101>(),
+  GVA = GenDCReg<0b011, 0b0100, 0b011>,
+  GZVA = GenDCReg<0b011, 0b0100, 0b100>,
+  CGVAC = GenDCReg<0b011, 0b1010, 0b011>,
+  CGDVAC = GenDCReg<0b011, 0b1010, 0b101>,
+  CGVAP = GenDCReg<0b011, 0b1100, 0b011>,
+  CGDVAP = GenDCReg<0b011, 0b1100, 0b101>,
+  CGVADP = GenDCReg<0b011, 0b1101, 0b011>,
+  CGDVADP = GenDCReg<0b011, 0b1101, 0b101>,
+  CIGVAC = GenDCReg<0b011, 0b1110, 0b011>,
+  CIGDVAC = GenDCReg<0b011, 0b1110, 0b101>,

  // DPB
-  CVAP = GenDCReg<0b011, 0b1100, 0b001>(),
+  CVAP = GenDCReg<0b011, 0b1100, 0b001>,

  // DPB2
-  CVADP = GenDCReg<0b011, 0b1101, 0b001>(),
+  CVADP = GenDCReg<0b011, 0b1101, 0b001>,
 };

 template<uint32_t CRm, uint32_t op2>
-constexpr uint32_t GenHintBarrierReg() {
-  return CRm << 8 | op2 << 5;
-}
+inline constexpr uint32_t GenHintBarrierReg = CRm << 8 | op2 << 5;

 // This `HintRegister` enum is used for the hint instruction.
 enum class HintRegister : uint32_t {
-  NOP = GenHintBarrierReg<0b0000, 0b000>(),
-  YIELD = GenHintBarrierReg<0b0000, 0b001>(),
-  WFE = GenHintBarrierReg<0b0000, 0b010>(),
-  WFI = GenHintBarrierReg<0b0000, 0b011>(),
-  SEV = GenHintBarrierReg<0b0000, 0b100>(),
-  SEVL = GenHintBarrierReg<0b0000, 0b101>(),
-  DGH = GenHintBarrierReg<0b0000, 0b110>(),
-  CSDB = GenHintBarrierReg<0b0010, 0b100>(),
+  NOP = GenHintBarrierReg<0b0000, 0b000>,
+  YIELD = GenHintBarrierReg<0b0000, 0b001>,
+  WFE = GenHintBarrierReg<0b0000, 0b010>,
+  WFI = GenHintBarrierReg<0b0000, 0b011>,
+  SEV = GenHintBarrierReg<0b0000, 0b100>,
+  SEVL = GenHintBarrierReg<0b0000, 0b101>,
+  DGH = GenHintBarrierReg<0b0000, 0b110>,
+  CSDB = GenHintBarrierReg<0b0010, 0b100>,
 };

 // This `BarrierRegister` enum is used for the various barrier instructions.
 enum class BarrierRegister : uint32_t {
-  CLREX = GenHintBarrierReg<0b0000, 0b010>(),
-  TCOMMIT = GenHintBarrierReg<0b0000, 0b011>(),
-  DSB = GenHintBarrierReg<0b0000, 0b100>(),
-  DMB = GenHintBarrierReg<0b0000, 0b101>(),
-  ISB = GenHintBarrierReg<0b0000, 0b110>(),
-  SB = GenHintBarrierReg<0b0000, 0b111>(),
+  CLREX = GenHintBarrierReg<0b0000, 0b010>,
+  TCOMMIT = GenHintBarrierReg<0b0000, 0b011>,
+  DSB = GenHintBarrierReg<0b0000, 0b100>,
+  DMB = GenHintBarrierReg<0b0000, 0b101>,
+  ISB = GenHintBarrierReg<0b0000, 0b110>,
+  SB = GenHintBarrierReg<0b0000, 0b111>,
 };

 // This `BarrierScope` enum is used for the dsb/dmb instructions.
@@ -512,7 +508,7 @@ enum class SVEFMaxMinImm : uint32_t {
  _1_0,
 };

-/* This `BackwardLabel` struct used for retaining a location for PC-Relative instructions.
+/* This `BackwardLabel` struct is used for retaining a location for PC-Relative instructions.
 * This is specifically a label for a target that is logically `below` an instruction that uses it.
 * Which means that a branch would jump backwards.
 */
@@ -520,13 +516,11 @@ struct BackwardLabel {
  uint8_t* Location {};
 };

-/* This `SingleUseForwardLabel` struct used for retaining a location for PC-Relative instructions.
+/* This `ForwardLabel` struct is used for retaining a location for PC-Relative instructions.
 * This is specifically a label for a target that is logically `above` an instruction that uses it.
 * Which means that a branch would jump forwards.
- *
- * The `ForwardLabel` struct can be bound to multiple instructions, so it needs a vector for each bind instruction type.
 */
-struct SingleUseForwardLabel {
+struct ForwardLabel {
  enum class InstType {
    UNKNOWN,
    ADR,
@@ -537,12 +531,16 @@ struct SingleUseForwardLabel {
    RELATIVE_LOAD,
    LONG_ADDRESS_GEN,
  };
-  uint8_t* Location {};
-  InstType Type = InstType::UNKNOWN;
-};

-struct ForwardLabel {
-  fextl::vector<SingleUseForwardLabel> Insts {};
+  struct Reference {
+    uint8_t* Location {};
+    InstType Type = InstType::UNKNOWN;
+  };
+
+  // The first element is stored separately to avoid allocations for simple cases
+  Reference FirstInst;
+
+  fextl::vector<Reference> Insts;
 };

 /* This `BiDirectionalLabel` struct used for retaining a location for PC-Relative instructions.
@@ -554,14 +552,12 @@ struct BiDirectionalLabel {
  ForwardLabel Forward;
 };

-static inline void AddLocationToLabel(SingleUseForwardLabel* Label, SingleUseForwardLabel&& Location) {
-  LOGMAN_THROW_A_FMT(Label->Type == SingleUseForwardLabel::InstType::UNKNOWN, "Trying to bind a SingleUseForwardLabel to multiple "
-                                                                              "locations. Use ForwardLabel instead.");
-  *Label = std::move(Location);
-}
-
-static inline void AddLocationToLabel(ForwardLabel* Label, SingleUseForwardLabel&& Location) {
-  Label->Insts.emplace_back(std::move(Location));
+static inline void AddLocationToLabel(ForwardLabel* Label, ForwardLabel::Reference&& Location) {
+  if (Label->FirstInst.Location == nullptr) {
+    Label->FirstInst = Location;
+  } else {
+    Label->Insts.push_back(Location);
+  }
 }

 // Some FCMA ASIMD instructions support a rotation argument.
@@ -577,6 +573,10 @@ enum class Rotation : uint32_t {
 template<typename T>
 concept IsXOrWRegister = std::is_same_v<T, XRegister> || std::is_same_v<T, WRegister>;

+// Concept for contraining some instructions to accept only a QRegister or DRegister.
+template<typename T>
+concept IsQOrDRegister = std::is_same_v<T, QRegister> || std::is_same_v<T, DRegister>;
+
 // 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)
 //
@@ -602,6 +602,18 @@ constexpr bool AreVectorsSequential(T first, const Args&... args) {
  return (fn(first, args) && ...);
 }

+// Returns if the immediate can fit in to add/sub immediate instruction encodings.
+constexpr bool IsImmAddSub(uint64_t imm) {
+  constexpr uint64_t U12Mask = 0xFFF;
+  auto FitsWithin12Bits = [](uint64_t imm) {
+    return (imm & ~U12Mask) == 0;
+  };
+  // Can fit in to the instruction encoding:
+  // - if only bits [11:0] are set.
+  // - if only bits [23:12] are set.
+  return FitsWithin12Bits(imm) || (FitsWithin12Bits(imm >> 12) && (imm & U12Mask) == 0);
+}
+
 // This is an emitter that is designed around the smallest code bloat as possible.
 // Eschewing most developer convenience in order to keep code as small as possible.

@@ -618,15 +630,15 @@ public:
  // Bind a backward label to an address.
  // Address that is bound is the current emitter location.
  void Bind(BackwardLabel* Label) {
-    LOGMAN_THROW_AA_FMT(Label->Location == nullptr, "Trying to bind a label twice");
+    LOGMAN_THROW_A_FMT(Label->Location == nullptr, "Trying to bind a label twice");
    Label->Location = GetCursorAddress<uint8_t*>();
  }

-  void Bind(const SingleUseForwardLabel* Label) {
+  void Bind(const ForwardLabel::Reference* Label) {
    uint8_t* CurrentAddress = GetCursorAddress<uint8_t*>();
    // Patch up the instructions
    switch (Label->Type) {
-    case SingleUseForwardLabel::InstType::ADR: {
+    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");
@@ -638,7 +650,7 @@ public:
      *Instruction = Inst;
      break;
    }
-    case SingleUseForwardLabel::InstType::ADRP: {
+    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");
@@ -652,7 +664,7 @@ public:
      break;
    }

-    case SingleUseForwardLabel::InstType::B: {
+    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");
@@ -666,7 +678,7 @@ public:
      break;
    }

-    case SingleUseForwardLabel::InstType::TEST_BRANCH: {
+    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");
@@ -679,8 +691,8 @@ public:

      break;
    }
-    case SingleUseForwardLabel::InstType::BC:
-    case SingleUseForwardLabel::InstType::RELATIVE_LOAD: {
+    case ForwardLabel::InstType::BC:
+    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");
@@ -692,7 +704,7 @@ public:
      *Instruction = Inst;
      break;
    }
-    case SingleUseForwardLabel::InstType::LONG_ADDRESS_GEN: {
+    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]);
@@ -737,10 +749,9 @@ public:
  // Bind a forward label to a location.
  // This walks all the instructions in the label's vector.
  // Then backpatching all instructions that have used the label.
-  template<bool WarnAboutEmpty = false>
  void Bind(ForwardLabel* Label) {
-    if constexpr (WarnAboutEmpty) {
-      LOGMAN_THROW_A_FMT(Label->Insts.empty() == false, "Binding forward label that didn't have any instructions using it");
+    if (Label->FirstInst.Location) {
+      Bind(&Label->FirstInst);
    }
    for (auto& Inst : Label->Insts) {
      Bind(&Inst);
@@ -753,12 +764,18 @@ public:
    if (!Label->Backward.Location) {
      Bind(&Label->Backward);
    }
-    Bind<false>(&Label->Forward);
+    Bind(&Label->Forward);
  }

 #include <CodeEmitter/VixlUtils.inl>

 public:
+
+// This symbol is used to allow external tooling (IDEs, clang-format, ...) to process the included files individually:
+// If defined, the files will inject member functions into this class.
+// If not, the files will wrap the member functions in a class so that tooling will process them properly.
+#define INCLUDED_BY_EMITTER
+
  // TODO: Implement SME when it matters.
 #include <CodeEmitter/ALUOps.inl>
 #include <CodeEmitter/BranchOps.inl>
@@ -768,7 +785,9 @@ public:
 #include <CodeEmitter/ASIMDOps.inl>
 #include <CodeEmitter/SVEOps.inl>

-private:
+#undef INCLUDED_BY_EMITTER
+
+protected:
  template<typename T>
  uint32_t Encode_ra(T Reg) const {
    return Reg.Idx() << 10;
@@ -780,7 +799,6 @@ private:
  uint32_t Encode_rt2(T Reg) const {
    return Reg.Idx() << 10;
  }
-  template<>
  uint32_t Encode_rt2(uint32_t Reg) const {
    return Reg << 10;
  }
@@ -816,7 +834,6 @@ private:
  uint32_t Encode_rt(T Reg) const {
    return Reg.Idx();
  }
-  template<>
  uint32_t Encode_rt(Prefetch Reg) const {
    return FEXCore::ToUnderlying(Reg);
  }
@@ -30,9 +30,9 @@ public:
 private:
  uint32_t Index;
 };
-static_assert(sizeof(Register) == sizeof(uint32_t), "Needs to be uint32_t");
-static_assert(std::is_trivial_v<Register>, "Needs to be trivial");
-static_assert(std::is_standard_layout_v<Register>, "Needs to be standard");
+static_assert(sizeof(Register) == sizeof(uint32_t));
+static_assert(std::is_trivially_copyable_v<Register>);
+static_assert(std::is_standard_layout_v<Register>);

 /* 32-bit GPR register class.
 * This class will imply a 32-bit register size being used.
@@ -58,9 +58,9 @@ public:
 private:
  uint32_t Index;
 };
-static_assert(sizeof(Register) == sizeof(uint32_t), "Needs to be uint32_t");
-static_assert(std::is_trivial_v<Register>, "Needs to be trivial");
-static_assert(std::is_standard_layout_v<Register>, "Needs to be standard");
+static_assert(sizeof(WRegister) == sizeof(uint32_t));
+static_assert(std::is_trivially_copyable_v<WRegister>);
+static_assert(std::is_standard_layout_v<WRegister>);

 /* 64-bit GPR register class.
 * This class will imply a 64-bit register size being used.
@@ -86,9 +86,9 @@ public:
 private:
  uint32_t Index;
 };
-static_assert(sizeof(Register) == sizeof(uint32_t), "Needs to be uint32_t");
-static_assert(std::is_trivial_v<Register>, "Needs to be trivial");
-static_assert(std::is_standard_layout_v<Register>, "Needs to be standard");
+static_assert(sizeof(XRegister) == sizeof(uint32_t));
+static_assert(std::is_trivially_copyable_v<XRegister>);
+static_assert(std::is_standard_layout_v<XRegister>);

 inline constexpr WRegister Register::W() const {
  return WRegister {Index};
@@ -283,9 +283,9 @@ public:
 private:
  uint32_t Index;
 };
-static_assert(sizeof(VRegister) == sizeof(uint32_t), "Needs to be uint32_t");
-static_assert(std::is_trivial_v<VRegister>, "Needs to be trivial");
-static_assert(std::is_standard_layout_v<VRegister>, "Needs to be standard");
+static_assert(sizeof(VRegister) == sizeof(uint32_t));
+static_assert(std::is_trivially_copyable_v<VRegister>);
+static_assert(std::is_standard_layout_v<VRegister>);

 /* 8-bit ASIMD register class
 * This class implies 8-bit scalar register.
@@ -315,9 +315,9 @@ public:
 private:
  uint32_t Index;
 };
-static_assert(sizeof(BRegister) == sizeof(uint32_t), "Needs to be uint32_t");
-static_assert(std::is_trivial_v<BRegister>, "Needs to be trivial");
-static_assert(std::is_standard_layout_v<BRegister>, "Needs to be standard");
+static_assert(sizeof(BRegister) == sizeof(uint32_t));
+static_assert(std::is_trivially_copyable_v<BRegister>);
+static_assert(std::is_standard_layout_v<BRegister>);

 /* 16-bit ASIMD register class
 * This class implies 16-bit scalar register.
@@ -347,9 +347,9 @@ public:
 private:
  uint32_t Index;
 };
-static_assert(sizeof(HRegister) == sizeof(uint32_t), "Needs to be uint32_t");
-static_assert(std::is_trivial_v<HRegister>, "Needs to be trivial");
-static_assert(std::is_standard_layout_v<HRegister>, "Needs to be standard");
+static_assert(sizeof(HRegister) == sizeof(uint32_t));
+static_assert(std::is_trivially_copyable_v<HRegister>);
+static_assert(std::is_standard_layout_v<HRegister>);

 /* 32-bit ASIMD register class
 * This class implies 32-bit scalar register.
@@ -379,9 +379,9 @@ public:
 private:
  uint32_t Index;
 };
-static_assert(sizeof(SRegister) == sizeof(uint32_t), "Needs to be uint32_t");
-static_assert(std::is_trivial_v<SRegister>, "Needs to be trivial");
-static_assert(std::is_standard_layout_v<SRegister>, "Needs to be standard");
+static_assert(sizeof(SRegister) == sizeof(uint32_t));
+static_assert(std::is_trivially_copyable_v<SRegister>);
+static_assert(std::is_standard_layout_v<SRegister>);

 /* 64-bit ASIMD register class
 * This class doesn't imply Vector or Scalar.
@@ -412,9 +412,9 @@ public:
 private:
  uint32_t Index;
 };
-static_assert(sizeof(DRegister) == sizeof(uint32_t), "Needs to be uint32_t");
-static_assert(std::is_trivial_v<DRegister>, "Needs to be trivial");
-static_assert(std::is_standard_layout_v<DRegister>, "Needs to be standard");
+static_assert(sizeof(DRegister) == sizeof(uint32_t));
+static_assert(std::is_trivially_copyable_v<DRegister>);
+static_assert(std::is_standard_layout_v<DRegister>);

 /* 128-bit ASIMD register class
 * This class doesn't imply Vector or Scalar.
@@ -445,9 +445,9 @@ public:
 private:
  uint32_t Index;
 };
-static_assert(sizeof(QRegister) == sizeof(uint32_t), "Needs to be uint32_t");
-static_assert(std::is_trivial_v<QRegister>, "Needs to be trivial");
-static_assert(std::is_standard_layout_v<QRegister>, "Needs to be standard");
+static_assert(sizeof(QRegister) == sizeof(uint32_t));
+static_assert(std::is_trivially_copyable_v<QRegister>);
+static_assert(std::is_standard_layout_v<QRegister>);

 /* Unsized SVE register class.
 * This class explicitly implies the instruction will operate using SVE.
@@ -474,9 +474,9 @@ public:
 private:
  uint32_t Index;
 };
-static_assert(sizeof(ZRegister) == sizeof(uint32_t), "Needs to be uint32_t");
-static_assert(std::is_trivial_v<ZRegister>, "Needs to be trivial");
-static_assert(std::is_standard_layout_v<ZRegister>, "Needs to be standard");
+static_assert(sizeof(ZRegister) == sizeof(uint32_t));
+static_assert(std::is_trivially_copyable_v<ZRegister>);
+static_assert(std::is_standard_layout_v<ZRegister>);

 // VRegister
 inline constexpr BRegister VRegister::B() const {
@@ -919,9 +919,9 @@ public:
 private:
  uint32_t Index;
 };
-static_assert(sizeof(PRegister) == sizeof(uint32_t), "Needs to be uint32_t");
-static_assert(std::is_trivial_v<PRegister>, "Needs to be trivial");
-static_assert(std::is_standard_layout_v<PRegister>, "Needs to be standard");
+static_assert(sizeof(PRegister) == sizeof(uint32_t));
+static_assert(std::is_trivially_copyable_v<PRegister>);
+static_assert(std::is_standard_layout_v<PRegister>);

 // Unsized predicate register for SVE with zeroing semantics.
 class PRegisterZero {
@@ -947,9 +947,9 @@ public:
 private:
  uint32_t Index;
 };
-static_assert(sizeof(PRegisterZero) == sizeof(uint32_t), "Needs to be uint32_t");
-static_assert(std::is_trivial_v<PRegisterZero>, "Needs to be trivial");
-static_assert(std::is_standard_layout_v<PRegisterZero>, "Needs to be standard");
+static_assert(sizeof(PRegisterZero) == sizeof(uint32_t));
+static_assert(std::is_trivially_copyable_v<PRegisterZero>);
+static_assert(std::is_standard_layout_v<PRegisterZero>);

 // Unsized predicate register for SVE with merging semantics.
 class PRegisterMerge {
@@ -975,9 +975,9 @@ public:
 private:
  uint32_t Index;
 };
-static_assert(sizeof(PRegisterZero) == sizeof(uint32_t), "Needs to be uint32_t");
-static_assert(std::is_trivial_v<PRegisterZero>, "Needs to be trivial");
-static_assert(std::is_standard_layout_v<PRegisterZero>, "Needs to be standard");
+static_assert(sizeof(PRegisterMerge) == sizeof(uint32_t));
+static_assert(std::is_trivially_copyable_v<PRegisterMerge>);
+static_assert(std::is_standard_layout_v<PRegisterMerge>);

 // PRegister
 inline constexpr PRegisterZero PRegister::Zeroing() const {
@@ -16,17 +16,25 @@
 * Exceptions to this rule will have asserts in the emitter implementation when misused.
 *
 */
+
+#pragma once
+#ifndef INCLUDED_BY_EMITTER
+#include <CodeEmitter/Emitter.h>
+namespace ARMEmitter {
+struct EmitterOps : Emitter {
+#endif
+
 public:
-// Advanced SIMD scalar copy
+  // Advanced SIMD scalar copy
  void dup(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Index) {
    constexpr uint32_t Op = 0b0101'1110'0000'0000'0000'01 << 10;

    const uint32_t SizeImm = FEXCore::ToUnderlying(size);
    const uint32_t IndexShift = SizeImm + 1;
    const uint32_t ElementSize = 1U << SizeImm;
-    const uint32_t MaxIndex = 128U / (ElementSize * 8);
+    [[maybe_unused]] const uint32_t MaxIndex = 128U / (ElementSize * 8);

-    LOGMAN_THROW_AA_FMT(Index < MaxIndex, "Index too large. Index={}, Max Index: {}", Index, MaxIndex);
+    LOGMAN_THROW_A_FMT(Index < MaxIndex, "Index too large. Index={}, Max Index: {}", Index, MaxIndex);

    const uint32_t imm5 = (Index << IndexShift) | ElementSize;

@@ -37,7 +45,7 @@ public:
    dup(size, rd, rn, Index);
  }

-// Advanced SIMD scalar three same FP16
+  // Advanced SIMD scalar three same FP16
  void fmulx(HRegister rd, HRegister rn, HRegister rm) {
    ASIMDScalarThreeSameFP16(0, 0, 0b011, rm, rn, rd);
  }
@@ -66,7 +74,7 @@ public:
    ASIMDScalarThreeSameFP16(1, 1, 0b101, rm, rn, rd);
  }

-// Advanced SIMD scalar two-register miscellaneous FP16
+  // Advanced SIMD scalar two-register miscellaneous FP16
  void fcvtns(HRegister rd, HRegister rn) {
    ASIMDScalarTwoRegMiscFP16(0, 0, 0b11010, rn, rd);
  }
@@ -128,9 +136,17 @@ public:
    ASIMDScalarTwoRegMiscFP16(1, 1, 0b11101, rn, rd);
  }

-// Advanced SIMD scalar three same extra
-// XXX:
-// Advanced SIMD scalar two-register miscellaneous
+  // Advanced SIMD scalar three same extra
+  void sqrdmlah(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i16Bit || size == ScalarRegSize::i32Bit, "Only supports 16/32-bit");
+    ASIMDScalarThreeSameExtra(1, size, 0b0000, rm, rn, rd);
+  }
+  void sqrdmlsh(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i16Bit || size == ScalarRegSize::i32Bit, "Only supports 16/32-bit");
+    ASIMDScalarThreeSameExtra(1, size, 0b0001, rm, rn, rd);
+  }
+
+  // Advanced SIMD scalar two-register miscellaneous
  void suqadd(ScalarRegSize size, VRegister rd, VRegister rn) {
    ASIMDScalar2RegMisc(0, 0, size, 0b00011, rd, rn);
  }
@@ -140,67 +156,55 @@ public:

  ///< Comparison against 0.0
  void cmgt(ScalarRegSize size, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMDScalar2RegMisc(0, 0, size, 0b01000, rd, rn);
  }
  ///< Comparison against 0.0
  void cmeq(ScalarRegSize size, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMDScalar2RegMisc(0, 0, size, 0b01001, rd, rn);
  }

  ///< Comparison against 0.0
  void cmlt(ScalarRegSize size, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMDScalar2RegMisc(0, 0, size, 0b01010, rd, rn);
  }
  void abs(ScalarRegSize size, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMDScalar2RegMisc(0, 0, size, 0b01011, rd, rn);
  }
  ///< size is destination size.
  void sqxtn(ScalarRegSize size, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size != ScalarRegSize::i64Bit, "64-bit destination not supported");
+    LOGMAN_THROW_A_FMT(size != ScalarRegSize::i64Bit, "64-bit destination not supported");
    ASIMDScalar2RegMisc(0, 0, size, 0b10100, rd, rn);
  }

  void fcvtns(ScalarRegSize size, VRegister rd, VRegister rn) {
    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");

-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i16Bit :
-        ScalarRegSize::i8Bit;
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i16Bit : ScalarRegSize::i8Bit;

    ASIMDScalar2RegMisc(0, 0, ConvertedSize, 0b11010, rd, rn);
  }
  void fcvtms(ScalarRegSize size, VRegister rd, VRegister rn) {
    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");

