Docs: Update for release FEX-2308

Merge pull request #2860 from lioncash/alias
ARMEmitter: Add missing atomic aliases
2026-10-18 05:00:37 +02:00 · 2023-08-06 02:34:55 -07:00 · 2023-08-06 02:05:30 -07:00 · 2023-08-06 05:05:07 -04:00 · 2023-08-04 21:49:59 -04:00 · 2023-08-04 19:47:03 -04:00
1353 changed files with 139362 additions and 44616 deletions
@@ -14,6 +14,7 @@ env:
  CC: clang
  CXX: clang++
  FEX_FORCE32BITALLOCATOR: 1
+  FEX_ENABLEAVX: 1

 jobs:
  build:
@@ -24,14 +25,29 @@ jobs:
      fail-fast: false

    steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v3

    - name: Set runner label
      run: echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV

+    - name: Set rootfs paths
+      run: |
+        echo "FEX_ROOTFS_MOUNT=/mnt/AutoNFS/rootfs/" >> $GITHUB_ENV
+        echo "FEX_ROOTFS_PATH=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "FEX_ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+
+    - name: Update RootFS cache
+      # Use a bash shell so we can use the same syntax for environment variable
+      # access regardless of the host operating system
+      shell: bash
+      run: $GITHUB_WORKSPACE/Scripts/CI_FetchRootFS.py
+
    - name : submodule checkout
      # Need to update submodules
-      run: git submodule update --init --depth 1
+      run: |
+        git submodule sync --recursive
+        git submodule update --init --depth 1

    - name: Clean Build Environment
      run: rm -Rf ${{runner.workspace}}/build
@@ -49,7 +65,7 @@ jobs:
      # Note the current convention is to use the -S and -B options here to specify source
      # and build directories, but this is only available with CMake 3.13 and higher.
      # The CMake binaries on the Github Actions machines are (as of this writing) 3.12
-      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True
+      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True -DENABLE_INTERPRETER=True -DBUILD_FEX_LINUX_TESTS=True -DBUILD_THUNKS=True -DCMAKE_INSTALL_PREFIX=${{runner.workspace}}/build/install

    - name: Build
      working-directory: ${{runner.workspace}}/build
@@ -57,6 +73,11 @@ jobs:
      # Execute the build.  You can specify a specific target with "--target <NAME>"
      run: cmake --build . --config $BUILD_TYPE

+    - name: Install
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      run: cmake --build . --config $BUILD_TYPE --target install
+
    - name: ASM Tests
      working-directory: ${{runner.workspace}}/build
      shell: bash
@@ -151,6 +172,78 @@ jobs:
      working-directory: ${{runner.workspace}}/build
      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_APITests.log || true

+    - name: FEXCore APITest tests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      run: cmake --build . --config $BUILD_TYPE --target fexcore_apitests
+
+    - name: FEXCore APITest Test Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_FEXCoreAPITests.log || true
+
+    - name: ARMEmitter tests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      run: cmake --build . --config $BUILD_TYPE --target emitter_tests
+
+    - name: ARMEmitter Test Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_ARMEmitterTests.log || true
+
+    - name: FEXLinuxTests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      run: cmake --build . --config $BUILD_TYPE --target fex_linux_tests_all
+
+    - name: FEXLinuxTests Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_FEXLinuxTests.log || true
+
+    - name: Thunkgen tests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      run: cmake --build . --config $BUILD_TYPE --target thunkgen_tests
+
+    - name: Thunkgen Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_ThunkgenTests.log || true
+
+    - name: Test GL No-Thunks
+      if: matrix.arch[1] == 'x64'
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      env:
+        DISPLAY: ":0"
+      run: cmake --build . --config $BUILD_TYPE --target thunk_functional_tests_nothunks
+
+    - name: No thunks Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_NoThunkResults.log || true
+
+    - name: Test GL Thunks
+      if: matrix.arch[1] == 'x64'
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      env:
+        DISPLAY: ":0"
+      run: cmake --build . --config $BUILD_TYPE --target thunk_functional_tests_thunks
+
+    - name: Thunks Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_ThunkResults.log || true
+
    - name: Truncate test results
      if: ${{ always() }}
      shell: bash
@@ -159,15 +252,21 @@ jobs:
      # ASM tests get quite close to 10MB
      run: truncate --size=<20M ${{runner.workspace}}/build/Testing/Temporary/LastTest_*.log || true

+    - name: Remove old SHM regions
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: cmake --build . --config $BUILD_TYPE --target remove_old_shm_regions
+
    - name: Set runner name
      if: ${{ always() }}
      run: echo "runner_name=$(hostname)" >> $GITHUB_ENV

    - name: Upload results
      if: ${{ always() }}
-      uses: 'actions/upload-artifact@v2'
+      uses: 'actions/upload-artifact@v3'
+      timeout-minutes: 1
      with:
        name: Results-${{ env.runner_name }}
        path: ${{runner.workspace}}/build/Testing/Temporary/LastTest_*.log
        retention-days: 3
-
@@ -0,0 +1,208 @@
+name: GLIBC fault test
+# This workflow file is the same as the `Build + Test` with some key differences
+# - Runs on any x86 and ARM64 runner
+# - Disables the glibc jemalloc compile option
+# - Enables the glibc allocator fault option
+# - Disables gvisor tests to reduce stress on CI machines (tmp/shm tests overwhelm them)
+# - Disables thunk tests since they are incompatible with glibc fault allocator
+# - Disables ARMEmitter tests (We don't want to fault test vixl's disassembler)
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - main
+
+env:
+  # Customize the CMake build type here (Release, Debug, RelWithDebInfo, etc.)
+  BUILD_TYPE: Release
+  CC: clang
+  CXX: clang++
+  FEX_FORCE32BITALLOCATOR: 1
+  FEX_ENABLEAVX: 1
+
+jobs:
+  build:
+    runs-on: ${{ matrix.arch }}
+    strategy:
+      matrix:
+        # Run on an x86 device and any ARM runner.
+        arch: [[self-hosted, x64], [self-hosted, ARM64]]
+      fail-fast: false
+
+    steps:
+    - uses: actions/checkout@v3
+
+    - name: Set runner label
+      run: echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
+
+    - name: Set rootfs paths
+      run: |
+        echo "FEX_ROOTFS_MOUNT=/mnt/AutoNFS/rootfs/" >> $GITHUB_ENV
+        echo "FEX_ROOTFS_PATH=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "FEX_ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+
+    - name: Update RootFS cache
+      # Use a bash shell so we can use the same syntax for environment variable
+      # access regardless of the host operating system
+      shell: bash
+      run: $GITHUB_WORKSPACE/Scripts/CI_FetchRootFS.py
+
+    - name : submodule checkout
+      # Need to update submodules
+      run: |
+        git submodule sync --recursive
+        git submodule update --init --depth 1
+
+    - name: Clean Build Environment
+      run: rm -Rf ${{runner.workspace}}/build
+
+    - name: Create Build Environment
+      # Some projects don't allow in-source building, so create a separate build directory
+      # We'll use this as our working directory for all subsequent commands
+      run: cmake -E make_directory ${{runner.workspace}}/build
+
+    - name: Configure CMake
+      # Use a bash shell so we can use the same syntax for environment variable
+      # access regardless of the host operating system
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      # Note the current convention is to use the -S and -B options here to specify source
+      # and build directories, but this is only available with CMake 3.13 and higher.
+      # The CMake binaries on the Github Actions machines are (as of this writing) 3.12
+      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True -DENABLE_INTERPRETER=True -DBUILD_FEX_LINUX_TESTS=True -DENABLE_GLIBC_ALLOCATOR_HOOK_FAULT=True -DENABLE_JEMALLOC_GLIBC_ALLOC=False -DCMAKE_INSTALL_PREFIX=${{runner.workspace}}/build/install
+
+    - name: Build
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      # Execute the build.  You can specify a specific target with "--target <NAME>"
+      run: cmake --build . --config $BUILD_TYPE
+
+    - name: Install
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      run: cmake --build . --config $BUILD_TYPE --target install
+
+    - name: ASM Tests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      # Execute the unit tests
+      run: cmake --build . --config $BUILD_TYPE --target asm_tests
+
+    - name: ASM Test Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_ASM.log || true
+
+    - name: IR Tests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      # Execute the unit tests
+      run: cmake --build . --config $BUILD_TYPE --target ir_tests
+
+    - name: IR Test Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_IR.log || true
+
+    - name: Posix Tests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      # Execute the posixtest
+      run: cmake --build . --config $BUILD_TYPE --target posix_tests
+
+    - name: Posix Test Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_Posix.log || true
+
+    - name: gcc target tests 64
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      # Execute the gvisor tests
+      run: cmake --build . --config $BUILD_TYPE --target gcc_target_tests_64
+
+    - name: GCC64 Test Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_GCC64.log || true
+
+    - name: gcc target tests 32
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      # Execute the gvisor tests
+      run: cmake --build . --config $BUILD_TYPE --target gcc_target_tests_32
+
+    - name: GCC32 Test Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_GCC32.log || true
+
+    - name: APITest tests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      run: cmake --build . --config $BUILD_TYPE --target api_tests
+
+    - name: APITest Test Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_APITests.log || true
+
+    - name: FEXCore APITest tests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      run: cmake --build . --config $BUILD_TYPE --target fexcore_apitests
+
+    - name: FEXCore APITest Test Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_FEXCoreAPITests.log || true
+
+    - name: FEXLinuxTests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      run: cmake --build . --config $BUILD_TYPE --target fex_linux_tests_all
+
+    - name: FEXLinuxTests Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_FEXLinuxTests.log || true
+
+    - name: Truncate test results
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      # Cap out the log files at 20M in case something crash spins and dumps fault text
+      # ASM tests get quite close to 10MB
+      run: truncate --size=<20M ${{runner.workspace}}/build/Testing/Temporary/LastTest_*.log || true
+
+    - name: Remove old SHM regions
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: cmake --build . --config $BUILD_TYPE --target remove_old_shm_regions
+
+    - name: Set runner name
+      if: ${{ always() }}
+      run: echo "runner_name=$(hostname)" >> $GITHUB_ENV
+
+    - name: Upload results
+      if: ${{ always() }}
+      uses: 'actions/upload-artifact@v3'
+      timeout-minutes: 1
+      with:
+        name: Results-${{ env.runner_name }}
+        path: ${{runner.workspace}}/build/Testing/Temporary/LastTest_*.log
+        retention-days: 3
+
@@ -0,0 +1,96 @@
+name: Mingw build
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - main
+
+env:
+  BUILD_TYPE: Debug
+  FEX_ENABLEAVX: 1
+
+jobs:
+  build:
+    runs-on: ${{ matrix.arch }}
+    strategy:
+      matrix:
+        arch: [[self-hosted, x64, mingw], [self-hosted, ARM64, mingw]]
+      fail-fast: false
+
+    steps:
+    - uses: actions/checkout@v3
+
+    - name: Set runner label
+      run: echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
+
+    - name: Set CC x86
+      if: matrix.arch[1] == 'x64'
+      run: |
+        echo "CC=$HOME/llvm-mingw/build/bin/x86_64-w64-mingw32-clang" >> $GITHUB_ENV
+        echo "CXX=$HOME/llvm-mingw/build/bin/x86_64-w64-mingw32-clang++" >> $GITHUB_ENV
+
+    - name: Set CC Arm64
+      if: matrix.arch[1] == 'ARM64'
+      run: |
+        echo "CC=$HOME/llvm-mingw/build/bin/aarch64-w64-mingw32-clang" >> $GITHUB_ENV
+        echo "CXX=$HOME/llvm-mingw/build/bin/aarch64-w64-mingw32-clang++" >> $GITHUB_ENV
+
+    - name: Set rootfs paths
+      run: |
+        echo "FEX_ROOTFS_MOUNT=/mnt/AutoNFS/rootfs/" >> $GITHUB_ENV
+        echo "FEX_ROOTFS_PATH=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "FEX_ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+
+    - name: Update RootFS cache
+      # Use a bash shell so we can use the same syntax for environment variable
+      # access regardless of the host operating system
+      shell: bash
+      run: $GITHUB_WORKSPACE/Scripts/CI_FetchRootFS.py
+
+    - name : submodule checkout
+      # Need to update submodules
+      run: |
+        git submodule sync --recursive
+        git submodule update --init --depth 1
+
+    - name: Clean Build Environment
+      run: rm -Rf ${{runner.workspace}}/build
+
+    - name: Create Build Environment
+      # Some projects don't allow in-source building, so create a separate build directory
+      # We'll use this as our working directory for all subsequent commands
+      run: cmake -E make_directory ${{runner.workspace}}/build
+
+    - name: Configure CMake
+      # Use a bash shell so we can use the same syntax for environment variable
+      # access regardless of the host operating system
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      # Note the current convention is to use the -S and -B options here to specify source
+      # and build directories, but this is only available with CMake 3.13 and higher.
+      # The CMake binaries on the Github Actions machines are (as of this writing) 3.12
+      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True -DENABLE_INTERPRETER=False -DBUILD_TESTS=False -DENABLE_JEMALLOC=False -DENABLE_JEMALLOC_GLIBC_ALLOC=False -DCMAKE_INSTALL_PREFIX=${{runner.workspace}}/build/install
+
+    - name: Build
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      # Execute the build.  You can specify a specific target with "--target <NAME>"
+      run: cmake --build . --config $BUILD_TYPE
+
+    - name: Set runner name
+      if: ${{ always() }}
+      run: echo "runner_name=$(hostname)" >> $GITHUB_ENV
+
+    - name: Upload results
+      if: ${{ always() }}
+      uses: 'actions/upload-artifact@v3'
+      timeout-minutes: 1
+      with:
+        name: Results-${{ env.runner_name }}
+        path: ${{runner.workspace}}/build/Testing/Temporary/LastTest_*.log
+        retention-days: 3
+
@@ -0,0 +1,120 @@
+name: Vixl Simulator run
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - main
+
+env:
+  # Customize the CMake build type here (Release, Debug, RelWithDebInfo, etc.)
+  BUILD_TYPE: Release
+  CC: clang
+  CXX: clang++
+  FEX_ENABLEAVX: 1
+
+jobs:
+  build:
+    runs-on: ${{ matrix.arch }}
+    strategy:
+      matrix:
+        # Only the x86-64 runner is fast enough to run this
+        arch: [[self-hosted, x64], [self-hosted, ARMv8.4]]
+      fail-fast: false
+
+    steps:
+    - uses: actions/checkout@v3
+
+    - name: Set runner label
+      run: echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
+
+    - name: Set rootfs paths
+      run: |
+        echo "FEX_ROOTFS_MOUNT=/mnt/AutoNFS/rootfs/" >> $GITHUB_ENV
+        echo "FEX_ROOTFS_PATH=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "FEX_ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+
+    - name: Update RootFS cache
+      # Use a bash shell so we can use the same syntax for environment variable
+      # access regardless of the host operating system
+      shell: bash
+      run: $GITHUB_WORKSPACE/Scripts/CI_FetchRootFS.py
+
+    - name : submodule checkout
+      # Need to update submodules
+      run: |
+        git submodule sync --recursive
+        git submodule update --init --depth 1
+
+    - name: Clean Build Environment
+      run: rm -Rf ${{runner.workspace}}/build
+
+    - name: Create Build Environment
+      # Some projects don't allow in-source building, so create a separate build directory
+      # We'll use this as our working directory for all subsequent commands
+      run: cmake -E make_directory ${{runner.workspace}}/build
+
+    - name: Configure CMake
+      # Use a bash shell so we can use the same syntax for environment variable
+      # access regardless of the host operating system
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      # Note the current convention is to use the -S and -B options here to specify source
+      # and build directories, but this is only available with CMake 3.13 and higher.
+      # The CMake binaries on the Github Actions machines are (as of this writing) 3.12
+      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DENABLE_VIXL_SIMULATOR=True -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True
+
+    - name: Build
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      # Execute the build.  You can specify a specific target with "--target <NAME>"
+      run: cmake --build . --config $BUILD_TYPE
+
+    - name: ASM Tests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      # Execute the unit tests
+      run: cmake --build . --config $BUILD_TYPE --target asm_tests
+
+    - name: ASM Test Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_ASM.log || true
+
+    - name: IR Tests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      # Execute the unit tests
+      run: cmake --build . --config $BUILD_TYPE --target ir_tests
+
+    - name: IR Test Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_IR.log || true
+
+    - name: Truncate test results
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      # Cap out the log files at 20M in case something crash spins and dumps fault text
+      # ASM tests get quite close to 10MB
+      run: truncate --size=<20M ${{runner.workspace}}/build/Testing/Temporary/LastTest_*.log || true
+
+    - name: Set runner name
+      if: ${{ always() }}
+      run: echo "runner_name=$(hostname)" >> $GITHUB_ENV
+
+    - name: Upload results
+      if: ${{ always() }}
+      uses: 'actions/upload-artifact@v3'
+      timeout-minutes: 1
+      with:
+        name: Results-${{ env.runner_name }}
+        path: ${{runner.workspace}}/build/Testing/Temporary/LastTest_*.log
+        retention-days: 3
+
@@ -10,3 +10,4 @@ out/
 .vscode/
 .vs/
 *.pyc
+.cache
@@ -1,9 +1,9 @@
 [submodule "External/vixl"]
 	shallow = true
 	path = External/vixl
-	url = https://github.com/Sonicadvance1/vixl.git
+	url = https://github.com/FEX-Emu/vixl.git
 [submodule "External/cpp-optparse"]
-	path = External/cpp-optparse
+	path = Source/Common/cpp-optparse
 	url = https://github.com/Sonicadvance1/cpp-optparse
 [submodule "External/imgui"]
 	path = External/imgui
@@ -45,3 +45,14 @@
 [submodule "External/Catch2"]
 	path = External/Catch2
 	url = https://github.com/catchorg/Catch2.git
+[submodule "External/robin-map"]
+	shallow = true
+	path = External/robin-map
+	url = https://github.com/FEX-Emu/robin-map.git
+[submodule "External/Vulkan-Headers"]
+	shallow = true
+	path = External/Vulkan-Headers
+	url = https://github.com/KhronosGroup/Vulkan-Headers.git
+[submodule "External/jemalloc_glibc"]
+	path = External/jemalloc_glibc
+	url = https://github.com/FEX-Emu/jemalloc.git
@@ -0,0 +1,5 @@
+{
+  "ThunksDB": {
+    "GL": 1
+  }
+}
@@ -0,0 +1,5 @@
+{
+  "ThunksDB": {
+    "Vulkan": 1
+  }
+}
@@ -0,0 +1,21 @@
+if(NOT EXISTS "@CMAKE_BINARY_DIR@/install_manifest.txt")
+  message(FATAL_ERROR "Cannot find install manifest: @CMAKE_BINARY_DIR@/install_manifest.txt")
+endif()
+
+file(READ "@CMAKE_BINARY_DIR@/install_manifest.txt" files)
+string(REGEX REPLACE "\n" ";" files "${files}")
+foreach(file ${files})
+  message(STATUS "Uninstalling $ENV{DESTDIR}${file}")
+  if(IS_SYMLINK "$ENV{DESTDIR}${file}" OR EXISTS "$ENV{DESTDIR}${file}")
+    exec_program(
+      "@CMAKE_COMMAND@" ARGS "-E remove \"$ENV{DESTDIR}${file}\""
+      OUTPUT_VARIABLE rm_out
+      RETURN_VALUE rm_retval
+      )
+    if(NOT "${rm_retval}" STREQUAL 0)
+      message(FATAL_ERROR "Problem when removing $ENV{DESTDIR}${file}")
+    endif()
+  else(IS_SYMLINK "$ENV{DESTDIR}${file}" OR EXISTS "$ENV{DESTDIR}${file}")
+    message(STATUS "File $ENV{DESTDIR}${file} does not exist.")
+  endif()
+endforeach()
@@ -0,0 +1,30 @@
+# Extracts a version from the passed in version string in the form of "<Major>.<Minor>.<Patch>".
+# If a part of the version is missing then it gets set as zero.
+# Version variables returned in:
+# ${Package}_VERSION_MAJOR
+# ${Package}_VERSION_MINOR
+# ${Package}_VERSION_PATCH
+function(version_to_variables VERSION _Package)
+  string(REPLACE "." ";" VERSION_LIST "${VERSION}")
+  list (LENGTH VERSION_LIST VERSION_LEN)
+  if (${VERSION_LEN} GREATER 0)
+    list(GET VERSION_LIST 0 VERSION_MAJOR)
+    set(${_Package}_VERSION_MAJOR ${VERSION_MAJOR} PARENT_SCOPE)
+  else()
+    set(${_Package}_VERSION_MAJOR 0 PARENT_SCOPE)
+  endif()
+
+  if (${VERSION_LEN} GREATER 1)
+    list(GET VERSION_LIST 1 VERSION_MINOR)
+    set(${_Package}_VERSION_MINOR ${VERSION_MINOR} PARENT_SCOPE)
+  else()
+    set(${_Package}_VERSION_MINOR 0 PARENT_SCOPE)
+  endif()
+
+  if (${VERSION_LEN} GREATER 2)
+    list(GET VERSION_LIST 2 VERSION_PATCH)
+    set(${_Package}_VERSION_PATCH ${VERSION_PATCH} PARENT_SCOPE)
+  else()
+    set(${_Package}_VERSION_PATCH 0 PARENT_SCOPE)
+  endif()
+endfunction()
@@ -1,31 +1,84 @@
 cmake_minimum_required(VERSION 3.14)
 project(FEX)

+INCLUDE (CheckIncludeFiles)
+CHECK_INCLUDE_FILES ("gdb/jit-reader.h" HAVE_GDB_JIT_READER_H)
+
 option(BUILD_TESTS "Build unit tests to ensure sanity" TRUE)
+option(BUILD_FEX_LINUX_TESTS "Build FEXLinuxTests, requires x86 compiler" FALSE)
 option(BUILD_THUNKS "Build thunks" FALSE)
+option(BUILD_FEXCONFIG "Build FEXConfig, requires SDL2 and X11" TRUE)
+option(ENABLE_CLANG_THUNKS "Build thunks with clang" FALSE)
 option(ENABLE_CLANG_FORMAT "Run clang format over the source" FALSE)
 option(ENABLE_IWYU "Enables include what you use program" FALSE)
 option(ENABLE_LTO "Enable LTO with compilation" TRUE)
 option(ENABLE_XRAY "Enable building with LLVM X-Ray" FALSE)
-option(ENABLE_LLD "Enable linking with LLD" FALSE)
+set(USE_LINKER "" CACHE STRING "Allow overriding the linker path directly")
 option(ENABLE_ASAN "Enables Clang ASAN" FALSE)
 option(ENABLE_TSAN "Enables Clang TSAN" FALSE)
 option(ENABLE_ASSERTIONS "Enables assertions in build" FALSE)
-option(ENABLE_VISUAL_DEBUGGER "Enables the visual debugger for compiling" FALSE)
+option(ENABLE_GDB_SYMBOLS "Enables GDBSymbols integration support" ${HAVE_GDB_JIT_READER_H})
 option(ENABLE_STRICT_WERROR "Enables stricter -Werror for CI" FALSE)
 option(ENABLE_WERROR "Enables -Werror" FALSE)
-option(ENABLE_STATIC_PIE "Enables static-pie build" FALSE)
 option(ENABLE_JEMALLOC "Enables jemalloc allocator" TRUE)
+option(ENABLE_JEMALLOC_GLIBC_ALLOC "Enables jemalloc glibc allocator" TRUE)
 option(ENABLE_OFFLINE_TELEMETRY "Enables FEX offline telemetry" TRUE)
 option(ENABLE_COMPILE_TIME_TRACE "Enables time trace compile option" FALSE)
 option(ENABLE_LIBCXX "Enables LLVM libc++" FALSE)
+option(ENABLE_INTERPRETER "Enables FEX's Interpreter" FALSE)
+option(ENABLE_CCACHE "Enables ccache for compile caching" TRUE)
+option(ENABLE_TERMUX_BUILD "Forces building for Termux on a non-Termux build machine" FALSE)
+option(ENABLE_VIXL_SIMULATOR "Forces the FEX JIT to use the VIXL simulator" FALSE)
+option(ENABLE_VIXL_DISASSEMBLER "Enables debug disassembler output with VIXL" FALSE)
+option(COMPILE_VIXL_DISASSEMBLER "Compiles the vixl disassembler in to vixl" FALSE)
+option(ENABLE_FEXCORE_PROFILER "Enables use of the FEXCore timeline profiling capabilities" FALSE)
+set (FEXCORE_PROFILER_BACKEND "gpuvis" CACHE STRING "Set which backend you want to use for the FEXCore profiler")
+option(ENABLE_GLIBC_ALLOCATOR_HOOK_FAULT "Enables glibc memory allocation hooking with fault for CI testing")

-set (X86_C_COMPILER "x86_64-linux-gnu-gcc" CACHE STRING "c compiler for compiling x86 guest libs")
-set (X86_CXX_COMPILER "x86_64-linux-gnu-g++" CACHE STRING "c++ compiler for compiling x86 guest libs")
+set (X86_32_TOOLCHAIN_FILE "${CMAKE_CURRENT_SOURCE_DIR}/toolchain_x86_32.cmake" CACHE FILEPATH "Toolchain file for the (cross-)compiler targeting i686")
+set (X86_64_TOOLCHAIN_FILE "${CMAKE_CURRENT_SOURCE_DIR}/toolchain_x86_64.cmake" CACHE FILEPATH "Toolchain file for the (cross-)compiler targeting x86_64")
 set (DATA_DIRECTORY "${CMAKE_INSTALL_PREFIX}/share/fex-emu" CACHE PATH "global data directory")

+string(FIND ${CMAKE_BASE_NAME} mingw CONTAINS_MINGW)
+if (NOT CONTAINS_MINGW EQUAL -1)
+  message (STATUS "Mingw build")
+  set (MINGW_BUILD TRUE)
+  set (ENABLE_JEMALLOC FALSE)
+endif()
+
+if (ENABLE_FEXCORE_PROFILER)
+  add_definitions(-DENABLE_FEXCORE_PROFILER=1)
+  string(TOUPPER "${FEXCORE_PROFILER_BACKEND}" FEXCORE_PROFILER_BACKEND)
+
+  if (FEXCORE_PROFILER_BACKEND STREQUAL "GPUVIS")
+    add_definitions(-DFEXCORE_PROFILER_BACKEND=1)
+  else()
+    message(FATAL_ERROR "Unknown FEXCore profiler backend ${FEXCORE_PROFILER_BACKEND}")
+  endif()
+endif()
+
+if (ENABLE_JEMALLOC_GLIBC_ALLOC AND ENABLE_GLIBC_ALLOCATOR_HOOK_FAULT)
+  message(FATAL_ERROR "Can't have both glibc fault allocator and jemalloc glibc allocator enabled at the same time")
+endif()
+
+if (ENABLE_GLIBC_ALLOCATOR_HOOK_FAULT)
+  add_definitions(-DGLIBC_ALLOCATOR_FAULT=1)
+endif()
+
+# uninstall target
+if(NOT TARGET uninstall)
+  configure_file(
+    "${CMAKE_CURRENT_SOURCE_DIR}/CMakeFiles/cmake_uninstall.cmake.in"
+    "${CMAKE_CURRENT_BINARY_DIR}/CMakeFiles/cmake_uninstall.cmake"
+    IMMEDIATE @ONLY)
+
+  add_custom_target(uninstall
+    COMMAND ${CMAKE_COMMAND} -P ${CMAKE_CURRENT_BINARY_DIR}/CMakeFiles/cmake_uninstall.cmake)
+endif()
+
 # These options are meant for package management
 set (TUNE_CPU "native" CACHE STRING "Override the CPU the build is tuned for")
+set (TUNE_ARCH "generic" CACHE STRING "Override the Arch the build is tuned for")
 set (OVERRIDE_VERSION "detect" CACHE STRING "Override the FEX version in the format of <MMYY>{.<REV>}")

 string(TOUPPER "${CMAKE_BUILD_TYPE}" CMAKE_BUILD_TYPE)
@@ -38,6 +91,17 @@ if (ENABLE_ASSERTIONS)
  add_definitions(-DASSERTIONS_ENABLED=1)
 endif()

+if (ENABLE_GDB_SYMBOLS)
+  message(STATUS "GDBSymbols support enabled")
+  add_definitions(-DGDB_SYMBOLS_ENABLED=1)
+endif()
+
+
+if (ENABLE_INTERPRETER)
+  message(STATUS "Interpreter enabled")
+  add_definitions(-DINTERPRETER_ENABLED=1)
+endif()
+
 set(CMAKE_CXX_STANDARD 20)
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/Bin)
@@ -67,15 +131,17 @@ if (CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64")
  set (CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mcx16")
 endif()

-if (CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64")
+if (CMAKE_SYSTEM_PROCESSOR MATCHES "^aarch64|^arm64|^armv8\.*")
  set(_M_ARM_64 1)
  add_definitions(-D_M_ARM_64=1)
 endif()

-find_program(CCACHE_PROGRAM ccache)
-if(CCACHE_PROGRAM)
-  message(STATUS "CCache enabled")
-  set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE "${CCACHE_PROGRAM}")
+if (ENABLE_CCACHE)
+  find_program(CCACHE_PROGRAM ccache)
+  if(CCACHE_PROGRAM)
+    message(STATUS "CCache enabled")
+    set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE "${CCACHE_PROGRAM}")
+  endif()
 endif()

 if (ENABLE_XRAY)
@@ -89,9 +155,10 @@ if (ENABLE_COMPILE_TIME_TRACE)
 endif()

 set (PTHREAD_LIB pthread)
-if (ENABLE_LLD)
-  set (LD_OVERRIDE "-fuse-ld=lld")
-  link_libraries(${LD_OVERRIDE})
+
+if (USE_LINKER)
+  message(STATUS "Overriding linker to: ${USE_LINKER}")
+  add_link_options("-fuse-ld=${USE_LINKER}")
 endif()

 if (ENABLE_LIBCXX)
@@ -105,128 +172,16 @@ if (NOT ENABLE_OFFLINE_TELEMETRY)
  add_definitions(-DFEX_DISABLE_TELEMETRY=1)
 endif()

-if (ENABLE_STATIC_PIE)
-  if (_M_ARM_64 AND ENABLE_LLD)
-    message (FATAL_ERROR "Static linking does not currently work with AArch64+LLD. Use GNU ld for now.")
-  endif()
+if(DEFINED ENV{TERMUX_VERSION} OR ENABLE_TERMUX_BUILD)
+  add_definitions(-DTERMUX_BUILD=1)
+  set(TERMUX_BUILD 1)

-  file(WRITE ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeTmp/Determine_iplt.c
-    "int main(int argc, char* argv[])
-      {
-        return 0;
-      }")
+  # Termux doesn't support Jemalloc due to bad interactions between emutls, jemalloc, and scudo
+  set(ENABLE_JEMALLOC FALSE)

-  # Compile the test application with our LD_OVERRIDE and static-pie options
-  try_compile(
-    COMPILE_RESULT
-    ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeTmp
-    ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeTmp/Determine_iplt.c
-    COMPILE_DEFINITIONS "-fPIE ${LD_OVERRIDE}"
-    LINK_LIBRARIES "-static-pie ${LD_OVERRIDE}"
-    COPY_FILE ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeTmp/Determine_iplt
-  )
-
-  if (${COMPILE_RESULT})
-    # Read the symbols from the elf
-    execute_process(COMMAND
-      readelf -s ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeTmp/Determine_iplt
-      OUTPUT_FILE ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeTmp/plt_out.txt
-      OUTPUT_VARIABLE PLT_SYMBOLS)
-
-    # Pull out the __rela_iplt_{start,end} symbols if they exist
-    execute_process(COMMAND
-      "grep" "__rela_iplt" ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeTmp/plt_out.txt
-      OUTPUT_VARIABLE PLT_SYMBOLS)
-
-    set (SYMBOLS_FINE TRUE)
-    set (HAS_IPLT -1)
-    # Check if we have any symbols in our grep output
-    # The symbols must either not exist at all OR the symbols are zero
-    if (PLT_SYMBOLS)
-      string(FIND ${PLT_SYMBOLS} "__rela_iplt_start" HAS_IPLT)
-    endif()
-
-    if (NOT HAS_IPLT EQUAL -1)
-      # We have some symbols from readelf. Let's parse the results to check if they are zero
-      # Format: '35: 0000000000000000     0 NOTYPE  LOCAL  HIDDEN   UND __rela_iplt_start'
-      string(REPLACE "\n" ";" SYMBOL_LIST ${PLT_SYMBOLS})
-      foreach (SYMBOL ${SYMBOL_LIST})
-        # strip any leading and trailing whitespace
-        string (STRIP ${SYMBOL} SYMBOL)
-        # Convert string to a list
-        string(REPLACE " " ";" SYMBOL_VALUES ${SYMBOL}})
-        # Pull out the address argument
-        list(GET SYMBOL_VALUES 1 OFFSET)
-
-        # Check against integer zero
-        if (NOT ${OFFSET} EQUAL 0)
-          # Symbol wasn't zero, this now fails
-          set (SYMBOLS_FINE FALSE)
-        endif()
-      endforeach()
-    endif()
-
-    if (SYMBOLS_FINE)
-      # We can now exnable static-pie
-      set (STATIC_PIE_OPTIONS "-static-pie")
-      # Pthreads has an issue with exposing symbols
-      # We need to make some concessions to the pthread gods
-      if (ENABLE_LLD)
-        set (PTHREAD_LIB
-          -Wl,--undefined-glob=pthread_*
-          -Wl,--undefined=__cxa_finalize
-          -Wl,--undefined=_pthread_cleanup_push_defer
-          -Wl,--undefined=_pthread_cleanup_pop_restore
-          -Wl,--undefined=__pthread_cleanup_upto
-          pthread)
-      else()
-        set (PTHREAD_LIB
-          -Wl,--undefined=pthread_join
-          -Wl,--undefined=pthread_attr_getdetachstate
-          -Wl,--undefined=pthread_sigmask
-          -Wl,--undefined=pthread_mutex_lock
-          -Wl,--undefined=pthread_cond_init
-          -Wl,--undefined=pthread_attr_init
-          -Wl,--undefined=pthread_mutex_unlock
-          -Wl,--undefined=pthread_mutexattr_destroy
-          -Wl,--undefined=pthread_detach
-          -Wl,--undefined=pthread_mutex_init
-          -Wl,--undefined=pthread_getattr_np
-          -Wl,--undefined=pthread_cond_timedwait
-          -Wl,--undefined=pthread_attr_destroy
-          -Wl,--undefined=pthread_mutexattr_settype
-          -Wl,--undefined=pthread_rwlock_unlock
-          -Wl,--undefined=pthread_rwlock_wrlock
-          -Wl,--undefined=pthread_setspecific
-          -Wl,--undefined=pthread_create
-          -Wl,--undefined=pthread_cond_clockwait
-          -Wl,--undefined=pthread_key_create
-          -Wl,--undefined=pthread_rwlock_rdlock
-          -Wl,--undefined=pthread_setname_np
-          -Wl,--undefined=pthread_cond_signal
-          -Wl,--undefined=pthread_mutexattr_init
-          -Wl,--undefined=pthread_attr_setstack
-          -Wl,--undefined=pthread_self
-          -Wl,--undefined=pthread_getaffinity_np
-          -Wl,--undefined=pthread_cond_wait
-          -Wl,--undefined=pthread_mutex_trylock
-          -Wl,--undefined=pthread_cond_broadcast
-          -Wl,--undefined=pthread_cond_destroy
-          -Wl,--undefined=pthread_getspecific
-          -Wl,--undefined=pthread_key_delete
-          -Wl,--undefined=pthread_once
-          -Wl,--undefined=__cxa_finalize
-          -Wl,--undefined=_pthread_cleanup_push_defer
-          -Wl,--undefined=_pthread_cleanup_pop_restore
-          -Wl,--undefined=__pthread_cleanup_upto
-          pthread)
-      endif()
-    else()
-      message (FATAL_ERROR "Application has __rela_iplt_{start,end} symbols. Which means static-pie can't be enabled")
-    endif()
-  else()
-    message (FATAL_ERROR "Couldn't compile static-pie test. Static-pie can't be enabled! Is your glibc compiled without static-pie?")
-  endif()
+  # Termux builds can't rely on X11 packages
+  # SDL2 isn't even compiled with GL support so our GUIs wouldn't even work
+  set(BUILD_FEXCONFIG FALSE)
 endif()

 if (ENABLE_ASAN)
@@ -240,7 +195,22 @@ if (ENABLE_TSAN)
  link_libraries(-fno-omit-frame-pointer -fsanitize=thread)
 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()
+  message (STATUS
+    " jemalloc glibc allocator disabled!\n"
+    " This is not a recommended configuration!\n"
+    " This will very explicitly break thunk execution!\n"
+    " Use at your own risk!")
+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/)
@@ -258,6 +228,13 @@ set (CMAKE_LINKER_FLAGS_RELWITHDEBINFO "${CMAKE_LINKER_FLAGS_RELWITHDEBINFO} -fn
 set (CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -fomit-frame-pointer")
 set (CMAKE_LINKER_FLAGS_RELEASE "${CMAKE_LINKER_FLAGS_RELEASE} -fomit-frame-pointer")

+include_directories(External/robin-map/include/)
+
+if (BUILD_TESTS)
+  # Enable vixl disassembler if tests are enabled.
+  set(COMPILE_VIXL_DISASSEMBLER TRUE)
+endif()
+
 add_subdirectory(External/vixl/)
 include_directories(External/vixl/src/)

@@ -268,13 +245,9 @@ endif()

 find_package(PkgConfig REQUIRED)
 find_package(Python 3.0 REQUIRED COMPONENTS Interpreter)
-pkg_check_modules(XXHASH libxxhash>=0.8.0 QUIET)

-if (NOT XXHASH_FOUND)
-  message(STATUS "xxHash not found. Using Externals")
-  add_subdirectory(External/xxhash/)
-  include_directories(External/xxhash/)
-endif()
+add_subdirectory(External/xxhash/)
+include_directories(External/xxhash/)

 add_definitions(-Wno-trigraphs)
 add_definitions(-DGLOBAL_DATA_DIRECTORY="${DATA_DIRECTORY}/")
@@ -289,9 +262,8 @@ if (BUILD_TESTS)
  include(Catch)
 endif()

-add_subdirectory(External/cpp-optparse/)
-include_directories(External/cpp-optparse/)
-
+# Disable fmt install
+set(FMT_INSTALL OFF)
 add_subdirectory(External/fmt/)

 add_subdirectory(External/imgui/)
@@ -335,6 +307,15 @@ if(ENABLE_WERROR OR ENABLE_STRICT_WERROR)
  endif()
 endif()

+if (NOT TUNE_ARCH STREQUAL "generic")
+  check_cxx_compiler_flag("-march=${TUNE_ARCH}" COMPILER_SUPPORTS_ARCH_TYPE)
+  if(COMPILER_SUPPORTS_ARCH_TYPE)
+    add_compile_options("-march=${TUNE_ARCH}")
+  else()
+    message(FATAL_ERROR "Trying to compile arch type '${TUNE_ARCH}' but the compiler doesn't support this")
+  endif()
+endif()
+
 if (TUNE_CPU STREQUAL "native")
  if(_M_ARM_64)
    if (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 999999.0)
@@ -446,8 +427,6 @@ if (BUILD_TESTS)
 endif()

 add_subdirectory(FEXHeaderUtils/)
-include_directories(FEXHeaderUtils/)
-
 add_subdirectory(External/FEXCore)

 # Binfmt_misc files must be installed prior to Source/ installs
@@ -457,51 +436,60 @@ add_subdirectory(Source/)
 add_subdirectory(Data/AppConfig/)

 # Install the ThunksDB file
-install(
-  FILES ${CMAKE_CURRENT_SOURCE_DIR}/Data/ThunksDB.json
-  DESTINATION ${DATA_DIRECTORY}/)
+file(GLOB CONFIG_SOURCES CONFIGURE_DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/Data/*.json)
+
+# Any application configuration json file gets installed
+foreach(CONFIG_SRC ${CONFIG_SOURCES})
+  install(FILES ${CONFIG_SRC}
+    DESTINATION ${DATA_DIRECTORY}/)
+endforeach()

 if (BUILD_TESTS)
  add_subdirectory(unittests/)
 endif()

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

-  include(ExternalProject)
+  # Thunk targets for both host libraries and IDE integration
+  add_subdirectory(ThunkLibs/HostLibs)

-  ExternalProject_Add(host-libs
-    PREFIX host-libs
-    SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/ThunkLibs/HostLibs"
-    BINARY_DIR "Host"
+  # Thunk targets for IDE integration of guest code, only
+  add_subdirectory(ThunkLibs/GuestLibs)
+
+  # Thunk targets for guest libraries
+  include(ExternalProject)
+  ExternalProject_Add(guest-libs
+    PREFIX guest-libs
+    SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/ThunkLibs/GuestLibs"
+    BINARY_DIR "Guest"
    CMAKE_ARGS
+      "-DBITNESS=64"
      "-DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}"
+      "-DENABLE_CLANG_THUNKS=${ENABLE_CLANG_THUNKS}"
+      "-DCMAKE_TOOLCHAIN_FILE:FILEPATH=${X86_64_TOOLCHAIN_FILE}"
      "-DCMAKE_INSTALL_PREFIX=${CMAKE_INSTALL_PREFIX}"
+      "-DSTRUCT_VERIFIER=${CMAKE_SOURCE_DIR}/Scripts/StructPackVerifier.py"
+      "-DFEX_PROJECT_SOURCE_DIR=${FEX_PROJECT_SOURCE_DIR}"
      "-DGENERATOR_EXE=$<TARGET_FILE:thunkgen>"
    INSTALL_COMMAND ""
    BUILD_ALWAYS ON
    DEPENDS thunkgen
  )

-  install(
-    CODE "MESSAGE(\"-- Installing: host-libs\")"
-    CODE "
-    EXECUTE_PROCESS(COMMAND ${CMAKE_COMMAND} --build . --target ThunkHostsInstall
-    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Host
-    )"
-    DEPENDS host-libs
-  )
-
-  ExternalProject_Add(guest-libs
-    PREFIX guest-libs
+  ExternalProject_Add(guest-libs-32
+    PREFIX guest-libs-32
    SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/ThunkLibs/GuestLibs"
-    BINARY_DIR "Guest"
+    BINARY_DIR "Guest_32"
    CMAKE_ARGS
+      "-DBITNESS=32"
      "-DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}"
-      "-DX86_C_COMPILER:STRING=${X86_C_COMPILER}"
-      "-DX86_CXX_COMPILER:STRING=${X86_CXX_COMPILER}"
+      "-DENABLE_CLANG_THUNKS=${ENABLE_CLANG_THUNKS}"
+      "-DCMAKE_TOOLCHAIN_FILE:FILEPATH=${X86_32_TOOLCHAIN_FILE}"
      "-DCMAKE_INSTALL_PREFIX=${CMAKE_INSTALL_PREFIX}"
      "-DSTRUCT_VERIFIER=${CMAKE_SOURCE_DIR}/Scripts/StructPackVerifier.py"
+      "-DFEX_PROJECT_SOURCE_DIR=${FEX_PROJECT_SOURCE_DIR}"
      "-DGENERATOR_EXE=$<TARGET_FILE:thunkgen>"
    INSTALL_COMMAND ""
    BUILD_ALWAYS ON
@@ -511,11 +499,33 @@ if (BUILD_THUNKS)
  install(
    CODE "MESSAGE(\"-- Installing: guest-libs\")"
    CODE "
-    EXECUTE_PROCESS(COMMAND ${CMAKE_COMMAND} --build . --target ThunkGuestsInstall
+    EXECUTE_PROCESS(COMMAND ${CMAKE_COMMAND} --build . --target install
    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest
    )"
    DEPENDS guest-libs
  )
+
+  install(
+    CODE "MESSAGE(\"-- Installing: guest-libs-32\")"
+    CODE "
+    EXECUTE_PROCESS(COMMAND ${CMAKE_COMMAND} --build . --target install
+    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest_32
+    )"
+    DEPENDS guest-libs-32
+  )
+
+  add_custom_target(uninstall_guest-libs
+    COMMAND ${CMAKE_COMMAND} "--build" "." "--target" "uninstall"
+    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest
+  )
+
+  add_custom_target(uninstall_guest-libs-32
+    COMMAND ${CMAKE_COMMAND} "--build" "." "--target" "uninstall"
+    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest_32
+  )
+
+  add_dependencies(uninstall uninstall_guest-libs)
+  add_dependencies(uninstall uninstall_guest-libs-32)
 endif()

 set(FEX_VERSION_MAJOR "0")
@@ -568,15 +578,9 @@ endif()

 # Package creation
 set (CPACK_GENERATOR "DEB")
-if (ENABLE_STATIC_PIE)
-  set (CPACK_PACKAGE_NAME fex-emu-static)
-  set (CPACK_DEBIAN_PACKAGE_CONFLICTS "fex-emu")
-else()
-  set (CPACK_PACKAGE_NAME fex-emu)
-  set (CPACK_DEBIAN_PACKAGE_CONFLICTS "fex-emu-static")
-endif()
+set (CPACK_PACKAGE_NAME fex-emu)
 set (CPACK_PACKAGE_FILE_NAME "${CPACK_PACKAGE_NAME}-${GIT_DESCRIBE_STRING}_${CMAKE_SYSTEM_PROCESSOR}")
-set (CPACK_PACKAGE_CONTACT "FEX-Emu Maintainers <team@fex-emu.org>")
+set (CPACK_PACKAGE_CONTACT "FEX-Emu Maintainers <team@fex-emu.com>")
 set (CPACK_PACKAGE_VERSION_MAJOR "${FEX_VERSION_MAJOR}")
 set (CPACK_PACKAGE_VERSION_MINOR "${FEX_VERSION_MINOR}")
 set (CPACK_PACKAGE_VERSION_PATCH "${FEX_VERSION_PATCH}")
@@ -55,7 +55,7 @@ further defined and clarified by project maintainers.
 ## Enforcement

 Instances of abusive, harassing, or otherwise unacceptable behavior may be
-reported by contacting the project team at team@fex-emu.org. All
+reported by contacting the project team at team@fex-emu.com. All
 complaints will be reviewed and investigated and will result in a response that
 is deemed necessary and appropriate to the circumstances. The project team is
 obligated to maintain confidentiality with regard to the reporter of an incident.
@@ -11,15 +11,15 @@ endforeach()
 # First generate then install it
 foreach(GEN_CONFIG_SRC ${GEN_CONFIG_SOURCES})
  # Get the filename only component
-  get_filename_component(CONFIG_NAME ${GEN_CONFIG_SRC} NAME_WE)
+  get_filename_component(CONFIG_NAME ${GEN_CONFIG_SRC} NAME_WLE)

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

  # Then install the configured json
  install(
-    FILES ${CMAKE_BINARY_DIR}/Data/AppConfig/${CONFIG_NAME}.json
+    FILES ${CMAKE_BINARY_DIR}/Data/AppConfig/${CONFIG_NAME}
    DESTINATION ${DATA_DIRECTORY}/AppConfig/)
 endforeach()
@@ -0,0 +1,5 @@
+{
+  "Config": {
+    "HideHypervisorBit": "1"
+  }
+}
@@ -0,0 +1,5 @@
+{
+  "Config": {
+    "AdditionalArguments": "--no-sandbox"
+  }
+}
@@ -6,18 +6,10 @@
        "X11"
      ],
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libGL.so",
-        "/usr/lib/x86_64-linux-gnu/libGL.so.1",
-        "/usr/lib/x86_64-linux-gnu/libGL.so.1.2.0",
-        "/usr/lib/x86_64-linux-gnu/libGL.so.1.7.0",
-        "/usr/local/lib/x86_64-linux-gnu/libGL.so",
-        "/usr/local/lib/x86_64-linux-gnu/libGL.so.1",
-        "/usr/local/lib/x86_64-linux-gnu/libGL.so.1.2.0",
-        "/usr/local/lib/x86_64-linux-gnu/libGL.so.1.7.0",
-        "/lib/x86_64-linux-gnu/libGL.so",
-        "/lib/x86_64-linux-gnu/libGL.so.1",
-        "/lib/x86_64-linux-gnu/libGL.so.1.2.0",
-        "/lib/x86_64-linux-gnu/libGL.so.1.7.0"
+        "@PREFIX_LIB@/libGL.so",
+        "@PREFIX_LIB@/libGL.so.1",
+        "@PREFIX_LIB@/libGL.so.1.2.0",
+        "@PREFIX_LIB@/libGL.so.1.7.0"
      ]
    },
    "GLESv2": {
@@ -26,322 +18,170 @@
        "X11"
      ],
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libGLESv2.so",
-        "/usr/lib/x86_64-linux-gnu/libGLESv2.so.2",
-        "/usr/lib/x86_64-linux-gnu/libGLESv2.so.2.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libGLESv2.so",
-        "/usr/local/lib/x86_64-linux-gnu/libGLESv2.so.2",
-        "/usr/local/lib/x86_64-linux-gnu/libGLESv2.so.2.0.0",
-        "/lib/x86_64-linux-gnu/libGLESv2.so",
-        "/lib/x86_64-linux-gnu/libGLESv2.so.2",
-        "/lib/x86_64-linux-gnu/libGLESv2.so.2.0.0"
+        "@PREFIX_LIB@/libGLESv2.so",
+        "@PREFIX_LIB@/libGLESv2.so.2",
+        "@PREFIX_LIB@/libGLESv2.so.2.0.0"
      ]
    },
    "X11": {
      "Library": "libX11-guest.so",
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libX11.so",
-        "/usr/lib/x86_64-linux-gnu/libX11.so.6",
-        "/usr/lib/x86_64-linux-gnu/libX11.so.6.4.0",
-        "/usr/local/lib/x86_64-linux-gnu/libX11.so",
-        "/usr/local/lib/x86_64-linux-gnu/libX11.so.6",
-        "/usr/local/lib/x86_64-linux-gnu/libX11.so.6.4.0",
-        "/lib/x86_64-linux-gnu/libX11.so",
-        "/lib/x86_64-linux-gnu/libX11.so.6",
-        "/lib/x86_64-linux-gnu/libX11.so.6.4.0"
+        "@PREFIX_LIB@/libX11.so",
+        "@PREFIX_LIB@/libX11.so.6",
+        "@PREFIX_LIB@/libX11.so.6.4.0"
      ]
    },
-    "Vulkan-radeon": {
-      "Library": "libvulkan_radeon-guest.so",
+    "Vulkan": {
+      "Library": "libvulkan-guest.so",
      "Depends": [
        "xcb"
      ],
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_radeon.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_radeon.so",
-        "/lib/x86_64-linux-gnu/libvulkan_radeon.so"
+        "@PREFIX_LIB@/libvulkan.so",
+        "@PREFIX_LIB@/libvulkan.so.1",
+        "@HOME@/.local/share/Steam/ubuntu12_32/steam-runtime/pinned_libs_64/libvulkan.so.1"
      ],
      "Comment": [
        "Vulkan library relies on xcb, otherwise it crashes with jemalloc"
      ]
    },
-    "Vulkan-lavapipe": {
-      "Library": "libvulkan_lvp-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_lvp.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_lvp.so",
-        "/lib/x86_64-linux-gnu/libvulkan_lvp.so"
-      ]
-    },
-    "Vulkan-freedreno": {
-      "Library": "libvulkan_freedreno-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_freedreno.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_freedreno.so",
-        "/lib/x86_64-linux-gnu/libvulkan_freedreno.so"
-      ]
-    },
-    "Vulkan-intel": {
-      "Library": "libvulkan_intel-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_intel.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_intel.so",
-        "/lib/x86_64-linux-gnu/libvulkan_intel.so"
-      ]
-    },
-    "Vulkan-panfrost": {
-      "Library": "libvulkan_panfrost-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_panfrost.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_panfrost.so",
-        "/lib/x86_64-linux-gnu/libvulkan_panfrost.so"
-      ]
-    },
-    "Vulkan-nvidia": {
-      "Library": "libvulkan_nvidia-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libGLX_nvidia.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libGLX_nvidia.so.0",
-        "/lib/x86_64-linux-gnu/libGLX_nvidia.so.0"
-      ],
-      "Comment": [
-        "Not currently wired up"
-      ]
-    },
-    "Vulkan-virtio": {
-      "Library": "libvulkan_virtio-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_virtio.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_virtio.so",
-        "/lib/x86_64-linux-gnu/libvulkan_virtio.so"
-      ]
-    },
    "xcb": {
+      "Depends": [
+        "X11"
+      ],
      "Library": "libxcb-guest.so",
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb.so.1",
-        "/usr/lib/x86_64-linux-gnu/libxcb.so.1.1.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb.so.1",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb.so.1.1.0",
-        "/lib/x86_64-linux-gnu/libxcb.so",
-        "/lib/x86_64-linux-gnu/libxcb.so.1",
-        "/lib/x86_64-linux-gnu/libxcb.so.1.1.0"
+        "@PREFIX_LIB@/libxcb.so",
+        "@PREFIX_LIB@/libxcb.so.1",
+        "@PREFIX_LIB@/libxcb.so.1.1.0"
      ]
    },
    "xcb-dri2": {
-      "Library": "libxcb_dri2-guest.so",
+      "Library": "libxcb-dri2-guest.so",
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri2.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri2.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri2.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri2.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri2.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri2.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-dri2.so",
-        "/lib/x86_64-linux-gnu/libxcb-dri2.so.0",
-        "/lib/x86_64-linux-gnu/libxcb-dri2.so.0.0.0"
+        "@PREFIX_LIB@/libxcb-dri2.so",
+        "@PREFIX_LIB@/libxcb-dri2.so.0",
+        "@PREFIX_LIB@/libxcb-dri2.so.0.0.0"
      ]
    },
    "xcb-dri3": {
-      "Library": "libxcb_dri3-guest.so",
+      "Library": "libxcb-dri3-guest.so",
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri3.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri3.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri3.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri3.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri3.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri3.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-dri3.so",
-        "/lib/x86_64-linux-gnu/libxcb-dri3.so.0",
-        "/lib/x86_64-linux-gnu/libxcb-dri3.so.0.0.0"
+        "@PREFIX_LIB@/libxcb-dri3.so",
+        "@PREFIX_LIB@/libxcb-dri3.so.0",
+        "@PREFIX_LIB@/libxcb-dri3.so.0.0.0"
      ]
    },
    "xcb-xfixes": {
-      "Library": "libxcb_xfixes-guest.so",
+      "Library": "libxcb-xfixes-guest.so",
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-xfixes.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-xfixes.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-xfixes.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-xfixes.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-xfixes.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-xfixes.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-xfixes.so",
-        "/lib/x86_64-linux-gnu/libxcb-xfixes.so.0",
-        "/lib/x86_64-linux-gnu/libxcb-xfixes.so.0.0.0"
+        "@PREFIX_LIB@/libxcb-xfixes.so",
+        "@PREFIX_LIB@/libxcb-xfixes.so.0",
+        "@PREFIX_LIB@/libxcb-xfixes.so.0.0.0"
      ]
    },
    "xcb-shm": {
-      "Library": "libxcb_shm-guest.so",
+      "Library": "libxcb-shm-guest.so",
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-shm.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-shm.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-shm.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-shm.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-shm.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-shm.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-shm.so",
-        "/lib/x86_64-linux-gnu/libxcb-shm.so.0",
-        "/lib/x86_64-linux-gnu/libxcb-shm.so.0.0.0"
+        "@PREFIX_LIB@/libxcb-shm.so",
+        "@PREFIX_LIB@/libxcb-shm.so.0",
+        "@PREFIX_LIB@/libxcb-shm.so.0.0.0"
      ]
    },
    "xcb-sync": {
-      "Library": "libxcb_sync-guest.so",
+      "Library": "libxcb-sync-guest.so",
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-sync.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-sync.so.1",
-        "/usr/lib/x86_64-linux-gnu/libxcb-sync.so.1.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-sync.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-sync.so.1",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-sync.so.1.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-sync.so",
-        "/lib/x86_64-linux-gnu/libxcb-sync.so.1",
-        "/lib/x86_64-linux-gnu/libxcb-sync.so.1.0.0"
+        "@PREFIX_LIB@/libxcb-sync.so",
+        "@PREFIX_LIB@/libxcb-sync.so.1",
+        "@PREFIX_LIB@/libxcb-sync.so.1.0.0"
      ]
    },
    "xcb-randr": {
-      "Library": "libxcb_randr-guest.so",
+      "Library": "libxcb-randr-guest.so",
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-randr.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-randr.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-randr.so.0.1.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-randr.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-randr.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-randr.so.0.1.0",
-        "/lib/x86_64-linux-gnu/libxcb-randr.so",
-        "/lib/x86_64-linux-gnu/libxcb-randr.so.0",
-        "/lib/x86_64-linux-gnu/libxcb-randr.so.0.1.0"
+        "@PREFIX_LIB@/libxcb-randr.so",
+        "@PREFIX_LIB@/libxcb-randr.so.0",
+        "@PREFIX_LIB@/libxcb-randr.so.0.1.0"
      ]
    },
    "xcb-present": {
-      "Library": "libxcb_present-guest.so",
+      "Library": "libxcb-present-guest.so",
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-present.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-present.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-present.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-present.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-present.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-present.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-present.so",
-        "/lib/x86_64-linux-gnu/libxcb-present.so.0",
-        "/lib/x86_64-linux-gnu/libxcb-present.so.0.0.0"
+        "@PREFIX_LIB@/libxcb-present.so",
+        "@PREFIX_LIB@/libxcb-present.so.0",
+        "@PREFIX_LIB@/libxcb-present.so.0.0.0"
      ]
    },
    "xcb-glx": {
-      "Library": "libxcb_glx-guest.so",
+      "Library": "libxcb-glx-guest.so",
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-glx.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-glx.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-glx.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-glx.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-glx.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-glx.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-glx.so",
-        "/lib/x86_64-linux-gnu/libxcb-glx.so.0",
-        "/lib/x86_64-linux-gnu/libxcb-glx.so.0.0.0"
+        "@PREFIX_LIB@/libxcb-glx.so",
+        "@PREFIX_LIB@/libxcb-glx.so.0",
+        "@PREFIX_LIB@/libxcb-glx.so.0.0.0"
      ]
    },
    "xshmfence": {
-      "Library": "libshmfence-guest.so",
+      "Library": "libxshmfence-guest.so",
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxshmfence.so",
-        "/usr/lib/x86_64-linux-gnu/libxshmfence.so.1",
-        "/usr/lib/x86_64-linux-gnu/libxshmfence.so.1.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxshmfence.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxshmfence.so.1",
-        "/usr/local/lib/x86_64-linux-gnu/libxshmfence.so.1.0.0",
-        "/lib/x86_64-linux-gnu/libxshmfence.so",
-        "/lib/x86_64-linux-gnu/libxshmfence.so.1",
-        "/lib/x86_64-linux-gnu/libxshmfence.so.1.0.0"
+        "@PREFIX_LIB@/libxshmfence.so",
+        "@PREFIX_LIB@/libxshmfence.so.1",
+        "@PREFIX_LIB@/libxshmfence.so.1.0.0"
      ]
    },
    "drm": {
      "Library": "libdrm-guest.so",
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libdrm.so",
-        "/usr/lib/x86_64-linux-gnu/libdrm.so.2",
-        "/usr/lib/x86_64-linux-gnu/libdrm.so.2.4.0",
-        "/usr/local/lib/x86_64-linux-gnu/libdrm.so",
-        "/usr/local/lib/x86_64-linux-gnu/libdrm.so.2",
-        "/usr/local/lib/x86_64-linux-gnu/libdrm.so.2.4.0",
-        "/lib/x86_64-linux-gnu/libdrm.so",
-        "/lib/x86_64-linux-gnu/libdrm.so.2",
-        "/lib/x86_64-linux-gnu/libdrm.so.2.4.0"
+        "@PREFIX_LIB@/libdrm.so",
+        "@PREFIX_LIB@/libdrm.so.2",
+        "@PREFIX_LIB@/libdrm.so.2.4.0"
      ]
    },
    "asound": {
      "Library": "libasound-guest.so",
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libasound.so",
-        "/usr/lib/x86_64-linux-gnu/libasound.so.2",
-        "/usr/lib/x86_64-linux-gnu/libasound.so.2.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libasound.so",
-        "/usr/local/lib/x86_64-linux-gnu/libasound.so.2",
-        "/usr/local/lib/x86_64-linux-gnu/libasound.so.2.0.0",
-        "/lib/x86_64-linux-gnu/libasound.so",
-        "/lib/x86_64-linux-gnu/libasound.so.2",
-        "/lib/x86_64-linux-gnu/libasound.so.2.0.0"
+        "@PREFIX_LIB@/libasound.so",
+        "@PREFIX_LIB@/libasound.so.2",
+        "@PREFIX_LIB@/libasound.so.2.0.0"
      ]
    },
    "Xrender": {
      "Library": "libXrender-guest.so",
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libXrender.so",
-        "/usr/lib/x86_64-linux-gnu/libXrender.so.1",
-        "/usr/lib/x86_64-linux-gnu/libXrender.so.1.3.0",
-        "/usr/local/lib/x86_64-linux-gnu/libXrender.so",
-        "/usr/local/lib/x86_64-linux-gnu/libXrender.so.1",
-        "/usr/local/lib/x86_64-linux-gnu/libXrender.so.1.3.0",
-        "/lib/x86_64-linux-gnu/libXrender.so",
-        "/lib/x86_64-linux-gnu/libXrender.so.1",
-        "/lib/x86_64-linux-gnu/libXrender.so.1.3.0"
+        "@PREFIX_LIB@/libXrender.so",
+        "@PREFIX_LIB@/libXrender.so.1",
+        "@PREFIX_LIB@/libXrender.so.1.3.0"
      ]
    },
    "Xext": {
      "Library": "libXext-guest.so",
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libXext.so",
-        "/usr/lib/x86_64-linux-gnu/libXext.so.6",
-        "/usr/lib/x86_64-linux-gnu/libXext.so.6.4.0",
-        "/usr/local/lib/x86_64-linux-gnu/libXext.so",
-        "/usr/local/lib/x86_64-linux-gnu/libXext.so.6",
-        "/usr/local/lib/x86_64-linux-gnu/libXext.so.6.4.0",
-        "/lib/x86_64-linux-gnu/libXext.so",
-        "/lib/x86_64-linux-gnu/libXext.so.6",
-        "/lib/x86_64-linux-gnu/libXext.so.6.4.0"
+        "@PREFIX_LIB@/libXext.so",
+        "@PREFIX_LIB@/libXext.so.6",
+        "@PREFIX_LIB@/libXext.so.6.4.0"
      ]
    },
    "Xfixes": {
      "Library": "libXfixes-guest.so",
      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libXfixes.so",
-        "/usr/lib/x86_64-linux-gnu/libXfixes.so.3",
-        "/usr/lib/x86_64-linux-gnu/libXfixes.so.3.1.0",
-        "/usr/local/lib/x86_64-linux-gnu/libXfixes.so",
-        "/usr/local/lib/x86_64-linux-gnu/libXfixes.so.3",
-        "/usr/local/lib/x86_64-linux-gnu/libXfixes.so.3.1.0",
-        "/lib/x86_64-linux-gnu/libXfixes.so",
-        "/lib/x86_64-linux-gnu/libXfixes.so.3",
-        "/lib/x86_64-linux-gnu/libXfixes.so.3.1.0"
+        "@PREFIX_LIB@/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": [
+        "@PREFIX_LIB@/libwayland-client.so",
+        "@PREFIX_LIB@/libwayland-client.so.0",
+        "@PREFIX_LIB@/libwayland-client.so.0.20.0"
      ]
    },
    "":{}
@@ -9,12 +9,19 @@ if (CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64")
  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mcx16")
 endif()

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

-set(ENABLE_JIT_X86_64 ${_M_X86_64} CACHE BOOL "Enable the x86_64 JIT")
-set(ENABLE_JIT_ARM64 ${_M_ARM_64} CACHE BOOL "Enable the ARM64 JIT")
+if (ENABLE_VIXL_SIMULATOR)
+  # If the vixl simulator is enabled then we are using the ARM64 JIT
+  option(ENABLE_JIT_X86_64 "Enable the x86_64 JIT" FALSE)
+  option(ENABLE_JIT_ARM64 "Enable the ARM64 JIT" TRUE)
+else()
+  option(ENABLE_JIT_X86_64 "Enable the x86_64 JIT" ${_M_X86_64})
+  option(ENABLE_JIT_ARM64 "Enable the ARM64 JIT" ${_M_ARM_64})
+endif()
+
 option(ENABLE_CLANG_FORMAT "Run clang format over the source" FALSE)

 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
@@ -27,7 +34,6 @@ set(CMAKE_INCLUDE_CURRENT_DIR ON)
 include(CheckCXXCompilerFlag)
 include(CheckIncludeFileCXX)

-
 if (EXISTS ${CMAKE_CURRENT_DIR}/External/vixl/)
    # Useful to have for freestanding libFEXCore
    add_subdirectory(External/vixl/)
@@ -46,14 +52,14 @@ if (OVERRIDE_VERSION STREQUAL "detect")

  if (GIT_FOUND)
    execute_process(
-      COMMAND ${GIT_EXECUTABLE} rev-parse --short HEAD
+      COMMAND ${GIT_EXECUTABLE} rev-parse --short=7 HEAD
      WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
      OUTPUT_VARIABLE GIT_SHORT_HASH
      ERROR_QUIET
      OUTPUT_STRIP_TRAILING_WHITESPACE
    )
    execute_process(
-      COMMAND ${GIT_EXECUTABLE} describe
+      COMMAND ${GIT_EXECUTABLE} describe --abbrev=7
      WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
      OUTPUT_VARIABLE GIT_DESCRIBE_STRING
      ERROR_QUIET
@@ -71,8 +77,13 @@ configure_file(

 include_directories(${CMAKE_BINARY_DIR}/generated)

+add_compile_options(-fno-exceptions)
 add_subdirectory(Source/)

 install (DIRECTORY include/FEXCore ${CMAKE_BINARY_DIR}/include/FEXCore
  DESTINATION include
  COMPONENT Development)
+
+if (BUILD_TESTS)
+  add_subdirectory(unittests/)
+endif()
@@ -18,14 +18,8 @@ This project aims to provide a fast and functional x86-64 emulation library that
 * Portable library implementation in order to support easy integration in to applications
 ### Target Host Architecture
 The target host architecture for this library is AArch64. Specifically the ARMv8.1 version or newer.
-The CPU IR is designed with AArch64 in mind but there is a desire to run the recompiled code on other architectures as well.
-Multiple architecture support is desired for easier bringup and debugging, performance isn't as much of a priority there (ex. x86-64(guest) translated to x86-64(host))
-### Not currently goals but will be in the future
-* 32bit x86 support
-  * This will be a desire in the future, but to lower the amount of work required, decided to push this off for now.
-* Integration in to WINE
-* Later generation of x86-64 instruction sets
-  * Including AVX, F16C, XOP, FMA, AVX2, etc
+The CPU IR is designed with AArch64 in mind but should allow for other architectures as well.
+x86-64 host support is available for ease of development, but is not a priority.
 ### Not desired
 * Kernel space emulation
 * CPL0-2 emulation
@@ -33,7 +27,3 @@ Multiple architecture support is desired for easier bringup and debugging, perfo
 * IRQs
 * SVM
 * "Cycle Accurate" emulation
-### Dependencies
- * clang-tidy if you want to ensure the code stays tidy
- * cmake
- * A C++17 compliant compiler (There are assumptions made about using Clang and LTO)
@@ -22,12 +22,14 @@ def print_header():
 #define OPT_UINT64(group, enum, json, default) OPT_BASE(uint64_t, group, enum, json, default)
 #endif
 #ifndef OPT_STR
-#define OPT_STR(group, enum, json, default) OPT_BASE(std::string, group, enum, json, default)
+#define OPT_STR(group, enum, json, default) OPT_BASE(fextl::string, group, enum, json, default)
 #endif
 #ifndef OPT_STRARRAY
-#define OPT_STRARRAY(group, enum, json, default) OPT_BASE(std::string, group, enum, json, default)
+#define OPT_STRARRAY(group, enum, json, default) OPT_BASE(fextl::string, group, enum, json, default)
+#endif
+#ifndef OPT_STRENUM
+#define OPT_STRENUM(group, enum, json, default) OPT_BASE(uint64_t, group, enum, json, default)
 #endif
-
 '''
    output_file.write(header)

@@ -40,6 +42,7 @@ def print_tail():
 #undef OPT_UINT64
 #undef OPT_STR
 #undef OPT_STRARRAY
+#undef OPT_STRENUM
 '''
    output_file.write(tail)

@@ -127,12 +130,13 @@ def print_man_options(options):
                short = op_vals["ShortArg"]

            default = op_vals["Default"]
+            value_type = op_vals["Type"]

            # Textual default rather than enum based
            if ("TextDefault" in op_vals):
                default = op_vals["TextDefault"]

-            if (op_vals["Type"] == "str" or op_vals["Type"] == "strarray"):
+            if (value_type == "str" or value_type == "strarray" or value_type == "strenum"):
                # Wrap the string argument in quotes
                default = "'" + default + "'"
            print_man_option(
@@ -141,6 +145,12 @@ def print_man_options(options):
                op_vals["Desc"],
                default
            )
+            if (value_type == "strenum"):
+                Enums = op_vals["Enums"]
+                output_man.write("\\fBAvailable Options:\\fR\n")
+                for enum_op_key, enum_op_vals in Enums.items():
+                    output_man.write("{}, ".format(enum_op_vals))
+                output_man.write("\n")

    output_man.write(".El\n")

@@ -150,12 +160,13 @@ def print_man_environment(options):
    for op_group, group_vals in options.items():
        for op_key, op_vals in group_vals.items():
            default = op_vals["Default"]
+            value_type = op_vals["Type"]

            # Textual default rather than enum based
            if ("TextDefault" in op_vals):
                default = op_vals["TextDefault"]

-            if (op_vals["Type"] == "str" or op_vals["Type"] == "strarray"):
+            if (value_type == "str" or value_type == "strarray" or value_type == "strenum"):
                # Wrap the string argument in quotes
                default = "'" + default + "'"
            print_man_env_option(
@@ -165,6 +176,13 @@ def print_man_environment(options):
                False
            )

+            if (value_type == "strenum"):
+                Enums = op_vals["Enums"]
+                output_man.write("\\fBAvailable Options:\\fR\n")
+                for enum_op_key, enum_op_vals in Enums.items():
+                    output_man.write("{}, ".format(enum_op_vals))
+                output_man.write("\n")
+
    print_man_environment_tail()
    output_man.write(".El\n")

@@ -334,7 +352,7 @@ def print_argloader_options(options):
        for op_key, op_vals in group_vals.items():
            default = op_vals["Default"]

-            if (op_vals["Type"] == "str" or op_vals["Type"] == "strarray"):
+            if (op_vals["Type"] == "str" or op_vals["Type"] == "strarray" or op_vals["Type"] == "strenum"):
                # Wrap the string argument in quotes
                default = "\"" + default + "\""

@@ -371,15 +389,21 @@ def print_parse_argloader_options(options):

            value_type = op_vals["Type"]
            NeedsString = False
-            conversion_func = "std::to_string"
+            conversion_func = "fextl::fmt::format(\"{}\", "
            if ("ArgumentHandler" in op_vals):
                NeedsString = True
-                conversion_func = "FEXCore::Config::Handler::{0}".format(op_vals["ArgumentHandler"])
+                conversion_func = "FEXCore::Config::Handler::{0}(".format(op_vals["ArgumentHandler"])
            if (value_type == "str"):
                NeedsString = True
-                conversion_func = ""
+                conversion_func = "("
+            if (value_type == "bool"):
+                # boolean values need a decimal specifier. Otherwise fmt prints strings.
+                conversion_func = "fextl::fmt::format(\"{:d}\", "

-            if (value_type == "strarray"):
+            if (value_type == "strenum"):
+                output_argloader.write("\tfextl::string UserValue = Options[\"{0}\"];\n".format(op_key))
+                output_argloader.write("\tSet(FEXCore::Config::ConfigOption::CONFIG_{}, FEXCore::Config::EnumParser(FEXCore::Config::{}_EnumPairs, UserValue));\n".format(op_key.upper(), op_key, op_key))
+            elif (value_type == "strarray"):
                # these need a bit more help
                output_argloader.write("\tauto Array = Options.all(\"{0}\");\n".format(op_key))
                output_argloader.write("\tfor (auto iter = Array.begin(); iter != Array.end(); ++iter) {\n")
@@ -387,11 +411,11 @@ def print_parse_argloader_options(options):
                output_argloader.write("\t}\n")
            else:
                if (NeedsString):
-                    output_argloader.write("\tstd::string UserValue = Options[\"{0}\"];\n".format(op_key))
+                    output_argloader.write("\tfextl::string UserValue = Options[\"{0}\"];\n".format(op_key))
                else:
                    output_argloader.write("\t{0} UserValue = Options.get(\"{1}\");\n".format(value_type, op_key))

-                output_argloader.write("\tSet(FEXCore::Config::ConfigOption::CONFIG_{0}, {1}(UserValue));\n".format(op_key.upper(), conversion_func))
+                output_argloader.write("\tSet(FEXCore::Config::ConfigOption::CONFIG_{0}, {1}UserValue));\n".format(op_key.upper(), conversion_func))
            output_argloader.write("}\n")

    output_argloader.write("#endif\n")
@@ -404,6 +428,12 @@ def print_parse_envloader_options(options):

    for op_group, group_vals in options.items():
        for op_key, op_vals in group_vals.items():
+            value_type = op_vals["Type"]
+            if (value_type == "strenum"):
+                output_argloader.write("else if (Key == \"FEX_{0}\") {{\n".format(op_key.upper()))
+                output_argloader.write("Value = FEXCore::Config::EnumParser(FEXCore::Config::{}_EnumPairs, Value);\n".format(op_key, op_key))
+                output_argloader.write("}\n")
+
            if ("ArgumentHandler" in op_vals):
                conversion_func = "FEXCore::Config::Handler::{0}".format(op_vals["ArgumentHandler"])
                output_argloader.write("else if (Key == \"FEX_{0}\") {{\n".format(op_key.upper()))
@@ -411,6 +441,41 @@ def print_parse_envloader_options(options):
                output_argloader.write("}\n")
    output_argloader.write("#endif\n")

+def print_parse_enum_options(options):
+    output_argloader.write("#ifdef ENUMDEFINES\n")
+    output_argloader.write("#undef ENUMDEFINES\n")
+    for op_group, group_vals in options.items():
+        for op_key, op_vals in group_vals.items():
+            if (op_vals["Type"] == "strenum"):
+                output_argloader.write("enum {} : uint64_t {{\n".format(op_key))
+                Enums = op_vals["Enums"]
+                i = 0
+                # Always have an OFF.
+                output_argloader.write("\tOFF = 0,\n")
+                for enum_op_key, enum_op_vals in Enums.items():
+                    output_argloader.write("\t{} = 1ULL << {},\n".format(enum_op_key.upper(), i))
+                    i += 1
+
+                output_argloader.write("};\n")
+
+    for op_group, group_vals in options.items():
+        for op_key, op_vals in group_vals.items():
+            if (op_vals["Type"] == "strenum"):
+                Enums = op_vals["Enums"]
+
+                output_argloader.write("using {}ConfigPair = std::pair<std::string_view, FEXCore::Config::{}>;\n".format(op_key, op_key))
+                output_argloader.write("constexpr static std::array<{}ConfigPair, {}> {}_EnumPairs = {{{{\n".format(op_key, len(Enums) + 1, op_key))
+                i = 0
+                # Always have an OFF.
+                output_argloader.write("\t{{ \"off\", FEXCore::Config::{}::OFF }},\n".format(op_key))
+                for enum_op_key, enum_op_vals in Enums.items():
+                    output_argloader.write("\t{{ \"{}\", FEXCore::Config::{}::{} }},\n".format(enum_op_vals, op_key, enum_op_key.upper()))
+                    i += 1
+
+                output_argloader.write("}};\n")
+
+    output_argloader.write("#endif\n")
+
 def check_for_duplicate_options(options):
    short_map = []
    long_map = []
@@ -489,4 +554,7 @@ print_parse_argloader_options(options);
 # Generate environment loader code
 print_parse_envloader_options(options);

+# Generate enum variable options
+print_parse_enum_options(options);
+
 output_argloader.close()
@@ -7,17 +7,12 @@ OpClasses = collections.OrderedDict()
 def get_ir_classes(ops, defines):
    global OpClasses

-    for op_key, op_vals in ops.items():
-        if not ("Last" in op_vals):
-            OpClass = "#Unknown"
+    for op_class, opslist in ops.items():
+        if not (op_class in OpClasses):
+            OpClasses[op_class] = []

-            if ("OpClass" in op_vals):
-                OpClass = op_vals["OpClass"]
-
-            if not (OpClass in OpClasses):
-                OpClasses[OpClass] = []
-
-            OpClasses[OpClass].append([op_key, op_vals])
+        for op, op_val in opslist.items():
+            OpClasses[op_class].append([op, op_val])

    # Sort the dictionary after we are done parsing it
    OpClasses = collections.OrderedDict(sorted(OpClasses.items()))
@@ -38,41 +33,9 @@ def print_ir_ops():
            op_key = op[0]
            op_vals = op[1]
            output_file.write("## %s\n" % (op_key))
-            HasDest = ("HasDest" in op_vals and op_vals["HasDest"] == True)
-            HasSSAArgs = ("SSAArgs" in op_vals and len(op_vals["SSAArgs"]) > 0)
-            HasSSAArgNames = "SSANames" in op_vals
-            HasArgs = "Args" in op_vals
-            SSAArgsCount = 0
-            ArgCount = 0
-            if (HasSSAArgs):
-                SSAArgsCount = int(op_vals["SSAArgs"])
-            if (HasArgs):
-                ArgCount = len(op_vals["Args"])
-
-            TotalArgsCount = SSAArgsCount + (ArgCount / 2)

            output_file.write(">")
-            if (HasDest):
-                output_file.write("%dest = ")
-
-            output_file.write("%s " % op_key)
-
-            ArgComma = (", ", "")
-            if (HasSSAArgs):
-                for i in range(0, SSAArgsCount):
-                    FinalArg = (i + 1) == TotalArgsCount
-                    if (HasSSAArgNames):
-                        output_file.write("%%%s%s" % (op_vals["SSANames"][i], ArgComma[FinalArg]))
-                    else:
-                        output_file.write("%%ssa%d%s" % (i, ArgComma[FinalArg]))
-
-            if (HasArgs):
-                Args = op_vals["Args"]
-                for i in range(0, ArgCount, 2):
-                    FinalArg = ((i / 2) + SSAArgsCount + 1) == TotalArgsCount
-                    data_type = Args[i]
-                    data_name = Args[i + 1]
-                    output_file.write("\<%s %s\>%s" % (data_type, data_name, ArgComma[FinalArg]))
+            output_file.write(op_key)

            output_file.write("\n\n")

@@ -1,13 +1,19 @@
-set (MAN_DIR ${CMAKE_INSTALL_PREFIX}/share/man CACHE PATH "MAN_DIR")
+set (MAN_DIR share/man CACHE PATH "MAN_DIR")

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

+if (NOT MINGW_BUILD)
+  list(APPEND FEXCORE_BASE_SRCS
+    Utils/Allocator/64BitAllocator.cpp)
+endif()
+
 set (SRCS
  Common/JitSymbols.cpp
  Common/SoftFloat-3e/extF80_add.c
@@ -80,39 +86,30 @@ set (SRCS
  Interface/Context/Context.cpp
  Interface/Core/LookupCache.cpp
  Interface/Core/BlockSamplingData.cpp
-  Interface/Core/CompileService.cpp
  Interface/Core/Core.cpp
+  Interface/Core/CPUBackend.cpp
  Interface/Core/CPUID.cpp
  Interface/Core/Frontend.cpp
  Interface/Core/GdbServer.cpp
  Interface/Core/HostFeatures.cpp
+  Interface/Core/ObjectCache/JobHandling.cpp
+  Interface/Core/ObjectCache/NamedRegionObjectHandler.cpp
+  Interface/Core/ObjectCache/ObjectCacheService.cpp
  Interface/Core/OpcodeDispatcher/Crypto.cpp
  Interface/Core/OpcodeDispatcher/Flags.cpp
  Interface/Core/OpcodeDispatcher/Vector.cpp
  Interface/Core/OpcodeDispatcher/X87.cpp
+  Interface/Core/OpcodeDispatcher/X87F64.cpp
  Interface/Core/OpcodeDispatcher.cpp
  Interface/Core/SignalDelegator.cpp
  Interface/Core/X86Tables.cpp
  Interface/Core/X86DebugInfo.cpp
  Interface/Core/X86HelperGen.cpp
-  Interface/Core/ArchHelpers/Arm64_stubs.cpp
  Interface/Core/ArchHelpers/Arm64Emitter.cpp
  Interface/Core/Dispatcher/Dispatcher.cpp
  Interface/Core/Dispatcher/X86Dispatcher.cpp
  Interface/Core/Dispatcher/Arm64Dispatcher.cpp
-  Interface/Core/Interpreter/InterpreterCore.cpp
-  Interface/Core/Interpreter/InterpreterOps.cpp
-  Interface/Core/Interpreter/ALUOps.cpp
-  Interface/Core/Interpreter/AtomicOps.cpp
-  Interface/Core/Interpreter/BranchOps.cpp
-  Interface/Core/Interpreter/ConversionOps.cpp
-  Interface/Core/Interpreter/EncryptionOps.cpp
-  Interface/Core/Interpreter/F80Ops.cpp
-  Interface/Core/Interpreter/FlagOps.cpp
-  Interface/Core/Interpreter/MemoryOps.cpp
-  Interface/Core/Interpreter/MiscOps.cpp
-  Interface/Core/Interpreter/MoveOps.cpp
-  Interface/Core/Interpreter/VectorOps.cpp
+  Interface/Core/Interpreter/Fallbacks/InterpreterFallbacks.cpp
  Interface/Core/X86Tables/BaseTables.cpp
  Interface/Core/X86Tables/DDDTables.cpp
  Interface/Core/X86Tables/EVEXTables.cpp
@@ -126,6 +123,7 @@ set (SRCS
  Interface/Core/X86Tables/X87Tables.cpp
  Interface/Core/X86Tables/XOPTables.cpp
  Interface/HLE/Thunks/Thunks.cpp
+  Interface/GDBJIT/GDBJIT.cpp
  Interface/IR/AOTIR.cpp
  Interface/IR/IRDumper.cpp
  Interface/IR/IRParser.cpp
@@ -142,19 +140,40 @@ set (SRCS
  Interface/IR/Passes/PhiValidation.cpp
  Interface/IR/Passes/RedundantFlagCalculationElimination.cpp
  Interface/IR/Passes/DeadStoreElimination.cpp
-  Interface/IR/Passes/StaticRegisterAllocationPass.cpp
  Interface/IR/Passes/RegisterAllocationPass.cpp
  Interface/IR/Passes/SyscallOptimization.cpp
-  Utils/Allocator.cpp
-  Utils/Allocator/64BitAllocator.cpp
  Utils/NetStream.cpp
  Utils/Telemetry.cpp
  Utils/Threads.cpp
+  Utils/Profiler.cpp
  )

-if(_M_ARM_64)
+if (_M_ARM_64)
+  list(APPEND SRCS Utils/ArchHelpers/Arm64.cpp)
+else()
+  list(APPEND SRCS Utils/ArchHelpers/Arm64_stubs.cpp)
+endif()
+
+if (ENABLE_GLIBC_ALLOCATOR_HOOK_FAULT)
+  list(APPEND FEXCORE_BASE_SRCS
+    Utils/AllocatorOverride.cpp)
+endif()
+
+if (ENABLE_INTERPRETER)
  list(APPEND SRCS
-    Interface/Core/ArchHelpers/Arm64.cpp)
+    Interface/Core/Interpreter/InterpreterCore.cpp
+    Interface/Core/Interpreter/InterpreterOps.cpp
+    Interface/Core/Interpreter/ALUOps.cpp
+    Interface/Core/Interpreter/AtomicOps.cpp
+    Interface/Core/Interpreter/BranchOps.cpp
+    Interface/Core/Interpreter/ConversionOps.cpp
+    Interface/Core/Interpreter/EncryptionOps.cpp
+    Interface/Core/Interpreter/F80Ops.cpp
+    Interface/Core/Interpreter/FlagOps.cpp
+    Interface/Core/Interpreter/MemoryOps.cpp
+    Interface/Core/Interpreter/MiscOps.cpp
+    Interface/Core/Interpreter/MoveOps.cpp
+    Interface/Core/Interpreter/VectorOps.cpp)
 endif()

 set(DEFINES -DTHREAD_LOCAL=_Thread_local)
@@ -167,6 +186,15 @@ if (_M_ARM_64)
  list(APPEND DEFINES -D_M_ARM_64=1)
 endif()

+if (ENABLE_VIXL_SIMULATOR)
+  # We can run the simulator on both x86-64 or AArch64 hosts
+  list(APPEND DEFINES -DVIXL_SIMULATOR=1 -DVIXL_INCLUDE_SIMULATOR_AARCH64=1)
+endif()
+
+if (ENABLE_VIXL_DISASSEMBLER)
+  list(APPEND DEFINES -DVIXL_DISASSEMBLER=1)
+endif()
+
 if (ENABLE_JIT_X86_64)
  list(APPEND SRCS
    Interface/Core/JIT/x86_64/JIT.cpp
@@ -179,7 +207,9 @@ if (ENABLE_JIT_X86_64)
    Interface/Core/JIT/x86_64/MemoryOps.cpp
    Interface/Core/JIT/x86_64/MiscOps.cpp
    Interface/Core/JIT/x86_64/MoveOps.cpp
-    Interface/Core/JIT/x86_64/VectorOps.cpp)
+    Interface/Core/JIT/x86_64/VectorOps.cpp
+    Interface/Core/JIT/x86_64/x64Relocations.cpp
+    )
  list(APPEND DEFINES -DJIT_X86_64)
 endif()

@@ -196,14 +226,27 @@ if (ENABLE_JIT_ARM64)
    Interface/Core/JIT/Arm64/MemoryOps.cpp
    Interface/Core/JIT/Arm64/MiscOps.cpp
    Interface/Core/JIT/Arm64/MoveOps.cpp
-    Interface/Core/JIT/Arm64/VectorOps.cpp)
+    Interface/Core/JIT/Arm64/VectorOps.cpp
+    Interface/Core/JIT/Arm64/Arm64Relocations.cpp
+    )
+endif()
+
+set (LIBS fmt::fmt vixl xxhash FEXHeaderUtils)
+
+if (NOT MINGW_BUILD)
+  list (APPEND LIBS dl)
+else()
+  list (APPEND LIBS synchronization)
 endif()

-set (LIBS vixl dl xxhash tiny-json)
 if (ENABLE_JEMALLOC)
  list (APPEND LIBS FEX_jemalloc)
 endif()

+if (ENABLE_JEMALLOC_GLIBC_ALLOC)
+  list (APPEND LIBS FEX_jemalloc_glibc)
+endif()
+
 # Generate config
 configure_file(
  ${CMAKE_CURRENT_SOURCE_DIR}/Interface/Config/Config.json.in
@@ -214,13 +257,11 @@ set(OUTPUT_IR_FOLDER "${CMAKE_BINARY_DIR}/include/FEXCore/IR")
 set(OUTPUT_NAME "${OUTPUT_IR_FOLDER}/IRDefines.inc")
 set(INPUT_NAME "${CMAKE_CURRENT_SOURCE_DIR}/Interface/IR/IR.json")

-add_custom_target(CREATE_IR_FOLDER ALL
-  COMMAND ${CMAKE_COMMAND} -E make_directory "${OUTPUT_IR_FOLDER}")
+file(MAKE_DIRECTORY "${OUTPUT_IR_FOLDER}")

 add_custom_command(
  OUTPUT "${OUTPUT_NAME}"
  DEPENDS "${INPUT_NAME}"
-  DEPENDS CREATE_IR_FOLDER
  DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_generator.py"
  COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_generator.py" "${INPUT_NAME}" "${OUTPUT_NAME}"
  )
@@ -234,7 +275,6 @@ set(OUTPUT_IR_DOC "${CMAKE_BINARY_DIR}/IR.md")
 add_custom_command(
  OUTPUT "${OUTPUT_IR_DOC}"
  DEPENDS "${INPUT_NAME}"
-  DEPENDS CREATE_IR_FOLDER
  DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_doc_generator.py"
  COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_doc_generator.py" "${INPUT_NAME}" "${OUTPUT_IR_DOC}"
  )
@@ -255,15 +295,13 @@ set(INPUT_CONFIG_NAME "${CMAKE_BINARY_DIR}/generated/Config/Config.json")
 set(OUTPUT_MAN_NAME "${CMAKE_BINARY_DIR}/generated/FEX.1")
 set(OUTPUT_MAN_NAME_COMPRESS "${CMAKE_BINARY_DIR}/generated/FEX.1.gz")

-add_custom_target(CREATE_CONFIG_FOLDER ALL
-  COMMAND ${CMAKE_COMMAND} -E make_directory "${OUTPUT_CONFIG_FOLDER}")
+file(MAKE_DIRECTORY "${OUTPUT_CONFIG_FOLDER}")

 add_custom_command(
  OUTPUT "${OUTPUT_CONFIG_NAME}"
  OUTPUT "${OUTPUT_CONFIG_OPTION_NAME}"
  OUTPUT "${OUTPUT_MAN_NAME}"
  DEPENDS "${INPUT_CONFIG_NAME}"
-  DEPENDS CREATE_CONFIG_FOLDER
  DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/config_generator.py"
  COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/config_generator.py" "${INPUT_CONFIG_NAME}" "${OUTPUT_CONFIG_NAME}" "${OUTPUT_MAN_NAME}"
  "${OUTPUT_CONFIG_OPTION_NAME}"
@@ -313,7 +351,7 @@ function(AddDefaultOptionsToTarget Name)
  target_include_directories(${Name} PUBLIC "${CMAKE_BINARY_DIR}/include/")

  target_compile_definitions(${Name} PRIVATE ${DEFINES})
-  add_dependencies(${Name} CONFIG_INC)
+  add_dependencies(${Name} CONFIG_INC IR_INC)

  target_compile_options(${Name}
    PRIVATE
@@ -324,6 +362,7 @@ function(AddDefaultOptionsToTarget Name)

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

  if (GCC_COLOR)
@@ -349,14 +388,13 @@ endfunction()

 # Build FEXCore_Config static library
 add_library(FEXCore_Base STATIC ${FEXCORE_BASE_SRCS})
-target_link_libraries(FEXCore_Base fmt::fmt tiny-json)
+target_link_libraries(FEXCore_Base ${LIBS})
 AddDefaultOptionsToTarget(FEXCore_Base)

 function(AddObject Name Type)
  add_library(${Name} ${Type} ${SRCS})
-  add_dependencies(${Name} IR_INC)

-  target_link_libraries(${Name} FEXCore_Base ${LIBS})
+  target_link_libraries(${Name} FEXCore_Base)
  AddDefaultOptionsToTarget(${Name})

  set_target_properties(${Name} PROPERTIES OUTPUT_NAME FEXCore)
@@ -364,8 +402,18 @@ endfunction()

 function(AddLibrary Name Type)
  add_library(${Name} ${Type} $<TARGET_OBJECTS:${PROJECT_NAME}_object>)
-  target_link_libraries(${Name} FEXCore_Base ${LIBS})
+  target_link_libraries(${Name} FEXCore_Base)
  set_target_properties(${Name} PROPERTIES OUTPUT_NAME FEXCore)
+  if (MINGW_BUILD)
+    # Mingw build isn't building a linux shared library, so it can't have a SONAME.
+    set_target_properties(${Name} PROPERTIES NO_SONAME ON)
+    # Change the suffixes otherwise cmake continues using .a and .so
+    if (${Type} STREQUAL SHARED)
+      set_target_properties(${Name} PROPERTIES SUFFIX ".dll")
+    elseif(${Type} STREQUAL STATIC)
+      set_target_properties(${Name} PROPERTIES SUFFIX ".lib")
+    endif()
+  endif()

  AddDefaultOptionsToTarget(${Name})
 endfunction()
@@ -374,10 +422,12 @@ AddObject(${PROJECT_NAME}_object OBJECT)
 AddLibrary(${PROJECT_NAME} STATIC)
 AddLibrary(${PROJECT_NAME}_shared SHARED)

-install(TARGETS ${PROJECT_NAME} ${PROJECT_NAME}_shared
-  LIBRARY
-    DESTINATION lib
-    COMPONENT Libraries
-  ARCHIVE
-    DESTINATION lib
-    COMPONENT Libraries)
+if (NOT MINGW_BUILD)
+  install(TARGETS ${PROJECT_NAME} ${PROJECT_NAME}_shared
+    LIBRARY
+      DESTINATION lib
+      COMPONENT Libraries
+    ARCHIVE
+      DESTINATION lib
+      COMPONENT Libraries)
+endif()
@@ -20,12 +20,12 @@ struct BitSet final {
  ElementType *Memory;
  void Allocate(size_t Elements) {
    size_t AllocateSize = AlignUp(Elements, MinimumSizeBits) / MinimumSize;
-    LOGMAN_THROW_A_FMT((AllocateSize * MinimumSize) >= Elements, "Fail");
+    LOGMAN_THROW_AA_FMT((AllocateSize * MinimumSize) >= Elements, "Fail");
    Memory = static_cast<ElementType*>(FEXCore::Allocator::malloc(AllocateSize));
  }
  void Realloc(size_t Elements) {
    size_t AllocateSize = AlignUp(Elements, MinimumSizeBits) / MinimumSize;
-    LOGMAN_THROW_A_FMT((AllocateSize * MinimumSize) >= Elements, "Fail");
+    LOGMAN_THROW_AA_FMT((AllocateSize * MinimumSize) >= Elements, "Fail");
    Memory = static_cast<ElementType*>(FEXCore::Allocator::realloc(Memory, AllocateSize));
  }
  void Free() {
@@ -64,7 +64,7 @@ struct BitSetView final {
  ElementType *Memory;

  void GetView(BitSet<T> &Set, uint64_t ElementOffset) {
-    LOGMAN_THROW_A_FMT((ElementOffset % MinimumSize) == 0,
+    LOGMAN_THROW_AA_FMT((ElementOffset % MinimumSize) == 0,
                       "Bitset view offset needs to be aligned to size of backing element");
    Memory = &Set.Memory[ElementOffset / MinimumSizeBits];
  }
@@ -1,53 +1,89 @@
+#include <FEXCore/fextl/fmt.h>
+
 #include "Common/JitSymbols.h"

-#include <string>
+#include <fcntl.h>
 #include <unistd.h>

-#include <fmt/format.h>
-
 namespace FEXCore {
-  JITSymbols::JITSymbols() : fp{nullptr, std::fclose} {
-    const auto PerfMap = fmt::format("/tmp/perf-{}.map", getpid());
+  JITSymbols::JITSymbols() {
+  }

-    fp.reset(fopen(PerfMap.c_str(), "wb"));
-    if (fp) {
-      // Disable buffering on this file
-      setvbuf(fp.get(), nullptr, _IONBF, 0);
+  JITSymbols::~JITSymbols() {
+    if (fd != -1) {
+      close(fd);
    }
  }

-  JITSymbols::~JITSymbols() = default;
+  void JITSymbols::InitFile() {
+    // We can't use FILE here since we must be robust against forking processes closing our FD from under us.
+#ifdef __ANDROID__
+    // Android simpleperf looks in /data/local/tmp instead of /tmp
+    const auto PerfMap = fextl::fmt::format("/data/local/tmp/perf-{}.map", getpid());
+#else
+    const auto PerfMap = fextl::fmt::format("/tmp/perf-{}.map", getpid());
+#endif
+    fd = open(PerfMap.c_str(), O_CREAT | O_TRUNC | O_WRONLY | O_APPEND, 0644);
+  }

  void JITSymbols::Register(const void *HostAddr, uint64_t GuestAddr, uint32_t CodeSize) {
-    if (!fp) return;
+    if (fd == -1) return;

    // Linux perf format is very straightforward
    // `<HostPtr> <Size> <Name>\n`
-    fmt::print(fp.get(), "{:x} {:x} JIT_0x{:x}_{:x}\n", HostAddr, CodeSize, GuestAddr, HostAddr);
+    const auto Buffer = fextl::fmt::format("{} {:x} JIT_0x{:x}_{}\n", HostAddr, CodeSize, GuestAddr, HostAddr);
+    auto Result = write(fd, Buffer.c_str(), Buffer.size());
+    if (Result == -1 && errno == EBADF) {
+      fd = -1;
+    }
  }

  void JITSymbols::Register(const void *HostAddr, uint32_t CodeSize, std::string_view Name) {
-    if (!fp) return;
+    if (fd == -1) return;

    // Linux perf format is very straightforward
    // `<HostPtr> <Size> <Name>\n`
-    fmt::print(fp.get(), "{:x} {:x} {}_{:x}\n", HostAddr, CodeSize, Name, HostAddr);
+    const auto Buffer = fextl::fmt::format("{} {:x} {}_{}\n", HostAddr, CodeSize, Name, HostAddr);
+    auto Result = write(fd, Buffer.c_str(), Buffer.size());
+    if (Result == -1 && errno == EBADF) {
+      fd = -1;
+    }
+  }
+
+  void JITSymbols::Register(const void *HostAddr, uint32_t CodeSize, std::string_view Name, uintptr_t Offset) {
+    if (fd == -1) return;
+
+    // Linux perf format is very straightforward
+    // `<HostPtr> <Size> <Name>\n`
+    const auto Buffer = fextl::fmt::format("{} {:x} {}+0x{:x} ({})\n", HostAddr, CodeSize, Name, Offset, HostAddr);
+    auto Result = write(fd, Buffer.c_str(), Buffer.size());
+    if (Result == -1 && errno == EBADF) {
+      fd = -1;
+    }
  }

  void JITSymbols::RegisterNamedRegion(const void *HostAddr, uint32_t CodeSize, std::string_view Name) {
-    if (!fp) return;
+    if (fd == -1) return;

    // Linux perf format is very straightforward
    // `<HostPtr> <Size> <Name>\n`
-    fmt::print(fp.get(), "{:x} {:x} {}\n", HostAddr, CodeSize, Name);
+    const auto Buffer = fextl::fmt::format("{} {:x} {}\n", HostAddr, CodeSize, Name);
+    auto Result = write(fd, Buffer.c_str(), Buffer.size());
+    if (Result == -1 && errno == EBADF) {
+      fd = -1;
+    }
  }

  void JITSymbols::RegisterJITSpace(const void *HostAddr, uint32_t CodeSize) {
-    if (!fp) return;
+    if (fd == -1) return;

    // Linux perf format is very straightforward
    // `<HostPtr> <Size> <Name>\n`
-    fmt::print(fp.get(), "{:x} {:x} FEXJIT\n", HostAddr, CodeSize);
+    const auto Buffer = fextl::fmt::format("{} {:x} FEXJIT\n", HostAddr, CodeSize);
+    auto Result = write(fd, Buffer.c_str(), Buffer.size());
+    if (Result == -1 && errno == EBADF) {
+      fd = -1;
+    }
  }

 } // namespace FEXCore
@@ -11,14 +11,14 @@ public:
  JITSymbols();
  ~JITSymbols();

+  void InitFile();
  void Register(const void *HostAddr, uint64_t GuestAddr, uint32_t CodeSize);
  void Register(const void *HostAddr, uint32_t CodeSize, std::string_view Name);
+  void Register(const void *HostAddr, uint32_t CodeSize, std::string_view Name, uintptr_t Offset);
  void RegisterNamedRegion(const void *HostAddr, uint32_t CodeSize, std::string_view Name);
  void RegisterJITSpace(const void *HostAddr, uint32_t CodeSize);

 private:
-  using FILEPtr = std::unique_ptr<FILE, decltype(&std::fclose)>;
-
-  FILEPtr fp;
+  int fd{-1};
 };
 }
@@ -1,91 +0,0 @@
-#include "Common/Paths.h"
-#include <FEXCore/Utils/LogManager.h>
-
-#include <cstdlib>
-#include <filesystem>
-#include <memory>
-#include <pwd.h>
-#include <system_error>
-#include <unistd.h>
-
-namespace FEXCore::Paths {
-  std::unique_ptr<std::string> CachePath;
-  std::unique_ptr<std::string> EntryCache;
-
-  char const* FindUserHomeThroughUID() {
-    auto passwd = getpwuid(geteuid());
-    if (passwd) {
-      return passwd->pw_dir;
-    }
-    return nullptr;
-  }
-
-  const char *GetHomeDirectory() {
-    char const *HomeDir = getenv("HOME");
-
-    // Try to get home directory from uid
-    if (!HomeDir) {
-      HomeDir = FindUserHomeThroughUID();
-    }
-
-    // try the PWD
-    if (!HomeDir) {
-      HomeDir = getenv("PWD");
-    }
-
-    // Still doesn't exit? You get local
-    if (!HomeDir) {
-      HomeDir = ".";
-    }
-
-    return HomeDir;
-  }
-
-  void InitializePaths() {
-    CachePath = std::make_unique<std::string>();
-    EntryCache = std::make_unique<std::string>();
-
-    char const *HomeDir = getenv("HOME");
-
-    if (!HomeDir) {
-      HomeDir = getenv("PWD");
-    }
-
-    if (!HomeDir) {
-      HomeDir = ".";
-    }
-
-    char *XDGDataDir = getenv("XDG_DATA_DIR");
-    if (XDGDataDir) {
-      *CachePath = XDGDataDir;
-    }
-    else {
-      if (HomeDir) {
-        *CachePath = HomeDir;
-      }
-    }
-
-    *CachePath += "/.fex-emu/";
-    *EntryCache = *CachePath + "/EntryCache/";
-
-    std::error_code ec{};
-    // Ensure the folder structure is created for our Data
-    if (!std::filesystem::exists(*EntryCache, ec) &&
-        !std::filesystem::create_directories(*EntryCache, ec)) {
-      LogMan::Msg::DFmt("Couldn't create EntryCache directory: '{}'", *EntryCache);
-    }
-  }
-
-  void ShutdownPaths() {
-    CachePath.reset();
-    EntryCache.reset();
-  }
-
-  std::string GetCachePath() {
-    return *CachePath;
-  }
-
-  std::string GetEntryCachePath() {
-    return *EntryCache;
-  }
-}
@@ -1,12 +0,0 @@
-#pragma once
-#include <string>
-
-namespace FEXCore::Paths {
-  void InitializePaths();
-  void ShutdownPaths();
-
-  const char *GetHomeDirectory();
-
-  std::string GetCachePath();
-  std::string GetEntryCachePath();
-}
@@ -2,19 +2,19 @@

 #include <FEXCore/Utils/BitUtils.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/fextl/sstream.h>
+#include <FEXCore/fextl/string.h>

 #include <cmath>
 #include <cstring>
 #include <stdint.h>
-#include <string>
-#include <sstream>

 extern "C" {
 #include "SoftFloat-3e/platform.h"
 #include "SoftFloat-3e/softfloat.h"
 }

-struct X80SoftFloat {
+struct FEX_PACKED X80SoftFloat {
 #ifdef _M_X86_64
 // Define this to push some operations to x87
 // Only useful to see if precision loss is killing something
@@ -32,22 +32,28 @@ struct X80SoftFloat {
 #else
 #error No 128bit float for this target!
 #endif
-  struct __attribute__((packed)) {
-    uint64_t Significand : 64;
-    uint16_t Exponent    : 15;
-    unsigned Sign        : 1;
-  };
+
+#ifndef _WIN32
+#define LIBRARY_PRECISION BIGFLOAT
+#else
+// Mingw Win32 libraries don't have `__float128` helpers. Needs to use a lower precision.
+#define LIBRARY_PRECISION double
+#endif
+
+  uint64_t Significand : 64;
+  uint16_t Exponent    : 15;
+  uint16_t Sign        : 1;

  X80SoftFloat() { memset(this, 0, sizeof(*this)); }
-  X80SoftFloat(unsigned _Sign, uint16_t _Exponent, uint64_t _Significand)
+  X80SoftFloat(uint16_t _Sign, uint16_t _Exponent, uint64_t _Significand)
    : Significand {_Significand}
    , Exponent {_Exponent}
    , Sign {_Sign}
    {
  }

-  std::string str() const {
-    std::ostringstream string;
+  fextl::string str() const {
+    fextl::ostringstream string;
    string << std::hex << Sign;
    string << "_" << Exponent;
    string << "_" << (Significand >> 63);
@@ -57,51 +63,183 @@ struct X80SoftFloat {

  // Ops
  static X80SoftFloat FADD(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    faddp;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
    return extF80_add(lhs, rhs);
+#endif
  }

  static X80SoftFloat FSUB(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fsubp;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
    return extF80_sub(lhs, rhs);
+#endif
  }

  static X80SoftFloat FMUL(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fmulp;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
    return extF80_mul(lhs, rhs);
+#endif
  }

  static X80SoftFloat FDIV(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fdivp;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
    return extF80_div(lhs, rhs);
+#endif
  }

  static X80SoftFloat FREM(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
-    X80SoftFloat Rem = extF80_rem(lhs, rhs);
-    if (SignBit(Rem)) {
-      Rem = extF80_add(Rem, rhs);
-    }
-    else {
-      Rem.Sign = SignBit(lhs);
-    }
+#if defined(DEBUG_X86_FLOAT)
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fprem;
+    fstpt %[result];
+    ffreep %%st(0);
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");

-    return Rem;
+    return Result;
+#else
+    return extF80_rem(lhs, rhs);
+#endif
  }

  static X80SoftFloat FREM1(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
+#if defined(DEBUG_X86_FLOAT)
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fprem1;
+    fstpt %[result];
+    ffreep %%st(0);
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
    return extF80_rem(lhs, rhs);
+#endif
  }

  static X80SoftFloat FRNDINT(X80SoftFloat const &lhs) {
    return extF80_roundToInt(lhs, softfloat_roundingMode, false);
  }

+  static X80SoftFloat FRNDINT(X80SoftFloat const &lhs, uint_fast8_t RoundMode) {
+    return extF80_roundToInt(lhs, RoundMode, false);
+  }
+
  static X80SoftFloat FXTRACT_SIG(X80SoftFloat const &lhs) {
+#if defined(DEBUG_X86_FLOAT)
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fxtract;
+    fstpt %[result];
+    ffreep %%st(0);
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
    X80SoftFloat Tmp = lhs;
    Tmp.Exponent = 0x3FFF;
    Tmp.Sign = lhs.Sign;
    return Tmp;
+#endif
  }

  static X80SoftFloat FXTRACT_EXP(X80SoftFloat const &lhs) {
+#if defined(DEBUG_X86_FLOAT)
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fxtract;
+    ffreep %%st(0);
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
    int32_t TrueExp = lhs.Exponent - ExponentBias;
    return i32_to_extF80(TrueExp);
+#endif
  }

  static void FCMP(X80SoftFloat const &lhs, X80SoftFloat const &rhs, bool *eq, bool *lt, bool *nan) {
@@ -112,61 +250,189 @@ struct X80SoftFloat {

  static X80SoftFloat FSCALE(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
    WARN_ONCE_FMT("x87: Application used FSCALE which may have accuracy problems");
-    X80SoftFloat Int = FRNDINT(rhs);
-    BIGFLOAT Src2_d = Int;
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fscale; # st0 = st0 * 2^(rdint(st1))
+    fstpt %[result];
+    ffreep %%st(0);
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
+    X80SoftFloat Int = FRNDINT(rhs, softfloat_round_minMag);
+    LIBRARY_PRECISION Src2_d = Int;
    Src2_d = exp2l(Src2_d);
    X80SoftFloat Src2_X80 = Src2_d;
    X80SoftFloat Result = extF80_mul(lhs, Src2_X80);
    return Result;
+#endif
  }

  static X80SoftFloat F2XM1(X80SoftFloat const &lhs) {
    WARN_ONCE_FMT("x87: Application used F2XM1 which may have accuracy problems");
-    BIGFLOAT Src1_d = lhs;
-    BIGFLOAT Result = exp2l(Src1_d);
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    f2xm1; # st0 = 2^st(0) - 1
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st");
+
+    return Result;
+#else
+    LIBRARY_PRECISION Src1_d = lhs;
+    LIBRARY_PRECISION Result = exp2l(Src1_d);
    Result -= 1.0;
    return Result;
+#endif
  }

  static X80SoftFloat FYL2X(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
    WARN_ONCE_FMT("x87: Application used FYL2X which may have accuracy problems");
-    BIGFLOAT Src1_d = lhs;
-    BIGFLOAT Src2_d = rhs;
-    BIGFLOAT Tmp = Src2_d * log2l(Src1_d);
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st(1)
+    fldt %[lhs]; # st(0)
+    fyl2x; # st(1) * log2l(st(0))
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
+    LIBRARY_PRECISION Src1_d = lhs;
+    LIBRARY_PRECISION Src2_d = rhs;
+    LIBRARY_PRECISION Tmp = Src2_d * log2l(Src1_d);
    return Tmp;
+#endif
  }

  static X80SoftFloat FATAN(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
    WARN_ONCE_FMT("x87: Application used FATAN which may have accuracy problems");
-    BIGFLOAT Src1_d = lhs;
-    BIGFLOAT Src2_d = rhs;
-    BIGFLOAT Tmp = atan2l(Src1_d, Src2_d);
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs];
+    fldt %[rhs];
+    fpatan;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
+    LIBRARY_PRECISION Src1_d = lhs;
+    LIBRARY_PRECISION Src2_d = rhs;
+    LIBRARY_PRECISION Tmp = atan2l(Src1_d, Src2_d);
    return Tmp;
+#endif
  }

  static X80SoftFloat FTAN(X80SoftFloat const &lhs) {
    WARN_ONCE_FMT("x87: Application used FTAN which may have accuracy problems");
-    BIGFLOAT Src_d = lhs;
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fptan;
+    ffreep %%st(0);
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st");
+
+    return Result;
+#else
+    LIBRARY_PRECISION Src_d = lhs;
    Src_d = tanl(Src_d);
    return Src_d;
+#endif
  }

  static X80SoftFloat FSIN(X80SoftFloat const &lhs) {
    WARN_ONCE_FMT("x87: Application used FSIN which may have accuracy problems");
-    BIGFLOAT Src_d = lhs;
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fsin;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st");
+
+    return Result;
+#else
+    LIBRARY_PRECISION Src_d = lhs;
    Src_d = sinl(Src_d);
    return Src_d;
+#endif
  }

  static X80SoftFloat FCOS(X80SoftFloat const &lhs) {
    WARN_ONCE_FMT("x87: Application used FCOS which may have accuracy problems");
-    BIGFLOAT Src_d = lhs;
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fcos;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st");
+
+    return Result;
+#else
+    LIBRARY_PRECISION Src_d = lhs;
    Src_d = cosl(Src_d);
    return Src_d;
+#endif
  }

  static X80SoftFloat FSQRT(X80SoftFloat const &lhs) {
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fsqrt;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st");
+
+    return Result;
+#else
    return extF80_sqrt(lhs);
+#endif
  }

  operator float() const {
@@ -179,6 +445,7 @@ struct X80SoftFloat {
    return FEXCore::BitCast<double>(Result);
  }

+#ifndef _WIN32
  operator BIGFLOAT() const {
 #if BIGFLOATSIZE == 16
    const float128_t Result = extF80_to_f128(*this);
@@ -189,6 +456,7 @@ struct X80SoftFloat {
    return result;
 #endif
  }
+#endif

  operator int16_t() const {
    auto rv = extF80_to_i32(*this, softfloat_roundingMode, false);
@@ -257,6 +525,7 @@ struct X80SoftFloat {
    *this = f64_to_extF80(FEXCore::BitCast<float64_t>(rhs));
  }

+#ifndef _WIN32
  X80SoftFloat(BIGFLOAT rhs) {
 #if BIGFLOATSIZE == 16
    *this = f128_to_extF80(FEXCore::BitCast<float128_t>(rhs));
@@ -264,6 +533,7 @@ struct X80SoftFloat {
    *this = FEXCore::BitCast<long double>(rhs);
 #endif
  }
+#endif

  X80SoftFloat(const int16_t rhs) {
    *this = i32_to_extF80(rhs);
@@ -302,4 +572,9 @@ private:
  static constexpr uint32_t ExponentBias = 16383;
 };

+#ifndef _WIN32
 static_assert(sizeof(X80SoftFloat) == 10, "tword must be 10bytes in size");
+#else
+// Padding on this extends to 16-bytes rather than 10-bytes on WIN32.
+static_assert(sizeof(X80SoftFloat) == 16, "tword must be 16bytes in size");
+#endif
@@ -1,48 +1,49 @@
 #pragma once
+#include <FEXCore/fextl/string.h>
+
 #include <cstdint>
-#include <string>
 #include <string_view>
 #include <optional>

 namespace FEXCore::StrConv {
  [[maybe_unused]] static bool Conv(std::string_view Value, bool *Result) {
-    *Result = std::stoi(std::string(Value), nullptr, 0);
+    *Result = std::strtoull(Value.data(), nullptr, 0);
    return true;
  }

  [[maybe_unused]] static bool Conv(std::string_view Value, uint8_t *Result) {
-    *Result = std::stoi(std::string(Value), nullptr, 0);
+    *Result = std::strtoul(Value.data(), nullptr, 0);
    return true;
  }

  [[maybe_unused]] static bool Conv(std::string_view Value, uint16_t *Result) {
-    *Result = std::stoi(std::string(Value), nullptr, 0);
+    *Result = std::strtoul(Value.data(), nullptr, 0);
    return true;
  }

  [[maybe_unused]] static bool Conv(std::string_view Value, uint32_t *Result) {
-    *Result = std::stoi(std::string(Value), nullptr, 0);
+    *Result = std::strtoul(Value.data(), nullptr, 0);
    return true;
  }

  [[maybe_unused]] static bool Conv(std::string_view Value, int32_t *Result) {
-    *Result = std::stoi(std::string(Value), nullptr, 0);
+    *Result = std::strtol(Value.data(), nullptr, 0);
    return true;
  }

  [[maybe_unused]] static bool Conv(std::string_view Value, uint64_t *Result) {
-    *Result = std::stoull(std::string(Value), nullptr, 0);
-    return true;
-  }
-  [[maybe_unused]] static bool Conv(std::string_view Value, std::string *Result) {
-    *Result = Value;
+    *Result = std::strtoull(Value.data(), nullptr, 0);
    return true;
  }
  template <typename T,
    typename = std::enable_if<std::is_enum<T>::value, T>>
  [[maybe_unused]] static bool Conv(std::string_view Value, T *Result) {
-    *Result = static_cast<T>(std::stoull(std::string(Value), nullptr, 0));
+    *Result = static_cast<T>(std::stoull(Value.data(), nullptr, 0));
    return true;
  }

+  [[maybe_unused]] static bool Conv(std::string_view Value, fextl::string *Result) {
+    *Result = Value;
+    return true;
+  }
 }
@@ -0,0 +1,30 @@
+#pragma once
+#include <FEXCore/fextl/string.h>
+
+namespace FEXCore::StringUtils {
+  // Trim the left side of the string of whitespace and new lines
+  [[maybe_unused]] static fextl::string LeftTrim(fextl::string String, std::string_view TrimTokens = " \t\n\r") {
+    size_t pos = fextl::string::npos;
+    if ((pos = String.find_first_not_of(TrimTokens)) != fextl::string::npos) {
+      String.erase(0, pos);
+    }
+
+    return String;
+  }
+
+  // Trim the right side of the string of whitespace and new lines
+  [[maybe_unused]] static fextl::string RightTrim(fextl::string String, std::string_view TrimTokens = " \t\n\r") {
+    size_t pos = fextl::string::npos;
+    if ((pos = String.find_last_not_of(TrimTokens)) != fextl::string::npos) {
+      String.erase(String.begin() + pos + 1, String.end());
+    }
+
+    return String;
+  }
+
+  // Trim both the left and right of the string of whitespace and new lines
+  [[maybe_unused]] static fextl::string Trim(fextl::string String, std::string_view TrimTokens = " \t\n\r") {
+    return RightTrim(LeftTrim(std::move(String), TrimTokens), TrimTokens);
+  }
+
+}
@@ -1,35 +1,34 @@
 #include "Common/StringConv.h"
-#include "Common/Paths.h"
-#include "Utils/FileLoading.h"
+#include "Common/StringUtils.h"
+#include "FEXCore/Utils/EnumUtils.h"

 #include <FEXCore/Config/Config.h>
+#include <FEXCore/Utils/Allocator.h>
+#include <FEXCore/Utils/CPUInfo.h>
+#include <FEXCore/Utils/FileLoading.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/fextl/fmt.h>
+#include <FEXCore/fextl/list.h>
+#include <FEXCore/fextl/map.h>
+#include <FEXCore/fextl/memory.h>
+#include <FEXCore/fextl/string.h>
+#include <FEXCore/fextl/unordered_map.h>
+#include <FEXCore/fextl/vector.h>
+#include <FEXHeaderUtils/Filesystem.h>

 #include <array>
-#include <assert.h>
 #include <cstdlib>
-#include <filesystem>
-#include <fstream>
 #include <functional>
-#include <map>
-#include <memory>
-#include <list>
 #include <optional>
 #include <stddef.h>
 #include <stdint.h>
-#include <string>
 #include <string_view>
-#include <sys/sysinfo.h>
 #include <system_error>
 #include <type_traits>
-#include <unordered_map>
 #include <utility>
-#include <vector>
-
-#include <tiny-json.h>

 namespace FEXCore::Context {
-  struct Context;
+  class Context;
 }

 namespace FEXCore::Config {
@@ -38,176 +37,87 @@ namespace DefaultValues {
 #define OPT_BASE(type, group, enum, json, default) const P(type) P(enum) = P(default);
 #define OPT_STR(group, enum, json, default) const std::string_view P(enum) = P(default);
 #define OPT_STRARRAY(group, enum, json, default) OPT_STR(group, enum, json, default)
+#define OPT_STRENUM(group, enum, json, default) const uint64_t P(enum) = FEXCore::ToUnderlying(P(default));
 #include <FEXCore/Config/ConfigValues.inl>
 }

-namespace JSON {
-  struct JsonAllocator {
-    jsonPool_t PoolObject;
-    std::unique_ptr<std::list<json_t>> json_objects;
+  enum Paths {
+    PATH_DATA_DIR = 0,
+    PATH_CONFIG_DIR_LOCAL,
+    PATH_CONFIG_DIR_GLOBAL,
+    PATH_CONFIG_FILE_LOCAL,
+    PATH_CONFIG_FILE_GLOBAL,
+    PATH_LAST,
  };
-  static_assert(offsetof(JsonAllocator, PoolObject) == 0, "This needs to be at offset zero");
+  static std::array<fextl::string, Paths::PATH_LAST> Paths;

-  json_t* PoolInit(jsonPool_t* Pool) {
-    JsonAllocator* alloc = reinterpret_cast<JsonAllocator*>(Pool);
-    alloc->json_objects = std::make_unique<std::list<json_t>>();
-    return &*alloc->json_objects->emplace(alloc->json_objects->end());
+  void SetDataDirectory(const std::string_view Path) {
+    Paths[PATH_DATA_DIR] = Path;
  }

-  json_t* PoolAlloc(jsonPool_t* Pool) {
-    JsonAllocator* alloc = reinterpret_cast<JsonAllocator*>(Pool);
-    return &*alloc->json_objects->emplace(alloc->json_objects->end());
+  void SetConfigDirectory(const std::string_view Path, bool Global) {
+    Paths[PATH_CONFIG_DIR_LOCAL + Global] = Path;
  }

-  static void LoadJSonConfig(const std::string &Config, std::function<void(const char *Name, const char *ConfigSring)> Func) {
-    std::vector<char> Data;
-    if (!FEXCore::FileLoading::LoadFile(Data, Config)) {
-      return;
-    }
-
-    JsonAllocator Pool {
-      .PoolObject = {
-        .init = PoolInit,
-        .alloc = PoolAlloc,
-      },
-    };
-
-    json_t const *json = json_createWithPool(&Data.at(0), &Pool.PoolObject);
-    if (!json) {
-      LogMan::Msg::EFmt("Couldn't create json");
-      return;
-    }
-
-    json_t const* ConfigList = json_getProperty(json, "Config");
-
-    if (!ConfigList) {
-      LogMan::Msg::EFmt("Couldn't get config list");
-      return;
-    }
-
-    for (json_t const* ConfigItem = json_getChild(ConfigList);
-      ConfigItem != nullptr;
-      ConfigItem = json_getSibling(ConfigItem)) {
-      const char* ConfigName = json_getName(ConfigItem);
-      const char* ConfigString = json_getValue(ConfigItem);
-
-      if (!ConfigName) {
-        LogMan::Msg::EFmt("Couldn't get config name");
-        return;
-      }
-
-      if (!ConfigString) {
-        LogMan::Msg::EFmt("Couldn't get ConfigString for '{}'", ConfigName);
-        return;
-      }
-
-      Func(ConfigName, ConfigString);
-    }
-  }
-}
-
-  std::string GetDataDirectory() {
-    std::string DataDir{};
-
-    char const *HomeDir = Paths::GetHomeDirectory();
-    char const *DataXDG = getenv("XDG_DATA_HOME");
-    char const *DataOverride = getenv("FEX_APP_DATA_LOCATION");
-    if (DataOverride) {
-      // Data override will override the complete directory
-      DataDir = DataOverride;
-    }
-    else {
-      DataDir = DataXDG ?: HomeDir;
-      DataDir += "/.fex-emu/";
-    }
-    return DataDir;
+  void SetConfigFileLocation(const std::string_view Path, bool Global) {
+    Paths[PATH_CONFIG_FILE_LOCAL + Global] = Path;
  }

-  std::string GetConfigDirectory(bool Global) {
-    std::string ConfigDir;
-    if (Global) {
-      ConfigDir = GLOBAL_DATA_DIRECTORY;
-    }
-    else {
-      char const *HomeDir = Paths::GetHomeDirectory();
-      char const *ConfigXDG = getenv("XDG_CONFIG_HOME");
-      char const *ConfigOverride = getenv("FEX_APP_CONFIG_LOCATION");
-      if (ConfigOverride) {
-        // Config override completely overrides the config directory
-        ConfigDir = ConfigOverride;
-      }
-      else {
-        ConfigDir = ConfigXDG ? ConfigXDG : HomeDir;
-        ConfigDir += "/.fex-emu/";
-      }
-
-      // Ensure the folder structure is created for our configuration
-      std::error_code ec{};
-      if (!std::filesystem::exists(ConfigDir, ec) &&
-          !std::filesystem::create_directories(ConfigDir, ec)) {
-        // Let's go local in this case
-        return "./";
-      }
-    }
-
-    return ConfigDir;
+  fextl::string const& GetDataDirectory() {
+    return Paths[PATH_DATA_DIR];
  }

-  std::string GetConfigFileLocation() {
-    std::string ConfigFile{};
-    const char *AppConfig = getenv("FEX_APP_CONFIG");
-    if (AppConfig) {
-      // App config environment variable overwrites only the config file
-      ConfigFile = AppConfig;
-    }
-    else {
-      ConfigFile = GetConfigDirectory(false) + "Config.json";
-    }
-    return ConfigFile;
+  fextl::string const& GetConfigDirectory(bool Global) {
+    return Paths[PATH_CONFIG_DIR_LOCAL + Global];
  }

-  std::string GetApplicationConfig(const std::string &Filename, bool Global) {
-    std::string ConfigFile = GetConfigDirectory(Global);
+  fextl::string const& GetConfigFileLocation(bool Global) {
+    return Paths[PATH_CONFIG_FILE_LOCAL + Global];
+  }
+
+  fextl::string GetApplicationConfig(const std::string_view Program, bool Global) {
+    fextl::string ConfigFile = GetConfigDirectory(Global);

-    std::error_code ec{};
    if (!Global &&
-        !std::filesystem::exists(ConfigFile, ec) &&
-        !std::filesystem::create_directories(ConfigFile, ec)) {
+        !FHU::Filesystem::Exists(ConfigFile) &&
+        !FHU::Filesystem::CreateDirectories(ConfigFile)) {
      LogMan::Msg::DFmt("Couldn't create config directory: '{}'", ConfigFile);
      // Let's go local in this case
-      return "./" + Filename + ".json";
+      return fextl::fmt::format("./{}.json", Program);
    }

    ConfigFile += "AppConfig/";

    // Attempt to create the local folder if it doesn't exist
    if (!Global &&
-        !std::filesystem::exists(ConfigFile, ec) &&
-        !std::filesystem::create_directories(ConfigFile, ec)) {
+        !FHU::Filesystem::Exists(ConfigFile) &&
+        !FHU::Filesystem::CreateDirectories(ConfigFile)) {
      // Let's go local in this case
-      return "./" + Filename + ".json";
+      return fextl::fmt::format("./{}.json", Program);
    }

-    ConfigFile += Filename + ".json";
-    return ConfigFile;
+    return fextl::fmt::format("{}{}.json", ConfigFile, Program);
  }

  void SetConfig(FEXCore::Context::Context *CTX, ConfigOption Option, uint64_t Config) {
  }

-  void SetConfig(FEXCore::Context::Context *CTX, ConfigOption Option, std::string const &Config) {
+  void SetConfig(FEXCore::Context::Context *CTX, ConfigOption Option, fextl::string const &Config) {
  }

  uint64_t GetConfig(FEXCore::Context::Context *CTX, ConfigOption Option) {
    return 0;
  }

-  static std::map<FEXCore::Config::LayerType, std::unique_ptr<FEXCore::Config::Layer>> ConfigLayers;
+  static fextl::map<FEXCore::Config::LayerType, fextl::unique_ptr<FEXCore::Config::Layer>> ConfigLayers;
  static FEXCore::Config::Layer *Meta{};

-  constexpr std::array<FEXCore::Config::LayerType, 6> LoadOrder = {
+  constexpr std::array<FEXCore::Config::LayerType, 9> LoadOrder = {
+    FEXCore::Config::LayerType::LAYER_GLOBAL_MAIN,
    FEXCore::Config::LayerType::LAYER_MAIN,
+    FEXCore::Config::LayerType::LAYER_GLOBAL_STEAM_APP,
    FEXCore::Config::LayerType::LAYER_GLOBAL_APP,
+    FEXCore::Config::LayerType::LAYER_LOCAL_STEAM_APP,
    FEXCore::Config::LayerType::LAYER_LOCAL_APP,
    FEXCore::Config::LayerType::LAYER_ARGUMENTS,
    FEXCore::Config::LayerType::LAYER_ENVIRONMENT,
@@ -259,17 +169,17 @@ namespace JSON {
    }

    // If an environment variable exists in both current meta and in the incoming layer then the meta layer value is overwritten
-    std::unordered_map<std::string, std::string> LookupMap;
+    fextl::unordered_map<fextl::string, fextl::string> LookupMap;
    const auto AddToMap = [&LookupMap](FEXCore::Config::LayerValue const &Value) {
      for (const auto &EnvVar : Value) {
        const auto ItEq = EnvVar.find_first_of('=');
-        if (ItEq == std::string::npos) {
+        if (ItEq == fextl::string::npos) {
          // Broken environment variable
          // Skip
          continue;
        }
-        auto Key = std::string(EnvVar.begin(), EnvVar.begin() + ItEq);
-        auto Value = std::string(EnvVar.begin() + ItEq + 1, EnvVar.end());
+        auto Key = fextl::string(EnvVar.begin(), EnvVar.begin() + ItEq);
+        auto Value = fextl::string(EnvVar.begin() + ItEq + 1, EnvVar.end());

        // Add the key to the map, overwriting whatever previous value was there
        LookupMap.insert_or_assign(std::move(Key), std::move(Value));
@@ -302,7 +212,7 @@ namespace JSON {
  }

  void Initialize() {
-    AddLayer(std::make_unique<MetaLayer>(FEXCore::Config::LayerType::LAYER_TOP));
+    AddLayer(fextl::make_unique<MetaLayer>(FEXCore::Config::LayerType::LAYER_TOP));
    Meta = ConfigLayers.begin()->second.get();
  }

@@ -320,16 +230,15 @@ namespace JSON {
    }
  }

-  std::string ExpandPath(std::string const &ContainerPrefix, std::string PathName) {
+  fextl::string ExpandPath(fextl::string const &ContainerPrefix, fextl::string PathName) {
    if (PathName.empty()) {
      return {};
    }

-    std::filesystem::path Path{PathName};

    // Expand home if it exists
-    if (Path.is_relative()) {
-      std::string Home = getenv("HOME") ?: "";
+    if (FHU::Filesystem::IsRelative(PathName)) {
+      fextl::string Home = getenv("HOME") ?: "";
      // Home expansion only works if it is the first character
      // This matches bash behaviour
      if (PathName.at(0) == '~') {
@@ -338,12 +247,15 @@ namespace JSON {
      }

      // Expand relative path to absolute
-      Path = std::filesystem::absolute(Path);
+      char ExistsTempPath[PATH_MAX];
+      char *RealPath = FHU::Filesystem::Absolute(PathName.c_str(), ExistsTempPath);
+      if (RealPath) {
+        PathName = RealPath;
+      }

      // Only return if it exists
-      std::error_code ec{};
-      if (std::filesystem::exists(Path, ec)) {
-        return Path;
+      if (FHU::Filesystem::Exists(PathName)) {
+        return PathName;
      }
    }
    else {
@@ -359,9 +271,9 @@ namespace JSON {
      // HostThunks: $CMAKE_INSTALL_PREFIX/lib/fex-emu/HostThunks/
      // GuestThunks: $CMAKE_INSTALL_PREFIX/share/fex-emu/GuestThunks/
      if (!ContainerPrefix.empty() && !PathName.empty()) {
-        if (!std::filesystem::exists(PathName)) {
+        if (!FHU::Filesystem::Exists(PathName)) {
          auto ContainerPath = ContainerPrefix + PathName;
-          if (std::filesystem::exists(ContainerPath)) {
+          if (FHU::Filesystem::Exists(ContainerPath)) {
            return ContainerPath;
          }
        }
@@ -370,38 +282,30 @@ namespace JSON {
    return {};
  }

-  std::string ltrim(std::string String) {
-    size_t pos = std::string::npos;
-    if ((pos = String.find_first_not_of(" \t\n\r")) != std::string::npos) {
-      String.erase(0, pos);
-    }
+  constexpr char ContainerManager[] = "/run/host/container-manager";

-    return String;
-  }
-
-  std::string rtrim(std::string String) {
-    size_t pos = std::string::npos;
-    if ((pos = String.find_last_not_of(" \t\n\r")) != std::string::npos) {
-      String.erase(String.begin() + pos + 1, String.end());
-    }
-
-    return String;
-  }
-
-  std::string trim(std::string String) {
-    return rtrim(ltrim(String));
-  }
-
-
-  std::string FindContainerPrefix() {
+  fextl::string FindContainer() {
    // We only support pressure-vessel at the moment
-    const static std::string ContainerManager = "/run/host/container-manager";
-    if (std::filesystem::exists(ContainerManager)) {
-      std::vector<char> Manager{};
+    if (FHU::Filesystem::Exists(ContainerManager)) {
+      fextl::vector<char> Manager{};
      if (FEXCore::FileLoading::LoadFile(Manager, ContainerManager)) {
        // Trim the whitespace, may contain a newline
-        std::string ManagerStr = Manager.data();
-        ManagerStr = trim(ManagerStr);
+        fextl::string ManagerStr = Manager.data();
+        ManagerStr = FEXCore::StringUtils::Trim(ManagerStr);
+        return ManagerStr;
+      }
+    }
+    return {};
+  }
+
+  fextl::string FindContainerPrefix() {
+    // We only support pressure-vessel at the moment
+    if (FHU::Filesystem::Exists(ContainerManager)) {
+      fextl::vector<char> Manager{};
+      if (FEXCore::FileLoading::LoadFile(Manager, ContainerManager)) {
+        // Trim the whitespace, may contain a newline
+        fextl::string ManagerStr = Manager.data();
+        ManagerStr = FEXCore::StringUtils::Trim(ManagerStr);
        if (strncmp(ManagerStr.data(), "pressure-vessel", Manager.size()) == 0) {
          // We are running inside of pressure vessel
          // Our $CMAKE_INSTALL_PREFIX paths are now inside of /run/host/$CMAKE_INSTALL_PREFIX
@@ -422,7 +326,7 @@ namespace JSON {
      FEX_CONFIG_OPT(Cores, THREADS);
      if (Cores == 0) {
        // When the number of emulated CPU cores is zero then auto detect
-        FEXCore::Config::EraseSet(FEXCore::Config::CONFIG_THREADS, std::to_string(get_nprocs_conf()));
+        FEXCore::Config::EraseSet(FEXCore::Config::CONFIG_THREADS, fextl::fmt::format("{}", FEXCore::CPUInfo::CalculateNumberOfCPUs()));
      }
    }