-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i16Bit :
-        ScalarRegSize::i8Bit;
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i16Bit : ScalarRegSize::i8Bit;

    ASIMDScalar2RegMisc(0, 0, ConvertedSize, 0b11011, rd, rn);
  }
  void fcvtas(ScalarRegSize size, VRegister rd, VRegister rn) {
    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");

-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i16Bit :
-        ScalarRegSize::i8Bit;
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i16Bit : ScalarRegSize::i8Bit;

    ASIMDScalar2RegMisc(0, 0, ConvertedSize, 0b11100, rd, rn);
  }
  void scvtf(ScalarRegSize size, VRegister rd, VRegister rn) {
    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");

-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i16Bit :
-        ScalarRegSize::i8Bit;
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i16Bit : ScalarRegSize::i8Bit;

    ASIMDScalar2RegMisc(0, 0, ConvertedSize, 0b11101, rd, rn);
  }
@@ -249,70 +253,58 @@ public:
  }
  ///< Comparison against 0.0
  void cmge(ScalarRegSize size, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMDScalar2RegMisc(0, 1, size, 0b01000, rd, rn);
  }
  ///< Comparison against 0.0
  void cmle(ScalarRegSize size, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMDScalar2RegMisc(0, 1, size, 0b01001, rd, rn);
  }
  void neg(ScalarRegSize size, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMDScalar2RegMisc(0, 1, size, 0b01011, rd, rn);
  }
  ///< size is destination.
  void sqxtun(ScalarRegSize size, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size != ScalarRegSize::i64Bit, "64-bit destination not supported");
+    LOGMAN_THROW_A_FMT(size != ScalarRegSize::i64Bit, "64-bit destination not supported");
    ASIMDScalar2RegMisc(0, 1, size, 0b10010, rd, rn);
  }
  ///< size is destination.
  void uqxtn(ScalarRegSize size, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size != ScalarRegSize::i64Bit, "64-bit destination not supported");
+    LOGMAN_THROW_A_FMT(size != ScalarRegSize::i64Bit, "64-bit destination not supported");
    ASIMDScalar2RegMisc(0, 1, size, 0b10100, rd, rn);
  }
  ///< size is destination.
  void fcvtxn(ScalarRegSize size, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
    ASIMDScalar2RegMisc(0, 1, ScalarRegSize::i16Bit, 0b10110, rd, rn);
  }
  void fcvtnu(ScalarRegSize size, VRegister rd, VRegister rn) {
    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");

-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i16Bit :
-        ScalarRegSize::i8Bit;
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i16Bit : ScalarRegSize::i8Bit;

    ASIMDScalar2RegMisc(0, 1, ConvertedSize, 0b11010, rd, rn);
  }
  void fcvtmu(ScalarRegSize size, VRegister rd, VRegister rn) {
    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");

-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i16Bit :
-        ScalarRegSize::i8Bit;
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i16Bit : ScalarRegSize::i8Bit;

    ASIMDScalar2RegMisc(0, 1, ConvertedSize, 0b11011, rd, rn);
  }
  void fcvtau(ScalarRegSize size, VRegister rd, VRegister rn) {
    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");

-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i16Bit :
-        ScalarRegSize::i8Bit;
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i16Bit : ScalarRegSize::i8Bit;

    ASIMDScalar2RegMisc(0, 1, ConvertedSize, 0b11100, rd, rn);
  }
  void ucvtf(ScalarRegSize size, VRegister rd, VRegister rn) {
    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");

-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i16Bit :
-        ScalarRegSize::i8Bit;
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i16Bit : ScalarRegSize::i8Bit;

    ASIMDScalar2RegMisc(0, 1, ConvertedSize, 0b11101, rd, rn);
  }
@@ -366,73 +358,55 @@ public:
  }

  void fmaxnmp(ScalarRegSize size, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");

-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i16Bit :
-        ScalarRegSize::i8Bit;
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i16Bit : ScalarRegSize::i8Bit;

    ASIMDScalar2RegMisc(1, 1, ConvertedSize, 0b01100, rd, rn);
  }
  void faddp(ScalarRegSize size, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");

-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i16Bit :
-        ScalarRegSize::i8Bit;
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i16Bit : ScalarRegSize::i8Bit;

    ASIMDScalar2RegMisc(1, 1, ConvertedSize, 0b01101, rd, rn);
  }
  void fmaxp(ScalarRegSize size, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");

-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i16Bit :
-        ScalarRegSize::i8Bit;
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i16Bit : ScalarRegSize::i8Bit;

    ASIMDScalar2RegMisc(1, 1, ConvertedSize, 0b01111, rd, rn);
  }
  void fminnmp(ScalarRegSize size, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
    ASIMDScalar2RegMisc(1, 1, size, 0b01100, rd, rn);
  }
  void fminp(ScalarRegSize size, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
    ASIMDScalar2RegMisc(1, 1, size, 0b01111, rd, rn);
  }
-// Advanced SIMD scalar three different
+  // Advanced SIMD scalar three different
  ///< size is destination.
  void sqdmlal(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i32Bit :
-        ScalarRegSize::i16Bit;
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i32Bit : ScalarRegSize::i16Bit;
    ASIMD3RegDifferent(0, ConvertedSize, 0b1001, rd, rn, rm);
  }
  ///< size is destination.
  void sqdmlsl(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i32Bit :
-        ScalarRegSize::i16Bit;
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i32Bit : ScalarRegSize::i16Bit;
    ASIMD3RegDifferent(0, ConvertedSize, 0b1011, rd, rn, rm);
  }

  ///< size is destination.
  void sqdmull(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i32Bit :
-        ScalarRegSize::i16Bit;
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i32Bit : ScalarRegSize::i16Bit;
    ASIMD3RegDifferent(0, ConvertedSize, 0b1101, rd, rn, rm);
  }
-// Advanced SIMD scalar three same
+  // Advanced SIMD scalar three same
  void sqadd(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
    ASIMD3RegSame(0, size, 0b00001, rd, rn, rm);
  }
@@ -440,71 +414,62 @@ public:
    ASIMD3RegSame(0, size, 0b00101, rd, rn, rm);
  }
  void cmgt(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMD3RegSame(0, size, 0b00110, rd, rn, rm);
  }
  void cmge(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMD3RegSame(0, size, 0b00111, rd, rn, rm);
  }
  void sshl(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMD3RegSame(0, size, 0b01000, rd, rn, rm);
  }
  void sqshl(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
    ASIMD3RegSame(0, size, 0b01001, rd, rn, rm);
  }
  void srshl(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMD3RegSame(0, size, 0b01010, rd, rn, rm);
  }
  void sqrshl(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
    ASIMD3RegSame(0, size, 0b01011, rd, rn, rm);
  }
  void add(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMD3RegSame(0, size, 0b10000, rd, rn, rm);
  }
  void cmtst(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMD3RegSame(0, size, 0b10001, rd, rn, rm);
  }
  void sqdmulh(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i32Bit || size == ScalarRegSize::i16Bit, "Invalid size");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i32Bit || size == ScalarRegSize::i16Bit, "Invalid size");
    ASIMD3RegSame(0, size, 0b10110, rd, rn, rm);
  }
  void fmulx(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");

-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i16Bit :
-        ScalarRegSize::i8Bit;
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i16Bit : ScalarRegSize::i8Bit;

    ASIMD3RegSame(0, ConvertedSize, 0b11011, rd, rn, rm);
  }
  void fcmeq(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");

-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i16Bit :
-        ScalarRegSize::i8Bit;
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i16Bit : ScalarRegSize::i8Bit;

    ASIMD3RegSame(0, ConvertedSize, 0b11100, rd, rn, rm);
  }
  void frecps(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");

-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i16Bit :
-        ScalarRegSize::i8Bit;
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i16Bit : ScalarRegSize::i8Bit;

    ASIMD3RegSame(0, ConvertedSize, 0b11111, rd, rn, rm);
  }
  void frsqrts(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
    ASIMD3RegSame(0, size, 0b11111, rd, rn, rm);
  }
  void uqadd(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
@@ -514,75 +479,69 @@ public:
    ASIMD3RegSame(1, size, 0b00101, rd, rn, rm);
  }
  void cmhi(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMD3RegSame(1, size, 0b00110, rd, rn, rm);
  }
  void cmhs(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMD3RegSame(1, size, 0b00111, rd, rn, rm);
  }
  void ushl(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMD3RegSame(1, size, 0b01000, rd, rn, rm);
  }
  void uqshl(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
    ASIMD3RegSame(1, size, 0b01001, rd, rn, rm);
  }
  void urshl(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMD3RegSame(1, size, 0b01010, rd, rn, rm);
  }
  void uqrshl(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
    ASIMD3RegSame(1, size, 0b01011, rd, rn, rm);
  }
  void sub(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMD3RegSame(1, size, 0b10000, rd, rn, rm);
  }
  void cmeq(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit, "Only supports 64-bit");
    ASIMD3RegSame(1, size, 0b10001, rd, rn, rm);
  }
  void sqrdmulh(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i32Bit || size == ScalarRegSize::i16Bit, "Invalid size");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i32Bit || size == ScalarRegSize::i16Bit, "Invalid size");
    ASIMD3RegSame(1, size, 0b10110, rd, rn, rm);
  }
  void fcmge(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");

-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i16Bit :
-        ScalarRegSize::i8Bit;
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i16Bit : ScalarRegSize::i8Bit;

    ASIMD3RegSame(1, ConvertedSize, 0b11100, rd, rn, rm);
  }
  void facge(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");

-    const ScalarRegSize ConvertedSize =
-      size == ScalarRegSize::i64Bit ?
-        ScalarRegSize::i16Bit :
-        ScalarRegSize::i8Bit;
+    const ScalarRegSize ConvertedSize = size == ScalarRegSize::i64Bit ? ScalarRegSize::i16Bit : ScalarRegSize::i8Bit;

    ASIMD3RegSame(1, ConvertedSize, 0b11101, rd, rn, rm);
  }
  void fabd(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
    ASIMD3RegSame(1, size, 0b11010, rd, rn, rm);
  }
  void fcmgt(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
    ASIMD3RegSame(1, size, 0b11100, rd, rn, rm);
  }
  void facgt(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for float convert");
    ASIMD3RegSame(1, size, 0b11101, rd, rn, rm);
  }
-// Advanced SIMD scalar shift by immediate
+  // Advanced SIMD scalar shift by immediate
  void sshr(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
-    LOGMAN_THROW_AA_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
+    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
+    LOGMAN_THROW_A_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
    const size_t SubregSizeInBits = ScalarRegSizeInBits(size);
    // Shift encoded in immh:immb, but inverted with 128-bit source
    // shift = (esize * 2) - immh:immb
@@ -592,8 +551,8 @@ public:
    ASIMDScalarShiftByImm(0, immh, immb, 0b00000, rd, rn);
  }
  void ssra(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
-    LOGMAN_THROW_AA_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
+    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
+    LOGMAN_THROW_A_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
    const size_t SubregSizeInBits = ScalarRegSizeInBits(size);
    // Shift encoded in immh:immb, but inverted with 128-bit source
    // shift = (esize * 2) - immh:immb
@@ -603,8 +562,8 @@ public:
    ASIMDScalarShiftByImm(0, immh, immb, 0b00010, rd, rn);
  }
  void srshr(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
-    LOGMAN_THROW_AA_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
+    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
+    LOGMAN_THROW_A_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
    const size_t SubregSizeInBits = ScalarRegSizeInBits(size);
    // Shift encoded in immh:immb, but inverted with 128-bit source
    // shift = (esize * 2) - immh:immb
@@ -614,8 +573,8 @@ public:
    ASIMDScalarShiftByImm(0, immh, immb, 0b00100, rd, rn);
  }
  void srsra(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
-    LOGMAN_THROW_AA_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
+    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
+    LOGMAN_THROW_A_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
    const size_t SubregSizeInBits = ScalarRegSizeInBits(size);
    // Shift encoded in immh:immb, but inverted with 128-bit source
    // shift = (esize * 2) - immh:immb
@@ -625,8 +584,8 @@ public:
    ASIMDScalarShiftByImm(0, immh, immb, 0b00110, rd, rn);
  }
  void shl(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
-    LOGMAN_THROW_AA_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
+    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
+    LOGMAN_THROW_A_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
    // Shift encoded a bit weirdly.
    // shift = immh:immb - elementsize but immh is /also/ used for element size.
    const uint32_t immh = 1 << FEXCore::ToUnderlying(size) | (Shift >> 3);
@@ -644,7 +603,7 @@ public:
  ///< size is destination
  void sqshrn(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < ScalarRegSizeInBits(size), "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size != ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sqshrn");
+    LOGMAN_THROW_A_FMT(size != ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sqshrn");
    const size_t SubregSizeInBits = ScalarRegSizeInBits(size);
    // Shift encoded in immh:immb, but inverted with 128-bit source
    // shift = (esize * 2) - immh:immb
@@ -655,7 +614,7 @@ public:
  }
  void sqrshrn(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < ScalarRegSizeInBits(size), "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size != ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sqshrn");
+    LOGMAN_THROW_A_FMT(size != ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sqshrn");
    const size_t SubregSizeInBits = ScalarRegSizeInBits(size);
    // Shift encoded in immh:immb, but inverted with 128-bit source
    // shift = (esize * 2) - immh:immb
@@ -666,8 +625,8 @@ public:
  }
  // TODO: SCVTF, FCVTZS
  void ushr(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
-    LOGMAN_THROW_AA_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
+    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
+    LOGMAN_THROW_A_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
    const size_t SubregSizeInBits = ScalarRegSizeInBits(size);
    // Shift encoded in immh:immb, but inverted with 128-bit source
    // shift = (esize * 2) - immh:immb
@@ -677,8 +636,8 @@ public:
    ASIMDScalarShiftByImm(1, immh, immb, 0b00000, rd, rn);
  }
  void usra(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
-    LOGMAN_THROW_AA_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
+    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
+    LOGMAN_THROW_A_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
    const size_t SubregSizeInBits = ScalarRegSizeInBits(size);
    // Shift encoded in immh:immb, but inverted with 128-bit source
    // shift = (esize * 2) - immh:immb
@@ -688,8 +647,8 @@ public:
    ASIMDScalarShiftByImm(1, immh, immb, 0b00010, rd, rn);
  }
  void urshr(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
-    LOGMAN_THROW_AA_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
+    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
+    LOGMAN_THROW_A_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
    const size_t SubregSizeInBits = ScalarRegSizeInBits(size);
    // Shift encoded in immh:immb, but inverted with 128-bit source
    // shift = (esize * 2) - immh:immb
@@ -699,8 +658,8 @@ public:
    ASIMDScalarShiftByImm(1, immh, immb, 0b00100, rd, rn);
  }
  void ursra(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
-    LOGMAN_THROW_AA_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
+    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
+    LOGMAN_THROW_A_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
    const size_t SubregSizeInBits = ScalarRegSizeInBits(size);
    // Shift encoded in immh:immb, but inverted with 128-bit source
    // shift = (esize * 2) - immh:immb
@@ -710,8 +669,8 @@ public:
    ASIMDScalarShiftByImm(1, immh, immb, 0b00110, rd, rn);
  }
  void sri(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
-    LOGMAN_THROW_AA_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
+    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
+    LOGMAN_THROW_A_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
    const size_t SubregSizeInBits = ScalarRegSizeInBits(size);
    // Shift encoded in immh:immb, but inverted with 128-bit source
    // shift = (esize * 2) - immh:immb
@@ -721,8 +680,8 @@ public:
    ASIMDScalarShiftByImm(1, immh, immb, 0b01000, rd, rn);
  }
  void sli(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
-    LOGMAN_THROW_AA_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
+    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < 64, "Invalid shift for sshr");
+    LOGMAN_THROW_A_FMT(size == ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sshr");
    // Shift encoded a bit weirdly.
    // shift = immh:immb - elementsize but immh is /also/ used for element size.
    const uint32_t immh = 1 << FEXCore::ToUnderlying(size) | (Shift >> 3);
@@ -748,7 +707,7 @@ public:
  ///< size is destination.
  void sqshrun(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < ScalarRegSizeInBits(size), "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size != ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sqshrun");
+    LOGMAN_THROW_A_FMT(size != ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sqshrun");
    const size_t SubregSizeInBits = ScalarRegSizeInBits(size);
    // Shift encoded in immh:immb, but inverted with 128-bit source
    // shift = (esize * 2) - immh:immb
@@ -760,7 +719,7 @@ public:
  ///< size is destination.
  void sqrshrun(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < ScalarRegSizeInBits(size), "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size != ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sqrshrun");
+    LOGMAN_THROW_A_FMT(size != ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sqrshrun");
    const size_t SubregSizeInBits = ScalarRegSizeInBits(size);
    // Shift encoded in immh:immb, but inverted with 128-bit source
    // shift = (esize * 2) - immh:immb
@@ -772,7 +731,7 @@ public:
  ///< size is destination.
  void uqshrn(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < ScalarRegSizeInBits(size), "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size != ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sqrshrun");
+    LOGMAN_THROW_A_FMT(size != ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sqrshrun");
    const size_t SubregSizeInBits = ScalarRegSizeInBits(size);
    // Shift encoded in immh:immb, but inverted with 128-bit source
    // shift = (esize * 2) - immh:immb
@@ -784,7 +743,7 @@ public:
  ///< size is destination.
  void uqrshrn(ScalarRegSize size, VRegister rd, VRegister rn, uint32_t Shift) {
    LOGMAN_THROW_A_FMT(Shift > 0 && Shift < ScalarRegSizeInBits(size), "Invalid shift for sshr");
-    LOGMAN_THROW_AA_FMT(size != ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sqrshrun");
+    LOGMAN_THROW_A_FMT(size != ARMEmitter::ScalarRegSize::i64Bit, "Invalid size selected for sqrshrun");
    const size_t SubregSizeInBits = ScalarRegSizeInBits(size);
    // Shift encoded in immh:immb, but inverted with 128-bit source
    // shift = (esize * 2) - immh:immb
@@ -793,11 +752,52 @@ public:
    const uint32_t immb = InvertedShift & 0b111;
    ASIMDScalarShiftByImm(1, immh, immb, 0b10011, rd, rn);
  }
+
  // TODO: UCVTF, FCVTZU
-// Advanced SIMD scalar x indexed element
-// XXX:
-//
-// Floating-point data-processing (1 source)
+
+  // Advanced SIMD scalar x indexed element
+  void sqdmlal(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm, uint32_t index) {
+    LOGMAN_THROW_A_FMT(size != ScalarRegSize::i64Bit, "Scalar size must not be 64-bit");
+    ASIMDScalarXIndexedElement(0, size, 0b0011, rm, rn, rd, index);
+  }
+  void sqdmlsl(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm, uint32_t index) {
+    LOGMAN_THROW_A_FMT(size != ScalarRegSize::i64Bit, "Scalar size must not be 64-bit");
+    ASIMDScalarXIndexedElement(0, size, 0b0111, rm, rn, rd, index);
+  }
+  void sqdmull(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm, uint32_t index) {
+    LOGMAN_THROW_A_FMT(size != ScalarRegSize::i64Bit, "Scalar size must not be 64-bit");
+    ASIMDScalarXIndexedElement(0, size, 0b1011, rm, rn, rd, index);
+  }
+  void sqdmulh(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm, uint32_t index) {
+    LOGMAN_THROW_A_FMT(size != ScalarRegSize::i64Bit, "Scalar size must not be 64-bit");
+    ASIMDScalarXIndexedElement(0, size, 0b1100, rm, rn, rd, index);
+  }
+  void sqrdmulh(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm, uint32_t index) {
+    LOGMAN_THROW_A_FMT(size != ScalarRegSize::i64Bit, "Scalar size must not be 64-bit");
+    ASIMDScalarXIndexedElement(0, size, 0b1101, rm, rn, rd, index);
+  }
+  void fmla(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm, uint32_t index) {
+    ASIMDScalarXIndexedElement(0, size, 0b0001, rm, rn, rd, index);
+  }
+  void fmls(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm, uint32_t index) {
+    ASIMDScalarXIndexedElement(0, size, 0b0101, rm, rn, rd, index);
+  }
+  void fmul(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm, uint32_t index) {
+    ASIMDScalarXIndexedElement(0, size, 0b1001, rm, rn, rd, index);
+  }
+  void sqrdmlah(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm, uint32_t index) {
+    LOGMAN_THROW_A_FMT(size != ScalarRegSize::i64Bit, "Scalar size must not be 64-bit");
+    ASIMDScalarXIndexedElement(1, size, 0b1101, rm, rn, rd, index);
+  }
+  void sqrdmlsh(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm, uint32_t index) {
+    LOGMAN_THROW_A_FMT(size != ScalarRegSize::i64Bit, "Scalar size must not be 64-bit");
+    ASIMDScalarXIndexedElement(1, size, 0b1111, rm, rn, rd, index);
+  }
+  void fmulx(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm, uint32_t index) {
+    ASIMDScalarXIndexedElement(1, size, 0b1001, rm, rn, rd, index);
+  }
+
+  // Floating-point data-processing (1 source)
  void fmov(ScalarRegSize size, VRegister rd, VRegister rn) {
    Float1Source(size, 0, 0, 0b000000, rd, rn);
  }
@@ -991,14 +991,14 @@ public:
    Float1Source(0, 0, 0b11, 0b001111, rd.V(), rn.V());
  }

-// Floating-point compare
+  // Floating-point compare
  void fcmp(ScalarRegSize Size, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(Size != ScalarRegSize::i8Bit, "8-bit destination not supported");
+    LOGMAN_THROW_A_FMT(Size != ScalarRegSize::i8Bit, "8-bit destination not supported");

-    const auto ConvertedSize =
-      Size == ARMEmitter::ScalarRegSize::i64Bit ? 0b01 :
-      Size == ARMEmitter::ScalarRegSize::i32Bit ? 0b00 :
-      Size == ARMEmitter::ScalarRegSize::i16Bit ? 0b11 : 0;
+    const auto ConvertedSize = Size == ARMEmitter::ScalarRegSize::i64Bit ? 0b01 :
+                               Size == ARMEmitter::ScalarRegSize::i32Bit ? 0b00 :
+                               Size == ARMEmitter::ScalarRegSize::i16Bit ? 0b11 :
+                                                                           0;

    FloatCompare(0, 0, ConvertedSize, 0b00, 0b00000, rn, rm);
  }
@@ -1051,7 +1051,7 @@ public:
    FloatCompare(0, 0, 0b11, 0b00, 0b11000, rn.V(), VReg::v0);
  }

-// Floating-point immediate
+  // Floating-point immediate
  void fmov(ARMEmitter::ScalarRegSize size, ARMEmitter::VRegister rd, float Value) {
    uint32_t M = 0;
    uint32_t S = 0;
@@ -1061,16 +1061,13 @@ public:
    if (size == ARMEmitter::ScalarRegSize::i16Bit) {
      LOGMAN_MSG_A_FMT("Unsupported");
      FEX_UNREACHABLE;
-    }
-    else if (size == ARMEmitter::ScalarRegSize::i32Bit) {
+    } else if (size == ARMEmitter::ScalarRegSize::i32Bit) {
      ptype = 0b00;
      imm8 = FP32ToImm8(Value);
-    }
-    else if (size == ARMEmitter::ScalarRegSize::i64Bit) {
+    } else if (size == ARMEmitter::ScalarRegSize::i64Bit) {
      ptype = 0b01;
      imm8 = FP64ToImm8(Value);
-    }
-    else {
+    } else {
      FEX_UNREACHABLE;
    }

@@ -1090,7 +1087,7 @@ public:
    dc32(Instr);
  }

-// Floating-point conditional compare
+  // Floating-point conditional compare
  void fccmp(SRegister rn, SRegister rm, StatusFlags flags, Condition Cond) {
    FloatConditionalCompare(0, 0, 0b00, 0b0, rn.V(), rm.V(), flags, Cond);
  }
@@ -1110,7 +1107,7 @@ public:
    FloatConditionalCompare(0, 0, 0b11, 0b1, rn.V(), rm.V(), flags, Cond);
  }

-// Floating-point data-processing (2 source)
+  // Floating-point data-processing (2 source)
  void fmul(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm) {
    Float2Source(size, 0, 0, 0b0000, rd, rn, rm);
  }
@@ -1225,11 +1222,10 @@ public:

  // Floating-point conditional select
  void fcsel(ScalarRegSize size, VRegister rd, VRegister rn, VRegister rm, Condition Cond) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i16Bit || size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for {}", __func__);
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i16Bit || size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit,
+                       "Invalid size selected for {}", __func__);

-    const uint32_t ConvertedSize =
-      size == ScalarRegSize::i64Bit ? 0b01 :
-      size == ScalarRegSize::i32Bit ? 0b00 : 0b11;
+    const uint32_t ConvertedSize = size == ScalarRegSize::i64Bit ? 0b01 : size == ScalarRegSize::i32Bit ? 0b00 : 0b11;

    FloatConditionalSelect(0, 0, ConvertedSize, rd, rn, rm, Cond);
  }
@@ -1244,7 +1240,7 @@ public:
    FloatConditionalSelect(0, 0, 0b11, rd.V(), rn.V(), rm.V(), Cond);
  }

-// Floating-point data-processing (3 source)
+  // Floating-point data-processing (3 source)
  void fmadd(SRegister rd, SRegister rn, SRegister rm, SRegister ra) {
    Float3Source(0, 0, 0b00, 0, 0, rd.V(), rn.V(), rm.V(), ra.V());
  }
@@ -1285,7 +1281,7 @@ public:
  }

 private:
-// Advanced SIMD scalar copy
+  // Advanced SIMD scalar copy
  void ASIMDScalarCopy(uint32_t Op, uint32_t Q, uint32_t imm5, uint32_t imm4, ARMEmitter::VRegister rd, ARMEmitter::VRegister rn) {
    uint32_t Instr = Op;

@@ -1297,7 +1293,7 @@ private:
    dc32(Instr);
  }

-// Advanced SIMD scalar three same FP16
+  // Advanced SIMD scalar three same FP16
  void ASIMDScalarThreeSameFP16(uint32_t U, uint32_t a, uint32_t opcode, HRegister rm, HRegister rn, HRegister rd) {
    uint32_t Instr = 0b0101'1110'0100'0000'0000'0100'0000'0000;

@@ -1309,7 +1305,7 @@ private:
    Instr |= rd.Idx();
    dc32(Instr);
  }
-// Advanced SIMD scalar two-register miscellaneous FP16
+  // Advanced SIMD scalar two-register miscellaneous FP16
  void ASIMDScalarTwoRegMiscFP16(uint32_t U, uint32_t a, uint32_t opcode, HRegister rn, HRegister rd) {
    uint32_t Instr = 0b0101'1110'0111'1000'0000'1000'0000'0000;

@@ -1321,9 +1317,19 @@ private:
    dc32(Instr);
  }

-// Advanced SIMD scalar three same extra
-// XXX:
-// Advanced SIMD scalar two-register miscellaneous
+  // Advanced SIMD scalar three same extra
+  void ASIMDScalarThreeSameExtra(uint32_t U, ScalarRegSize size, uint32_t opcode, VRegister rm, VRegister rn, VRegister rd) {
+    uint32_t Instr = 0b0101'1110'0000'0000'1000'0100'0000'0000;
+    Instr |= U << 29;
+    Instr |= FEXCore::ToUnderlying(size) << 22;
+    Instr |= rm.Idx() << 16;
+    Instr |= opcode << 11;
+    Instr |= rn.Idx() << 5;
+    Instr |= rd.Idx();
+    dc32(Instr);
+  }
+
+  // Advanced SIMD scalar two-register miscellaneous
  void ASIMDScalar2RegMisc(uint32_t b20, uint32_t U, ScalarRegSize size, uint32_t opcode, VRegister rd, VRegister rn) {
    uint32_t Instr = 0b0101'1110'0010'0000'0000'1000'0000'0000;

@@ -1336,9 +1342,7 @@ private:
    dc32(Instr);
  }

-// Advanced SIMD scalar pairwise
-// XXX:
-// Advanced SIMD scalar three different
+  // Advanced SIMD scalar three different
  void ASIMD3RegDifferent(uint32_t U, ScalarRegSize size, uint32_t opcode, VRegister rd, VRegister rn, VRegister rm) {
    uint32_t Instr = 0b0101'1110'0010'0000'0000'0000'0000'0000;

@@ -1350,7 +1354,7 @@ private:
    Instr |= Encode_rd(rd);
    dc32(Instr);
  }
-// Advanced SIMD scalar three same
+  // Advanced SIMD scalar three same
  void ASIMD3RegSame(uint32_t U, ScalarRegSize size, uint32_t opcode, VRegister rd, VRegister rn, VRegister rm) {
    uint32_t Instr = 0b0101'1110'0010'0000'0000'0100'0000'0000;

@@ -1362,7 +1366,7 @@ private:
    Instr |= Encode_rd(rd);
    dc32(Instr);
  }
-// Advanced SIMD scalar shift by immediate
+  // Advanced SIMD scalar shift by immediate
  void ASIMDScalarShiftByImm(uint32_t U, uint32_t immh, uint32_t immb, uint32_t opcode, VRegister rd, VRegister rn) {
    uint32_t Instr = 0b0101'1111'0000'0000'0000'0100'0000'0000;

@@ -1374,9 +1378,51 @@ private:
    Instr |= Encode_rd(rd);
    dc32(Instr);
  }
-// Advanced SIMD scalar x indexed element
-// XXX:
-// Floating-point data-processing (1 source)
+
+  // Advanced SIMD scalar x indexed element
+  void ASIMDScalarXIndexedElement(uint32_t U, ScalarRegSize size, uint32_t opcode, VRegister rm, VRegister rn, VRegister rd, uint32_t index) {
+    LOGMAN_THROW_A_FMT(size != ScalarRegSize::i8Bit, "Scalar size must not be 8-bit");
+
+    [[maybe_unused]] const auto invalid_bound = 16U >> FEXCore::ToUnderlying(size);
+    LOGMAN_THROW_A_FMT(index < invalid_bound, "Index ({}) must be within [0-{}]", index, invalid_bound - 1);
+
+    uint32_t Instr = 0b0101'1111'0000'0000'0000'0000'0000'0000;
+
+    // FMUL/FMLA/FMLS indexed variants deal with size differently.
+    if (opcode == 0b0001 || opcode == 0b0101 || opcode == 0b1001) {
+      // Unlike other instructions in the group, 16-bit is encoded as zero
+      // and 32/64-bit are encoded with the top bit always set to one.
+      if (size != ScalarRegSize::i16Bit) {
+        Instr |= (0b10 | (FEXCore::ToUnderlying(size) & 1)) << 22;
+      }
+    } else {
+      Instr |= FEXCore::ToUnderlying(size) << 22;
+    }
+
+    uint32_t H = 0;
+    uint32_t LM = 0;
+    if (size == ScalarRegSize::i16Bit) {
+      LOGMAN_THROW_A_FMT(rm <= VReg::v15, "rm ({}) must be within [v0-v15]", rm.Idx());
+      H = (index >> 2) & 1;
+      LM = index & 0b11;
+    } else if (size == ScalarRegSize::i32Bit) {
+      H = (index >> 1) & 1;
+      LM = (index & 0b01) << 1;
+    } else {
+      H = index & 1;
+    }
+
+    Instr |= U << 29;
+    Instr |= LM << 20;
+    Instr |= rm.Idx() << 16;
+    Instr |= opcode << 12;
+    Instr |= H << 11;
+    Instr |= rn.Idx() << 5;
+    Instr |= rd.Idx();
+    dc32(Instr);
+  }
+
+  // Floating-point data-processing (1 source)
  void Float1Source(uint32_t M, uint32_t S, uint32_t ptype, uint32_t opcode, VRegister rd, VRegister rn) {
    uint32_t Instr = 0b0001'1110'0010'0000'0100'0000'0000'0000;

@@ -1390,16 +1436,15 @@ private:
    dc32(Instr);
  }
  void Float1Source(ScalarRegSize size, uint32_t M, uint32_t S, uint32_t opcode, VRegister rd, VRegister rn) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i16Bit || size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for {}", __func__);
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i16Bit || size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit,
+                       "Invalid size selected for {}", __func__);

-    const uint32_t ConvertedSize =
-      size == ScalarRegSize::i64Bit ? 0b01 :
-      size == ScalarRegSize::i32Bit ? 0b00 : 0b11;
+    const uint32_t ConvertedSize = size == ScalarRegSize::i64Bit ? 0b01 : size == ScalarRegSize::i32Bit ? 0b00 : 0b11;

    Float1Source(M, S, ConvertedSize, opcode, rd, rn);
  }

-// Floating-point compare
+  // Floating-point compare
  void FloatCompare(uint32_t M, uint32_t S, uint32_t ftype, uint32_t op, uint32_t opcode2, VRegister rn, VRegister rm) {
    uint32_t Instr = 0b0001'1110'0010'0000'0010'0000'0000'0000;

@@ -1413,9 +1458,9 @@ private:

    dc32(Instr);
  }
-// Floating-point immediate
-// XXX:
-// Floating-point conditional compare
+  // Floating-point immediate
+  // XXX:
+  // Floating-point conditional compare
  void FloatConditionalCompare(uint32_t M, uint32_t S, uint32_t ptype, uint32_t op, VRegister rn, VRegister rm, StatusFlags flags, Condition Cond) {
    uint32_t Instr = 0b0001'1110'0010'0000'0000'0100'0000'0000;

@@ -1430,7 +1475,7 @@ private:

    dc32(Instr);
  }
-// Floating-point data-processing (2 source)
+  // Floating-point data-processing (2 source)

  void Float2Source(uint32_t M, uint32_t S, uint32_t ptype, uint32_t opcode, VRegister rd, VRegister rn, VRegister rm) {
    uint32_t Instr = 0b0001'1110'0010'0000'0000'1000'0000'0000;
@@ -1447,16 +1492,15 @@ private:
  }

  void Float2Source(ScalarRegSize size, uint32_t M, uint32_t S, uint32_t opcode, VRegister rd, VRegister rn, VRegister rm) {
-    LOGMAN_THROW_AA_FMT(size == ScalarRegSize::i16Bit || size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit, "Invalid size selected for {}", __func__);
+    LOGMAN_THROW_A_FMT(size == ScalarRegSize::i16Bit || size == ScalarRegSize::i64Bit || size == ScalarRegSize::i32Bit,
+                       "Invalid size selected for {}", __func__);

-    const uint32_t ConvertedSize =
-      size == ScalarRegSize::i64Bit ? 0b01 :
-      size == ScalarRegSize::i32Bit ? 0b00 : 0b11;
+    const uint32_t ConvertedSize = size == ScalarRegSize::i64Bit ? 0b01 : size == ScalarRegSize::i32Bit ? 0b00 : 0b11;

    Float2Source(M, S, ConvertedSize, opcode, rd, rn, rm);
  }

-// Floating-point conditional select
+  // Floating-point conditional select
  void FloatConditionalSelect(uint32_t M, uint32_t S, uint32_t ptype, VRegister rd, VRegister rn, VRegister rm, Condition Cond) {
    uint32_t Instr = 0b0001'1110'0010'0000'0000'1100'0000'0000;

@@ -1470,7 +1514,7 @@ private:
    dc32(Instr);
  }

-// Floating-point data-processing (3 source)
+  // Floating-point data-processing (3 source)
  void Float3Source(uint32_t M, uint32_t S, uint32_t ptype, uint32_t o1, uint32_t o0, VRegister rd, VRegister rn, VRegister rm, VRegister ra) {
    uint32_t Instr = 0b0001'1111'0000'0000'0000'0000'0000'0000;

@@ -1485,3 +1529,8 @@ private:
    Instr |= Encode_rd(rd);
    dc32(Instr);
  }
+
+#ifndef INCLUDED_BY_EMITTER
+}; // struct LoadstoreEmitterOps
+} // namespace ARMEmitter
+#endif
@@ -4,173 +4,191 @@
 * This is mostly a mashup of various instruction types.
 * Nothing follows an explicit pattern since they are mostly different.
 */
+
+#pragma once
+#ifndef INCLUDED_BY_EMITTER
+#include <CodeEmitter/Emitter.h>
+namespace ARMEmitter {
+struct EmitterOps : Emitter {
+#endif
+
 public:
-    // System with result
-    // TODO: SYSL
-    // System Instruction
-    // TODO: AT
-    // TODO: CFP
-    // TODO: CPP
-    void dc(ARMEmitter::DataCacheOperation DCOp, ARMEmitter::Register rt) {
-      constexpr uint32_t Op = 0b1101'0101'0000'1000'0111 << 12;
-      SystemInstruction(Op, 0, FEXCore::ToUnderlying(DCOp), rt);
-    }
-    // TODO: DVP
-    // TODO: IC
-    // TODO: TLBI
+  // Reserved
+  void udf(uint32_t Imm) {
+    LOGMAN_THROW_A_FMT(Imm < 0x1'0000, "Immediate needs to be 16-bit");
+    dc32(Imm);
+  }

-    // Exception generation
-    void svc(uint32_t Imm) {
-      ExceptionGeneration(0b000, 0b000, 0b01, Imm);
-    }
-    void hvc(uint32_t Imm) {
-      ExceptionGeneration(0b000, 0b000, 0b10, Imm);
-    }
-    void smc(uint32_t Imm) {
-      ExceptionGeneration(0b000, 0b000, 0b11, Imm);
-    }
-    void brk(uint32_t Imm) {
-      ExceptionGeneration(0b001, 0b000, 0b00, Imm);
-    }
-    void hlt(uint32_t Imm) {
-      ExceptionGeneration(0b010, 0b000, 0b00, Imm);
-    }
-    void tcancel(uint32_t Imm) {
-      ExceptionGeneration(0b011, 0b000, 0b00, Imm);
-    }
-    void dcps1(uint32_t Imm) {
-      ExceptionGeneration(0b101, 0b000, 0b01, Imm);
-    }
-    void dcps2(uint32_t Imm) {
-      ExceptionGeneration(0b101, 0b000, 0b10, Imm);
-    }
-    void dcps3(uint32_t Imm) {
-      ExceptionGeneration(0b101, 0b000, 0b11, Imm);
-    }
-    // System instructions with register argument
-    void wfet(ARMEmitter::Register rt) {
-      SystemInstructionWithReg(0b0000, 0b000, rt);
-    }
-    void wfit(ARMEmitter::Register rt) {
-      SystemInstructionWithReg(0b0000, 0b001, rt);
-    }
+  // System with result
+  // TODO: SYSL
+  // System Instruction
+  // TODO: AT
+  // TODO: CFP
+  // TODO: CPP
+  void dc(ARMEmitter::DataCacheOperation DCOp, ARMEmitter::Register rt) {
+    constexpr uint32_t Op = 0b1101'0101'0000'1000'0111 << 12;
+    SystemInstruction(Op, 0, FEXCore::ToUnderlying(DCOp), rt);
+  }
+  // TODO: DVP
+  // TODO: IC
+  // TODO: TLBI

-    // Hints
-    void nop() {
-      Hint(ARMEmitter::HintRegister::NOP);
-    }
-    void yield() {
-      Hint(ARMEmitter::HintRegister::YIELD);
-    }
-    void wfe() {
-      Hint(ARMEmitter::HintRegister::WFE);
-    }
-    void wfi() {
-      Hint(ARMEmitter::HintRegister::WFI);
-    }
-    void sev() {
-      Hint(ARMEmitter::HintRegister::SEV);
-    }
-    void sevl() {
-      Hint(ARMEmitter::HintRegister::SEVL);
-    }
-    void dgh() {
-      Hint(ARMEmitter::HintRegister::DGH);
-    }
-    void csdb() {
-      Hint(ARMEmitter::HintRegister::CSDB);
-    }
+  // Exception generation
+  void svc(uint32_t Imm) {
+    ExceptionGeneration(0b000, 0b000, 0b01, Imm);
+  }
+  void hvc(uint32_t Imm) {
+    ExceptionGeneration(0b000, 0b000, 0b10, Imm);
+  }
+  void smc(uint32_t Imm) {
+    ExceptionGeneration(0b000, 0b000, 0b11, Imm);
+  }
+  void brk(uint32_t Imm) {
+    ExceptionGeneration(0b001, 0b000, 0b00, Imm);
+  }
+  void hlt(uint32_t Imm) {
+    ExceptionGeneration(0b010, 0b000, 0b00, Imm);
+  }
+  void tcancel(uint32_t Imm) {
+    ExceptionGeneration(0b011, 0b000, 0b00, Imm);
+  }
+  void dcps1(uint32_t Imm) {
+    ExceptionGeneration(0b101, 0b000, 0b01, Imm);
+  }
+  void dcps2(uint32_t Imm) {
+    ExceptionGeneration(0b101, 0b000, 0b10, Imm);
+  }
+  void dcps3(uint32_t Imm) {
+    ExceptionGeneration(0b101, 0b000, 0b11, Imm);
+  }
+  // System instructions with register argument
+  void wfet(ARMEmitter::Register rt) {
+    SystemInstructionWithReg(0b0000, 0b000, rt);
+  }
+  void wfit(ARMEmitter::Register rt) {
+    SystemInstructionWithReg(0b0000, 0b001, rt);
+  }

-    // Barriers
-    void clrex(uint32_t imm = 15) {
-      LOGMAN_THROW_AA_FMT(imm < 16, "Immediate out of range");
-      Barrier(ARMEmitter::BarrierRegister::CLREX, imm);
-    }
-    void dsb(ARMEmitter::BarrierScope Scope) {
-      Barrier(ARMEmitter::BarrierRegister::DSB, FEXCore::ToUnderlying(Scope));
-    }
-    void dmb(ARMEmitter::BarrierScope Scope) {
-      Barrier(ARMEmitter::BarrierRegister::DMB, FEXCore::ToUnderlying(Scope));
-    }
-    void isb() {
-      Barrier(ARMEmitter::BarrierRegister::ISB, FEXCore::ToUnderlying(ARMEmitter::BarrierScope::SY));
-    }
-    void sb() {
-      Barrier(ARMEmitter::BarrierRegister::SB, 0);
-    }
-    void tcommit() {
-      Barrier(ARMEmitter::BarrierRegister::TCOMMIT, 0);
-    }
+  // Hints
+  void nop() {
+    Hint(ARMEmitter::HintRegister::NOP);
+  }
+  void yield() {
+    Hint(ARMEmitter::HintRegister::YIELD);
+  }
+  void wfe() {
+    Hint(ARMEmitter::HintRegister::WFE);
+  }
+  void wfi() {
+    Hint(ARMEmitter::HintRegister::WFI);
+  }
+  void sev() {
+    Hint(ARMEmitter::HintRegister::SEV);
+  }
+  void sevl() {
+    Hint(ARMEmitter::HintRegister::SEVL);
+  }
+  void dgh() {
+    Hint(ARMEmitter::HintRegister::DGH);
+  }
+  void csdb() {
+    Hint(ARMEmitter::HintRegister::CSDB);
+  }

-    // System register move
-    void msr(ARMEmitter::SystemRegister reg, ARMEmitter::Register rt) {
-      constexpr uint32_t Op = 0b1101'0101'0001 << 20;
-      SystemRegisterMove(Op, rt, reg);
-    }
+  // Barriers
+  void clrex(uint32_t imm = 15) {
+    LOGMAN_THROW_A_FMT(imm < 16, "Immediate out of range");
+    Barrier(ARMEmitter::BarrierRegister::CLREX, imm);
+  }
+  void dsb(ARMEmitter::BarrierScope Scope) {
+    Barrier(ARMEmitter::BarrierRegister::DSB, FEXCore::ToUnderlying(Scope));
+  }
+  void dmb(ARMEmitter::BarrierScope Scope) {
+    Barrier(ARMEmitter::BarrierRegister::DMB, FEXCore::ToUnderlying(Scope));
+  }
+  void isb() {
+    Barrier(ARMEmitter::BarrierRegister::ISB, FEXCore::ToUnderlying(ARMEmitter::BarrierScope::SY));
+  }
+  void sb() {
+    Barrier(ARMEmitter::BarrierRegister::SB, 0);
+  }
+  void tcommit() {
+    Barrier(ARMEmitter::BarrierRegister::TCOMMIT, 0);
+  }

-    void mrs(ARMEmitter::Register rd, ARMEmitter::SystemRegister reg) {
-      constexpr uint32_t Op = 0b1101'0101'0011 << 20;
-      SystemRegisterMove(Op, rd, reg);
-    }
+  // System register move
+  void msr(ARMEmitter::SystemRegister reg, ARMEmitter::Register rt) {
+    constexpr uint32_t Op = 0b1101'0101'0001 << 20;
+    SystemRegisterMove(Op, rt, reg);
+  }
+
+  void mrs(ARMEmitter::Register rd, ARMEmitter::SystemRegister reg) {
+    constexpr uint32_t Op = 0b1101'0101'0011 << 20;
+    SystemRegisterMove(Op, rd, reg);
+  }

 private:

-    // Exception Generation
-    void ExceptionGeneration(uint32_t opc, uint32_t op2, uint32_t LL, uint32_t Imm) {
-      LOGMAN_THROW_AA_FMT((Imm & 0xFFFF'0000) == 0, "Imm amount too large");
+  // Exception Generation
+  void ExceptionGeneration(uint32_t opc, uint32_t op2, uint32_t LL, uint32_t Imm) {
+    LOGMAN_THROW_A_FMT((Imm & 0xFFFF'0000) == 0, "Imm amount too large");

-      uint32_t Instr = 0b1101'0100 << 24;
+    uint32_t Instr = 0b1101'0100 << 24;

-      Instr |= opc << 21;
-      Instr |= Imm << 5;
-      Instr |= op2 << 2;
-      Instr |= LL;
+    Instr |= opc << 21;
+    Instr |= Imm << 5;
+    Instr |= op2 << 2;
+    Instr |= LL;

-      dc32(Instr);
-    }
+    dc32(Instr);
+  }

-    // System instructions with register argument
-    void SystemInstructionWithReg(uint32_t CRm, uint32_t op2, ARMEmitter::Register rt) {
-      uint32_t Instr = 0b1101'0101'0000'0011'0001 << 12;
+  // System instructions with register argument
+  void SystemInstructionWithReg(uint32_t CRm, uint32_t op2, ARMEmitter::Register rt) {
+    uint32_t Instr = 0b1101'0101'0000'0011'0001 << 12;

-      Instr |= CRm << 8;
-      Instr |= op2 << 5;
-      Instr |= Encode_rt(rt);
-      dc32(Instr);
-    }
+    Instr |= CRm << 8;
+    Instr |= op2 << 5;
+    Instr |= Encode_rt(rt);
+    dc32(Instr);
+  }

-    // Hints
-    void Hint(ARMEmitter::HintRegister Reg) {
-      uint32_t Instr = 0b1101'0101'0000'0011'0010'0000'0001'1111U;
-      Instr |= FEXCore::ToUnderlying(Reg);
-      dc32(Instr);
-    }
-    // Barriers
-    void Barrier(ARMEmitter::BarrierRegister Reg, uint32_t CRm) {
-      uint32_t Instr = 0b1101'0101'0000'0011'0011'0000'0001'1111U;
-      Instr |= CRm << 8;
-      Instr |= FEXCore::ToUnderlying(Reg);
-      dc32(Instr);
-    }
+  // Hints
+  void Hint(ARMEmitter::HintRegister Reg) {
+    uint32_t Instr = 0b1101'0101'0000'0011'0010'0000'0001'1111U;
+    Instr |= FEXCore::ToUnderlying(Reg);
+    dc32(Instr);
+  }
+  // Barriers
+  void Barrier(ARMEmitter::BarrierRegister Reg, uint32_t CRm) {
+    uint32_t Instr = 0b1101'0101'0000'0011'0011'0000'0001'1111U;
+    Instr |= CRm << 8;
+    Instr |= FEXCore::ToUnderlying(Reg);
+    dc32(Instr);
+  }

-    // System Instruction
-    void SystemInstruction(uint32_t Op, uint32_t L, uint32_t SubOp, ARMEmitter::Register rt) {
-      uint32_t Instr = Op;
+  // System Instruction
+  void SystemInstruction(uint32_t Op, uint32_t L, uint32_t SubOp, ARMEmitter::Register rt) {
+    uint32_t Instr = Op;