@@ -434,14 +338,29 @@ namespace JSON {
 #else
      constexpr uint32_t MaxCoreNumber = 1;
 #endif
-      if (Core > MaxCoreNumber) {
+#ifdef INTERPRETER_ENABLED
+      constexpr uint32_t MinCoreNumber = 0;
+#else
+      constexpr uint32_t MinCoreNumber = 1;
+#endif
+      if (Core > MaxCoreNumber || Core < MinCoreNumber) {
        // Sanitize the core option by setting the core to the JIT if invalid
-        FEXCore::Config::EraseSet(FEXCore::Config::CONFIG_CORE, std::to_string(FEXCore::Config::CONFIG_IRJIT));
+        FEXCore::Config::EraseSet(FEXCore::Config::CONFIG_CORE, fextl::fmt::format("{}", static_cast<uint32_t>(FEXCore::Config::CONFIG_IRJIT)));
      }
    }

-    std::string ContainerPrefix { FindContainerPrefix() };
-    auto ExpandPathIfExists = [&ContainerPrefix](FEXCore::Config::ConfigOption Config, std::string PathName) {
+    if (FEXCore::Config::Exists(FEXCore::Config::CONFIG_CACHEOBJECTCODECOMPILATION)) {
+      FEX_CONFIG_OPT(CacheObjectCodeCompilation, CACHEOBJECTCODECOMPILATION);
+      FEX_CONFIG_OPT(Core, CORE);
+
+      if (CacheObjectCodeCompilation() && Core() == FEXCore::Config::CONFIG_INTERPRETER) {
+        // If running the interpreter then disable cache code compilation
+        FEXCore::Config::Erase(FEXCore::Config::CONFIG_CACHEOBJECTCODECOMPILATION);
+      }
+    }
+
+    fextl::string ContainerPrefix { FindContainerPrefix() };
+    auto ExpandPathIfExists = [&ContainerPrefix](FEXCore::Config::ConfigOption Config, fextl::string PathName) {
      auto NewPath = ExpandPath(ContainerPrefix, PathName);
      if (!NewPath.empty()) {
        FEXCore::Config::EraseSet(Config, NewPath);
@@ -450,16 +369,15 @@ namespace JSON {

    if (FEXCore::Config::Exists(FEXCore::Config::CONFIG_ROOTFS)) {
      FEX_CONFIG_OPT(PathName, ROOTFS);
-      auto ExpandedString = ExpandPath(ContainerPrefix, PathName());
+      auto ExpandedString = ExpandPath(ContainerPrefix,PathName());
      if (!ExpandedString.empty()) {
        // Adjust the path if it ended up being relative
        FEXCore::Config::EraseSet(FEXCore::Config::CONFIG_ROOTFS, ExpandedString);
      }
      else if (!PathName().empty()) {
        // If the filesystem doesn't exist then let's see if it exists in the fex-emu folder
-        std::string NamedRootFS = GetDataDirectory() + "RootFS/" + PathName();
-        std::error_code ec{};
-        if (std::filesystem::exists(NamedRootFS, ec)) {
+        fextl::string NamedRootFS = GetDataDirectory() + "RootFS/" + PathName();
+        if (FHU::Filesystem::Exists(NamedRootFS)) {
          FEXCore::Config::EraseSet(FEXCore::Config::CONFIG_ROOTFS, NamedRootFS);
        }
      }
@@ -481,9 +399,8 @@ namespace JSON {
      }
      else if (!PathName().empty()) {
        // If the filesystem doesn't exist then let's see if it exists in the fex-emu folder
-        std::string NamedConfig = GetDataDirectory() + "ThunkConfigs/" + PathName();
-        std::error_code ec{};
-        if (std::filesystem::exists(NamedConfig, ec)) {
+        fextl::string NamedConfig = GetDataDirectory() + "ThunkConfigs/" + PathName();
+        if (FHU::Filesystem::Exists(NamedConfig)) {
          FEXCore::Config::EraseSet(FEXCore::Config::CONFIG_THUNKCONFIG, NamedConfig);
        }
      }
@@ -497,11 +414,11 @@ namespace JSON {

    if (FEXCore::Config::Exists(FEXCore::Config::CONFIG_SINGLESTEP)) {
      // Single stepping also enforces single instruction size blocks
-      Set(FEXCore::Config::ConfigOption::CONFIG_MAXINST, std::to_string(1u));
+      Set(FEXCore::Config::ConfigOption::CONFIG_MAXINST, "1");
    }
  }

-  void AddLayer(std::unique_ptr<FEXCore::Config::Layer> _Layer) {
+  void AddLayer(fextl::unique_ptr<FEXCore::Config::Layer> _Layer) {
    ConfigLayers.emplace(_Layer->GetLayerType(), std::move(_Layer));
  }

@@ -513,7 +430,7 @@ namespace JSON {
    return Meta->All(Option);
  }

-  std::optional<std::string*> Get(ConfigOption Option) {
+  std::optional<fextl::string*> Get(ConfigOption Option) {
    return Meta->Get(Option);
  }

@@ -535,7 +452,7 @@ namespace JSON {
    auto Value = FEXCore::Config::Get(Option);

    if (!FEXCore::StrConv::Conv(**Value, &Result)) {
-      assert(0 && "Attempted to convert invalid value");
+      LOGMAN_MSG_A_FMT("Attempted to convert invalid value");
    }
    return Result;
  }
@@ -554,7 +471,7 @@ namespace JSON {
  }

  template<>
-  std::string Value<std::string>::GetIfExists(FEXCore::Config::ConfigOption Option, std::string Default) {
+  fextl::string Value<fextl::string>::GetIfExists(FEXCore::Config::ConfigOption Option, fextl::string Default) {
    auto Value = FEXCore::Config::Get(Option);
    if (Value) {
      return **Value;
@@ -565,13 +482,13 @@ namespace JSON {
  }

  template<>
-  std::string Value<std::string>::GetIfExists(FEXCore::Config::ConfigOption Option, std::string_view Default) {
+  fextl::string Value<fextl::string>::GetIfExists(FEXCore::Config::ConfigOption Option, std::string_view Default) {
    auto Value = FEXCore::Config::Get(Option);
    if (Value) {
      return **Value;
    }
    else {
-      return std::string(Default);
+      return fextl::string(Default);
    }
  }

@@ -586,161 +503,19 @@ namespace JSON {
  template uint64_t Value<uint64_t>::GetIfExists(FEXCore::Config::ConfigOption Option, uint64_t Default);

  // Constructor
-  template Value<std::string>::Value(FEXCore::Config::ConfigOption _Option, std::string Default);
+  template Value<fextl::string>::Value(FEXCore::Config::ConfigOption _Option, fextl::string Default);
  template Value<bool>::Value(FEXCore::Config::ConfigOption _Option, bool Default);
  template Value<uint8_t>::Value(FEXCore::Config::ConfigOption _Option, uint8_t Default);
  template Value<uint64_t>::Value(FEXCore::Config::ConfigOption _Option, uint64_t Default);

  template<typename T>
-  void Value<T>::GetListIfExists(FEXCore::Config::ConfigOption Option, std::list<std::string> *List) {
+  void Value<T>::GetListIfExists(FEXCore::Config::ConfigOption Option, fextl::list<fextl::string> *List) {
    auto Value = FEXCore::Config::All(Option);
    List->clear();
    if (Value) {
      *List = **Value;
    }
  }
-  template void Value<std::string>::GetListIfExists(FEXCore::Config::ConfigOption Option, std::list<std::string> *List);
-
-  // Application loaders
-  class MainLoader final : public FEXCore::Config::OptionMapper {
-  public:
-    explicit MainLoader();
-    explicit MainLoader(std::string ConfigFile);
-    void Load() override;
-
-  private:
-    std::string Config;
-  };
-
-  class AppLoader final : public FEXCore::Config::OptionMapper {
-  public:
-    explicit AppLoader(const std::string& Filename, bool Global);
-    void Load();
-
-  private:
-    std::string Config;
-  };
-
-  class EnvLoader final : public FEXCore::Config::Layer {
-  public:
-    explicit EnvLoader(char *const _envp[]);
-    void Load() override;
-
-  private:
-    char *const *envp;
-  };
-
-  static const std::map<std::string, FEXCore::Config::ConfigOption, std::less<>> ConfigLookup = {{
-#define OPT_BASE(type, group, enum, json, default) {#json, FEXCore::Config::ConfigOption::CONFIG_##enum},
-#include <FEXCore/Config/ConfigValues.inl>
-  }};
-  static const std::vector<std::pair<const char*, FEXCore::Config::ConfigOption>> EnvConfigLookup = {{
-#define OPT_BASE(type, group, enum, json, default) {"FEX_" #enum, FEXCore::Config::ConfigOption::CONFIG_##enum},
-#include <FEXCore/Config/ConfigValues.inl>
-  }};
-
-  OptionMapper::OptionMapper(FEXCore::Config::LayerType Layer)
-    : FEXCore::Config::Layer(Layer) {
-  }
-
-  void OptionMapper::MapNameToOption(const char *ConfigName, const char *ConfigString) {
-    auto it = ConfigLookup.find(ConfigName);
-    if (it != ConfigLookup.end()) {
-      Set(it->second, ConfigString);
-    }
-  }
-
-  MainLoader::MainLoader()
-    : FEXCore::Config::OptionMapper(FEXCore::Config::LayerType::LAYER_MAIN)
-    , Config{FEXCore::Config::GetConfigFileLocation()} {
-  }
-
-  MainLoader::MainLoader(std::string ConfigFile)
-    : FEXCore::Config::OptionMapper(FEXCore::Config::LayerType::LAYER_MAIN)
-    , Config{std::move(ConfigFile)} {
-  }
-
-  void MainLoader::Load() {
-    JSON::LoadJSonConfig(Config, [this](const char *Name, const char *ConfigString) {
-      MapNameToOption(Name, ConfigString);
-    });
-  }
-
-  AppLoader::AppLoader(const std::string& Filename, bool Global)
-    : FEXCore::Config::OptionMapper(Global ? FEXCore::Config::LayerType::LAYER_GLOBAL_APP : FEXCore::Config::LayerType::LAYER_LOCAL_APP) {
-    Config = FEXCore::Config::GetApplicationConfig(Filename, Global);
-
-    // Immediately load so we can reload the meta layer
-    Load();
-  }
-
-  void AppLoader::Load() {
-    JSON::LoadJSonConfig(Config, [this](const char *Name, const char *ConfigString) {
-      MapNameToOption(Name, ConfigString);
-    });
-  }
-
-  EnvLoader::EnvLoader(char *const _envp[])
-    : FEXCore::Config::Layer(FEXCore::Config::LayerType::LAYER_ENVIRONMENT)
-    , envp {_envp} {
-  }
-
-  void EnvLoader::Load() {
-    std::unordered_map<std::string_view, std::string_view> EnvMap;
-
-    for(const char *const *pvar=envp; pvar && *pvar; pvar++) {
-      std::string_view Var(*pvar);
-      size_t pos = Var.rfind('=');
-      if (std::string::npos == pos)
-        continue;
-
-      std::string_view Key = Var.substr(0,pos);
-      std::string_view Value {Var.substr(pos+1)};
-
-#define ENVLOADER
-#include <FEXCore/Config/ConfigOptions.inl>
-
-      EnvMap[Key]=Value;
-    }
-
-    std::function GetVar = [=](const std::string_view id)  -> std::optional<std::string_view> {
-      if (EnvMap.find(id) != EnvMap.end())
-        return EnvMap.at(id);
-
-      // If envp[] was empty, search using std::getenv()
-      const char* vs = std::getenv(id.data());
-      if (vs) {
-        return vs;
-      }
-      else {
-        return std::nullopt;
-      }
-    };
-
-    std::optional<std::string_view> Value;
-
-    for (auto &it : EnvConfigLookup) {
-      if ((Value = GetVar(it.first)).has_value()) {
-        Set(it.second, std::string(*Value));
-      }
-    }
-  }
-
-  std::unique_ptr<FEXCore::Config::Layer> CreateMainLayer(std::string const *File) {
-    if (File) {
-      return std::make_unique<FEXCore::Config::MainLoader>(*File);
-    }
-    else {
-      return std::make_unique<FEXCore::Config::MainLoader>();
-    }
-  }
-
-  std::unique_ptr<FEXCore::Config::Layer> CreateAppLayer(const std::string& Filename, bool Global) {
-    return std::make_unique<FEXCore::Config::AppLoader>(Filename, Global);
-  }
-
-  std::unique_ptr<FEXCore::Config::Layer> CreateEnvironmentLayer(char *const _envp[]) {
-    return std::make_unique<FEXCore::Config::EnvLoader>(_envp);
-  }
+  template void Value<fextl::string>::GetListIfExists(FEXCore::Config::ConfigOption Option, fextl::list<fextl::string> *List);
 }

@@ -16,10 +16,11 @@
      },
      "Multiblock": {
        "Type": "bool",
-        "Default": "true",
+        "Default": "false",
        "ShortArg": "m",
        "Desc": [
-          "Controls multiblock code compilation"
+          "Controls multiblock code compilation",
+          "Can cause long JIT compilation times and stutter"
        ]
      },
      "MaxInst": {
@@ -38,6 +39,24 @@
          "Number of physical hardware threads to tell the process we have.",
          "0 will auto detect."
        ]
+      },
+      "CacheObjectCodeCompilation": {
+        "Type": "uint32",
+        "Default": "FEXCore::Config::ConfigObjectCodeHandler::CONFIG_NONE",
+        "TextDefault": "none",
+        "Choices": [ "none", "read", "readwrite" ],
+        "ArgumentHandler": "CacheObjectCodeHandler",
+        "Desc": [
+          "Cache JIT object code to drive.",
+          "Allows JIT code to be shared between applications"
+        ]
+      },
+      "EnableAVX": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Determines whether or not we use the expanded register file for AVX or not"
+        ]
      }
    },
    "Emulation": {
@@ -72,6 +91,20 @@
          "Folder to find the guest-side thunking libraries."
        ]
      },
+      "ThunkHostLibs32": {
+        "Type": "str",
+        "Default": "@CMAKE_INSTALL_PREFIX@/lib/fex-emu/HostThunks_32/",
+        "Desc": [
+          "Folder to find the 32-bit host-side thunking libraries."
+        ]
+      },
+      "ThunkGuestLibs32": {
+        "Type": "str",
+        "Default": "@CMAKE_INSTALL_PREFIX@/share/fex-emu/GuestThunks_32/",
+        "Desc": [
+          "Folder to find the 32-bit guest-side thunking libraries."
+        ]
+      },
      "ThunkConfig": {
        "Type": "str",
        "Default": "",
@@ -104,6 +137,13 @@
          "This can be useful for setting environment variables that thunks can pick up.",
          "Typically isn't necessary since the guest libc isn't thunked. But is possible."
        ]
+      },
+      "AdditionalArguments": {
+        "Type": "strarray",
+        "Default": "",
+        "Desc": [
+          "Allows the user to pass additional arguments to the application"
+        ]
      }
    },
    "Debug": {
@@ -190,6 +230,41 @@
          "Useful for determining hot blocks of code",
          "Has some file writing overhead per JIT block"
        ]
+      },
+      "GDBSymbols": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Integrates with GDB using the JIT interface.",
+          "Needs the fex jit loader in GDB, which can be loaded via `jit-reader-load libFEXGDBReader.so.`",
+          "Also needs x86_64-linux-gnu-objdump in PATH.",
+          "Can be very slow."
+        ]
+      },
+      "InjectLibSegFault": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Sets the environment variable LD_PRELOAD=libSegFault.so",
+          "This allows the user to very easily enable libSegFault without dealing with environment variables",
+          "Very useful for applications that have launch scripts that set the variable to nothing at launch",
+          "Set this in an application configuration for injecting in to only specific applications.",
+          "\tNote: If x86/x86_64 libSegFault.so isn't installed then this option won't work."
+        ]
+      },
+      "Disassemble": {
+        "Type": "strenum",
+        "Default": "FEXCore::Config::Disassemble::OFF",
+        "Enums": {
+          "DISPATCHER": "dispatcher",
+          "BLOCKS": "blocks"
+        },
+        "Desc": [
+          "Allows controlling of the vixl disassembler.",
+          "\toff: No disassembly will be output",
+          "\tdispatcher: Will enable disassembly of the JIT dispatcher loop",
+          "\tblocks: Will enable disassembly of the translated instruction code blocks"
+        ]
      }
    },
    "Logging": {
@@ -201,36 +276,28 @@
          "Disables logging"
        ]
      },
-      "OutputSocket": {
-        "Type": "str",
-        "Default": "",
-        "Desc": [
-          "Socket to connect to",
-          "eg: localhost:8087",
-          "If set will override the OutputLog location"
-        ]
-      },
      "OutputLog": {
        "Type": "str",
-        "Default": "stderr",
+        "Default": "server",
        "ShortArg": "o",
        "Desc": [
          "File to write FEX output to.",
-          "[stdout, stderr, <Filename>]"
+          "[stdout, stderr, server, <Filename>]"
        ]
      }
    },
    "Hacks": {
      "SMCChecks": {
        "Type": "uint8",
-        "Default": "FEXCore::Config::CONFIG_SMC_MMAN",
-        "TextDefault": "mman",
+        "Default": "FEXCore::Config::CONFIG_SMC_MTRACK",
+        "TextDefault": "mtrack",
        "ArgumentHandler": "SMCCheckHandler",
        "Desc": [
          "Checks code for modification before execution.",
          "\tnone: No checks",
-          "\tmman: Invalidate on mmap, mprotect, munmap",
-          "\tfull: Validate code before every run (slow)"
+          "\tmtrack: Page tracking based invalidation",
+          "\tfull: Validate code before every run (slow)",
+          "\tmman: Invalidate on mmap, mprotect, munmap (deprecated, use mtrack)"
        ]
      },
      "TSOEnabled": {
@@ -241,6 +308,21 @@
          "Highly likely to break any multithreaded application if disabled."
        ]
      },
+      "TSOAutoMigration": {
+        "Type": "bool",
+        "Default": "true",
+        "Desc": [
+          "Automatically enables TSO when shared memory is used.",
+          "Should work without issues in most cases."
+        ]
+      },
+      "X87ReducedPrecision": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Emulates X87 floating point using 64-bit precision. This reduces emulation accuracy and may result in rendering bugs."
+        ]
+      },
      "ABILocalFlags": {
        "Type": "bool",
        "Default": "false",
@@ -274,6 +356,14 @@
          "Forces a process to stall out on initialization",
          "Useful for a process that keeps restarting and doesn't work"
        ]
+      },
+      "HideHypervisorBit": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Hides the hypervisor CPUID bit when set.",
+          "Should only be used for applications that have issues with this set."
+        ]
      }
    },
    "Misc": {
@@ -299,6 +389,13 @@
        "Desc": [
          "Loads an AOT IR cache for the loaded executable."
        ]
+      },
+      "ServerSocketPath": {
+        "Type": "str",
+        "Default": "",
+        "Desc": [
+          "Override for a FEXServer socket path. Only useful for chroots."
+        ]
      }
    }
  },
@@ -316,6 +413,17 @@
        "Type": "str",
        "Default": ""
      },
+      "APP_CONFIG_NAME": {
+        "Type": "str",
+        "Default": "",
+        "Desc": [
+          "This is the application config name that has been loaded.",
+          "This differs from APP_FILENAME in two ways",
+          "Where APP_FILENAME always points to the executable path that FEX-Emu is executing.",
+          "This matches what is used to load the AppLayer configuration name.",
+          "When running through a compatibility layer like wine, this will only be the exe name, instead of wine full path."
+        ]
+      },
      "IS64BIT_MODE": {
        "Type": "bool",
        "Default": "false"
@@ -1,6 +1,4 @@
-#include "Common/Paths.h"
 #include "Interface/Context/Context.h"
-#include "Interface/Core/Core.h"
 #include "Interface/Core/OpcodeDispatcher.h"
 #include "Interface/Core/X86Tables/X86Tables.h"

@@ -19,208 +17,78 @@ namespace FEXCore::HLE {

 namespace FEXCore::Context {
  void InitializeStaticTables(OperatingMode Mode) {
-    FEXCore::Paths::InitializePaths();
    X86Tables::InitializeInfoTables(Mode);
    IR::InstallOpcodeHandlers(Mode);
  }

-  void ShutdownStaticTables() {
-    FEXCore::Paths::ShutdownPaths();
+  fextl::unique_ptr<FEXCore::Context::Context> FEXCore::Context::Context::CreateNewContext() {
+    return fextl::make_unique<FEXCore::Context::ContextImpl>();
  }

-  FEXCore::Context::Context *CreateNewContext() {
-    return new FEXCore::Context::Context{};
+  bool FEXCore::Context::ContextImpl::InitializeContext() {
+    // This should be used for generating things that are shared between threads
+    CPUID.Init(this);
+    return true;
  }

-  bool InitializeContext(FEXCore::Context::Context *CTX) {
-    return FEXCore::CPU::CreateCPUCore(CTX);
+  void FEXCore::Context::ContextImpl::SetExitHandler(ExitHandler handler) {
+    CustomExitHandler = std::move(handler);
  }

-  void DestroyContext(FEXCore::Context::Context *CTX) {
-    if (CTX->ParentThread) {
-      CTX->DestroyThread(CTX->ParentThread);
-    }
-    delete CTX;
+  ExitHandler FEXCore::Context::ContextImpl::GetExitHandler() const {
+    return CustomExitHandler;
  }

-  FEXCore::Core::InternalThreadState* InitCore(FEXCore::Context::Context *CTX, FEXCore::CodeLoader *Loader) {
-    return CTX->InitCore(Loader);
+  void FEXCore::Context::ContextImpl::Stop() {
+    Stop(false);
  }

-  void SetExitHandler(FEXCore::Context::Context *CTX, ExitHandler handler) {
-    CTX->CustomExitHandler = std::move(handler);
+  void FEXCore::Context::ContextImpl::CompileRIP(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP) {
+    CompileBlock(Thread->CurrentFrame, GuestRIP);
  }

-  ExitHandler GetExitHandler(const FEXCore::Context::Context *CTX) {
-    return CTX->CustomExitHandler;
+  FEXCore::Context::ExitReason FEXCore::Context::ContextImpl::GetExitReason() {
+    return ParentThread->ExitReason;
  }

-  void Run(FEXCore::Context::Context *CTX) {
-    CTX->Run();
+  bool FEXCore::Context::ContextImpl::IsDone() const {
+    return IsPaused();
  }

-  void Step(FEXCore::Context::Context *CTX) {
-    CTX->Step();
+  void FEXCore::Context::ContextImpl::GetCPUState(FEXCore::Core::CPUState *State) const {
+    memcpy(State, ParentThread->CurrentFrame, sizeof(FEXCore::Core::CPUState));
  }

-  void CompileRIP(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP) {
-    Thread->CTX->CompileBlock(Thread->CurrentFrame, GuestRIP);
+  void FEXCore::Context::ContextImpl::SetCPUState(const FEXCore::Core::CPUState *State) {
+    memcpy(ParentThread->CurrentFrame, State, sizeof(FEXCore::Core::CPUState));
  }

-  FEXCore::Context::ExitReason RunUntilExit(FEXCore::Context::Context *CTX) {
-    return CTX->RunUntilExit();
+  void FEXCore::Context::ContextImpl::SetCustomCPUBackendFactory(CustomCPUFactoryType Factory) {
+    CustomCPUFactory = std::move(Factory);
  }

-  int GetProgramStatus(const FEXCore::Context::Context *CTX) {
-    return CTX->GetProgramStatus();
+  HostFeatures FEXCore::Context::ContextImpl::GetHostFeatures() const {
+    return HostFeatures;
  }

-  FEXCore::Context::ExitReason GetExitReason(const FEXCore::Context::Context *CTX) {
-    return CTX->ParentThread->ExitReason;
+  void FEXCore::Context::ContextImpl::SetSignalDelegator(FEXCore::SignalDelegator *_SignalDelegation) {
+    SignalDelegation = _SignalDelegation;
  }

-  bool IsDone(const FEXCore::Context::Context *CTX) {
-    return CTX->IsPaused();
+  void FEXCore::Context::ContextImpl::SetSyscallHandler(FEXCore::HLE::SyscallHandler *Handler) {
+    SyscallHandler = Handler;
+    SourcecodeResolver = Handler->GetSourcecodeResolver();
  }

-  void GetCPUState(const FEXCore::Context::Context *CTX, FEXCore::Core::CPUState *State) {
-    memcpy(State, CTX->ParentThread->CurrentFrame, sizeof(FEXCore::Core::CPUState));
+  FEXCore::CPUID::FunctionResults FEXCore::Context::ContextImpl::RunCPUIDFunction(uint32_t Function, uint32_t Leaf) {
+    return CPUID.RunFunction(Function, Leaf);
  }

-  void SetCPUState(FEXCore::Context::Context *CTX, const FEXCore::Core::CPUState *State) {
-    memcpy(CTX->ParentThread->CurrentFrame, State, sizeof(FEXCore::Core::CPUState));
+  FEXCore::CPUID::XCRResults FEXCore::Context::ContextImpl::RunXCRFunction(uint32_t Function) {
+    return CPUID.RunXCRFunction(Function);
  }

-  void Pause(FEXCore::Context::Context *CTX) {
-    CTX->Pause();
+  FEXCore::CPUID::FunctionResults FEXCore::Context::ContextImpl::RunCPUIDFunctionName(uint32_t Function, uint32_t Leaf, uint32_t CPU) {
+    return CPUID.RunFunctionName(Function, Leaf, CPU);
  }
-
-  void Stop(FEXCore::Context::Context *CTX) {
-    CTX->Stop(false);
-  }
-
-  void SetCustomCPUBackendFactory(FEXCore::Context::Context *CTX, CustomCPUFactoryType Factory) {
-    CTX->CustomCPUFactory = std::move(Factory);
-  }
-
-  bool AddVirtualMemoryMapping([[maybe_unused]] FEXCore::Context::Context *CTX, [[maybe_unused]] uint64_t VirtualAddress, [[maybe_unused]] uint64_t PhysicalAddress, [[maybe_unused]] uint64_t Size) {
-    return false;
-  }
-
-  void RegisterExternalSyscallVisitor(FEXCore::Context::Context *CTX, [[maybe_unused]] uint64_t Syscall, [[maybe_unused]] FEXCore::HLE::SyscallVisitor *Visitor) {
-  }
-
-  void HandleCallback(FEXCore::Context::Context *CTX, FEXCore::Core::InternalThreadState *Thread, uint64_t RIP) {
-    CTX->HandleCallback(Thread, RIP);
-  }
-
-  void RegisterHostSignalHandler(FEXCore::Context::Context *CTX, int Signal, HostSignalDelegatorFunction Func, bool Required) {
-      CTX->RegisterHostSignalHandler(Signal, std::move(Func), Required);
-  }
-
-  void RegisterFrontendHostSignalHandler(FEXCore::Context::Context *CTX, int Signal, HostSignalDelegatorFunction Func, bool Required) {
-    CTX->RegisterFrontendHostSignalHandler(Signal, std::move(Func), Required);
-  }
-
-  FEXCore::Core::InternalThreadState* CreateThread(FEXCore::Context::Context *CTX, FEXCore::Core::CPUState *NewThreadState, uint64_t ParentTID) {
-    return CTX->CreateThread(NewThreadState, ParentTID);
-  }
-
-  void ExecutionThread(FEXCore::Context::Context *CTX, FEXCore::Core::InternalThreadState *Thread) {
-    return CTX->ExecutionThread(Thread);
-  }
-
-  void InitializeThread(FEXCore::Context::Context *CTX, FEXCore::Core::InternalThreadState *Thread) {
-    return CTX->InitializeThread(Thread);
-  }
-
-  void RunThread(FEXCore::Context::Context *CTX, FEXCore::Core::InternalThreadState *Thread) {
-    CTX->RunThread(Thread);
-  }
-
-  void StopThread(FEXCore::Context::Context *CTX, FEXCore::Core::InternalThreadState *Thread) {
-    CTX->StopThread(Thread);
-  }
-
-  void DestroyThread(FEXCore::Context::Context *CTX, FEXCore::Core::InternalThreadState *Thread) {
-    CTX->DestroyThread(Thread);
-  }
-
-  void CleanupAfterFork(FEXCore::Context::Context *CTX, FEXCore::Core::InternalThreadState *Thread) {
-    CTX->CleanupAfterFork(Thread);
-  }
-
-  void SetSignalDelegator(FEXCore::Context::Context *CTX, FEXCore::SignalDelegator *SignalDelegation) {
-    CTX->SignalDelegation = SignalDelegation;
-  }
-
-  void SetSyscallHandler(FEXCore::Context::Context *CTX, FEXCore::HLE::SyscallHandler *Handler) {
-    CTX->SyscallHandler = Handler;
-  }
-
-  FEXCore::CPUID::FunctionResults RunCPUIDFunction(FEXCore::Context::Context *CTX, uint32_t Function, uint32_t Leaf) {
-    return CTX->CPUID.RunFunction(Function, Leaf);
-  }
-
-  FEX_DEFAULT_VISIBILITY FEXCore::CPUID::FunctionResults RunCPUIDFunctionName(FEXCore::Context::Context *CTX, uint32_t Function, uint32_t Leaf, uint32_t CPU) {
-    return CTX->CPUID.RunFunctionName(Function, Leaf, CPU);
-  }
-
-  void SetAOTIRLoader(FEXCore::Context::Context *CTX, std::function<int(const std::string&)> CacheReader) {
-    CTX->SetAOTIRLoader(CacheReader);
-  }
-
-  void SetAOTIRWriter(FEXCore::Context::Context *CTX, std::function<std::unique_ptr<std::ofstream>(const std::string&)> CacheWriter) {
-    CTX->SetAOTIRWriter(CacheWriter);
-  }
-
-  void SetAOTIRRenamer(FEXCore::Context::Context *CTX, std::function<void(const std::string&)> CacheRenamer) {
-    CTX->SetAOTIRRenamer(CacheRenamer);
-  }
-
-  void FinalizeAOTIRCache(FEXCore::Context::Context *CTX) {
-    CTX->FinalizeAOTIRCache();
-  }
-
-  void WriteFilesWithCode(FEXCore::Context::Context *CTX, std::function<void(const std::string& fileid, const std::string& filename)> Writer) {
-    CTX->WriteFilesWithCode(Writer);
-  }
-
-  void AddNamedRegion(FEXCore::Context::Context *CTX, uintptr_t Base, uintptr_t Length, uintptr_t Offset, const std::string& Name) {
-    return CTX->AddNamedRegion(Base, Length, Offset, Name);
-  }
-  void RemoveNamedRegion(FEXCore::Context::Context *CTX, uintptr_t Base, uintptr_t Length) {
-    return CTX->RemoveNamedRegion(Base, Length);
-  }
-
-namespace Debug {
-  void CompileRIP(FEXCore::Context::Context *CTX, uint64_t RIP) {
-    CTX->CompileRIP(CTX->ParentThread, RIP);
-  }
-  uint64_t GetThreadCount(FEXCore::Context::Context *CTX) {
-    return CTX->GetThreadCount();
-  }
-
-  FEXCore::Core::RuntimeStats *GetRuntimeStatsForThread(FEXCore::Context::Context *CTX, uint64_t Thread) {
-    return CTX->GetRuntimeStatsForThread(Thread);
-  }
-
-  bool GetDebugDataForRIP(FEXCore::Context::Context *CTX, uint64_t RIP, FEXCore::Core::DebugData *Data) {
-    return CTX->GetDebugDataForRIP(RIP, Data);
-  }
-
-  bool FindHostCodeForRIP(FEXCore::Context::Context *CTX, uint64_t RIP, uint8_t **Code) {
-    return CTX->FindHostCodeForRIP(RIP, Code);
-  }
-
-  // XXX:
-  // bool FindIRForRIP(FEXCore::Context::Context *CTX, uint64_t RIP, FEXCore::IR::IntrusiveIRList **ir) {
-  //   return CTX->FindIRForRIP(RIP, ir);
-  // }
-
-  // void SetIRForRIP(FEXCore::Context::Context *CTX, uint64_t RIP, FEXCore::IR::IntrusiveIRList *const ir) {
-  //   CTX->SetIRForRIP(RIP, ir);
-  // }
-}
-
 }
@@ -2,45 +2,57 @@

 #include "Common/JitSymbols.h"
 #include "Interface/Core/CPUID.h"
-#include "Interface/Core/HostFeatures.h"
 #include "Interface/Core/X86HelperGen.h"
+#include "Interface/Core/ObjectCache/ObjectCacheService.h"
+#include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/IR/AOTIR.h"
 #include <FEXCore/Config/Config.h>
 #include <FEXCore/Core/Context.h>
 #include <FEXCore/Core/CoreState.h>
+#include <FEXCore/Core/HostFeatures.h>
 #include <FEXCore/Core/SignalDelegator.h>
 #include <FEXCore/Debug/InternalThreadState.h>
 #include <FEXCore/Utils/CompilerDefs.h>
+#include <FEXCore/Utils/DeferredSignalMutex.h>
 #include <FEXCore/Utils/Event.h>
+#include <FEXCore/fextl/memory.h>
+#include <FEXCore/fextl/set.h>
+#include <FEXCore/fextl/string.h>
+#include <FEXCore/fextl/unordered_map.h>
+#include <FEXCore/fextl/vector.h>
+#include <FEXHeaderUtils/Syscalls.h>
 #include <stdint.h>

-
 #include <atomic>
 #include <condition_variable>
 #include <functional>
 #include <istream>
-#include <map>
 #include <memory>
 #include <mutex>
 #include <shared_mutex>
 #include <stddef.h>
-#include <string>
-#include <unordered_map>
 #include <queue>
-#include <vector>

 namespace FEXCore {
 class CodeLoader;
 class ThunkHandler;
 class GdbServer;

+namespace CodeSerialize {
+  class CodeObjectSerializeService;
+}
+
 namespace CPU {
  class Arm64JITCore;
  class X86JITCore;
+  class InterpreterCore;
+  class Dispatcher;
 }
 namespace HLE {
 struct SyscallArguments;
 class SyscallHandler;
+class SourcecodeResolver;
+struct SourcecodeMap;
 }
 }

@@ -58,7 +70,130 @@ namespace FEXCore::Context {
    MODE_SINGLESTEP = 1,
  };

-  struct Context {
+  class ContextImpl final : public FEXCore::Context::Context {
+    public:
+      // Context base class implementation.
+      bool InitializeContext() override;
+
+      FEXCore::Core::InternalThreadState* InitCore(uint64_t InitialRIP, uint64_t StackPointer) override;
+
+      void SetExitHandler(ExitHandler handler) override;
+      ExitHandler GetExitHandler() const override;
+
+      void Pause() override;
+      void Run() override;
+      void Stop() override;
+      void Step() override;
+
+      ExitReason RunUntilExit() override;
+
+      void CompileRIP(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP) override;
+
+      int GetProgramStatus() const override;
+
+      ExitReason GetExitReason() override;
+
+      bool IsDone() const override;
+
+      void GetCPUState(FEXCore::Core::CPUState *State) const override;
+      void SetCPUState(const FEXCore::Core::CPUState *State) override;
+
+      void SetCustomCPUBackendFactory(CustomCPUFactoryType Factory) override;
+
+      HostFeatures GetHostFeatures() const override;
+
+      void HandleCallback(FEXCore::Core::InternalThreadState *Thread, uint64_t RIP) override;
+
+      uint64_t RestoreRIPFromHostPC(FEXCore::Core::InternalThreadState *Thread, uint64_t HostPC) override;
+
+      /**
+       * @brief Used to create FEX thread objects in preparation for creating a true OS thread. Does set a TID or PID.
+       *
+       * @param NewThreadState The initial thread state to setup for our state
+       * @param ParentTID The PID that was the parent thread that created this
+       *
+       * @return The InternalThreadState object that tracks all of the emulated thread's state
+       *
+       * Usecases:
+       *  OS thread Creation:
+       *    - Thread = CreateThread(NewState, PPID);
+       *    - InitializeThread(Thread);
+       *  OS fork (New thread created with a clone of thread state):
+       *    - clone{2, 3}
+       *    - Thread = CreateThread(CopyOfThreadState, PPID);
+       *    - ExecutionThread(Thread); // Starts executing without creating another host thread
+       *  Thunk callback executing guest code from native host thread
+       *    - Thread = CreateThread(NewState, PPID);
+       *    - InitializeThreadTLSData(Thread);
+       *    - HandleCallback(Thread, RIP);
+       */
+      FEXCore::Core::InternalThreadState* CreateThread(FEXCore::Core::CPUState *NewThreadState, uint64_t ParentTID) override;
+
+      // Public for threading
+      void ExecutionThread(FEXCore::Core::InternalThreadState *Thread) override;
+      /**
+       * @brief Initializes the OS thread object and prepares to start executing on that new OS thread
+       *
+       * @param Thread The internal FEX thread state object
+       *
+       * The OS thread will wait until RunThread is executed
+       */
+      void InitializeThread(FEXCore::Core::InternalThreadState *Thread) override;
+      /**
+       * @brief Starts the OS thread object to start executing guest code
+       *
+       * @param Thread The internal FEX thread state object
+       */
+      void RunThread(FEXCore::Core::InternalThreadState *Thread) override;
+      void StopThread(FEXCore::Core::InternalThreadState *Thread) override;
+      /**
+       * @brief Destroys this FEX thread object and stops tracking it internally
+       *
+       * @param Thread The internal FEX thread state object
+       */
+      void DestroyThread(FEXCore::Core::InternalThreadState *Thread) override;
+
+#ifndef _WIN32
+      void LockBeforeFork(FEXCore::Core::InternalThreadState *Thread) override;
+      void UnlockAfterFork(FEXCore::Core::InternalThreadState *Thread, bool Child) override;
+#endif
+      void SetSignalDelegator(FEXCore::SignalDelegator *SignalDelegation) override;
+      void SetSyscallHandler(FEXCore::HLE::SyscallHandler *Handler) override;
+
+      FEXCore::CPUID::FunctionResults RunCPUIDFunction(uint32_t Function, uint32_t Leaf) override;
+      FEXCore::CPUID::XCRResults RunXCRFunction(uint32_t Function) override;
+      FEXCore::CPUID::FunctionResults RunCPUIDFunctionName(uint32_t Function, uint32_t Leaf, uint32_t CPU) override;
+
+      FEXCore::IR::AOTIRCacheEntry *LoadAOTIRCacheEntry(const fextl::string& Name) override;
+      void UnloadAOTIRCacheEntry(FEXCore::IR::AOTIRCacheEntry *Entry) override;
+
+      void SetAOTIRLoader(AOTIRLoaderCBFn CacheReader) override {
+        IRCaptureCache.SetAOTIRLoader(std::move(CacheReader));
+      }
+      void SetAOTIRWriter(AOTIRWriterCBFn CacheWriter) override {
+        IRCaptureCache.SetAOTIRWriter(std::move(CacheWriter));
+      }
+      void SetAOTIRRenamer(AOTIRRenamerCBFn CacheRenamer) override {
+        IRCaptureCache.SetAOTIRRenamer(std::move(CacheRenamer));
+      }
+
+      void FinalizeAOTIRCache() override {
+        IRCaptureCache.FinalizeAOTIRCache();
+      }
+      void WriteFilesWithCode(AOTIRCodeFileWriterFn Writer) override {
+        IRCaptureCache.WriteFilesWithCode(Writer);
+      }
+      void InvalidateGuestCodeRange(FEXCore::Core::InternalThreadState *Thread, uint64_t Start, uint64_t Length) override;
+      void InvalidateGuestCodeRange(FEXCore::Core::InternalThreadState *Thread, uint64_t Start, uint64_t Length, CodeRangeInvalidationFn callback) override;
+      void MarkMemoryShared() override;
+
+      void ConfigureAOTGen(FEXCore::Core::InternalThreadState *Thread, fextl::set<uint64_t> *ExternalBranches, uint64_t SectionMaxAddress) override;
+      // returns false if a handler was already registered
+      CustomIRResult AddCustomIREntrypoint(uintptr_t Entrypoint, CustomIREntrypointHandler Handler, void *Creator = nullptr, void *Data = nullptr) override;
+
+      void AppendThunkDefinitions(fextl::vector<FEXCore::IR::ThunkDefinition> const& Definitions) override;
+
+    public:
    friend class FEXCore::HLE::SyscallHandler;
  #ifdef JIT_ARM64
    friend class FEXCore::CPU::Arm64JITCore;
@@ -67,6 +202,7 @@ namespace FEXCore::Context {
    friend class FEXCore::CPU::X86JITCore;
  #endif

+    friend class FEXCore::CPU::InterpreterCore;
    friend class FEXCore::IR::Validation::IRValidation;

    struct {
@@ -81,6 +217,7 @@ namespace FEXCore::Context {
      FEX_CONFIG_OPT(GdbServer, GDBSERVER);
      FEX_CONFIG_OPT(Is64BitMode, IS64BIT_MODE);
      FEX_CONFIG_OPT(TSOEnabled, TSOENABLED);
+      FEX_CONFIG_OPT(TSOAutoMigration, TSOAUTOMIGRATION);
      FEX_CONFIG_OPT(ABILocalFlags, ABILOCALFLAGS);
      FEX_CONFIG_OPT(ABINoPF, ABINOPF);
      FEX_CONFIG_OPT(AOTIRCapture, AOTIRCAPTURE);
@@ -91,25 +228,26 @@ namespace FEXCore::Context {
      FEX_CONFIG_OPT(MaxInstPerBlock, MAXINST);
      FEX_CONFIG_OPT(RootFSPath, ROOTFS);
      FEX_CONFIG_OPT(ThunkHostLibsPath, THUNKHOSTLIBS);
+      FEX_CONFIG_OPT(ThunkHostLibsPath32, THUNKHOSTLIBS32);
      FEX_CONFIG_OPT(ThunkConfigFile, THUNKCONFIG);
      FEX_CONFIG_OPT(DumpIR, DUMPIR);
      FEX_CONFIG_OPT(StaticRegisterAllocation, SRA);
      FEX_CONFIG_OPT(GlobalJITNaming, GLOBALJITNAMING);
      FEX_CONFIG_OPT(LibraryJITNaming, LIBRARYJITNAMING);
      FEX_CONFIG_OPT(BlockJITNaming, BLOCKJITNAMING);
+      FEX_CONFIG_OPT(GDBSymbols, GDBSYMBOLS);
      FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
+      FEX_CONFIG_OPT(CacheObjectCodeCompilation, CACHEOBJECTCODECOMPILATION);
+      FEX_CONFIG_OPT(x87ReducedPrecision, X87REDUCEDPRECISION);
    } Config;

-    using IntCallbackReturn =  FEX_NAKED void(*)(FEXCore::Core::InternalThreadState *Thread, volatile void *Host_RSP);
-    IntCallbackReturn InterpreterCallbackReturn;
-
    FEXCore::HostFeatures HostFeatures;

    std::mutex ThreadCreationMutex;
-    uint64_t ThreadID{};
-    FEXCore::Core::InternalThreadState* ParentThread;
-    std::vector<FEXCore::Core::InternalThreadState*> Threads;
+    FEXCore::Core::InternalThreadState* ParentThread{};
+    fextl::vector<FEXCore::Core::InternalThreadState*> Threads;
    std::atomic_bool CoreShuttingDown{false};
+    bool NeedToCheckXID{true};

    std::mutex IdleWaitMutex;
    std::condition_variable IdleWaitCV;
@@ -118,74 +256,76 @@ namespace FEXCore::Context {
    Event PauseWait;
    bool Running{};

+    FEXCore::ForkableSharedMutex CodeInvalidationMutex;
+
    FEXCore::CPUIDEmu CPUID;
    FEXCore::HLE::SyscallHandler *SyscallHandler{};
-    std::unique_ptr<FEXCore::ThunkHandler> ThunkHandler;
+    FEXCore::HLE::SourcecodeResolver *SourcecodeResolver{};
+    fextl::unique_ptr<FEXCore::ThunkHandler> ThunkHandler;
+    fextl::unique_ptr<FEXCore::CPU::Dispatcher> Dispatcher;

    CustomCPUFactoryType CustomCPUFactory;
    FEXCore::Context::ExitHandler CustomExitHandler;

 #ifdef BLOCKSTATS
-    std::unique_ptr<FEXCore::BlockSamplingData> BlockData;
+    fextl::unique_ptr<FEXCore::BlockSamplingData> BlockData;
 #endif

    SignalDelegator *SignalDelegation{};
    X86GeneratedCode X86CodeGen;

-    Context();
-    ~Context();
+    ContextImpl();
+    ~ContextImpl();

-    FEXCore::Core::InternalThreadState* InitCore(FEXCore::CodeLoader *Loader);
-    FEXCore::Context::ExitReason RunUntilExit();
-    int GetProgramStatus() const;
    bool IsPaused() const { return !Running; }
-    void Pause();
-    void Run();
    void WaitForThreadsToRun();
-    void Step();
    void Stop(bool IgnoreCurrentThread);
    void WaitForIdle();
-    void StopThread(FEXCore::Core::InternalThreadState *Thread);
    void SignalThread(FEXCore::Core::InternalThreadState *Thread, FEXCore::Core::SignalEvent Event);

    bool GetGdbServerStatus() const { return DebugServer != nullptr; }
    void StartGdbServer();
    void StopGdbServer();
-    void HandleCallback(FEXCore::Core::InternalThreadState *Thread, uint64_t RIP);
-    void RegisterHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required);
-    void RegisterFrontendHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required);

-    static void RemoveCodeEntry(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP);
+    static void ThreadRemoveCodeEntry(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP);
+    static void ThreadAddBlockLink(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestDestination, uintptr_t HostLink, const std::function<void()> &delinker);

-    // Wrapper which takes CpuStateFrame instead of InternalThreadState
-    static void RemoveCodeEntryFromJit(FEXCore::Core::CpuStateFrame *Frame, uint64_t GuestRIP) {
-      RemoveCodeEntry(Frame->Thread, GuestRIP);
+    template<auto Fn>
+    static uint64_t ThreadExitFunctionLink(FEXCore::Core::CpuStateFrame *Frame, uint64_t *record) {
+      auto Thread = Frame->Thread;
+      ScopedDeferredSignalWithForkableSharedLock lk(static_cast<ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);
+
+      return Fn(Frame, record);
    }

-    // Debugger interface
-    void CompileRIP(FEXCore::Core::InternalThreadState *Thread, uint64_t RIP);
-    uint64_t GetThreadCount() const;
-    FEXCore::Core::RuntimeStats *GetRuntimeStatsForThread(uint64_t Thread);
-    bool GetDebugDataForRIP(uint64_t RIP, FEXCore::Core::DebugData *Data);
-    bool FindHostCodeForRIP(uint64_t RIP, uint8_t **Code);
+    // Wrapper which takes CpuStateFrame instead of InternalThreadState and unique_locks CodeInvalidationMutex
+    // Must be called from owning thread
+    static void ThreadRemoveCodeEntryFromJit(FEXCore::Core::CpuStateFrame *Frame, uint64_t GuestRIP) {
+      auto Thread = Frame->Thread;
+
+      LogMan::Throw::AFmt(Thread->ThreadManager.GetTID() == FHU::Syscalls::gettid(), "Must be called from owning thread {}, not {}", Thread->ThreadManager.GetTID(), FHU::Syscalls::gettid());
+      ScopedDeferredSignalWithForkableUniqueLock lk(static_cast<ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);
+
+      ThreadRemoveCodeEntry(Thread, GuestRIP);
+    }
+
+    void RemoveCustomIREntrypoint(uintptr_t Entrypoint);

    struct GenerateIRResult {
      FEXCore::IR::IRListView* IRList;
-      // User's responsibility to deallocate this.
-      FEXCore::IR::RegisterAllocationData* RAData;
+      FEXCore::IR::RegisterAllocationData::UniquePtr RAData;
      uint64_t TotalInstructions;
      uint64_t TotalInstructionsLength;
      uint64_t StartAddr;
      uint64_t Length;
    };
-    [[nodiscard]] GenerateIRResult GenerateIR(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP);
+    [[nodiscard]] GenerateIRResult GenerateIR(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP, bool ExtendedDebugInfo);

    struct CompileCodeResult {
      void* CompiledCode;
      FEXCore::IR::IRListView* IRData;
      FEXCore::Core::DebugData* DebugData;
-      // User's responsibility to deallocate this.
-      FEXCore::IR::RegisterAllocationData* RAData;
+      FEXCore::IR::RegisterAllocationData::UniquePtr RAData;
      bool GeneratedIR;
      uint64_t StartAddr;
      uint64_t Length;
@@ -196,109 +336,66 @@ namespace FEXCore::Context {
    // same as CompileBlock, but aborts on failure
    void CompileBlockJit(FEXCore::Core::CpuStateFrame *Frame, uint64_t GuestRIP);

-    /**
-     * @brief Initializes the JIT compilers for the thread
-     *
-     * @param State The internal FEX thread state object
-     * @param CompileThread Is this for the compile service or not?
-     *
-     * InitializeCompiler is called inside of CreateThread, so you likely don't need this
-     * This is exposed because the CompileService needs to initialize compilers while copying data from
-     * the paired InternalThreadState that it is compiling code for
-     */
-    void InitializeCompiler(FEXCore::Core::InternalThreadState* State, bool CompileThread);
-
    // Used for thread creation from syscalls
    /**
-     * @brief Used to create FEX thread objects in preparation for creating a true OS thread
-     *
-     * @param NewThreadState The initial thread state to setup for our state
-     * @param ParentTID The PID that was the parent thread that created this
-     *
-     * @return The InternalThreadState object that tracks all of the emulated thread's state
-     *
-     * Usecases:
-     *  OS thread Creation:
-     *    - Thread = CreateThread(NewState, PPID);
-     *    - InitializeThread(Thread);
-     *  OS fork (New thread created with a clone of thread state):
-     *    - clone{2, 3}
-     *    - Thread = CreateThread(CopyOfThreadState, PPID);
-     *    - ExecutionThread(Thread); // Starts executing without creating another host thread
-     *  Thunk callback executing guest code from native host thread
-     *    - Thread = CreateThread(NewState, PPID);
-     *    - InitializeThreadTLSData(Thread);
-     *    - HandleCallback(Thread, RIP);
-     */
-    FEXCore::Core::InternalThreadState* CreateThread(FEXCore::Core::CPUState *NewThreadState, uint64_t ParentTID);
-
-    /**
-     * @brief Initializes the TLS data for a thread
+     * @brief Initializes TID, PID and TLS data for a thread
     *
     * @param Thread The internal FEX thread state object
     */
    void InitializeThreadTLSData(FEXCore::Core::InternalThreadState *Thread);

-    /**
-     * @brief Initializes the OS thread object and prepares to start executing on that new OS thread
-     *
-     * @param Thread The internal FEX thread state object
-     *
-     * The OS thread will wait until RunThread is executed
-     */
-    void InitializeThread(FEXCore::Core::InternalThreadState *Thread);
-
-    /**
-     * @brief Starts the OS thread object to start executing guest code
-     *
-     * @param Thread The internal FEX thread state object
-     */
-    void RunThread(FEXCore::Core::InternalThreadState *Thread);
-
-    /**
-     * @brief Destroys this FEX thread object and stops tracking it internally
-     *
-     * @param Thread The internal FEX thread state object
-     */
-    void DestroyThread(FEXCore::Core::InternalThreadState *Thread);
    void CopyMemoryMapping(FEXCore::Core::InternalThreadState *ParentThread, FEXCore::Core::InternalThreadState *ChildThread);

-    void CleanupAfterFork(FEXCore::Core::InternalThreadState *ExceptForThread);
-
-    std::vector<FEXCore::Core::InternalThreadState*>* GetThreads() { return &Threads; }
+    fextl::vector<FEXCore::Core::InternalThreadState*>* GetThreads() { return &Threads; }

    uint8_t GetGPRSize() const { return Config.Is64BitMode ? 8 : 4; }

-    void AddNamedRegion(uintptr_t Base, uintptr_t Size, uintptr_t Offset, const std::string &filename);
-    void RemoveNamedRegion(uintptr_t Base, uintptr_t Size);
-
    FEXCore::JITSymbols Symbols;

-    // Public for threading
-    void ExecutionThread(FEXCore::Core::InternalThreadState *Thread);
+    void GetVDSOSigReturn(VDSOSigReturn *VDSOPointers) override {
+      if (VDSOPointers->VDSO_kernel_sigreturn == nullptr) {
+        VDSOPointers->VDSO_kernel_sigreturn = reinterpret_cast<void*>(X86CodeGen.sigreturn_32);
+      }

-    void FinalizeAOTIRCache() {
-      IRCaptureCache.FinalizeAOTIRCache();
+      if (VDSOPointers->VDSO_kernel_rt_sigreturn == nullptr) {
+        VDSOPointers->VDSO_kernel_rt_sigreturn = reinterpret_cast<void*>(X86CodeGen.rt_sigreturn_32);
+      }
    }

-    void WriteFilesWithCode(std::function<void(const std::string& fileid, const std::string& filename)> Writer) {
-      IRCaptureCache.WriteFilesWithCode(Writer);
+    void IncrementIdleRefCount() override {
+      ++IdleWaitRefCount;
    }

-    void SetAOTIRLoader(std::function<int(const std::string&)> CacheReader) {
-      IRCaptureCache.SetAOTIRLoader(CacheReader);
+    FEXCore::Utils::PooledAllocatorVirtual OpDispatcherAllocator;
+    FEXCore::Utils::PooledAllocatorVirtual FrontendAllocator;
+
+    // If Atomic-based TSO emulation is enabled or not.
+    bool IsAtomicTSOEnabled() const { return AtomicTSOEmulationEnabled; }
+
+    void SetHardwareTSOSupport(bool HardwareTSOSupported) override {
+      SupportsHardwareTSO = HardwareTSOSupported;
+      UpdateAtomicTSOEmulationConfig();
    }

-    void SetAOTIRWriter(std::function<std::unique_ptr<std::ofstream>(const std::string&)> CacheWriter) {
-      IRCaptureCache.SetAOTIRWriter(CacheWriter);
-    }
+    void EnableExitOnHLT() override { ExitOnHLT = true; }

-    void SetAOTIRRenamer(std::function<void(const std::string&)> CacheRenamer) {
-      IRCaptureCache.SetAOTIRRenamer(CacheRenamer);
-    }
+    bool ExitOnHLTEnabled() const { return ExitOnHLT; }
+
+    FEXCore::CPU::CPUBackendFeatures BackendFeatures;

  protected:
-    void ClearCodeCache(FEXCore::Core::InternalThreadState *Thread, bool AlsoClearIRCache);
+    void ClearCodeCache(FEXCore::Core::InternalThreadState *Thread);
+
+    void UpdateAtomicTSOEmulationConfig() {
+      if (SupportsHardwareTSO) {
+        // If the hardware supports TSO then we don't need to emulate it through atomics.
+        AtomicTSOEmulationEnabled = false;
+      }
+      else {
+        // Atomic TSO emulation only enabled if the config option is enabled.
+        AtomicTSOEmulationEnabled = (IsMemoryShared || !Config.TSOAutoMigration) && Config.TSOEnabled;
+      }
+    }

  private:
    /**
@@ -310,22 +407,38 @@ namespace FEXCore::Context {
     */
    void InitializeThreadData(FEXCore::Core::InternalThreadState *Thread);

+    /**
+     * @brief Initializes the JIT compilers for the thread
+     *
+     * @param State The internal FEX thread state object
+     *
+     * InitializeCompiler is called inside of CreateThread, so you likely don't need this
+     */
+    void InitializeCompiler(FEXCore::Core::InternalThreadState* Thread);
+
    void WaitForIdleWithTimeout();

    void NotifyPause();

-    void AddBlockMapping(FEXCore::Core::InternalThreadState *Thread, uint64_t Address, void *Ptr, uint64_t Start, uint64_t Length);
-    FEXCore::CodeLoader *LocalLoader{};
+    void AddBlockMapping(FEXCore::Core::InternalThreadState *Thread, uint64_t Address, void *Ptr);

    // Entry Cache
-    uint64_t StartingRIP;
    std::mutex ExitMutex;
-    std::unique_ptr<GdbServer> DebugServer;
+    fextl::unique_ptr<GdbServer> DebugServer;

    IR::AOTIRCaptureCache IRCaptureCache;
+    fextl::unique_ptr<FEXCore::CodeSerialize::CodeObjectSerializeService> CodeObjectCacheService;

    bool StartPaused = false;
+    bool IsMemoryShared = false;
+    bool SupportsHardwareTSO = false;
+    bool AtomicTSOEmulationEnabled = true;
+    bool ExitOnHLT = false;
    FEX_CONFIG_OPT(AppFilename, APP_FILENAME);
+
+    std::shared_mutex CustomIRMutex;
+    fextl::unordered_map<uint64_t, std::tuple<CustomIREntrypointHandler, void *, void *>> CustomIRHandlers;
+    FEXCore::CPU::DispatcherConfig DispatcherConfig;
  };

  uint64_t HandleSyscall(FEXCore::HLE::SyscallHandler *Handler, FEXCore::Core::CpuStateFrame *Frame, FEXCore::HLE::SyscallArguments *Args);
@@ -1,47 +1,272 @@
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
+#include "FEXCore/Utils/AllocatorHooks.h"
+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
+#include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Context/Context.h"
+#include "Interface/HLE/Thunks/Thunks.h"

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

-#include "aarch64/cpu-aarch64.h"
-#include "cpu-features.h"
-#include "aarch64/instructions-aarch64.h"
-#include "utils-vixl.h"
+#include <aarch64/cpu-aarch64.h>
+#include <aarch64/instructions-aarch64.h>
+#include <cpu-features.h>
+#include <utils-vixl.h>

+#include <array>
 #include <tuple>
+#include <utility>

 namespace FEXCore::CPU {
-#define STATE x28
+// Register x18 is unused in the current configuration.
+// This is due to it being a platform register on wine platforms.
+// TODO: Allow x18 register allocation in the future to gain one more register.

-// We want vixl to not allocate a default buffer. Jit and dispatcher will manually create one.
-Arm64Emitter::Arm64Emitter(FEXCore::Context::Context *ctx, size_t size) : vixl::aarch64::Assembler(size, vixl::aarch64::PositionDependentCode) {
-  CPU.SetUp();
+namespace x64 {
+  // All but x19 and x29 are caller saved
+  constexpr std::array<FEXCore::ARMEmitter::Register, 16> SRA = {
+    FEXCore::ARMEmitter::Reg::r4, FEXCore::ARMEmitter::Reg::r5,
+    FEXCore::ARMEmitter::Reg::r6, FEXCore::ARMEmitter::Reg::r7,
+    FEXCore::ARMEmitter::Reg::r8, FEXCore::ARMEmitter::Reg::r9,
+    FEXCore::ARMEmitter::Reg::r10, FEXCore::ARMEmitter::Reg::r11,
+    FEXCore::ARMEmitter::Reg::r12, FEXCore::ARMEmitter::Reg::r13,
+    FEXCore::ARMEmitter::Reg::r14, FEXCore::ARMEmitter::Reg::r15,
+    FEXCore::ARMEmitter::Reg::r16, FEXCore::ARMEmitter::Reg::r17,
+    FEXCore::ARMEmitter::Reg::r19, FEXCore::ARMEmitter::Reg::r29
+  };

-  auto Features = vixl::CPUFeatures::InferFromOS();
-  if (ctx->HostFeatures.SupportsAtomics) {
-    // Hypervisor can hide this on the c630?
-    Features.Combine(vixl::CPUFeatures::Feature::kLORegions);
-  }
+  constexpr std::array<FEXCore::ARMEmitter::Register, 9> RA = {
+    // All these callee saved
+    FEXCore::ARMEmitter::Reg::r20, FEXCore::ARMEmitter::Reg::r21,
+    FEXCore::ARMEmitter::Reg::r22, FEXCore::ARMEmitter::Reg::r23,
+    FEXCore::ARMEmitter::Reg::r24, FEXCore::ARMEmitter::Reg::r25,
+    FEXCore::ARMEmitter::Reg::r26, FEXCore::ARMEmitter::Reg::r27,
+    FEXCore::ARMEmitter::Reg::r30,
+  };

-  SetCPUFeatures(Features);
+  constexpr std::array<std::pair<FEXCore::ARMEmitter::Register, FEXCore::ARMEmitter::Register>, 4> RAPair = {{
+    {FEXCore::ARMEmitter::Reg::r20, FEXCore::ARMEmitter::Reg::r21},
+    {FEXCore::ARMEmitter::Reg::r22, FEXCore::ARMEmitter::Reg::r23},
+    {FEXCore::ARMEmitter::Reg::r24, FEXCore::ARMEmitter::Reg::r25},
+    {FEXCore::ARMEmitter::Reg::r26, FEXCore::ARMEmitter::Reg::r27},
+  }};
+
+  // All are caller saved
+  constexpr std::array<FEXCore::ARMEmitter::VRegister, 16> SRAFPR = {
+    FEXCore::ARMEmitter::VReg::v16, FEXCore::ARMEmitter::VReg::v17,
+    FEXCore::ARMEmitter::VReg::v18, FEXCore::ARMEmitter::VReg::v19,
+    FEXCore::ARMEmitter::VReg::v20, FEXCore::ARMEmitter::VReg::v21,
+    FEXCore::ARMEmitter::VReg::v22, FEXCore::ARMEmitter::VReg::v23,
+    FEXCore::ARMEmitter::VReg::v24, FEXCore::ARMEmitter::VReg::v25,
+    FEXCore::ARMEmitter::VReg::v26, FEXCore::ARMEmitter::VReg::v27,
+    FEXCore::ARMEmitter::VReg::v28, FEXCore::ARMEmitter::VReg::v29,
+    FEXCore::ARMEmitter::VReg::v30, FEXCore::ARMEmitter::VReg::v31
+  };
+
+  //  v8..v15 = (lower 64bits) Callee saved
+  constexpr std::array<FEXCore::ARMEmitter::VRegister, 12> RAFPR = {
+    // v0 ~ v3 are used as temps.
+    // FEXCore::ARMEmitter::VReg::v0, FEXCore::ARMEmitter::VReg::v1,
+    // FEXCore::ARMEmitter::VReg::v2, FEXCore::ARMEmitter::VReg::v3,
+
+    FEXCore::ARMEmitter::VReg::v4, FEXCore::ARMEmitter::VReg::v5,
+    FEXCore::ARMEmitter::VReg::v6, FEXCore::ARMEmitter::VReg::v7,
+    FEXCore::ARMEmitter::VReg::v8, FEXCore::ARMEmitter::VReg::v9,
+    FEXCore::ARMEmitter::VReg::v10, FEXCore::ARMEmitter::VReg::v11,
+    FEXCore::ARMEmitter::VReg::v12, FEXCore::ARMEmitter::VReg::v13,
+    FEXCore::ARMEmitter::VReg::v14, FEXCore::ARMEmitter::VReg::v15,
+  };
 }

-void Arm64Emitter::LoadConstant(vixl::aarch64::Register Reg, uint64_t Constant) {
-  bool Is64Bit = Reg.IsX();
+namespace x32 {
+  // All but x19 and x29 are caller saved
+  constexpr std::array<FEXCore::ARMEmitter::Register, 8> SRA = {
+    FEXCore::ARMEmitter::Reg::r4, FEXCore::ARMEmitter::Reg::r5,
+    FEXCore::ARMEmitter::Reg::r6, FEXCore::ARMEmitter::Reg::r7,
+    FEXCore::ARMEmitter::Reg::r8, FEXCore::ARMEmitter::Reg::r9,
+    FEXCore::ARMEmitter::Reg::r10, FEXCore::ARMEmitter::Reg::r11,
+  };
+
+  constexpr std::array<FEXCore::ARMEmitter::Register, 17> RA = {
+    // All these callee saved
+    FEXCore::ARMEmitter::Reg::r20, FEXCore::ARMEmitter::Reg::r21,
+    FEXCore::ARMEmitter::Reg::r22, FEXCore::ARMEmitter::Reg::r23,
+    FEXCore::ARMEmitter::Reg::r24, FEXCore::ARMEmitter::Reg::r25,
+    FEXCore::ARMEmitter::Reg::r26, FEXCore::ARMEmitter::Reg::r27,
+
+    // Registers only available on 32-bit
+    // All these are caller saved (except for r19).
+    FEXCore::ARMEmitter::Reg::r12, FEXCore::ARMEmitter::Reg::r13,
+    FEXCore::ARMEmitter::Reg::r14, FEXCore::ARMEmitter::Reg::r15,
+    FEXCore::ARMEmitter::Reg::r16, FEXCore::ARMEmitter::Reg::r17,
+    FEXCore::ARMEmitter::Reg::r29, FEXCore::ARMEmitter::Reg::r30,
+
+    FEXCore::ARMEmitter::Reg::r19,
+  };
+
+  constexpr std::array<std::pair<FEXCore::ARMEmitter::Register, FEXCore::ARMEmitter::Register>, 8> RAPair = {{
+    {FEXCore::ARMEmitter::Reg::r20, FEXCore::ARMEmitter::Reg::r21},
+    {FEXCore::ARMEmitter::Reg::r22, FEXCore::ARMEmitter::Reg::r23},
+    {FEXCore::ARMEmitter::Reg::r24, FEXCore::ARMEmitter::Reg::r25},
+    {FEXCore::ARMEmitter::Reg::r26, FEXCore::ARMEmitter::Reg::r27},
+
+    {FEXCore::ARMEmitter::Reg::r12, FEXCore::ARMEmitter::Reg::r13},
+    {FEXCore::ARMEmitter::Reg::r14, FEXCore::ARMEmitter::Reg::r15},
+    {FEXCore::ARMEmitter::Reg::r16, FEXCore::ARMEmitter::Reg::r17},
+    {FEXCore::ARMEmitter::Reg::r29, FEXCore::ARMEmitter::Reg::r30},
+  }};
+
+  // All are caller saved
+  constexpr std::array<FEXCore::ARMEmitter::VRegister, 8> SRAFPR = {
+    FEXCore::ARMEmitter::VReg::v16, FEXCore::ARMEmitter::VReg::v17,
+    FEXCore::ARMEmitter::VReg::v18, FEXCore::ARMEmitter::VReg::v19,
+    FEXCore::ARMEmitter::VReg::v20, FEXCore::ARMEmitter::VReg::v21,
+    FEXCore::ARMEmitter::VReg::v22, FEXCore::ARMEmitter::VReg::v23,
+  };
+
+  //  v8..v15 = (lower 64bits) Callee saved
+  constexpr std::array<FEXCore::ARMEmitter::VRegister, 20> RAFPR = {
+    // v0 ~ v3 are used as temps.
+    // FEXCore::ARMEmitter::VReg::v0, FEXCore::ARMEmitter::VReg::v1,
+    // FEXCore::ARMEmitter::VReg::v2, FEXCore::ARMEmitter::VReg::v3,
+
+    FEXCore::ARMEmitter::VReg::v4, FEXCore::ARMEmitter::VReg::v5,
+    FEXCore::ARMEmitter::VReg::v6, FEXCore::ARMEmitter::VReg::v7,
+    FEXCore::ARMEmitter::VReg::v8, FEXCore::ARMEmitter::VReg::v9,
+    FEXCore::ARMEmitter::VReg::v10, FEXCore::ARMEmitter::VReg::v11,
+    FEXCore::ARMEmitter::VReg::v12, FEXCore::ARMEmitter::VReg::v13,
+    FEXCore::ARMEmitter::VReg::v14, FEXCore::ARMEmitter::VReg::v15,
+
+    FEXCore::ARMEmitter::VReg::v24, FEXCore::ARMEmitter::VReg::v25,
+    FEXCore::ARMEmitter::VReg::v26, FEXCore::ARMEmitter::VReg::v27,
+    FEXCore::ARMEmitter::VReg::v28, FEXCore::ARMEmitter::VReg::v29,
+    FEXCore::ARMEmitter::VReg::v30, FEXCore::ARMEmitter::VReg::v31
+  };
+}
+
+// We want vixl to not allocate a default buffer. Jit and dispatcher will manually create one.
+Arm64Emitter::Arm64Emitter(FEXCore::Context::ContextImpl *ctx, size_t size)
+  : Emitter(size ? (uint8_t*)FEXCore::Allocator::VirtualAlloc(size, true) : nullptr, size)
+  , EmitterCTX {ctx} {
+  CPU.SetUp();
+
+  // Number of register available is dependent on what operating mode the proccess is in.
+  if (EmitterCTX->Config.Is64BitMode()) {
+    StaticRegisters = x64::SRA;
+    GeneralRegisters = x64::RA;
+    GeneralPairRegisters = x64::RAPair;
+    StaticFPRegisters = x64::SRAFPR;
+    GeneralFPRegisters = x64::RAFPR;
+  }
+  else {
+    ConfiguredDynamicRegisterBase = std::span(x32::RA.begin() + 8, 8);
+
+    StaticRegisters = x32::SRA;
+    GeneralRegisters = x32::RA;
+    GeneralPairRegisters = x32::RAPair;
+
+    StaticFPRegisters = x32::SRAFPR;
+    GeneralFPRegisters = x32::RAFPR;
+  }
+}
+
+Arm64Emitter::~Arm64Emitter() {
+  auto BufferSize = GetBufferSize();
+  if (BufferSize) {
+    FEXCore::Allocator::VirtualFree(GetBufferBase(), BufferSize);
+  }
+}
+
+void Arm64Emitter::LoadConstant(ARMEmitter::Size s, ARMEmitter::Register Reg, uint64_t Constant, bool NOPPad) {
+  bool Is64Bit = s == ARMEmitter::Size::i64Bit;
  int Segments = Is64Bit ? 4 : 2;

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

-  movz(Reg, (Constant) & 0xFFFF, 0);
-  for (int i = 1; i < Segments; ++i) {
+  if ((Constant >> 32) == 0) {
+    // If the upper 32-bits is all zero, we can now switch to a 32-bit move.
+    s = ARMEmitter::Size::i32Bit;
+    Is64Bit = false;
+    Segments = 2;
+  }
+
+  // If this can be loaded with a mov bitmask.
+  const auto IsImm = vixl::aarch64::Assembler::IsImmLogical(Constant, RegSizeInBits(s));
+  if (IsImm) {
+    orr(s, Reg, ARMEmitter::Reg::zr, Constant);
+    if (NOPPad) {
+      nop(); nop(); nop();
+    }
+    return;
+  }
+
+  int NumMoves = 1;
+  int RequiredMoveSegments{};
+
+  // Count the number of move segments
+  // We only want to use ADRP+ADD if we have more than 1 segment
+  for (size_t i = 0; i < Segments; ++i) {
    uint16_t Part = (Constant >> (i * 16)) & 0xFFFF;
-    if (Part) {
-      movk(Reg, Part, i * 16);
+    if (Part != 0) {
+      ++RequiredMoveSegments;
+    }
+  }
+
+  // ADRP+ADD is specifically optimized in hardware
+  // Check if we can use this
+  auto PC = GetCursorAddress<uint64_t>();
+
+  // PC aligned to page
+  uint64_t AlignedPC = PC & ~0xFFFULL;
+
+  // Offset from aligned PC
+  int64_t AlignedOffset = static_cast<int64_t>(Constant) - static_cast<int64_t>(AlignedPC);
+
+  // If the aligned offset is within the 4GB window then we can use ADRP+ADD
+  // and the number of move segments more than 1
+  if (RequiredMoveSegments > 1 && vixl::IsInt32(AlignedOffset)) {
+    // If this is 4k page aligned then we only need ADRP
+    if ((AlignedOffset & 0xFFF) == 0) {
+      adrp(Reg, AlignedOffset >> 12);
+    }
+    else {
+      // If the constant is within 1MB of PC then we can still use ADR to load in a single instruction
+      // 21-bit signed integer here
+      int64_t SmallOffset = static_cast<int64_t>(Constant) - static_cast<int64_t>(PC);
+      if (vixl::IsInt21(SmallOffset)) {
+        adr(Reg, SmallOffset);
+      }
+      else {
+        // Need to use ADRP + ADD
+        adrp(Reg, AlignedOffset >> 12);
+        add(s, Reg, Reg, Constant & 0xFFF);
+        NumMoves = 2;
+      }
+    }
+  }
+  else {
+    movz(s, Reg, (Constant) & 0xFFFF, 0);
+    for (int i = 1; i < Segments; ++i) {
+      uint16_t Part = (Constant >> (i * 16)) & 0xFFFF;
+      if (Part) {
+        movk(s, Reg, Part, i * 16);
+        ++NumMoves;
+      }
+    }
+  }
+
+  if (NOPPad) {
+    for (int i = NumMoves; i < Segments; ++i) {
+      nop();
    }
  }
 }
@@ -49,198 +274,306 @@ void Arm64Emitter::LoadConstant(vixl::aarch64::Register Reg, uint64_t Constant)
 void Arm64Emitter::PushCalleeSavedRegisters() {
  // We need to save pairs of registers
  // We save r19-r30
-  MemOperand PairOffset(sp, -16, PreIndex);
-  const std::array<std::pair<vixl::aarch64::XRegister, vixl::aarch64::XRegister>, 6> CalleeSaved = {{
-    {x19, x20},
-    {x21, x22},
-    {x23, x24},
-    {x25, x26},
-    {x27, x28},
-    {x29, x30},
+  const fextl::vector<std::pair<ARMEmitter::XRegister, ARMEmitter::XRegister>> CalleeSaved = {{
+#ifdef _WIN32
+    // Platform register, Just save it twice to make logic easy.
+    {ARMEmitter::XReg::x18, ARMEmitter::XReg::x18},
+#endif
+    {ARMEmitter::XReg::x19, ARMEmitter::XReg::x20},
+    {ARMEmitter::XReg::x21, ARMEmitter::XReg::x22},
+    {ARMEmitter::XReg::x23, ARMEmitter::XReg::x24},
+    {ARMEmitter::XReg::x25, ARMEmitter::XReg::x26},
+    {ARMEmitter::XReg::x27, ARMEmitter::XReg::x28},
+    {ARMEmitter::XReg::x29, ARMEmitter::XReg::x30},
  }};

  for (auto &RegPair : CalleeSaved) {
-    stp(RegPair.first, RegPair.second, PairOffset);
+    stp<ARMEmitter::IndexType::PRE>(RegPair.first, RegPair.second, ARMEmitter::Reg::rsp, -16);
  }

  // Additionally we need to store the lower 64bits of v8-v15
  // Here's a fun thing, we can use two ST4 instructions to store everything
  // We just need a single sub to sp before that
  const std::array<
-    std::tuple<vixl::aarch64::VRegister,
-               vixl::aarch64::VRegister,
-               vixl::aarch64::VRegister,
-               vixl::aarch64::VRegister>, 2> FPRs = {{
-    {v8, v9, v10, v11},
-    {v12, v13, v14, v15},
+    std::tuple<ARMEmitter::DRegister,
+               ARMEmitter::DRegister,
+               ARMEmitter::DRegister,
+               ARMEmitter::DRegister>, 2> FPRs = {{
+    {ARMEmitter::DReg::d8, ARMEmitter::DReg::d9, ARMEmitter::DReg::d10, ARMEmitter::DReg::d11},
+    {ARMEmitter::DReg::d12, ARMEmitter::DReg::d13, ARMEmitter::DReg::d14, ARMEmitter::DReg::d15},
  }};

  uint32_t VectorSaveSize = sizeof(uint64_t) * 8;
-  sub(sp, sp, VectorSaveSize);
+  sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, VectorSaveSize);
  // SP supporting move
  // We just saved x19 so it is safe
-  add(x19, sp, 0);
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r19, ARMEmitter::Reg::rsp, 0);

-  MemOperand QuadOffset(x19, 32, PostIndex);
  for (auto &RegQuad : FPRs) {
-    st4(std::get<0>(RegQuad).D(),
-        std::get<1>(RegQuad).D(),
-        std::get<2>(RegQuad).D(),
-        std::get<3>(RegQuad).D(),
+    st4(ARMEmitter::SubRegSize::i64Bit,
+        std::get<0>(RegQuad),
+        std::get<1>(RegQuad),
+        std::get<2>(RegQuad),
+        std::get<3>(RegQuad),
        0,
-        QuadOffset);
+        ARMEmitter::Reg::r19,
+        32);
  }
 }

 void Arm64Emitter::PopCalleeSavedRegisters() {
  const std::array<
-    std::tuple<vixl::aarch64::VRegister,
-               vixl::aarch64::VRegister,
-               vixl::aarch64::VRegister,
-               vixl::aarch64::VRegister>, 2> FPRs = {{
-    {v12, v13, v14, v15},
-    {v8, v9, v10, v11},
+    std::tuple<ARMEmitter::DRegister,
+               ARMEmitter::DRegister,
+               ARMEmitter::DRegister,
+               ARMEmitter::DRegister>, 2> FPRs = {{
+    {ARMEmitter::DReg::d12, ARMEmitter::DReg::d13, ARMEmitter::DReg::d14, ARMEmitter::DReg::d15},
+    {ARMEmitter::DReg::d8, ARMEmitter::DReg::d9, ARMEmitter::DReg::d10, ARMEmitter::DReg::d11},
  }};

-  MemOperand QuadOffset(sp, 32, PostIndex);
  for (auto &RegQuad : FPRs) {
-    ld4(std::get<0>(RegQuad).D(),
-        std::get<1>(RegQuad).D(),
-        std::get<2>(RegQuad).D(),
-        std::get<3>(RegQuad).D(),
+    ld4(ARMEmitter::SubRegSize::i64Bit,
+        std::get<0>(RegQuad),
+        std::get<1>(RegQuad),
+        std::get<2>(RegQuad),
+        std::get<3>(RegQuad),
        0,
-        QuadOffset);
+        ARMEmitter::Reg::rsp,
+        32);
  }

-  MemOperand PairOffset(sp, 16, PostIndex);
-  const std::array<std::pair<vixl::aarch64::XRegister, vixl::aarch64::XRegister>, 6> CalleeSaved = {{
-    {x29, x30},
-    {x27, x28},
-    {x25, x26},
-    {x23, x24},
-    {x21, x22},
-    {x19, x20},
+  const fextl::vector<std::pair<ARMEmitter::XRegister, ARMEmitter::XRegister>> CalleeSaved = {{
+    {ARMEmitter::XReg::x29, ARMEmitter::XReg::x30},
+    {ARMEmitter::XReg::x27, ARMEmitter::XReg::x28},
+    {ARMEmitter::XReg::x25, ARMEmitter::XReg::x26},
+    {ARMEmitter::XReg::x23, ARMEmitter::XReg::x24},
+    {ARMEmitter::XReg::x21, ARMEmitter::XReg::x22},
+    {ARMEmitter::XReg::x19, ARMEmitter::XReg::x20},
+#ifdef _WIN32
+    // Platform register.
+    {ARMEmitter::XReg::x18, ARMEmitter::XReg::zr},
+#endif
  }};

  for (auto &RegPair : CalleeSaved) {
-    ldp(RegPair.first, RegPair.second, PairOffset);
+    ldp<ARMEmitter::IndexType::POST>(RegPair.first, RegPair.second, ARMEmitter::Reg::rsp, 16);
  }
 }

+void Arm64Emitter::SpillStaticRegs(FEXCore::ARMEmitter::Register TmpReg, bool FPRs, uint32_t GPRSpillMask, uint32_t FPRSpillMask) {
+  if (!StaticRegisterAllocation()) {
+    return;
+  }

-void Arm64Emitter::SpillStaticRegs(bool FPRs, uint32_t SpillMask) {
-  if (StaticRegisterAllocation()) {
-    for (size_t i = 0; i < SRA64.size(); i+=2) {
-      auto Reg1 = SRA64[i];
-      auto Reg2 = SRA64[i+1];
-      if (((1U << Reg1.GetCode()) & SpillMask) &&
-          ((1U << Reg2.GetCode()) & SpillMask)) {
-        stp(Reg1, Reg2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i])));
-      }
-      else if (((1U << Reg1.GetCode()) & SpillMask)) {
-        str(Reg1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i])));
-      }
-      else if (((1U << Reg2.GetCode()) & SpillMask)) {
-        str(Reg2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i+1])));
-      }
+  for (size_t i = 0; i < StaticRegisters.size(); i+=2) {
+    auto Reg1 = StaticRegisters[i];
+    auto Reg2 = StaticRegisters[i+1];
+    if (((1U << Reg1.Idx()) & GPRSpillMask) &&
+        ((1U << Reg2.Idx()) & GPRSpillMask)) {
+      stp<ARMEmitter::IndexType::OFFSET>(Reg1.X(), Reg2.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i]));
    }
+    else if (((1U << Reg1.Idx()) & GPRSpillMask)) {
+      str(Reg1.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i]));
+    }
+    else if (((1U << Reg2.Idx()) & GPRSpillMask)) {
+      str(Reg2.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i+1]));
+    }
+  }

-    if (FPRs) {
-      for (size_t i = 0; i < SRAFPR.size(); i+=2) {
-        stp(SRAFPR[i].Q(), SRAFPR[i+1].Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[i][0])));
+  if (FPRs) {
+    if (EmitterCTX->HostFeatures.SupportsAVX) {
+      for (size_t i = 0; i < StaticFPRegisters.size(); i++) {
+        const auto Reg = StaticFPRegisters[i];
+
+        if (((1U << Reg.Idx()) & FPRSpillMask) != 0) {
+          mov(ARMEmitter::Size::i64Bit, TMP4.R(), offsetof(Core::CpuStateFrame, State.xmm.avx.data[i][0]));
+          st1b<ARMEmitter::SubRegSize::i8Bit>(Reg.Z(), PRED_TMP_32B, STATE.R(), TMP4.R());
+        }
+      }
+    } else {
+      if (GPRSpillMask && FPRSpillMask == ~0U) {
+        // Optimize the common case where we can spill four registers per instruction
+        // Load the sse offset in to the temporary register
+        add(ARMEmitter::Size::i64Bit, TmpReg, STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[0][0]));
+        for (size_t i = 0; i < StaticFPRegisters.size(); i += 4) {
+          const auto Reg1 = StaticFPRegisters[i];
+          const auto Reg2 = StaticFPRegisters[i + 1];
+          const auto Reg3 = StaticFPRegisters[i + 2];
+          const auto Reg4 = StaticFPRegisters[i + 3];
+          st1<ARMEmitter::SubRegSize::i64Bit>(Reg1.Q(), Reg2.Q(), Reg3.Q(), Reg4.Q(), TmpReg, 64);
+        }
+      }
+      else {
+        for (size_t i = 0; i < StaticFPRegisters.size(); i += 2) {
+          const auto Reg1 = StaticFPRegisters[i];
+          const auto Reg2 = StaticFPRegisters[i + 1];
+
+          if (((1U << Reg1.Idx()) & FPRSpillMask) &&
+              ((1U << Reg2.Idx()) & FPRSpillMask)) {
+            stp<ARMEmitter::IndexType::OFFSET>(Reg1.Q(), Reg2.Q(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i][0]));
+          }
+          else if (((1U << Reg1.Idx()) & FPRSpillMask)) {
+            str(Reg1.Q(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i][0]));
+          }
+          else if (((1U << Reg2.Idx()) & FPRSpillMask)) {
+            str(Reg2.Q(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i+1][0]));
+          }
+        }
      }
    }
  }
 }

-void Arm64Emitter::FillStaticRegs(bool FPRs, uint32_t FillMask) {
-  if (StaticRegisterAllocation()) {
-    for (size_t i = 0; i < SRA64.size(); i+=2) {
-      auto Reg1 = SRA64[i];
-      auto Reg2 = SRA64[i+1];
-      if (((1U << Reg1.GetCode()) & FillMask) &&
-          ((1U << Reg2.GetCode()) & FillMask)) {
-        ldp(Reg1, Reg2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i])));
+void Arm64Emitter::FillStaticRegs(bool FPRs, uint32_t GPRFillMask, uint32_t FPRFillMask) {
+  if (!StaticRegisterAllocation()) {
+    return;
+  }
+
+  if (FPRs) {
+    if (EmitterCTX->HostFeatures.SupportsAVX) {
+      // Set up predicate registers.
+      // We don't bother spilling these in SpillStaticRegs,
+      // since all that matters is we restore them on a fill.
+      // It's not a concern if they get trounced by something else.
+      ptrue<ARMEmitter::SubRegSize::i8Bit>(PRED_TMP_16B, ARMEmitter::PredicatePattern::SVE_VL16);
+      ptrue<ARMEmitter::SubRegSize::i8Bit>(PRED_TMP_32B, ARMEmitter::PredicatePattern::SVE_VL32);
+
+      for (size_t i = 0; i < StaticFPRegisters.size(); i++) {
+        const auto Reg = StaticFPRegisters[i];
+        if (((1U << Reg.Idx()) & FPRFillMask) != 0) {
+          mov(ARMEmitter::Size::i64Bit, TMP4.R(), offsetof(Core::CpuStateFrame, State.xmm.avx.data[i][0]));
+          ld1b<ARMEmitter::SubRegSize::i8Bit>(Reg.Z(), PRED_TMP_32B.Zeroing(), STATE.R(), TMP4.R());
+        }
      }
-      else if (((1U << Reg1.GetCode()) & FillMask)) {
-        ldr(Reg1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i])));
+    } else {
+      if (GPRFillMask && FPRFillMask == ~0U) {
+        // Optimize the common case where we can fill four registers per instruction.
+        // Use one of the filling static registers before we fill it.
+        auto TmpReg = StaticRegisters[FindFirstSetBit(GPRFillMask)];
+
+        // Load the sse offset in to the temporary register
+        add(ARMEmitter::Size::i64Bit, TmpReg, STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[0][0]));
+        for (size_t i = 0; i < StaticFPRegisters.size(); i += 4) {
+          const auto Reg1 = StaticFPRegisters[i];
+          const auto Reg2 = StaticFPRegisters[i + 1];
+          const auto Reg3 = StaticFPRegisters[i + 2];
+          const auto Reg4 = StaticFPRegisters[i + 3];
+          ld1<ARMEmitter::SubRegSize::i64Bit>(Reg1.Q(), Reg2.Q(), Reg3.Q(), Reg4.Q(), TmpReg, 64);
+        }
      }
-      else if (((1U << Reg2.GetCode()) & FillMask)) {
-        ldr(Reg2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i+1])));
+      else {
+        for (size_t i = 0; i < StaticFPRegisters.size(); i += 2) {
+          const auto Reg1 = StaticFPRegisters[i];
+          const auto Reg2 = StaticFPRegisters[i + 1];
+
+          if (((1U << Reg1.Idx()) & FPRFillMask) &&
+              ((1U << Reg2.Idx()) & FPRFillMask)) {
+            ldp<ARMEmitter::IndexType::OFFSET>(Reg1.Q(), Reg2.Q(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i][0]));
+          }
+          else if (((1U << Reg1.Idx()) & FPRFillMask)) {
+            ldr(Reg1.Q(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i][0]));
+          }
+          else if (((1U << Reg2.Idx()) & FPRFillMask)) {
+            ldr(Reg2.Q(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i+1][0]));
+          }
+        }
      }
    }
+  }

-    if (FPRs) {
-      for (size_t i = 0; i < SRAFPR.size(); i+=2) {
-        ldp(SRAFPR[i].Q(), SRAFPR[i+1].Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[i][0])));
-      }
+  for (size_t i = 0; i < StaticRegisters.size(); i+=2) {
+    auto Reg1 = StaticRegisters[i];
+    auto Reg2 = StaticRegisters[i+1];
+    if (((1U << Reg1.Idx()) & GPRFillMask) &&
+        ((1U << Reg2.Idx()) & GPRFillMask)) {
+      ldp<ARMEmitter::IndexType::OFFSET>(Reg1.X(), Reg2.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i]));
+    }
+    else if ((1U << Reg1.Idx()) & GPRFillMask) {
+      ldr(Reg1.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i]));
+    }
+    else if ((1U << Reg2.Idx()) & GPRFillMask) {
+      ldr(Reg2.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i+1]));
    }
  }
 }

-void Arm64Emitter::PushDynamicRegsAndLR() {
-  uint64_t SPOffset = AlignUp((RA64.size() + 1) * 8 + RAFPR.size() * 16, 16);
+void Arm64Emitter::PushDynamicRegsAndLR(FEXCore::ARMEmitter::Register TmpReg) {
+  const auto CanUseSVE = EmitterCTX->HostFeatures.SupportsAVX;
+  const auto GPRSize = (ConfiguredDynamicRegisterBase.size() + 1) * Core::CPUState::GPR_REG_SIZE;
+  const auto FPRRegSize = CanUseSVE ? Core::CPUState::XMM_AVX_REG_SIZE
+                                    : Core::CPUState::XMM_SSE_REG_SIZE;
+  const auto FPRSize = GeneralFPRegisters.size() * FPRRegSize;
+  const uint64_t SPOffset = AlignUp(GPRSize + FPRSize, 16);

-  sub(sp, sp, SPOffset);
-  int i = 0;
+  sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, SPOffset);

-  for (auto RA : RAFPR)
-  {
-    str(RA.Q(), MemOperand(sp, i * 8));
-    i+=2;
+  // rsp capable move
+  add(ARMEmitter::Size::i64Bit, TmpReg, ARMEmitter::Reg::rsp, 0);
+
+  if (CanUseSVE) {
+    for (size_t i = 0; i < GeneralFPRegisters.size(); i += 4) {
+      const auto Reg1 = GeneralFPRegisters[i];
+      const auto Reg2 = GeneralFPRegisters[i + 1];
+      const auto Reg3 = GeneralFPRegisters[i + 2];
+      const auto Reg4 = GeneralFPRegisters[i + 3];
+      st4b(Reg1.Z(), Reg2.Z(), Reg3.Z(), Reg4.Z(), PRED_TMP_32B, TmpReg, 0);
+      add(ARMEmitter::Size::i64Bit, TmpReg, TmpReg, 32 * 4);
+    }
+  } else {
+    LOGMAN_THROW_A_FMT(GeneralFPRegisters.size() % 4 == 0, "Needs to have multiple of 4 FPRs for RA");
+    for (size_t i = 0; i < GeneralFPRegisters.size(); i += 4) {
+      const auto Reg1 = GeneralFPRegisters[i];
+      const auto Reg2 = GeneralFPRegisters[i + 1];
+      const auto Reg3 = GeneralFPRegisters[i + 2];
+      const auto Reg4 = GeneralFPRegisters[i + 3];
+      st1<ARMEmitter::SubRegSize::i64Bit>(Reg1.Q(), Reg2.Q(), Reg3.Q(), Reg4.Q(), TmpReg, 64);
+    }
  }

-#if 0 // All GPRs should be caller saved
-  for (auto RA : RA64)
-  {
-    str(RA, MemOperand(sp, i * 8));
-    i++;
+  for (size_t i = 0; i < ConfiguredDynamicRegisterBase.size(); i += 2) {
+    const auto Reg1 = ConfiguredDynamicRegisterBase[i];
+    const auto Reg2 = ConfiguredDynamicRegisterBase[i + 1];
+    stp<ARMEmitter::IndexType::POST>(Reg1.X(), Reg2.X(), TmpReg, 16);
  }
-#endif

-  str(lr, MemOperand(sp, i * 8));
+  str(ARMEmitter::XReg::lr, TmpReg, 0);
 }

 void Arm64Emitter::PopDynamicRegsAndLR() {
-  uint64_t SPOffset = AlignUp((RA64.size() + 1) * 8 + RAFPR.size() * 16, 16);
-  int i = 0;
+  const auto CanUseSVE = EmitterCTX->HostFeatures.SupportsAVX;

-  for (auto RA : RAFPR)
-  {
-    ldr(RA.Q(), MemOperand(sp, i * 8));
-    i+=2;
-  }
-
-#if 0 // All GPRs should be caller saved
-  for (auto RA : RA64)
-  {
-    ldr(RA, MemOperand(sp, i * 8));
-    i++;
-  }
-#endif
-
-  ldr(lr, MemOperand(sp, i * 8));
-
-  add(sp, sp, SPOffset);
-}
-
-void Arm64Emitter::ResetStack() {
-  if (SpillSlots == 0)
-    return;
-
-  if (IsImmAddSub(SpillSlots * 16)) {
-    add(sp, sp, SpillSlots * 16);
+  if (CanUseSVE) {
+    for (size_t i = 0; i < GeneralFPRegisters.size(); i += 4) {
+      const auto Reg1 = GeneralFPRegisters[i];
+      const auto Reg2 = GeneralFPRegisters[i + 1];
+      const auto Reg3 = GeneralFPRegisters[i + 2];
+      const auto Reg4 = GeneralFPRegisters[i + 3];
+      ld4b(Reg1.Z(), Reg2.Z(), Reg3.Z(), Reg4.Z(), PRED_TMP_32B.Zeroing(), ARMEmitter::Reg::rsp);
+      add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, 32 * 4);
+    }
  } else {
-   // Too big to fit in a 12bit immediate
-   LoadConstant(x0, SpillSlots * 16);
-   add(sp, sp, x0);
+    for (size_t i = 0; i < GeneralFPRegisters.size(); i += 4) {
+      const auto Reg1 = GeneralFPRegisters[i];
+      const auto Reg2 = GeneralFPRegisters[i + 1];
+      const auto Reg3 = GeneralFPRegisters[i + 2];
+      const auto Reg4 = GeneralFPRegisters[i + 3];
+      ld1<ARMEmitter::SubRegSize::i64Bit>(Reg1.Q(), Reg2.Q(), Reg3.Q(), Reg4.Q(), ARMEmitter::Reg::rsp, 64);
+    }
  }
+
+  for (size_t i = 0; i < ConfiguredDynamicRegisterBase.size(); i += 2) {
+    const auto Reg1 = ConfiguredDynamicRegisterBase[i];
+    const auto Reg2 = ConfiguredDynamicRegisterBase[i + 1];
+    ldp<ARMEmitter::IndexType::POST>(Reg1.X(), Reg2.X(), ARMEmitter::Reg::rsp, 16);
+  }
+
+  ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
 }

 void Arm64Emitter::Align16B() {
  uint64_t CurrentOffset = GetCursorAddress<uint64_t>();
  for (uint64_t i = (16 - (CurrentOffset & 0xF)); i != 0; i -= 4) {
-      nop();
+    nop();
  }
 }

@@ -1,81 +1,182 @@
 #pragma once

-#include "aarch64/assembler-aarch64.h"
-#include "aarch64/constants-aarch64.h"
-#include "aarch64/cpu-aarch64.h"
-#include "aarch64/operands-aarch64.h"
-#include "platform-vixl.h"
-#include "FEXCore/Config/Config.h"
+#include "FEXCore/Utils/EnumUtils.h"
+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
+#include "Interface/Core/ArchHelpers/CodeEmitter/Registers.h"
+
+#include "Interface/Core/Dispatcher/Dispatcher.h"
+#include "Interface/Core/ObjectCache/Relocations.h"
+
+#include <aarch64/assembler-aarch64.h>
+#include <aarch64/constants-aarch64.h>
+#include <aarch64/cpu-aarch64.h>
+#include <aarch64/operands-aarch64.h>
+#include <platform-vixl.h>
+#ifdef VIXL_DISASSEMBLER
+#include <aarch64/disasm-aarch64.h>
+#endif
+#ifdef VIXL_SIMULATOR
+#include <aarch64/simulator-aarch64.h>
+#include <aarch64/simulator-constants-aarch64.h>
+#endif
+
+#include <FEXCore/Config/Config.h>

 #include <array>
-#include <stddef.h>
-#include <stdint.h>
+#include <cstddef>
+#include <cstdint>
 #include <utility>
+#include <span>

 namespace FEXCore::CPU {
-using namespace vixl;
-using namespace vixl::aarch64;
+// Contains the address to the currently available CPU state
+constexpr auto STATE = FEXCore::ARMEmitter::XReg::x28;

-// All but x29 are caller saved
-const std::array<aarch64::Register, 16> SRA64 = {
-  x4, x5, x6, x7, x8, x9, x10, x11,
-  x12, x18, x17, x16, x15, x14, x13, x29
-};
+// GPR temporaries. Only x3 can be used across spill boundaries
+// so if these ever need to change, be very careful about that.
+constexpr auto TMP1 = FEXCore::ARMEmitter::XReg::x0;
+constexpr auto TMP2 = FEXCore::ARMEmitter::XReg::x1;
+constexpr auto TMP3 = FEXCore::ARMEmitter::XReg::x2;
+constexpr auto TMP4 = FEXCore::ARMEmitter::XReg::x3;

-// All are callee saved
-const std::array<aarch64::Register, 9> RA64 = {
-  x20, x21, x22, x23, x24, x25, x26, x27,
-  x19
-};
+// Vector temporaries
+constexpr auto VTMP1 = FEXCore::ARMEmitter::VReg::v0;
+constexpr auto VTMP2 = FEXCore::ARMEmitter::VReg::v1;
+constexpr auto VTMP3 = FEXCore::ARMEmitter::VReg::v2;
+constexpr auto VTMP4 = FEXCore::ARMEmitter::VReg::v3;

-const std::array<std::pair<aarch64::Register, aarch64::Register>, 4>  RA64Pair = {{
-  {x20, x21},
-  {x22, x23},
-  {x24, x25},
-  {x26, x27},
-}};
-
-const std::array<std::pair<aarch64::Register, aarch64::Register>, 4> RA32Pair = {{
-  {w20, w21},
-  {w22, w23},
-  {w24, w25},
-  {w26, w27},
-}};
-
-// All are caller saved
-const std::array<aarch64::VRegister, 16> SRAFPR = {
-  v16, v17, v18, v19, v20, v21, v22, v23,
-  v24, v25, v26, v27, v28, v29, v30, v31
-};
-
-//  v8..v15 = (lower 64bits) Callee saved
-const std::array<aarch64::VRegister, 12> RAFPR = {
-/*v0,  v1,  v2,  v3,*/v4,  v5,  v6,  v7,  // v0 ~ v3 are used as temps
-  v8,  v9,  v10, v11, v12, v13, v14, v15
-};
+// Predicate register temporaries (used when AVX support is enabled)
+// PRED_TMP_16B indicates a predicate register that indicates the first 16 bytes set to 1.
+// PRED_TMP_32B indicates a predicate register that indicates the first 32 bytes set to 1.
+constexpr FEXCore::ARMEmitter::PRegister PRED_TMP_16B = FEXCore::ARMEmitter::PReg::p6;
+constexpr FEXCore::ARMEmitter::PRegister PRED_TMP_32B = FEXCore::ARMEmitter::PReg::p7;

 // This class contains common emitter utility functions that can
 // be used by both Arm64 JIT and ARM64 Dispatcher
-class Arm64Emitter : public vixl::aarch64::Assembler {
+class Arm64Emitter : public FEXCore::ARMEmitter::Emitter {
 protected:
-  Arm64Emitter(FEXCore::Context::Context *ctx, size_t size);
+  Arm64Emitter(FEXCore::Context::ContextImpl *ctx, size_t size);
+  ~Arm64Emitter();

+  FEXCore::Context::ContextImpl *EmitterCTX;
  vixl::aarch64::CPU CPU;
-  void LoadConstant(vixl::aarch64::Register Reg, uint64_t Constant);
-  void SpillStaticRegs(bool FPRs = true, uint32_t SpillMask = ~0U);
-  void FillStaticRegs(bool FPRs = true, uint32_t FillMask = ~0U);

-  void PushDynamicRegsAndLR();
+  std::span<const FEXCore::ARMEmitter::Register> ConfiguredDynamicRegisterBase{};
+  std::span<const FEXCore::ARMEmitter::Register> StaticRegisters{};
+  std::span<const FEXCore::ARMEmitter::Register> GeneralRegisters{};
+  std::span<const std::pair<FEXCore::ARMEmitter::Register, FEXCore::ARMEmitter::Register>> GeneralPairRegisters{};
+  std::span<const FEXCore::ARMEmitter::VRegister> StaticFPRegisters{};
+  std::span<const FEXCore::ARMEmitter::VRegister> GeneralFPRegisters{};
+
+  /**
+   * @name Register Allocation
+   * @{ */
+  constexpr static uint32_t RegisterClasses = 6;
+
+  constexpr static uint64_t GPRBase = (0ULL << 32);
+  constexpr static uint64_t FPRBase = (1ULL << 32);
+  constexpr static uint64_t GPRPairBase = (2ULL << 32);
+
+  /**  @} */
+
+  constexpr static uint8_t RA_32 = 0;
+  constexpr static uint8_t RA_64 = 1;
+  constexpr static uint8_t RA_FPR = 2;
+
+  void LoadConstant(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register Reg, uint64_t Constant, bool NOPPad = false);
+
+
+  // NOTE: These functions WILL clobber the register TMP4 if AVX support is enabled
+  //       and FPRs are being spilled or filled. If only GPRs are spilled/filled, then
+  //       TMP4 is left alone.
+  void SpillStaticRegs(FEXCore::ARMEmitter::Register TmpReg, bool FPRs = true, uint32_t GPRSpillMask = ~0U, uint32_t FPRSpillMask = ~0U);
+  void FillStaticRegs(bool FPRs = true, uint32_t GPRFillMask = ~0U, uint32_t FPRFillMask = ~0U);
+
+  // Register 0-18 + 29 + 30 are caller saved
+  static constexpr uint32_t CALLER_GPR_MASK = 0b0110'0000'0000'0111'1111'1111'1111'1111U;
+
+  // This isn't technically true because the lower 64-bits of v8..v15 are callee saved
+  // We can't guarantee only the lower 64bits are used so flush everything
+  static constexpr uint32_t CALLER_FPR_MASK = ~0U;
+
+  void PushDynamicRegsAndLR(FEXCore::ARMEmitter::Register TmpReg);
  void PopDynamicRegsAndLR();

  void PushCalleeSavedRegisters();
  void PopCalleeSavedRegisters();

-  void ResetStack();
  void Align16B();

-  uint32_t SpillSlots{};
+#ifdef VIXL_SIMULATOR
+  // Generates a vixl simulator runtime call.
+  //
+  // This matches behaviour of vixl's macro assembler, but we need to reimplement it since we aren't using the macro assembler.
+  // This isn't too complex with how vixl emits this.
+  //
+  // Emit:
+  // 1) hlt(kRuntimeCallOpcode)
+  // 2) Simulator wrapper handler
+  // 3) Function to call
+  // 4) Style of the function call (Call versus tail-call)

+  template<typename R, typename... P>
+  void GenerateRuntimeCall(R (*Function)(P...)) {
+    uintptr_t SimulatorWrapperAddress = reinterpret_cast<uintptr_t>(
+      &(vixl::aarch64::Simulator::RuntimeCallStructHelper<R, P...>::Wrapper));
+
+    uintptr_t FunctionAddress = reinterpret_cast<uintptr_t>(Function);
+
+    hlt(vixl::aarch64::kRuntimeCallOpcode);
+
+    // Simulator wrapper address pointer.
+    dc64(SimulatorWrapperAddress);
+
+    // Runtime function address to call
+    dc64(FunctionAddress);
+
+    // Call type
+    dc32(vixl::aarch64::kCallRuntime);
+  }
+
+  template<typename R, typename... P>
+  void GenerateIndirectRuntimeCall(ARMEmitter::Register Reg) {
+    uintptr_t SimulatorWrapperAddress = reinterpret_cast<uintptr_t>(
+      &(vixl::aarch64::Simulator::RuntimeCallStructHelper<R, P...>::Wrapper));
+
+    hlt(vixl::aarch64::kIndirectRuntimeCallOpcode);
+
+    // Simulator wrapper address pointer.
+    dc64(SimulatorWrapperAddress);
+
+    // Register that contains the function to call
+    dc32(Reg.Idx());
+
+    // Call type
+    dc32(vixl::aarch64::kCallRuntime);
+  }
+
+  template<>
+  void GenerateIndirectRuntimeCall<float, __uint128_t>(ARMEmitter::Register Reg) {
+    uintptr_t SimulatorWrapperAddress = reinterpret_cast<uintptr_t>(
+      &(vixl::aarch64::Simulator::RuntimeCallStructHelper<float, __uint128_t>::Wrapper));
+
+    hlt(vixl::aarch64::kIndirectRuntimeCallOpcode);
+
+    // Simulator wrapper address pointer.
+    dc64(SimulatorWrapperAddress);
+
+    // Register that contains the function to call
+    dc32(Reg.Idx());
+
+    // Call type
+    dc32(vixl::aarch64::kCallRuntime);
+  }
+
+#endif
+#ifdef VIXL_DISASSEMBLER
+  vixl::aarch64::PrintDisassembler Disasm {stderr};
+  FEX_CONFIG_OPT(Disassemble, DISASSEMBLE);
+#endif
  FEX_CONFIG_OPT(StaticRegisterAllocation, SRA);
 };