-      Instr |= L << 21;
-      Instr |= SubOp;
-      Instr |= Encode_rt(rt);
+    Instr |= L << 21;
+    Instr |= SubOp;
+    Instr |= Encode_rt(rt);

-      dc32(Instr);
-    }
+    dc32(Instr);
+  }

-    // System register move
-    void SystemRegisterMove(uint32_t Op, ARMEmitter::Register rt, ARMEmitter::SystemRegister reg) {
-      uint32_t Instr = Op;
+  // System register move
+  void SystemRegisterMove(uint32_t Op, ARMEmitter::Register rt, ARMEmitter::SystemRegister reg) {
+    uint32_t Instr = Op;

-      Instr |= FEXCore::ToUnderlying(reg);
-      Instr |= Encode_rt(rt);
+    Instr |= FEXCore::ToUnderlying(reg);
+    Instr |= Encode_rt(rt);

-      dc32(Instr);
-    }
+    dc32(Instr);
+  }

+#ifndef INCLUDED_BY_EMITTER
+}; // struct LoadstoreEmitterOps
+} // namespace ARMEmitter
+#endif
@@ -34,11 +34,7 @@
 // by the corresponding fields in the logical instruction.
 // If it can not be encoded, the function returns false, and the values pointed
 // to by n, imm_s and imm_r are undefined.
-static bool IsImmLogical(uint64_t value,
-                             unsigned width,
-                             unsigned* n,
-                             unsigned* imm_s,
-                             unsigned* imm_r) {
+static bool IsImmLogical(uint64_t value, unsigned width, unsigned* n = nullptr, unsigned* imm_s = nullptr, unsigned* imm_r = nullptr) {
  [[maybe_unused]] constexpr auto kBRegSize = 8;
  [[maybe_unused]] constexpr auto kHRegSize = 16;
  [[maybe_unused]] constexpr auto kSRegSize = 32;
@@ -47,8 +43,7 @@ static bool IsImmLogical(uint64_t value,
  constexpr auto kWRegSize = 32;
  constexpr auto kXRegSize = 64;

-  LOGMAN_THROW_A_FMT((width == kBRegSize) || (width == kHRegSize) ||
-              (width == kSRegSize) || (width == kDRegSize), "Unexpected imm size");
+  LOGMAN_THROW_A_FMT((width == kBRegSize) || (width == kHRegSize) || (width == kSRegSize) || (width == kDRegSize), "Unexpected imm size");

  bool negate = false;

@@ -182,12 +177,7 @@ static bool IsImmLogical(uint64_t value,
  // (1 + 2^d + 2^(2d) + ...), i.e. 0x0001000100010001 or similar. These can
  // be derived using a table lookup on CLZ(d).
  static const uint64_t multipliers[] = {
-      0x0000000000000001UL,
-      0x0000000100000001UL,
-      0x0001000100010001UL,
-      0x0101010101010101UL,
-      0x1111111111111111UL,
-      0x5555555555555555UL,
+    0x0000000000000001UL, 0x0000000100000001UL, 0x0001000100010001UL, 0x0101010101010101UL, 0x1111111111111111UL, 0x5555555555555555UL,
  };
  uint64_t multiplier = multipliers[CountLeadingZeros(d, kXRegSize) - 57];
  uint64_t candidate = (b - a) * multiplier;
@@ -243,16 +233,62 @@ static bool IsImmLogical(uint64_t value,
  return true;
 }

+static inline bool IsIntN(unsigned n, int64_t x) {
+  if (n == 64) {
+    return true;
+  }
+  int64_t limit = INT64_C(1) << (n - 1);
+  return (-limit <= x) && (x < limit);
+}
+
+static inline bool IsUintN(unsigned n, int64_t x) {
+  // Convert to an unsigned integer to avoid implementation-defined behavior.
+  return !(static_cast<uint64_t>(x) >> n);
+}
+
+// clang-format off
+#define INT_1_TO_32_LIST(V)                                                    \
+V(1)  V(2)  V(3)  V(4)  V(5)  V(6)  V(7)  V(8)                                 \
+V(9)  V(10) V(11) V(12) V(13) V(14) V(15) V(16)                                \
+V(17) V(18) V(19) V(20) V(21) V(22) V(23) V(24)                                \
+V(25) V(26) V(27) V(28) V(29) V(30) V(31) V(32)
+
+#define INT_33_TO_63_LIST(V)                                                   \
+V(33) V(34) V(35) V(36) V(37) V(38) V(39) V(40)                                \
+V(41) V(42) V(43) V(44) V(45) V(46) V(47) V(48)                                \
+V(49) V(50) V(51) V(52) V(53) V(54) V(55) V(56)                                \
+V(57) V(58) V(59) V(60) V(61) V(62) V(63)
+
+#define INT_1_TO_63_LIST(V) INT_1_TO_32_LIST(V) INT_33_TO_63_LIST(V)
+
+// clang-format on
+
+#define DECLARE_IS_INT_N(N)                \
+  static inline bool IsInt##N(int64_t x) { \
+    return IsIntN(N, x);                   \
+  }
+
+#define DECLARE_IS_UINT_N(N)                \
+  static inline bool IsUint##N(int64_t x) { \
+    return IsUintN(N, x);                   \
+  }
+
+INT_1_TO_63_LIST(DECLARE_IS_INT_N)
+INT_1_TO_63_LIST(DECLARE_IS_UINT_N)
+
+#undef DECLARE_IS_INT_N
+#undef DECLARE_IS_UINT_N
+
 private:

-template <typename V>
+template<typename V>
 static inline bool IsPowerOf2(V value) {
  return (value != 0) && ((value & (value - 1)) == 0);
 }

 // Some compilers dislike negating unsigned integers,
 // so we provide an equivalent.
-template <typename T>
+template<typename T>
 static inline T UnsignedNegate(T value) {
  static_assert(std::is_unsigned<T>::value);
  return ~value + 1;
@@ -262,7 +298,7 @@ static inline uint64_t LowestSetBit(uint64_t value) {
  return value & UnsignedNegate(value);
 }

-template <typename V>
+template<typename V>
 static inline int CountLeadingZeros(V value, int width = (sizeof(V) * 8)) {
 #if COMPILER_HAS_BUILTIN_CLZ
  if (width == 32) {
@@ -0,0 +1,3 @@
+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.
@@ -0,0 +1,35 @@
+# This is a reference AArch64 cross compile script
+# Pass in to cmake when building:
+# eg: cmake --toolchain ../Data/CMake/toolchain_aarch64.cmake ..
+if (NOT DEFINED ENV{SYSROOT})
+  message(FATAL_ERROR "Need to have SYSROOT environment variable set")
+endif()
+
+set(CMAKE_SYSTEM_NAME Linux)
+set(CMAKE_SYSTEM_PROCESSOR aarch64)
+set(CMAKE_CROSSCOMPILING TRUE)
+
+# Target triple needs to match the binutils exactly
+set(TARGET_TRIPLE aarch64-linux-gnu)
+set(CMAKE_C_COMPILER "clang")
+set(CMAKE_CXX_COMPILER "clang++")
+set(CMAKE_C_COMPILER_AR "llvm-ar")
+set(CMAKE_CXX_COMPILER_AR "llvm-ar")
+set(CMAKE_C_COMPILER_RANLIB "llvm-ranlib")
+set(CMAKE_CXX_COMPILER_RANLIB "llvm-ranlib")
+set(CMAKE_LINKER "ld.lld")
+
+set(CMAKE_C_COMPILER_TARGET ${TARGET_TRIPLE})
+set(CMAKE_CXX_COMPILER_TARGET ${TARGET_TRIPLE})
+
+# Set the environment variable SYSROOT to the aarch64 rootfs
+set(CMAKE_FIND_ROOT_PATH "$ENV{SYSROOT}")
+set(CMAKE_SYSROOT "$ENV{SYSROOT}")
+
+list(APPEND CMAKE_PREFIX_PATH "$ENV{SYSROOT}/usr/lib/${TARGET_TRIPLE}/cmake/")
+
+set(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
+
+set(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
+set(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)
+set(CMAKE_FIND_ROOT_PATH_MODE_PACKAGE ONLY)
@@ -10,6 +10,10 @@ set(CMAKE_DLLTOOL ${MINGW_TRIPLE}-dlltool)
 # (https://github.com/llvm/llvm-project/issues/47432)
 set(CMAKE_SHARED_LINKER_FLAGS_INIT "-static -static-libgcc -static-libstdc++ -Wl,--file-alignment=4096,/mllvm:-align-loops=1")
 set(CMAKE_EXE_LINKER_FLAGS_INIT "-static -static-libgcc -static-libstdc++ -Wl,--file-alignment=4096,/mllvm:-align-loops=1")
-
+set(CMAKE_C_STANDARD_LIBRARIES "" CACHE STRING "" FORCE)
+set(CMAKE_CXX_STANDARD_LIBRARIES "" CACHE STRING "" FORCE)
+set(CMAKE_STANDARD_LIBRARIES "" CACHE STRING "" FORCE)
 set(CMAKE_SYSTEM_NAME Windows)
 set(CMAKE_SYSTEM_PROCESSOR ${MINGW_TRIPLE})
+
+set(CMAKE_FIND_ROOT_PATH_MODE_PACKAGE ONLY)
@@ -0,0 +1,31 @@
+# --- Stage 1: Builder ---
+FROM ubuntu:22.04 as builder
+
+RUN DEBIAN_FRONTEND="noninteractive" apt-get update
+RUN DEBIAN_FRONTEND="noninteractive" apt install -y cmake \
+clang-13 llvm-13 nasm ninja-build pkg-config \
+libcap-dev libglfw3-dev libepoxy-dev python3-dev libsdl2-dev \
+python3 linux-headers-generic  \
+git  qtbase5-dev qtdeclarative5-dev lld
+
+RUN git clone --recurse-submodules https://github.com/FEX-Emu/FEX.git
+WORKDIR /FEX
+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 ninja
+
+WORKDIR /FEX/build
+
+# --- Stage 2: Runner ---
+FROM builder as runner
+
+RUN DEBIAN_FRONTEND="noninteractive" apt-get update
+RUN DEBIAN_FRONTEND="noninteractive" apt install -y \
+libcap-dev libglfw3-dev libepoxy-dev
+
+COPY --from=builder /FEX/Bin/* /usr/bin/
+
+WORKDIR /
@@ -9,25 +9,6 @@
        "@PREFIX_LIB@/libGL.so.1.7.0"
      ]
    },
-    "GLESv2": {
-      "Library": "libGLESv2-guest.so",
-      "Depends": [
-        "X11"
-      ],
-      "Overlay": [
-        "@PREFIX_LIB@/libGLESv2.so",
-        "@PREFIX_LIB@/libGLESv2.so.2",
-        "@PREFIX_LIB@/libGLESv2.so.2.0.0"
-      ]
-    },
-    "X11": {
-      "Library": "libX11-guest.so",
-      "Overlay": [
-        "@PREFIX_LIB@/libX11.so",
-        "@PREFIX_LIB@/libX11.so.6",
-        "@PREFIX_LIB@/libX11.so.6.4.0"
-      ]
-    },
    "Vulkan": {
      "Library": "libvulkan-guest.so",
      "Overlay": [
@@ -36,89 +17,6 @@
        "@HOME@/.local/share/Steam/ubuntu12_32/steam-runtime/pinned_libs_64/libvulkan.so.1"
      ]
    },
-    "xcb": {
-      "Depends": [
-        "X11"
-      ],
-      "Library": "libxcb-guest.so",
-      "Overlay": [
-        "@PREFIX_LIB@/libxcb.so",
-        "@PREFIX_LIB@/libxcb.so.1",
-        "@PREFIX_LIB@/libxcb.so.1.1.0"
-      ]
-    },
-    "xcb-dri2": {
-      "Library": "libxcb-dri2-guest.so",
-      "Overlay": [
-        "@PREFIX_LIB@/libxcb-dri2.so",
-        "@PREFIX_LIB@/libxcb-dri2.so.0",
-        "@PREFIX_LIB@/libxcb-dri2.so.0.0.0"
-      ]
-    },
-    "xcb-dri3": {
-      "Library": "libxcb-dri3-guest.so",
-      "Overlay": [
-        "@PREFIX_LIB@/libxcb-dri3.so",
-        "@PREFIX_LIB@/libxcb-dri3.so.0",
-        "@PREFIX_LIB@/libxcb-dri3.so.0.0.0"
-      ]
-    },
-    "xcb-xfixes": {
-      "Library": "libxcb-xfixes-guest.so",
-      "Overlay": [
-        "@PREFIX_LIB@/libxcb-xfixes.so",
-        "@PREFIX_LIB@/libxcb-xfixes.so.0",
-        "@PREFIX_LIB@/libxcb-xfixes.so.0.0.0"
-      ]
-    },
-    "xcb-shm": {
-      "Library": "libxcb-shm-guest.so",
-      "Overlay": [
-        "@PREFIX_LIB@/libxcb-shm.so",
-        "@PREFIX_LIB@/libxcb-shm.so.0",
-        "@PREFIX_LIB@/libxcb-shm.so.0.0.0"
-      ]
-    },
-    "xcb-sync": {
-      "Library": "libxcb-sync-guest.so",
-      "Overlay": [
-        "@PREFIX_LIB@/libxcb-sync.so",
-        "@PREFIX_LIB@/libxcb-sync.so.1",
-        "@PREFIX_LIB@/libxcb-sync.so.1.0.0"
-      ]
-    },
-    "xcb-randr": {
-      "Library": "libxcb-randr-guest.so",
-      "Overlay": [
-        "@PREFIX_LIB@/libxcb-randr.so",
-        "@PREFIX_LIB@/libxcb-randr.so.0",
-        "@PREFIX_LIB@/libxcb-randr.so.0.1.0"
-      ]
-    },
-    "xcb-present": {
-      "Library": "libxcb-present-guest.so",
-      "Overlay": [
-        "@PREFIX_LIB@/libxcb-present.so",
-        "@PREFIX_LIB@/libxcb-present.so.0",
-        "@PREFIX_LIB@/libxcb-present.so.0.0.0"
-      ]
-    },
-    "xcb-glx": {
-      "Library": "libxcb-glx-guest.so",
-      "Overlay": [
-        "@PREFIX_LIB@/libxcb-glx.so",
-        "@PREFIX_LIB@/libxcb-glx.so.0",
-        "@PREFIX_LIB@/libxcb-glx.so.0.0.0"
-      ]
-    },
-    "xshmfence": {
-      "Library": "libxshmfence-guest.so",
-      "Overlay": [
-        "@PREFIX_LIB@/libxshmfence.so",
-        "@PREFIX_LIB@/libxshmfence.so.1",
-        "@PREFIX_LIB@/libxshmfence.so.1.0.0"
-      ]
-    },
    "drm": {
      "Library": "libdrm-guest.so",
      "Overlay": [
@@ -141,38 +39,6 @@
        "@PREFIX_LIB@/libfex_thunk_test.so"
      ]
    },
-    "Xrender": {
-      "Library": "libXrender-guest.so",
-      "Overlay": [
-        "@PREFIX_LIB@/libXrender.so",
-        "@PREFIX_LIB@/libXrender.so.1",
-        "@PREFIX_LIB@/libXrender.so.1.3.0"
-      ]
-    },
-    "Xext": {
-      "Library": "libXext-guest.so",
-      "Overlay": [
-        "@PREFIX_LIB@/libXext.so",
-        "@PREFIX_LIB@/libXext.so.6",
-        "@PREFIX_LIB@/libXext.so.6.4.0"
-      ]
-    },
-    "Xfixes": {
-      "Library": "libXfixes-guest.so",
-      "Overlay": [
-        "@PREFIX_LIB@/libXfixes.so",
-        "@PREFIX_LIB@/libXfixes.so.3",
-        "@PREFIX_LIB@/libXfixes.so.3.1.0"
-      ]
-    },
-    "OpenCL": {
-      "Library" : "libOpenCL-guest.so",
-      "Overlay": [
-        "@PREFIX_LIB@/libOpenCL.so",
-        "@PREFIX_LIB@/libOpenCL.so.1",
-        "@PREFIX_LIB@/libOpenCL.so.1.0.0"
-      ]
-    },
    "WaylandClient": {
      "Library" : "libwayland-client-guest.so",
      "Overlay": [
@@ -180,7 +46,6 @@
        "@PREFIX_LIB@/libwayland-client.so.0",
        "@PREFIX_LIB@/libwayland-client.so.0.20.0"
      ]
-    },
-    "":{}
+    }
  }
 }
@@ -13,5 +13,14 @@ function(GenBinFmt Name)
    DESTINATION ${CMAKE_INSTALL_PREFIX}/share/binfmts/)
 endfunction()