@@ -0,0 +1,322 @@
+/* Branch instruction emitters.
+ *
+ * Most of these instructions will use `BackwardLabel`, `ForwardLabel`, or `BiDirectionLabel` to determine where a branch targets.
+ */
+public:
+  // Branches, Exception Generating and System instructions
+  public:
+    // Conditional branch immediate
+    ///< Branch conditional
+    void b(FEXCore::ARMEmitter::Condition Cond, uint32_t Imm) {
+      constexpr uint32_t Op = 0b0101'010 << 25;
+      Branch_Conditional(Op, 0, 0, Cond, Imm);
+    }
+    void b(FEXCore::ARMEmitter::Condition Cond, BackwardLabel const* Label) {
+      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+      LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+      constexpr uint32_t Op = 0b0101'010 << 25;
+      Branch_Conditional(Op, 0, 0, Cond, Imm >> 2);
+    }
+    void b(FEXCore::ARMEmitter::Condition Cond, ForwardLabel *Label) {
+      Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::BC });
+      constexpr uint32_t Op = 0b0101'010 << 25;
+      Branch_Conditional(Op, 0, 0, Cond, 0);
+    }
+
+    void b(FEXCore::ARMEmitter::Condition Cond, BiDirectionalLabel *Label) {
+      if (Label->Backward.Location) {
+        b(Cond, &Label->Backward);
+      }
+      else {
+        b(Cond, &Label->Forward);
+      }
+    }
+
+    ///< Branch consistent conditional
+    void bc(FEXCore::ARMEmitter::Condition Cond, uint32_t Imm) {
+      constexpr uint32_t Op = 0b0101'010 << 25;
+      Branch_Conditional(Op, 0, 1, Cond, Imm);
+    }
+    void bc(FEXCore::ARMEmitter::Condition Cond, BackwardLabel const* Label) {
+      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+      LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+      constexpr uint32_t Op = 0b0101'010 << 25;
+      Branch_Conditional(Op, 0, 1, Cond, Imm >> 2);
+    }
+
+    void bc(FEXCore::ARMEmitter::Condition Cond, ForwardLabel *Label) {
+      Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::BC });
+      constexpr uint32_t Op = 0b0101'010 << 25;
+      Branch_Conditional(Op, 0, 1, Cond, 0);
+    }
+
+    void bc(FEXCore::ARMEmitter::Condition Cond, BiDirectionalLabel *Label) {
+      if (Label->Backward.Location) {
+        bc(Cond, &Label->Backward);
+      }
+      else {
+        bc(Cond, &Label->Forward);
+      }
+    }
+
+    // Unconditional branch register
+    void br(FEXCore::ARMEmitter::Register rn) {
+      constexpr uint32_t Op = 0b1101011 << 25 |
+                              0b0'000 << 21 |   // opc
+                              0b1'1111 << 16 |  // op2
+                              0b0000'00 << 10 | // op3
+                              0b0'0000;         // op4
+
+      UnconditionalBranch(Op, rn);
+    }
+    void blr(FEXCore::ARMEmitter::Register rn) {
+      constexpr uint32_t Op = 0b1101011 << 25 |
+                              0b0'001 << 21 |   // opc
+                              0b1'1111 << 16 |  // op2
+                              0b0000'00 << 10 | // op3
+                              0b0'0000;         // op4
+
+      UnconditionalBranch(Op, rn);
+    }
+    void ret(FEXCore::ARMEmitter::Register rn = FEXCore::ARMEmitter::Reg::r30) {
+      constexpr uint32_t Op = 0b1101011 << 25 |
+                              0b0'010 << 21 |   // opc
+                              0b1'1111 << 16 |  // op2
+                              0b0000'00 << 10 | // op3
+                              0b0'0000;         // op4
+
+      UnconditionalBranch(Op, rn);
+    }
+
+    // Unconditional branch immediate
+    void b(uint32_t Imm) {
+      constexpr uint32_t Op = 0b0001'01 << 26;
+
+      UnconditionalBranch(Op, Imm);
+    }
+    void b(BackwardLabel const* Label) {
+      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+      LOGMAN_THROW_A_FMT(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+      constexpr uint32_t Op = 0b0001'01 << 26;
+
+      UnconditionalBranch(Op, Imm >> 2);
+    }
+    void b(ForwardLabel *Label) {
+      Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::B });
+      constexpr uint32_t Op = 0b0001'01 << 26;
+
+      UnconditionalBranch(Op, 0);
+    }
+
+    void b(BiDirectionalLabel *Label) {
+      if (Label->Backward.Location) {
+        b(&Label->Backward);
+      }
+      else {
+        b(&Label->Forward);
+      }
+    }
+
+    void bl(uint32_t Imm) {
+      constexpr uint32_t Op = 0b1001'01 << 26;
+
+      UnconditionalBranch(Op, Imm);
+    }
+
+    void bl(BackwardLabel const* Label) {
+      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+      LOGMAN_THROW_A_FMT(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+      constexpr uint32_t Op = 0b1001'01 << 26;
+
+      UnconditionalBranch(Op, Imm >> 2);
+    }
+    void bl(ForwardLabel *Label) {
+      Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::B });
+      constexpr uint32_t Op = 0b1001'01 << 26;
+
+      UnconditionalBranch(Op, 0);
+    }
+
+    void bl(BiDirectionalLabel *Label) {
+      if (Label->Backward.Location) {
+        bl(&Label->Backward);
+      }
+      else {
+        bl(&Label->Forward);
+      }
+    }
+
+    // Compare and branch
+    void cbz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, uint32_t Imm) {
+      constexpr uint32_t Op = 0b0011'0100 << 24;
+
+      CompareAndBranch(Op, s, rt, Imm);
+    }
+
+    void cbz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, BackwardLabel const* Label) {
+      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+      LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+
+      constexpr uint32_t Op = 0b0011'0100 << 24;
+
+      CompareAndBranch(Op, s, rt, Imm >> 2);
+    }
+
+    void cbz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, ForwardLabel *Label) {
+      Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::BC });
+
+      constexpr uint32_t Op = 0b0011'0100 << 24;
+
+      CompareAndBranch(Op, s, rt, 0);
+    }
+
+    void cbz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, BiDirectionalLabel *Label) {
+      if (Label->Backward.Location) {
+        cbz(s, rt, &Label->Backward);
+      }
+      else {
+        cbz(s, rt, &Label->Forward);
+      }
+    }
+
+    void cbnz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, uint32_t Imm) {
+      constexpr uint32_t Op = 0b0011'0101 << 24;
+
+      CompareAndBranch(Op, s, rt, Imm);
+    }
+
+    void cbnz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, BackwardLabel const* Label) {
+      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+      LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+
+      constexpr uint32_t Op = 0b0011'0101 << 24;
+
+      CompareAndBranch(Op, s, rt, Imm >> 2);
+    }
+
+    void cbnz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, ForwardLabel *Label) {
+      Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::BC });
+
+      constexpr uint32_t Op = 0b0011'0101 << 24;
+
+      CompareAndBranch(Op, s, rt, 0);
+    }
+
+    void cbnz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, BiDirectionalLabel *Label) {
+      if (Label->Backward.Location) {
+        cbnz(s, rt, &Label->Backward);
+      }
+      else {
+        cbnz(s, rt, &Label->Forward);
+      }
+    }
+
+    // Test and branch immediate
+    void tbz(FEXCore::ARMEmitter::Register rt, uint32_t Bit, uint32_t Imm) {
+      constexpr uint32_t Op = 0b0011'0110 << 24;
+
+      TestAndBranch(Op, rt, Bit, Imm);
+    }
+    void tbz(FEXCore::ARMEmitter::Register rt, uint32_t Bit, BackwardLabel const* Label) {
+      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+      LOGMAN_THROW_A_FMT(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+
+      constexpr uint32_t Op = 0b0011'0110 << 24;
+
+      TestAndBranch(Op, rt, Bit, Imm >> 2);
+    }
+    void tbz(FEXCore::ARMEmitter::Register rt, uint32_t Bit, ForwardLabel *Label) {
+      Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::TEST_BRANCH });
+
+      constexpr uint32_t Op = 0b0011'0110 << 24;
+
+      TestAndBranch(Op, rt, Bit, 0);
+    }
+
+    void tbz(FEXCore::ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel *Label) {
+      if (Label->Backward.Location) {
+        tbz(rt, Bit, &Label->Backward);
+      }
+      else {
+        tbz(rt, Bit, &Label->Forward);
+      }
+    }
+
+    void tbnz(FEXCore::ARMEmitter::Register rt, uint32_t Bit, uint32_t Imm) {
+      constexpr uint32_t Op = 0b0011'0111 << 24;
+
+      TestAndBranch(Op, rt, Bit, Imm);
+    }
+    void tbnz(FEXCore::ARMEmitter::Register rt, uint32_t Bit, BackwardLabel const* Label) {
+      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+      LOGMAN_THROW_A_FMT(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+
+      constexpr uint32_t Op = 0b0011'0111 << 24;
+
+      TestAndBranch(Op, rt, Bit, Imm >> 2);
+    }
+    void tbnz(FEXCore::ARMEmitter::Register rt, uint32_t Bit, ForwardLabel *Label) {
+      Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::TEST_BRANCH });
+      constexpr uint32_t Op = 0b0011'0111 << 24;
+
+      TestAndBranch(Op, rt, Bit, 0);
+    }
+
+    void tbnz(FEXCore::ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel *Label) {
+      if (Label->Backward.Location) {
+        tbnz(rt, Bit, &Label->Backward);
+      }
+      else {
+        tbnz(rt, Bit, &Label->Forward);
+      }
+    }
+
+private:
+    // Conditional branch immediate
+    void Branch_Conditional(uint32_t Op, uint32_t Op1, uint32_t Op0, FEXCore::ARMEmitter::Condition Cond, uint32_t Imm) {
+      uint32_t Instr = Op;
+
+      Instr |= Op1 << 24;
+      Instr |= (Imm & 0x7'FFFF) << 5;
+      Instr |= Op0 << 4;
+      Instr |= FEXCore::ToUnderlying(Cond);
+
+      dc32(Instr);
+    }
+
+    // Unconditional branch register
+    void UnconditionalBranch(uint32_t Op, FEXCore::ARMEmitter::Register rn) {
+      uint32_t Instr = Op;
+      Instr |= Encode_rn(rn);
+      dc32(Instr);
+    }
+
+    // Unconditional branch - immediate
+    void UnconditionalBranch(uint32_t Op, uint32_t Imm) {
+      uint32_t Instr = Op;
+      Instr |= Imm & 0x3FF'FFFF;
+      dc32(Instr);
+    }
+
+    // Compare and branch
+    void CompareAndBranch(uint32_t Op, FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, uint32_t Imm) {
+      const uint32_t SF = s == FEXCore::ARMEmitter::Size::i64Bit ? (1U << 31) : 0;
+
+      uint32_t Instr = Op;
+
+      Instr |= SF;
+      Instr |= (Imm & 0x7'FFFF) << 5;
+      Instr |= Encode_rt(rt);
+      dc32(Instr);
+    }
+
+    // Test and branch - immediate
+    void TestAndBranch(uint32_t Op, FEXCore::ARMEmitter::Register rt, uint32_t Bit, uint32_t Imm) {
+      uint32_t Instr = Op;
+
+      Instr |= (Bit >> 5) << 31;
+      Instr |= (Bit & 0b1'1111) << 19;
+      Instr |= (Imm & 0x3FFF) << 5;
+      Instr |= Encode_rt(rt);
+      dc32(Instr);
+    }
@@ -0,0 +1,105 @@
+#pragma once
+#include <cstddef>
+#include <cstdint>
+#include <cstring>
+
+namespace FEXCore::ARMEmitter {
+  class Buffer {
+    public:
+      Buffer() {
+        SetBuffer(nullptr, 0);
+      }
+
+      Buffer(uint8_t* Base, uint64_t BaseSize) {
+        SetBuffer(Base, BaseSize);
+      }
+
+      void SetBuffer(uint8_t* Base, uint64_t BaseSize) {
+        BufferBase = Base;
+        CurrentOffset = BufferBase;
+        Size = BaseSize;
+      }
+
+      void dc8(uint8_t Data) {
+        decltype(Data) *Memory = reinterpret_cast<decltype(Data)*>(CurrentOffset);
+        *Memory = Data;
+        CurrentOffset += sizeof(Data);
+      }
+
+      void dc16(uint16_t Data) {
+        decltype(Data) *Memory = reinterpret_cast<decltype(Data)*>(CurrentOffset);
+        *Memory = Data;
+        CurrentOffset += sizeof(Data);
+      }
+
+      void dc32(uint32_t Data) {
+        decltype(Data) *Memory = reinterpret_cast<decltype(Data)*>(CurrentOffset);
+        *Memory = Data;
+        CurrentOffset += sizeof(Data);
+      }
+
+      void dc64(uint64_t Data) {
+        decltype(Data) *Memory = reinterpret_cast<decltype(Data)*>(CurrentOffset);
+        *Memory = Data;
+        CurrentOffset += sizeof(Data);
+      }
+      void EmitString(const char *String) {
+        const auto StringLength = strlen(String);
+        memcpy(CurrentOffset, String, StringLength);
+        CurrentOffset += StringLength;
+      }
+
+      void Align() {
+        // Align the buffer to instruction size
+        auto CurrentAlignment = reinterpret_cast<uint64_t>(CurrentOffset) & 0b11;
+        if (!CurrentAlignment) {
+          return;
+        }
+        CurrentOffset += 4 - CurrentAlignment;
+      }
+
+      template<typename T>
+      T GetCursorAddress() const {
+        return reinterpret_cast<T>(CurrentOffset);
+      }
+
+      static void ClearICache(void* Begin, std::size_t Length) {
+        __builtin___clear_cache(static_cast<char*>(Begin), static_cast<char*>(Begin) + Length);
+      }
+
+      size_t GetCursorOffset() const {
+        return static_cast<size_t>(CurrentOffset - BufferBase);
+      }
+
+      uint8_t *GetBufferBase() const {
+        return BufferBase;
+      }
+
+      void CursorIncrement(size_t Size) {
+        CurrentOffset += Size;
+      }
+
+      void SetCursorOffset(size_t Offset) {
+        CurrentOffset = BufferBase + Offset;
+      }
+
+      uint64_t GetBufferSize() const {
+        return Size;
+      }
+
+      template<typename T>
+      size_t GetCursorOffsetFromAddress(const T* Address) const {
+        return static_cast<size_t>(reinterpret_cast<const uint8_t*>(Address) - BufferBase);
+      }
+
+    protected:
+
+      void ResetBuffer() {
+        CurrentOffset = BufferBase;
+      }
+
+      uint8_t* BufferBase;
+      uint8_t* CurrentOffset;
+      uint64_t Size;
+  };
+}
@@ -0,0 +1,831 @@
+#pragma once
+
+#include "Interface/Core/ArchHelpers/CodeEmitter/Buffer.h"
+#include "Interface/Core/ArchHelpers/CodeEmitter/Registers.h"
+
+#include <FEXCore/Utils/CompilerDefs.h>
+#include <FEXCore/Utils/EnumUtils.h>
+#include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/MathUtils.h>
+#include <FEXCore/fextl/vector.h>
+
+#include <aarch64/assembler-aarch64.h>
+
+#include <cstdint>
+#include <utility>
+#include <type_traits>
+
+/*
+ * Welcome to FEX-Emu's custom AArch64 emitter.
+ * This was written specifically to avoid the performance cost of the vixl emitter.
+ *
+ * There are some specific design constraints in this design to target a couple features:
+ *   - High performance
+ *   - Low CPU cache performance hit
+ *   - Significantly reduced code footprint
+ *   - Low number of branches
+ *
+ * These requirements are mostly achieved by removing a bunch of developer conveniences
+ * that vixl provides. The developer needs to take a lot of care to not shoot themselves in the foot.
+ *
+ * Misc design decisions:
+ * - Registers are encoded as basic uint32_t enums.
+ *   - Converting between different registers is zero-cost.
+ *   - Passing around as arguments are as cheap as registers
+ *     - Contrast to vixl where every register requires living on the stack.
+ *   - Registers can get encoded in to instructions with a simple `BFM` instruction.
+ *
+ * - Instructions are very simply emitted, allowing direct inlining most of the time.
+ *   - These are simple enough that multiple back-to-back instructions get optimized to 128-bit load-store operations.
+ *     - Contrast to vixl where pretty much no instruction emitter gets inlined.
+ *
+ * - Instruction emitters are /mostly/ unsized. Most instructions take a size argument first, which gets encoded
+ *   directly in to the instruction.
+ *   - Contrast to vixl where the register arguments are how the instructions determine operating size.
+ *   - Size argument allows FEX to use `CSEL` to select a size at runtime, instead of branching.
+ *   - Some instructions are explicitly sized based on register type. Read comments in the respective `inl` files to
+ *     see why.
+ *     Some scalar/vector operations are an example of this.
+ *
+ * - Almost zero helper functions.
+ *   - Primary exception to this rule is load-store operations. These will use a helper to make
+ *     it easier to select the correct load-store instruction. Mostly because these are a nightmare selecting
+ *     the right instruction.
+ */
+namespace FEXCore::ARMEmitter {
+  /*
+   * This `Size` enum is used for most ALU operations.
+   * These follow the AArch64 encoding style in most cases.
+   */
+  enum class Size : uint32_t {
+    i32Bit = 0,
+    i64Bit,
+  };
+
+  // This allows us to get the `Size` enum in bits.
+  [[nodiscard]]
+  constexpr size_t RegSizeInBits(Size size) {
+    return size_t{32} << FEXCore::ToUnderlying(size);
+  }
+
+  /* This `SubRegSize` enum is used for most ASIMD operations.
+   * These follow the AArch64 encoding style in most cases.
+   */
+  enum class SubRegSize : uint32_t {
+    i8Bit   = 0b00,
+    i16Bit  = 0b01,
+    i32Bit  = 0b10,
+    i64Bit  = 0b11,
+    i128Bit = 0b100,
+  };
+
+  // This allows us to get the `SubRegSize` in bits.
+  [[nodiscard]]
+  constexpr size_t SubRegSizeInBits(SubRegSize size) {
+    return size_t{8} << FEXCore::ToUnderlying(size);
+  }
+
+  /* This `ScalarRegSize` enum is used for most scalar float
+   * operations.
+   *
+   * This is specifically duplicated from `SubRegSize` to have strongly
+   * typed functions.
+   *
+   * `ScalarRegSize` specifically doesn't have `i128Bit` because scalar operations
+   * can't operate at 128-bit.
+   */
+  enum class ScalarRegSize : uint32_t {
+    i8Bit   = 0b00,
+    i16Bit  = 0b01,
+    i32Bit  = 0b10,
+    i64Bit  = 0b11,
+  };
+
+  // This allows us to get the `ScalarRegSize` in bits.
+  [[nodiscard]]
+  constexpr size_t ScalarRegSizeInBits(ScalarRegSize size) {
+    return size_t{8} << FEXCore::ToUnderlying(size);
+  }
+
+  /* This `VectorRegSizePair` union allows us to have an overlapping type
+   * to select a scalar operation or a vector depending on which operation
+   * we pass in.
+   * Useful in FEX's vector operations that behave as scalar or vector
+   * depending on various factors. But since the operation will have the sa,e
+   * element size, we want to choose the operation more easily
+   */
+  union VectorRegSizePair {
+    ScalarRegSize Scalar;
+    SubRegSize Vector;
+  };
+
+  // This allows us to create a `VectorRegSizePair` union.
+  [[nodiscard]]
+  constexpr VectorRegSizePair ToVectorSizePair(SubRegSize size) {
+    return VectorRegSizePair {.Vector = size};
+  }
+  [[nodiscard]]
+  constexpr VectorRegSizePair ToVectorSizePair(ScalarRegSize size) {
+    return VectorRegSizePair {.Scalar = size};
+  }
+
+  // This `ShiftType` enum is used for ALU shift-register encoded instructions.
+  enum class ShiftType : uint32_t {
+    LSL = 0,
+    LSR,
+    ASR,
+    ROR,
+  };
+
+  // This `ExtendedType` enum is used for ALU extended-register encoded instructions.
+  enum class ExtendedType : uint32_t {
+    UXTB = 0b000,
+    UXTH = 0b001,
+    UXTW = 0b010,
+    UXTX = 0b011,
+    SXTB = 0b100,
+    SXTH = 0b101,
+    SXTW = 0b110,
+    SXTX = 0b111,
+    LSL_32 = UXTW,
+    LSL_64 = UXTX,
+  };
+
+  // This `Condition` enum is used for various conditional instructions.
+  enum class Condition : uint32_t {
+    // Meaning:   Int                    - Float
+    CC_EQ = 0, // Equal                  - Equal
+    CC_NE,     // Not Eq                 - Not Eq or unordered
+    CC_CS,     // Carry set              - Greater than, equal, or unordered
+    CC_CC,     // Carry clear            - Less than
+    CC_MI,     // Minus/Negative         - Less than
+    CC_PL,     // Plus, positive or zero - GT, equal, or unordered
+    CC_VS,     // Overflow               - Unordered
+    CC_VC,     // No Overflow            - Ordered
+    CC_HI,     // Unsigned higher        - GT, or unordered
+    CC_LS,     // Unsigned lower or same - LT or EQ
+    CC_GE,     // Signed GT or EQ        - GT or EQ
+    CC_LT,     // Signed LT              - LT or Unordered
+    CC_GT,     // Signed GT              - GT
+    CC_LE,     // Signed LT or EQ        - LT, EQ, or Unordered
+    CC_AL,     // Always                 - Always
+    CC_NV,     // Always                 - Always
+
+    // Aliases
+    CC_HS = CC_CS,
+    CC_LO = CC_CC,
+  };
+
+  /*
+   * This `StatusFlags` enum is used for conditional compare encoded instructions.
+   * These directly encode to the `nzcv` flags.
+   */
+  enum class StatusFlags : uint32_t {
+    None = 0,
+    Flag_V = 0b0001,
+    Flag_C = 0b0010,
+    Flag_Z = 0b0100,
+    Flag_N = 0b1000,
+
+    Flag_NZCV = Flag_N | Flag_Z | Flag_C | Flag_V,
+  };
+
+
+  /*
+   * This `IndexType` enum is used for load-store instructions.
+   * Not all load-store instructions use this, so the user needs to be careful.
+   */
+  enum class IndexType {
+    POST,
+    OFFSET,
+    PRE,
+
+    UNPRIVILEGED,
+  };
+
+  /* This `SVEMemOperand` class is used for the helper SVE load-store instructions.
+   * Load-store instructions are quite expressive, so having a helper that handles these differences is worth it.
+   */
+  class SVEMemOperand final {
+    public:
+      // Used for scalar + vector variants to determine
+      // extension behavior on the index values.
+      enum class ModType : uint8_t {
+        MOD_UXTW,
+        MOD_SXTW,
+        MOD_LSL,
+        MOD_NONE,
+      };
+
+      enum class Type {
+        ScalarPlusScalar,
+        ScalarPlusImm,
+        ScalarPlusVector,
+        VectorPlusImm,
+      };
+
+      SVEMemOperand(XRegister rn, XRegister rm = XReg::zr)
+        : rn {rn}
+        , MemType{Type::ScalarPlusScalar}
+        , MetaType {
+          .ScalarScalarType {
+            .rm = rm,
+          }
+        } {}
+      SVEMemOperand(XRegister rn, int32_t imm = 0)
+        : rn {rn}
+        , MemType{Type::ScalarPlusImm}
+        , MetaType {
+          .ScalarImmType {
+            .Imm = imm,
+          }
+        } {}
+      SVEMemOperand(XRegister rn, ZRegister zm, ModType mod = ModType::MOD_NONE, uint8_t scale = 0)
+        : rn{rn}
+        , MemType{Type::ScalarPlusVector}
+        , MetaType {
+          .ScalarVectorType {
+            .zm = zm,
+            .mod = mod,
+            .scale = scale,
+          }
+        } {}
+      SVEMemOperand(ZRegister zn, uint32_t imm)
+        : rn{Register{zn.Idx()}}
+        , MemType{Type::VectorPlusImm}
+        , MetaType {
+          .VectorImmType{
+            .Imm = imm,
+          }
+        } {}
+
+      [[nodiscard]] bool IsScalarPlusScalar() const {
+        return MemType == Type::ScalarPlusScalar;
+      }
+      [[nodiscard]] bool IsScalarPlusImm() const {
+        return MemType == Type::ScalarPlusImm;
+      }
+      [[nodiscard]] bool IsScalarPlusVector() const {
+        return MemType == Type::ScalarPlusVector;
+      }
+      [[nodiscard]] bool IsVectorPlusImm() const {
+        return MemType == Type::VectorPlusImm;
+      }
+
+      union Data {
+        struct {
+          Register rm;
+        } ScalarScalarType;
+
+        struct {
+          int32_t Imm;
+        } ScalarImmType;
+
+        struct {
+          ZRegister zm;
+          ModType mod;
+          uint8_t scale;
+        } ScalarVectorType;
+
+        struct {
+          // rn will be a ZRegister
+          uint32_t Imm;
+        } VectorImmType;
+      };
+
+      Register rn;
+      Type MemType;
+      Data MetaType;
+  };
+
+  /* This `ExtendedMemOperand` class is used for the helper load-store instructions.
+   * Load-store instructions are quite expressive, so having a helper that handles these differences is worth it.
+   */
+  class ExtendedMemOperand final {
+    public:
+      ExtendedMemOperand(XRegister rn, XRegister rm = XReg::zr, ExtendedType Option = ExtendedType::LSL_64, uint32_t Shift = 0)
+        : rn {rn}
+        , MetaType {
+          .ExtendedType {
+            .Header = { .MemType = TYPE_EXTENDED },
+            .rm = rm,
+            .Option = Option,
+            .Shift = Shift,
+          }
+        } {}
+      ExtendedMemOperand(XRegister rn, IndexType Index = IndexType::OFFSET, int32_t Imm = 0)
+        : rn {rn}
+        , MetaType {
+          .ImmType {
+            .Header = { .MemType = TYPE_IMM },
+            .Index = Index,
+            .Imm = Imm,
+          }
+        } {}
+
+      Register rn;
+      enum Type {
+        TYPE_EXTENDED,
+        TYPE_IMM,
+      };
+      struct HeaderStruct {
+        Type MemType;
+      };
+      union {
+        HeaderStruct Header;
+        struct {
+          HeaderStruct Header;
+          Register rm;
+          ExtendedType Option;
+          uint32_t Shift;
+        } ExtendedType;
+        struct {
+          HeaderStruct Header;
+          IndexType Index;
+          int32_t Imm;
+        } ImmType;
+      } MetaType;
+  };
+
+  template<uint32_t op0, uint32_t op1, uint32_t CRn, uint32_t CRm, uint32_t op2>
+  constexpr uint32_t GenSystemReg() {
+    return op0 << 19 |
+      op1 << 16 |
+      CRn << 12 |
+      CRm << 8 |
+      op2 << 5;
+  };
+
+  // This `SystemRegister` enum is used for the mrs/msr instructions.
+  enum class SystemRegister : uint32_t {
+    CTR_EL0    = GenSystemReg<0b11, 0b011, 0b0000, 0b0000, 0b001>(),
+    DCZID_EL0  = GenSystemReg<0b11, 0b011, 0b0000, 0b0000, 0b111>(),
+    TPIDR_EL0  = GenSystemReg<0b11, 0b011, 0b1101, 0b0000, 0b010>(),
+    RNDR       = GenSystemReg<0b11, 0b011, 0b0010, 0b0100, 0b000>(),
+    RNDRRS     = GenSystemReg<0b11, 0b011, 0b0010, 0b0100, 0b001>(),
+    NZCV       = GenSystemReg<0b11, 0b011, 0b0100, 0b0010, 0b000>(),
+    FPCR       = GenSystemReg<0b11, 0b011, 0b0100, 0b0100, 0b000>(),
+    CNTFRQ_EL0 = GenSystemReg<0b11, 0b011, 0b1110, 0b0000, 0b000>(),
+    CNTVCT_EL0 = GenSystemReg<0b11, 0b011, 0b1110, 0b0000, 0b010>(),
+  };
+
+  template<uint32_t op1, uint32_t CRm, uint32_t op2>
+  constexpr uint32_t GenDCReg() {
+    return op1 << 16 |
+      CRm << 8 |
+      op2 << 5;
+  };
+
+  // This `DataCacheOperation` enum is used for the dc instruction.
+  enum class DataCacheOperation : uint32_t {
+    IVAC  = GenDCReg<0b000, 0b0110, 0b001>(),
+    ISW   = GenDCReg<0b000, 0b0110, 0b010>(),
+    CSW   = GenDCReg<0b000, 0b1010, 0b010>(),
+    CISW  = GenDCReg<0b000, 0b1110, 0b010>(),
+    ZVA   = GenDCReg<0b011, 0b0100, 0b001>(),
+    CVAC  = GenDCReg<0b011, 0b1010, 0b001>(),
+    CVAU  = GenDCReg<0b011, 0b1011, 0b001>(),
+    CIVAC = GenDCReg<0b011, 0b1110, 0b001>(),
+
+    // MTE2
+    IGVAC  = GenDCReg<0b000, 0b0110, 0b011>(),
+    IGSW   = GenDCReg<0b000, 0b0110, 0b100>(),
+    IGDVAC = GenDCReg<0b000, 0b0110, 0b101>(),
+    IGDSW  = GenDCReg<0b000, 0b0110, 0b110>(),
+    CGSW   = GenDCReg<0b000, 0b1010, 0b100>(),
+    CGDSW  = GenDCReg<0b000, 0b1010, 0b110>(),
+    CIGSW  = GenDCReg<0b000, 0b1110, 0b100>(),
+    CIGDSW = GenDCReg<0b000, 0b1110, 0b110>(),
+
+    // MTE
+    GVA      = GenDCReg<0b011, 0b0100, 0b011>(),
+    GZVA     = GenDCReg<0b011, 0b0100, 0b100>(),
+    CGVAC    = GenDCReg<0b011, 0b1010, 0b011>(),
+    CGDVAC   = GenDCReg<0b011, 0b1010, 0b101>(),
+    CGVAP    = GenDCReg<0b011, 0b1100, 0b011>(),
+    CGDVAP   = GenDCReg<0b011, 0b1100, 0b101>(),
+    CGVADP   = GenDCReg<0b011, 0b1101, 0b011>(),
+    CGDVADP  = GenDCReg<0b011, 0b1101, 0b101>(),
+    CIGVAC   = GenDCReg<0b011, 0b1110, 0b011>(),
+    CIGDVAC  = GenDCReg<0b011, 0b1110, 0b101>(),
+
+    // DPB
+    CVAP = GenDCReg<0b011, 0b1100, 0b001>(),
+
+    // DPB2
+    CVADP = GenDCReg<0b011, 0b1101, 0b001>(),
+  };
+
+  template<uint32_t CRm, uint32_t op2>
+  constexpr uint32_t GenHintBarrierReg() {
+    return CRm << 8 |
+      op2 << 5;
+  }
+
+  // This `HintRegister` enum is used for the hint instruction.
+  enum class HintRegister : uint32_t {
+    NOP   = GenHintBarrierReg<0b0000, 0b000>(),
+    YIELD = GenHintBarrierReg<0b0000, 0b001>(),
+    WFE   = GenHintBarrierReg<0b0000, 0b010>(),
+    WFI   = GenHintBarrierReg<0b0000, 0b011>(),
+    SEV   = GenHintBarrierReg<0b0000, 0b100>(),
+    SEVL  = GenHintBarrierReg<0b0000, 0b101>(),
+    DGH   = GenHintBarrierReg<0b0000, 0b110>(),
+    CSDB  = GenHintBarrierReg<0b0010, 0b100>(),
+  };
+
+  // This `BarrierRegister` enum is used for the various barrier instructions.
+  enum class BarrierRegister : uint32_t {
+    CLREX   = GenHintBarrierReg<0b0000, 0b010>(),
+    TCOMMIT = GenHintBarrierReg<0b0000, 0b011>(),
+    DSB     = GenHintBarrierReg<0b0000, 0b100>(),
+    DMB     = GenHintBarrierReg<0b0000, 0b101>(),
+    ISB     = GenHintBarrierReg<0b0000, 0b110>(),
+    SB      = GenHintBarrierReg<0b0000, 0b111>(),
+  };
+
+  // This `BarrierScope` enum is used for the dsb/dmb instructions.
+  enum class BarrierScope : uint32_t {
+    // Outer shareable
+    OSHLD  = 0b0001,
+    OSHST  = 0b0010,
+    OSH    = 0b0011,
+    // Non shareable
+    NSHLD  = 0b0101,
+    NSHST  = 0b0110,
+    NSH    = 0b0111,
+    // Inner shareable
+    ISHLD  = 0b1001,
+    ISHST  = 0b1010,
+    ISH    = 0b1011,
+    // Full System visibility
+    LD     = 0b1101,
+    ST     = 0b1110,
+    SY     = 0b1111,
+  };
+
+  // This `Prefetch` enum is used for prefetch instructions.
+  enum class Prefetch : uint32_t {
+    // Prefetch for load
+    PLDL1KEEP = 0b00000,
+    PLDL1STRM = 0b00001,
+    PLDL2KEEP = 0b00010,
+    PLDL2STRM = 0b00011,
+    PLDL3KEEP = 0b00100,
+    PLDL3STRM = 0b00101,
+
+    // Preload instructions
+    PLIL1KEEP = 0b01000,
+    PLIL1STRM = 0b01001,
+    PLIL2KEEP = 0b01010,
+    PLIL2STRM = 0b01011,
+    PLIL3KEEP = 0b01100,
+    PLIL3STRM = 0b01101,
+
+    // Preload for store
+    PSTL1KEEP = 0b10000,
+    PSTL1STRM = 0b10001,
+    PSTL2KEEP = 0b10010,
+    PSTL2STRM = 0b10011,
+    PSTL3KEEP = 0b10100,
+    PSTL3STRM = 0b10101,
+  };
+
+  // This `PredicatePattern` enun is used for some SVE instructions.
+  enum class PredicatePattern : uint32_t {
+    SVE_POW2  = 0b00000,
+    SVE_VL1   = 0b00001,
+    SVE_VL2   = 0b00010,
+    SVE_VL3   = 0b00011,
+    SVE_VL4   = 0b00100,
+    SVE_VL5   = 0b00101,
+    SVE_VL6   = 0b00110,
+    SVE_VL7   = 0b00111,
+    SVE_VL8   = 0b01000,
+    SVE_VL16  = 0b01001,
+    SVE_VL32  = 0b01010,
+    SVE_VL64  = 0b01011,
+    SVE_VL128 = 0b01100,
+    SVE_VL256 = 0b01101,
+    SVE_MUL4  = 0b11101,
+    SVE_MUL3  = 0b11110,
+    SVE_ALL   = 0b11111,
+  };
+
+  // Used with SVE FP immediate arithmetic instructions
+  enum class SVEFAddSubImm : uint32_t {
+    _0_5,
+    _1_0,
+  };
+  enum class SVEFMulImm : uint32_t {
+    _0_5,
+    _2_0,
+  };
+  enum class SVEFMaxMinImm : uint32_t {
+    _0_0,
+    _1_0,
+  };
+
+  /* This `BackwardLabel` struct used for retaining a location for PC-Relative instructions.
+   * This is specifically a label for a target that is logically `below` an instruction that uses it.
+   * Which means that a branch would jump backwards.
+   */
+  struct BackwardLabel {
+    uint8_t *Location{};
+  };
+
+  /* This `ForwardLabel` struct used for retaining a location for PC-Relative instructions.
+   * This is specifically a label for a target that is logically `above` an instruction that uses it.
+   * Which means that a branch would jump forwards.
+   *
+   * This can be bound to multiple instructions, so it needs a vector for each bind instruction type.
+   */
+  struct ForwardLabel {
+    struct Instructions {
+      enum class InstType {
+        ADR,
+        ADRP,
+        B,
+        BC,
+        TEST_BRANCH,
+        RELATIVE_LOAD,
+        LONG_ADDRESS_GEN,
+      };
+      uint8_t *Location{};
+      InstType Type;
+    };
+    fextl::vector<Instructions> Insts{};
+  };
+
+  /* This `BiDirectionalLabel` struct used for retaining a location for PC-Relative instructions.
+   * This is specifically a label for a target that is in either direction of an instruction that uses it.
+   * Which means a branch could jump backwards or forwards depending on situation.
+   */
+  struct BiDirectionalLabel {
+    BackwardLabel Backward;
+    ForwardLabel Forward;
+  };
+
+  // Some FCMA ASIMD instructions support a rotation argument.
+  enum class Rotation : uint32_t {
+    ROTATE_0   = 0b00,
+    ROTATE_90  = 0b01,
+    ROTATE_180 = 0b10,
+    ROTATE_270 = 0b11,
+  };
+
+  // Concept for contraining some instructions to accept only an XRegister or WRegister.
+  // Particularly for operations that differ encodings depending on which one is used.
+  template <typename T>
+  concept IsXOrWRegister = std::is_same_v<T, XRegister> || std::is_same_v<T, WRegister>;
+
+  // 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)
+  //
+  // For example, a set of registers like:
+  //
+  // v1,  v2, v3 and
+  // v31, v0, v1
+  //
+  // would both be considered sequential sequences, and some instructions in particular
+  // limit register lists to these kind of sequences.
+  //
+  template <typename T, typename... Args>
+  constexpr bool AreVectorsSequential(T first, const Args&... args) {
+    // Ensure we always have a pair of registers to compare against.
+    static_assert(sizeof...(args) >= 1, "Number of arguments must be greater than 1");
+
+    const auto fn = [](auto& lhs, const auto& rhs) {
+      const auto result = ((lhs.Idx() + 1) % 32) == rhs.Idx();
+      lhs = rhs;
+      return result;
+    };
+
+    return (fn(first, args) && ...);
+  }
+
+  // This is an emitter that is designed around the smallest code bloat as possible.
+  // Eschewing most developer convenience in order to keep code as small as possible.
+
+  // Choices:
+  // - Size of ops passed as an argument rather than template to let the compiler use csel instead of branching.
+  // - Registers are unsized so they can be passed in a GPR and not need conversion operations
+  class Emitter : public FEXCore::ARMEmitter::Buffer {
+  public:
+    Emitter() = default;
+
+    Emitter(uint8_t* Base, uint64_t BaseSize)
+      : Buffer (Base, BaseSize) {
+    }
+
+    // Bind a backward label to an address.
+    // Address that is bound is the current emitter location.
+    void Bind(BackwardLabel *Label) {
+      LOGMAN_THROW_AA_FMT(Label->Location == nullptr, "Trying to bind a label twice");
+      Label->Location = GetCursorAddress<uint8_t*>();
+    }
+
+    // Bind a forward label to a location.
+    // This walks all the instructions in the label's vector.
+    // Then backpatching all instructions that have used the label.
+    template<bool WarnAboutEmpty = false>
+    void Bind(ForwardLabel *Label) {
+      if constexpr (WarnAboutEmpty) {
+        LOGMAN_THROW_A_FMT(Label->Insts.empty() == false, "Binding forward label that didn't have any instructions using it");
+      }
+      uint8_t *CurrentAddress = GetCursorAddress<uint8_t*>();
+      for (const auto &Inst : Label->Insts) {
+        // Patch up the instructions
+        switch (Inst.Type) {
+          case ForwardLabel::Instructions::InstType::ADR: {
+            uint32_t *Instruction = reinterpret_cast<uint32_t*>(Inst.Location);
+            int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
+            LOGMAN_THROW_A_FMT(IsADRRange(Imm), "Unscaled offset too large");
+            uint32_t InstMask = 0b11 << 29 | 0b1111'1111'1111'1111'111 << 5;
+            uint32_t Offset = static_cast<uint32_t>(Imm) & 0x3F'FFFF;
+            uint32_t Inst = *Instruction & ~InstMask;
+            Inst |= (Offset & 0b11) << 29;
+            Inst |= (Offset >> 2) << 5;
+            *Instruction = Inst;
+            break;
+          }
+          case ForwardLabel::Instructions::InstType::ADRP: {
+            uint32_t *Instruction = reinterpret_cast<uint32_t*>(Inst.Location);
+            int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
+            LOGMAN_THROW_A_FMT(IsADRPRange(Imm) && IsADRPAligned(Imm), "Unscaled offset too large");
+            Imm >>= 12;
+            uint32_t InstMask = 0b11 << 29 | 0b1111'1111'1111'1111'111 << 5;
+            uint32_t Offset = static_cast<uint32_t>(Imm) & 0x3F'FFFF;
+            uint32_t Inst = *Instruction & ~InstMask;
+            Inst |= (Offset & 0b11) << 29;
+            Inst |= (Offset >> 2) << 5;
+            *Instruction = Inst;
+            break;
+          }
+
+          case ForwardLabel::Instructions::InstType::B: {
+            uint32_t *Instruction = reinterpret_cast<uint32_t*>(Inst.Location);
+            int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
+            LOGMAN_THROW_A_FMT(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+            Imm >>= 2;
+            uint32_t InstMask = 0x3FF'FFFF;
+            uint32_t Offset = static_cast<uint32_t>(Imm) & InstMask;
+            uint32_t Inst = *Instruction & ~InstMask;
+            Inst |= Offset;
+            *Instruction = Inst;
+
+            break;
+          }
+
+          case ForwardLabel::Instructions::InstType::TEST_BRANCH: {
+            uint32_t *Instruction = reinterpret_cast<uint32_t*>(Inst.Location);
+            int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
+            LOGMAN_THROW_A_FMT(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+            Imm >>= 2;
+            uint32_t InstMask = 0x3FFF;
+            uint32_t Offset = static_cast<uint32_t>(Imm) & InstMask;
+            uint32_t Inst = *Instruction & ~(InstMask << 5);
+            Inst |= Offset << 5;
+            *Instruction = Inst;
+
+            break;
+          }
+          case ForwardLabel::Instructions::InstType::BC:
+          case ForwardLabel::Instructions::InstType::RELATIVE_LOAD: {
+            uint32_t *Instruction = reinterpret_cast<uint32_t*>(Inst.Location);
+            int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
+            LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+            Imm >>= 2;
+            uint32_t InstMask = 0x7'FFFF;
+            uint32_t Offset = static_cast<uint32_t>(Imm) & InstMask;
+            uint32_t Inst = *Instruction & ~(InstMask << 5);
+            Inst |= Offset << 5;
+            *Instruction = Inst;
+            break;
+          }
+          case ForwardLabel::Instructions::InstType::LONG_ADDRESS_GEN: {
+            uint32_t *Instructions = reinterpret_cast<uint32_t*>(Inst.Location);
+            int64_t ImmInstOne = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[0]);
+            int64_t ImmInstTwo = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[1]);
+            auto OriginalOffset = GetCursorOffset();
+
+            auto InstOffset = GetCursorOffsetFromAddress(Instructions);
+            SetCursorOffset(InstOffset);
+
+            // We encoded the destination register in to the first instruction space.
+            // Read it back.
+            ARMEmitter::Register DestReg(Instructions[0]);
+
+            if (IsADRRange(ImmInstTwo)) {
+              // If within ADR range from the second instruction, then we can emit NOP+ADR
+              nop();
+              adr(DestReg, static_cast<uint32_t>(ImmInstTwo) & 0x7FFF);
+            }
+            else if (IsADRPRange(ImmInstOne)) {
+
+              // If within ADRP range from the first instruction, then we are /definitely/ in range for the second instruction.
+              // First check if we are in non-offset range for second instruction.
+              if (IsADRPAligned(reinterpret_cast<uint64_t>(CurrentAddress))) {
+                // We can emit nop + adrp
+                nop();
+                adrp(DestReg, static_cast<uint32_t>(ImmInstTwo >> 12) & 0x7FFF);
+              }
+              else {
+                // Not aligned, need adrp + add
+                adrp(DestReg, static_cast<uint32_t>(ImmInstOne >> 12) & 0x7FFF);
+                add(ARMEmitter::Size::i64Bit, DestReg, DestReg, ImmInstOne & 0xFFF);
+              }
+            }
+            else {
+              LOGMAN_MSG_A_FMT("Unscaled offset is too large");
+              FEX_UNREACHABLE;
+            }
+
+            SetCursorOffset(OriginalOffset);
+            break;
+          }
+          default: LOGMAN_MSG_A_FMT("Unexpected inst type in label fixup");
+        }
+      }
+    }
+
+    // Bind a bidirectional location to a location.
+    // Binds both forwards and backwards depending on how the label was used.
+    void Bind(BiDirectionalLabel *Label) {
+      if (!Label->Backward.Location) {
+        Bind(&Label->Backward);
+      }
+      Bind<false>(&Label->Forward);
+    }
+
+  public:
+  // TODO: Implement SME when it matters.
+#include "Interface/Core/ArchHelpers/CodeEmitter/ALUOps.inl"
+#include "Interface/Core/ArchHelpers/CodeEmitter/BranchOps.inl"
+#include "Interface/Core/ArchHelpers/CodeEmitter/LoadstoreOps.inl"
+#include "Interface/Core/ArchHelpers/CodeEmitter/SystemOps.inl"
+#include "Interface/Core/ArchHelpers/CodeEmitter/ScalarOps.inl"
+#include "Interface/Core/ArchHelpers/CodeEmitter/ASIMDOps.inl"
+#include "Interface/Core/ArchHelpers/CodeEmitter/SVEOps.inl"
+
+  private:
+    template<typename T>
+    uint32_t Encode_ra(T Reg) const {
+      return Reg.Idx() << 10;
+    }
+    uint32_t Encode_ra(uint32_t Reg) const {
+      return Reg << 10;
+    }
+    template<typename T>
+    uint32_t Encode_rt2(T Reg) const {
+      return Reg.Idx() << 10;
+    }
+    template<>
+    uint32_t Encode_rt2(uint32_t Reg) const {
+      return Reg << 10;
+    }
+    template<typename T>
+    uint32_t Encode_rm(T Reg) const {
+      return Reg.Idx() << 16;
+    }
+    uint32_t Encode_rm(uint32_t Reg) const {
+      return Reg << 16;
+    }
+    template<typename T>
+    uint32_t Encode_rs(T Reg) const {
+      return Reg.Idx() << 16;
+    }
+    uint32_t Encode_rs(uint32_t Reg) const {
+      return Reg << 16;
+    }
+    template<typename T>
+    uint32_t Encode_rn(T Reg) const {
+      return Reg.Idx() << 5;
+    }
+    uint32_t Encode_rn(uint32_t Reg) const {
+      return Reg << 5;
+    }
+    template<typename T>
+    uint32_t Encode_rd(T Reg) const {
+      return Reg.Idx();
+    }
+    uint32_t Encode_rd(uint32_t Reg) const {
+      return Reg;
+    }
+    template<typename T>
+    uint32_t Encode_rt(T Reg) const {
+      return Reg.Idx();
+    }
+    template<>
+    uint32_t Encode_rt(Prefetch Reg) const {
+      return FEXCore::ToUnderlying(Reg);
+    }
+    uint32_t Encode_rt(uint32_t Reg) const {
+      return Reg;
+    }
+    template<typename T>
+    uint32_t Encode_pd(T Reg) const {
+      return FEXCore::ToUnderlying(Reg);
+    }
+  };
+}
@@ -0,0 +1,175 @@
+/* System instruction emitters.
+ *
+ * This is mostly a mashup of various instruction types.
+ * Nothing follows an explicit pattern since they are mostly different.
+ */
+public:
+    // System with result
+    // TODO: SYSL
+    // System Instruction
+    // TODO: AT
+    // TODO: CFP
+    // TODO: CPP
+    void dc(FEXCore::ARMEmitter::DataCacheOperation DCOp, FEXCore::ARMEmitter::Register rt) {
+      constexpr uint32_t Op = 0b1101'0101'0000'1000'0111 << 12;
+      SystemInstruction(Op, 0, FEXCore::ToUnderlying(DCOp), rt);
+    }
+    // TODO: DVP
+    // TODO: IC
+    // TODO: TLBI
+
+    // Exception generation
+    void svc(uint32_t Imm) {
+      ExceptionGeneration(0b000, 0b000, 0b01, Imm);
+    }
+    void hvc(uint32_t Imm) {
+      ExceptionGeneration(0b000, 0b000, 0b10, Imm);
+    }
+    void smc(uint32_t Imm) {
+      ExceptionGeneration(0b000, 0b000, 0b11, Imm);
+    }
+    void brk(uint32_t Imm) {
+      ExceptionGeneration(0b001, 0b000, 0b00, Imm);
+    }
+    void hlt(uint32_t Imm) {
+      ExceptionGeneration(0b010, 0b000, 0b00, Imm);
+    }
+    void tcancel(uint32_t Imm) {
+      ExceptionGeneration(0b011, 0b000, 0b00, Imm);
+    }
+    void dcps1(uint32_t Imm) {
+      ExceptionGeneration(0b101, 0b000, 0b01, Imm);
+    }
+    void dcps2(uint32_t Imm) {
+      ExceptionGeneration(0b101, 0b000, 0b10, Imm);
+    }
+    void dcps3(uint32_t Imm) {
+      ExceptionGeneration(0b101, 0b000, 0b11, Imm);
+    }
+    // System instructions with register argument
+    void wfet(FEXCore::ARMEmitter::Register rt) {
+      SystemInstructionWithReg(0b0000, 0b000, rt);
+    }
+    void wfit(FEXCore::ARMEmitter::Register rt) {
+      SystemInstructionWithReg(0b0000, 0b001, rt);
+    }
+
+    // Hints
+    void nop() {
+      Hint(FEXCore::ARMEmitter::HintRegister::NOP);
+    }
+    void yield() {
+      Hint(FEXCore::ARMEmitter::HintRegister::YIELD);
+    }
+    void wfe() {
+      Hint(FEXCore::ARMEmitter::HintRegister::WFE);
+    }
+    void wfi() {
+      Hint(FEXCore::ARMEmitter::HintRegister::WFI);
+    }
+    void sev() {
+      Hint(FEXCore::ARMEmitter::HintRegister::SEV);
+    }
+    void sevl() {
+      Hint(FEXCore::ARMEmitter::HintRegister::SEVL);
+    }
+    void dgh() {
+      Hint(FEXCore::ARMEmitter::HintRegister::DGH);
+    }
+    void csdb() {
+      Hint(FEXCore::ARMEmitter::HintRegister::CSDB);
+    }
+
+    // Barriers
+    void clrex(uint32_t imm = 15) {
+      LOGMAN_THROW_AA_FMT(imm < 16, "Immediate out of range");
+      Barrier(FEXCore::ARMEmitter::BarrierRegister::CLREX, imm);
+    }
+    void dsb(FEXCore::ARMEmitter::BarrierScope Scope) {
+      Barrier(FEXCore::ARMEmitter::BarrierRegister::DSB, FEXCore::ToUnderlying(Scope));
+    }
+    void dmb(FEXCore::ARMEmitter::BarrierScope Scope) {
+      Barrier(FEXCore::ARMEmitter::BarrierRegister::DMB, FEXCore::ToUnderlying(Scope));
+    }
+    void isb() {
+      Barrier(FEXCore::ARMEmitter::BarrierRegister::ISB, FEXCore::ToUnderlying(FEXCore::ARMEmitter::BarrierScope::SY));
+    }
+    void sb() {
+      Barrier(FEXCore::ARMEmitter::BarrierRegister::SB, 0);
+    }
+    void tcommit() {
+      Barrier(FEXCore::ARMEmitter::BarrierRegister::TCOMMIT, 0);
+    }
+
+    // System register move
+    void msr(FEXCore::ARMEmitter::SystemRegister reg, FEXCore::ARMEmitter::Register rt) {
+      constexpr uint32_t Op = 0b1101'0101'0001 << 20;
+      SystemRegisterMove(Op, rt, reg);
+    }
+
+    void mrs(FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::SystemRegister reg) {
+      constexpr uint32_t Op = 0b1101'0101'0011 << 20;
+      SystemRegisterMove(Op, rd, reg);
+    }
+
+private:
+
+    // Exception Generation
+    void ExceptionGeneration(uint32_t opc, uint32_t op2, uint32_t LL, uint32_t Imm) {
+      LOGMAN_THROW_AA_FMT((Imm & 0xFFFF'0000) == 0, "Imm amount too large");
+
+      uint32_t Instr = 0b1101'0100 << 24;
+
+      Instr |= opc << 21;
+      Instr |= Imm << 5;
+      Instr |= op2 << 2;
+      Instr |= LL;
+
+      dc32(Instr);
+    }
+
+    // System instructions with register argument
+    void SystemInstructionWithReg(uint32_t CRm, uint32_t op2, FEXCore::ARMEmitter::Register rt) {
+      uint32_t Instr = 0b1101'0101'0000'0011'0001 << 12;
+
+      Instr |= CRm << 8;
+      Instr |= op2 << 5;
+      Instr |= Encode_rt(rt);
+      dc32(Instr);
+    }
+
+    // Hints
+    void Hint(FEXCore::ARMEmitter::HintRegister Reg) {
+      uint32_t Instr = 0b1101'0101'0000'0011'0010'0000'0001'1111U;
+      Instr |= FEXCore::ToUnderlying(Reg);
+      dc32(Instr);
+    }
+    // Barriers
+    void Barrier(FEXCore::ARMEmitter::BarrierRegister Reg, uint32_t CRm) {
+      uint32_t Instr = 0b1101'0101'0000'0011'0011'0000'0001'1111U;
+      Instr |= CRm << 8;
+      Instr |= FEXCore::ToUnderlying(Reg);
+      dc32(Instr);
+    }
+
+    // System Instruction
+    void SystemInstruction(uint32_t Op, uint32_t L, uint32_t SubOp, FEXCore::ARMEmitter::Register rt) {
+      uint32_t Instr = Op;
+
+      Instr |= L << 21;
+      Instr |= SubOp;
+      Instr |= Encode_rt(rt);
+
+      dc32(Instr);
+    }
+
+    // System register move
+    void SystemRegisterMove(uint32_t Op, FEXCore::ARMEmitter::Register rt, FEXCore::ARMEmitter::SystemRegister reg) {
+      uint32_t Instr = Op;
+
+      Instr |= FEXCore::ToUnderlying(reg);
+      Instr |= Encode_rt(rt);
+
+      dc32(Instr);
+    }
+
@@ -0,0 +1,88 @@
+#include "FEXCore/Utils/AllocatorHooks.h"
+#include "Interface/Context/Context.h"
+#include "Interface/Core/Dispatcher/Dispatcher.h"
+#include <FEXCore/Core/CPUBackend.h>
+
+namespace FEXCore {
+namespace CPU {
+
+CPUBackend::CPUBackend(FEXCore::Core::InternalThreadState *ThreadState, size_t InitialCodeSize, size_t MaxCodeSize)
+    : ThreadState(ThreadState), InitialCodeSize(InitialCodeSize), MaxCodeSize(MaxCodeSize) {}
+
+CPUBackend::~CPUBackend() {
+  for (auto CodeBuffer : CodeBuffers) {
+    FreeCodeBuffer(CodeBuffer);
+  }
+  CodeBuffers.clear();
+}
+
+auto CPUBackend::GetEmptyCodeBuffer() -> CodeBuffer * {
+  if (ThreadState->CurrentFrame->SignalHandlerRefCounter == 0) {
+    if (CodeBuffers.empty()) {
+      auto NewCodeBuffer = AllocateNewCodeBuffer(InitialCodeSize);
+      EmplaceNewCodeBuffer(NewCodeBuffer);
+    } else {
+      if (CodeBuffers.size() > 1) {
+        // If we have more than one code buffer we are tracking then walk them and delete
+        // This is a cleanup step
+        for (size_t i = 1; i < CodeBuffers.size(); i++) {
+          FreeCodeBuffer(CodeBuffers[i]);
+        }
+        CodeBuffers.resize(1);
+      }
+      // Set the current code buffer to the initial
+      CurrentCodeBuffer = &CodeBuffers[0];
+
+      if (CurrentCodeBuffer->Size != MaxCodeSize) {
+        FreeCodeBuffer(*CurrentCodeBuffer);
+
+        // Resize the code buffer and reallocate our code size
+        CurrentCodeBuffer->Size *= 1.5;
+        CurrentCodeBuffer->Size = std::min(CurrentCodeBuffer->Size, MaxCodeSize);
+
+        *CurrentCodeBuffer = AllocateNewCodeBuffer(CurrentCodeBuffer->Size);
+      }
+    }
+  } else {
+    // We have signal handlers that have generated code
+    // This means that we can not safely clear the code at this point in time
+    // Allocate some new code buffers that we can switch over to instead
+    auto NewCodeBuffer = AllocateNewCodeBuffer(InitialCodeSize);
+    EmplaceNewCodeBuffer(NewCodeBuffer);
+  }
+
+  return CurrentCodeBuffer;
+}
+
+auto CPUBackend::AllocateNewCodeBuffer(size_t Size) -> CodeBuffer {
+  CodeBuffer Buffer;
+  Buffer.Size = Size;
+  Buffer.Ptr = static_cast<uint8_t *>(
+      FEXCore::Allocator::VirtualAlloc(Buffer.Size, true));
+  LOGMAN_THROW_AA_FMT(!!Buffer.Ptr, "Couldn't allocate code buffer");
+
+  if (static_cast<Context::ContextImpl*>(ThreadState->CTX)->Config.GlobalJITNaming()) {
+    static_cast<Context::ContextImpl*>(ThreadState->CTX)->Symbols.RegisterJITSpace(Buffer.Ptr, Buffer.Size);
+  }
+  return Buffer;
+}
+
+void CPUBackend::FreeCodeBuffer(CodeBuffer Buffer) {
+  FEXCore::Allocator::VirtualFree(Buffer.Ptr, Buffer.Size);
+}
+
+bool CPUBackend::IsAddressInCodeBuffer(uintptr_t Address) const {
+  for (auto &Buffer: CodeBuffers) {
+    auto start = (uintptr_t)Buffer.Ptr;
+    auto end = start + Buffer.Size;
+
+    if (Address >= start && Address < end) {
+      return true;
+    }
+  }
+
+  return false;
+}
+
+}
+}
@@ -8,18 +8,20 @@ $end_info$
 #include "Common/StringConv.h"
 #include "Interface/Context/Context.h"
 #include "Interface/Core/CPUID.h"
-#include "Interface/Core/HostFeatures.h"
-#include "Utils/FileLoading.h"

 #include <FEXCore/Config/Config.h>
 #include <FEXCore/Core/CPUID.h>
+#include <FEXCore/Core/HostFeatures.h>
+#include <FEXCore/Utils/CPUInfo.h>
+#include <FEXCore/Utils/FileLoading.h>
+#include <FEXCore/fextl/string.h>
 #include <FEXHeaderUtils/Syscalls.h>

 #include "git_version.h"

 #include <cstring>
 #ifdef _M_X86_64
-#include <cpuid.h>
+#include "Interface/Core/Dispatcher/X86Dispatcher.h"
 #endif

 namespace FEXCore {
@@ -39,6 +41,7 @@ namespace ProductNames {
  static const char ARM_A78C[] = "Cortex-A78C";
  static const char ARM_A710[] = "Cortex-A710";
  static const char ARM_X1[] = "Cortex-X1";
+  static const char ARM_X1C[] = "Cortex-X1C";
  static const char ARM_X2[] = "Cortex-X2";
  static const char ARM_N1[] = "Neoverse N1";
  static const char ARM_N2[] = "Neoverse N2";
@@ -73,18 +76,6 @@ static uint32_t GetCPUID() {
  return CPU;
 }

-static uint32_t CalculateNumberOfCPUs() {
-  size_t CPUs = 1;
-
-  while(std::filesystem::exists("/sys/devices/system/cpu/cpu" + std::to_string(CPUs))) {
-    CPUs++;
-  }
-
-  return CPUs;
-}
-
-constexpr uint32_t SUPPORTS_AVX = 0;
-// #define CPUID_AMD
 #ifdef CPUID_AMD
 constexpr uint32_t FAMILY_IDENTIFIER =
  0 |          // Stepping
@@ -112,30 +103,30 @@ static uint32_t GetCycleCounterFrequency() {
 }

 void CPUIDEmu::SetupHostHybridFlag() {
-  size_t CPUs = CalculateNumberOfCPUs();
+  size_t CPUs = FEXCore::CPUInfo::CalculateNumberOfCPUs();
  PerCPUData.resize(CPUs);

  uint64_t MIDR{};
  for (size_t i = 0; i < CPUs; ++i) {
    std::error_code ec{};
-    std::string MIDRPath = "/sys/devices/system/cpu/cpu" + std::to_string(i) + "/regs/identification/midr_el1";
-    if (std::filesystem::exists(MIDRPath, ec)) {
-      std::vector<char> Data{};
-      // Needs to be a fixed size since depending on kernel it will try to read a full page of data and fail
-      // Only read 18 bytes for a 64bit value prefixed with 0x
-      if (FEXCore::FileLoading::LoadFile(Data, MIDRPath, 18)) {
-        uint64_t NewMIDR{};
-        if (FEXCore::StrConv::Conv(&Data.at(0), &NewMIDR)) {
-          if (MIDR != 0 && MIDR != NewMIDR) {
-            // CPU mismatch, claim hybrid
-            Hybrid = true;
-          }
+    fextl::string MIDRPath = fextl::fmt::format("/sys/devices/system/cpu/cpu{}/regs/identification/midr_el1", i);

-          // Truncate to 32-bits, top 32-bits are all reserved in MIDR
-          PerCPUData[i].ProductName = ProductNames::ARM_UNKNOWN;
-          PerCPUData[i].MIDR = NewMIDR;
-          MIDR = NewMIDR;
+    std::array<char, 18> Data;
+    // Needs to be a fixed size since depending on kernel it will try to read a full page of data and fail
+    // Only read 18 bytes for a 64bit value prefixed with 0x
+    if (FEXCore::FileLoading::LoadFileToBuffer(MIDRPath, Data) == sizeof(Data)) {
+      uint64_t NewMIDR{};
+      std::string_view MIDRView(Data.data(), sizeof(Data));
+      if (FEXCore::StrConv::Conv(MIDRView, &NewMIDR)) {
+        if (MIDR != 0 && MIDR != NewMIDR) {
+          // CPU mismatch, claim hybrid
+          Hybrid = true;
        }
+
+        // Truncate to 32-bits, top 32-bits are all reserved in MIDR
+        PerCPUData[i].ProductName = ProductNames::ARM_UNKNOWN;
+        PerCPUData[i].MIDR = NewMIDR;
+        MIDR = NewMIDR;
      }
    }
  }
@@ -150,7 +141,7 @@ void CPUIDEmu::SetupHostHybridFlag() {
  // CPU priority order
  // This is mostly arbitrary but will sort by some sort of CPU priority by performance
  // Relative list so things they will commonly end up in big.little configurations sort of relate
-  static constexpr std::array<CPUMIDR, 35> CPUMIDRs = {{
+  static constexpr std::array<CPUMIDR, 36> CPUMIDRs = {{
    // Typically big CPU cores
    {0x61, 0x023, 1, ProductNames::ARM_Firestorm}, // Apple M1 Firestorm

@@ -160,6 +151,7 @@ void CPUIDEmu::SetupHostHybridFlag() {
    {0x41, 0xd49, 1, ProductNames::ARM_N2}, // N2
    {0x41, 0xd48, 1, ProductNames::ARM_X2}, // X2
    {0x41, 0xd47, 1, ProductNames::ARM_A710}, // A710
+    {0x41, 0xd4C, 1, ProductNames::ARM_X1C}, // X1C
    {0x41, 0xd44, 1, ProductNames::ARM_X1}, // X1
    {0x41, 0xd42, 1, ProductNames::ARM_A78AE}, // A78AE
    {0x41, 0xd41, 1, ProductNames::ARM_A78}, // A78
@@ -213,8 +205,8 @@ void CPUIDEmu::SetupHostHybridFlag() {

  if (Hybrid) {
    // Walk the MIDRs and calculate big little designs
-    std::vector<const CPUMIDR*> BigCores;
-    std::vector<const CPUMIDR*> LittleCores;
+    fextl::vector<const CPUMIDR*> BigCores;
+    fextl::vector<const CPUMIDR*> LittleCores;

    // Separate CPU cores out to big or little selected
    for (size_t i = 0; i < CPUs; ++i) {
@@ -350,28 +342,28 @@ void CPUIDEmu::SetupHostHybridFlag() {

 #else
 static uint32_t GetCycleCounterFrequency() {
-  uint32_t eax, ebx, ecx, edx;
-  __cpuid(0, eax, ebx, ecx, edx);
-  if (eax >= 0x15) {
-    __cpuid(0x15, eax, ebx, ecx, edx);
+  uint32_t data[4];
+  Xbyak::util::Cpu::getCpuid(0, data);
+  if (data[0] >= 0x15) {
+    Xbyak::util::Cpu::getCpuid(0x15, data);

-    if (eax && ebx && ecx) {
-      return ecx * ebx / eax;
+    if (data[0] && data[1] && data[2]) {
+      return data[2] * data[1] / data[0];
    }
  }
  return 0;
 }

 void CPUIDEmu::SetupHostHybridFlag() {
-  uint32_t eax, ebx, ecx, edx;
-  __cpuid(0, eax, ebx, ecx, edx);
-  if (eax >= 0x7) {
-    __cpuid(0x7, eax, ebx, ecx, edx);
+  uint32_t data[4];
+  Xbyak::util::Cpu::getCpuid(0, data);
+  if (data[0] >= 0x7) {
+    Xbyak::util::Cpu::getCpuid(0x7, data);
    // Bit 15 of edx claims hybrid CPU
-    Hybrid = (edx & (1U << 15)) != 0;
+    Hybrid = (data[3] & (1U << 15)) != 0;
  }

-  size_t CPUs = CalculateNumberOfCPUs();
+  size_t CPUs = FEXCore::CPUInfo::CalculateNumberOfCPUs();
  PerCPUData.resize(CPUs);
  for (size_t i = 0; i < CPUs; ++i) {
    PerCPUData[i].IsBig = true;
@@ -404,6 +396,9 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) {
  FEXCore::CPUID::FunctionResults Res{};
  uint32_t CoreCount = Cores();

+  // Hypervisor bit is normally set but some applications have issues with it.
+  uint32_t Hypervisor = HideHypervisorBit() ? 0 : 1;
+
  Res.eax = FAMILY_IDENTIFIER;

  Res.ebx = 0 | // Brand index
@@ -413,7 +408,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) {

  Res.ecx =
    (1 <<  0) | // SSE3
-    (0 <<  1) | // PCLMULQDQ
+    (CTX->HostFeatures.SupportsPMULL_128Bit <<  1) | // PCLMULQDQ
    (1 <<  2) | // DS area supports 64bit layout
    (1 <<  3) | // MWait
    (0 <<  4) | // DS-CPL
@@ -432,18 +427,18 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) {
    (0 << 17) | // Process-context identifiers
    (0 << 18) | // Prefetching from memory mapped device
    (1 << 19) | // SSE4.1
-    (0 << 20) | // SSE4.2
+    (CTX->HostFeatures.SupportsCRC << 20) | // SSE4.2
    (0 << 21) | // X2APIC
    (1 << 22) | // MOVBE
    (1 << 23) | // POPCNT
    (0 << 24) | // APIC TSC-Deadline
    (CTX->HostFeatures.SupportsAES << 25) | // AES
-    (0 << 26) | // XSAVE
-    (0 << 27) | // OSXSAVE
-    (SUPPORTS_AVX << 28) | // AVX
+    (SupportsAVX() << 26) | // XSAVE
+    (SupportsAVX() << 27) | // OSXSAVE
+    (SupportsAVX() << 28) | // AVX
    (0 << 29) | // F16C
-    (0 << 30) | // RDRAND
-    (0 << 31);  // Hypervisor always returns zero
+    (CTX->HostFeatures.SupportsRAND << 30) | // RDRAND
+    (Hypervisor << 31);

  Res.edx =
    (1 <<  0) | // FPU
@@ -629,12 +624,12 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_07h(uint32_t Leaf) {
      (1 <<  0) | // FS/GS support
      (0 <<  1) | // TSC adjust MSR
      (0 <<  2) | // SGX
-      (1 <<  3) | // BMI1
+      (SupportsAVX() << 3) | // BMI1
      (0 <<  4) | // Intel Hardware Lock Elison
      (0 <<  5) | // AVX2 support
      (1 <<  6) | // FPU data pointer updated only on exception
      (1 <<  7) | // SMEP support
-      (1 <<  8) | // BMI2
+      (SupportsAVX() << 8) | // BMI2
      (0 <<  9) | // Enhanced REP MOVSB/STOSB
      (1 << 10) | // INVPCID for system software control of process-context
      (0 << 11) | // Restricted transactional memory
@@ -644,18 +639,18 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_07h(uint32_t Leaf) {
      (0 << 15) | // Intel Resource Directory Technology Allocation
      (0 << 16) | // Reserved
      (0 << 17) | // Reserved
-      (0 << 18) | // RDSEED
+      (CTX->HostFeatures.SupportsRAND << 18) | // RDSEED
      (1 << 19) | // ADCX and ADOX instructions
      (0 << 20) | // SMAP Supervisor mode access prevention and CLAC/STAC instructions
      (0 << 21) | // Reserved
      (0 << 22) | // Reserved
-      (0 << 23) | // CLFLUSHOPT instruction
-      (0 << 24) | // CLWB instruction
+      (1 << 23) | // CLFLUSHOPT instruction
+      (CTX->HostFeatures.SupportsCLWB << 24) | // CLWB instruction
      (0 << 25) | // Intel processor trace
      (0 << 26) | // Reserved
      (0 << 27) | // Reserved
      (0 << 28) | // Reserved
-      (0 << 29) | // SHA instructions
+      (1 << 29) | // SHA instructions
      (0 << 30) | // Reserved
      (0 << 31);  // Reserved

@@ -698,7 +693,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_07h(uint32_t Leaf) {
      (0 <<  1) | // Reserved
      (0 <<  2) | // AVX512_4VNNIW
      (0 <<  3) | // AVX512_4FMAPS
-      (0 <<  4) | // Fast Short Rep Mov
+      (1 <<  4) | // Fast Short Rep Mov
      (0 <<  5) | // Reserved
      (0 <<  6) | // Reserved
      (0 <<  7) | // Reserved
@@ -735,13 +730,13 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_0Dh(uint32_t Leaf) {
  // Leaf 0
  FEXCore::CPUID::FunctionResults Res{};

-  uint32_t XFeatureSupportedSizeMax = SUPPORTS_AVX ? 0x0000'0340 : 0x0000'0240; // XFeatureEnabledSizeMax: Legacy Header + FPU/SSE + AVX
+  uint32_t XFeatureSupportedSizeMax = SupportsAVX() ? 0x0000'0340 : 0x0000'0240; // XFeatureEnabledSizeMax: Legacy Header + FPU/SSE + AVX
  if (Leaf == 0) {
    // XFeatureSupportedMask[31:0]
    Res.eax =
      (1 << 0) |            // X87 support
      (1 << 1) |            // 128-bit SSE support
-      (SUPPORTS_AVX << 2) | // 256-bit AVX support
+      (SupportsAVX() << 2) | // 256-bit AVX support
      (0b00 << 3) |         // MPX State
      (0b000 << 5) |        // AVX-512 state
      (0 << 8) |            // "Used for IA32_XSS" ... Used for what?
@@ -775,8 +770,8 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_0Dh(uint32_t Leaf) {
    Res.edx = 0;
  }
  else if (Leaf == 2) {
-    Res.eax = SUPPORTS_AVX ? 0x0000'0100 : 0; // YmmSaveStateSize
-    Res.ebx = SUPPORTS_AVX ? 0x0000'0240 : 0; // YmmSaveStateOffset
+    Res.eax = SupportsAVX() ? 0x0000'0100 : 0; // YmmSaveStateSize
+    Res.ebx = SupportsAVX() ? 0x0000'0240 : 0; // YmmSaveStateOffset

    // Reserved
    Res.ecx = 0;
@@ -809,6 +804,47 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_1Ah(uint32_t Leaf) {
  return Res;
 }

+// Hypervisor CPUID information leaf
+FEXCore::CPUID::FunctionResults CPUIDEmu::Function_4000_0000h(uint32_t Leaf) {
+  FEXCore::CPUID::FunctionResults Res{};
+  // Maximum supported hypervisor leafs
+  // We only expose the information leaf
+  //
+  // Common courtesy to follow VMWare's "Hypervisor CPUID Interface proposal"
+  // 4000_0000h - Information leaf. Advertising to the software which hypervisor this is
+  // 4000_0001h - 4000_000Fh - Hypervisor specific leafs. FEX can use these for anything
+  // 4000_0010h - 4000_00FFh - "Generic Leafs" - Try not to overwrite, other hypervisors might expect information in these
+  //
+  // CPUID documentation information:
+  // 4000_0000h - 4FFF_FFFFh - No existing or future CPU will return information in this range
+  // Reserved entirely for VMs to do whatever they want.
+  Res.eax = 0x40000001;
+
+  // EBX, EDX, ECX become the hypervisor ID signature
+  constexpr static char HypervisorID[12] = "FEXIFEXIEMU";
+  memcpy(&Res.ebx, HypervisorID, sizeof(HypervisorID));
+  return Res;
+}
+
+// Hypervisor CPUID information leaf
+FEXCore::CPUID::FunctionResults CPUIDEmu::Function_4000_0001h(uint32_t Leaf) {
+  FEXCore::CPUID::FunctionResults Res{};
+  if (Leaf == 0) {
+    // EAX[3:0] Is the host architecture that FEX is running under
+#ifdef _M_X86_64
+    // EAX[3:0] = 1 = x86_64 host architecture
+    Res.eax |= 0b0001;
+#elif defined(_M_ARM_64)
+    // EAX[3:0] = 2 = AArch64 host architecture
+    Res.eax |= 0b0010;
+#else
+    // EAX[3:0] = 0 = Unknown architecture
+#endif
+  }
+
+  return Res;
+}
+
 // Highest extended function implemented
 FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0000h(uint32_t Leaf) {
  FEXCore::CPUID::FunctionResults Res{};
@@ -830,6 +866,13 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0000h(uint32_t Leaf) {

 // Extended processor and feature bits
 FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0001h(uint32_t Leaf) {
+
+  // RDTSCP is disabled on WIN32/Wine because there is no sane way to query processor ID.
+#ifndef _WIN32
+  constexpr uint32_t SUPPORTS_RDTSCP = 0;
+#else
+  constexpr uint32_t SUPPORTS_RDTSCP = 1;
+#endif
  FEXCore::CPUID::FunctionResults Res{};

  Res.eax = FAMILY_IDENTIFIER;
@@ -896,11 +939,11 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0001h(uint32_t Leaf) {
    (1 << 24) | // FXSAVE/FXRSTOR
    (1 << 25) | // FXSAVE/FXRSTOR Optimizations
    (0 << 26) | // 1 gigabit pages
-    (1 << 27) | // RDTSCP
+    (SUPPORTS_RDTSCP << 27) | // RDTSCP
    (0 << 28) | // Reserved
    (1 << 29) | // Long Mode
-    (0 << 30) | // 3DNow! Extensions
-    (0 << 31);  // 3DNow!
+    (1 << 30) | // 3DNow! Extensions
+    (1 << 31);  // 3DNow!
  return Res;
 }

@@ -927,14 +970,14 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0004h(uint32_t Leaf) {
 FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0002h(uint32_t Leaf, uint32_t CPU) {
  FEXCore::CPUID::FunctionResults Res{};
  memset(&Res, ' ', sizeof(FEXCore::CPUID::FunctionResults));
-  memcpy(&Res, &ProcessorBrand[0], std::min(16L, DESCRIBE_STR_SIZE));
+  memcpy(&Res, &ProcessorBrand[0], std::min(ssize_t{16L}, DESCRIBE_STR_SIZE));
  return Res;
 }

 FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0003h(uint32_t Leaf, uint32_t CPU) {
  FEXCore::CPUID::FunctionResults Res{};
  memset(&Res, ' ', sizeof(FEXCore::CPUID::FunctionResults));
-  memcpy(&Res, &ProcessorBrand[16], std::max(0L, DESCRIBE_STR_SIZE - 16));
+  memcpy(&Res, &ProcessorBrand[16], std::max(ssize_t{0L}, DESCRIBE_STR_SIZE - 16));
  return Res;
 }

@@ -1163,85 +1206,26 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_Reserved(uint32_t Leaf) {
  return Res;
 }

-void CPUIDEmu::Init(FEXCore::Context::Context *ctx) {
+FEXCore::CPUID::XCRResults CPUIDEmu::XCRFunction_0h() {
+  // This just returns XCR0
+  FEXCore::CPUID::XCRResults Res{
+    .eax = static_cast<uint32_t>(XCR0),
+    .edx = static_cast<uint32_t>(XCR0 >> 32),
+  };
+
+  return Res;
+}
+
+void CPUIDEmu::Init(FEXCore::Context::ContextImpl *ctx) {
  CTX = ctx;

-  RegisterFunction(0, &CPUIDEmu::Function_0h);
-  RegisterFunction(1, &CPUIDEmu::Function_01h);
-  RegisterFunction(2, &CPUIDEmu::Function_02h);
-  // 3: Serial Number(previously), now reserved
-#ifndef CPUID_AMD
-  // Deterministic cache parameters for each level
-  RegisterFunction(0x4, &CPUIDEmu::Function_04h);
-#endif
-  // 5: Monitor/mwait
-  // Thermal and power management
-  RegisterFunction(6, &CPUIDEmu::Function_06h);
-  // Extended feature flags
-  RegisterFunction(7, &CPUIDEmu::Function_07h);
-  // 9: Direct Cache Access information
-  // 0x0A: Architectural performance monitoring
-  // 0x0B: Extended topology enumeration
-  // 0x0D: Processor extended state enumeration
-  RegisterFunction(0x0D, &CPUIDEmu::Function_0Dh);
-  // 0x0F: Intel RDT monitoring
-  // 0x10: Intel RDT allocation enumeration
-  // 0x12: Intel SGX capability enumeration
-  // 0x13: Reserved
-  // 0x14: Intel Processor trace
-#ifndef CPUID_AMD
-  // Timestamp counter information
-  // Doesn't exist on AMD hardware
-  RegisterFunction(0x15, &CPUIDEmu::Function_15h);
-#endif
-  // 0x16: Processor frequency information
-  // 0x17: SoC vendor attribute enumeration
-
-  // 0x1A: Hybrid Information Sub-leaf
-#ifndef CPUID_AMD
-  RegisterFunction(0x1A, &CPUIDEmu::Function_1Ah);
-#endif
-  // Largest extended function number
-  RegisterFunction(0x8000'0000, &CPUIDEmu::Function_8000_0000h);
-  // Processor vendor
-  RegisterFunction(0x8000'0001, &CPUIDEmu::Function_8000_0001h);
-  // Processor brand string
-  RegisterFunction(0x8000'0002, &CPUIDEmu::Function_8000_0002h);
-  // Processor brand string continued
-  RegisterFunction(0x8000'0003, &CPUIDEmu::Function_8000_0003h);
-  // Processor brand string continued
-  RegisterFunction(0x8000'0004, &CPUIDEmu::Function_8000_0004h);
-  // 0x8000'0005: L1 Cache and TLB identifiers
-#ifdef CPUID_AMD
-  RegisterFunction(0x8000'0005, &CPUIDEmu::Function_8000_0005h);
-#else
-  // This is full reserved on Intel platforms
-  RegisterFunction(0x8000'0005, &CPUIDEmu::Function_Reserved);
-#endif
-  // 0x8000'0006: L2 Cache identifiers
-  RegisterFunction(0x8000'0006, &CPUIDEmu::Function_8000_0006h);
-  // Advanced power management information
-  RegisterFunction(0x8000'0007, &CPUIDEmu::Function_8000_0007h);
-  // Virtual and physical address sizes
-  RegisterFunction(0x8000'0008, &CPUIDEmu::Function_8000_0008h);
-
-  // 0x8000'000A: SVM Revision
-  // TLB 1GB page identifiers
-  RegisterFunction(0x8000'0019, &CPUIDEmu::Function_8000_0019h);
-
-  // 0x8000'001A: Performance optimization identifiers
-  // 0x8000'001B: Instruction based sampling identifiers
-  // 0x8000'001C: Lightweight profiling capabilities
-  // 0x8000'001D: Cache properties
-#ifdef CPUID_AMD
-  // Deterministic cache parameters for each level
-  RegisterFunction(0x8000'001D, &CPUIDEmu::Function_8000_001Dh);
-#endif
-  // 0x8000'001E: Extended APIC ID
-  // 0x8000'001F: AMD Secure Encryption
-
  // Setup some state tracking
  SetupHostHybridFlag();
+
+  // TODO: Enable once AVX is supported.
+  if (false && CTX->HostFeatures.SupportsAVX) {
+    XCR0 |= XCR0_AVX;
+  }
 }
 }

@@ -1,17 +1,21 @@
 #pragma once

+#include <FEXCore/Core/CPUID.h>
+#include <FEXCore/Config/Config.h>
+#include <FEXCore/fextl/vector.h>
+
 #include <cstdint>
 #include <unordered_map>
 #include <utility>

-#include <FEXCore/Core/CPUID.h>
-#include <FEXCore/Config/Config.h>
-
 namespace FEXCore {
 namespace Context {
-  struct Context;
+  class ContextImpl;
 }

+// Debugging define to switch what family of CPU we execute as.
+// Might be useful if an application makes an assumption about a CPU.
+// #define CPUID_AMD
 class CPUIDEmu final {
 private:
  constexpr static uint32_t CPUID_VENDOR_INTEL1 = 0x756E6547; // "Genu"
@@ -27,16 +31,27 @@ public:
  // if we report anything differently then applications are likely to break
  constexpr static uint64_t CACHELINE_SIZE = 64;

-  void Init(FEXCore::Context::Context *ctx);
+  void Init(FEXCore::Context::ContextImpl *ctx);

  FEXCore::CPUID::FunctionResults RunFunction(uint32_t Function, uint32_t Leaf) {
-    const auto Handler = FunctionHandlers.find(Function);
-
-    if (Handler == FunctionHandlers.end()) {
-      return Function_Reserved(Leaf);
+    if (Function < Primary.size()) {
+      const auto Handler = Primary[Function];
+      return (this->*Handler)(Leaf);
    }

-    return (this->*Handler->second)(Leaf);
+    constexpr uint32_t HypervisorBase = 0x4000'0000;
+    if (Function >= HypervisorBase && Function < (HypervisorBase + Hypervisor.size())) {
+      const auto Handler = Hypervisor[Function - HypervisorBase];
+      return (this->*Handler)(Leaf);
+    }
+
+    constexpr uint32_t ExtendedBase = 0x8000'0000;
+    if (Function >= ExtendedBase && Function < (ExtendedBase + Extended.size())) {
+      const auto Handler = Extended[Function - ExtendedBase];
+      return (this->*Handler)(Leaf);
+    }
+
+    return Function_Reserved(Leaf);
  }

  FEXCore::CPUID::FunctionResults RunFunctionName(uint32_t Function, uint32_t Leaf, uint32_t CPU) {
@@ -48,17 +63,52 @@ public:
      return Function_8000_0004h(Leaf, CPU % PerCPUData.size());
  }

-private:
-  FEXCore::Context::Context *CTX;
-  bool Hybrid{};
-  FEX_CONFIG_OPT(Cores, THREADS);
+  FEXCore::CPUID::XCRResults RunXCRFunction(uint32_t Function) {
+    if (Function >= 1) {
+      // XCR function 1 is not yet supported.
+      return {};
+    }

-  using FunctionHandler = FEXCore::CPUID::FunctionResults (CPUIDEmu::*)(uint32_t Leaf);
-  void RegisterFunction(uint32_t Function, FunctionHandler Handler) {
-    FunctionHandlers.insert_or_assign(Function, Handler);
+    return XCRFunction_0h();
  }

-  std::unordered_map<uint32_t, FunctionHandler> FunctionHandlers;
+private:
+  FEXCore::Context::ContextImpl *CTX;
+  bool Hybrid{};
+  FEX_CONFIG_OPT(Cores, THREADS);
+  FEX_CONFIG_OPT(HideHypervisorBit, HIDEHYPERVISORBIT);
+
+  // XFEATURE_ENABLED_MASK
+  // Mask that configures what features are enabled on the CPU.
+  // Affects XSAVE and XRSTOR when modified.
+  // Bit layout is as follows.
+  // [0]     - x87 enabled
+  // [1]     - SSE enabled
+  // [2]     - YMM enabled (256-bit SSE)
+  // [8:3]   - Reserved. MBZ.
+  // [9]     - MPK
+  // [10]    - Reserved. MBZ.
+  // [11]    - CET_U
+  // [12]    - CET_S
+  // [61:13] - Reserved. MBZ.
+  // [62]    - LWP (Lightweight profiling)
+  // [63]    - Reserved for XCR bit vector expansion. MBZ.
+  // Always enable x87 and SSE by default.
+  constexpr static uint64_t XCR0_X87 = 1ULL << 0;
+  constexpr static uint64_t XCR0_SSE = 1ULL << 1;
+  constexpr static uint64_t XCR0_AVX = 1ULL << 2;
+
+  uint64_t XCR0 {
+    XCR0_X87 |
+    XCR0_SSE
+  };
+
+  uint32_t SupportsAVX() const {
+    return (XCR0 & XCR0_AVX) ? 1 : 0;
+  }
+
+  using FunctionHandler = FEXCore::CPUID::FunctionResults (CPUIDEmu::*)(uint32_t Leaf);
+
  struct CPUData {
    const char *ProductName{};
 #ifdef _M_ARM_64
@@ -66,7 +116,7 @@ private:
 #endif
    bool IsBig{};
  };
-  std::vector<CPUData> PerCPUData{};
+  fextl::vector<CPUData> PerCPUData{};

  // Functions
  FEXCore::CPUID::FunctionResults Function_0h(uint32_t Leaf);
@@ -78,6 +128,8 @@ private:
  FEXCore::CPUID::FunctionResults Function_0Dh(uint32_t Leaf);
  FEXCore::CPUID::FunctionResults Function_15h(uint32_t Leaf);
  FEXCore::CPUID::FunctionResults Function_1Ah(uint32_t Leaf);
+  FEXCore::CPUID::FunctionResults Function_4000_0000h(uint32_t Leaf);
+  FEXCore::CPUID::FunctionResults Function_4000_0001h(uint32_t Leaf);
  FEXCore::CPUID::FunctionResults Function_8000_0000h(uint32_t Leaf);
  FEXCore::CPUID::FunctionResults Function_8000_0001h(uint32_t Leaf);
  FEXCore::CPUID::FunctionResults Function_8000_0002h(uint32_t Leaf);
@@ -92,12 +144,163 @@ private:
  FEXCore::CPUID::FunctionResults Function_8000_0006h(uint32_t Leaf);
  FEXCore::CPUID::FunctionResults Function_8000_0007h(uint32_t Leaf);
  FEXCore::CPUID::FunctionResults Function_8000_0008h(uint32_t Leaf);
-  FEXCore::CPUID::FunctionResults Function_8000_0009h(uint32_t Leaf);
  FEXCore::CPUID::FunctionResults Function_8000_0019h(uint32_t Leaf);
  FEXCore::CPUID::FunctionResults Function_8000_001Dh(uint32_t Leaf);
  FEXCore::CPUID::FunctionResults Function_Reserved(uint32_t Leaf);

-  void SetupHostHybridFlag();
+  FEXCore::CPUID::XCRResults XCRFunction_0h();

+  void SetupHostHybridFlag();
+  static constexpr std::array<FunctionHandler, 27> Primary = {
+    // 0: Highest function parameter and ID
+    &CPUIDEmu::Function_0h,
+    // 1: Processor info
+    &CPUIDEmu::Function_01h,
+    // 2: Cache and TLB info
+    &CPUIDEmu::Function_02h,
+    // 3: Serial Number(previously), now reserved
+    &CPUIDEmu::Function_Reserved,
+#ifndef CPUID_AMD
+    // 4: Deterministic cache parameters for each level
+    &CPUIDEmu::Function_04h,
+#else
+    &CPUIDEmu::Function_Reserved,
+#endif
+    // 5: Monitor/mwait
+    &CPUIDEmu::Function_Reserved,
+    // 6: Thermal and power management
+    &CPUIDEmu::Function_06h,
+    // 7: Extended feature flags
+    &CPUIDEmu::Function_07h,
+    // 0x08: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 9: Direct Cache Access information
+    &CPUIDEmu::Function_Reserved,
+    // 0x0A: Architectural performance monitoring
+    &CPUIDEmu::Function_Reserved,
+    // 0x0B: Extended topology enumeration
+    &CPUIDEmu::Function_Reserved,
+    // 0x0C: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x0D: Processor extended state enumeration
+    &CPUIDEmu::Function_0Dh,
+    // 0x0E: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x0F: Intel RDT monitoring
+    &CPUIDEmu::Function_Reserved,
+    // 0x10: Intel RDT allocation enumeration
+    &CPUIDEmu::Function_Reserved,
+    // 0x12: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x12: Intel SGX capability enumeration
+    &CPUIDEmu::Function_Reserved,
+    // 0x13: Reserved
+    &CPUIDEmu::Function_Reserved,
+    // 0x14: Intel Processor trace
+    &CPUIDEmu::Function_Reserved,
+#ifndef CPUID_AMD
+    // Timestamp counter information
+    // Doesn't exist on AMD hardware
+    &CPUIDEmu::Function_15h,
+#else
+    &CPUIDEmu::Function_Reserved,
+#endif
+    // 0x16: Processor frequency information
+    &CPUIDEmu::Function_Reserved,
+    // 0x17: SoC vendor attribute enumeration
+    &CPUIDEmu::Function_Reserved,
+    // 0x18: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x19: Reserved?
+    &CPUIDEmu::Function_Reserved,
+#ifndef CPUID_AMD
+    // 0x1A: Hybrid Information Sub-leaf
+    &CPUIDEmu::Function_1Ah,
+#else
+    &CPUIDEmu::Function_Reserved,
+#endif
+  };
+
+  static constexpr std::array<FunctionHandler, 2> Hypervisor = {
+    // Hypervisor CPUID information leaf
+    &CPUIDEmu::Function_4000_0000h,
+    // FEX-Emu specific leaf
+    &CPUIDEmu::Function_4000_0001h,
+  };
+
+  static constexpr std::array<FunctionHandler, 32> Extended = {
+    // Largest extended function number
+    &CPUIDEmu::Function_8000_0000h,
+    // Processor vendor
+    &CPUIDEmu::Function_8000_0001h,
+    // Processor brand string
+    &CPUIDEmu::Function_8000_0002h,
+    // Processor brand string continued
+    &CPUIDEmu::Function_8000_0003h,
+    // Processor brand string continued
+    &CPUIDEmu::Function_8000_0004h,
+#ifdef CPUID_AMD
+    // 0x8000'0005: L1 Cache and TLB identifiers
+    &CPUIDEmu::Function_8000_0005h,
+#else
+    &CPUIDEmu::Function_Reserved,
+#endif
+    // 0x8000'0006: L2 Cache identifiers
+    &CPUIDEmu::Function_8000_0006h,
+    // 0x8000'0007: Advanced power management information
+    &CPUIDEmu::Function_8000_0007h,
+    // 0x8000'0008: Virtual and physical address sizes
+    &CPUIDEmu::Function_8000_0008h,
+    // 0x8000'0009: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'000A: SVM Revision
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'000B: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'000C: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'000D: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'000E: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'000F: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'0010: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'0011: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'0012: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'0013: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'0014: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'0015: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'0016: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'0017: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'0018: Reserved?
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'0019: TLB 1GB page identifiers
+    &CPUIDEmu::Function_8000_0019h,
+    // 0x8000'001A: Performance optimization identifiers
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'001B: Instruction based sampling identifiers
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'001C: Lightweight profiling capabilities
+    &CPUIDEmu::Function_Reserved,
+#ifdef CPUID_AMD
+    // 0x8000'001D: Cache properties
+    &CPUIDEmu::Function_8000_001Dh,
+#else
+    &CPUIDEmu::Function_Reserved,
+#endif
+    // 0x8000'001E: Extended APIC ID
+    &CPUIDEmu::Function_Reserved,
+    // 0x8000'001F: AMD Secure Encryption
+    &CPUIDEmu::Function_Reserved,
+  };
 };
 }
@@ -1,165 +0,0 @@
-#include "Interface/Context/Context.h"
-#include "Interface/Core/LookupCache.h"
-#include "Interface/Core/CompileService.h"
-#include "Interface/Core/OpcodeDispatcher.h"
-#include "FEXCore/Debug/InternalThreadState.h"
-#include "FEXCore/HLE/Linux/ThreadManagement.h"
-#include "Interface/IR/PassManager.h"
-
-#include <FEXCore/Core/CPUBackend.h>
-#include <FEXCore/Core/CoreState.h>
-#include <FEXCore/Core/SignalDelegator.h>
-#include <FEXCore/Utils/Event.h>
-#include <FEXCore/Utils/LogManager.h>
-#include <FEXCore/Utils/Threads.h>
-
-#include <memory>
-#include <pthread.h>
-#include <stdio.h>
-
-namespace FEXCore {
-  static void* ThreadHandler(void *Arg) {
-    FEXCore::CompileService *This = reinterpret_cast<FEXCore::CompileService*>(Arg);
-    This->ExecutionThread();
-    return nullptr;
-  }
-
-  CompileService::CompileService(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread)
-    : CTX {ctx}
-    , ParentThread {Thread} {
-
-    CompileThreadData = std::make_unique<FEXCore::Core::InternalThreadState>();
-    CompileThreadData->IsCompileService = true;
-
-    // We need a compiler for this work thread
-    CTX->InitializeCompiler(CompileThreadData.get(), true);
-    CompileThreadData->CPUBackend->CopyNecessaryDataForCompileThread(ParentThread->CPUBackend.get());
-
-    uint64_t OldMask = FEXCore::Threads::SetSignalMask(~0ULL);
-    WorkerThread = FEXCore::Threads::Thread::Create(ThreadHandler, this);
-    FEXCore::Threads::SetSignalMask(OldMask);
-  }
-
-  void CompileService::Initialize() {
-    // Share CompileService which = this
-    CompileThreadData->CompileService = ParentThread->CompileService;
-  }
-
-  void CompileService::Shutdown() {
-    ShuttingDown = true;
-    // Kick the working thread
-    StartWork.NotifyAll();
-    WorkerThread->join(nullptr);
-  }
-
-  void CompileService::ClearCache(FEXCore::Core::InternalThreadState *Thread) {
-    // On cache clear we need to spin down the execution thread to ensure it isn't trying to give us more work items
-    if (CompileMutex.try_lock()) {
-      // We can only clear these things if we pulled the compile mutex
-
-      // Grab the work queue and clear it
-      // We don't need to grab the queue mutex since this thread will no longer receive any work events
-      // Threads are bounded 1:1
-      while (!WorkQueue.empty()) {
-        WorkQueue.pop();
-      }
-
-      // Go through the garbage collection array and clear it
-      // It's safe to clear things that aren't marked safe since we are clearing cache
-      GCArray.clear();
-
-      LOGMAN_THROW_A_FMT(CompileThreadData->LocalIRCache.empty(), "Compile service must never have LocalIRCache");
-
-      CompileMutex.unlock();
-    }
-
-    // Clear the inverse cache of what is calling us from the Context ClearCache routine
-    auto SelectedThread = Thread->IsCompileService ? ParentThread : Thread;
-    SelectedThread->LookupCache->ClearCache();
-    SelectedThread->CPUBackend->ClearCache();
-  }
-
-  CompileService::WorkItem *CompileService::CompileCode(uint64_t RIP) {
-    WorkItem* ResultItem = nullptr;
-
-    {
-      // Tell the worker thread to compile code for us
-      auto Item = std::make_unique<WorkItem>();
-      Item->RIP = RIP;
-
-      // Fill the threads work queue
-      std::scoped_lock lk(QueueMutex);
-      ResultItem = WorkQueue.emplace(std::move(Item)).get();
-    }
-
-    // Notify the thread that it has more work
-    StartWork.NotifyAll();
-
-    return ResultItem;
-  }
-
-  void CompileService::ExecutionThread() {
-    // Set our thread name so we can see its relation
-    char ThreadName[16]{};
-    snprintf(ThreadName, 16, "%ld-CS", ParentThread->ThreadManager.TID.load());
-    pthread_setname_np(pthread_self(), ThreadName);
-
-    while (true) {
-      // Wait for work
-      StartWork.Wait();
-      if (ShuttingDown.load()) {
-        break;
-      }
-
-      std::scoped_lock lk(CompileMutex);
-      size_t WorkItems{};
-
-      do {
-        // Grab a work item
-        std::unique_ptr<WorkItem> Item{};
-        {
-          std::scoped_lock lk(QueueMutex);
-          WorkItems = WorkQueue.size();
-          if (WorkItems != 0) {
-            Item = std::move(WorkQueue.front());
-            WorkQueue.pop();
-          }
-        }
-
-        // If we had a work item then work on it
-        if (Item) {
-          // Make sure it's not in lookup cache by accident
-          LOGMAN_THROW_A_FMT(CompileThreadData->LookupCache->FindBlock(Item->RIP) == 0, "Compile Service must never have entries in the LookupCache");
-
-          // Code isn't in cache, compile now
-          // Set our thread state's RIP
-          CompileThreadData->CurrentFrame->State.rip = Item->RIP;
-
-          auto [CodePtr, IRList, DebugData, RAData, Generated, StartAddr, Length] = CTX->CompileCode(CompileThreadData.get(), Item->RIP);
-
-          LOGMAN_THROW_A_FMT(Generated == true, "Compile Service doesn't have IR Cache");
-
-          if (!CodePtr) {
-            // XXX: We currently have the expectation that compile service code will be significantly smaller than regular thread's code
-            ERROR_AND_DIE_FMT("Couldn't compile code for thread at RIP: 0x{:x}", Item->RIP);
-          }
-
-          Item->CodePtr = CodePtr;
-          Item->IRList = IRList;
-          Item->DebugData = DebugData;
-          Item->RAData = RAData;
-          Item->StartAddr = StartAddr;
-          Item->Length = Length;
-
-          auto& GCItem = GCArray.emplace_back(std::move(Item));
-          GCItem->ServiceWorkDone.NotifyAll();
-        }
-      } while (WorkItems != 0);
-
-      // Clean up any safe entries in our GC array if we have any.
-      std::erase_if(GCArray, [](const auto& Entry) {
-        return Entry->SafeToClear.load(std::memory_order_relaxed);
-      });
-    }
-  }
-}
@@ -1,68 +0,0 @@
-#pragma once
-
-#include <FEXCore/Debug/InternalThreadState.h>
-#include <FEXCore/Utils/Event.h>
-#include <FEXCore/Utils/Threads.h>
-
-#include <atomic>
-#include <memory>
-#include <mutex>
-#include <queue>
-#include <stdint.h>
-#include <vector>
-
-namespace FEXCore {
-namespace Context {
-  struct Context;
-}
-namespace IR {
-  class IRListView;
-  class RegisterAllocationData;
-};
-class CompileService final {
-  public:
-    CompileService(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread);
-    void Initialize();
-    void Shutdown();
-
-    struct WorkItem {
-      // Incoming
-      uint64_t RIP{};
-
-      // Outgoing
-      void *CodePtr{};
-      FEXCore::IR::IRListView *IRList{};
-      FEXCore::IR::RegisterAllocationData *RAData{};
-      FEXCore::Core::DebugData *DebugData{};
-      uint64_t StartAddr;
-      uint64_t Length;
-
-      // Communication
-      Event ServiceWorkDone{};
-      std::atomic_bool SafeToClear{};
-    };
-
-    WorkItem *CompileCode(uint64_t RIP);
-    void ClearCache(FEXCore::Core::InternalThreadState *Thread);
-
-    // Public for threading
-    void ExecutionThread();
-
-    bool IsAddressInJITCode(uint64_t Address) const {
-      return CompileThreadData->CPUBackend->IsAddressInJITCode(Address, false, false);
-    }
-  private:
-    FEXCore::Context::Context *CTX;
-    FEXCore::Core::InternalThreadState *ParentThread;
-
-    std::unique_ptr<FEXCore::Threads::Thread> WorkerThread;
-    std::unique_ptr<FEXCore::Core::InternalThreadState> CompileThreadData;
-
-    std::mutex QueueMutex{};
-    std::mutex CompileMutex{};
-    std::queue<std::unique_ptr<WorkItem>> WorkQueue{};
-    std::vector<std::unique_ptr<WorkItem>> GCArray{};
-    Event StartWork{};
-    std::atomic_bool ShuttingDown{false};
-};
-}
@@ -1,23 +0,0 @@
-#pragma once
-
-namespace FEXCore {
-  class CodeLoader;
-}
-
-namespace FEXCore::Context {
-  struct Context;
-}
-
-namespace FEXCore::CPU {
-
-  /**
-   * @brief Create the CPU core backend for the context passed in
-   *
-   * @param CTX
-   *
-   * @return true if core was able to be create
-   */
-  bool CreateCPUCore(FEXCore::Context::Context *CTX);
-
-  bool LoadCode(FEXCore::Context::Context *CTX, FEXCore::CodeLoader *Loader);
-}
@@ -1,6 +1,6 @@
+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
 #include "Interface/Core/LookupCache.h"

-#include "Interface/Core/ArchHelpers/MContext.h"
 #include "Interface/Core/Dispatcher/Arm64Dispatcher.h"
 #include "Interface/Core/Interpreter/InterpreterClass.h"

@@ -11,39 +11,49 @@
 #include <FEXCore/Core/CoreState.h>
 #include <FEXCore/Core/X86Enums.h>
 #include <FEXCore/Debug/InternalThreadState.h>
+#include <FEXCore/fextl/memory.h>
 #include <FEXHeaderUtils/Syscalls.h>

 #include <array>
 #include <bit>
 #include <cmath>
+#include <cstddef>
 #include <cstdint>
 #include <memory>
-#include <stddef.h>

-#include "aarch64/assembler-aarch64.h"
-#include "aarch64/constants-aarch64.h"
-#include "aarch64/operands-aarch64.h"
-#include "aarch64/cpu-aarch64.h"
-#include "code-buffer-vixl.h"
-#include "platform-vixl.h"
+#include <aarch64/assembler-aarch64.h>
+#include <aarch64/constants-aarch64.h>
+#include <aarch64/cpu-aarch64.h>
+#include <aarch64/operands-aarch64.h>
+#include <code-buffer-vixl.h>
+#include <platform-vixl.h>

-#include <sys/syscall.h>
 #include <unistd.h>

 namespace FEXCore::CPU {

-using namespace vixl;
-using namespace vixl::aarch64;
+constexpr size_t MAX_DISPATCHER_CODE_SIZE = 4096;

-static constexpr size_t MAX_DISPATCHER_CODE_SIZE = 4096;
-#define STATE x28
+Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::ContextImpl *ctx, const DispatcherConfig &config)
+  : FEXCore::CPU::Dispatcher(ctx, config), Arm64Emitter(ctx, MAX_DISPATCHER_CODE_SIZE)
+#ifdef VIXL_SIMULATOR
+  , Simulator {&Decoder}
+#endif
+{
+#ifdef VIXL_SIMULATOR
+  // Hardcode a 256-bit vector width if we are running in the simulator.
+  Simulator.SetVectorLengthInBits(256);
+#endif

-Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, DispatcherConfig &config)
-  : Dispatcher(ctx, Thread), Arm64Emitter(ctx, MAX_DISPATCHER_CODE_SIZE) {
-  SRAEnabled = config.StaticRegisterAssignment;
-  SetAllowAssembler(true);
+  EmitDispatcher();
+}

-  DispatchPtr = GetCursorAddress<CPUBackend::AsmDispatch>();
+void Arm64Dispatcher::EmitDispatcher() {
+#ifdef VIXL_DISASSEMBLER
+  const auto DisasmBegin = GetCursorAddress<const vixl::aarch64::Instruction*>();
+#endif
+
+  DispatchPtr = GetCursorAddress<AsmDispatch>();

  // while (true) {
  //    Ptr = FindBlock(RIP)
@@ -53,15 +63,9 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
  //    Ptr();
  // }

-  uint64_t VirtualMemorySize = Thread->LookupCache->GetVirtualMemorySize();
-  Literal l_VirtualMemory {VirtualMemorySize};
-  Literal l_PagePtr {Thread->LookupCache->GetPagePointer()};
-  Literal l_L1Ptr {Thread->LookupCache->GetL1Pointer()};
-  Literal l_CTX {reinterpret_cast<uintptr_t>(CTX)};
-  Literal l_Sleep {reinterpret_cast<uint64_t>(SleepThread)};
-  Literal l_CompileBlock {GetCompileBlockPtr()};
-  Literal l_ExitFunctionLink {config.ExitFunctionLink};
-  Literal l_ExitFunctionLinkThis {config.ExitFunctionLinkThis};
+  ARMEmitter::ForwardLabel l_CTX;
+  ARMEmitter::ForwardLabel l_Sleep;
+  ARMEmitter::ForwardLabel l_CompileBlock;

  // Push all the register we need to save
  PushCalleeSavedRegisters();
@@ -69,135 +73,110 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
  // Push our memory base to the correct register
  // Move our thread pointer to the correct register
  // This is passed in to parameter 0 (x0)
-  mov(STATE, x0);
+  mov(STATE, ARMEmitter::XReg::x0);

  // Save this stack pointer so we can cleanly shutdown the emulation with a long jump
  // regardless of where we were in the stack
-  add(x0, sp, 0);
-  str(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, ReturningStackLocation)));
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, ARMEmitter::Reg::rsp, 0);
+  str(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, ReturningStackLocation));

  AbsoluteLoopTopAddressFillSRA = GetCursorAddress<uint64_t>();

-  if (SRAEnabled) {
+  if (config.StaticRegisterAllocation) {
    FillStaticRegs();
  }

  // We want to ensure that we are 16 byte aligned at the top of this loop
  Align16B();
-  aarch64::Label FullLookup{};
-  aarch64::Label CallBlock{};
-  aarch64::Label LoopTop{};
-  aarch64::Label ExitSpillSRA{};
-  aarch64::Label ThreadPauseHandler{};
+  ARMEmitter::BiDirectionalLabel FullLookup{};
+  ARMEmitter::BiDirectionalLabel CallBlock{};
+  ARMEmitter::BackwardLabel LoopTop{};

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

  // Load in our RIP
  // Don't modify x2 since it contains our RIP once the block doesn't exist
-  ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip)));
-  auto RipReg = x2;
+
+  auto RipReg = ARMEmitter::XReg::x2;
+  ldr(RipReg, STATE_PTR(CpuStateFrame, State.rip));

  // L1 Cache
-  ldr(x0, &l_L1Ptr);
+  ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));

-  and_(x3, RipReg, LookupCache::L1_ENTRIES_MASK);
-  add(x0, x0, Operand(x3, Shift::LSL, 4));
-  ldp(x3, x0, MemOperand(x0));
-  cmp(x0, RipReg);
-  b(&FullLookup, Condition::ne);
+  and_(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, RipReg.R(), LookupCache::L1_ENTRIES_MASK);
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, ARMEmitter::Reg::r0, ARMEmitter::Reg::r3, ARMEmitter::ShiftType::LSL , 4);
+  ldp<ARMEmitter::IndexType::OFFSET>(ARMEmitter::XReg::x3, ARMEmitter::XReg::x0, ARMEmitter::Reg::r0, 0);
+  cmp(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, RipReg.R());
+  b(ARMEmitter::Condition::CC_NE, &FullLookup);

-  if (!config.ExecuteBlocksWithCall) {
-    br(x3);
-  } else {
-    b(&CallBlock);
-  }
+  br(ARMEmitter::Reg::r3);

  // L1C check failed, do a full lookup
-  bind(&FullLookup);
+  Bind(&FullLookup);

  // This is the block cache lookup routine
  // It matches what is going on it LookupCache.h::FindBlock
-  ldr(x0, &l_PagePtr);
+  ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Pointers.Common.L2Pointer));

  // Mask the address by the virtual address size so we can check for aliases
+  uint64_t VirtualMemorySize = CTX->Config.VirtualMemSize;
  if (std::popcount(VirtualMemorySize) == 1) {
-    and_(x3, RipReg, Thread->LookupCache->GetVirtualMemorySize() - 1);
+    and_(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, RipReg.R(), VirtualMemorySize - 1);
  }
  else {
-    ldr(x3, &l_VirtualMemory);
-    and_(x3, RipReg, x3);
+    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, VirtualMemorySize);
+    and_(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, RipReg.R(), ARMEmitter::Reg::r3);
  }

-  aarch64::Label NoBlock;
+  ARMEmitter::ForwardLabel NoBlock;
+
  {
    // Offset the address and add to our page pointer
-    lsr(x1, x3, 12);
+    lsr(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, ARMEmitter::Reg::r3, 12);

    // Load the pointer from the offset
-    ldr(x0, MemOperand(x0, x1, Shift::LSL, 3));
+    ldr(ARMEmitter::XReg::x0, ARMEmitter::Reg::r0, ARMEmitter::Reg::r1, ARMEmitter::ExtendedType::LSL_64, 3);

    // If page pointer is zero then we have no block
-    cbz(x0, &NoBlock);
+    cbz(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, &NoBlock);

    // Steal the page offset
-    and_(x1, x3, 0x0FFF);
+    and_(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, ARMEmitter::Reg::r3, 0x0FFF);

    // Shift the offset by the size of the block cache entry
-    add(x0, x0, Operand(x1, Shift::LSL, (int)log2(sizeof(FEXCore::LookupCache::LookupCacheEntry))));
+    add(ARMEmitter::XReg::x0, ARMEmitter::XReg::x0, ARMEmitter::XReg::x1, ARMEmitter::ShiftType::LSL, (int)log2(sizeof(FEXCore::LookupCache::LookupCacheEntry)));

-    // Load the guest address first to ensure it maps to the address we are currently at
+    // The the full LookupCacheEntry with a single LDP.
+    // Check the guest address first to ensure it maps to the address we are currently at.
    // This fixes aliasing problems
-    ldr(x1, MemOperand(x0, offsetof(FEXCore::LookupCache::LookupCacheEntry, GuestCode)));
-    cmp(x1, RipReg);
-    b(&NoBlock, Condition::ne);
+    ldp<ARMEmitter::IndexType::OFFSET>(ARMEmitter::XReg::x3, ARMEmitter::XReg::x1, ARMEmitter::Reg::r0, 0);

-    // Now load the actual host block to execute if we can
-    ldr(x3, MemOperand(x0, offsetof(FEXCore::LookupCache::LookupCacheEntry, HostCode)));
-    cbz(x3, &NoBlock);
+    // If the guest address doesn't match, Compile the block.
+    cmp(ARMEmitter::XReg::x1, RipReg);
+    b(ARMEmitter::Condition::CC_NE, &NoBlock);
+
+    // Check the host address to see if it matches, else compile the block.
+    cbz(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, &NoBlock);

    // If we've made it here then we have a real compiled block
    {
      // update L1 cache
-      ldr(x0, &l_L1Ptr);
+      ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));

-      and_(x1, RipReg, LookupCache::L1_ENTRIES_MASK);
-      add(x0, x0, Operand(x1, Shift::LSL, 4));
-      stp(x3, x2, MemOperand(x0));
+      and_(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, RipReg.R(), LookupCache::L1_ENTRIES_MASK);
+      add(ARMEmitter::XReg::x0, ARMEmitter::XReg::x0, ARMEmitter::XReg::x1, ARMEmitter::ShiftType::LSL, 4);
+      stp<ARMEmitter::IndexType::OFFSET>(ARMEmitter::XReg::x3, ARMEmitter::XReg::x2, ARMEmitter::Reg::r0);

      // Jump to the block
-      if (!config.ExecuteBlocksWithCall) {
-        br(x3);
-      } else {
-        bind(&CallBlock);
-        mov(x0, STATE);
-        blr(x3);
-
-        if (CTX->GetGdbServerStatus()) {
-          // If we have a gdb server running then run in a less efficient mode that checks if we need to exit
-          // This happens when single stepping
-
-          static_assert(sizeof(CTX->Config.RunningMode) == 4, "This is expected to be size of 4");
-          ldr(x0, &l_CTX);
-          ldr(w0, MemOperand(x0, offsetof(FEXCore::Context::Context, Config.RunningMode)));
-          // If the value == 0 then branch to the top
-          cbz(x0, &LoopTop);
-          // Else we need to pause now
-          b(&ThreadPauseHandler);
-        } else {
-          // Unconditionally loop to the top
-          // We will only stop on error when compiling a block or signal
-          b(&LoopTop);
-        }
-      }
+      br(ARMEmitter::Reg::r3);
    }
  }

  {
-    bind(&ExitSpillSRA);
    ThreadStopHandlerAddressSpillSRA = GetCursorAddress<uint64_t>();
-    if (SRAEnabled)
-      SpillStaticRegs();
+    if (config.StaticRegisterAllocation)
+      SpillStaticRegs(TMP1);

    ThreadStopHandlerAddress = GetCursorAddress<uint64_t>();

@@ -208,115 +187,72 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
    ret();
  }

-  constexpr bool SignalSafeCompile = true;
  {
    ExitFunctionLinkerAddress = GetCursorAddress<uint64_t>();
-    if (SRAEnabled)
-      SpillStaticRegs();
+    if (config.StaticRegisterAllocation)
+      SpillStaticRegs(TMP1);

-    if (SignalSafeCompile) {
-      // When compiling code, mask all signals to reduce the chance of reentrant allocations
-      // Args:
-      // X0: SETMASK
-      // X1: Pointer to mask value (uint64_t)
-      // X2: Pointer to old mask value (uint64_t)
-      // X3: Size of mask, sizeof(uint64_t)
-      // X8: Syscall
+    ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CPUState, DeferredSignalRefCount));
+    add(ARMEmitter::Size::i64Bit, ARMEmitter::XReg::x0, ARMEmitter::XReg::x0, 1);
+    str(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CPUState, DeferredSignalRefCount));

-      LoadConstant(x0, ~0ULL);
-      stp(x0, x0, MemOperand(sp, -16, PreIndex));
-      LoadConstant(x0, SIG_SETMASK);
-      add(x1, sp, 0);
-      add(x2, sp, 0);
-      LoadConstant(x3, 8);
-      LoadConstant(x8, SYS_rt_sigprocmask);
-      svc(0);
-    }
+    mov(ARMEmitter::XReg::x0, STATE);
+    mov(ARMEmitter::XReg::x1, ARMEmitter::XReg::lr);

-    ldr(x0, &l_ExitFunctionLinkThis);
-    mov(x1, STATE);
-    mov(x2, lr);
+    ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionLink));
+#ifdef VIXL_SIMULATOR
+    GenerateIndirectRuntimeCall<uintptr_t, void *, void *>(ARMEmitter::Reg::r2);
+#else
+    blr(ARMEmitter::Reg::r2);
+#endif

-    ldr(x3, &l_ExitFunctionLink);
-    blr(x3);
-
-    if (SignalSafeCompile) {
-      // Now restore the signal mask
-      // Living in the same location
-
-      mov(x4, x0);
-      LoadConstant(x0, SIG_SETMASK);
-      add(x1, sp, 0);
-      LoadConstant(x2, 0);
-      LoadConstant(x3, 8);
-      LoadConstant(x8, SYS_rt_sigprocmask);
-      svc(0);
-
-      // Bring stack back
-      add(sp, sp, 16);
-
-      mov(x0, x4);
-    }
-
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
      FillStaticRegs();
-    br(x0);
+
+    ldr(ARMEmitter::XReg::x1, STATE, offsetof(FEXCore::Core::CPUState, DeferredSignalRefCount));
+    subs(ARMEmitter::Size::i64Bit, ARMEmitter::XReg::x1, ARMEmitter::XReg::x1, 1);
+    str(ARMEmitter::XReg::x1, STATE, offsetof(FEXCore::Core::CPUState, DeferredSignalRefCount));
+
+    // Trigger segfault if any deferred signals are pending
+    ldr(ARMEmitter::XReg::x1, STATE, offsetof(FEXCore::Core::CPUState, DeferredSignalFaultAddress));
+    str(ARMEmitter::XReg::zr, ARMEmitter::XReg::x1, 0);
+
+    br(ARMEmitter::Reg::r0);
  }

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

-    if (SRAEnabled)
-      SpillStaticRegs();
+    if (config.StaticRegisterAllocation)
+      SpillStaticRegs(TMP1);

-    if (SignalSafeCompile) {
-      // When compiling code, mask all signals to reduce the chance of reentrant allocations
-      // Args:
-      // X0: SETMASK
-      // X1: Pointer to mask value (uint64_t)
-      // X2: Pointer to old mask value (uint64_t)
-      // X3: Size of mask, sizeof(uint64_t)
-      // X8: Syscall
+    ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CPUState, DeferredSignalRefCount));
+    add(ARMEmitter::Size::i64Bit, ARMEmitter::XReg::x0, ARMEmitter::XReg::x0, 1);
+    str(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CPUState, DeferredSignalRefCount));

-      LoadConstant(x0, ~0ULL);
-      stp(x0, x2, MemOperand(sp, -16, PreIndex));
-      LoadConstant(x0, SIG_SETMASK);
-      add(x1, sp, 0);
-      add(x2, sp, 0);
-      LoadConstant(x3, 8);
-      LoadConstant(x8, SYS_rt_sigprocmask);
-      svc(0);
-
-      // Reload x2 to bring back RIP
-      ldr(x2, MemOperand(sp, 8, Offset));
-    }
-
-    ldr(x0, &l_CTX);
-    mov(x1, STATE);
-    ldr(x3, &l_CompileBlock);
+    ldr(ARMEmitter::XReg::x0, &l_CTX);
+    mov(ARMEmitter::XReg::x1, STATE);
+    ldr(ARMEmitter::XReg::x3, &l_CompileBlock);

    // X2 contains our guest RIP
-    blr(x3); // { CTX, Frame, RIP}
+#ifdef VIXL_SIMULATOR
+    GenerateIndirectRuntimeCall<void, void *, uint64_t, void *>(ARMEmitter::Reg::r3);
+#else
+    blr(ARMEmitter::Reg::r3); // { CTX, Frame, RIP}
+#endif

-
-    if (SignalSafeCompile) {
-      // Now restore the signal mask
-      // Living in the same location
-      LoadConstant(x0, SIG_SETMASK);
-      add(x1, sp, 0);
-      LoadConstant(x2, 0);
-      LoadConstant(x3, 8);
-      LoadConstant(x8, SYS_rt_sigprocmask);
-      svc(0);
-
-      // Bring stack back
-      add(sp, sp, 16);
-    }
-
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
      FillStaticRegs();

+    ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CPUState, DeferredSignalRefCount));
+    subs(ARMEmitter::Size::i64Bit, ARMEmitter::XReg::x0, ARMEmitter::XReg::x0, 1);
+    str(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CPUState, DeferredSignalRefCount));
+
+    // Trigger segfault if any deferred signals are pending
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CPUState, DeferredSignalFaultAddress));
+    str(ARMEmitter::XReg::zr, TMP1, 0);
+
    b(&LoopTop);
  }

@@ -328,57 +264,78 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
    hlt(0);
  }

+  {
+    SignalHandlerReturnAddressRT = GetCursorAddress<uint64_t>();
+
+    // Now to get back to our old location we need to do a fault dance
+    // We can't use SIGTRAP here since gdb catches it and never gives it to the application!
+    hlt(0);
+  }
+
  {
    // Guest SIGILL handler
    // Needs to be distinct from the SignalHandlerReturnAddress
-    UnimplementedInstructionAddress = GetCursorAddress<uint64_t>();
+    GuestSignal_SIGILL = GetCursorAddress<uint64_t>();

-    if (SRAEnabled)
-      SpillStaticRegs();
+    if (config.StaticRegisterAllocation)
+      SpillStaticRegs(TMP1);

    hlt(0);
  }

+  {
+    // Guest SIGTRAP handler
+    // Needs to be distinct from the SignalHandlerReturnAddress
+    GuestSignal_SIGTRAP  = GetCursorAddress<uint64_t>();
+
+    if (config.StaticRegisterAllocation)
+      SpillStaticRegs(TMP1);
+
+    brk(0);
+  }
+
  {
    // Guest Overflow handler
    // Needs to be distinct from the SignalHandlerReturnAddress
-    OverflowExceptionInstructionAddress = GetCursorAddress<uint64_t>();
+    GuestSignal_SIGSEGV = GetCursorAddress<uint64_t>();

-    if (SRAEnabled)
-      SpillStaticRegs();
-
-    LoadConstant(x0, reinterpret_cast<uint64_t>(&SynchronousFaultData));
-    LoadConstant(w1, 1);
-    strb(w1, MemOperand(x0, offsetof(Dispatcher::SynchronousFaultDataStruct, FaultToTopAndGeneratedException)));
-    LoadConstant(w1, X86State::X86_TRAPNO_OF);
-    str(w1, MemOperand(x0, offsetof(Dispatcher::SynchronousFaultDataStruct, TrapNo)));
-    LoadConstant(w1, 0x80);
-    str(w1, MemOperand(x0, offsetof(Dispatcher::SynchronousFaultDataStruct, si_code)));
-    LoadConstant(x1, 0);
-    str(w1, MemOperand(x0, offsetof(Dispatcher::SynchronousFaultDataStruct, err_code)));
+    if (config.StaticRegisterAllocation)
+      SpillStaticRegs(TMP1);

    // hlt/udf = SIGILL
    // brk = SIGTRAP
    // ??? = SIGSEGV
    // Force a SIGSEGV by loading zero
-    ldr(x1, MemOperand(x1));
+    if (CTX->ExitOnHLTEnabled()) {
+      ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, ReturningStackLocation));
+      add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::r0, 0);
+      PopCalleeSavedRegisters();
+      ret();
+    }
+    else {
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, 0);
+      ldr(ARMEmitter::XReg::x1, ARMEmitter::Reg::r1);
+    }
  }

  {
    ThreadPauseHandlerAddressSpillSRA = GetCursorAddress<uint64_t>();
-    if (SRAEnabled)
-      SpillStaticRegs();
+    if (config.StaticRegisterAllocation)
+      SpillStaticRegs(TMP1);

-    bind(&ThreadPauseHandler);
    ThreadPauseHandlerAddress = GetCursorAddress<uint64_t>();
    // We are pausing, this means the frontend should be waiting for this thread to idle
    // We will have faulted and jumped to this location at this point

    // Call our sleep handler
-    ldr(x0, &l_CTX);
-    mov(x1, STATE);
-    ldr(x2, &l_Sleep);
-    blr(x2);
+    ldr(ARMEmitter::XReg::x0, &l_CTX);
+    mov(ARMEmitter::XReg::x1, STATE);
+    ldr(ARMEmitter::XReg::x2, &l_Sleep);
+#ifdef VIXL_SIMULATOR
+    GenerateIndirectRuntimeCall<void, void *, void *>(ARMEmitter::Reg::r2);
+#else
+    blr(ARMEmitter::Reg::r2);
+#endif

    PauseReturnInstruction = GetCursorAddress<uint64_t>();
    // Fault to start running again
@@ -401,97 +358,254 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
    // On return to the thunk, the thunk can get whatever its return value is from the thread context depending on ABI handling on its end
    // When the thunk itself returns, it'll do its regular return logic there
    // void ReentrantCallback(FEXCore::Core::InternalThreadState *Thread, uint64_t RIP);
-    CallbackPtr = GetCursorAddress<CPUBackend::JITCallback>();
+    CallbackPtr = GetCursorAddress<JITCallback>();

    // We expect the thunk to have previously pushed the registers it was using
    PushCalleeSavedRegisters();

    // First thing we need to move the thread state pointer back in to our register
-    mov(STATE, x0);
+    mov(STATE, ARMEmitter::XReg::x0);

    // Make sure to adjust the refcounter so we don't clear the cache now
-    LoadConstant(x0, reinterpret_cast<uint64_t>(&SignalHandlerRefCounter));
-    ldr(w2, MemOperand(x0));
-    add(w2, w2, 1);
-    str(w2, MemOperand(x0));
+    ldr(ARMEmitter::WReg::w2, STATE_PTR(CpuStateFrame, SignalHandlerRefCounter));
+    add(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r2, 1);
+    str(ARMEmitter::WReg::w2, STATE_PTR(CpuStateFrame, SignalHandlerRefCounter));

    // Now push the callback return trampoline to the guest stack
    // Guest will be misaligned because calling a thunk won't correct the guest's stack once we call the callback from the host
-    LoadConstant(x0, CTX->X86CodeGen.CallbackReturn);
+    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, CTX->X86CodeGen.CallbackReturn);

-    ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])));
-    sub(x2, x2, 16);
-    str(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])));
+    ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]));
+    sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r2, 16);
+    str(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]));

    // Store the trampoline to the guest stack
    // Guest stack is now correctly misaligned after a regular call instruction
-    str(x0, MemOperand(x2));
+    str(ARMEmitter::XReg::x0, ARMEmitter::Reg::r2, 0);

    // Store RIP to the context state
-    str(x1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip)));
+    str(ARMEmitter::XReg::x1, STATE_PTR(CpuStateFrame, State.rip));

    // load static regs
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
      FillStaticRegs();

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

-  place(&l_VirtualMemory);
-  place(&l_PagePtr);
-  place(&l_L1Ptr);
-  place(&l_CTX);
-  place(&l_Sleep);
-  place(&l_CompileBlock);
-  place(&l_ExitFunctionLink);
-  place(&l_ExitFunctionLinkThis);
+  {
+    LUDIVHandlerAddress = GetCursorAddress<uint64_t>();

+    PushDynamicRegsAndLR(ARMEmitter::Reg::r3);
+    SpillStaticRegs(ARMEmitter::Reg::r3);
+
+    ldr(ARMEmitter::XReg::x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LUDIV));
+#ifdef VIXL_SIMULATOR
+    GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t>(ARMEmitter::Reg::r3);
+#else
+    blr(ARMEmitter::Reg::r3);
+#endif
+    FillStaticRegs();
+
+    // Result is now in x0
+    // Fix the stack and any values that were stepped on
+    PopDynamicRegsAndLR();
+
+    // Go back to our code block
+    ret();
+  }
+
+  {
+    LDIVHandlerAddress = GetCursorAddress<uint64_t>();
+
+    PushDynamicRegsAndLR(ARMEmitter::Reg::r3);
+    SpillStaticRegs(ARMEmitter::Reg::r3);
+
+    ldr(ARMEmitter::XReg::x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LDIV));
+#ifdef VIXL_SIMULATOR
+    GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t>(ARMEmitter::Reg::r3);
+#else
+    blr(ARMEmitter::Reg::r3);
+#endif
+    FillStaticRegs();
+
+    // Result is now in x0
+    // Fix the stack and any values that were stepped on
+    PopDynamicRegsAndLR();
+
+    // Go back to our code block
+    ret();
+  }
+
+  {
+    LUREMHandlerAddress = GetCursorAddress<uint64_t>();
+
+    PushDynamicRegsAndLR(ARMEmitter::Reg::r3);
+    SpillStaticRegs(ARMEmitter::Reg::r3);
+
+    ldr(ARMEmitter::XReg::x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LUREM));
+#ifdef VIXL_SIMULATOR
+    GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t>(ARMEmitter::Reg::r3);
+#else
+    blr(ARMEmitter::Reg::r3);
+#endif
+    FillStaticRegs();
+
+    // Result is now in x0
+    // Fix the stack and any values that were stepped on
+    PopDynamicRegsAndLR();
+
+    // Go back to our code block
+    ret();
+  }
+
+  {
+    LREMHandlerAddress = GetCursorAddress<uint64_t>();
+
+    PushDynamicRegsAndLR(ARMEmitter::Reg::r3);
+    SpillStaticRegs(ARMEmitter::Reg::r3);
+
+    ldr(ARMEmitter::XReg::x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LREM));
+
+#ifdef VIXL_SIMULATOR
+    GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t>(ARMEmitter::Reg::r3);
+#else
+    blr(ARMEmitter::Reg::r3);
+#endif
+    FillStaticRegs();
+
+    // Result is now in x0
+    // Fix the stack and any values that were stepped on
+    PopDynamicRegsAndLR();
+
+    // Go back to our code block
+    ret();
+  }
+
+  Bind(&l_CTX);
+  dc64(reinterpret_cast<uintptr_t>(CTX));
+  Bind(&l_Sleep);
+  dc64(reinterpret_cast<uint64_t>(SleepThread));
+  Bind(&l_CompileBlock);
+  dc64(GetCompileBlockPtr());

-  FinalizeCode();
  Start = reinterpret_cast<uint64_t>(DispatchPtr);
  End = GetCursorAddress<uint64_t>();
-  vixl::aarch64::CPU::EnsureIAndDCacheCoherency(reinterpret_cast<void*>(DispatchPtr), End - reinterpret_cast<uint64_t>(DispatchPtr));
-  GetBuffer()->SetExecutable();
+  ClearICache(reinterpret_cast<void*>(DispatchPtr), End - reinterpret_cast<uint64_t>(DispatchPtr));

  if (CTX->Config.BlockJITNaming()) {
-    std::string Name = "Dispatch_" + std::to_string(FHU::Syscalls::gettid());
+    fextl::string Name = fextl::fmt::format("Dispatch_{}", FHU::Syscalls::gettid());
    CTX->Symbols.Register(reinterpret_cast<void*>(DispatchPtr), End - reinterpret_cast<uint64_t>(DispatchPtr), Name);
  }
  if (CTX->Config.GlobalJITNaming()) {
    CTX->Symbols.RegisterJITSpace(reinterpret_cast<void*>(DispatchPtr), End - reinterpret_cast<uint64_t>(DispatchPtr));
  }
+
+#ifdef VIXL_DISASSEMBLER
+  if (Disassemble() & FEXCore::Config::Disassemble::DISPATCHER) {
+    const auto DisasmEnd = GetCursorAddress<const vixl::aarch64::Instruction*>();
+    Disasm.DisassembleBuffer(DisasmBegin, DisasmEnd);
+  }
+#endif
 }

-void Arm64Dispatcher::SpillSRA(void *ucontext, uint32_t IgnoreMask) {
-  for(int i = 0; i < SRA64.size(); i++) {
-    if (IgnoreMask & (1U << SRA64[i].GetCode())) {
-      // Skip this one, it's already spilled
-      continue;
-    }
-    ThreadState->CurrentFrame->State.gregs[i] = ArchHelpers::Context::GetArmReg(ucontext, SRA64[i].GetCode());
-  }
-
-  for(int i = 0; i < SRAFPR.size(); i++) {
-    auto FPR = ArchHelpers::Context::GetArmFPR(ucontext, SRAFPR[i].GetCode());
-    memcpy(&ThreadState->CurrentFrame->State.xmm[i][0], &FPR, sizeof(__uint128_t));
-  }
+#ifdef VIXL_SIMULATOR
+void Arm64Dispatcher::ExecuteDispatch(FEXCore::Core::CpuStateFrame *Frame) {
+  Simulator.WriteXRegister(0, reinterpret_cast<int64_t>(Frame));
+  Simulator.RunFrom(reinterpret_cast<vixl::aarch64::Instruction const*>(DispatchPtr));
 }

-#ifdef _M_ARM_64
-
-void InterpreterCore::CreateAsmDispatch(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread) {
-  DispatcherConfig config;
-  config.ExecuteBlocksWithCall = true;
-
-  Dispatcher = std::make_unique<Arm64Dispatcher>(ctx, Thread, config);
-  DispatchPtr = Dispatcher->DispatchPtr;
-  CallbackPtr = Dispatcher->CallbackPtr;
-
-  // TODO: It feels wrong to initialize this way
-  ctx->InterpreterCallbackReturn = Dispatcher->ReturnPtr;
+void Arm64Dispatcher::ExecuteJITCallback(FEXCore::Core::CpuStateFrame *Frame, uint64_t RIP) {
+  Simulator.WriteXRegister(0, reinterpret_cast<int64_t>(Frame));
+  Simulator.WriteXRegister(1, RIP);
+  Simulator.RunFrom(reinterpret_cast<vixl::aarch64::Instruction const*>(CallbackPtr));
 }

 #endif

+size_t Arm64Dispatcher::GenerateGDBPauseCheck(uint8_t *CodeBuffer, uint64_t GuestRIP) {
+  FEXCore::ARMEmitter::Emitter emit{CodeBuffer, MaxGDBPauseCheckSize};
+
+  ARMEmitter::ForwardLabel RunBlock;
+
+  // If we have a gdb server running then run in a less efficient mode that checks if we need to exit
+  // This happens when single stepping
+
+  static_assert(sizeof(FEXCore::Context::ContextImpl::Config.RunningMode) == 4, "This is expected to be size of 4");
+  emit.ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Thread));
+  emit.ldr(ARMEmitter::XReg::x0, ARMEmitter::Reg::r0, offsetof(FEXCore::Core::InternalThreadState, CTX)); // Get Context
+  emit.ldr(ARMEmitter::WReg::w0, ARMEmitter::Reg::r0, offsetof(FEXCore::Context::ContextImpl, Config.RunningMode));
+
+  // If the value == 0 then we don't need to stop
+  emit.cbz(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r0, &RunBlock);
+  {
+    ARMEmitter::ForwardLabel l_GuestRIP;
+    // Make sure RIP is syncronized to the context
+    emit.ldr(ARMEmitter::XReg::x0, &l_GuestRIP);
+    emit.str(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, State.rip));
+
+    // Stop the thread
+    emit.ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Pointers.Common.ThreadPauseHandlerSpillSRA));
+    emit.br(ARMEmitter::Reg::r0);
+    emit.Bind(&l_GuestRIP);
+    emit.dc64(GuestRIP);
+  }
+  emit.Bind(&RunBlock);
+
+  auto UsedBytes = emit.GetCursorOffset();
+  emit.ClearICache(CodeBuffer, UsedBytes);
+  return UsedBytes;
+}
+
+size_t Arm64Dispatcher::GenerateInterpreterTrampoline(uint8_t *CodeBuffer) {
+  LOGMAN_THROW_AA_FMT(!config.StaticRegisterAllocation, "GenerateInterpreterTrampoline dispatcher does not support SRA");
+
+  FEXCore::ARMEmitter::Emitter emit{CodeBuffer, MaxInterpreterTrampolineSize};
+  ARMEmitter::ForwardLabel InlineIRData;
+
+  emit.mov(ARMEmitter::XReg::x0, STATE);
+  emit.adr(ARMEmitter::Reg::r1, &InlineIRData);
+
+  emit.ldr(ARMEmitter::XReg::x3, STATE_PTR(CpuStateFrame, Pointers.Interpreter.FragmentExecuter));
+  emit.blr(ARMEmitter::Reg::r3);
+
+  emit.ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Pointers.Common.DispatcherLoopTop));
+  emit.br(ARMEmitter::Reg::r0);
+
+  emit.Bind(&InlineIRData);
+
+  auto UsedBytes = emit.GetCursorOffset();
+  emit.ClearICache(CodeBuffer, UsedBytes);
+  return UsedBytes;
+}
+
+void Arm64Dispatcher::InitThreadPointers(FEXCore::Core::InternalThreadState *Thread) {
+  // Setup dispatcher specific pointers that need to be accessed from JIT code
+  {
+    auto &Common = Thread->CurrentFrame->Pointers.Common;
+
+    Common.DispatcherLoopTop = AbsoluteLoopTopAddress;
+    Common.DispatcherLoopTopFillSRA = AbsoluteLoopTopAddressFillSRA;
+    Common.ExitFunctionLinker = ExitFunctionLinkerAddress;
+    Common.ThreadStopHandlerSpillSRA = ThreadStopHandlerAddressSpillSRA;
+    Common.ThreadPauseHandlerSpillSRA = ThreadPauseHandlerAddressSpillSRA;
+    Common.GuestSignal_SIGILL = GuestSignal_SIGILL;
+    Common.GuestSignal_SIGTRAP = GuestSignal_SIGTRAP;
+    Common.GuestSignal_SIGSEGV = GuestSignal_SIGSEGV;
+    Common.SignalReturnHandler = SignalHandlerReturnAddress;
+    Common.SignalReturnHandlerRT = SignalHandlerReturnAddressRT;
+
+    auto &AArch64 = Thread->CurrentFrame->Pointers.AArch64;
+    AArch64.LUDIVHandler = LUDIVHandlerAddress;
+    AArch64.LDIVHandler = LDIVHandlerAddress;
+    AArch64.LUREMHandler = LUREMHandlerAddress;
+    AArch64.LREMHandler = LREMHandlerAddress;
+  }
+}
+
+fextl::unique_ptr<Dispatcher> Dispatcher::CreateArm64(FEXCore::Context::ContextImpl *CTX, const DispatcherConfig &Config) {
+  return fextl::make_unique<Arm64Dispatcher>(CTX, Config);
+}
+
 }
@@ -3,22 +3,64 @@
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"

-namespace FEXCore::Context {
-struct Context;
-}
+#ifdef VIXL_SIMULATOR
+#include <aarch64/simulator-aarch64.h>
+#endif

 namespace FEXCore::Core {
 struct InternalThreadState;
 }

+#define STATE_PTR(STATE_TYPE, FIELD) \
+  STATE.R(), offsetof(FEXCore::Core::STATE_TYPE, FIELD)
+
 namespace FEXCore::CPU {

 class Arm64Dispatcher final : public Dispatcher, public Arm64Emitter {
  public:
-    Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, DispatcherConfig &config);
+    Arm64Dispatcher(FEXCore::Context::ContextImpl *ctx, const DispatcherConfig &config);
+    void InitThreadPointers(FEXCore::Core::InternalThreadState *Thread) override;
+    size_t GenerateGDBPauseCheck(uint8_t *CodeBuffer, uint64_t GuestRIP) override;
+    size_t GenerateInterpreterTrampoline(uint8_t *CodeBuffer) override;

-  protected:
-    void SpillSRA(void *ucontext, uint32_t IgnoreMask) override;
+#ifdef VIXL_SIMULATOR
+  void ExecuteDispatch(FEXCore::Core::CpuStateFrame *Frame) override;
+  void ExecuteJITCallback(FEXCore::Core::CpuStateFrame *Frame, uint64_t RIP) override;
+#endif
+
+  void EmitDispatcher();
+
+  uint16_t GetSRAGPRCount() const override {
+    return StaticRegisters.size();
+  }
+
+  uint16_t GetSRAFPRCount() const override {
+    return StaticFPRegisters.size();
+  }
+
+  void GetSRAGPRMapping(uint8_t Mapping[16]) const override {
+    for (size_t i = 0; i < StaticRegisters.size(); ++i) {
+      Mapping[i] = StaticRegisters[i].Idx();
+    }
+  }
+
+  void GetSRAFPRMapping(uint8_t Mapping[16]) const override {
+    for (size_t i = 0; i < StaticFPRegisters.size(); ++i) {
+      Mapping[i] = StaticFPRegisters[i].Idx();
+    }
+  }
+
+  private:
+    // Long division helpers
+    uint64_t LUDIVHandlerAddress{};
+    uint64_t LDIVHandlerAddress{};
+    uint64_t LUREMHandlerAddress{};
+    uint64_t LREMHandlerAddress{};
+
+#ifdef VIXL_SIMULATOR
+    vixl::aarch64::Decoder Decoder;
+    vixl::aarch64::Simulator Simulator;
+#endif
 };

 }
@@ -1,13 +1,11 @@

-#include "Interface/Core/ArchHelpers/MContext.h"
+#include "Interface/Context/Context.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Core/X86HelperGen.h"
-#include "Interface/Core/CompileService.h"

 #include <FEXCore/Config/Config.h>
 #include <FEXCore/Core/CoreState.h>
 #include <FEXCore/Core/SignalDelegator.h>
-#include <FEXCore/Core/UContext.h>
 #include <FEXCore/Core/X86Enums.h>
 #include <FEXCore/Debug/InternalThreadState.h>
 #include <FEXCore/Utils/Event.h>
@@ -16,13 +14,13 @@

 #include <atomic>
 #include <condition_variable>
-#include <bits/types/siginfo_t.h>
 #include <csignal>
 #include <cstring>
+#include <signal.h>

 namespace FEXCore::CPU {

-void Dispatcher::SleepThread(FEXCore::Context::Context *ctx, FEXCore::Core::CpuStateFrame *Frame) {
+void Dispatcher::SleepThread(FEXCore::Context::ContextImpl *ctx, FEXCore::Core::CpuStateFrame *Frame) {
  auto Thread = Frame->Thread;

  --ctx->IdleWaitRefCount;
@@ -40,740 +38,16 @@ void Dispatcher::SleepThread(FEXCore::Context::Context *ctx, FEXCore::Core::CpuS
  ctx->IdleWaitCV.notify_all();
 }

-ArchHelpers::Context::ContextBackup* Dispatcher::StoreThreadState(int Signal, void *ucontext) {
-  // We can end up getting a signal at any point in our host state
-  // Jump to a handler that saves all state so we can safely return
-  uint64_t OldSP = ArchHelpers::Context::GetSp(ucontext);
-  uintptr_t NewSP = OldSP;
-
-  size_t StackOffset = sizeof(ArchHelpers::Context::ContextBackup);
-
-  // We need to back up behind the host's red zone
-  // We do this on the guest side as well
-  // (does nothing on arm hosts)
-  NewSP -= ArchHelpers::Context::ContextBackup::RedZoneSize;
-
-  NewSP -= StackOffset;
-  NewSP = AlignDown(NewSP, 16);
-
-  auto Context = reinterpret_cast<ArchHelpers::Context::ContextBackup*>(NewSP);
-  ArchHelpers::Context::BackupContext(ucontext, Context);
-
-  // Retain the action pointer so we can see it when we return
-  Context->Signal = Signal;
-
-  // Save guest state
-  // We can't guarantee if registers are in context or host GPRs
-  // So we need to save everything
-  memcpy(&Context->GuestState, ThreadState->CurrentFrame, sizeof(FEXCore::Core::CPUState));
-
-  // Set the new SP
-  ArchHelpers::Context::SetSp(ucontext, NewSP);
-
-  // Signal frames are only used on the interpreter
-  // The JITS require the stack to be setup correctly on rt_sigreturn
-  if (CTX->Config.Core() == FEXCore::Config::CONFIG_INTERPRETER) {
-    SignalFrames.push(NewSP);
-  }
-
-  Context->Flags = 0;
-  Context->FPStateLocation = 0;
-  Context->UContextLocation = 0;
-  Context->SigInfoLocation = 0;
-
-  // Store fault to top status and then reset it
-  Context->FaultToTopAndGeneratedException = SynchronousFaultData.FaultToTopAndGeneratedException;
-  SynchronousFaultData.FaultToTopAndGeneratedException = false;
-
-  return Context;
-}
-
-void Dispatcher::RestoreThreadState(void *ucontext) {
-  uint64_t OldSP{};
-  if (CTX->Config.Core() == FEXCore::Config::CONFIG_IRJIT) {
-    OldSP = ArchHelpers::Context::GetSp(ucontext);
-  }
-  else {
-    LOGMAN_THROW_A_FMT(!SignalFrames.empty(), "Trying to restore a signal frame when we don't have any");
-    OldSP = SignalFrames.top();
-    SignalFrames.pop();
-  }
-
-  uintptr_t NewSP = OldSP;
-  auto Context = reinterpret_cast<ArchHelpers::Context::ContextBackup*>(NewSP);
-
-  // First thing, reset the guest state
-  memcpy(ThreadState->CurrentFrame, &Context->GuestState, sizeof(FEXCore::Core::CPUState));
-
-  // Now restore host state
-  ArchHelpers::Context::RestoreContext(ucontext, Context);
-
-  if (Context->UContextLocation) {
-    auto Frame = ThreadState->CurrentFrame;
-
-    if (Context->Flags &ArchHelpers::Context::ContextFlags::CONTEXT_FLAG_INJIT) {
-      // XXX: Unsupported since it needs state reconstruction
-      // If we are in the JIT then SRA might need to be restored to values from the context
-      // We can't currently support this since it might result in tearing without real state reconstruction
-    }
-
-    if (!(Context->Flags & ArchHelpers::Context::ContextFlags::CONTEXT_FLAG_32BIT)) {
-      auto *guest_uctx = reinterpret_cast<FEXCore::x86_64::ucontext_t*>(Context->UContextLocation);
-      [[maybe_unused]] auto *guest_siginfo = reinterpret_cast<siginfo_t*>(Context->SigInfoLocation);
-
-      // If the guest modified the RIP then we need to take special precautions here
-      if (Context->OriginalRIP != guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_RIP] ||
-          Context->FaultToTopAndGeneratedException) {
-        // Hack! Go back to the top of the dispatcher top
-        // This is only safe inside the JIT rather than anything outside of it
-        ArchHelpers::Context::SetPc(ucontext, AbsoluteLoopTopAddressFillSRA);
-        // Set our state register to point to our guest thread data
-        ArchHelpers::Context::SetState(ucontext, reinterpret_cast<uint64_t>(Frame));
-
-        Frame->State.rip = guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_RIP];
-        // XXX: Full context setting
-        uint32_t eflags = guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_EFL];
-        for (size_t i = 0; i < 32; ++i) {
-          Frame->State.flags[i] = (eflags & (1U << i)) ? 1 : 0;
-        }
-
-        Frame->State.flags[1] = 1;
-        Frame->State.flags[9] = 1;
-
-#define COPY_REG(x) \
-            Frame->State.gregs[X86State::REG_##x] = guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_##x];
-            COPY_REG(R8);
-            COPY_REG(R9);
-            COPY_REG(R10);
-            COPY_REG(R11);
-            COPY_REG(R12);
-            COPY_REG(R13);
-            COPY_REG(R14);
-            COPY_REG(R15);
-            COPY_REG(RDI);
-            COPY_REG(RSI);
-            COPY_REG(RBP);
-            COPY_REG(RBX);
-            COPY_REG(RDX);
-            COPY_REG(RAX);
-            COPY_REG(RCX);
-            COPY_REG(RSP);
-#undef COPY_REG
-        FEXCore::x86_64::_libc_fpstate *fpstate = reinterpret_cast<FEXCore::x86_64::_libc_fpstate*>(guest_uctx->uc_mcontext.fpregs);
-        // Copy float registers
-        memcpy(Frame->State.mm, fpstate->_st, sizeof(Frame->State.mm));
-        memcpy(Frame->State.xmm, fpstate->_xmm, sizeof(Frame->State.xmm));
-
-        // FCW store default
-        Frame->State.FCW = fpstate->fcw;
-        Frame->State.FTW = fpstate->ftw;
-
-        // Deconstruct FSW
-        Frame->State.flags[FEXCore::X86State::X87FLAG_C0_LOC] = (fpstate->fsw >> 8) & 1;
-        Frame->State.flags[FEXCore::X86State::X87FLAG_C1_LOC] = (fpstate->fsw >> 9) & 1;
-        Frame->State.flags[FEXCore::X86State::X87FLAG_C2_LOC] = (fpstate->fsw >> 10) & 1;
-        Frame->State.flags[FEXCore::X86State::X87FLAG_C3_LOC] = (fpstate->fsw >> 14) & 1;
-        Frame->State.flags[FEXCore::X86State::X87FLAG_TOP_LOC] = (fpstate->fsw >> 11) & 0b111;
-      }
-    }
-    else {
-      auto *guest_uctx = reinterpret_cast<FEXCore::x86::ucontext_t*>(Context->UContextLocation);
-      [[maybe_unused]] auto *guest_siginfo = reinterpret_cast<FEXCore::x86::siginfo_t*>(Context->SigInfoLocation);
-      // If the guest modified the RIP then we need to take special precautions here
-      if (Context->OriginalRIP != guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_EIP] ||
-          Context->FaultToTopAndGeneratedException) {
-        // Hack! Go back to the top of the dispatcher top
-        // This is only safe inside the JIT rather than anything outside of it
-        ArchHelpers::Context::SetPc(ucontext, AbsoluteLoopTopAddressFillSRA);
-        // Set our state register to point to our guest thread data
-        ArchHelpers::Context::SetState(ucontext, reinterpret_cast<uint64_t>(Frame));
-
-        // XXX: Full context setting
-        // First 32-bytes of flags is EFLAGS broken out
-        uint32_t eflags = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_EFL];
-        for (size_t i = 0; i < 32; ++i) {
-          Frame->State.flags[i] = (eflags & (1U << i)) ? 1 : 0;
-        }
-
-        Frame->State.flags[1] = 1;
-        Frame->State.flags[9] = 1;
-
-        Frame->State.rip = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_EIP];
-        Frame->State.cs = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_CS];
-        Frame->State.ds = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_DS];
-        Frame->State.es = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ES];
-        Frame->State.fs = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_FS];
-        Frame->State.gs = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_GS];
-        Frame->State.ss = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_SS];
-#define COPY_REG(x) \
-      Frame->State.gregs[X86State::REG_##x] = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_##x];
-        COPY_REG(RDI);
-        COPY_REG(RSI);
-        COPY_REG(RBP);
-        COPY_REG(RBX);
-        COPY_REG(RDX);
-        COPY_REG(RAX);
-        COPY_REG(RCX);
-        COPY_REG(RSP);
-#undef COPY_REG
-        FEXCore::x86::_libc_fpstate *fpstate = reinterpret_cast<FEXCore::x86::_libc_fpstate*>(guest_uctx->uc_mcontext.fpregs);
-
-        // Copy float registers
-        for (size_t i = 0; i < 8; ++i) {
-          // 32-bit st register size is only 10 bytes. Not padded to 16byte like x86-64
-          memcpy(&Frame->State.mm[i], &fpstate->_st[i], 10);
-        }
-
-        // Extended XMM state
-        memcpy(fpstate->_xmm, Frame->State.xmm, sizeof(Frame->State.xmm));
-
-        // FCW store default
-        Frame->State.FCW = fpstate->fcw;
-        Frame->State.FTW = fpstate->ftw;
-
-        // Deconstruct FSW
-        Frame->State.flags[FEXCore::X86State::X87FLAG_C0_LOC] = (fpstate->fsw >> 8) & 1;
-        Frame->State.flags[FEXCore::X86State::X87FLAG_C1_LOC] = (fpstate->fsw >> 9) & 1;
-        Frame->State.flags[FEXCore::X86State::X87FLAG_C2_LOC] = (fpstate->fsw >> 10) & 1;
-        Frame->State.flags[FEXCore::X86State::X87FLAG_C3_LOC] = (fpstate->fsw >> 14) & 1;
-        Frame->State.flags[FEXCore::X86State::X87FLAG_TOP_LOC] = (fpstate->fsw >> 11) & 0b111;
-      }
-    }
-  }
-}
-
-static uint32_t ConvertSignalToTrapNo(int Signal, siginfo_t *HostSigInfo) {
-  switch (Signal) {
-    case SIGSEGV:
-      if (HostSigInfo->si_code == SEGV_MAPERR ||
-          HostSigInfo->si_code == SEGV_ACCERR) {
-        // Protection fault
-        return X86State::X86_TRAPNO_PF;
-      }
-      break;
-  }
-
-  // Unknown mapping, fall back to old behaviour and just pass signal
-  return Signal;
-}
-
-static uint32_t ConvertSignalToError(int Signal, siginfo_t *HostSigInfo) {
-  switch (Signal) {
-    case SIGSEGV:
-      if (HostSigInfo->si_code == SEGV_MAPERR ||
-          HostSigInfo->si_code == SEGV_ACCERR) {
-        // Protection fault
-        // Always a user fault for us
-        // XXX: PF_PROT and PF_WRITE
-        return X86State::X86_PF_USER;
-      }
-      break;
-  }
-
-  // Not a page fault issue
-  return 0;
-}
-
-bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack) {
-  auto ContextBackup = StoreThreadState(Signal, ucontext);
-
-  auto Frame = ThreadState->CurrentFrame;
-
-  // Ref count our faults
-  // We use this to track if it is safe to clear cache
-  ++SignalHandlerRefCounter;
-
-  uint64_t OldPC = ArchHelpers::Context::GetPc(ucontext);
-  // Set the new PC
-  ArchHelpers::Context::SetPc(ucontext, AbsoluteLoopTopAddressFillSRA);
-  // Set our state register to point to our guest thread data
-  ArchHelpers::Context::SetState(ucontext, reinterpret_cast<uint64_t>(Frame));
-
-  uint64_t OldGuestSP = Frame->State.gregs[X86State::REG_RSP];
-  uint64_t NewGuestSP = OldGuestSP;
-
-  // Pulling from context here
-  bool Is64BitMode = CTX->Config.Is64BitMode;
-  uint64_t SignalReturn = CTX->X86CodeGen.SignalReturn;
-
-  // Spill the SRA regardless of signal handler type
-  // We are going to be returning to the top of the dispatcher which will fill again
-  // Otherwise we might load garbage
-  if (SRAEnabled) {
-    if (IsAddressInJITCode(OldPC, false)) {
-      uint32_t IgnoreMask{};
-#ifdef _M_ARM_64
-      if (Frame->InSyscallInfo != 0) {
-        // We are in a syscall, this means we are in a weird register state
-        // We need to spill SRA but only some of it, since some values have already been spilled
-        // Lower 16 bits tells us which registers are already spilled to the context
-        // So we ignore spilling those ones
-        uint16_t NumRegisters = std::popcount(Frame->InSyscallInfo & 0xFFFF);
-        if (NumRegisters >= 4) {
-          // Unhandled case
-          IgnoreMask = 0;
-        }
-        else {
-          IgnoreMask = Frame->InSyscallInfo & 0xFFFF;
-        }
-      }
-      else {
-        // We must spill everything
-        IgnoreMask = 0;
-      }
-#endif
-
-      // We are in jit, SRA must be spilled
-      SpillSRA(ucontext, IgnoreMask);
-
-      ContextBackup->Flags |= ArchHelpers::Context::ContextFlags::CONTEXT_FLAG_INJIT;
-    } else {
-      if (!IsAddressInJITCode(OldPC, true)) {
-        // This is likely to cause issues but in some cases it isn't fatal
-        // This can also happen if we have put a signal on hold, then we just reenabled the signal
-        // So we are in the syscall handler
-        // Only throw a log message in this case
-        if constexpr (false) {
-          // XXX: Messages in the signal handler can cause us to crash
-          LogMan::Msg::EFmt("Signals in dispatcher have unsynchronized context");
-        }
-      }
-    }
-  }
-
-  // altstack is only used if the signal handler was setup with SA_ONSTACK
-  if (GuestAction->sa_flags & SA_ONSTACK) {
-    // Additionally the altstack is only used if the enabled (SS_DISABLE flag is not set)
-    if (!(GuestStack->ss_flags & SS_DISABLE)) {
-      // If our guest is already inside of the alternative stack
-      // Then that means we are hitting recursive signals and we need to walk back the stack correctly
-      uint64_t AltStackBase = reinterpret_cast<uint64_t>(GuestStack->ss_sp);
-      uint64_t AltStackEnd = AltStackBase + GuestStack->ss_size;
-      if (OldGuestSP >= AltStackBase &&
-          OldGuestSP <= AltStackEnd) {
-        // We are already in the alt stack, the rest of the code will handle adjusting this
-      }
-      else {
-        NewGuestSP = AltStackEnd;
-      }
-    }
-  }
-
-  if (Is64BitMode) {
-    // Back up past the redzone, which is 128bytes
-    // 32-bit doesn't have a redzone
-    NewGuestSP -= 128;
-  }
-
-  // siginfo_t
-  siginfo_t *HostSigInfo = reinterpret_cast<siginfo_t*>(info);
-
-  // Backup where we think the RIP currently is
-  ContextBackup->OriginalRIP = Frame->State.rip;
-
-  if (GuestAction->sa_flags & SA_SIGINFO) {
-    // Setup ucontext a bit
-    if (Is64BitMode) {
-      NewGuestSP -= sizeof(FEXCore::x86_64::_libc_fpstate);
-      NewGuestSP = AlignDown(NewGuestSP, alignof(FEXCore::x86_64::_libc_fpstate));
-      uint64_t FPStateLocation = NewGuestSP;
-
-      NewGuestSP -= sizeof(FEXCore::x86_64::ucontext_t);
-      NewGuestSP = AlignDown(NewGuestSP, alignof(FEXCore::x86_64::ucontext_t));
-      uint64_t UContextLocation = NewGuestSP;
-
-      NewGuestSP -= sizeof(siginfo_t);
-      NewGuestSP = AlignDown(NewGuestSP, alignof(siginfo_t));
-      uint64_t SigInfoLocation = NewGuestSP;
-
-      ContextBackup->FPStateLocation = FPStateLocation;
-      ContextBackup->UContextLocation = UContextLocation;
-      ContextBackup->SigInfoLocation = SigInfoLocation;
-
-      FEXCore::x86_64::ucontext_t *guest_uctx = reinterpret_cast<FEXCore::x86_64::ucontext_t*>(UContextLocation);
-      siginfo_t *guest_siginfo = reinterpret_cast<siginfo_t*>(SigInfoLocation);
-
-      // We have extended float information
-      guest_uctx->uc_flags = FEXCore::x86_64::UC_FP_XSTATE;
-
-      // Pointer to where the fpreg memory is
-      guest_uctx->uc_mcontext.fpregs = reinterpret_cast<FEXCore::x86_64::_libc_fpstate*>(FPStateLocation);
-      FEXCore::x86_64::_libc_fpstate *fpstate = reinterpret_cast<FEXCore::x86_64::_libc_fpstate*>(FPStateLocation);
-
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_RIP] = Frame->State.rip;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_EFL] = 0;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_CSGSFS] = 0;
-
-      // aarch64 and x86_64 siginfo_t matches. We can just copy this over
-      // SI_USER could also potentially have random data in it, needs to be bit perfect
-      // For guest faults we don't have a real way to reconstruct state to a real guest RIP
-      *guest_siginfo = *HostSigInfo;
-
-      if (ContextBackup->FaultToTopAndGeneratedException) {
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_TRAPNO] = SynchronousFaultData.TrapNo;
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_ERR] = SynchronousFaultData.err_code;
-
-        // Overwrite si_code
-        guest_siginfo->si_code = SynchronousFaultData.si_code;
-      }
-      else {
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_TRAPNO] = ConvertSignalToTrapNo(Signal, HostSigInfo);
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_ERR] = ConvertSignalToError(Signal, HostSigInfo);
-      }
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_OLDMASK] = 0;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_CR2] = 0;
-
-#define COPY_REG(x) \
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_##x] = Frame->State.gregs[X86State::REG_##x];
-      COPY_REG(R8);
-      COPY_REG(R9);
-      COPY_REG(R10);
-      COPY_REG(R11);
-      COPY_REG(R12);
-      COPY_REG(R13);
-      COPY_REG(R14);
-      COPY_REG(R15);
-      COPY_REG(RDI);
-      COPY_REG(RSI);
-      COPY_REG(RBP);
-      COPY_REG(RBX);
-      COPY_REG(RDX);
-      COPY_REG(RAX);
-      COPY_REG(RCX);
-      COPY_REG(RSP);
-#undef COPY_REG
-
-      // Copy float registers
-      memcpy(fpstate->_st, Frame->State.mm, sizeof(Frame->State.mm));
-      memcpy(fpstate->_xmm, Frame->State.xmm, sizeof(Frame->State.xmm));
-
-      // FCW store default
-      fpstate->fcw = Frame->State.FCW;
-      fpstate->ftw = Frame->State.FTW;
-
-      // Reconstruct FSW
-      fpstate->fsw =
-        (Frame->State.flags[FEXCore::X86State::X87FLAG_TOP_LOC] << 11) |
-        (Frame->State.flags[FEXCore::X86State::X87FLAG_C0_LOC] << 8) |
-        (Frame->State.flags[FEXCore::X86State::X87FLAG_C1_LOC] << 9) |
-        (Frame->State.flags[FEXCore::X86State::X87FLAG_C2_LOC] << 10) |
-        (Frame->State.flags[FEXCore::X86State::X87FLAG_C3_LOC] << 14);
-
-      // Copy over signal stack information
-      guest_uctx->uc_stack.ss_flags = GuestStack->ss_flags;
-      guest_uctx->uc_stack.ss_sp = GuestStack->ss_sp;
-      guest_uctx->uc_stack.ss_size = GuestStack->ss_size;
-
-      Frame->State.gregs[X86State::REG_RSI] = SigInfoLocation;
-      Frame->State.gregs[X86State::REG_RDX] = UContextLocation;
-    }
-    else {
-      ContextBackup->Flags |= ArchHelpers::Context::ContextFlags::CONTEXT_FLAG_32BIT;
-
-      NewGuestSP -= sizeof(FEXCore::x86::_libc_fpstate);
-      NewGuestSP = AlignDown(NewGuestSP, alignof(FEXCore::x86::_libc_fpstate));
-      uint64_t FPStateLocation = NewGuestSP;
-
-      NewGuestSP -= sizeof(FEXCore::x86::ucontext_t);
-      NewGuestSP = AlignDown(NewGuestSP, alignof(FEXCore::x86::ucontext_t));
-      uint64_t UContextLocation = NewGuestSP;
-
-      NewGuestSP -= sizeof(FEXCore::x86::siginfo_t);
-      NewGuestSP = AlignDown(NewGuestSP, alignof(FEXCore::x86::siginfo_t));
-      uint64_t SigInfoLocation = NewGuestSP;
-
-      ContextBackup->FPStateLocation = FPStateLocation;
-      ContextBackup->UContextLocation = UContextLocation;
-      ContextBackup->SigInfoLocation = SigInfoLocation;
-
-      FEXCore::x86::ucontext_t *guest_uctx = reinterpret_cast<FEXCore::x86::ucontext_t*>(UContextLocation);
-      FEXCore::x86::siginfo_t *guest_siginfo = reinterpret_cast<FEXCore::x86::siginfo_t*>(SigInfoLocation);
-
-      // We have extended float information
-      guest_uctx->uc_flags = FEXCore::x86::UC_FP_XSTATE;
-
-      // Pointer to where the fpreg memory is
-      guest_uctx->uc_mcontext.fpregs = static_cast<uint32_t>(FPStateLocation);
-      FEXCore::x86::_libc_fpstate *fpstate = reinterpret_cast<FEXCore::x86::_libc_fpstate*>(FPStateLocation);
-
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_GS] = Frame->State.gs;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_FS] = Frame->State.fs;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ES] = Frame->State.es;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_DS] = Frame->State.ds;
-      if (ContextBackup->FaultToTopAndGeneratedException) {
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_TRAPNO] = SynchronousFaultData.TrapNo;
-        guest_siginfo->si_code = SynchronousFaultData.si_code;
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ERR] = SynchronousFaultData.err_code;
-      }
-      else {
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_TRAPNO] = ConvertSignalToTrapNo(Signal, HostSigInfo);
-        guest_siginfo->si_code = HostSigInfo->si_code;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ERR] = ConvertSignalToError(Signal, HostSigInfo);
-      }
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_EIP] = Frame->State.rip;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_CS] = Frame->State.cs;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_EFL] = 0;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_UESP] = 0;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_SS] = Frame->State.ss;
-
-#define COPY_REG(x) \
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_##x] = Frame->State.gregs[X86State::REG_##x];
-      COPY_REG(RDI);
-      COPY_REG(RSI);
-      COPY_REG(RBP);
-      COPY_REG(RBX);
-      COPY_REG(RDX);
-      COPY_REG(RAX);
-      COPY_REG(RCX);
-      COPY_REG(RSP);
-#undef COPY_REG
-
-      // Copy float registers
-      for (size_t i = 0; i < 8; ++i) {
-        // 32-bit st register size is only 10 bytes. Not padded to 16byte like x86-64
-        memcpy(&fpstate->_st[i], &Frame->State.mm[i], 10);
-      }
-
-      // Extended XMM state
-      fpstate->status = FEXCore::x86::fpstate_magic::MAGIC_XFPSTATE;
-      memcpy(fpstate->_xmm, Frame->State.xmm, sizeof(Frame->State.xmm));
-
-      // FCW store default
-      fpstate->fcw = Frame->State.FCW;
-      fpstate->ftw = Frame->State.FTW;
-      // Reconstruct FSW
-      fpstate->fsw =
-        (Frame->State.flags[FEXCore::X86State::X87FLAG_TOP_LOC] << 11) |
-        (Frame->State.flags[FEXCore::X86State::X87FLAG_C0_LOC] << 8) |
-        (Frame->State.flags[FEXCore::X86State::X87FLAG_C1_LOC] << 9) |
-        (Frame->State.flags[FEXCore::X86State::X87FLAG_C2_LOC] << 10) |
-        (Frame->State.flags[FEXCore::X86State::X87FLAG_C3_LOC] << 14);
-
-      // Copy over signal stack information
-      guest_uctx->uc_stack.ss_flags = GuestStack->ss_flags;
-      guest_uctx->uc_stack.ss_sp = static_cast<uint32_t>(reinterpret_cast<uint64_t>(GuestStack->ss_sp));
-      guest_uctx->uc_stack.ss_size = GuestStack->ss_size;
-
-      // These three elements are in every siginfo
-      guest_siginfo->si_signo = HostSigInfo->si_signo;
-      guest_siginfo->si_errno = HostSigInfo->si_errno;
-
-      switch (Signal) {
-        case SIGSEGV:
-        case SIGBUS:
-          // Macro expansion to get the si_addr
-          // This is the address trying to be accessed, not the RIP
-          guest_siginfo->_sifields._sigfault.addr = static_cast<uint32_t>(reinterpret_cast<uintptr_t>(HostSigInfo->si_addr));
-          break;
-        case SIGFPE:
-        case SIGILL:
-          // Macro expansion to get the si_addr
-          // Can't really give a real result here. Pull from the context for now
-          guest_siginfo->_sifields._sigfault.addr = Frame->State.rip;
-          break;
-        case SIGCHLD:
-          guest_siginfo->_sifields._sigchld.pid = HostSigInfo->si_pid;
-          guest_siginfo->_sifields._sigchld.uid = HostSigInfo->si_uid;
-          guest_siginfo->_sifields._sigchld.status = HostSigInfo->si_status;
-          guest_siginfo->_sifields._sigchld.utime = HostSigInfo->si_utime;
-          guest_siginfo->_sifields._sigchld.stime = HostSigInfo->si_stime;
-          break;
-        case SIGALRM:
-        case SIGVTALRM:
-          guest_siginfo->_sifields._timer.tid = HostSigInfo->si_timerid;
-          guest_siginfo->_sifields._timer.overrun = HostSigInfo->si_overrun;
-          guest_siginfo->_sifields._timer.sigval.sival_int = HostSigInfo->si_int;
-          break;
-        default:
-          LogMan::Msg::EFmt("Unhandled siginfo_t for signal: {}\n", Signal);
-          break;
-      }
-
-      NewGuestSP -= 4;
-      *(uint32_t*)NewGuestSP = UContextLocation;
-      NewGuestSP -= 4;
-      *(uint32_t*)NewGuestSP = SigInfoLocation;
-      NewGuestSP -= 4;
-      *(uint32_t*)NewGuestSP = Signal;
-    }
-
-    Frame->State.rip = reinterpret_cast<uint64_t>(GuestAction->sigaction_handler.sigaction);
-  }
-  else {
-    if (!Is64BitMode) {
-      NewGuestSP -= 4;
-      *(uint32_t*)NewGuestSP = Signal;
-    }
-
-    Frame->State.rip = reinterpret_cast<uint64_t>(GuestAction->sigaction_handler.handler);
-  }
-
-  if (Is64BitMode) {
-    Frame->State.gregs[FEXCore::X86State::REG_RDI] = Signal;
-
-    // Set up the new SP for stack handling
-    NewGuestSP -= 8;
-    *(uint64_t*)NewGuestSP = SignalReturn;
-    Frame->State.gregs[FEXCore::X86State::REG_RSP] = NewGuestSP;
-  }
-  else {
-    NewGuestSP -= 4;
-    *(uint32_t*)NewGuestSP = SignalReturn;
-    LOGMAN_THROW_A_FMT(SignalReturn < 0x1'0000'0000ULL, "This needs to be below 4GB");
-    Frame->State.gregs[FEXCore::X86State::REG_RSP] = NewGuestSP;
-  }
-
-  // The guest starts its signal frame with a zero initialized FPU
-  // Set that up now. Little bit costly but it's a requirement
-  // This state will be restored on rt_sigreturn
-  memset(Frame->State.xmm, 0, sizeof(Frame->State.xmm));
-  memset(Frame->State.mm, 0, sizeof(Frame->State.mm));
-  Frame->State.FCW = 0x37F;
-  Frame->State.FTW = 0xFFFF;
-
-  return true;
-}
-
-bool Dispatcher::HandleSIGILL(int Signal, void *info, void *ucontext) {
-
-  if (ArchHelpers::Context::GetPc(ucontext) == SignalHandlerReturnAddress) {
-    RestoreThreadState(ucontext);
-
-    // Ref count our faults
-    // We use this to track if it is safe to clear cache
-    --SignalHandlerRefCounter;
-    return true;
-  }
-
-  if (ArchHelpers::Context::GetPc(ucontext) == PauseReturnInstruction) {
-    RestoreThreadState(ucontext);
-
-    // Ref count our faults
-    // We use this to track if it is safe to clear cache
-    --SignalHandlerRefCounter;
-    return true;
-  }
-
-  return false;
-}
-
-bool Dispatcher::HandleSignalPause(int Signal, void *info, void *ucontext) {
-  FEXCore::Core::SignalEvent SignalReason = ThreadState->SignalReason.load();
-  auto Frame = ThreadState->CurrentFrame;
-
-  if (SignalReason == FEXCore::Core::SignalEvent::Pause) {
-    // Store our thread state so we can come back to this
-    StoreThreadState(Signal, ucontext);
-
-    if (SRAEnabled && IsAddressInJITCode(ArchHelpers::Context::GetPc(ucontext), false)) {
-      // We are in jit, SRA must be spilled
-      ArchHelpers::Context::SetPc(ucontext, ThreadPauseHandlerAddressSpillSRA);
-    } else {
-      if (SRAEnabled) {
-        // We are in non-jit, SRA is already spilled
-        LOGMAN_THROW_A_FMT(!IsAddressInJITCode(ArchHelpers::Context::GetPc(ucontext), true),
-                           "Signals in dispatcher have unsynchronized context");
-      }
-      ArchHelpers::Context::SetPc(ucontext, ThreadPauseHandlerAddress);
-    }
-
-    // Set our state register to point to our guest thread data
-    ArchHelpers::Context::SetState(ucontext, reinterpret_cast<uint64_t>(Frame));
-
-    // Ref count our faults
-    // We use this to track if it is safe to clear cache
-    ++SignalHandlerRefCounter;
-
-    ThreadState->SignalReason.store(FEXCore::Core::SignalEvent::Nothing);
-    return true;
-  }
-
-  if (SignalReason == FEXCore::Core::SignalEvent::Stop) {
-    // Our thread is stopping
-    // We don't care about anything at this point
-    // Set the stack to our starting location when we entered the core and get out safely
-    ArchHelpers::Context::SetSp(ucontext, Frame->ReturningStackLocation);
-
-    // Our ref counting doesn't matter anymore
-    SignalHandlerRefCounter = 0;
-
-    // Set the new PC
-    if (SRAEnabled && IsAddressInJITCode(ArchHelpers::Context::GetPc(ucontext), false)) {
-      // We are in jit, SRA must be spilled
-      ArchHelpers::Context::SetPc(ucontext, ThreadStopHandlerAddressSpillSRA);
-    } else {
-      if (SRAEnabled) {
-        // We are in non-jit, SRA is already spilled
-        LOGMAN_THROW_A_FMT(!IsAddressInJITCode(ArchHelpers::Context::GetPc(ucontext), true),
-                           "Signals in dispatcher have unsynchronized context");
-      }
-      ArchHelpers::Context::SetPc(ucontext, ThreadStopHandlerAddress);
-    }
-
-    // We need to be a little bit careful here
-    // If we were already paused (due to GDB) and we are immediately stopping (due to gdb kill)
-    // Then we need to ensure we don't double decrement our idle thread counter
-    if (ThreadState->RunningEvents.ThreadSleeping) {
-      // If the thread was sleeping then its idle counter was decremented
-      // Reincrement it here to not break logic
-      ++ThreadState->CTX->IdleWaitRefCount;
-    }
-
-    ThreadState->SignalReason.store(FEXCore::Core::SignalEvent::Nothing);
-    return true;
-  }
-
-  if (SignalReason == FEXCore::Core::SignalEvent::Return) {
-    RestoreThreadState(ucontext);
-
-    // Ref count our faults
-    // We use this to track if it is safe to clear cache
-    --SignalHandlerRefCounter;
-
-    ThreadState->SignalReason.store(FEXCore::Core::SignalEvent::Nothing);
-    return true;
-  }
-
-  return false;
-}
-
 uint64_t Dispatcher::GetCompileBlockPtr() {
-  using ClassPtrType = void (FEXCore::Context::Context::*)(FEXCore::Core::CpuStateFrame *, uint64_t);
+  using ClassPtrType = void (FEXCore::Context::ContextImpl::*)(FEXCore::Core::CpuStateFrame *, uint64_t);
  union PtrCast {
    ClassPtrType ClassPtr;
    uintptr_t Data;
  };