-GenBinFmt(FEX-x86.in)
-GenBinFmt(FEX-x86_64.in)
+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/)
+else()
+  GenBinFmt(FEX-x86.in)
+  GenBinFmt(FEX-x86_64.in)
+endif()
@@ -0,0 +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
@@ -6,4 +6,3 @@ mask \xff\xff\xff\xff\xff\xfe\xfe\x00\x00\x00\x00\xff\xff\xff\xff\xff\xfe\xff\xf
 credentials yes
 fix_binary yes
 preserve yes
-expose_interpreter optional
@@ -0,0 +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
@@ -6,4 +6,3 @@ mask \xff\xff\xff\xff\xff\xfe\xfe\x00\x00\x00\x00\xff\xff\xff\xff\xff\xfe\xff\xf
 credentials yes
 fix_binary yes
 preserve yes
-expose_interpreter optional
@@ -0,0 +1,35 @@
+{ pkgs ? import <nixpkgs> { } }:
+
+let
+  pkgsCross32 = pkgs.pkgsCross.gnu32;
+  pkgsCross64 = pkgs.pkgsCross.gnu64;
+
+  gcc32 = pkgs.writeText "toolchain_nix_gcc_x86_32.txt" ''
+    set(CMAKE_SYSTEM_PROCESSOR i686)
+    set(CMAKE_C_COMPILER ${pkgsCross32.buildPackages.gcc}/bin/i686-unknown-linux-gnu-gcc)
+    set(CMAKE_CXX_COMPILER ${pkgsCross32.buildPackages.gcc}/bin/i686-unknown-linux-gnu-g++)
+  '';
+
+  gcc64 = pkgs.writeText "toolchain_nix_gcc_x86_64.txt" ''
+    set(CMAKE_SYSTEM_PROCESSOR x86_64)
+    set(CMAKE_C_COMPILER ${pkgsCross64.buildPackages.gcc}/bin/x86_64-unknown-linux-gnu-gcc)
+    set(CMAKE_CXX_COMPILER ${pkgsCross64.buildPackages.gcc}/bin/x86_64-unknown-linux-gnu-g++)
+  '';
+in
+pkgs.mkShell {
+  buildInputs = [
+    pkgsCross64.buildPackages.clang
+    pkgsCross32.buildPackages.clang
+  ];
+
+  shellHook = ''
+    if [[ $- == *i* ]]; then
+      echo "toolchain32: ${gcc32}"
+      echo "toolchain64: ${gcc64}"
+      echo ""
+      echo "Use \$FEX_CMAKE_TOOLCHAINS to configure CMake."
+    fi
+  '';
+
+  FEX_CMAKE_TOOLCHAINS = "-DX86_32_TOOLCHAIN_FILE=${gcc32} -DX86_64_TOOLCHAIN_FILE=${gcc64}";
+}
@@ -0,0 +1,83 @@
+{ pkgs ? import <nixpkgs> { } }:
+
+let
+  pkgsCross32 = pkgs.pkgsCross.gnu32;
+  pkgsCross64 = pkgs.pkgsCross.gnu64;
+
+  devRootFS = pkgs.buildEnv {
+    name = "fex-dev-rootfs";
+    paths = [
+      pkgsCross64.stdenv.cc.libc_dev
+      pkgsCross32.stdenv.cc.libc_dev
+      pkgsCross64.stdenv.cc.cc
+      pkgsCross32.stdenv.cc.cc
+
+      pkgs.alsa-lib.dev
+      pkgs.libdrm.dev
+      pkgs.libGL.dev
+      pkgs.wayland.dev
+      pkgs.xorg.libX11.dev
+      pkgs.xorg.libxcb.dev
+      pkgs.xorg.libXrandr.dev
+      pkgs.xorg.libXrender.dev
+      pkgs.xorg.xorgproto
+    ];
+    ignoreCollisions = true;
+    pathsToLink = [
+      "/include"
+      "/lib"
+    ];
+
+    postBuild = ''
+      mkdir -p $out/usr
+      ln -s $out/include $out/usr/
+    '';
+  };
+
+  toolchain32 = pkgs.writeText "toolchain_nix_x86_32.txt" ''
+    set(CMAKE_EXE_LINKER_FLAGS_INIT "-fuse-ld=lld")
+    set(CMAKE_MODULE_LINKER_FLAGS_INIT "-fuse-ld=lld")
+    set(CMAKE_SHARED_LINKER_FLAGS_INIT "-fuse-ld=lld")
+    set(CMAKE_SYSTEM_PROCESSOR i686)
+    set(CMAKE_C_COMPILER clang)
+    set(CMAKE_CXX_COMPILER clang++)
+    set(CMAKE_C_COMPILER ${pkgsCross32.buildPackages.clang}/bin/i686-unknown-linux-gnu-clang)
+    set(CMAKE_CXX_COMPILER ${pkgsCross32.buildPackages.clang}/bin/i686-unknown-linux-gnu-clang++)
+    set(CLANG_FLAGS "-nodefaultlibs -nostartfiles -lstdc++ -target i686-linux-gnu -msse2 -mfpmath=sse --sysroot=${devRootFS} -iwithsysroot/include")
+    set(CMAKE_C_FLAGS "''${CMAKE_C_FLAGS} ''${CLANG_FLAGS}")
+    set(CMAKE_CXX_FLAGS "''${CMAKE_CXX_FLAGS} ''${CLANG_FLAGS}")
+  '';
+
+  toolchain64 = pkgs.writeText "toolchain_nix_x86_64.txt" ''
+    set(CMAKE_EXE_LINKER_FLAGS_INIT "-fuse-ld=lld")
+    set(CMAKE_MODULE_LINKER_FLAGS_INIT "-fuse-ld=lld")
+    set(CMAKE_SHARED_LINKER_FLAGS_INIT "-fuse-ld=lld")
+    set(CMAKE_SYSTEM_PROCESSOR x86_64)
+    set(CMAKE_C_COMPILER clang)
+    set(CMAKE_CXX_COMPILER clang++)
+    set(CMAKE_C_COMPILER ${pkgsCross64.buildPackages.clang}/bin/x86_64-unknown-linux-gnu-clang)
+    set(CMAKE_CXX_COMPILER ${pkgsCross64.buildPackages.clang}/bin/x86_64-unknown-linux-gnu-clang++)
+    set(CLANG_FLAGS "-nodefaultlibs -nostartfiles -lstdc++ -target x86_64-linux-gnu --sysroot=${devRootFS} -iwithsysroot/usr/include")
+    set(CMAKE_C_FLAGS "''${CMAKE_C_FLAGS} ''${CLANG_FLAGS}")
+    set(CMAKE_CXX_FLAGS "''${CMAKE_CXX_FLAGS} ''${CLANG_FLAGS}")
+  '';
+in
+pkgs.mkShell {
+  buildInputs = [
+    pkgsCross64.buildPackages.clang
+    pkgsCross32.buildPackages.clang
+  ];
+
+  shellHook = ''
+    if [[ $- == *i* ]]; then
+      echo "Set up dev RootFS at ${devRootFS}"
+      echo "toolchain32: ${toolchain32}"
+      echo "toolchain64: ${toolchain64}"
+      echo ""
+      echo "Use \$FEX_CMAKE_TOOLCHAINS to configure CMake."
+    fi
+  '';
+
+  FEX_CMAKE_TOOLCHAINS = "-DX86_32_TOOLCHAIN_FILE=${toolchain32} -DX86_64_TOOLCHAIN_FILE=${toolchain64} -DX86_DEV_ROOTFS=${devRootFS}";
+  ROOTFS = "${devRootFS}";
+}
@@ -0,0 +1,52 @@
+{ pkgs ? import <nixpkgs> { } }:
+
+let
+  toolchain = pkgs.fetchzip {
+    url = "https://github.com/bylaws/llvm-mingw/releases/download/20250305/llvm-mingw-20250305-ucrt-ubuntu-20.04-aarch64.tar.xz";
+    sha256 = "sha256-cA03/ab9O61eO9+S2JzIXD4V0HzTXK5/AYyxW2d73Po=";
+  };
+
+  cmakeToolchainFile = pkgs.substitute {
+    # Use absolute paths that are discoverable outside of the nix shell
+    src = ../../CMake/toolchain_mingw.cmake;
+    substitutions = ["--replace-fail" "\${MINGW_TRIPLE}-" "${toolchain}/bin/\${MINGW_TRIPLE}-"];
+  };
+
+  mesonCrossFile = pkgs.writeText "crossfile_llvm_mingw.txt" ''
+    [binaries]
+    ar = '${toolchain}/bin/arm64ec-w64-mingw32-ar'
+    c = '${toolchain}/bin/arm64ec-w64-mingw32-gcc'
+    cpp = '${toolchain}/bin/arm64ec-w64-mingw32-g++'
+    ld = '${toolchain}/bin/arm64ec-w64-mingw32-ld'
+    windres = '${toolchain}/bin/arm64ec-w64-mingw32-windres'
+    strip = '${toolchain}/bin/strip'
+    widl = '${toolchain}/bin/arm64ec-w64-mingw32-widl'
+    pkgconfig = 'aarch64-linux-gnu-pkg-config'
+    [host_machine]
+    system = 'windows'
+    cpu_family = 'aarch64'
+    cpu = 'aarch64'
+    endian = 'little'
+  '';
+in
+pkgs.mkShell {
+  buildInputs = [
+    toolchain
+  ];
+
+  shellHook = ''
+    if [[ $- == *i* ]]; then
+      echo "llvm-mingw set up at ${toolchain}."
+      echo ""
+      echo "To configure DXVK/vkd3d-proton: meson setup \$FEX_MESON_CROSSFILE"
+      echo ""
+      echo "To configure 32-bit FEX build: cmake \$FEX_CMAKE_TOOLCHAIN_WOW64"
+      echo "To configure 64-bit FEX build: cmake \$FEX_CMAKE_TOOLCHAIN_ARM64EC"
+    fi
+  '';
+
+  # E.g. cmake $FEX_CMAKE_TOOLCHAIN_ARM64EC -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr -DENABLE_LTO=False -DBUILD_TESTS=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}";
+}
@@ -0,0 +1,21 @@
+#! /usr/bin/env nix-shell
+#! nix-shell -i bash WineOnArm/shell.nix
+
+# Helper script to configure CMake for building FEX as library for emulation
+# of 32-bit applications in Wine/Proton.
+# The required cross-toolchains will be set up and managed by nix.
+
+if [ $# -eq 0 ]
+then
+  echo "Expected CMake argument list"
+  exit 1
+fi
+
+if [ -f CMakeCache.txt ]
+then
+  echo "Expected empty build folder"
+  exit 1
+fi
+
+set -o xtrace
+cmake $FEX_CMAKE_TOOLCHAIN_WOW64 -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr -DENABLE_LTO=False -DBUILD_TESTS=False $@
@@ -0,0 +1,21 @@
+#! /usr/bin/env nix-shell
+#! nix-shell -i bash WineOnArm/shell.nix
+
+# Helper script to configure CMake for building FEX as library for emulation
+# of 64-bit applications in Wine/Proton
+# Nix is used to install and manage the required cross-toolchains.
+
+if [ $# -eq 0 ]
+then
+  echo "Expected CMake argument list"
+  exit 1
+fi
+
+if [ -f CMakeCache.txt ]
+then
+  echo "Expected empty build folder"
+  exit 1
+fi
+
+set -o xtrace
+cmake $FEX_CMAKE_TOOLCHAIN_ARM64EC -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr -DENABLE_LTO=False -DBUILD_TESTS=False $@
@@ -0,0 +1,17 @@
+#! /usr/bin/env nix-shell
+#! nix-shell -i bash FEXLinuxTests/shell.nix
+
+# Helper script to configure CMake for building FEXLinuxTests.
+# Nix is used to install and manage the required cross-toolchains.
+
+if [ ! -f CMakeCache.txt ]
+then
+  echo "Must be run from a pre-configured CMake build folder"
+  exit 1
+fi
+
+# Remove previous build to ensure the new toolchain is applied
+rm -rf unittests/FEXLinuxTests
+
+set -o xtrace
+cmake . $FEX_CMAKE_TOOLCHAINS -DBUILD_TESTS=ON -DBUILD_FEX_LINUX_TESTS=ON
@@ -0,0 +1,22 @@
+# Helper script to configure CMake for library forwarding in FEX.
+# Nix is used to install and manage the required cross-toolchains.
+
+if [ ! -f CMakeCache.txt ]
+then
+  echo "Must be run from a pre-configured CMake build folder"
+  exit 1
+fi
+
+# Remove previous build to ensure the new toolchain is applied
+rm -rf guest-libs guest-libs-32 Guest Guest_32
+
+# Set clang executable path manually since the one from the nix store
+# will be picked up otherwise
+CLANG_EXEC_PATH=""
+if ! grep -q CLANG_EXEC_PATH CMakeCache.txt
+then
+  CLANG_EXEC_PATH="-DCLANG_EXEC_PATH=`which clang`"
+fi
+
+nix-shell `dirname -- "$0"`/LibraryForwarding/shell.nix \
+  --run "set -o xtrace; cmake . \$FEX_CMAKE_TOOLCHAINS -DBUILD_THUNKS=ON $CLANG_EXEC_PATH; set +o xtrace"
@@ -1,31 +0,0 @@
-# --- Stage 1: Builder ---
-FROM ubuntu:20.04 as builder
-
-RUN DEBIAN_FRONTEND="noninteractive" apt-get update
-RUN DEBIAN_FRONTEND="noninteractive" apt install -y cmake \
-clang-10 llvm-10 nasm ninja-build pkg-config \
-libcap-dev libglfw3-dev libepoxy-dev python3-dev libsdl2-dev \
-python3 linux-headers-generic \
-git
-
-RUN git clone --recurse-submodules https://github.com/FEX-Emu/FEX.git
-
-CMD [ "mkdir /opt/FEX/build" ]
-
-WORKDIR /opt/FEX/build
-
-ARG CC=clang-10
-ARG CXX=clang++-10
-RUN cmake -G Ninja .. -DCMAKE_BUILD_TYPE=Release
-RUN ninja
-
-# --- Stage 2: Runner ---
-FROM ubuntu:20.04
-
-RUN DEBIAN_FRONTEND="noninteractive" apt-get update
-RUN DEBIAN_FRONTEND="noninteractive" apt install -y \
-libcap-dev libglfw3-dev libepoxy-dev
-
-COPY --from=builder /opt/FEX/build/Bin/* /usr/bin/
-
-WORKDIR /root
@@ -0,0 +1,98 @@
+
+set (SRCS
+  # F80 support
+  src/extF80_add.c
+  src/extF80_div.c
+  src/extF80_sub.c
+  src/extF80_mul.c
+  src/extF80_rem.c
+  src/extF80_sqrt.c
+  src/extF80_le.c
+  src/extF80_to_i32.c
+  src/extF80_to_i64.c
+  src/extF80_to_ui64.c
+  src/extF80_to_f32.c
+  src/extF80_to_f64.c
+  src/i32_to_extF80.c
+  src/ui64_to_extF80.c
+  src/extF80_to_f128.c
+  src/f128_to_extF80.c
+
+  # F128 support
+  src/f128_add.c
+  src/f128_div.c
+  src/f128_eq.c
+  src/f128_eq_signaling.c
+  src/f128_isSignalingNaN.c
+  src/f128_le.c
+  src/f128_le_quiet.c
+  src/f128_lt.c
+  src/f128_lt_quiet.c
+  src/f128_mulAdd.c
+  src/f128_mul.c
+  src/f128_rem.c
+  src/f128_sqrt.c
+  src/f128_sub.c
+  src/f128_to_f16.c
+  src/f128_to_f32.c
+  src/f128_to_f64.c
+  src/f128_to_i32.c
+  src/f128_to_i64.c
+  src/f128_to_ui32.c
+  src/f128_to_ui64.c
+  src/s_addMagsF128.c
+  src/s_subMagsF128.c
+  src/s_normRoundPackToF128.c
+  src/s_roundPackToF128.c
+  src/s_propagateNaNF128UI.c
+
+  # Conversion
+  src/f32_to_f128.c
+  src/i32_to_f128.c
+
+  src/s_roundToUI64.c
+  src/s_f128UIToCommonNaN.c
+  src/s_commonNaNToF128UI.c
+  src/s_normSubnormalF128Sig.c
+  src/s_roundToI32.c
+  src/s_roundToI64.c
+  src/s_roundPackToF32.c
+  src/s_addMagsExtF80.c
+  src/s_extF80UIToCommonNaN.c
+  src/s_commonNaNToF32UI.c
+  src/s_commonNaNToF64UI.c
+  src/s_roundPackToF64.c
+  src/s_propagateNaNExtF80UI.c
+  src/s_roundPackToExtF80.c
+  src/s_normSubnormalExtF80Sig.c
+  src/s_subMagsExtF80.c
+  src/s_shiftRightJam128.c
+  src/s_shiftRightJam128Extra.c
+  src/s_normRoundPackToExtF80.c
+  src/s_approxRecip_1Ks.c
+  src/s_approxRecipSqrt32_1.c
+  src/s_approxRecipSqrt_1Ks.c
+  src/softfloat_raiseFlags.c
+  src/f64_to_extF80.c
+  src/s_commonNaNToExtF80UI.c
+  src/s_normSubnormalF64Sig.c
+  src/s_f64UIToCommonNaN.c
+  src/extF80_roundToInt.c
+  src/extF80_eq.c
+  src/extF80_lt.c
+  src/f32_to_extF80.c
+  src/s_normSubnormalF32Sig.c
+  src/s_f32UIToCommonNaN.c)
+
+if (_M_ARM_64 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()
+
+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})
@@ -50,50 +50,11 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include <stdint.h>
 #include "softfloat_types.h"

-#ifndef THREAD_LOCAL
-#define THREAD_LOCAL
-#endif
-
-/*----------------------------------------------------------------------------
-| Software floating-point underflow tininess-detection mode.
-*----------------------------------------------------------------------------*/
-extern THREAD_LOCAL uint_fast8_t softfloat_detectTininess;
-enum {
-    softfloat_tininess_beforeRounding = 0,
-    softfloat_tininess_afterRounding  = 1
-};
-
-/*----------------------------------------------------------------------------
-| Software floating-point rounding mode.  (Mode "odd" is supported only if
-| SoftFloat is compiled with macro 'SOFTFLOAT_ROUND_ODD' defined.)
-*----------------------------------------------------------------------------*/
-extern THREAD_LOCAL uint_fast8_t softfloat_roundingMode;
-enum {
-    softfloat_round_near_even   = 0,
-    softfloat_round_minMag      = 1,
-    softfloat_round_min         = 2,
-    softfloat_round_max         = 3,
-    softfloat_round_near_maxMag = 4,
-    softfloat_round_odd         = 6
-};
-
-/*----------------------------------------------------------------------------
-| Software floating-point exception flags.
-*----------------------------------------------------------------------------*/
-extern THREAD_LOCAL uint_fast8_t softfloat_exceptionFlags;
-enum {
-    softfloat_flag_inexact   =  1,
-    softfloat_flag_underflow =  2,
-    softfloat_flag_overflow  =  4,
-    softfloat_flag_infinite  =  8,
-    softfloat_flag_invalid   = 16
-};
-
 /*----------------------------------------------------------------------------
 | Routine to raise any or all of the software floating-point exception flags.
 *----------------------------------------------------------------------------*/
 FEXCORE_PRESERVE_ALL_ATTR
-void softfloat_raiseFlags( uint_fast8_t );
+void softfloat_raiseFlags( struct softfloat_state *, uint_fast8_t );

 /*----------------------------------------------------------------------------
 | Integer-to-floating-point conversion routines.
@@ -187,8 +148,8 @@ float16_t f32_to_f16( float32_t );
 float64_t f32_to_f64( float32_t );
 #ifdef SOFTFLOAT_FAST_INT64
 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t f32_to_extF80( float32_t );
-float128_t f32_to_f128( float32_t );
+extFloat80_t f32_to_extF80( struct softfloat_state *, float32_t );
+float128_t f32_to_f128( struct softfloat_state *, float32_t );
 #endif
 void f32_to_extF80M( float32_t, extFloat80_t * );
 void f32_to_f128M( float32_t, float128_t * );
@@ -223,7 +184,7 @@ float16_t f64_to_f16( float64_t );
 float32_t f64_to_f32( float64_t );
 #ifdef SOFTFLOAT_FAST_INT64
 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t f64_to_extF80( float64_t );
+extFloat80_t f64_to_extF80( struct softfloat_state *, float64_t );
 float128_t f64_to_f128( float64_t );
 #endif
 void f64_to_extF80M( float64_t, extFloat80_t * );
@@ -244,57 +205,55 @@ bool f64_le_quiet( float64_t, float64_t );
 bool f64_lt_quiet( float64_t, float64_t );
 bool f64_isSignalingNaN( float64_t );

-/*----------------------------------------------------------------------------
-| Rounding precision for 80-bit extended double-precision floating-point.
-| Valid values are 32, 64, and 80.
-*----------------------------------------------------------------------------*/
-extern THREAD_LOCAL uint_fast8_t extF80_roundingPrecision;
-
 /*----------------------------------------------------------------------------
 | 80-bit extended double-precision floating-point operations.
 *----------------------------------------------------------------------------*/
 #ifdef SOFTFLOAT_FAST_INT64
 uint_fast32_t extF80_to_ui32( extFloat80_t, uint_fast8_t, bool );
 FEXCORE_PRESERVE_ALL_ATTR
-uint_fast64_t extF80_to_ui64( extFloat80_t, uint_fast8_t, bool );
+uint_fast64_t extF80_to_ui64( struct softfloat_state *, extFloat80_t, uint_fast8_t, bool );
 FEXCORE_PRESERVE_ALL_ATTR
-int_fast32_t extF80_to_i32( extFloat80_t, uint_fast8_t, bool );
+int_fast32_t extF80_to_i32( struct softfloat_state *, extFloat80_t, uint_fast8_t, bool );
 FEXCORE_PRESERVE_ALL_ATTR
-int_fast64_t extF80_to_i64( extFloat80_t, uint_fast8_t, bool );
+int_fast64_t extF80_to_i64( struct softfloat_state *, extFloat80_t, uint_fast8_t, bool );
 uint_fast32_t extF80_to_ui32_r_minMag( extFloat80_t, bool );
 uint_fast64_t extF80_to_ui64_r_minMag( extFloat80_t, bool );
 int_fast32_t extF80_to_i32_r_minMag( extFloat80_t, bool );
 int_fast64_t extF80_to_i64_r_minMag( extFloat80_t, bool );
 float16_t extF80_to_f16( extFloat80_t );
 FEXCORE_PRESERVE_ALL_ATTR
-float32_t extF80_to_f32( extFloat80_t );
+float32_t extF80_to_f32( struct softfloat_state *, extFloat80_t );
 FEXCORE_PRESERVE_ALL_ATTR
-float64_t extF80_to_f64( extFloat80_t );
+float64_t extF80_to_f64( struct softfloat_state *, extFloat80_t );
 FEXCORE_PRESERVE_ALL_ATTR
-float128_t extF80_to_f128( extFloat80_t );
+float128_t extF80_to_f128( struct softfloat_state *, extFloat80_t );
 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t extF80_roundToInt( extFloat80_t, uint_fast8_t, bool );
+extFloat80_t extF80_roundToInt( struct softfloat_state *, extFloat80_t, uint_fast8_t, bool );
 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t extF80_add( extFloat80_t, extFloat80_t );
+extFloat80_t extF80_add( struct softfloat_state *, extFloat80_t, extFloat80_t );
 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t extF80_sub( extFloat80_t, extFloat80_t );
+extFloat80_t extF80_sub( struct softfloat_state *, extFloat80_t, extFloat80_t );
 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t extF80_mul( extFloat80_t, extFloat80_t );
+extFloat80_t extF80_mul( struct softfloat_state *, extFloat80_t, extFloat80_t );
 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t extF80_div( extFloat80_t, extFloat80_t );
+extFloat80_t extF80_div( struct softfloat_state *, extFloat80_t, extFloat80_t );
 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t extF80_rem( extFloat80_t, extFloat80_t );
+extFloat80_t extF80_rem( struct softfloat_state *, extFloat80_t, extFloat80_t );
 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t extF80_sqrt( extFloat80_t );
+extFloat80_t extF80_sqrt( struct softfloat_state *, extFloat80_t );
 FEXCORE_PRESERVE_ALL_ATTR
-bool extF80_eq( extFloat80_t, extFloat80_t );
-bool extF80_le( extFloat80_t, extFloat80_t );
+bool extF80_eq( struct softfloat_state *, extFloat80_t, extFloat80_t );
+bool extF80_le( struct softfloat_state *, extFloat80_t, extFloat80_t );
 FEXCORE_PRESERVE_ALL_ATTR
-bool extF80_lt( extFloat80_t, extFloat80_t );
+bool extF80_lt( struct softfloat_state *, extFloat80_t, extFloat80_t );
 bool extF80_eq_signaling( extFloat80_t, extFloat80_t );
 bool extF80_le_quiet( extFloat80_t, extFloat80_t );
 bool extF80_lt_quiet( extFloat80_t, extFloat80_t );
 bool extF80_isSignalingNaN( extFloat80_t );
+static inline extFloat80_t extF80_complement_sign(extFloat80_t a) {
+  a.signExp ^= 1ULL << 15;
+  return a;
+}
 #endif
 uint_fast32_t extF80M_to_ui32( const extFloat80_t *, uint_fast8_t, bool );
 uint_fast64_t extF80M_to_ui64( const extFloat80_t *, uint_fast8_t, bool );
@@ -329,34 +288,38 @@ bool extF80M_isSignalingNaN( const extFloat80_t * );
 | 128-bit (quadruple-precision) floating-point operations.
 *----------------------------------------------------------------------------*/
 #ifdef SOFTFLOAT_FAST_INT64
-uint_fast32_t f128_to_ui32( float128_t, uint_fast8_t, bool );
-uint_fast64_t f128_to_ui64( float128_t, uint_fast8_t, bool );
-int_fast32_t f128_to_i32( float128_t, uint_fast8_t, bool );
-int_fast64_t f128_to_i64( float128_t, uint_fast8_t, bool );
+uint_fast32_t f128_to_ui32( struct softfloat_state *, float128_t, uint_fast8_t, bool );
+uint_fast64_t f128_to_ui64( struct softfloat_state *, float128_t, uint_fast8_t, bool );
+int_fast32_t f128_to_i32( struct softfloat_state *, float128_t, uint_fast8_t, bool );
+int_fast64_t f128_to_i64( struct softfloat_state *, float128_t, uint_fast8_t, bool );
 uint_fast32_t f128_to_ui32_r_minMag( float128_t, bool );
 uint_fast64_t f128_to_ui64_r_minMag( float128_t, bool );
 int_fast32_t f128_to_i32_r_minMag( float128_t, bool );
 int_fast64_t f128_to_i64_r_minMag( float128_t, bool );
-float16_t f128_to_f16( float128_t );
-float32_t f128_to_f32( float128_t );
-float64_t f128_to_f64( float128_t );
+float16_t f128_to_f16( struct softfloat_state *, float128_t );
+float32_t f128_to_f32( struct softfloat_state *, float128_t );
+float64_t f128_to_f64( struct softfloat_state *, float128_t );
 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t f128_to_extF80( float128_t );
+extFloat80_t f128_to_extF80( struct softfloat_state *, float128_t );
 float128_t f128_roundToInt( float128_t, uint_fast8_t, bool );
-float128_t f128_add( float128_t, float128_t );
-float128_t f128_sub( float128_t, float128_t );
-float128_t f128_mul( float128_t, float128_t );
-float128_t f128_mulAdd( float128_t, float128_t, float128_t );
-float128_t f128_div( float128_t, float128_t );
-float128_t f128_rem( float128_t, float128_t );
-float128_t f128_sqrt( float128_t );
-bool f128_eq( float128_t, float128_t );
-bool f128_le( float128_t, float128_t );
-bool f128_lt( float128_t, float128_t );
-bool f128_eq_signaling( float128_t, float128_t );
-bool f128_le_quiet( float128_t, float128_t );
-bool f128_lt_quiet( float128_t, float128_t );
+float128_t f128_add( struct softfloat_state *, float128_t, float128_t );
+float128_t f128_sub( struct softfloat_state *, float128_t, float128_t );
+float128_t f128_mul( struct softfloat_state *, float128_t, float128_t );
+float128_t f128_mulAdd( struct softfloat_state *, float128_t, float128_t, float128_t );
+float128_t f128_div( struct softfloat_state *, float128_t, float128_t );
+float128_t f128_rem( struct softfloat_state *, float128_t, float128_t );
+float128_t f128_sqrt( struct softfloat_state *, float128_t );
+bool f128_eq( struct softfloat_state *, float128_t, float128_t );
+bool f128_le( struct softfloat_state *, float128_t, float128_t );
+bool f128_lt( struct softfloat_state *, float128_t, float128_t );
+bool f128_eq_signaling( struct softfloat_state *, float128_t, float128_t );
+bool f128_le_quiet( struct softfloat_state *, float128_t, float128_t );
+bool f128_lt_quiet( struct softfloat_state *, float128_t, float128_t );
 bool f128_isSignalingNaN( float128_t );
+static inline float128_t f128_complement_sign(float128_t a) {
+  a.v[1] ^= 1ULL << 63;
+  return a;
+}
 #endif
 uint_fast32_t f128M_to_ui32( const float128_t *, uint_fast8_t, bool );
 uint_fast64_t f128M_to_ui64( const float128_t *, uint_fast8_t, bool );
@@ -77,5 +77,50 @@ struct extFloat80M { uint16_t signExp; uint64_t signif; };
 *----------------------------------------------------------------------------*/
 typedef struct extFloat80M extFloat80_t;

+enum {
+    softfloat_tininess_beforeRounding = 0,
+    softfloat_tininess_afterRounding  = 1
+};
+
+enum {
+    softfloat_round_near_even   = 0,
+    softfloat_round_minMag      = 1,
+    softfloat_round_min         = 2,
+    softfloat_round_max         = 3,
+    softfloat_round_near_maxMag = 4,
+    softfloat_round_odd         = 6
+};
+
+enum {
+    softfloat_flag_inexact   =  1,
+    softfloat_flag_underflow =  2,
+    softfloat_flag_overflow  =  4,
+    softfloat_flag_infinite  =  8,
+    softfloat_flag_invalid   = 16
+};
+
+struct softfloat_state {
+/*----------------------------------------------------------------------------
+| Software floating-point underflow tininess-detection mode.
+*----------------------------------------------------------------------------*/
+uint8_t detectTininess; /* = init_detectTininess */
+/*----------------------------------------------------------------------------
+| Software floating-point rounding mode.  (Mode "odd" is supported only if
+| SoftFloat is compiled with macro 'SOFTFLOAT_ROUND_ODD' defined.)
+*----------------------------------------------------------------------------*/
+uint8_t roundingMode; /* = softfloat_round_near_even */
+
+/*----------------------------------------------------------------------------
+| Software floating-point exception flags.
+*----------------------------------------------------------------------------*/
+uint8_t exceptionFlags; /* = 0 */
+
+/*----------------------------------------------------------------------------
+| Rounding precision for 80-bit extended double-precision floating-point.
+| Valid values are 32, 64, and 80.
+*----------------------------------------------------------------------------*/
+uint8_t roundingPrecision; /* = 80 */
+};
+
 #endif

@@ -41,7 +41,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "softfloat.h"