  PtrCast CompileBlockPtr;
-  CompileBlockPtr.ClassPtr = &FEXCore::Context::Context::CompileBlockJit;
+  CompileBlockPtr.ClassPtr = &FEXCore::Context::ContextImpl::CompileBlockJit;
  return CompileBlockPtr.Data;
 }

-void Dispatcher::RemoveCodeBuffer(uint8_t* start_to_remove) {
-  for (auto iter = CodeBuffers.begin(); iter != CodeBuffers.end(); ++iter) {
-    auto [start, end] = *iter;
-    if (start == reinterpret_cast<uint64_t>(start_to_remove)) {
-      CodeBuffers.erase(iter);
-      return;
-    }
-  }
-}
-
-bool Dispatcher::IsAddressInJITCode(uint64_t Address, bool IncludeDispatcher, bool IncludeCompileService) const {
-  for (auto [start, end] : CodeBuffers) {
-    if (Address >= start && Address < end) {
-      return true;
-    }
-  }
-
-  if (IncludeDispatcher && IsAddressInDispatcher(Address)) {
-    return true;
-  }
-
-  if (IncludeCompileService &&  ThreadState->CompileService && ThreadState->CompileService->IsAddressInJITCode(Address)) {
-    return true;
-  }
-  return false;
-}
-
 }
@@ -1,16 +1,13 @@
 #pragma once

 #include <FEXCore/Core/CPUBackend.h>
+#include <FEXCore/fextl/memory.h>

-#include "Interface/Context/Context.h"
-#include "Interface/Core/ArchHelpers/MContext.h"
-
-#include <bits/types/stack_t.h>
 #include <cstdint>
+#include <signal.h>
 #include <stddef.h>
 #include <stack>
 #include <tuple>
-#include <vector>

 namespace FEXCore {
 struct GuestSigAction;
@@ -21,21 +18,19 @@ struct CpuStateFrame;
 struct InternalThreadState;
 }

+namespace FEXCore::Context {
+class ContextImpl;
+}
+
 namespace FEXCore::CPU {

 struct DispatcherConfig {
-  bool ExecuteBlocksWithCall = false;
-  uintptr_t ExitFunctionLink = 0;
-  uintptr_t ExitFunctionLinkThis = 0;
-  bool StaticRegisterAssignment = false;
+  bool StaticRegisterAllocation = false;
 };

 class Dispatcher {
 public:
  virtual ~Dispatcher() = default;
-  CPUBackend::AsmDispatch DispatchPtr;
-  CPUBackend::JITCallback CallbackPtr;
-  FEXCore::Context::Context::IntCallbackReturn ReturnPtr;

  /**
   * @name Dispatch Helper functions
@@ -48,61 +43,73 @@ public:
  uint64_t ThreadPauseHandlerAddressSpillSRA{};
  uint64_t ExitFunctionLinkerAddress{};
  uint64_t SignalHandlerReturnAddress{};
-  uint64_t UnimplementedInstructionAddress{};
-  uint64_t OverflowExceptionInstructionAddress{};
+  uint64_t SignalHandlerReturnAddressRT{};
+  uint64_t GuestSignal_SIGILL{};
+  uint64_t GuestSignal_SIGTRAP{};
+  uint64_t GuestSignal_SIGSEGV{};
+  uint64_t IntCallbackReturnAddress{};

  uint64_t PauseReturnInstruction{};

  /**  @} */

-  uint32_t SignalHandlerRefCounter{};
-  struct SynchronousFaultDataStruct {
-    bool FaultToTopAndGeneratedException{};
-    uint32_t TrapNo;
-    uint32_t err_code;
-    uint32_t si_code;
-  } SynchronousFaultData;
-
  uint64_t Start{};
  uint64_t End{};

-  bool HandleGuestSignal(int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack);
-  bool HandleSIGILL(int Signal, void *info, void *ucontext);
-  bool HandleSignalPause(int Signal, void *info, void *ucontext);
+  virtual void InitThreadPointers(FEXCore::Core::InternalThreadState *Thread) = 0;

-  void RegisterCodeBuffer(uint8_t* start, size_t size) {
-    CodeBuffers.emplace_back(reinterpret_cast<uint64_t>(start),
-        reinterpret_cast<uint64_t>(start + size));
+  // These are across all arches for now
+  static constexpr size_t MaxGDBPauseCheckSize = 128;
+  static constexpr size_t MaxInterpreterTrampolineSize = 128;
+
+  virtual size_t GenerateGDBPauseCheck(uint8_t *CodeBuffer, uint64_t GuestRIP) = 0;
+  virtual size_t GenerateInterpreterTrampoline(uint8_t *CodeBuffer) = 0;
+
+  static fextl::unique_ptr<Dispatcher> CreateX86(FEXCore::Context::ContextImpl *CTX, const DispatcherConfig &Config);
+  static fextl::unique_ptr<Dispatcher> CreateArm64(FEXCore::Context::ContextImpl *CTX, const DispatcherConfig &Config);
+
+  virtual void ExecuteDispatch(FEXCore::Core::CpuStateFrame *Frame) {
+    DispatchPtr(Frame);
  }

-  void RemoveCodeBuffer(uint8_t* start);
-
-  bool IsAddressInJITCode(uint64_t Address, bool IncludeDispatcher = true, bool IncludeCompileService = true) const;
-  bool IsAddressInDispatcher(uint64_t Address) const {
-    return Address >= Start && Address < End;
+  virtual void ExecuteJITCallback(FEXCore::Core::CpuStateFrame *Frame, uint64_t RIP) {
+    CallbackPtr(Frame, RIP);
  }

+  virtual uint16_t GetSRAGPRCount() const {
+    return 0U;
+  }
+
+  virtual uint16_t GetSRAFPRCount() const {
+    return 0U;
+  }
+
+  virtual void GetSRAGPRMapping(uint8_t Mapping[16]) const {
+  }
+
+  virtual void GetSRAFPRMapping(uint8_t Mapping[16]) const {
+  }
+
+  const DispatcherConfig& GetConfig() const { return config; }
+
 protected:
-  Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread)
+  Dispatcher(FEXCore::Context::ContextImpl *ctx, const DispatcherConfig &Config)
    : CTX {ctx}
-    , ThreadState {Thread} {}
+    , config {Config}
+    {}

-  ArchHelpers::Context::ContextBackup* StoreThreadState(int Signal, void *ucontext);
-  void RestoreThreadState(void *ucontext);
-  std::stack<uint64_t, std::vector<uint64_t>> SignalFrames;
+  FEXCore::Context::ContextImpl *CTX;
+  DispatcherConfig config;

-  bool SRAEnabled = false;
-  virtual void SpillSRA(void *ucontext, uint32_t IgnoreMask) {}
-
-  FEXCore::Context::Context *CTX;
-  FEXCore::Core::InternalThreadState *ThreadState;
-
-  static void SleepThread(FEXCore::Context::Context *ctx, FEXCore::Core::CpuStateFrame *Frame);
+  static void SleepThread(FEXCore::Context::ContextImpl *ctx, FEXCore::Core::CpuStateFrame *Frame);

  static uint64_t GetCompileBlockPtr();

-private:
-  std::vector<std::tuple<uint64_t, uint64_t>> CodeBuffers; // Start, End
+  using AsmDispatch = void(*)(FEXCore::Core::CpuStateFrame *Frame);
+  using JITCallback = void(*)(FEXCore::Core::CpuStateFrame *Frame, uint64_t RIP);
+
+  AsmDispatch DispatchPtr;
+  JITCallback CallbackPtr;
 };

 }
@@ -1,3 +1,4 @@
+#include "FEXCore/Utils/AllocatorHooks.h"
 #include "Interface/Core/LookupCache.h"

 #include "Interface/Core/Dispatcher/X86Dispatcher.h"
@@ -11,28 +12,33 @@
 #include <FEXCore/Core/CPUBackend.h>
 #include <FEXCore/Debug/InternalThreadState.h>
 #include <FEXCore/Utils/Allocator.h>
+#include <FEXCore/fextl/memory.h>
+#include <FEXCore/fextl/string.h>
 #include <FEXHeaderUtils/Syscalls.h>

 #include <cmath>
 #include <memory>
 #include <stddef.h>
 #include <stdint.h>
-#include <sys/mman.h>
-#include "xbyak/xbyak.h"
+
+#define STATE_PTR(STATE_TYPE, FIELD) \
+  [STATE + offsetof(FEXCore::Core::STATE_TYPE, FIELD)]

 namespace FEXCore::CPU {
 static constexpr size_t MAX_DISPATCHER_CODE_SIZE = 4096;
 #define STATE r14

-X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, DispatcherConfig &config)
-  : Dispatcher(ctx, Thread)
+X86Dispatcher::X86Dispatcher(FEXCore::Context::ContextImpl *ctx, const DispatcherConfig &config)
+  : Dispatcher(ctx, config)
  , Xbyak::CodeGenerator(MAX_DISPATCHER_CODE_SIZE,
-      FEXCore::Allocator::mmap(nullptr, MAX_DISPATCHER_CODE_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0),
+      FEXCore::Allocator::VirtualAlloc(MAX_DISPATCHER_CODE_SIZE, true),
      nullptr) {

+  LOGMAN_THROW_AA_FMT(!config.StaticRegisterAllocation, "X86 dispatcher does not support SRA");
+
  using namespace Xbyak;
  using namespace Xbyak::util;
-  DispatchPtr = getCurr<CPUBackend::AsmDispatch>();
+  DispatchPtr = getCurr<AsmDispatch>();

  // Temp registers
  // rax, rcx, rdx, rsi, r8, r9,
@@ -78,11 +84,10 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte

  // Save this stack pointer so we can cleanly shutdown the emulation with a long jump
  // regardless of where we were in the stack
-  mov(qword [rdi + offsetof(FEXCore::Core::CpuStateFrame, ReturningStackLocation)], rsp);
+  mov(qword STATE_PTR(CpuStateFrame, ReturningStackLocation), rsp);

  Label LoopTop;
  Label FullLookup;
-  Label CallBlock;
  Label NoBlock;
  Label ExitBlock;
  Label ThreadPauseHandler;
@@ -92,30 +97,26 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte

  {
    // Load our RIP
-    mov(rdx, qword [STATE + offsetof(FEXCore::Core::CPUState, rip)]);
+    mov(rdx, qword STATE_PTR(CPUState, rip));

    // L1 Cache
-    mov(r13, Thread->LookupCache->GetL1Pointer());
+    mov(r13, qword STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));
    mov(rax, rdx);

    and_(rax, LookupCache::L1_ENTRIES_MASK);
    shl(rax, 4);
-    cmp(qword[r13 + rax + 8], rdx);
+    cmp(qword[r13 + rax + offsetof(FEXCore::LookupCache::LookupCacheEntry, GuestCode)], rdx);
    jne(FullLookup);

-    if (!config.ExecuteBlocksWithCall) {
-      jmp(qword[r13 + rax + 0]);
-    } else {
-      mov(rax, qword[r13 + rax + 0]);
-      jmp(CallBlock);
-    }
+    jmp(qword[r13 + rax + offsetof(FEXCore::LookupCache::LookupCacheEntry, HostCode)]);

    L(FullLookup);
-    mov(r13, Thread->LookupCache->GetPagePointer());
+    mov(r13, qword STATE_PTR(CpuStateFrame, Pointers.Common.L2Pointer));

    // Full lookup
+    uint64_t VirtualMemorySize = CTX->Config.VirtualMemSize;
    mov(rax, rdx);
-    mov(rbx, Thread->LookupCache->GetVirtualMemorySize() - 1);
+    mov(rbx, VirtualMemorySize - 1);
    and_(rax, rbx);
    shr(rax, 12);

@@ -142,8 +143,7 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
    je(NoBlock);

    // Update L1
-
-    mov(r13, Thread->LookupCache->GetL1Pointer());
+    mov(r13, qword STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));
    mov(rcx, rdx);
    and_(rcx, LookupCache::L1_ENTRIES_MASK);
    shl(rcx, 1);
@@ -151,30 +151,7 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
    mov(qword[r13 + rcx*8 + 0], rax);

    // Real block if we made it here
-    if (!config.ExecuteBlocksWithCall) {
-      jmp(rax);
-    } else {
-      L(CallBlock);
-      mov(rdi, STATE);
-      call(rax);
-
-      if (CTX->GetGdbServerStatus()) {
-        // If we have a gdb server running then run in a less efficient mode that checks if we need to exit
-        // This happens when single stepping
-        static_assert(sizeof(CTX->Config.RunningMode) == 4, "This is expected to be size of 4");
-        mov(rax, qword [STATE + (offsetof(FEXCore::Core::InternalThreadState, CTX))]);
-
-        // If the value == 0 then branch to the top
-        cmp(dword [rax + (offsetof(FEXCore::Context::Context, Config.RunningMode))], 0);
-        je(LoopTop);
-        // Else we need to pause now
-        jmp(ThreadPauseHandler);
-        ud2();
-      }
-      else {
-        jmp(LoopTop);
-      }
-    }
+    jmp(rax);
  }

  {
@@ -197,6 +174,8 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
  {
    L(NoBlock);

+    inc(qword [STATE + offsetof(FEXCore::Core::CPUState, DeferredSignalRefCount)]);
+
    // {rdi, rsi, rdx}
    mov(rdi, reinterpret_cast<uint64_t>(CTX));
    mov(rsi, STATE);
@@ -204,19 +183,40 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte

    call(rax);

+    dec(qword [STATE + offsetof(FEXCore::Core::CPUState, DeferredSignalRefCount)]);
+
+    Label AfterStore;
+    // Skip the deferred fault address if the refcount isn't zero
+    jne(AfterStore);
+    mov(rax, qword [STATE + offsetof(FEXCore::Core::CPUState, DeferredSignalFaultAddress)]);
+    mov(rax, qword [rax]);
+
+    L(AfterStore);
+
    // rdx already contains RIP here
    jmp(LoopTop);
  }

  {
    ExitFunctionLinkerAddress = getCurr<uint64_t>();
-    // {rdi, rsi, rdx}
-    mov(rdi, config.ExitFunctionLinkThis);
-    mov(rsi, STATE);
-    mov(rdx, rax); // rax is set at the block end
+    inc(qword [STATE + offsetof(FEXCore::Core::CPUState, DeferredSignalRefCount)]);
+
+    // {rdi, rsi}
+    mov(rdi, STATE);
+    mov(rsi, rax); // rax is set at the block end
+
+    call(qword STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionLink));
+
+    dec(qword [STATE + offsetof(FEXCore::Core::CPUState, DeferredSignalRefCount)]);
+
+    Label AfterStore;
+    // Skip the deferred fault address if the refcount isn't zero
+    jne(AfterStore);
+    mov(rbx, qword [STATE + offsetof(FEXCore::Core::CPUState, DeferredSignalFaultAddress)]);
+    mov(qword [rbx], rbx);
+
+    L(AfterStore);

-    mov(rax, config.ExitFunctionLink);
-    call(rax);
    jmp(rax);
  }

@@ -237,7 +237,7 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
  }

  {
-    CallbackPtr = getCurr<CPUBackend::JITCallback>();
+    CallbackPtr = getCurr<JITCallback>();

    push(rbx);
    push(rbp);
@@ -252,8 +252,7 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
    // XXX: XMM?

    // Make sure to adjust the refcounter so we don't clear the cache now
-    mov(rax, reinterpret_cast<uint64_t>(&SignalHandlerRefCounter));
-    add(dword [rax], 1);
+    add(qword STATE_PTR(CpuStateFrame, SignalHandlerRefCounter), 1);

    // Now push the callback return trampoline to the guest stack
    // Guest will be misaligned because calling a thunk won't correct the guest's stack once we call the callback from the host
@@ -261,12 +260,12 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte

    // Store the trampoline to the guest stack
    // Guest stack is now correctly misaligned after a regular call instruction
-    sub(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])], 16);
-    mov(rbx, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])]);
+    sub(qword STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]), 16);
+    mov(rbx, qword STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]));
    mov(qword [rbx], rax);

    // Store RIP to the context state
-    mov(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, State.rip)], rsi);
+    mov(qword STATE_PTR(CpuStateFrame, State.rip), rsi);

    // Back to the loop top now
    jmp(LoopTop);
@@ -279,32 +278,42 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
  }

  {
-    // Guest SIGILL handler
-    // Needs to be distinct from the SignalHandlerReturnAddress
-    UnimplementedInstructionAddress = getCurr<uint64_t>();
+    // RT Signal return handler
+    SignalHandlerReturnAddressRT = getCurr<uint64_t>();
    ud2();
  }

  {
-    // Guest Overflow handler
+    // Guest SIGILL handler
    // Needs to be distinct from the SignalHandlerReturnAddress
-    OverflowExceptionInstructionAddress = getCurr<uint64_t>();
+    GuestSignal_SIGILL = getCurr<uint64_t>();
+    ud2();
+  }
+
+  {
+    // Guest SIGTRAP handler
+    // Needs to be distinct from the SignalHandlerReturnAddress
+    GuestSignal_SIGTRAP = getCurr<uint64_t>();

    // ud2 = SIGILL
    // int3 = SIGTRAP
    // hlt = SIGSEGV
+    int3();
+  }

-    mov(rax, reinterpret_cast<uint64_t>(&SynchronousFaultData));
-    add(byte [rax + offsetof(Dispatcher::SynchronousFaultDataStruct, FaultToTopAndGeneratedException)], 1);
-    mov(dword [rax + offsetof(Dispatcher::SynchronousFaultDataStruct, TrapNo)], X86State::X86_TRAPNO_OF);
-    mov(dword [rax + offsetof(Dispatcher::SynchronousFaultDataStruct, err_code)], 0);
-    mov(dword [rax + offsetof(Dispatcher::SynchronousFaultDataStruct, si_code)], 0x80);
+  {
+    // Guest SIGSEGV handler
+    // Needs to be distinct from the SignalHandlerReturnAddress
+    GuestSignal_SIGSEGV = getCurr<uint64_t>();

+    // ud2 = SIGILL
+    // int3 = SIGTRAP
+    // hlt = SIGSEGV
    hlt();
  }

  {
-    ReturnPtr = getCurr<FEXCore::Context::Context::IntCallbackReturn>();
+    IntCallbackReturnAddress = getCurr<uint64_t>();
 //  using CallbackReturn =  FEX_NAKED void(*)(FEXCore::Core::InternalThreadState *Thread, volatile void *Host_RSP);

    // rdi = thread
@@ -331,32 +340,99 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
  End = Start + getSize();

  if (CTX->Config.BlockJITNaming()) {
-    std::string Name = "Dispatch_" + std::to_string(FHU::Syscalls::gettid());
+    fextl::string Name = fextl::fmt::format("Dispatch_{}", FHU::Syscalls::gettid());
    CTX->Symbols.Register(reinterpret_cast<void*>(Start), End-Start, Name);
  }
  if (CTX->Config.GlobalJITNaming()) {
    CTX->Symbols.RegisterJITSpace(reinterpret_cast<void*>(Start), End-Start);
  }
+
+}
+
+size_t X86Dispatcher::GenerateGDBPauseCheck(uint8_t *CodeBuffer, uint64_t GuestRIP) {
+  using namespace Xbyak;
+  using namespace Xbyak::util;
+
+  Xbyak::CodeGenerator emit(1, &emit); // actual emit target set with setNewBuffer
+  emit.setNewBuffer(CodeBuffer, MaxGDBPauseCheckSize);
+
+  Label RunBlock;
+
+  // If we have a gdb server running then run in a less efficient mode that checks if we need to exit
+  // This happens when single stepping
+  static_assert(sizeof(CTX->Config.RunningMode) == 4, "This is expected to be size of 4");
+  emit.mov(rax, reinterpret_cast<uint64_t>(CTX));
+
+  // If the value == 0 then we don't need to stop
+  emit.cmp(dword [rax + (offsetof(FEXCore::Context::ContextImpl, Config.RunningMode))], 0);
+  emit.je(RunBlock);
+  {
+    // Make sure RIP is syncronized to the context
+    emit.mov(rax, GuestRIP);
+    emit.mov(qword STATE_PTR(CpuStateFrame, State.rip), rax);
+
+    // Stop the thread
+    emit.mov(rax, qword STATE_PTR(CpuStateFrame, Pointers.Common.ThreadPauseHandlerSpillSRA));
+    emit.jmp(rax);
+  }
+
+  emit.L(RunBlock);
+
+  emit.ready();
+
+  return emit.getSize();
+}
+
+
+size_t X86Dispatcher::GenerateInterpreterTrampoline(uint8_t *CodeBuffer) {
+  using namespace Xbyak;
+  using namespace Xbyak::util;
+
+  Xbyak::CodeGenerator emit(1, &emit); // actual emit target set with setNewBuffer
+  emit.setNewBuffer(CodeBuffer, MaxInterpreterTrampolineSize);
+
+  Label InlineIRData;
+
+  emit.mov(rdi, STATE);
+  emit.lea(rsi, ptr[rip + InlineIRData]);
+  emit.call(qword STATE_PTR(CpuStateFrame, Pointers.Interpreter.FragmentExecuter));
+
+  emit.jmp(qword STATE_PTR(CpuStateFrame, Pointers.Common.DispatcherLoopTop));
+
+  emit.L(InlineIRData);
+
+  emit.ready();
+
+  return emit.getSize();
 }

 X86Dispatcher::~X86Dispatcher() {
-  FEXCore::Allocator::munmap(top_, MAX_DISPATCHER_CODE_SIZE);
+  FEXCore::Allocator::VirtualFree(top_, MAX_DISPATCHER_CODE_SIZE);
 }

-#ifdef _M_X86_64
+void X86Dispatcher::InitThreadPointers(FEXCore::Core::InternalThreadState *Thread) {
+  // Setup dispatcher specific pointers that need to be accessed from JIT code
+  {
+    auto &Common = Thread->CurrentFrame->Pointers.Common;

-void InterpreterCore::CreateAsmDispatch(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread) {
-  DispatcherConfig config;
-  config.ExecuteBlocksWithCall = true;
+    Common.DispatcherLoopTop = AbsoluteLoopTopAddress;
+    Common.DispatcherLoopTopFillSRA = AbsoluteLoopTopAddressFillSRA;
+    Common.ExitFunctionLinker = ExitFunctionLinkerAddress;
+    Common.ThreadStopHandlerSpillSRA = ThreadStopHandlerAddress;
+    Common.ThreadPauseHandlerSpillSRA = ThreadPauseHandlerAddress;
+    Common.GuestSignal_SIGILL = GuestSignal_SIGILL;
+    Common.GuestSignal_SIGTRAP = GuestSignal_SIGTRAP;
+    Common.GuestSignal_SIGSEGV = GuestSignal_SIGSEGV;
+    Common.SignalReturnHandler = SignalHandlerReturnAddress;
+    Common.SignalReturnHandlerRT = SignalHandlerReturnAddressRT;

-  Dispatcher = std::make_unique<X86Dispatcher>(ctx, Thread, config);
-  DispatchPtr = Dispatcher->DispatchPtr;
-  CallbackPtr = Dispatcher->CallbackPtr;
-
-  // TODO: It feels wrong to initialize this way
-  ctx->InterpreterCallbackReturn = Dispatcher->ReturnPtr;
+    auto &Interpreter = Thread->CurrentFrame->Pointers.Interpreter;
+    (uintptr_t&)Interpreter.CallbackReturn = IntCallbackReturnAddress;
+  }
 }

-#endif
+fextl::unique_ptr<Dispatcher> Dispatcher::CreateX86(FEXCore::Context::ContextImpl *CTX, const DispatcherConfig &Config) {
+  return fextl::make_unique<X86Dispatcher>(CTX, Config);
+}

 }
@@ -1,13 +1,24 @@
 #pragma once

+#include <FEXCore/fextl/list.h>
+#include <FEXCore/fextl/unordered_map.h>
+#include <FEXCore/fextl/unordered_set.h>
+
 #include "Interface/Core/Dispatcher/Dispatcher.h"

 #define XBYAK64
-#include <xbyak/xbyak.h>
+#define XBYAK_CUSTOM_ALLOC
+#define XBYAK_CUSTOM_MALLOC FEXCore::Allocator::malloc
+#define XBYAK_CUSTOM_FREE FEXCore::Allocator::free
+#define XBYAK_CUSTOM_SETS
+#define XBYAK_STD_UNORDERED_SET fextl::unordered_set
+#define XBYAK_STD_UNORDERED_MAP fextl::unordered_map
+#define XBYAK_STD_UNORDERED_MULTIMAP fextl::unordered_multimap
+#define XBYAK_STD_LIST fextl::list
+#define XBYAK_NO_EXCEPTION

-namespace FEXCore::Context {
-  struct Context;
-}
+#include <xbyak/xbyak.h>
+#include <xbyak/xbyak_util.h>

 namespace FEXCore::Core {
  struct InternalThreadState;
@@ -17,7 +28,10 @@ namespace FEXCore::CPU {

 class X86Dispatcher final : public Dispatcher, public Xbyak::CodeGenerator {
  public:
-    X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, DispatcherConfig &config);
+    X86Dispatcher(FEXCore::Context::ContextImpl *ctx, const DispatcherConfig &config);
+    void InitThreadPointers(FEXCore::Core::InternalThreadState *Thread) override;
+    size_t GenerateGDBPauseCheck(uint8_t *CodeBuffer, uint64_t GuestRIP) override;
+    size_t GenerateInterpreterTrampoline(uint8_t *CodeBuffer) override;

    virtual ~X86Dispatcher() override;
 };
@@ -7,20 +7,20 @@ $end_info$

 #include "Interface/Context/Context.h"
 #include "Interface/Core/Frontend.h"
+#include "Interface/Core/X86Tables/X86Tables.h"

 #include <array>
-#include <assert.h>
 #include <algorithm>
 #include <cstring>
 #include <FEXCore/Config/Config.h>
 #include <FEXCore/Core/X86Enums.h>
-#include <FEXCore/Debug/X86Tables.h>
 #include <FEXCore/HLE/SyscallHandler.h>
 #include <FEXCore/Utils/Allocator.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/Profiler.h>
 #include <FEXCore/Utils/Telemetry.h>
-#include <set>
-#include <sys/mman.h>
+#include <FEXCore/fextl/set.h>
+#include <FEXHeaderUtils/TypeDefines.h>

 namespace FEXCore::Frontend {
 #include "Interface/Core/VSyscall/VSyscall.inc"
@@ -30,26 +30,6 @@ using namespace FEXCore::X86Tables;
 static uint32_t MapModRMToReg(uint8_t REX, uint8_t bits, bool HighBits, bool HasREX, bool HasXMM, bool HasMM, uint8_t InvalidOffset = 16) {
  using GPRArray = std::array<uint32_t, 16>;

-  static constexpr GPRArray GPRIndexes = {
-    // Classical ordering?
-    FEXCore::X86State::REG_RAX,
-    FEXCore::X86State::REG_RCX,
-    FEXCore::X86State::REG_RDX,
-    FEXCore::X86State::REG_RBX,
-    FEXCore::X86State::REG_RSP,
-    FEXCore::X86State::REG_RBP,
-    FEXCore::X86State::REG_RSI,
-    FEXCore::X86State::REG_RDI,
-    FEXCore::X86State::REG_R8,
-    FEXCore::X86State::REG_R9,
-    FEXCore::X86State::REG_R10,
-    FEXCore::X86State::REG_R11,
-    FEXCore::X86State::REG_R12,
-    FEXCore::X86State::REG_R13,
-    FEXCore::X86State::REG_R14,
-    FEXCore::X86State::REG_R15,
-  };
-
  static constexpr GPRArray GPR8BitHighIndexes = {
    // Classical ordering?
    FEXCore::X86State::REG_RAX,
@@ -70,129 +50,46 @@ static uint32_t MapModRMToReg(uint8_t REX, uint8_t bits, bool HighBits, bool Has
    FEXCore::X86State::REG_R15,
  };

-  static constexpr GPRArray XMMIndexes = {
-    FEXCore::X86State::REG_XMM_0,
-    FEXCore::X86State::REG_XMM_1,
-    FEXCore::X86State::REG_XMM_2,
-    FEXCore::X86State::REG_XMM_3,
-    FEXCore::X86State::REG_XMM_4,
-    FEXCore::X86State::REG_XMM_5,
-    FEXCore::X86State::REG_XMM_6,
-    FEXCore::X86State::REG_XMM_7,
-    FEXCore::X86State::REG_XMM_8,
-    FEXCore::X86State::REG_XMM_9,
-    FEXCore::X86State::REG_XMM_10,
-    FEXCore::X86State::REG_XMM_11,
-    FEXCore::X86State::REG_XMM_12,
-    FEXCore::X86State::REG_XMM_13,
-    FEXCore::X86State::REG_XMM_14,
-    FEXCore::X86State::REG_XMM_15,
-  };
-
-  static constexpr GPRArray MMIndexes = {
-    FEXCore::X86State::REG_MM_0,
-    FEXCore::X86State::REG_MM_1,
-    FEXCore::X86State::REG_MM_2,
-    FEXCore::X86State::REG_MM_3,
-    FEXCore::X86State::REG_MM_4,
-    FEXCore::X86State::REG_MM_5,
-    FEXCore::X86State::REG_MM_6,
-    FEXCore::X86State::REG_MM_7,
-    FEXCore::X86State::REG_INVALID,
-    FEXCore::X86State::REG_INVALID,
-    FEXCore::X86State::REG_INVALID,
-    FEXCore::X86State::REG_INVALID,
-    FEXCore::X86State::REG_INVALID,
-    FEXCore::X86State::REG_INVALID,
-    FEXCore::X86State::REG_INVALID,
-    FEXCore::X86State::REG_INVALID
-  };
-
-  const GPRArray *GPRs = &GPRIndexes;
-  if (HasXMM) {
-    GPRs = &XMMIndexes;
-  }
-  else if (HasMM) {
-    GPRs = &MMIndexes;
-  }
-  else if (HighBits && !HasREX) {
-    GPRs = &GPR8BitHighIndexes;
-  }
-
  uint8_t Offset = (REX << 3) | bits;

  if (Offset == InvalidOffset) {
    return FEXCore::X86State::REG_INVALID;
  }
-  return (*GPRs)[(REX << 3) | bits];
+
+  if (HasXMM) {
+    return FEXCore::X86State::REG_XMM_0 + Offset;
+  }
+  else if (HasMM) {
+    return FEXCore::X86State::REG_MM_0 + Offset;
+  }
+  else if (!(HighBits && !HasREX)) {
+    return FEXCore::X86State::REG_RAX + Offset;
+  }
+
+  return GPR8BitHighIndexes[Offset];
 }

 static uint32_t MapVEXToReg(uint8_t vvvv, bool HasXMM) {
-  using GPRArray = std::array<uint32_t, 16>;
-
-  static constexpr GPRArray GPRIndexes = {
-    FEXCore::X86State::REG_RAX,
-    FEXCore::X86State::REG_RCX,
-    FEXCore::X86State::REG_RDX,
-    FEXCore::X86State::REG_RBX,
-    FEXCore::X86State::REG_RSP,
-    FEXCore::X86State::REG_RBP,
-    FEXCore::X86State::REG_RSI,
-    FEXCore::X86State::REG_RDI,
-    FEXCore::X86State::REG_R8,
-    FEXCore::X86State::REG_R9,
-    FEXCore::X86State::REG_R10,
-    FEXCore::X86State::REG_R11,
-    FEXCore::X86State::REG_R12,
-    FEXCore::X86State::REG_R13,
-    FEXCore::X86State::REG_R14,
-    FEXCore::X86State::REG_R15,
-  };
-
-  static constexpr GPRArray XMMIndexes = {
-    FEXCore::X86State::REG_XMM_0,
-    FEXCore::X86State::REG_XMM_1,
-    FEXCore::X86State::REG_XMM_2,
-    FEXCore::X86State::REG_XMM_3,
-    FEXCore::X86State::REG_XMM_4,
-    FEXCore::X86State::REG_XMM_5,
-    FEXCore::X86State::REG_XMM_6,
-    FEXCore::X86State::REG_XMM_7,
-    FEXCore::X86State::REG_XMM_8,
-    FEXCore::X86State::REG_XMM_9,
-    FEXCore::X86State::REG_XMM_10,
-    FEXCore::X86State::REG_XMM_11,
-    FEXCore::X86State::REG_XMM_12,
-    FEXCore::X86State::REG_XMM_13,
-    FEXCore::X86State::REG_XMM_14,
-    FEXCore::X86State::REG_XMM_15,
-  };
-
  if (HasXMM) {
-    return XMMIndexes[vvvv];
+    return FEXCore::X86State::REG_XMM_0 + vvvv;
  } else {
-    return GPRIndexes[vvvv];
+    return FEXCore::X86State::REG_RAX + vvvv;
  }
 }

-Decoder::Decoder(FEXCore::Context::Context *ctx)
+Decoder::Decoder(FEXCore::Context::ContextImpl *ctx)
  : CTX {ctx}
-  , OSABI { ctx->SyscallHandler ? ctx->SyscallHandler->GetOSABI() : FEXCore::HLE::SyscallOSABI::OS_UNKNOWN } {
-  // Using mmap is a start-up time optimization
-  // Take advantage of page faulting to reduce startup time for minimal runtime cost
-  DecodedBuffer =
-    reinterpret_cast<FEXCore::X86Tables::DecodedInst *>(
-      FEXCore::Allocator::mmap(0, sizeof(FEXCore::X86Tables::DecodedInst) * DefaultDecodedBufferSize,
-      PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0));
+  , OSABI { ctx->SyscallHandler ? ctx->SyscallHandler->GetOSABI() : FEXCore::HLE::SyscallOSABI::OS_UNKNOWN }
+  , PoolObject {ctx->FrontendAllocator, sizeof(FEXCore::X86Tables::DecodedInst) * DefaultDecodedBufferSize} {
 }

 Decoder::~Decoder() {
-  FEXCore::Allocator::munmap(DecodedBuffer, sizeof(FEXCore::X86Tables::DecodedInst) * DefaultDecodedBufferSize);
+  PoolObject.UnclaimBuffer();
 }

 uint8_t Decoder::ReadByte() {
  uint8_t Byte = InstStream[InstructionSize];
-  LOGMAN_THROW_A_FMT(InstructionSize < MAX_INST_SIZE, "Max instruction size exceeded!");
+  LOGMAN_THROW_AA_FMT(InstructionSize < MAX_INST_SIZE, "Max instruction size exceeded!");
  Instruction[InstructionSize] = Byte;
  InstructionSize++;
  return Byte;
@@ -204,19 +101,12 @@ uint8_t Decoder::PeekByte(uint8_t Offset) const {
 }

 uint64_t Decoder::ReadData(uint8_t Size) {
-  if (Size == 0) {
-    return 0;
-  }
-
-  if (Size > sizeof(uint64_t)) {
-    LOGMAN_MSG_A_FMT("Unknown data size to read");
-    return 0;
-  }
+  LOGMAN_THROW_AA_FMT(Size != 0 && Size <= sizeof(uint64_t), "Unknown data size to read");

  uint64_t Res = 0;
  std::memcpy(&Res, &InstStream[InstructionSize], Size);

-#ifndef NDEBUG
+#if defined(ASSERTIONS_ENABLED) && ASSERTIONS_ENABLED
  for(size_t i = 0; i < Size; ++i) {
    ReadByte();
  }
@@ -348,16 +238,29 @@ void Decoder::DecodeModRM_64(X86Tables::DecodedOperand *Operand, X86Tables::ModR
    Operand->Data.SIB.Scale = 1 << SIB.scale;

    // The invalid encoding types are described at Table 1-12. "promoted nsigned is always non-zero"
-    Operand->Data.SIB.Index = MapModRMToReg(DecodeInst->Flags & DecodeFlags::FLAG_REX_XGPR_X ? 1 : 0, SIB.index, false, false, false, false, 0b100);
-    Operand->Data.SIB.Base  = MapModRMToReg(DecodeInst->Flags & DecodeFlags::FLAG_REX_XGPR_B ? 1 : 0, SIB.base, false, false, false, false, ModRM.mod == 0 ? 0b101 : 16);
+    {
+      // If we have a VSIB byte (as opposed to SIB), then the index register is a vector.
+      const bool IsIndexVector = (DecodeInst->TableInfo->Flags & InstFlags::FLAGS_VEX_VSIB) != 0;
+      if (IsIndexVector) {
+        DecodeInst->Flags |= X86Tables::DecodeFlags::FLAG_VSIB_BYTE;
+      }

-    LOGMAN_THROW_A_FMT(Displacement <= 4, "Number of bytes should be <= 4 for literal src");
+      const uint8_t IndexREX = (DecodeInst->Flags & DecodeFlags::FLAG_REX_XGPR_X) != 0 ? 1 : 0;
+      const uint8_t BaseREX = (DecodeInst->Flags & DecodeFlags::FLAG_REX_XGPR_B) != 0 ? 1 : 0;

-    uint64_t Literal = ReadData(Displacement);
-    if (Displacement == 1) {
-      Literal = static_cast<int8_t>(Literal);
+      Operand->Data.SIB.Index = MapModRMToReg(IndexREX, SIB.index, false, false, IsIndexVector, false, 0b100);
+      Operand->Data.SIB.Base  = MapModRMToReg(BaseREX, SIB.base, false, false, false, false, ModRM.mod == 0 ? 0b101 : 16);
+    }
+
+    LOGMAN_THROW_AA_FMT(Displacement <= 4, "Number of bytes should be <= 4 for literal src");
+
+    if (Displacement) {
+      uint64_t Literal = ReadData(Displacement);
+      if (Displacement == 1) {
+        Literal = static_cast<int8_t>(Literal);
+      }
+      Operand->Data.SIB.Offset = Literal;
    }
-    Operand->Data.SIB.Offset = Literal;
  }
  else if (ModRM.mod == 0) {
    // Explained in Table 1-14. "Operand Addressing Using ModRM and SIB Bytes"
@@ -392,12 +295,6 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
  DecodeInst->OP = Op;
  DecodeInst->TableInfo = Info;

-  // XXX: Once we support 32bit x86 then this will be necessary to support
-  if (Info->Type == FEXCore::X86Tables::TYPE_LEGACY_PREFIX) {
-    LogMan::Msg::DFmt("Legacy Prefix");
-    return false;
-  }
-
  if (Info->Type == FEXCore::X86Tables::TYPE_UNKNOWN) {
    LogMan::Msg::DFmt("Unknown instruction: {} 0x{:04x} 0x{:x}", Info->Name ?: "UND", Op, DecodeInst->PC);
    return false;
@@ -408,7 +305,7 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
    return false;
  }

-  LOGMAN_THROW_A_FMT(!(Info->Type >= FEXCore::X86Tables::TYPE_GROUP_1 && Info->Type <= FEXCore::X86Tables::TYPE_GROUP_P),
+  LOGMAN_THROW_AA_FMT(!(Info->Type >= FEXCore::X86Tables::TYPE_GROUP_1 && Info->Type <= FEXCore::X86Tables::TYPE_GROUP_P),
                     "Group Ops should have been decoded before this!");

  uint8_t DestSize{};
@@ -416,27 +313,35 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
                                       (Options.w && CTX->Config.Is64BitMode);
  const bool HasNarrowingDisplacement = (FEXCore::X86Tables::DecodeFlags::GetOpAddr(DecodeInst->Flags, 0) & FEXCore::X86Tables::DecodeFlags::FLAG_OPERAND_SIZE_LAST) != 0;

-  bool HasXMMSrc = !!(Info->Flags & FEXCore::X86Tables::InstFlags::FLAGS_XMM_FLAGS) &&
-    !HAS_XMM_SUBFLAG(Info->Flags, FEXCore::X86Tables::InstFlags::FLAGS_SF_SRC_GPR) &&
-    !HAS_XMM_SUBFLAG(Info->Flags, FEXCore::X86Tables::InstFlags::FLAGS_SF_MMX_SRC);
-  bool HasXMMDst = !!(Info->Flags & FEXCore::X86Tables::InstFlags::FLAGS_XMM_FLAGS) &&
-    !HAS_XMM_SUBFLAG(Info->Flags, FEXCore::X86Tables::InstFlags::FLAGS_SF_DST_GPR) &&
-    !HAS_XMM_SUBFLAG(Info->Flags, FEXCore::X86Tables::InstFlags::FLAGS_SF_MMX_DST);
-  bool HasMMSrc = !!(Info->Flags & FEXCore::X86Tables::InstFlags::FLAGS_XMM_FLAGS) &&
-    !HAS_XMM_SUBFLAG(Info->Flags, FEXCore::X86Tables::InstFlags::FLAGS_SF_SRC_GPR) &&
-    HAS_XMM_SUBFLAG(Info->Flags, FEXCore::X86Tables::InstFlags::FLAGS_SF_MMX_SRC);
-  bool HasMMDst = !!(Info->Flags & FEXCore::X86Tables::InstFlags::FLAGS_XMM_FLAGS) &&
-    !HAS_XMM_SUBFLAG(Info->Flags, FEXCore::X86Tables::InstFlags::FLAGS_SF_DST_GPR) &&
-    HAS_XMM_SUBFLAG(Info->Flags, FEXCore::X86Tables::InstFlags::FLAGS_SF_MMX_DST);
+  const bool HasXMMFlags = (Info->Flags & InstFlags::FLAGS_XMM_FLAGS) != 0;
+  bool HasXMMSrc = HasXMMFlags &&
+    !HAS_XMM_SUBFLAG(Info->Flags, InstFlags::FLAGS_SF_SRC_GPR) &&
+    !HAS_XMM_SUBFLAG(Info->Flags, InstFlags::FLAGS_SF_MMX_SRC);
+  bool HasXMMDst = HasXMMFlags &&
+    !HAS_XMM_SUBFLAG(Info->Flags, InstFlags::FLAGS_SF_DST_GPR) &&
+    !HAS_XMM_SUBFLAG(Info->Flags, InstFlags::FLAGS_SF_MMX_DST);
+  bool HasMMSrc = HasXMMFlags &&
+    !HAS_XMM_SUBFLAG(Info->Flags, InstFlags::FLAGS_SF_SRC_GPR) &&
+    HAS_XMM_SUBFLAG(Info->Flags, InstFlags::FLAGS_SF_MMX_SRC);
+  bool HasMMDst = HasXMMFlags &&
+    !HAS_XMM_SUBFLAG(Info->Flags, InstFlags::FLAGS_SF_DST_GPR) &&
+    HAS_XMM_SUBFLAG(Info->Flags, InstFlags::FLAGS_SF_MMX_DST);

  // Is ModRM present via explicit instruction encoded or REX?
  const bool HasMODRM = !!(Info->Flags & FEXCore::X86Tables::InstFlags::FLAGS_MODRM);

  const bool HasREX = !!(DecodeInst->Flags & DecodeFlags::FLAG_REX_PREFIX);
-  const bool HasHighXMM = HAS_XMM_SUBFLAG(Info->Flags, FEXCore::X86Tables::InstFlags::FLAGS_SF_HIGH_XMM_REG);
  const bool Has16BitAddressing = !CTX->Config.Is64BitMode &&
    DecodeInst->Flags & DecodeFlags::FLAG_ADDRESS_SIZE;

+  // This is used for ModRM register modification
+  // For both modrm.reg and modrm.rm(when mod == 0b11) when value is >= 0b100
+  // then it changes from expected registers to the high 8bits of the lower registers
+  // Bit annoying to support
+  // In the case of no modrm (REX in byte situation) then it is unaffected
+  bool Is8BitSrc{};
+  bool Is8BitDest{};
+
  // If we require ModRM and haven't decoded it yet, do it now
  // Some instructions have to read modrm upfront, others do it later
  if (HasMODRM && !DecodeInst->DecodedModRM) {
@@ -453,14 +358,24 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
    if (DstSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_8BIT) {
      DecodeInst->Flags |= DecodeFlags::GenSizeDstSize(DecodeFlags::SIZE_8BIT);
      DestSize = 1;
+      Is8BitDest = true;
    }
    else if (DstSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_16BIT) {
      DecodeInst->Flags |= DecodeFlags::GenSizeDstSize(DecodeFlags::SIZE_16BIT);
      DestSize = 2;
    }
    else if (DstSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_128BIT) {
-      DecodeInst->Flags |= DecodeFlags::GenSizeDstSize(DecodeFlags::SIZE_128BIT);
-      DestSize = 16;
+      if (Options.L) {
+        DecodeInst->Flags |= DecodeFlags::GenSizeDstSize(DecodeFlags::SIZE_256BIT);
+        DestSize = 32;
+      } else {
+        DecodeInst->Flags |= DecodeFlags::GenSizeDstSize(DecodeFlags::SIZE_128BIT);
+        DestSize = 16;
+      }
+    }
+    else if (DstSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_256BIT) {
+      DecodeInst->Flags |= DecodeFlags::GenSizeDstSize(DecodeFlags::SIZE_256BIT);
+      DestSize = 32;
    }
    else if (HasNarrowingDisplacement &&
      (DstSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_DEF ||
@@ -486,12 +401,20 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
    // Decode sources
    if (SrcSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_8BIT) {
      DecodeInst->Flags |= DecodeFlags::GenSizeSrcSize(DecodeFlags::SIZE_8BIT);
+      Is8BitSrc = true;
    }
    else if (SrcSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_16BIT) {
      DecodeInst->Flags |= DecodeFlags::GenSizeSrcSize(DecodeFlags::SIZE_16BIT);
    }
    else if (SrcSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_128BIT) {
-      DecodeInst->Flags |= DecodeFlags::GenSizeSrcSize(DecodeFlags::SIZE_128BIT);
+      if (Options.L) {
+        DecodeInst->Flags |= DecodeFlags::GenSizeSrcSize(DecodeFlags::SIZE_256BIT);
+      } else {
+        DecodeInst->Flags |= DecodeFlags::GenSizeSrcSize(DecodeFlags::SIZE_128BIT);
+      }
+    }
+    else if (SrcSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_256BIT) {
+      DecodeInst->Flags |= DecodeFlags::GenSizeSrcSize(DecodeFlags::SIZE_256BIT);
    }
    else if (HasNarrowingDisplacement &&
      (SrcSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_DEF ||
@@ -512,14 +435,6 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
    }
  }

-  // This is used for ModRM register modification
-  // For both modrm.reg and modrm.rm(when mod == 0b11) when value is >= 0b100
-  // then it changes from expected registers to the high 8bits of the lower registers
-  // Bit annoying to support
-  // In the case of no modrm (REX in byte situation) then it is unaffected
-  const bool Is8BitSrc = (DecodeFlags::GetSizeSrcFlags(DecodeInst->Flags) == DecodeFlags::SIZE_8BIT);
-  const bool Is8BitDest = (DecodeFlags::GetSizeDstFlags(DecodeInst->Flags) == DecodeFlags::SIZE_8BIT);
-
  auto *CurrentDest = &DecodeInst->Dest;

  if (HAS_NON_XMM_SUBFLAG(Info->Flags, FEXCore::X86Tables::InstFlags::FLAGS_SF_DST_RAX) ||
@@ -530,16 +445,15 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
    CurrentDest->Data.GPR.GPR = HAS_NON_XMM_SUBFLAG(Info->Flags, FEXCore::X86Tables::InstFlags::FLAGS_SF_DST_RAX) ? FEXCore::X86State::REG_RAX : FEXCore::X86State::REG_RDX;
    CurrentDest = &DecodeInst->Src[0];
  }
-
-  if (HAS_NON_XMM_SUBFLAG(Info->Flags, FEXCore::X86Tables::InstFlags::FLAGS_SF_REX_IN_BYTE)) {
-    LOGMAN_THROW_A_FMT(!HasMODRM, "This instruction shouldn't have ModRM!");
+  else if (HAS_NON_XMM_SUBFLAG(Info->Flags, FEXCore::X86Tables::InstFlags::FLAGS_SF_REX_IN_BYTE)) {
+    LOGMAN_THROW_AA_FMT(!HasMODRM, "This instruction shouldn't have ModRM!");

    // If the REX is in the byte that means the lower nibble of the OP contains the destination GPR
    // This also means that the destination is always a GPR on these ones
    // ADDITIONALLY:
    // If there is a REX prefix then that allows extended GPR usage
    CurrentDest->Type = DecodedOperand::OpType::GPR;
-    DecodeInst->Dest.Data.GPR.HighBits = (Is8BitDest && !HasREX && (Op & 0b111) >= 0b100) || HasHighXMM;
+    DecodeInst->Dest.Data.GPR.HighBits = (Is8BitDest && !HasREX && (Op & 0b111) >= 0b100);
    CurrentDest->Data.GPR.GPR = MapModRMToReg(DecodeInst->Flags & DecodeFlags::FLAG_REX_XGPR_B ? 1 : 0, Op & 0b111, Is8BitDest, HasREX, false, false);

    if (CurrentDest->Data.GPR.GPR == FEXCore::X86State::REG_INVALID)
@@ -567,7 +481,7 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,

    // Decode the GPR source first
    GPR.Type = DecodedOperand::OpType::GPR;
-    GPR.Data.GPR.HighBits = (GPR8Bit && ModRM.reg >= 0b100 && !HasREX) || HasHighXMM;
+    GPR.Data.GPR.HighBits = (GPR8Bit && ModRM.reg >= 0b100 && !HasREX);
    GPR.Data.GPR.GPR = MapModRMToReg(DecodeInst->Flags & DecodeFlags::FLAG_REX_XGPR_R ? 1 : 0, ModRM.reg, GPR8Bit, HasREX, HasXMMGPR, HasMMGPR);

    if (GPR.Data.GPR.GPR == FEXCore::X86State::REG_INVALID)
@@ -577,14 +491,17 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
    // ModRM.Mod != 0b11 == Register-direct addressing
    if (ModRM.mod == 0b11) {
      NonGPR.Type = DecodedOperand::OpType::GPR;
-      NonGPR.Data.GPR.HighBits = (NonGPR8Bit && ModRM.rm >= 0b100 && !HasREX) || HasHighXMM;
+      NonGPR.Data.GPR.HighBits = (NonGPR8Bit && ModRM.rm >= 0b100 && !HasREX);
      NonGPR.Data.GPR.GPR = MapModRMToReg(DecodeInst->Flags & DecodeFlags::FLAG_REX_XGPR_B ? 1 : 0, ModRM.rm, NonGPR8Bit, HasREX, HasXMMNonGPR, HasMMNonGPR);
      if (NonGPR.Data.GPR.GPR == FEXCore::X86State::REG_INVALID)
        return false;
    }
    else {
-      auto Disp = DecodeModRMs_Disp[Has16BitAddressing];
-      (this->*Disp)(&NonGPR, ModRM);
+      // Only decode if we haven't pre-decoded
+      if (NonGPR.IsNone()) {
+        auto Disp = DecodeModRMs_Disp[Has16BitAddressing];
+        (this->*Disp)(&NonGPR, ModRM);
+      }
    }

    return true;
@@ -595,7 +512,12 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
  if ((Info->Flags & FEXCore::X86Tables::InstFlags::FLAGS_VEX_1ST_SRC) != 0) {
    DecodeInst->Src[CurrentSrc].Type = DecodedOperand::OpType::GPR;
    DecodeInst->Src[CurrentSrc].Data.GPR.HighBits = false;
-    DecodeInst->Src[CurrentSrc].Data.GPR.GPR = MapVEXToReg(Options.vvvv, HasXMMSrc);
+
+    // If we have XMM flags at all, then SRC 1 cannot be a GPR. The only case where
+    // this is possible is with BMI1 and BMI2 instructions (which are all GPR-based
+    // and don't use XMM flags)
+    DecodeInst->Src[CurrentSrc].Data.GPR.GPR = MapVEXToReg(Options.vvvv, HasXMMFlags);
+
    ++CurrentSrc;
  }

@@ -638,7 +560,7 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
  }

  if (Bytes != 0) {
-    LOGMAN_THROW_A_FMT(Bytes <= 8, "Number of bytes should be <= 8 for literal src");
+    LOGMAN_THROW_AA_FMT(Bytes <= 8, "Number of bytes should be <= 8 for literal src");

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

@@ -663,7 +585,7 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
    DecodeInst->Src[CurrentSrc].Data.Literal.Value = Literal;
  }

-  LOGMAN_THROW_A_FMT(Bytes == 0, "Inst at 0x{:x}: 0x{:04x} '{}' Had an instruction of size {} with {} remaining",
+  LOGMAN_THROW_AA_FMT(Bytes == 0, "Inst at 0x{:x}: 0x{:04x} '{}' Had an instruction of size {} with {} remaining",
                     DecodeInst->PC, DecodeInst->OP, DecodeInst->TableInfo->Name ?: "UND", InstructionSize, Bytes);
  DecodeInst->InstSize = InstructionSize;
  return true;
@@ -673,12 +595,6 @@ bool Decoder::NormalOpHeader(FEXCore::X86Tables::X86InstInfo const *Info, uint16
  DecodeInst->OP = Op;
  DecodeInst->TableInfo = Info;

-  // XXX: Once we support 32bit x86 then this will be necessary to support
-  if (Info->Type == FEXCore::X86Tables::TYPE_LEGACY_PREFIX) {
-    LogMan::Msg::DFmt("Legacy Prefix");
-    return false;
-  }
-
  if (Info->Type == FEXCore::X86Tables::TYPE_UNKNOWN) {
    LogMan::Msg::DFmt("Unknown instruction: {} 0x{:04x} 0x{:x}", Info->Name ?: "UND", Op, DecodeInst->PC);
    return false;
@@ -689,10 +605,14 @@ bool Decoder::NormalOpHeader(FEXCore::X86Tables::X86InstInfo const *Info, uint16
    return false;
  }

-  LOGMAN_THROW_A_FMT(Info->Type != FEXCore::X86Tables::TYPE_REX_PREFIX,
+  LOGMAN_THROW_AA_FMT(Info->Type != FEXCore::X86Tables::TYPE_REX_PREFIX,
                     "REX PREFIX should have been decoded before this!");

-  if (Info->Type >= FEXCore::X86Tables::TYPE_GROUP_1 &&
+  // A normal instruction is the most likely.
+  if (Info->Type == FEXCore::X86Tables::TYPE_INST) [[likely]] {
+    return NormalOp(Info, Op);
+  }
+  else if (Info->Type >= FEXCore::X86Tables::TYPE_GROUP_1 &&
      Info->Type <= FEXCore::X86Tables::TYPE_GROUP_11) {
    uint8_t ModRMByte = ReadByte();
    DecodeInst->ModRM = ModRMByte;
@@ -746,7 +666,7 @@ bool Decoder::NormalOpHeader(FEXCore::X86Tables::X86InstInfo const *Info, uint16
        3,
      };
      uint8_t Field = RegToField[ModRM.reg];
-      LOGMAN_THROW_A_FMT(Field != 255, "Invalid field selected!");
+      LOGMAN_THROW_AA_FMT(Field != 255, "Invalid field selected!");

      LocalOp = (Field << 3) | ModRM.rm;
      return NormalOp(&SecondModRMTableOps[LocalOp], LocalOp);
@@ -761,9 +681,6 @@ bool Decoder::NormalOpHeader(FEXCore::X86Tables::X86InstInfo const *Info, uint16
    DecodeInst->ModRM = ModRMByte;
    DecodeInst->DecodedModRM = true;