 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t extF80_add( extFloat80_t a, extFloat80_t b )
+extFloat80_t extF80_add( struct softfloat_state *state, extFloat80_t a, extFloat80_t b )
 {
    union { struct extFloat80M s; extFloat80_t f; } uA;
    uint_fast16_t uiA64;
@@ -53,7 +53,7 @@ extFloat80_t extF80_add( extFloat80_t a, extFloat80_t b )
    bool signB;
    extFloat80_t
        (*magsFuncPtr)(
-            uint_fast16_t, uint_fast64_t, uint_fast16_t, uint_fast64_t, bool );
+            struct softfloat_state *, uint_fast16_t, uint_fast64_t, uint_fast16_t, uint_fast64_t, bool );

    uA.f = a;
    uiA64 = uA.s.signExp;
@@ -65,6 +65,6 @@ extFloat80_t extF80_add( extFloat80_t a, extFloat80_t b )
    signB = signExtF80UI64( uiB64 );
    magsFuncPtr =
        (signA == signB) ? softfloat_addMagsExtF80 : softfloat_subMagsExtF80;
-    return (*magsFuncPtr)( uiA64, uiA0, uiB64, uiB0, signA );
+    return (*magsFuncPtr)( state, uiA64, uiA0, uiB64, uiB0, signA );
 }

@@ -42,7 +42,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "softfloat.h"

 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t extF80_div( extFloat80_t a, extFloat80_t b )
+extFloat80_t extF80_div( struct softfloat_state *state, extFloat80_t a, extFloat80_t b )
 {
    union { struct extFloat80M s; extFloat80_t f; } uA;
    uint_fast16_t uiA64;
@@ -107,7 +107,7 @@ extFloat80_t extF80_div( extFloat80_t a, extFloat80_t b )
    if ( ! (sigB & UINT64_C( 0x8000000000000000 )) ) {
        if ( ! sigB ) {
            if ( ! sigA ) goto invalid;
-            softfloat_raiseFlags( softfloat_flag_infinite );
+            softfloat_raiseFlags( state, softfloat_flag_infinite );
            goto infinity;
        }
        normExpSig = softfloat_normSubnormalExtF80Sig( sigB );
@@ -169,18 +169,18 @@ extFloat80_t extF80_div( extFloat80_t a, extFloat80_t b )
    sigZExtra = (uint64_t) ((uint_fast64_t) q<<41);
    return
        softfloat_roundPackToExtF80(
-            signZ, expZ, sigZ, sigZExtra, extF80_roundingPrecision );
+            state, signZ, expZ, sigZ, sigZExtra, state->roundingPrecision );
    /*------------------------------------------------------------------------
    *------------------------------------------------------------------------*/
 propagateNaN:
-    uiZ = softfloat_propagateNaNExtF80UI( uiA64, uiA0, uiB64, uiB0 );
+    uiZ = softfloat_propagateNaNExtF80UI( state, uiA64, uiA0, uiB64, uiB0 );
    uiZ64 = uiZ.v64;
    uiZ0  = uiZ.v0;
    goto uiZ;
    /*------------------------------------------------------------------------
    *------------------------------------------------------------------------*/
 invalid:
-    softfloat_raiseFlags( softfloat_flag_invalid );
+    softfloat_raiseFlags( state, softfloat_flag_invalid );
    uiZ64 = defaultNaNExtF80UI64;
    uiZ0  = defaultNaNExtF80UI0;
    goto uiZ;
@@ -42,7 +42,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "softfloat.h"

 FEXCORE_PRESERVE_ALL_ATTR
-bool extF80_eq( extFloat80_t a, extFloat80_t b )
+bool extF80_eq( struct softfloat_state *state, extFloat80_t a, extFloat80_t b )
 {
    union { struct extFloat80M s; extFloat80_t f; } uA;
    uint_fast16_t uiA64;
@@ -62,7 +62,7 @@ bool extF80_eq( extFloat80_t a, extFloat80_t b )
               softfloat_isSigNaNExtF80UI( uiA64, uiA0 )
            || softfloat_isSigNaNExtF80UI( uiB64, uiB0 )
        ) {
-            softfloat_raiseFlags( softfloat_flag_invalid );
+            softfloat_raiseFlags( state, softfloat_flag_invalid );
        }
        return false;
    }
@@ -0,0 +1,73 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+bool extF80_le( struct softfloat_state *state, extFloat80_t a, extFloat80_t b )
+{
+    union { struct extFloat80M s; extFloat80_t f; } uA;
+    uint_fast16_t uiA64;
+    uint_fast64_t uiA0;
+    union { struct extFloat80M s; extFloat80_t f; } uB;
+    uint_fast16_t uiB64;
+    uint_fast64_t uiB0;
+    bool signA, signB;
+
+    uA.f = a;
+    uiA64 = uA.s.signExp;
+    uiA0  = uA.s.signif;
+    uB.f = b;
+    uiB64 = uB.s.signExp;
+    uiB0  = uB.s.signif;
+    if ( isNaNExtF80UI( uiA64, uiA0 ) || isNaNExtF80UI( uiB64, uiB0 ) ) {
+        softfloat_raiseFlags( state, softfloat_flag_invalid );
+        return false;
+    }
+    signA = signExtF80UI64( uiA64 );
+    signB = signExtF80UI64( uiB64 );
+    return
+        (signA != signB)
+            ? signA || ! (((uiA64 | uiB64) & 0x7FFF) | uiA0 | uiB0)
+            : ((uiA64 == uiB64) && (uiA0 == uiB0))
+                  || (signA ^ softfloat_lt128( uiA64, uiA0, uiB64, uiB0 ));
+
+}
+
@@ -42,7 +42,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "softfloat.h"

 FEXCORE_PRESERVE_ALL_ATTR
-bool extF80_lt( extFloat80_t a, extFloat80_t b )
+bool extF80_lt( struct softfloat_state *state, extFloat80_t a, extFloat80_t b )
 {
    union { struct extFloat80M s; extFloat80_t f; } uA;
    uint_fast16_t uiA64;
@@ -59,7 +59,7 @@ bool extF80_lt( extFloat80_t a, extFloat80_t b )
    uiB64 = uB.s.signExp;
    uiB0  = uB.s.signif;
    if ( isNaNExtF80UI( uiA64, uiA0 ) || isNaNExtF80UI( uiB64, uiB0 ) ) {
-        softfloat_raiseFlags( softfloat_flag_invalid );
+        softfloat_raiseFlags( state, softfloat_flag_invalid );
        return false;
    }
    signA = signExtF80UI64( uiA64 );
@@ -42,7 +42,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "softfloat.h"

 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t extF80_mul( extFloat80_t a, extFloat80_t b )
+extFloat80_t extF80_mul( struct softfloat_state *state, extFloat80_t a, extFloat80_t b )
 {
    union { struct extFloat80M s; extFloat80_t f; } uA;
    uint_fast16_t uiA64;
@@ -125,11 +125,11 @@ extFloat80_t extF80_mul( extFloat80_t a, extFloat80_t b )
    }
    return
        softfloat_roundPackToExtF80(
-            signZ, expZ, sig128Z.v64, sig128Z.v0, extF80_roundingPrecision );
+            state, signZ, expZ, sig128Z.v64, sig128Z.v0, state->roundingPrecision );
    /*------------------------------------------------------------------------
    *------------------------------------------------------------------------*/
 propagateNaN:
-    uiZ = softfloat_propagateNaNExtF80UI( uiA64, uiA0, uiB64, uiB0 );
+    uiZ = softfloat_propagateNaNExtF80UI( state, uiA64, uiA0, uiB64, uiB0 );
    uiZ64 = uiZ.v64;
    uiZ0  = uiZ.v0;
    goto uiZ;
@@ -137,7 +137,7 @@ extFloat80_t extF80_mul( extFloat80_t a, extFloat80_t b )
    *------------------------------------------------------------------------*/
 infArg:
    if ( ! magBits ) {
-        softfloat_raiseFlags( softfloat_flag_invalid );
+        softfloat_raiseFlags( state, softfloat_flag_invalid );
        uiZ64 = defaultNaNExtF80UI64;
        uiZ0  = defaultNaNExtF80UI0;
    } else {
@@ -42,7 +42,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "softfloat.h"

 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t extF80_rem( extFloat80_t a, extFloat80_t b )
+extFloat80_t extF80_rem( struct softfloat_state *state, extFloat80_t a, extFloat80_t b )
 {
    union { struct extFloat80M s; extFloat80_t f; } uA;
    uint_fast16_t uiA64;
@@ -193,18 +193,18 @@ extFloat80_t extF80_rem( extFloat80_t a, extFloat80_t b )
    }
    return
        softfloat_normRoundPackToExtF80(
-            signRem, rem.v64 | rem.v0 ? expB + 32 : 0, rem.v64, rem.v0, 80 );
+            state, signRem, rem.v64 | rem.v0 ? expB + 32 : 0, rem.v64, rem.v0, 80 );
    /*------------------------------------------------------------------------
    *------------------------------------------------------------------------*/
 propagateNaN:
-    uiZ = softfloat_propagateNaNExtF80UI( uiA64, uiA0, uiB64, uiB0 );
+    uiZ = softfloat_propagateNaNExtF80UI( state, uiA64, uiA0, uiB64, uiB0 );
    uiZ64 = uiZ.v64;
    uiZ0  = uiZ.v0;
    goto uiZ;
    /*------------------------------------------------------------------------
    *------------------------------------------------------------------------*/
 invalid:
-    softfloat_raiseFlags( softfloat_flag_invalid );
+    softfloat_raiseFlags( state, softfloat_flag_invalid );
    uiZ64 = defaultNaNExtF80UI64;
    uiZ0  = defaultNaNExtF80UI0;
    goto uiZ;
@@ -43,7 +43,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

 FEXCORE_PRESERVE_ALL_ATTR
 extFloat80_t
- extF80_roundToInt( extFloat80_t a, uint_fast8_t roundingMode, bool exact )
+ extF80_roundToInt( struct softfloat_state *state, extFloat80_t a, uint_fast8_t roundingMode, bool exact )
 {
    union { struct extFloat80M s; extFloat80_t f; } uA;
    uint_fast16_t uiA64, signUI64;
@@ -80,7 +80,7 @@ extFloat80_t
    if ( 0x403E <= exp ) {
        if ( exp == 0x7FFF ) {
            if ( sigA & UINT64_C( 0x7FFFFFFFFFFFFFFF ) ) {
-                uiZ = softfloat_propagateNaNExtF80UI( uiA64, sigA, 0, 0 );
+                uiZ = softfloat_propagateNaNExtF80UI( state, uiA64, sigA, 0, 0 );
                uiZ64 = uiZ.v64;
                sigZ  = uiZ.v0;
                goto uiZ;
@@ -93,7 +93,7 @@ extFloat80_t
        goto uiZ;
    }
    if ( exp <= 0x3FFE ) {
-        if ( exact ) softfloat_exceptionFlags |= softfloat_flag_inexact;
+        if ( exact ) state->exceptionFlags |= softfloat_flag_inexact;
        switch ( roundingMode ) {
         case softfloat_round_near_even:
            if ( !(sigA & UINT64_C( 0x7FFFFFFFFFFFFFFF )) ) break;
@@ -145,7 +145,7 @@ extFloat80_t
 #ifdef SOFTFLOAT_ROUND_ODD
        if ( roundingMode == softfloat_round_odd ) sigZ |= lastBitMask;
 #endif
-        if ( exact ) softfloat_exceptionFlags |= softfloat_flag_inexact;
+        if ( exact ) state->exceptionFlags |= softfloat_flag_inexact;
    }
 uiZ:
    uZ.s.signExp = uiZ64;
@@ -42,7 +42,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "softfloat.h"

 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t extF80_sqrt( extFloat80_t a )
+extFloat80_t extF80_sqrt( struct softfloat_state *state, extFloat80_t a )
 {
    union { struct extFloat80M s; extFloat80_t f; } uA;
    uint_fast16_t uiA64;
@@ -74,7 +74,7 @@ extFloat80_t extF80_sqrt( extFloat80_t a )
    *------------------------------------------------------------------------*/
    if ( expA == 0x7FFF ) {
        if ( sigA & UINT64_C( 0x7FFFFFFFFFFFFFFF ) ) {
-            uiZ = softfloat_propagateNaNExtF80UI( uiA64, uiA0, 0, 0 );
+            uiZ = softfloat_propagateNaNExtF80UI( state, uiA64, uiA0, 0, 0 );
            uiZ64 = uiZ.v64;
            uiZ0  = uiZ.v0;
            goto uiZ;
@@ -155,11 +155,11 @@ extFloat80_t extF80_sqrt( extFloat80_t a )
    }
    return
        softfloat_roundPackToExtF80(
-            0, expZ, sigZ, sigZExtra, extF80_roundingPrecision );
+            state, 0, expZ, sigZ, sigZExtra, state->roundingPrecision );
    /*------------------------------------------------------------------------
    *------------------------------------------------------------------------*/
 invalid:
-    softfloat_raiseFlags( softfloat_flag_invalid );
+    softfloat_raiseFlags( state, softfloat_flag_invalid );
    uiZ64 = defaultNaNExtF80UI64;
    uiZ0  = defaultNaNExtF80UI0;
    goto uiZ;
@@ -41,7 +41,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "softfloat.h"

 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t extF80_sub( extFloat80_t a, extFloat80_t b )
+extFloat80_t extF80_sub( struct softfloat_state *state, extFloat80_t a, extFloat80_t b )
 {
    union { struct extFloat80M s; extFloat80_t f; } uA;
    uint_fast16_t uiA64;
@@ -54,7 +54,7 @@ extFloat80_t extF80_sub( extFloat80_t a, extFloat80_t b )
 #if ! defined INLINE_LEVEL || (INLINE_LEVEL < 2)
    extFloat80_t
        (*magsFuncPtr)(
-            uint_fast16_t, uint_fast64_t, uint_fast16_t, uint_fast64_t, bool );
+            struct softfloat_state *, uint_fast16_t, uint_fast64_t, uint_fast16_t, uint_fast64_t, bool );
 #endif

    uA.f = a;
@@ -67,14 +67,14 @@ extFloat80_t extF80_sub( extFloat80_t a, extFloat80_t b )
    signB = signExtF80UI64( uiB64 );
 #if defined INLINE_LEVEL && (2 <= INLINE_LEVEL)
    if ( signA == signB ) {
-        return softfloat_subMagsExtF80( uiA64, uiA0, uiB64, uiB0, signA );
+        return softfloat_subMagsExtF80( state, uiA64, uiA0, uiB64, uiB0, signA );
    } else {
-        return softfloat_addMagsExtF80( uiA64, uiA0, uiB64, uiB0, signA );
+        return softfloat_addMagsExtF80( state, uiA64, uiA0, uiB64, uiB0, signA );
    }
 #else
    magsFuncPtr =
        (signA == signB) ? softfloat_subMagsExtF80 : softfloat_addMagsExtF80;
-    return (*magsFuncPtr)( uiA64, uiA0, uiB64, uiB0, signA );
+    return (*magsFuncPtr)( state, uiA64, uiA0, uiB64, uiB0, signA );
 #endif

 }
@@ -42,7 +42,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "softfloat.h"

 FEXCORE_PRESERVE_ALL_ATTR
-float128_t extF80_to_f128( extFloat80_t a )
+float128_t extF80_to_f128( struct softfloat_state *state, extFloat80_t a )
 {
    union { struct extFloat80M s; extFloat80_t f; } uA;
    uint_fast16_t uiA64;
@@ -61,7 +61,7 @@ float128_t extF80_to_f128( extFloat80_t a )
    exp = expExtF80UI64( uiA64 );
    frac = uiA0 & UINT64_C( 0x7FFFFFFFFFFFFFFF );
    if ( (exp == 0x7FFF) && frac ) {
-        softfloat_extF80UIToCommonNaN( uiA64, uiA0, &commonNaN );
+        softfloat_extF80UIToCommonNaN( state, uiA64, uiA0, &commonNaN );
        uiZ = softfloat_commonNaNToF128UI( &commonNaN );
    } else {
        sign = signExtF80UI64( uiA64 );
@@ -42,7 +42,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "softfloat.h"

 FEXCORE_PRESERVE_ALL_ATTR
-float32_t extF80_to_f32( extFloat80_t a )
+float32_t extF80_to_f32( struct softfloat_state *state, extFloat80_t a )
 {
    union { struct extFloat80M s; extFloat80_t f; } uA;
    uint_fast16_t uiA64;
@@ -66,7 +66,7 @@ float32_t extF80_to_f32( extFloat80_t a )
    *------------------------------------------------------------------------*/
    if ( exp == 0x7FFF ) {
        if ( sig & UINT64_C( 0x7FFFFFFFFFFFFFFF ) ) {
-            softfloat_extF80UIToCommonNaN( uiA64, uiA0, &commonNaN );
+            softfloat_extF80UIToCommonNaN( state, uiA64, uiA0, &commonNaN );
            uiZ = softfloat_commonNaNToF32UI( &commonNaN );
        } else {
            uiZ = packToF32UI( sign, 0xFF, 0 );
@@ -86,7 +86,7 @@ float32_t extF80_to_f32( extFloat80_t a )
    if ( sizeof (int_fast16_t) < sizeof (int_fast32_t) ) {
        if ( exp < -0x1000 ) exp = -0x1000;
    }
-    return softfloat_roundPackToF32( sign, exp, sig32 );
+    return softfloat_roundPackToF32( state, sign, exp, sig32 );
    /*------------------------------------------------------------------------
    *------------------------------------------------------------------------*/
 uiZ:
@@ -42,7 +42,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "softfloat.h"

 FEXCORE_PRESERVE_ALL_ATTR
-float64_t extF80_to_f64( extFloat80_t a )
+float64_t extF80_to_f64( struct softfloat_state *state, extFloat80_t a )
 {
    union { struct extFloat80M s; extFloat80_t f; } uA;
    uint_fast16_t uiA64;
@@ -72,7 +72,7 @@ float64_t extF80_to_f64( extFloat80_t a )
    *------------------------------------------------------------------------*/
    if ( exp == 0x7FFF ) {
        if ( sig & UINT64_C( 0x7FFFFFFFFFFFFFFF ) ) {
-            softfloat_extF80UIToCommonNaN( uiA64, uiA0, &commonNaN );
+            softfloat_extF80UIToCommonNaN( state, uiA64, uiA0, &commonNaN );
            uiZ = softfloat_commonNaNToF64UI( &commonNaN );
        } else {
            uiZ = packToF64UI( sign, 0x7FF, 0 );
@@ -86,7 +86,7 @@ float64_t extF80_to_f64( extFloat80_t a )
    if ( sizeof (int_fast16_t) < sizeof (int_fast32_t) ) {
        if ( exp < -0x1000 ) exp = -0x1000;
    }
-    return softfloat_roundPackToF64( sign, exp, sig );
+    return softfloat_roundPackToF64( state, sign, exp, sig );
    /*------------------------------------------------------------------------
    *------------------------------------------------------------------------*/
 uiZ:
@@ -43,7 +43,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

 FEXCORE_PRESERVE_ALL_ATTR
 int_fast32_t
- extF80_to_i32( extFloat80_t a, uint_fast8_t roundingMode, bool exact )
+ extF80_to_i32( struct softfloat_state *state, extFloat80_t a, uint_fast8_t roundingMode, bool exact )
 {
    union { struct extFloat80M s; extFloat80_t f; } uA;
    uint_fast16_t uiA64;
@@ -68,7 +68,7 @@ int_fast32_t
 #elif (i32_fromNaN == i32_fromNegOverflow)
        sign = 1;
 #else
-        softfloat_raiseFlags( softfloat_flag_invalid );
+        softfloat_raiseFlags( state, softfloat_flag_invalid );
        return i32_fromNaN;
 #endif
    }
@@ -78,7 +78,7 @@ int_fast32_t
    shiftDist = 0x4032 - exp;
    if ( shiftDist <= 0 ) shiftDist = 1;
    sig = softfloat_shiftRightJam64( sig, shiftDist );
-    return softfloat_roundToI32( sign, sig, roundingMode, exact );
+    return softfloat_roundToI32( state, sign, sig, roundingMode, exact );

 }

@@ -43,7 +43,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

 FEXCORE_PRESERVE_ALL_ATTR
 int_fast64_t
- extF80_to_i64( extFloat80_t a, uint_fast8_t roundingMode, bool exact )
+ extF80_to_i64( struct softfloat_state *state, extFloat80_t a, uint_fast8_t roundingMode, bool exact )
 {
    union { struct extFloat80M s; extFloat80_t f; } uA;
    uint_fast16_t uiA64;
@@ -68,7 +68,7 @@ int_fast64_t
        /*--------------------------------------------------------------------
        *--------------------------------------------------------------------*/
        if ( shiftDist ) {
-            softfloat_raiseFlags( softfloat_flag_invalid );
+            softfloat_raiseFlags( state, softfloat_flag_invalid );
            return
                (exp == 0x7FFF) && (sig & UINT64_C( 0x7FFFFFFFFFFFFFFF ))
                    ? i64_fromNaN
@@ -84,7 +84,7 @@ int_fast64_t
        sig = sig64Extra.v;
        sigExtra = sig64Extra.extra;
    }
-    return softfloat_roundToI64( sign, sig, sigExtra, roundingMode, exact );
+    return softfloat_roundToI64( state, sign, sig, sigExtra, roundingMode, exact );

 }

@@ -43,7 +43,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

 FEXCORE_PRESERVE_ALL_ATTR
 uint_fast64_t
- extF80_to_ui64( extFloat80_t a, uint_fast8_t roundingMode, bool exact )
+ extF80_to_ui64( struct softfloat_state *state, extFloat80_t a, uint_fast8_t roundingMode, bool exact )
 {
    union { struct extFloat80M s; extFloat80_t f; } uA;
    uint_fast16_t uiA64;
@@ -65,7 +65,7 @@ uint_fast64_t
    *------------------------------------------------------------------------*/
    shiftDist = 0x403E - exp;
    if ( shiftDist < 0 ) {
-        softfloat_raiseFlags( softfloat_flag_invalid );
+        softfloat_raiseFlags( state, softfloat_flag_invalid );
        return
            (exp == 0x7FFF) && (sig & UINT64_C( 0x7FFFFFFFFFFFFFFF ))
                ? ui64_fromNaN
@@ -79,7 +79,7 @@ uint_fast64_t
        sig = sig64Extra.v;
        sigExtra = sig64Extra.extra;
    }
-    return softfloat_roundToUI64( sign, sig, sigExtra, roundingMode, exact );
+    return softfloat_roundToUI64( state, sign, sig, sigExtra, roundingMode, exact );

 }

@@ -0,0 +1,78 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "softfloat.h"
+
+float128_t f128_add( struct softfloat_state *state, float128_t a, float128_t b )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    bool signA;
+    union ui128_f128 uB;
+    uint_fast64_t uiB64, uiB0;
+    bool signB;
+#if ! defined INLINE_LEVEL || (INLINE_LEVEL < 2)
+    float128_t
+        (*magsFuncPtr)(
+            uint_fast64_t, uint_fast64_t, uint_fast64_t, uint_fast64_t, bool );
+#endif
+
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    signA = signF128UI64( uiA64 );
+    uB.f = b;
+    uiB64 = uB.ui.v64;
+    uiB0  = uB.ui.v0;
+    signB = signF128UI64( uiB64 );
+#if defined INLINE_LEVEL && (2 <= INLINE_LEVEL)
+    if ( signA == signB ) {
+        return softfloat_addMagsF128( state, uiA64, uiA0, uiB64, uiB0, signA );
+    } else {
+        return softfloat_subMagsF128( state, uiA64, uiA0, uiB64, uiB0, signA );
+    }
+#else
+    magsFuncPtr =
+        (signA == signB) ? softfloat_addMagsF128 : softfloat_subMagsF128;
+    return (*magsFuncPtr)( uiA64, uiA0, uiB64, uiB0, signA );
+#endif
+
+}
+
@@ -0,0 +1,199 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+float128_t f128_div( struct softfloat_state *state, float128_t a, float128_t b )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    bool signA;
+    int_fast32_t expA;
+    struct uint128 sigA;
+    union ui128_f128 uB;
+    uint_fast64_t uiB64, uiB0;
+    bool signB;
+    int_fast32_t expB;
+    struct uint128 sigB;
+    bool signZ;
+    struct exp32_sig128 normExpSig;
+    int_fast32_t expZ;
+    struct uint128 rem;
+    uint_fast32_t recip32;
+    int ix;
+    uint_fast64_t q64;
+    uint_fast32_t q;
+    struct uint128 term;
+    uint_fast32_t qs[3];
+    uint_fast64_t sigZExtra;
+    struct uint128 sigZ, uiZ;
+    union ui128_f128 uZ;
+
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    signA = signF128UI64( uiA64 );
+    expA  = expF128UI64( uiA64 );
+    sigA.v64 = fracF128UI64( uiA64 );
+    sigA.v0  = uiA0;
+    uB.f = b;
+    uiB64 = uB.ui.v64;
+    uiB0  = uB.ui.v0;
+    signB = signF128UI64( uiB64 );
+    expB  = expF128UI64( uiB64 );
+    sigB.v64 = fracF128UI64( uiB64 );
+    sigB.v0  = uiB0;
+    signZ = signA ^ signB;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( expA == 0x7FFF ) {
+        if ( sigA.v64 | sigA.v0 ) goto propagateNaN;
+        if ( expB == 0x7FFF ) {
+            if ( sigB.v64 | sigB.v0 ) goto propagateNaN;
+            goto invalid;
+        }
+        goto infinity;
+    }
+    if ( expB == 0x7FFF ) {
+        if ( sigB.v64 | sigB.v0 ) goto propagateNaN;
+        goto zero;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( ! expB ) {
+        if ( ! (sigB.v64 | sigB.v0) ) {
+            if ( ! (expA | sigA.v64 | sigA.v0) ) goto invalid;
+            softfloat_raiseFlags( state, softfloat_flag_infinite );
+            goto infinity;
+        }
+        normExpSig = softfloat_normSubnormalF128Sig( sigB.v64, sigB.v0 );
+        expB = normExpSig.exp;
+        sigB = normExpSig.sig;
+    }
+    if ( ! expA ) {
+        if ( ! (sigA.v64 | sigA.v0) ) goto zero;
+        normExpSig = softfloat_normSubnormalF128Sig( sigA.v64, sigA.v0 );
+        expA = normExpSig.exp;
+        sigA = normExpSig.sig;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    expZ = expA - expB + 0x3FFE;
+    sigA.v64 |= UINT64_C( 0x0001000000000000 );
+    sigB.v64 |= UINT64_C( 0x0001000000000000 );
+    rem = sigA;
+    if ( softfloat_lt128( sigA.v64, sigA.v0, sigB.v64, sigB.v0 ) ) {
+        --expZ;
+        rem = softfloat_add128( sigA.v64, sigA.v0, sigA.v64, sigA.v0 );
+    }
+    recip32 = softfloat_approxRecip32_1( sigB.v64>>17 );
+    ix = 3;
+    for (;;) {
+        q64 = (uint_fast64_t) (uint32_t) (rem.v64>>19) * recip32;
+        q = (q64 + 0x80000000)>>32;
+        --ix;
+        if ( ix < 0 ) break;
+        rem = softfloat_shortShiftLeft128( rem.v64, rem.v0, 29 );
+        term = softfloat_mul128By32( sigB.v64, sigB.v0, q );
+        rem = softfloat_sub128( rem.v64, rem.v0, term.v64, term.v0 );
+        if ( rem.v64 & UINT64_C( 0x8000000000000000 ) ) {
+            --q;
+            rem = softfloat_add128( rem.v64, rem.v0, sigB.v64, sigB.v0 );
+        }
+        qs[ix] = q;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( ((q + 1) & 7) < 2 ) {
+        rem = softfloat_shortShiftLeft128( rem.v64, rem.v0, 29 );
+        term = softfloat_mul128By32( sigB.v64, sigB.v0, q );
+        rem = softfloat_sub128( rem.v64, rem.v0, term.v64, term.v0 );
+        if ( rem.v64 & UINT64_C( 0x8000000000000000 ) ) {
+            --q;
+            rem = softfloat_add128( rem.v64, rem.v0, sigB.v64, sigB.v0 );
+        } else if ( softfloat_le128( sigB.v64, sigB.v0, rem.v64, rem.v0 ) ) {
+            ++q;
+            rem = softfloat_sub128( rem.v64, rem.v0, sigB.v64, sigB.v0 );
+        }
+        if ( rem.v64 | rem.v0 ) q |= 1;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    sigZExtra = (uint64_t) ((uint_fast64_t) q<<60);
+    term = softfloat_shortShiftLeft128( 0, qs[1], 54 );
+    sigZ =
+        softfloat_add128(
+            (uint_fast64_t) qs[2]<<19, ((uint_fast64_t) qs[0]<<25) + (q>>4),
+            term.v64, term.v0
+        );
+    return
+        softfloat_roundPackToF128( state, signZ, expZ, sigZ.v64, sigZ.v0, sigZExtra );
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+ propagateNaN:
+    uiZ = softfloat_propagateNaNF128UI( state, uiA64, uiA0, uiB64, uiB0 );
+    goto uiZ;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+ invalid:
+    softfloat_raiseFlags( state, softfloat_flag_invalid );
+    uiZ.v64 = defaultNaNF128UI64;
+    uiZ.v0  = defaultNaNF128UI0;
+    goto uiZ;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+ infinity:
+    uiZ.v64 = packToF128UI64( signZ, 0x7FFF, 0 );
+    goto uiZ0;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+ zero:
+    uiZ.v64 = packToF128UI64( signZ, 0, 0 );
+ uiZ0:
+    uiZ.v0 = 0;
+ uiZ:
+    uZ.ui = uiZ;
+    return uZ.f;
+
+}
+
@@ -0,0 +1,73 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+bool f128_eq( struct softfloat_state *state, float128_t a, float128_t b )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    union ui128_f128 uB;
+    uint_fast64_t uiB64, uiB0;
+
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    uB.f = b;
+    uiB64 = uB.ui.v64;
+    uiB0  = uB.ui.v0;
+    if ( isNaNF128UI( uiA64, uiA0 ) || isNaNF128UI( uiB64, uiB0 ) ) {
+        if (
+               softfloat_isSigNaNF128UI( uiA64, uiA0 )
+            || softfloat_isSigNaNF128UI( uiB64, uiB0 )
+        ) {
+            softfloat_raiseFlags( state, softfloat_flag_invalid );
+        }
+        return false;
+    }
+    return
+           (uiA0 == uiB0)
+        && (   (uiA64 == uiB64)
+            || (! uiA0 && ! ((uiA64 | uiB64) & UINT64_C( 0x7FFFFFFFFFFFFFFF )))
+           );
+
+}
+
@@ -0,0 +1,67 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "softfloat.h"
+
+bool f128_eq_signaling( struct softfloat_state *state, float128_t a, float128_t b )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    union ui128_f128 uB;
+    uint_fast64_t uiB64, uiB0;
+
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    uB.f = b;
+    uiB64 = uB.ui.v64;
+    uiB0  = uB.ui.v0;
+    if ( isNaNF128UI( uiA64, uiA0 ) || isNaNF128UI( uiB64, uiB0 ) ) {
+        softfloat_raiseFlags( state, softfloat_flag_invalid );
+        return false;
+    }
+    return
+           (uiA0 == uiB0)
+        && (   (uiA64 == uiB64)
+            || (! uiA0 && ! ((uiA64 | uiB64) & UINT64_C( 0x7FFFFFFFFFFFFFFF )))
+           );
+
+}
+
@@ -4,8 +4,8 @@
 This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
 Package, Release 3e, by John R. Hauser.

-Copyright 2011, 2012, 2013, 2014, 2015, 2016 The Regents of the University of
-California.  All Rights Reserved.
+Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
+All rights reserved.

 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are met:
@@ -34,19 +34,18 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

 =============================================================================*/

-#include <stdint.h>
+#include <stdbool.h>
 #include "platform.h"
 #include "internals.h"
 #include "specialize.h"
 #include "softfloat.h"

-#ifndef THREAD_LOCAL
-#define THREAD_LOCAL
-#endif
+bool f128_isSignalingNaN( float128_t a )
+{
+    union ui128_f128 uA;

-THREAD_LOCAL uint_fast8_t softfloat_roundingMode = softfloat_round_near_even;
-THREAD_LOCAL uint_fast8_t softfloat_detectTininess = init_detectTininess;
-THREAD_LOCAL uint_fast8_t softfloat_exceptionFlags = 0;
+    uA.f = a;
+    return softfloat_isSigNaNF128UI( uA.ui.v64, uA.ui.v0 );

-THREAD_LOCAL uint_fast8_t extF80_roundingPrecision = 80;
+}

@@ -0,0 +1,72 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "softfloat.h"
+
+bool f128_le( struct softfloat_state *state, float128_t a, float128_t b )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    union ui128_f128 uB;
+    uint_fast64_t uiB64, uiB0;
+    bool signA, signB;
+
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    uB.f = b;
+    uiB64 = uB.ui.v64;
+    uiB0  = uB.ui.v0;
+    if ( isNaNF128UI( uiA64, uiA0 ) || isNaNF128UI( uiB64, uiB0 ) ) {
+        softfloat_raiseFlags( state, softfloat_flag_invalid );
+        return false;
+    }
+    signA = signF128UI64( uiA64 );
+    signB = signF128UI64( uiB64 );
+    return
+        (signA != signB)
+            ? signA
+                  || ! (((uiA64 | uiB64) & UINT64_C( 0x7FFFFFFFFFFFFFFF ))
+                            | uiA0 | uiB0)
+            : ((uiA64 == uiB64) && (uiA0 == uiB0))
+                  || (signA ^ softfloat_lt128( uiA64, uiA0, uiB64, uiB0 ));
+
+}
+
@@ -0,0 +1,78 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+bool f128_le_quiet( struct softfloat_state *state, float128_t a, float128_t b )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    union ui128_f128 uB;
+    uint_fast64_t uiB64, uiB0;
+    bool signA, signB;
+
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    uB.f = b;
+    uiB64 = uB.ui.v64;
+    uiB0  = uB.ui.v0;
+    if ( isNaNF128UI( uiA64, uiA0 ) || isNaNF128UI( uiB64, uiB0 ) ) {
+        if (
+               softfloat_isSigNaNF128UI( uiA64, uiA0 )
+            || softfloat_isSigNaNF128UI( uiB64, uiB0 )
+        ) {
+            softfloat_raiseFlags( state, softfloat_flag_invalid );
+        }
+        return false;
+    }
+    signA = signF128UI64( uiA64 );
+    signB = signF128UI64( uiB64 );
+    return
+        (signA != signB)
+            ? signA
+                  || ! (((uiA64 | uiB64) & UINT64_C( 0x7FFFFFFFFFFFFFFF ))
+                            | uiA0 | uiB0)
+            : ((uiA64 == uiB64) && (uiA0 == uiB0))
+                  || (signA ^ softfloat_lt128( uiA64, uiA0, uiB64, uiB0 ));
+
+}
+
@@ -0,0 +1,72 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "softfloat.h"
+
+bool f128_lt( struct softfloat_state *state, float128_t a, float128_t b )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    union ui128_f128 uB;
+    uint_fast64_t uiB64, uiB0;
+    bool signA, signB;
+
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    uB.f = b;
+    uiB64 = uB.ui.v64;
+    uiB0  = uB.ui.v0;
+    if ( isNaNF128UI( uiA64, uiA0 ) || isNaNF128UI( uiB64, uiB0 ) ) {
+        softfloat_raiseFlags( state, softfloat_flag_invalid );
+        return false;
+    }
+    signA = signF128UI64( uiA64 );
+    signB = signF128UI64( uiB64 );
+    return
+        (signA != signB)
+            ? signA
+                  && (((uiA64 | uiB64) & UINT64_C( 0x7FFFFFFFFFFFFFFF ))
+                          | uiA0 | uiB0)
+            : ((uiA64 != uiB64) || (uiA0 != uiB0))
+                  && (signA ^ softfloat_lt128( uiA64, uiA0, uiB64, uiB0 ));
+
+}
+
@@ -0,0 +1,78 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+bool f128_lt_quiet( struct softfloat_state *state, float128_t a, float128_t b )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    union ui128_f128 uB;
+    uint_fast64_t uiB64, uiB0;
+    bool signA, signB;
+
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    uB.f = b;
+    uiB64 = uB.ui.v64;
+    uiB0  = uB.ui.v0;
+    if ( isNaNF128UI( uiA64, uiA0 ) || isNaNF128UI( uiB64, uiB0 ) ) {
+        if (
+               softfloat_isSigNaNF128UI( uiA64, uiA0 )
+            || softfloat_isSigNaNF128UI( uiB64, uiB0 )
+        ) {
+            softfloat_raiseFlags( state, softfloat_flag_invalid );
+        }
+        return false;
+    }
+    signA = signF128UI64( uiA64 );
+    signB = signF128UI64( uiB64 );
+    return
+        (signA != signB)
+            ? signA
+                  && (((uiA64 | uiB64) & UINT64_C( 0x7FFFFFFFFFFFFFFF ))
+                          | uiA0 | uiB0)
+            : ((uiA64 != uiB64) || (uiA0 != uiB0))
+                  && (signA ^ softfloat_lt128( uiA64, uiA0, uiB64, uiB0 ));
+
+}
+
@@ -0,0 +1,163 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+float128_t f128_mul( struct softfloat_state *state, float128_t a, float128_t b )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    bool signA;
+    int_fast32_t expA;
+    struct uint128 sigA;
+    union ui128_f128 uB;
+    uint_fast64_t uiB64, uiB0;
+    bool signB;
+    int_fast32_t expB;
+    struct uint128 sigB;
+    bool signZ;
+    uint_fast64_t magBits;
+    struct exp32_sig128 normExpSig;
+    int_fast32_t expZ;
+    uint64_t sig256Z[4];
+    uint_fast64_t sigZExtra;
+    struct uint128 sigZ;
+    struct uint128_extra sig128Extra;
+    struct uint128 uiZ;
+    union ui128_f128 uZ;
+
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    signA = signF128UI64( uiA64 );
+    expA  = expF128UI64( uiA64 );
+    sigA.v64 = fracF128UI64( uiA64 );
+    sigA.v0  = uiA0;
+    uB.f = b;
+    uiB64 = uB.ui.v64;
+    uiB0  = uB.ui.v0;
+    signB = signF128UI64( uiB64 );
+    expB  = expF128UI64( uiB64 );
+    sigB.v64 = fracF128UI64( uiB64 );
+    sigB.v0  = uiB0;
+    signZ = signA ^ signB;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( expA == 0x7FFF ) {
+        if (
+            (sigA.v64 | sigA.v0) || ((expB == 0x7FFF) && (sigB.v64 | sigB.v0))
+        ) {
+            goto propagateNaN;
+        }
+        magBits = expB | sigB.v64 | sigB.v0;
+        goto infArg;
+    }
+    if ( expB == 0x7FFF ) {
+        if ( sigB.v64 | sigB.v0 ) goto propagateNaN;
+        magBits = expA | sigA.v64 | sigA.v0;
+        goto infArg;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( ! expA ) {
+        if ( ! (sigA.v64 | sigA.v0) ) goto zero;
+        normExpSig = softfloat_normSubnormalF128Sig( sigA.v64, sigA.v0 );
+        expA = normExpSig.exp;
+        sigA = normExpSig.sig;
+    }
+    if ( ! expB ) {
+        if ( ! (sigB.v64 | sigB.v0) ) goto zero;
+        normExpSig = softfloat_normSubnormalF128Sig( sigB.v64, sigB.v0 );
+        expB = normExpSig.exp;
+        sigB = normExpSig.sig;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    expZ = expA + expB - 0x4000;
+    sigA.v64 |= UINT64_C( 0x0001000000000000 );
+    sigB = softfloat_shortShiftLeft128( sigB.v64, sigB.v0, 16 );
+    softfloat_mul128To256M( sigA.v64, sigA.v0, sigB.v64, sigB.v0, sig256Z );
+    sigZExtra = sig256Z[indexWord( 4, 1 )] | (sig256Z[indexWord( 4, 0 )] != 0);
+    sigZ =
+        softfloat_add128(
+            sig256Z[indexWord( 4, 3 )], sig256Z[indexWord( 4, 2 )],
+            sigA.v64, sigA.v0
+        );
+    if ( UINT64_C( 0x0002000000000000 ) <= sigZ.v64 ) {
+        ++expZ;
+        sig128Extra =
+            softfloat_shortShiftRightJam128Extra(
+                sigZ.v64, sigZ.v0, sigZExtra, 1 );
+        sigZ = sig128Extra.v;
+        sigZExtra = sig128Extra.extra;
+    }
+    return
+        softfloat_roundPackToF128( state, signZ, expZ, sigZ.v64, sigZ.v0, sigZExtra );
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+ propagateNaN:
+    uiZ = softfloat_propagateNaNF128UI( state, uiA64, uiA0, uiB64, uiB0 );
+    goto uiZ;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+ infArg:
+    if ( ! magBits ) {
+        softfloat_raiseFlags( state, softfloat_flag_invalid );
+        uiZ.v64 = defaultNaNF128UI64;
+        uiZ.v0  = defaultNaNF128UI0;
+        goto uiZ;
+    }
+    uiZ.v64 = packToF128UI64( signZ, 0x7FFF, 0 );
+    goto uiZ0;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+ zero:
+    uiZ.v64 = packToF128UI64( signZ, 0, 0 );
+ uiZ0:
+    uiZ.v0 = 0;
+ uiZ:
+    uZ.ui = uiZ;
+    return uZ.f;
+
+}
+
@@ -0,0 +1,63 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "softfloat.h"
+
+float128_t f128_mulAdd( struct softfloat_state *state, float128_t a, float128_t b, float128_t c )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    union ui128_f128 uB;
+    uint_fast64_t uiB64, uiB0;
+    union ui128_f128 uC;
+    uint_fast64_t uiC64, uiC0;
+
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    uB.f = b;
+    uiB64 = uB.ui.v64;
+    uiB0  = uB.ui.v0;
+    uC.f = c;
+    uiC64 = uC.ui.v64;
+    uiC0  = uC.ui.v0;
+    return softfloat_mulAddF128( uiA64, uiA0, uiB64, uiB0, uiC64, uiC0, 0 );
+
+}
+
@@ -0,0 +1,190 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014, 2015, 2016 The Regents of the University of
+California.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+float128_t f128_rem( struct softfloat_state *state, float128_t a, float128_t b )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    bool signA;
+    int_fast32_t expA;
+    struct uint128 sigA;
+    union ui128_f128 uB;
+    uint_fast64_t uiB64, uiB0;
+    int_fast32_t expB;
+    struct uint128 sigB;
+    struct exp32_sig128 normExpSig;
+    struct uint128 rem;
+    int_fast32_t expDiff;
+    uint_fast32_t q, recip32;
+    uint_fast64_t q64;
+    struct uint128 term, altRem, meanRem;
+    bool signRem;
+    struct uint128 uiZ;
+    union ui128_f128 uZ;
+
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    signA = signF128UI64( uiA64 );
+    expA  = expF128UI64( uiA64 );
+    sigA.v64 = fracF128UI64( uiA64 );
+    sigA.v0  = uiA0;
+    uB.f = b;
+    uiB64 = uB.ui.v64;
+    uiB0  = uB.ui.v0;
+    expB  = expF128UI64( uiB64 );
+    sigB.v64 = fracF128UI64( uiB64 );
+    sigB.v0  = uiB0;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( expA == 0x7FFF ) {
+        if (
+            (sigA.v64 | sigA.v0) || ((expB == 0x7FFF) && (sigB.v64 | sigB.v0))
+        ) {
+            goto propagateNaN;
+        }
+        goto invalid;
+    }
+    if ( expB == 0x7FFF ) {
+        if ( sigB.v64 | sigB.v0 ) goto propagateNaN;
+        return a;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( ! expB ) {
+        if ( ! (sigB.v64 | sigB.v0) ) goto invalid;
+        normExpSig = softfloat_normSubnormalF128Sig( sigB.v64, sigB.v0 );
+        expB = normExpSig.exp;
+        sigB = normExpSig.sig;
+    }
+    if ( ! expA ) {
+        if ( ! (sigA.v64 | sigA.v0) ) return a;
+        normExpSig = softfloat_normSubnormalF128Sig( sigA.v64, sigA.v0 );
+        expA = normExpSig.exp;
+        sigA = normExpSig.sig;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    sigA.v64 |= UINT64_C( 0x0001000000000000 );
+    sigB.v64 |= UINT64_C( 0x0001000000000000 );
+    rem = sigA;
+    expDiff = expA - expB;
+    if ( expDiff < 1 ) {
+        if ( expDiff < -1 ) return a;
+        if ( expDiff ) {
+            --expB;
+            sigB = softfloat_add128( sigB.v64, sigB.v0, sigB.v64, sigB.v0 );
+            q = 0;
+        } else {
+            q = softfloat_le128( sigB.v64, sigB.v0, rem.v64, rem.v0 );
+            if ( q ) {
+                rem = softfloat_sub128( rem.v64, rem.v0, sigB.v64, sigB.v0 );
+            }
+        }
+    } else {
+        recip32 = softfloat_approxRecip32_1( sigB.v64>>17 );
+        expDiff -= 30;
+        for (;;) {
+            q64 = (uint_fast64_t) (uint32_t) (rem.v64>>19) * recip32;
+            if ( expDiff < 0 ) break;
+            q = (q64 + 0x80000000)>>32;
+            rem = softfloat_shortShiftLeft128( rem.v64, rem.v0, 29 );
+            term = softfloat_mul128By32( sigB.v64, sigB.v0, q );
+            rem = softfloat_sub128( rem.v64, rem.v0, term.v64, term.v0 );
+            if ( rem.v64 & UINT64_C( 0x8000000000000000 ) ) {
+                rem = softfloat_add128( rem.v64, rem.v0, sigB.v64, sigB.v0 );
+            }
+            expDiff -= 29;
+        }
+        /*--------------------------------------------------------------------
+        | (`expDiff' cannot be less than -29 here.)
+        *--------------------------------------------------------------------*/
+        q = (uint32_t) (q64>>32)>>(~expDiff & 31);
+        rem = softfloat_shortShiftLeft128( rem.v64, rem.v0, expDiff + 30 );
+        term = softfloat_mul128By32( sigB.v64, sigB.v0, q );
+        rem = softfloat_sub128( rem.v64, rem.v0, term.v64, term.v0 );
+        if ( rem.v64 & UINT64_C( 0x8000000000000000 ) ) {
+            altRem = softfloat_add128( rem.v64, rem.v0, sigB.v64, sigB.v0 );
+            goto selectRem;
+        }
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    do {
+        altRem = rem;
+        ++q;
+        rem = softfloat_sub128( rem.v64, rem.v0, sigB.v64, sigB.v0 );
+    } while ( ! (rem.v64 & UINT64_C( 0x8000000000000000 )) );
+ selectRem:
+    meanRem = softfloat_add128( rem.v64, rem.v0, altRem.v64, altRem.v0 );
+    if (
+        (meanRem.v64 & UINT64_C( 0x8000000000000000 ))
+            || (! (meanRem.v64 | meanRem.v0) && (q & 1))
+    ) {
+        rem = altRem;
+    }
+    signRem = signA;
+    if ( rem.v64 & UINT64_C( 0x8000000000000000 ) ) {
+        signRem = ! signRem;
+        rem = softfloat_sub128( 0, 0, rem.v64, rem.v0 );
+    }
+    return softfloat_normRoundPackToF128( state, signRem, expB - 1, rem.v64, rem.v0 );
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+ propagateNaN:
+    uiZ = softfloat_propagateNaNF128UI( state, uiA64, uiA0, uiB64, uiB0 );
+    goto uiZ;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+ invalid:
+    softfloat_raiseFlags( state, softfloat_flag_invalid );
+    uiZ.v64 = defaultNaNF128UI64;
+    uiZ.v0  = defaultNaNF128UI0;
+ uiZ:
+    uZ.ui = uiZ;
+    return uZ.f;
+
+}
+
@@ -0,0 +1,201 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014, 2015, 2016, 2017 The Regents of the
+University of California.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+float128_t f128_sqrt( struct softfloat_state *state, float128_t a )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    bool signA;
+    int_fast32_t expA;
+    struct uint128 sigA, uiZ;
+    struct exp32_sig128 normExpSig;
+    int_fast32_t expZ;
+    uint_fast32_t sig32A, recipSqrt32, sig32Z;
+    struct uint128 rem;
+    uint32_t qs[3];
+    uint_fast32_t q;
+    uint_fast64_t x64, sig64Z;
+    struct uint128 y, term;
+    uint_fast64_t sigZExtra;
+    struct uint128 sigZ;
+    union ui128_f128 uZ;
+
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    signA = signF128UI64( uiA64 );
+    expA  = expF128UI64( uiA64 );
+    sigA.v64 = fracF128UI64( uiA64 );
+    sigA.v0  = uiA0;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( expA == 0x7FFF ) {
+        if ( sigA.v64 | sigA.v0 ) {
+            uiZ = softfloat_propagateNaNF128UI( state, uiA64, uiA0, 0, 0 );
+            goto uiZ;
+        }
+        if ( ! signA ) return a;
+        goto invalid;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( signA ) {
+        if ( ! (expA | sigA.v64 | sigA.v0) ) return a;
+        goto invalid;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( ! expA ) {
+        if ( ! (sigA.v64 | sigA.v0) ) return a;
+        normExpSig = softfloat_normSubnormalF128Sig( sigA.v64, sigA.v0 );
+        expA = normExpSig.exp;
+        sigA = normExpSig.sig;
+    }
+    /*------------------------------------------------------------------------
+    | (`sig32Z' is guaranteed to be a lower bound on the square root of
+    | `sig32A', which makes `sig32Z' also a lower bound on the square root of
+    | `sigA'.)
+    *------------------------------------------------------------------------*/
+    expZ = ((expA - 0x3FFF)>>1) + 0x3FFE;
+    expA &= 1;
+    sigA.v64 |= UINT64_C( 0x0001000000000000 );
+    sig32A = sigA.v64>>17;
+    recipSqrt32 = softfloat_approxRecipSqrt32_1( expA, sig32A );
+    sig32Z = ((uint_fast64_t) sig32A * recipSqrt32)>>32;
+    if ( expA ) {
+        sig32Z >>= 1;
+        rem = softfloat_shortShiftLeft128( sigA.v64, sigA.v0, 12 );
+    } else {
+        rem = softfloat_shortShiftLeft128( sigA.v64, sigA.v0, 13 );
+    }
+    qs[2] = sig32Z;
+    rem.v64 -= (uint_fast64_t) sig32Z * sig32Z;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    q = ((uint32_t) (rem.v64>>2) * (uint_fast64_t) recipSqrt32)>>32;
+    x64 = (uint_fast64_t) sig32Z<<32;
+    sig64Z = x64 + ((uint_fast64_t) q<<3);
+    y = softfloat_shortShiftLeft128( rem.v64, rem.v0, 29 );
+    /*------------------------------------------------------------------------
+    | (Repeating this loop is a rare occurrence.)
+    *------------------------------------------------------------------------*/
+    for (;;) {
+        term = softfloat_mul64ByShifted32To128( x64 + sig64Z, q );
+        rem = softfloat_sub128( y.v64, y.v0, term.v64, term.v0 );
+        if ( ! (rem.v64 & UINT64_C( 0x8000000000000000 )) ) break;
+        --q;
+        sig64Z -= 1<<3;
+    }
+    qs[1] = q;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    q = ((rem.v64>>2) * recipSqrt32)>>32;
+    y = softfloat_shortShiftLeft128( rem.v64, rem.v0, 29 );
+    sig64Z <<= 1;
+    /*------------------------------------------------------------------------
+    | (Repeating this loop is a rare occurrence.)
+    *------------------------------------------------------------------------*/
+    for (;;) {
+        term = softfloat_shortShiftLeft128( 0, sig64Z, 32 );
+        term = softfloat_add128( term.v64, term.v0, 0, (uint_fast64_t) q<<6 );
+        term = softfloat_mul128By32( term.v64, term.v0, q );
+        rem = softfloat_sub128( y.v64, y.v0, term.v64, term.v0 );
+        if ( ! (rem.v64 & UINT64_C( 0x8000000000000000 )) ) break;
+        --q;
+    }
+    qs[0] = q;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    q = (((rem.v64>>2) * recipSqrt32)>>32) + 2;
+    sigZExtra = (uint64_t) ((uint_fast64_t) q<<59);
+    term = softfloat_shortShiftLeft128( 0, qs[1], 53 );
+    sigZ =
+        softfloat_add128(
+            (uint_fast64_t) qs[2]<<18, ((uint_fast64_t) qs[0]<<24) + (q>>5),
+            term.v64, term.v0
+        );
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( (q & 0xF) <= 2 ) {
+        q &= ~3;
+        sigZExtra = (uint64_t) ((uint_fast64_t) q<<59);
+        y = softfloat_shortShiftLeft128( sigZ.v64, sigZ.v0, 6 );
+        y.v0 |= sigZExtra>>58;
+        term = softfloat_sub128( y.v64, y.v0, 0, q );
+        y    = softfloat_mul64ByShifted32To128( term.v0,  q );
+        term = softfloat_mul64ByShifted32To128( term.v64, q );
+        term = softfloat_add128( term.v64, term.v0, 0, y.v64 );
+        rem = softfloat_shortShiftLeft128( rem.v64, rem.v0, 20 );
+        term = softfloat_sub128( term.v64, term.v0, rem.v64, rem.v0 );
+        /*--------------------------------------------------------------------
+        | The concatenation of `term' and `y.v0' is now the negative remainder
+        | (3 words altogether).
+        *--------------------------------------------------------------------*/
+        if ( term.v64 & UINT64_C( 0x8000000000000000 ) ) {
+            sigZExtra |= 1;
+        } else {
+            if ( term.v64 | term.v0 | y.v0 ) {
+                if ( sigZExtra ) {
+                    --sigZExtra;
+                } else {
+                    sigZ = softfloat_sub128( sigZ.v64, sigZ.v0, 0, 1 );
+                    sigZExtra = ~0;
+                }
+            }
+        }
+    }
+    return softfloat_roundPackToF128( state, 0, expZ, sigZ.v64, sigZ.v0, sigZExtra );
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+ invalid:
+    softfloat_raiseFlags( state, softfloat_flag_invalid );
+    uiZ.v64 = defaultNaNF128UI64;
+    uiZ.v0  = defaultNaNF128UI0;
+ uiZ:
+    uZ.ui = uiZ;
+    return uZ.f;
+
+}
+
@@ -0,0 +1,78 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014 The Regents of the University of California.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "softfloat.h"
+
+float128_t f128_sub( struct softfloat_state *state, float128_t a, float128_t b )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    bool signA;
+    union ui128_f128 uB;
+    uint_fast64_t uiB64, uiB0;
+    bool signB;
+#if ! defined INLINE_LEVEL || (INLINE_LEVEL < 2)
+    float128_t
+        (*magsFuncPtr)(
+            uint_fast64_t, uint_fast64_t, uint_fast64_t, uint_fast64_t, bool );
+#endif
+
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    signA = signF128UI64( uiA64 );
+    uB.f = b;
+    uiB64 = uB.ui.v64;
+    uiB0  = uB.ui.v0;
+    signB = signF128UI64( uiB64 );
+#if defined INLINE_LEVEL && (2 <= INLINE_LEVEL)
+    if ( signA == signB ) {
+        return softfloat_subMagsF128( state, uiA64, uiA0, uiB64, uiB0, signA );
+    } else {
+        return softfloat_addMagsF128( state, uiA64, uiA0, uiB64, uiB0, signA );
+    }
+#else
+    magsFuncPtr =
+        (signA == signB) ? softfloat_subMagsF128 : softfloat_addMagsF128;
+    return (*magsFuncPtr)( uiA64, uiA0, uiB64, uiB0, signA );
+#endif
+
+}
+
@@ -42,7 +42,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "softfloat.h"

 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t f128_to_extF80( float128_t a )
+extFloat80_t f128_to_extF80( struct softfloat_state *state, float128_t a )
 {
    union ui128_f128 uA;
    uint_fast64_t uiA64, uiA0;
@@ -70,7 +70,7 @@ extFloat80_t f128_to_extF80( float128_t a )
    *------------------------------------------------------------------------*/
    if ( exp == 0x7FFF ) {
        if ( frac64 | frac0 ) {
-            softfloat_f128UIToCommonNaN( uiA64, uiA0, &commonNaN );
+            softfloat_f128UIToCommonNaN( state, uiA64, uiA0, &commonNaN );
            uiZ = softfloat_commonNaNToExtF80UI( &commonNaN );
            uiZ64 = uiZ.v64;
            uiZ0  = uiZ.v0;
@@ -98,7 +98,7 @@ extFloat80_t f128_to_extF80( float128_t a )
    sig128 =
        softfloat_shortShiftLeft128(
            frac64 | UINT64_C( 0x0001000000000000 ), frac0, 15 );
-    return softfloat_roundPackToExtF80( sign, exp, sig128.v64, sig128.v0, 80 );
+    return softfloat_roundPackToExtF80( state, sign, exp, sig128.v64, sig128.v0, 80 );
    /*------------------------------------------------------------------------
    *------------------------------------------------------------------------*/
 uiZ:
@@ -0,0 +1,95 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
+California.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+float16_t f128_to_f16( struct softfloat_state *state, float128_t a )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    bool sign;
+    int_fast32_t exp;
+    uint_fast64_t frac64;
+    struct commonNaN commonNaN;
+    uint_fast16_t uiZ, frac16;
+    union ui16_f16 uZ;
+
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    sign  = signF128UI64( uiA64 );
+    exp   = expF128UI64( uiA64 );
+    frac64 = fracF128UI64( uiA64 ) | (uiA0 != 0);
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( exp == 0x7FFF ) {
+        if ( frac64 ) {
+            softfloat_f128UIToCommonNaN( state, uiA64, uiA0, &commonNaN );
+            uiZ = softfloat_commonNaNToF16UI( &commonNaN );
+        } else {
+            uiZ = packToF16UI( sign, 0x1F, 0 );
+        }
+        goto uiZ;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    frac16 = softfloat_shortShiftRightJam64( frac64, 34 );
+    if ( ! (exp | frac16) ) {
+        uiZ = packToF16UI( sign, 0, 0 );
+        goto uiZ;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    exp -= 0x3FF1;
+    if ( sizeof (int_fast16_t) < sizeof (int_fast32_t) ) {
+        if ( exp < -0x40 ) exp = -0x40;
+    }
+    return softfloat_roundPackToF16( sign, exp, frac16 | 0x4000 );
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+ uiZ:
+    uZ.ui = uiZ;
+    return uZ.f;
+
+}
+
@@ -0,0 +1,95 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
+California.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+float32_t f128_to_f32( struct softfloat_state *state, float128_t a )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    bool sign;
+    int_fast32_t exp;
+    uint_fast64_t frac64;
+    struct commonNaN commonNaN;
+    uint_fast32_t uiZ, frac32;
+    union ui32_f32 uZ;
+
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    sign  = signF128UI64( uiA64 );
+    exp   = expF128UI64( uiA64 );
+    frac64 = fracF128UI64( uiA64 ) | (uiA0 != 0);
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( exp == 0x7FFF ) {
+        if ( frac64 ) {
+            softfloat_f128UIToCommonNaN( state, uiA64, uiA0, &commonNaN );
+            uiZ = softfloat_commonNaNToF32UI( &commonNaN );
+        } else {
+            uiZ = packToF32UI( sign, 0xFF, 0 );
+        }
+        goto uiZ;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    frac32 = softfloat_shortShiftRightJam64( frac64, 18 );
+    if ( ! (exp | frac32) ) {
+        uiZ = packToF32UI( sign, 0, 0 );
+        goto uiZ;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    exp -= 0x3F81;
+    if ( sizeof (int_fast16_t) < sizeof (int_fast32_t) ) {
+        if ( exp < -0x1000 ) exp = -0x1000;
+    }
+    return softfloat_roundPackToF32( state, sign, exp, frac32 | 0x40000000 );
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+ uiZ:
+    uZ.ui = uiZ;
+    return uZ.f;
+
+}
+
@@ -0,0 +1,100 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
+California.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+float64_t f128_to_f64( struct softfloat_state *state, float128_t a )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    bool sign;
+    int_fast32_t exp;
+    uint_fast64_t frac64, frac0;
+    struct commonNaN commonNaN;
+    uint_fast64_t uiZ;
+    struct uint128 frac128;
+    union ui64_f64 uZ;
+
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    sign  = signF128UI64( uiA64 );
+    exp   = expF128UI64( uiA64 );
+    frac64 = fracF128UI64( uiA64 );
+    frac0  = uiA0;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( exp == 0x7FFF ) {
+        if ( frac64 | frac0 ) {
+            softfloat_f128UIToCommonNaN( state, uiA64, uiA0, &commonNaN );
+            uiZ = softfloat_commonNaNToF64UI( &commonNaN );
+        } else {
+            uiZ = packToF64UI( sign, 0x7FF, 0 );
+        }
+        goto uiZ;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    frac128 = softfloat_shortShiftLeft128( frac64, frac0, 14 );
+    frac64 = frac128.v64 | (frac128.v0 != 0);
+    if ( ! (exp | frac64) ) {
+        uiZ = packToF64UI( sign, 0, 0 );
+        goto uiZ;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    exp -= 0x3C01;
+    if ( sizeof (int_fast16_t) < sizeof (int_fast32_t) ) {
+        if ( exp < -0x1000 ) exp = -0x1000;
+    }
+    return
+        softfloat_roundPackToF64(
+            state, sign, exp, frac64 | UINT64_C( 0x4000000000000000 ) );
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+ uiZ:
+    uZ.ui = uiZ;
+    return uZ.f;
+
+}
+
@@ -0,0 +1,85 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014, 2015, 2016, 2017 The Regents of the
+University of California.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+int_fast32_t f128_to_i32( struct softfloat_state *state, float128_t a, uint_fast8_t roundingMode, bool exact )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    bool sign;
+    int_fast32_t exp;
+    uint_fast64_t sig64, sig0;
+    int_fast32_t shiftDist;
+
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    sign  = signF128UI64( uiA64 );
+    exp   = expF128UI64( uiA64 );
+    sig64 = fracF128UI64( uiA64 );
+    sig0  = uiA0;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+#if (i32_fromNaN != i32_fromPosOverflow) || (i32_fromNaN != i32_fromNegOverflow)
+    if ( (exp == 0x7FFF) && (sig64 | sig0) ) {
+#if (i32_fromNaN == i32_fromPosOverflow)
+        sign = 0;
+#elif (i32_fromNaN == i32_fromNegOverflow)
+        sign = 1;
+#else
+        softfloat_raiseFlags( softfloat_flag_invalid );
+        return i32_fromNaN;
+#endif
+    }
+#endif
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( exp ) sig64 |= UINT64_C( 0x0001000000000000 );
+    sig64 |= (sig0 != 0);
+    shiftDist = 0x4023 - exp;
+    if ( 0 < shiftDist ) sig64 = softfloat_shiftRightJam64( sig64, shiftDist );
+    return softfloat_roundToI32( state, sign, sig64, roundingMode, exact );
+
+}
+
@@ -0,0 +1,95 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014, 2015, 2016, 2017 The Regents of the
+University of California.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+int_fast64_t f128_to_i64( struct softfloat_state *state, float128_t a, uint_fast8_t roundingMode, bool exact )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    bool sign;
+    int_fast32_t exp;
+    uint_fast64_t sig64, sig0;
+    int_fast32_t shiftDist;
+    struct uint128 sig128;
+    struct uint64_extra sigExtra;
+
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    sign  = signF128UI64( uiA64 );
+    exp   = expF128UI64( uiA64 );
+    sig64 = fracF128UI64( uiA64 );
+    sig0  = uiA0;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    shiftDist = 0x402F - exp;
+    if ( shiftDist <= 0 ) {
+        /*--------------------------------------------------------------------
+        *--------------------------------------------------------------------*/
+        if ( shiftDist < -15 ) {
+            softfloat_raiseFlags( state, softfloat_flag_invalid );
+            return
+                (exp == 0x7FFF) && (sig64 | sig0) ? i64_fromNaN
+                    : sign ? i64_fromNegOverflow : i64_fromPosOverflow;
+        }
+        /*--------------------------------------------------------------------
+        *--------------------------------------------------------------------*/
+        sig64 |= UINT64_C( 0x0001000000000000 );
+        if ( shiftDist ) {
+            sig128 = softfloat_shortShiftLeft128( sig64, sig0, -shiftDist );
+            sig64 = sig128.v64;
+            sig0  = sig128.v0;
+        }
+    } else {
+        /*--------------------------------------------------------------------
+        *--------------------------------------------------------------------*/
+        if ( exp ) sig64 |= UINT64_C( 0x0001000000000000 );
+        sigExtra = softfloat_shiftRightJam64Extra( sig64, sig0, shiftDist );
+        sig64 = sigExtra.v;
+        sig0  = sigExtra.extra;
+    }
+    return softfloat_roundToI64( state, sign, sig64, sig0, roundingMode, exact );
+
+}
+
@@ -0,0 +1,86 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014, 2015, 2016, 2017 The Regents of the
+University of California.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+uint_fast32_t
+ f128_to_ui32( struct softfloat_state *state, float128_t a, uint_fast8_t roundingMode, bool exact )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    bool sign;
+    int_fast32_t exp;
+    uint_fast64_t sig64;
+    int_fast32_t shiftDist;
+
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    sign  = signF128UI64( uiA64 );
+    exp   = expF128UI64( uiA64 );
+    sig64 = fracF128UI64( uiA64 ) | (uiA0 != 0);
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+#if (ui32_fromNaN != ui32_fromPosOverflow) || (ui32_fromNaN != ui32_fromNegOverflow)
+    if ( (exp == 0x7FFF) && sig64 ) {
+#if (ui32_fromNaN == ui32_fromPosOverflow)
+        sign = 0;
+#elif (ui32_fromNaN == ui32_fromNegOverflow)
+        sign = 1;
+#else
+        softfloat_raiseFlags( softfloat_flag_invalid );
+        return ui32_fromNaN;
+#endif
+    }
+#endif
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( exp ) sig64 |= UINT64_C( 0x0001000000000000 );
+    shiftDist = 0x4023 - exp;
+    if ( 0 < shiftDist ) {
+        sig64 = softfloat_shiftRightJam64( sig64, shiftDist );
+    }
+    return softfloat_roundToUI32( sign, sig64, roundingMode, exact );
+
+}
+
@@ -0,0 +1,96 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014, 2015, 2016, 2017 The Regents of the
+University of California.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+uint_fast64_t
+ f128_to_ui64( struct softfloat_state *state, float128_t a, uint_fast8_t roundingMode, bool exact )
+{
+    union ui128_f128 uA;
+    uint_fast64_t uiA64, uiA0;
+    bool sign;
+    int_fast32_t exp;
+    uint_fast64_t sig64, sig0;
+    int_fast32_t shiftDist;
+    struct uint128 sig128;
+    struct uint64_extra sigExtra;
+
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    uA.f = a;
+    uiA64 = uA.ui.v64;
+    uiA0  = uA.ui.v0;
+    sign  = signF128UI64( uiA64 );
+    exp   = expF128UI64( uiA64 );
+    sig64 = fracF128UI64( uiA64 );
+    sig0  = uiA0;
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    shiftDist = 0x402F - exp;
+    if ( shiftDist <= 0 ) {
+        /*--------------------------------------------------------------------
+        *--------------------------------------------------------------------*/
+        if ( shiftDist < -15 ) {
+            softfloat_raiseFlags( state, softfloat_flag_invalid );
+            return
+                (exp == 0x7FFF) && (sig64 | sig0) ? ui64_fromNaN
+                    : sign ? ui64_fromNegOverflow : ui64_fromPosOverflow;
+        }
+        /*--------------------------------------------------------------------
+        *--------------------------------------------------------------------*/
+        sig64 |= UINT64_C( 0x0001000000000000 );
+        if ( shiftDist ) {
+            sig128 = softfloat_shortShiftLeft128( sig64, sig0, -shiftDist );
+            sig64 = sig128.v64;
+            sig0  = sig128.v0;
+        }
+    } else {
+        /*--------------------------------------------------------------------
+        *--------------------------------------------------------------------*/
+        if ( exp ) sig64 |= UINT64_C( 0x0001000000000000 );
+        sigExtra = softfloat_shiftRightJam64Extra( sig64, sig0, shiftDist );
+        sig64 = sigExtra.v;
+        sig0  = sigExtra.extra;
+    }
+    return softfloat_roundToUI64( state, sign, sig64, sig0, roundingMode, exact );
+
+}
+
@@ -42,7 +42,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "softfloat.h"

 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t f32_to_extF80( float32_t a )
+extFloat80_t f32_to_extF80( struct softfloat_state *state, float32_t a )
 {
    union ui32_f32 uA;
    uint_fast32_t uiA;
@@ -67,7 +67,7 @@ extFloat80_t f32_to_extF80( float32_t a )
    *------------------------------------------------------------------------*/
    if ( exp == 0xFF ) {
        if ( frac ) {
-            softfloat_f32UIToCommonNaN( uiA, &commonNaN );
+            softfloat_f32UIToCommonNaN( state, uiA, &commonNaN );
            uiZ = softfloat_commonNaNToExtF80UI( &commonNaN );
            uiZ64 = uiZ.v64;
            uiZ0  = uiZ.v0;
@@ -0,0 +1,96 @@
+
+/*============================================================================
+
+This C source file is part of the SoftFloat IEEE Floating-Point Arithmetic
+Package, Release 3e, by John R. Hauser.
+
+Copyright 2011, 2012, 2013, 2014, 2015 The Regents of the University of
+California.  All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+ 1. Redistributions of source code must retain the above copyright notice,
+    this list of conditions, and the following disclaimer.
+
+ 2. Redistributions in binary form must reproduce the above copyright notice,
+    this list of conditions, and the following disclaimer in the documentation
+    and/or other materials provided with the distribution.
+
+ 3. Neither the name of the University nor the names of its contributors may
+    be used to endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
+EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
+DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+=============================================================================*/
+
+#include <stdbool.h>
+#include <stdint.h>
+#include "platform.h"
+#include "internals.h"
+#include "specialize.h"
+#include "softfloat.h"
+
+float128_t f32_to_f128( struct softfloat_state *state, float32_t a )
+{
+    union ui32_f32 uA;
+    uint_fast32_t uiA;
+    bool sign;
+    int_fast16_t exp;
+    uint_fast32_t frac;
+    struct commonNaN commonNaN;
+    struct uint128 uiZ;
+    struct exp16_sig32 normExpSig;
+    union ui128_f128 uZ;
+
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    uA.f = a;
+    uiA = uA.ui;
+    sign = signF32UI( uiA );
+    exp  = expF32UI( uiA );
+    frac = fracF32UI( uiA );
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( exp == 0xFF ) {
+        if ( frac ) {
+            softfloat_f32UIToCommonNaN( state, uiA, &commonNaN );
+            uiZ = softfloat_commonNaNToF128UI( &commonNaN );
+        } else {
+            uiZ.v64 = packToF128UI64( sign, 0x7FFF, 0 );
+            uiZ.v0  = 0;
+        }
+        goto uiZ;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    if ( ! exp ) {
+        if ( ! frac ) {
+            uiZ.v64 = packToF128UI64( sign, 0, 0 );
+            uiZ.v0  = 0;
+            goto uiZ;
+        }
+        normExpSig = softfloat_normSubnormalF32Sig( frac );
+        exp = normExpSig.exp - 1;
+        frac = normExpSig.sig;
+    }
+    /*------------------------------------------------------------------------
+    *------------------------------------------------------------------------*/
+    uiZ.v64 = packToF128UI64( sign, exp + 0x3F80, (uint_fast64_t) frac<<25 );
+    uiZ.v0  = 0;
+ uiZ:
+    uZ.ui = uiZ;
+    return uZ.f;
+
+}
+
@@ -42,7 +42,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "softfloat.h"

 FEXCORE_PRESERVE_ALL_ATTR
-extFloat80_t f64_to_extF80( float64_t a )
+extFloat80_t f64_to_extF80( struct softfloat_state *state, float64_t a )
 {
    union ui64_f64 uA;
    uint_fast64_t uiA;
@@ -67,7 +67,7 @@ extFloat80_t f64_to_extF80( float64_t a )
    *------------------------------------------------------------------------*/
    if ( exp == 0x7FF ) {
        if ( frac ) {
-            softfloat_f64UIToCommonNaN( uiA, &commonNaN );
+            softfloat_f64UIToCommonNaN( state, uiA, &commonNaN );
            uiZ = softfloat_commonNaNToExtF80UI( &commonNaN );
            uiZ64 = uiZ.v64;
            uiZ0  = uiZ.v0;
				`@@ -0,0 +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`
				`@@ -0,0 +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`