-    FEXCore::X86Tables::ModRMDecoded ModRM;
-    ModRM.Hex = DecodeInst->ModRM;
-
    uint16_t X87Op = ((Op - 0xD8) << 8) | ModRMByte;
    return NormalOp(&X87Ops[X87Op], X87Op);
  }
@@ -782,6 +699,7 @@ bool Decoder::NormalOpHeader(FEXCore::X86Tables::X86InstInfo const *Info, uint16
    if (Op == 0xC5) { // Two byte VEX
      pp = Byte1 & 0b11;
      options.vvvv = 15 - ((Byte1 & 0b01111000) >> 3);
+      options.L = (Byte1 & 0b100) != 0;
    }
    else { // 0xC4 = Three byte VEX
      const uint8_t Byte2 = ReadByte();
@@ -789,6 +707,7 @@ bool Decoder::NormalOpHeader(FEXCore::X86Tables::X86InstInfo const *Info, uint16
      map_select = Byte1 & 0b11111;
      options.vvvv = 15 - ((Byte2 & 0b01111000) >> 3);
      options.w = (Byte2 & 0b10000000) != 0;
+      options.L = (Byte2 & 0b100) != 0;
      if ((Byte1 & 0b01000000) == 0) {
        LOGMAN_THROW_A_FMT(CTX->Config.Is64BitMode, "VEX.X shouldn't be 0 in 32-bit mode!");
        DecodeInst->Flags |= DecodeFlags::FLAG_REX_XGPR_X;
@@ -838,7 +757,8 @@ bool Decoder::NormalOpHeader(FEXCore::X86Tables::X86InstInfo const *Info, uint16
    return NormalOp(&EVEXTableOps[EVEXOp], EVEXOp);
  }

-  return NormalOp(Info, Op);
+  LOGMAN_MSG_A_FMT("Invalid instruction decoding type");
+  FEX_UNREACHABLE;
 }

 bool Decoder::DecodeInstruction(uint64_t PC) {
@@ -857,102 +777,106 @@ bool Decoder::DecodeInstruction(uint64_t PC) {
    case 0x0F: {// Escape Op
      uint8_t EscapeOp = ReadByte();
      switch (EscapeOp) {
-      case 0x0F: { // 3DNow!
-        // 3DNow! Instruction Encoding: 0F 0F [ModRM] [SIB] [Displacement] [Opcode]
-        // Decode ModRM
-        uint8_t ModRMByte = ReadByte();
-        DecodeInst->ModRM = ModRMByte;
-        DecodeInst->DecodedModRM = true;
+        case 0x0F: [[unlikely]] { // 3DNow!
+          // 3DNow! Instruction Encoding: 0F 0F [ModRM] [SIB] [Displacement] [Opcode]
+          // Decode ModRM
+          uint8_t ModRMByte = ReadByte();
+          DecodeInst->ModRM = ModRMByte;
+          DecodeInst->DecodedModRM = true;

-        FEXCore::X86Tables::ModRMDecoded ModRM;
-        ModRM.Hex = DecodeInst->ModRM;
+          FEXCore::X86Tables::ModRMDecoded ModRM;
+          ModRM.Hex = DecodeInst->ModRM;

-        const bool Has16BitAddressing = !CTX->Config.Is64BitMode &&
-          DecodeInst->Flags & DecodeFlags::FLAG_ADDRESS_SIZE;
+          const bool Has16BitAddressing = !CTX->Config.Is64BitMode &&
+            DecodeInst->Flags & DecodeFlags::FLAG_ADDRESS_SIZE;

-        auto Disp = DecodeModRMs_Disp[Has16BitAddressing];
-        (this->*Disp)(&DecodeInst->Src[0], ModRM);
+          // All 3DNow! instructions have the second argument as the rm handler
+          // We need to decode it upfront to get the displacement out of the way
+          if (ModRM.mod != 0b11) {
+            auto Disp = DecodeModRMs_Disp[Has16BitAddressing];
+            (this->*Disp)(&DecodeInst->Src[0], ModRM);
+          }

-        // Take a peek at the op just past the displacement
-        uint8_t LocalOp = ReadByte();
-        return NormalOpHeader(&FEXCore::X86Tables::DDDNowOps[LocalOp], LocalOp);
-      break;
-      }
-      case 0x38: { // F38 Table!
-        constexpr uint16_t PF_38_NONE = 0;
-        constexpr uint16_t PF_38_66 = 1;
-        constexpr uint16_t PF_38_F2 = 2;
-        constexpr uint16_t PF_38_F3 = 3;
-
-        uint16_t Prefix = PF_38_NONE;
-        if (DecodeInst->LastEscapePrefix == 0xF2) {
-          // Repeat prefix or instruction-specific
-          Prefix = PF_38_F2;
-        } else if (DecodeInst->LastEscapePrefix == 0xF3) {
-          // Repeat prefix or instruction-specific
-          Prefix = PF_38_F3;
-        } else if (DecodeInst->LastEscapePrefix == 0x66) {
-          // Operand size
-          Prefix = PF_38_66;
+          // Take a peek at the op just past the displacement
+          uint8_t LocalOp = ReadByte();
+          return NormalOpHeader(&FEXCore::X86Tables::DDDNowOps[LocalOp], LocalOp);
+        break;
        }
+        case 0x38: { // F38 Table!
+          constexpr uint16_t PF_38_NONE = 0;
+          constexpr uint16_t PF_38_66   = (1U << 0);
+          constexpr uint16_t PF_38_F2   = (1U << 1);
+          constexpr uint16_t PF_38_F3   = (1U << 2);

-        uint16_t LocalOp = (Prefix << 8) | ReadByte();
-        return NormalOpHeader(&FEXCore::X86Tables::H0F38TableOps[LocalOp], LocalOp);
-      break;
-      }
-      case 0x3A: { // F3A Table!
-        constexpr uint16_t PF_3A_NONE = 0;
-        constexpr uint16_t PF_3A_66   = (1 << 0);
-        constexpr uint16_t PF_3A_REX  = (1 << 1);
+          uint16_t Prefix = PF_38_NONE;
+          if (DecodeInst->Flags & DecodeFlags::FLAG_OPERAND_SIZE) {
+            Prefix |= PF_38_66;
+          }
+          if (DecodeInst->Flags & DecodeFlags::FLAG_REPNE_PREFIX) {
+            Prefix |= PF_38_F2;
+          }
+          if (DecodeInst->Flags & DecodeFlags::FLAG_REP_PREFIX) {
+            Prefix |= PF_38_F3;
+          }

-        uint16_t Prefix = PF_3A_NONE;
-        if (DecodeInst->LastEscapePrefix == 0x66) // Operand Size
-          Prefix = PF_3A_66;
-
-        if (DecodeInst->Flags & DecodeFlags::FLAG_REX_WIDENING)
-          Prefix |= PF_3A_REX;
-
-        uint16_t LocalOp = (Prefix << 8) | ReadByte();
-        return NormalOpHeader(&FEXCore::X86Tables::H0F3ATableOps[LocalOp], LocalOp);
-      break;
-      }
-      default: // Two byte table!
-        // x86-64 abuses three legacy prefixes to extend the table encodings
-        // 0x66 - Operand Size prefix
-        // 0xF2 - REPNE prefix
-        // 0xF3 - REP prefix
-        // If any of these three prefixes are used then it falls down the subtable
-        // Additionally: If you hit repeat of differnt prefixes then only the LAST one before this one works for subtable selection
-
-        bool NoOverlay = (FEXCore::X86Tables::SecondBaseOps[EscapeOp].Flags & InstFlags::FLAGS_NO_OVERLAY) != 0;
-        bool NoOverlay66 = (FEXCore::X86Tables::SecondBaseOps[EscapeOp].Flags & InstFlags::FLAGS_NO_OVERLAY66) != 0;
-
-        if (NoOverlay) { // This section of the table ignores prefix extention
-          return NormalOpHeader(&FEXCore::X86Tables::SecondBaseOps[EscapeOp], EscapeOp);
+          uint16_t LocalOp = (Prefix << 8) | ReadByte();
+          return NormalOpHeader(&FEXCore::X86Tables::H0F38TableOps[LocalOp], LocalOp);
+        break;
        }
-        else if (DecodeInst->LastEscapePrefix == 0xF3) { // REP
-          // Remove prefix so it doesn't effect calculations.
-          // This is only an escape prefix rather tan modifier now
-          DecodeInst->Flags &= ~DecodeFlags::FLAG_REP_PREFIX;
-          return NormalOpHeader(&FEXCore::X86Tables::RepModOps[EscapeOp], EscapeOp);
+        case 0x3A: { // F3A Table!
+          constexpr uint16_t PF_3A_NONE = 0;
+          constexpr uint16_t PF_3A_66   = (1 << 0);
+          constexpr uint16_t PF_3A_REX  = (1 << 1);
+
+          uint16_t Prefix = PF_3A_NONE;
+          if (DecodeInst->LastEscapePrefix == 0x66) // Operand Size
+            Prefix = PF_3A_66;
+
+          if (DecodeInst->Flags & DecodeFlags::FLAG_REX_WIDENING)
+            Prefix |= PF_3A_REX;
+
+          uint16_t LocalOp = (Prefix << 8) | ReadByte();
+          return NormalOpHeader(&FEXCore::X86Tables::H0F3ATableOps[LocalOp], LocalOp);
+        break;
        }
-        else if (DecodeInst->LastEscapePrefix == 0xF2) { // REPNE
-          // Remove prefix so it doesn't effect calculations.
-          // This is only an escape prefix rather tan modifier now
-          DecodeInst->Flags &= ~DecodeFlags::FLAG_REPNE_PREFIX;
-          return NormalOpHeader(&FEXCore::X86Tables::RepNEModOps[EscapeOp], EscapeOp);
+        default: [[likely]] { // Two byte table!
+          // x86-64 abuses three legacy prefixes to extend the table encodings
+          // 0x66 - Operand Size prefix
+          // 0xF2 - REPNE prefix
+          // 0xF3 - REP prefix
+          // If any of these three prefixes are used then it falls down the subtable
+          // Additionally: If you hit repeat of differnt prefixes then only the LAST one before this one works for subtable selection
+
+          bool NoOverlay = (FEXCore::X86Tables::SecondBaseOps[EscapeOp].Flags & InstFlags::FLAGS_NO_OVERLAY) != 0;
+          bool NoOverlay66 = (FEXCore::X86Tables::SecondBaseOps[EscapeOp].Flags & InstFlags::FLAGS_NO_OVERLAY66) != 0;
+
+          if (NoOverlay) { // This section of the table ignores prefix extention
+            return NormalOpHeader(&FEXCore::X86Tables::SecondBaseOps[EscapeOp], EscapeOp);
+          }
+          else if (DecodeInst->LastEscapePrefix == 0xF3) { // REP
+            // Remove prefix so it doesn't effect calculations.
+            // This is only an escape prefix rather tan modifier now
+            DecodeInst->Flags &= ~DecodeFlags::FLAG_REP_PREFIX;
+            return NormalOpHeader(&FEXCore::X86Tables::RepModOps[EscapeOp], EscapeOp);
+          }
+          else if (DecodeInst->LastEscapePrefix == 0xF2) { // REPNE
+            // Remove prefix so it doesn't effect calculations.
+            // This is only an escape prefix rather tan modifier now
+            DecodeInst->Flags &= ~DecodeFlags::FLAG_REPNE_PREFIX;
+            return NormalOpHeader(&FEXCore::X86Tables::RepNEModOps[EscapeOp], EscapeOp);
+          }
+          else if (DecodeInst->LastEscapePrefix == 0x66 && !NoOverlay66) { // Operand Size
+            // Remove prefix so it doesn't effect calculations.
+            // This is only an escape prefix rather tan modifier now
+            DecodeInst->Flags &= ~DecodeFlags::FLAG_OPERAND_SIZE;
+            DecodeFlags::PopOpAddrIf(&DecodeInst->Flags, DecodeFlags::FLAG_OPERAND_SIZE_LAST);
+            return NormalOpHeader(&FEXCore::X86Tables::OpSizeModOps[EscapeOp], EscapeOp);
+          }
+          else {
+            return NormalOpHeader(&FEXCore::X86Tables::SecondBaseOps[EscapeOp], EscapeOp);
+          }
+        break;
        }
-        else if (DecodeInst->LastEscapePrefix == 0x66 && !NoOverlay66) { // Operand Size
-          // Remove prefix so it doesn't effect calculations.
-          // This is only an escape prefix rather tan modifier now
-          DecodeInst->Flags &= ~DecodeFlags::FLAG_OPERAND_SIZE;
-          DecodeFlags::PopOpAddrIf(&DecodeInst->Flags, DecodeFlags::FLAG_OPERAND_SIZE_LAST);
-          return NormalOpHeader(&FEXCore::X86Tables::OpSizeModOps[EscapeOp], EscapeOp);
-        }
-        else {
-          return NormalOpHeader(&FEXCore::X86Tables::SecondBaseOps[EscapeOp], EscapeOp);
-        }
-      break;
      }
    break;
    }
@@ -1005,7 +929,7 @@ bool Decoder::DecodeInstruction(uint64_t PC) {
    case 0x65: // GS prefix
      DecodeInst->Flags |= DecodeFlags::FLAG_GS_PREFIX;
    break;
-    default: { // Default base table
+    default: [[likely]] { // Default base table
      auto Info = &FEXCore::X86Tables::BaseOps[Op];

      if (Info->Type == FEXCore::X86Tables::TYPE_REX_PREFIX) {
@@ -1040,7 +964,7 @@ bool Decoder::DecodeInstruction(uint64_t PC) {
  }

  if (DecodeInst->Dest.IsGPR()) {
-    assert(DecodeInst->Dest.Data.GPR.GPR != FEXCore::X86State::REG_INVALID);
+    LOGMAN_THROW_A_FMT(DecodeInst->Dest.Data.GPR.GPR != FEXCore::X86State::REG_INVALID, "Destination GPR was invalid");
  }

  return true;
@@ -1098,15 +1022,13 @@ void Decoder::BranchTargetInMultiblockRange() {

      // If we are conditional then a target can be the instruction past the conditional instruction
      uint64_t FallthroughRIP = DecodeInst->PC + DecodeInst->InstSize;
-      if (HasBlocks.find(FallthroughRIP) == HasBlocks.end() &&
-          BlocksToDecode.find(FallthroughRIP) == BlocksToDecode.end()) {
-        BlocksToDecode.emplace(FallthroughRIP);
+      if (!HasBlocks.contains(FallthroughRIP)) {
+        BlocksToDecode.insert(FallthroughRIP);
      }
    }

-    if (HasBlocks.find(TargetRIP) == HasBlocks.end() &&
-        BlocksToDecode.find(TargetRIP) == BlocksToDecode.end()) {
-      BlocksToDecode.emplace(TargetRIP);
+    if (!HasBlocks.contains(TargetRIP)) {
+      BlocksToDecode.insert(TargetRIP);
    }
  } else {
    if (ExternalBranches) {
@@ -1115,6 +1037,35 @@ void Decoder::BranchTargetInMultiblockRange() {
  }
 }

+bool Decoder::BranchTargetCanContinue(bool FinalInstruction) const {
+  if (FinalInstruction) {
+    return false;
+  }
+
+  uint64_t TargetRIP = 0;
+  const uint8_t GPRSize = CTX->GetGPRSize();
+
+  if (DecodeInst->OP == 0xE8) { // Call - immediate target
+    const uint64_t NextRIP = DecodeInst->PC + DecodeInst->InstSize;
+    LOGMAN_THROW_A_FMT(DecodeInst->Src[0].IsLiteral(), "Had wrong operand type");
+    TargetRIP = DecodeInst->PC + DecodeInst->InstSize + DecodeInst->Src[0].Data.Literal.Value;
+
+    if (GPRSize == 4) {
+      // If we are running a 32bit guest then wrap around addresses that go above 32bit
+      TargetRIP &= 0xFFFFFFFFU;
+    }
+
+    if (TargetRIP == NextRIP) {
+      // Optimize the case that the instruction is jumping just after itself.
+      // This is a GOT calculation which we can optimize out.
+      // Optimization occurs inside of the OpDispatcher implementation
+      return true;
+    }
+  }
+
+  return false;
+}
+
 const uint8_t *Decoder::AdjustAddrForSpecialRegion(uint8_t const* _InstStream, uint64_t EntryPoint, uint64_t RIP) {
  constexpr uint64_t VSyscall_Base = 0xFFFF'FFFF'FF60'0000ULL;
  constexpr uint64_t VSyscall_End = VSyscall_Base + 0x1000;
@@ -1134,7 +1085,8 @@ const uint8_t *Decoder::AdjustAddrForSpecialRegion(uint8_t const* _InstStream, u
  return _InstStream - EntryPoint + RIP;
 }

-void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC) {
+void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC, std::function<void(uint64_t BlockEntry, uint64_t Start, uint64_t Length)> AddContainedCodePage) {
+  FEXCORE_PROFILE_SCOPED("DecodeInstructions");
  Blocks.clear();
  BlocksToDecode.clear();
  HasBlocks.clear();
@@ -1142,6 +1094,7 @@ void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC)
  DecodedSize = 0;
  MaxCondBranchForward = 0;
  MaxCondBranchBackwards = ~0ULL;
+  DecodedBuffer = PoolObject.ReownOrClaimBuffer();

  // XXX: Load symbol data
  SymbolAvailable = false;
@@ -1163,6 +1116,12 @@ void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC)
  // Entry is a jump target
  BlocksToDecode.emplace(PC);

+  uint64_t CurrentCodePage = PC & FHU::FEX_PAGE_MASK;
+
+  fextl::set<uint64_t> CodePages = { CurrentCodePage };
+
+  AddContainedCodePage(PC, CurrentCodePage, FHU::FEX_PAGE_SIZE);
+
  while (!BlocksToDecode.empty()) {
    auto BlockDecodeIt = BlocksToDecode.begin();
    uint64_t RIPToDecode = *BlockDecodeIt;
@@ -1179,10 +1138,28 @@ void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC)
    InstStream = AdjustAddrForSpecialRegion(_InstStream, EntryPoint, RIPToDecode);

    while (1) {
+
+      // MAX_INST_SIZE assumes worst case
+      auto OpMinAddress = RIPToDecode + PCOffset;
+      auto OpMaxAddress = OpMinAddress + MAX_INST_SIZE;
+
+      auto OpMinPage = OpMinAddress & FHU::FEX_PAGE_MASK;
+      auto OpMaxPage = OpMaxAddress & FHU::FEX_PAGE_MASK;
+
+      if (OpMinPage != CurrentCodePage) {
+        CurrentCodePage = OpMinPage;
+        CodePages.insert(CurrentCodePage);
+      }
+
+      if (OpMaxPage != CurrentCodePage) {
+        CurrentCodePage = OpMaxPage;
+        CodePages.insert(CurrentCodePage);
+      }
+
      bool ErrorDuringDecoding = !DecodeInstruction(RIPToDecode + PCOffset);

-      if (ErrorDuringDecoding) {
-        LogMan::Msg::DFmt("Couldn't Decode something at 0x{:x}, Started at 0x{:x}", PC + PCOffset, PC);
+      if (ErrorDuringDecoding) [[unlikely]] {
+        LogMan::Msg::DFmt("Couldn't Decode something at 0x{:x}, Started at 0x{:x}", RIPToDecode + PCOffset, PC);
        // Put an invalid instruction in the stream so the core can raise SIGILL if hit
        CurrentBlockDecoding.HasInvalidInstruction = true;
        // Error while decoding instruction. We don't know the table or instruction size
@@ -1208,23 +1185,21 @@ void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC)
        CanContinue = true;
      }

+      bool FinalInstruction = DecodedSize >= CTX->Config.MaxInstPerBlock ||
+          DecodedSize >= DefaultDecodedBufferSize ||
+          TotalInstructions >= CTX->Config.MaxInstPerBlock;
+
      if (DecodeInst->TableInfo->Flags & FEXCore::X86Tables::InstFlags::FLAGS_SETS_RIP) {
        // If we have multiblock enabled
        // If the branch target is within our multiblock range then we can keep going on
        // We don't want to short circuit this since we want to calculate our ranges still
        BranchTargetInMultiblockRange();
+
+        // Bypass branches if we can continue through them in some cases.
+        CanContinue |= BranchTargetCanContinue(FinalInstruction);
      }

-      if (!CanContinue) {
-        break;
-      }
-
-      if (DecodedSize >= CTX->Config.MaxInstPerBlock ||
-          DecodedSize >= DefaultDecodedBufferSize) {
-        break;
-      }
-
-      if (TotalInstructions >= CTX->Config.MaxInstPerBlock) {
+      if (FinalInstruction || !CanContinue) {
        break;
      }

@@ -1240,6 +1215,9 @@ void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC)
    CurrentBlockDecoding.DecodedInstructions = &DecodedBuffer[BlockStartOffset];
  }

+  for (auto CodePage : CodePages) {
+    AddContainedCodePage(PC, CodePage, FHU::FEX_PAGE_SIZE);
+  }

  // sort for better branching
  std::sort(Blocks.begin(), Blocks.end(), [](const FEXCore::Frontend::Decoder::DecodedBlocks& a, const FEXCore::Frontend::Decoder::DecodedBlocks& b) {
@@ -1,17 +1,18 @@
 #pragma once

-#include <FEXCore/Debug/X86Tables.h>
+#include "Interface/Core/X86Tables/X86Tables.h"
+
 #include <FEXCore/HLE/SyscallHandler.h>
 #include <FEXCore/Utils/Telemetry.h>
+#include <FEXCore/fextl/set.h>
+#include <FEXCore/fextl/vector.h>

 #include <array>
 #include <cstdint>
-#include <set>
 #include <stddef.h>
-#include <vector>

 namespace FEXCore::Context {
-struct Context;
+class ContextImpl;
 }

 namespace FEXCore::Frontend {
@@ -25,33 +26,40 @@ public:
    bool HasInvalidInstruction{};
  };

-  Decoder(FEXCore::Context::Context *ctx);
+  Decoder(FEXCore::Context::ContextImpl *ctx);
  ~Decoder();
-  void DecodeInstructionsAtEntry(uint8_t const* InstStream, uint64_t PC);
+  void DecodeInstructionsAtEntry(uint8_t const* InstStream, uint64_t PC, std::function<void(uint64_t BlockEntry, uint64_t Start, uint64_t Length)> AddContainedCodePage);

-  std::vector<DecodedBlocks> const *GetDecodedBlocks() const {
+  fextl::vector<DecodedBlocks> const *GetDecodedBlocks() const {
    return &Blocks;
  }

  uint64_t DecodedMinAddress {};
  uint64_t DecodedMaxAddress {~0ULL};
-  
+
  void SetSectionMaxAddress(uint64_t v) { SectionMaxAddress = v; }
-  void SetExternalBranches(std::set<uint64_t> *v) { ExternalBranches = v; }
+  void SetExternalBranches(fextl::set<uint64_t> *v) { ExternalBranches = v; }
+
+  void DelayedDisownBuffer() {
+    PoolObject.DelayedDisownBuffer();
+  }
+
 private:
  // To pass any information from instruction prefixes
  // down into the actual instruction handling machinery.
  struct DecodedHeader {
    uint8_t vvvv; // Encoded operand in a VEX prefix.
    bool w;       // VEX.W bit.
+    bool L;       // VEX.L bit (if set then 256 bit operation, if unset then scalar or 128-bit operation)
  };

-  FEXCore::Context::Context *CTX;
+  FEXCore::Context::ContextImpl *CTX;
  const FEXCore::HLE::SyscallOSABI OSABI{};

  bool DecodeInstruction(uint64_t PC);

  void BranchTargetInMultiblockRange();
+  bool BranchTargetCanContinue(bool FinalInstruction) const;

  uint8_t ReadByte();
  uint8_t PeekByte(uint8_t Offset) const;
@@ -63,6 +71,7 @@ private:

  static constexpr size_t DefaultDecodedBufferSize = 0x10000;
  FEXCore::X86Tables::DecodedInst *DecodedBuffer{};
+  Utils::FixedSizePooledAllocation<FEXCore::X86Tables::DecodedInst*, 5000, 500> PoolObject;
  size_t DecodedSize {};

  uint8_t const *InstStream;
@@ -81,10 +90,10 @@ private:
  uint64_t SymbolMinAddress {~0ULL};
  uint64_t SectionMaxAddress {~0ULL};

-  std::vector<DecodedBlocks> Blocks;
-  std::set<uint64_t> BlocksToDecode;
-  std::set<uint64_t> HasBlocks;
-  std::set<uint64_t> *ExternalBranches {nullptr};
+  fextl::vector<DecodedBlocks> Blocks;
+  fextl::set<uint64_t> BlocksToDecode;
+  fextl::set<uint64_t> HasBlocks;
+  fextl::set<uint64_t> *ExternalBranches {nullptr};

  // ModRM rm decoding
  using DecodeModRMPtr = void (FEXCore::Frontend::Decoder::*)(X86Tables::DecodedOperand *Operand, X86Tables::ModRMDecoded ModRM);
@@ -6,40 +6,50 @@ $end_info$
 #pragma once

 #include <FEXCore/Config/Config.h>
+#include <FEXCore/Utils/Event.h>
 #include <FEXCore/Utils/Threads.h>
+#include <FEXCore/fextl/memory.h>
+#include <FEXCore/fextl/string.h>

+#include <atomic>
 #include <istream>
 #include <memory>
 #include <mutex>
 #include <stdint.h>
-#include <string>

 namespace FEXCore {

 namespace Context {
-  struct Context;
+  class ContextImpl;
 }

 class GdbServer {
 public:
-    GdbServer(FEXCore::Context::Context *ctx);
+    GdbServer(FEXCore::Context::ContextImpl *ctx);

    // Public for threading
    void GdbServerLoop();

+    void AlertLibrariesChanged() {
+      LibraryMapChanged = true;
+    }
+
 private:
    void Break(int signal);

    void OpenListenSocket();
-    std::unique_ptr<std::iostream> OpenSocket();
+    fextl::unique_ptr<std::iostream> OpenSocket();
    void StartThread();
-    std::string ReadPacket(std::iostream &stream);
-    void SendPacket(std::ostream &stream, const std::string& packet);
+    fextl::string ReadPacket(std::iostream &stream);
+    void SendPacket(std::ostream &stream, const fextl::string& packet);

    void SendACK(std::ostream &stream, bool NACK);

+    Event ThreadBreakEvent{};
+    void WaitForThreadWakeup();
+
    struct HandledPacketType {
-      std::string Response{};
+      fextl::string Response{};
      enum ResponseType {
        TYPE_NONE,
        TYPE_UNKNOWN,
@@ -52,31 +62,35 @@ private:
    };

    void SendPacketPair(const HandledPacketType& packetPair);
-    HandledPacketType ProcessPacket(const std::string &packet);
-    HandledPacketType handleQuery(const std::string &packet);
-    HandledPacketType handleXfer(const std::string &packet);
-    HandledPacketType handleMemory(const std::string &packet);
-    HandledPacketType handleV(const std::string& packet);
-    HandledPacketType handleThreadOp(const std::string &packet);
-    HandledPacketType handleBreakpoint(const std::string &packet);
+    HandledPacketType ProcessPacket(const fextl::string &packet);
+    HandledPacketType handleQuery(const fextl::string &packet);
+    HandledPacketType handleXfer(const fextl::string &packet);
+    HandledPacketType handleMemory(const fextl::string &packet);
+    HandledPacketType handleV(const fextl::string& packet);
+    HandledPacketType handleThreadOp(const fextl::string &packet);
+    HandledPacketType handleBreakpoint(const fextl::string &packet);
    HandledPacketType handleProgramOffsets();

    HandledPacketType ThreadAction(char action, uint32_t tid);

-    std::string readRegs();
-    HandledPacketType readReg(const std::string& packet);
+    fextl::string readRegs();
+    HandledPacketType readReg(const fextl::string& packet);

-    FEXCore::Context::Context *CTX;
-    std::unique_ptr<FEXCore::Threads::Thread> gdbServerThread;
-    std::unique_ptr<std::iostream> CommsStream;
+    FEXCore::Context::ContextImpl *CTX;
+    fextl::unique_ptr<FEXCore::Threads::Thread> gdbServerThread;
+    fextl::unique_ptr<std::iostream> CommsStream;
    std::mutex sendMutex;
    bool SettingNoAckMode{false};
    bool NoAckMode{false};
    bool NonStopMode{false};
-    std::string ThreadString{};
-    std::string MemoryMapString{};
-    std::string OSDataString{};
+    fextl::string ThreadString{};
+    fextl::string OSDataString{};
+    void buildLibraryMap();
+    std::atomic<bool> LibraryMapChanged = true;
+    fextl::string LibraryMapString{};

+    // Used to keep track of which signals to pass to the guest
+    std::array<bool, SignalDelegator::MAX_SIGNALS + 1> PassSignals{};
    uint32_t CurrentDebuggingThread{};
    int ListenSocket{};
    FEX_CONFIG_OPT(Filename, APP_FILENAME);
@@ -1,7 +1,7 @@
 #include "Interface/Core/CPUID.h"
-#include "Interface/Core/HostFeatures.h"
+#include <FEXCore/Core/HostFeatures.h>

-#ifdef _M_ARM_64
+#if defined(_M_ARM_64) || defined(VIXL_SIMULATOR)
 #include "aarch64/assembler-aarch64.h"
 #include "aarch64/cpu-aarch64.h"
 #include "aarch64/disasm-aarch64.h"
@@ -9,7 +9,7 @@
 #endif

 #ifdef _M_X86_64
-#include <xbyak/xbyak_util.h>
+#include "Interface/Core/Dispatcher/X86Dispatcher.h"
 #endif

 namespace FEXCore {
@@ -17,12 +17,12 @@ namespace FEXCore {
 // Data Zero Prohibited flag
 // 0b0 = ZVA/GVA/GZVA permitted
 // 0b1 = ZVA/GVA/GZVA prohibited
-constexpr uint32_t DCZID_DZP_MASK = 0b1'0000;
+[[maybe_unused]] constexpr uint32_t DCZID_DZP_MASK = 0b1'0000;
 // Log2 of the blocksize in 32-bit words
-constexpr uint32_t DCZID_BS_MASK = 0b0'1111;
+[[maybe_unused]] constexpr uint32_t DCZID_BS_MASK = 0b0'1111;

 #ifdef _M_ARM_64
-static uint32_t GetDCZID() {
+[[maybe_unused]] static uint32_t GetDCZID() {
  uint64_t Result{};
  __asm("mrs %[Res], DCZID_EL0"
      : [Res] "=r" (Result));
@@ -50,18 +50,48 @@ static uint32_t GetDCZID() {


 HostFeatures::HostFeatures() {
-#ifdef _M_ARM_64
+#if defined(_M_ARM_64) || defined(VIXL_SIMULATOR)
+#ifdef VIXL_SIMULATOR
+  auto Features = vixl::CPUFeatures::All();
+#else
+#ifndef _WIN32
  auto Features = vixl::CPUFeatures::InferFromOS();
+#else
+  // Need to use ID registers in WINE.
+  auto Features = vixl::CPUFeatures::InferFromIDRegisters();
+#endif
+#endif
  SupportsAES = Features.Has(vixl::CPUFeatures::Feature::kAES);
+  SupportsCRC = Features.Has(vixl::CPUFeatures::Feature::kCRC32);
  SupportsAtomics = Features.Has(vixl::CPUFeatures::Feature::kAtomics);
+  SupportsRAND = Features.Has(vixl::CPUFeatures::Feature::kRNG);
+
  // Only supported when FEAT_AFP is supported
  SupportsFlushInputsToZero = Features.Has(vixl::CPUFeatures::Feature::kAFP);
+  SupportsRCPC = Features.Has(vixl::CPUFeatures::Feature::kRCpc);
+  SupportsTSOImm9 = Features.Has(vixl::CPUFeatures::Feature::kRCpcImm);
+  SupportsPMULL_128Bit = Features.Has(vixl::CPUFeatures::Feature::kPmull1Q);
+  SupportsCSSC = Features.Has(vixl::CPUFeatures::Feature::kCSSC);

-  // RCPC is bugged on Snapdragon 865
-  // Causes glibc cond16 test to immediately throw assert
-  // __pthread_mutex_cond_lock: Assertion `mutex->__data.__owner == 0'
-  SupportsRCPC = false; //Features.Has(vixl::CPUFeatures::Feature::kRCpc);
+  Supports3DNow = true;
+  SupportsSSE4A = true;
+#ifdef VIXL_SIMULATOR
+  // Hardcode enable SVE with 256-bit wide registers.
+  SupportsAVX = true;
+#else
+  SupportsAVX = Features.Has(vixl::CPUFeatures::Feature::kSVE2) &&
+                vixl::aarch64::CPU::ReadSVEVectorLengthInBits() >= 256;
+#endif
+  SupportsSHA = true;
+  SupportsBMI1 = true;
+  SupportsBMI2 = true;
+  SupportsCLWB = true;

+  if (!SupportsAtomics) {
+    WARN_ONCE_FMT("Host CPU doesn't support atomics. Expect bad performance");
+  }
+
+#ifdef _M_ARM_64
  // We need to get the CPU's cache line size
  // We expect sane targets that have correct cacheline sizes across clusters
  uint64_t CTR;
@@ -71,16 +101,6 @@ HostFeatures::HostFeatures() {
  DCacheLineSize = 4 << ((CTR >> 16) & 0xF);
  ICacheLineSize = 4 << (CTR & 0xF);

-  if (!SupportsAtomics) {
-    WARN_ONCE_FMT("Host CPU doesn't support atomics. Expect bad performance");
-  }
-#endif
-#ifdef _M_X86_64
-  Xbyak::util::Cpu Features{};
-  SupportsAES = Features.has(Xbyak::util::Cpu::tAESNI);
-  SupportsFlushInputsToZero = true;
-  SupportsFloatExceptions = true;
-#else
  // Test if this CPU supports float exception trapping by attempting to enable
  // On unsupported these bits are architecturally defined as RAZ/WI
  constexpr uint32_t ExceptionEnableTraps =
@@ -101,6 +121,42 @@ HostFeatures::HostFeatures() {
  SetFPCR(OriginalFPCR);
 #endif

+#endif
+#if defined(_M_X86_64) && !defined(VIXL_SIMULATOR)
+  Xbyak::util::Cpu Features{};
+  SupportsAES = Features.has(Xbyak::util::Cpu::tAESNI);
+  SupportsCRC = Features.has(Xbyak::util::Cpu::tSSE42);
+  SupportsRAND = Features.has(Xbyak::util::Cpu::tRDRAND) && Features.has(Xbyak::util::Cpu::tRDSEED);
+  SupportsRCPC = true;
+  SupportsTSOImm9 = true;
+  Supports3DNow = Features.has(Xbyak::util::Cpu::t3DN) && Features.has(Xbyak::util::Cpu::tE3DN);
+  SupportsSSE4A = Features.has(Xbyak::util::Cpu::tSSE4a);
+  SupportsAVX = true;
+  SupportsSHA = Features.has(Xbyak::util::Cpu::tSHA);
+  SupportsBMI1 = Features.has(Xbyak::util::Cpu::tBMI1);
+  SupportsBMI2 = Features.has(Xbyak::util::Cpu::tBMI2);
+  SupportsBMI2 = Features.has(Xbyak::util::Cpu::tCLWB);
+  SupportsPMULL_128Bit = Features.has(Xbyak::util::Cpu::tPCLMULQDQ);
+
+  // xbyak doesn't know how to check for CLZero
+  // First ensure we support a new enough extended CPUID function range
+
+  uint32_t data[4];
+  Xbyak::util::Cpu::getCpuid(0x8000'0000, data);
+  if (data[0] >= 0x8000'0008U) {
+    // CLZero defined in 8000_00008_EBX[bit 0]
+    Xbyak::util::Cpu::getCpuid(0x8000'0008, data);
+    SupportsCLZERO = data[1] & 1;
+  }
+
+  SupportsFlushInputsToZero = true;
+  SupportsFloatExceptions = true;
+#endif
+
+#ifdef VIXL_SIMULATOR
+  // simulator doesn't support dc(ZVA)
+  SupportsCLZERO = false;
+#else
  // Check if we can support cacheline clears
  uint32_t DCZID = GetDCZID();
  if ((DCZID & DCZID_DZP_MASK) == 0) {
@@ -110,5 +166,12 @@ HostFeatures::HostFeatures() {
    // This means we can use the instruction
    SupportsCLZERO = DCZID_Bytes == CPUIDEmu::CACHELINE_SIZE;
  }
+#endif
+
+  // Disable AVX if the configuration explicitly has disabled it.
+  FEX_CONFIG_OPT(EnableAVX, ENABLEAVX);
+  if (!EnableAVX) {
+    SupportsAVX = false;
+  }
 }
 }
@@ -18,16 +18,16 @@ namespace FEXCore::CPU {
 DEF_OP(TruncElementPair) {
  auto Op = IROp->C<IR::IROp_TruncElementPair>();

-  switch (Op->Size) {
+  switch (IROp->Size) {
    case 4: {
-      uint64_t *Src = GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
+      uint64_t *Src = GetSrc<uint64_t*>(Data->SSAData, Op->Pair);
      uint64_t Result{};
      Result = Src[0] & ~0U;
      Result |= Src[1] << 32;
      GD = Result;
      break;
    }
-    default: LOGMAN_MSG_A_FMT("Unhandled Truncation size: {}", Op->Size); break;
+    default: LOGMAN_MSG_A_FMT("Unhandled Truncation size: {}", IROp->Size); break;
  }
 }

@@ -38,7 +38,13 @@ DEF_OP(Constant) {

 DEF_OP(EntrypointOffset) {
  auto Op = IROp->C<IR::IROp_EntrypointOffset>();
-  GD = Data->CurrentEntry + Op->Offset;
+  uint64_t Mask = ~0ULL;
+  uint8_t OpSize = IROp->Size;
+  if (OpSize == 4) {
+    Mask = 0xFFFF'FFFFULL;
+  }
+
+  GD = (Data->CurrentEntry + Op->Offset) & Mask;
 }

 DEF_OP(InlineConstant) {
@@ -63,11 +69,11 @@ DEF_OP(CycleCounter) {

 DEF_OP(Add) {
  auto Op = IROp->C<IR::IROp_Add>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Header.Args[1]);
-  auto Func = [](auto a, auto b) { return a + b; };
+  auto *Src1 = GetSrc<void*>(Data->SSAData, Op->Src1);
+  auto *Src2 = GetSrc<void*>(Data->SSAData, Op->Src2);
+  const auto Func = [](auto a, auto b) { return a + b; };

  switch (OpSize) {
    DO_OP(4, uint32_t, Func)
@@ -78,11 +84,11 @@ DEF_OP(Add) {

 DEF_OP(Sub) {
  auto Op = IROp->C<IR::IROp_Sub>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Header.Args[1]);
-  auto Func = [](auto a, auto b) { return a - b; };
+  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Src1);
+  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Src2);
+  const auto Func = [](auto a, auto b) { return a - b; };

  switch (OpSize) {
    DO_OP(4, uint32_t, Func)
@@ -93,9 +99,9 @@ DEF_OP(Sub) {

 DEF_OP(Neg) {
  auto Op = IROp->C<IR::IROp_Neg>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  uint64_t Src = *GetSrc<int64_t*>(Data->SSAData, Op->Header.Args[0]);
+  const uint64_t Src = *GetSrc<int64_t*>(Data->SSAData, Op->Src);
  switch (OpSize) {
    case 4:
      GD = -static_cast<int32_t>(Src);
@@ -107,12 +113,28 @@ DEF_OP(Neg) {
  }
 }

+DEF_OP(Abs) {
+  auto Op = IROp->C<IR::IROp_Abs>();
+  const uint8_t OpSize = IROp->Size;
+
+  const int64_t Src = *GetSrc<int64_t*>(Data->SSAData, Op->Src);
+  switch (OpSize) {
+    case 4:
+      GD = std::abs(static_cast<int32_t>(Src));
+      break;
+    case 8:
+      GD = std::abs(static_cast<int64_t>(Src));
+      break;
+    default: LOGMAN_MSG_A_FMT("Unknown Abs Size: {}\n", OpSize); break;
+  }
+}
+
 DEF_OP(Mul) {
  auto Op = IROp->C<IR::IROp_Mul>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);

  switch (OpSize) {
    case 4:
@@ -132,10 +154,10 @@ DEF_OP(Mul) {

 DEF_OP(UMul) {
  auto Op = IROp->C<IR::IROp_UMul>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);

  switch (OpSize) {
    case 4:
@@ -155,9 +177,9 @@ DEF_OP(UMul) {

 DEF_OP(Div) {
  auto Op = IROp->C<IR::IROp_Div>();
-  uint8_t OpSize = IROp->Size;
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint8_t OpSize = IROp->Size;
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);

  switch (OpSize) {
    case 1:
@@ -173,7 +195,7 @@ DEF_OP(Div) {
      GD = static_cast<int64_t>(Src1) / static_cast<int64_t>(Src2);
      break;
    case 16: {
-      __int128_t Tmp = *GetSrc<__int128_t*>(Data->SSAData, Op->Header.Args[0]) / *GetSrc<__int128_t*>(Data->SSAData, Op->Header.Args[1]);
+      __int128_t Tmp = *GetSrc<__int128_t*>(Data->SSAData, Op->Src1) / *GetSrc<__int128_t*>(Data->SSAData, Op->Src2);
      memcpy(GDP, &Tmp, 16);
      break;
    }
@@ -183,10 +205,10 @@ DEF_OP(Div) {

 DEF_OP(UDiv) {
  auto Op = IROp->C<IR::IROp_UDiv>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);

  switch (OpSize) {
    case 1:
@@ -202,7 +224,7 @@ DEF_OP(UDiv) {
      GD = static_cast<uint64_t>(Src1) / static_cast<uint64_t>(Src2);
      break;
    case 16: {
-      __uint128_t Tmp = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]) / *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
+      __uint128_t Tmp = *GetSrc<__uint128_t*>(Data->SSAData, Op->Src1) / *GetSrc<__uint128_t*>(Data->SSAData, Op->Src2);
      memcpy(GDP, &Tmp, 16);
      break;
    }
@@ -212,10 +234,10 @@ DEF_OP(UDiv) {

 DEF_OP(Rem) {
  auto Op = IROp->C<IR::IROp_Rem>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);

  switch (OpSize) {
    case 1:
@@ -231,7 +253,7 @@ DEF_OP(Rem) {
      GD = static_cast<int64_t>(Src1) % static_cast<int64_t>(Src2);
      break;
    case 16: {
-      __int128_t Tmp = *GetSrc<__int128_t*>(Data->SSAData, Op->Header.Args[0]) % *GetSrc<__int128_t*>(Data->SSAData, Op->Header.Args[1]);
+      __int128_t Tmp = *GetSrc<__int128_t*>(Data->SSAData, Op->Src1) % *GetSrc<__int128_t*>(Data->SSAData, Op->Src2);
      memcpy(GDP, &Tmp, 16);
      break;
    }
@@ -241,10 +263,10 @@ DEF_OP(Rem) {

 DEF_OP(URem) {
  auto Op = IROp->C<IR::IROp_URem>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);

  switch (OpSize) {
    case 1:
@@ -260,7 +282,7 @@ DEF_OP(URem) {
      GD = static_cast<uint64_t>(Src1) % static_cast<uint64_t>(Src2);
      break;
    case 16: {
-      __uint128_t Tmp = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]) % *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
+      __uint128_t Tmp = *GetSrc<__uint128_t*>(Data->SSAData, Op->Src1) % *GetSrc<__uint128_t*>(Data->SSAData, Op->Src2);
      memcpy(GDP, &Tmp, 16);
      break;
    }
@@ -270,10 +292,10 @@ DEF_OP(URem) {

 DEF_OP(MulH) {
  auto Op = IROp->C<IR::IROp_MulH>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);

  switch (OpSize) {
    case 4: {
@@ -292,10 +314,10 @@ DEF_OP(MulH) {

 DEF_OP(UMulH) {
  auto Op = IROp->C<IR::IROp_UMulH>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
  switch (OpSize) {
    case 4:
      GD = static_cast<uint64_t>(Src1) * static_cast<uint64_t>(Src2);
@@ -318,11 +340,11 @@ DEF_OP(UMulH) {

 DEF_OP(Or) {
  auto Op = IROp->C<IR::IROp_Or>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Header.Args[1]);
-  auto Func = [](auto a, auto b) { return a | b; };
+  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Src1);
+  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Src2);
+  const auto Func = [](auto a, auto b) { return a | b; };

  switch (OpSize) {
    DO_OP(1, uint8_t,  Func)
@@ -336,11 +358,11 @@ DEF_OP(Or) {

 DEF_OP(And) {
  auto Op = IROp->C<IR::IROp_And>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Header.Args[1]);
-  auto Func = [](auto a, auto b) { return a & b; };
+  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Src1);
+  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Src2);
+  const auto Func = [](auto a, auto b) { return a & b; };

  switch (OpSize) {
    DO_OP(1, uint8_t,  Func)
@@ -355,8 +377,8 @@ DEF_OP(Andn) {
  auto Op = IROp->C<IR::IROp_Andn>();
  const uint8_t OpSize = IROp->Size;

-  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Header.Args[1]);
+  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Src1);
+  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Src2);
  constexpr auto Func = [](auto a, auto b) {
    using Type = decltype(a);
    return static_cast<Type>(a & static_cast<Type>(~b));
@@ -373,11 +395,11 @@ DEF_OP(Andn) {

 DEF_OP(Xor) {
  auto Op = IROp->C<IR::IROp_Xor>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Header.Args[1]);
-  auto Func = [](auto a, auto b) { return a ^ b; };
+  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Src1);
+  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Src2);
+  const auto Func = [](auto a, auto b) { return a ^ b; };

  switch (OpSize) {
    DO_OP(1, uint8_t,  Func)
@@ -390,11 +412,11 @@ DEF_OP(Xor) {

 DEF_OP(Lshl) {
  auto Op = IROp->C<IR::IROp_Lshl>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-  uint8_t Mask = OpSize * 8 - 1;
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
+  const uint8_t Mask = OpSize * 8 - 1;
  switch (OpSize) {
    case 4:
      GD = static_cast<uint32_t>(Src1) << (Src2 & Mask);
@@ -408,11 +430,11 @@ DEF_OP(Lshl) {

 DEF_OP(Lshr) {
  auto Op = IROp->C<IR::IROp_Lshr>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-  uint8_t Mask = OpSize * 8 - 1;
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
+  const uint8_t Mask = OpSize * 8 - 1;
  switch (OpSize) {
    case 4:
      GD = static_cast<uint32_t>(Src1) >> (Src2 & Mask);
@@ -426,11 +448,11 @@ DEF_OP(Lshr) {

 DEF_OP(Ashr) {
  auto Op = IROp->C<IR::IROp_Ashr>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-  uint8_t Mask = OpSize * 8 - 1;
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
+  const uint8_t Mask = OpSize * 8 - 1;
  switch (OpSize) {
    case 4:
      GD = (uint32_t)(static_cast<int32_t>(Src1) >> (Src2 & Mask));
@@ -444,12 +466,12 @@ DEF_OP(Ashr) {

 DEF_OP(Ror) {
  auto Op = IROp->C<IR::IROp_Ror>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-  auto Ror = [] (auto In, auto R) {
-    auto RotateMask = sizeof(In) * 8 - 1;
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
+  const auto Ror = [] (auto In, auto R) {
+    const auto RotateMask = sizeof(In) * 8 - 1;
    R &= RotateMask;
    return (In >> R) | (In << (sizeof(In) * 8 - R));
  };
@@ -468,11 +490,11 @@ DEF_OP(Ror) {

 DEF_OP(Extr) {
  auto Op = IROp->C<IR::IROp_Extr>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-  auto Extr = [] (auto Src1, auto Src2, uint8_t lsb) -> decltype(Src1) {
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Upper);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Lower);
+  const auto Extr = [] (auto Src1, auto Src2, uint8_t lsb) -> decltype(Src1) {
    __uint128_t Result{};
    Result = Src1;
    Result <<= sizeof(Src1) * 8;
@@ -494,7 +516,7 @@ DEF_OP(Extr) {
 }

 DEF_OP(PDep) {
-  const auto Op = IROp->C<IR::IROp_PExt>();
+  const auto Op = IROp->C<IR::IROp_PDep>();
  const auto OpSize = IROp->Size;

  if (OpSize != 4 && OpSize != 8) {
@@ -502,10 +524,10 @@ DEF_OP(PDep) {
    return;
  }

-  const uint64_t Input = OpSize == 4 ? *GetSrc<uint32_t*>(Data->SSAData, Op->Args(0))
-                                     : *GetSrc<uint64_t*>(Data->SSAData, Op->Args(0));
-  uint64_t Mask = OpSize == 4 ? *GetSrc<uint32_t*>(Data->SSAData, Op->Args(1))
-                              : *GetSrc<uint64_t*>(Data->SSAData, Op->Args(1));
+  const uint64_t Input = OpSize == 4 ? *GetSrc<uint32_t*>(Data->SSAData, Op->Input)
+                                     : *GetSrc<uint64_t*>(Data->SSAData, Op->Input);
+  uint64_t Mask = OpSize == 4 ? *GetSrc<uint32_t*>(Data->SSAData, Op->Mask)
+                              : *GetSrc<uint64_t*>(Data->SSAData, Op->Mask);

  uint64_t Result = 0;
  for (uint64_t Index = 0; Mask > 0; Index++) {
@@ -526,10 +548,10 @@ DEF_OP(PExt) {
    return;
  }

-  const uint64_t Input = OpSize == 4 ? *GetSrc<uint32_t*>(Data->SSAData, Op->Args(0))
-                                     : *GetSrc<uint64_t*>(Data->SSAData, Op->Args(0));
-  uint64_t Mask = OpSize == 4 ? *GetSrc<uint32_t*>(Data->SSAData, Op->Args(1))
-                              : *GetSrc<uint64_t*>(Data->SSAData, Op->Args(1));
+  const uint64_t Input = OpSize == 4 ? *GetSrc<uint32_t*>(Data->SSAData, Op->Input)
+                                     : *GetSrc<uint64_t*>(Data->SSAData, Op->Input);
+  uint64_t Mask = OpSize == 4 ? *GetSrc<uint32_t*>(Data->SSAData, Op->Mask)
+                              : *GetSrc<uint64_t*>(Data->SSAData, Op->Mask);

  uint64_t Result = 0;
  for (uint64_t Offset = 0; Mask > 0; Offset++) {
@@ -543,39 +565,39 @@ DEF_OP(PExt) {

 DEF_OP(LDiv) {
  auto Op = IROp->C<IR::IROp_LDiv>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

  // Each source is OpSize in size
  // So you can have up to a 128bit divide from x86-64
  switch (OpSize) {
    case 2: {
-      uint16_t SrcLow = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t SrcHigh = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
-      int16_t Divisor = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[2]);
-      int32_t Source = (static_cast<uint32_t>(SrcHigh) << 16) | SrcLow;
-      int32_t Res = Source / Divisor;
+      const uint16_t SrcLow = *GetSrc<uint16_t*>(Data->SSAData, Op->Lower);
+      const uint16_t SrcHigh = *GetSrc<uint16_t*>(Data->SSAData, Op->Upper);
+      const int16_t Divisor = *GetSrc<uint16_t*>(Data->SSAData, Op->Divisor);
+      const int32_t Source = (static_cast<uint32_t>(SrcHigh) << 16) | SrcLow;
+      const int32_t Res = Source / Divisor;

      // We only store the lower bits of the result
      GD = static_cast<int16_t>(Res);
      break;
    }
    case 4: {
-      uint32_t SrcLow = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t SrcHigh = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
-      int32_t Divisor = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[2]);
-      int64_t Source = (static_cast<uint64_t>(SrcHigh) << 32) | SrcLow;
-      int64_t Res = Source / Divisor;
+      const uint32_t SrcLow = *GetSrc<uint32_t*>(Data->SSAData, Op->Lower);
+      const uint32_t SrcHigh = *GetSrc<uint32_t*>(Data->SSAData, Op->Upper);
+      const int32_t Divisor = *GetSrc<uint32_t*>(Data->SSAData, Op->Divisor);
+      const int64_t Source = (static_cast<uint64_t>(SrcHigh) << 32) | SrcLow;
+      const int64_t Res = Source / Divisor;

      // We only store the lower bits of the result
      GD = static_cast<int32_t>(Res);
      break;
    }
    case 8: {
-      uint64_t SrcLow = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t SrcHigh = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-      int64_t Divisor = *GetSrc<int64_t*>(Data->SSAData, Op->Header.Args[2]);
-      __int128_t Source = (static_cast<__int128_t>(SrcHigh) << 64) | SrcLow;
-      __int128_t Res = Source / Divisor;
+      const uint64_t SrcLow = *GetSrc<uint64_t*>(Data->SSAData, Op->Lower);
+      const uint64_t SrcHigh = *GetSrc<uint64_t*>(Data->SSAData, Op->Upper);
+      const int64_t Divisor = *GetSrc<int64_t*>(Data->SSAData, Op->Divisor);
+      const __int128_t Source = (static_cast<__int128_t>(SrcHigh) << 64) | SrcLow;
+      const __int128_t Res = Source / Divisor;

      // We only store the lower bits of the result
      memcpy(GDP, &Res, OpSize);
@@ -587,39 +609,39 @@ DEF_OP(LDiv) {

 DEF_OP(LUDiv) {
  auto Op = IROp->C<IR::IROp_LUDiv>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

  // Each source is OpSize in size
  // So you can have up to a 128bit divide from x86-64
  switch (OpSize) {
    case 2: {
-      uint16_t SrcLow = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t SrcHigh = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
-      uint16_t Divisor = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[2]);
-      uint32_t Source = (static_cast<uint32_t>(SrcHigh) << 16) | SrcLow;
-      uint32_t Res = Source / Divisor;
+      const uint16_t SrcLow = *GetSrc<uint16_t*>(Data->SSAData, Op->Lower);
+      const uint16_t SrcHigh = *GetSrc<uint16_t*>(Data->SSAData, Op->Upper);
+      const uint16_t Divisor = *GetSrc<uint16_t*>(Data->SSAData, Op->Divisor);
+      const uint32_t Source = (static_cast<uint32_t>(SrcHigh) << 16) | SrcLow;
+      const uint32_t Res = Source / Divisor;

      // We only store the lower bits of the result
      GD = static_cast<uint16_t>(Res);
      break;
    }
    case 4: {
-      uint32_t SrcLow = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t SrcHigh = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
-      uint32_t Divisor = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[2]);
-      uint64_t Source = (static_cast<uint64_t>(SrcHigh) << 32) | SrcLow;
-      uint64_t Res = Source / Divisor;
+      const uint32_t SrcLow = *GetSrc<uint32_t*>(Data->SSAData, Op->Lower);
+      const uint32_t SrcHigh = *GetSrc<uint32_t*>(Data->SSAData, Op->Upper);
+      const uint32_t Divisor = *GetSrc<uint32_t*>(Data->SSAData, Op->Divisor);
+      const uint64_t Source = (static_cast<uint64_t>(SrcHigh) << 32) | SrcLow;
+      const uint64_t Res = Source / Divisor;

      // We only store the lower bits of the result
      GD = static_cast<uint32_t>(Res);
      break;
    }
    case 8: {
-      uint64_t SrcLow = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t SrcHigh = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-      uint64_t Divisor = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[2]);
-      __uint128_t Source = (static_cast<__uint128_t>(SrcHigh) << 64) | SrcLow;
-      __uint128_t Res = Source / Divisor;
+      const uint64_t SrcLow = *GetSrc<uint64_t*>(Data->SSAData, Op->Lower);
+      const uint64_t SrcHigh = *GetSrc<uint64_t*>(Data->SSAData, Op->Upper);
+      const uint64_t Divisor = *GetSrc<uint64_t*>(Data->SSAData, Op->Divisor);
+      const __uint128_t Source = (static_cast<__uint128_t>(SrcHigh) << 64) | SrcLow;
+      const __uint128_t Res = Source / Divisor;

      // We only store the lower bits of the result
      memcpy(GDP, &Res, OpSize);
@@ -631,39 +653,39 @@ DEF_OP(LUDiv) {

 DEF_OP(LRem) {
  auto Op = IROp->C<IR::IROp_LRem>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

  // Each source is OpSize in size
  // So you can have up to a 128bit Remainder from x86-64
  switch (OpSize) {
    case 2: {
-      uint16_t SrcLow = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t SrcHigh = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
-      int16_t Divisor = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[2]);
-      int32_t Source = (static_cast<uint32_t>(SrcHigh) << 16) | SrcLow;
-      int32_t Res = Source % Divisor;
+      const uint16_t SrcLow = *GetSrc<uint16_t*>(Data->SSAData, Op->Lower);
+      const uint16_t SrcHigh = *GetSrc<uint16_t*>(Data->SSAData, Op->Upper);
+      const int16_t Divisor = *GetSrc<uint16_t*>(Data->SSAData, Op->Divisor);
+      const int32_t Source = (static_cast<uint32_t>(SrcHigh) << 16) | SrcLow;
+      const int32_t Res = Source % Divisor;

      // We only store the lower bits of the result
      GD = static_cast<int16_t>(Res);
      break;
    }
    case 4: {
-      uint32_t SrcLow = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t SrcHigh = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
-      int32_t Divisor = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[2]);
-      int64_t Source = (static_cast<uint64_t>(SrcHigh) << 32) | SrcLow;
-      int64_t Res = Source % Divisor;
+      const uint32_t SrcLow = *GetSrc<uint32_t*>(Data->SSAData, Op->Lower);
+      const uint32_t SrcHigh = *GetSrc<uint32_t*>(Data->SSAData, Op->Upper);
+      const int32_t Divisor = *GetSrc<uint32_t*>(Data->SSAData, Op->Divisor);
+      const int64_t Source = (static_cast<uint64_t>(SrcHigh) << 32) | SrcLow;
+      const int64_t Res = Source % Divisor;

      // We only store the lower bits of the result
      GD = static_cast<int32_t>(Res);
      break;
    }
    case 8: {
-      uint64_t SrcLow = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t SrcHigh = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-      int64_t Divisor = *GetSrc<int64_t*>(Data->SSAData, Op->Header.Args[2]);
-      __int128_t Source = (static_cast<__int128_t>(SrcHigh) << 64) | SrcLow;
-      __int128_t Res = Source % Divisor;
+      const uint64_t SrcLow = *GetSrc<uint64_t*>(Data->SSAData, Op->Lower);
+      const uint64_t SrcHigh = *GetSrc<uint64_t*>(Data->SSAData, Op->Upper);
+      const int64_t Divisor = *GetSrc<int64_t*>(Data->SSAData, Op->Divisor);
+      const __int128_t Source = (static_cast<__int128_t>(SrcHigh) << 64) | SrcLow;
+      const __int128_t Res = Source % Divisor;
      // We only store the lower bits of the result
      memcpy(GDP, &Res, OpSize);
      break;
@@ -674,39 +696,39 @@ DEF_OP(LRem) {

 DEF_OP(LURem) {
  auto Op = IROp->C<IR::IROp_LURem>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

  // Each source is OpSize in size
  // So you can have up to a 128bit Remainder from x86-64
  switch (OpSize) {
    case 2: {
-      uint16_t SrcLow = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t SrcHigh = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
-      uint16_t Divisor = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[2]);
-      uint32_t Source = (static_cast<uint32_t>(SrcHigh) << 16) | SrcLow;
-      uint32_t Res = Source % Divisor;
+      const uint16_t SrcLow = *GetSrc<uint16_t*>(Data->SSAData, Op->Lower);
+      const uint16_t SrcHigh = *GetSrc<uint16_t*>(Data->SSAData, Op->Upper);
+      const uint16_t Divisor = *GetSrc<uint16_t*>(Data->SSAData, Op->Divisor);
+      const uint32_t Source = (static_cast<uint32_t>(SrcHigh) << 16) | SrcLow;
+      const uint32_t Res = Source % Divisor;

      // We only store the lower bits of the result
      GD = static_cast<uint16_t>(Res);
      break;
    }
    case 4: {
-      uint32_t SrcLow = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t SrcHigh = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
-      uint32_t Divisor = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[2]);
-      uint64_t Source = (static_cast<uint64_t>(SrcHigh) << 32) | SrcLow;
-      uint64_t Res = Source % Divisor;
+      const uint32_t SrcLow = *GetSrc<uint32_t*>(Data->SSAData, Op->Lower);
+      const uint32_t SrcHigh = *GetSrc<uint32_t*>(Data->SSAData, Op->Upper);
+      const uint32_t Divisor = *GetSrc<uint32_t*>(Data->SSAData, Op->Divisor);
+      const uint64_t Source = (static_cast<uint64_t>(SrcHigh) << 32) | SrcLow;
+      const uint64_t Res = Source % Divisor;

      // We only store the lower bits of the result
      GD = static_cast<uint32_t>(Res);
      break;
    }
    case 8: {
-      uint64_t SrcLow = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t SrcHigh = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-      uint64_t Divisor = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[2]);
-      __uint128_t Source = (static_cast<__uint128_t>(SrcHigh) << 64) | SrcLow;
-      __uint128_t Res = Source % Divisor;
+      const uint64_t SrcLow = *GetSrc<uint64_t*>(Data->SSAData, Op->Lower);
+      const uint64_t SrcHigh = *GetSrc<uint64_t*>(Data->SSAData, Op->Upper);
+      const uint64_t Divisor = *GetSrc<uint64_t*>(Data->SSAData, Op->Divisor);
+      const __uint128_t Source = (static_cast<__uint128_t>(SrcHigh) << 64) | SrcLow;
+      const __uint128_t Res = Source % Divisor;
      // We only store the lower bits of the result
      memcpy(GDP, &Res, OpSize);
      break;
@@ -717,62 +739,62 @@ DEF_OP(LURem) {

 DEF_OP(Not) {
  auto Op = IROp->C<IR::IROp_Not>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
+  const uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Src);
  const uint64_t mask[9]= { 0, 0xFF, 0xFFFF, 0, 0xFFFFFFFF, 0, 0, 0, 0xFFFFFFFFFFFFFFFFULL };
-  uint64_t Mask = mask[OpSize];
+  const uint64_t Mask = mask[OpSize];
  GD = (~Src) & Mask;
 }

 DEF_OP(Popcount) {
  auto Op = IROp->C<IR::IROp_Popcount>();
-  uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
+  const uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Src);
  GD = std::popcount(Src);
 }

 DEF_OP(FindLSB) {
  auto Op = IROp->C<IR::IROp_FindLSB>();
-  uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Result = FindFirstSetBit(Src);
+  const uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Src);
+  const uint64_t Result = FindFirstSetBit(Src);
  GD = Result - 1;
 }

 DEF_OP(FindMSB) {
  auto Op = IROp->C<IR::IROp_FindMSB>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

  switch (OpSize) {
-    case 1: GD = (OpSize * 8 - std::countl_zero(*GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[0]))) - 1; break;
-    case 2: GD = (OpSize * 8 - std::countl_zero(*GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0]))) - 1; break;
-    case 4: GD = (OpSize * 8 - std::countl_zero(*GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0]))) - 1; break;
-    case 8: GD = (OpSize * 8 - std::countl_zero(*GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]))) - 1; break;
+    case 1: GD = (OpSize * 8 - std::countl_zero(*GetSrc<uint8_t*>(Data->SSAData, Op->Src))) - 1; break;
+    case 2: GD = (OpSize * 8 - std::countl_zero(*GetSrc<uint16_t*>(Data->SSAData, Op->Src))) - 1; break;
+    case 4: GD = (OpSize * 8 - std::countl_zero(*GetSrc<uint32_t*>(Data->SSAData, Op->Src))) - 1; break;
+    case 8: GD = (OpSize * 8 - std::countl_zero(*GetSrc<uint64_t*>(Data->SSAData, Op->Src))) - 1; break;
    default: LOGMAN_MSG_A_FMT("Unknown FindMSB size: {}", OpSize); break;
  }
 }

 DEF_OP(FindTrailingZeros) {
  auto Op = IROp->C<IR::IROp_FindTrailingZeros>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

  switch (OpSize) {
    case 1: {
-      auto Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[0]);
+      const auto Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Src);
      GD = std::countr_zero(Src);
      break;
    }
    case 2: {
-      auto Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0]);
+      const auto Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Src);
      GD = std::countr_zero(Src);
      break;
    }
    case 4: {
-      auto Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0]);
+      const auto Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Src);
      GD = std::countr_zero(Src);
      break;
    }
    case 8: {
-      auto Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
+      const auto Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Src);
      GD = std::countr_zero(Src);
      break;
    }
@@ -782,26 +804,26 @@ DEF_OP(FindTrailingZeros) {

 DEF_OP(CountLeadingZeroes) {
  auto Op = IROp->C<IR::IROp_CountLeadingZeroes>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

  switch (OpSize) {
    case 1: {
-      auto Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[0]);
+      const auto Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Src);
      GD = std::countl_zero(Src);
      break;
    }
    case 2: {
-      auto Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0]);
+      const auto Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Src);
      GD = std::countl_zero(Src);
      break;
    }
    case 4: {
-      auto Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0]);
+      const auto Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Src);
      GD = std::countl_zero(Src);
      break;
    }
    case 8: {
-      auto Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
+      const auto Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Src);
      GD = std::countl_zero(Src);
      break;
    }
@@ -811,12 +833,12 @@ DEF_OP(CountLeadingZeroes) {

 DEF_OP(Rev) {
  auto Op = IROp->C<IR::IROp_Rev>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

  switch (OpSize) {
-    case 2: GD = BSwap16(*GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0])); break;
-    case 4: GD = BSwap32(*GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0])); break;
-    case 8: GD = BSwap64(*GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0])); break;
+    case 2: GD = BSwap16(*GetSrc<uint16_t*>(Data->SSAData, Op->Src)); break;
+    case 4: GD = BSwap32(*GetSrc<uint32_t*>(Data->SSAData, Op->Src)); break;
+    case 8: GD = BSwap64(*GetSrc<uint64_t*>(Data->SSAData, Op->Src)); break;
    default: LOGMAN_MSG_A_FMT("Unknown REV size: {}", OpSize); break;
  }
 }
@@ -824,36 +846,36 @@ DEF_OP(Rev) {
 DEF_OP(Bfi) {
  auto Op = IROp->C<IR::IROp_Bfi>();
  uint64_t SourceMask = (1ULL << Op->Width) - 1;
-  if (Op->Width == 64)
+  if (Op->Width == 64) {
    SourceMask = ~0ULL;
-  uint64_t DestMask = ~(SourceMask << Op->lsb);
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-  uint64_t Res = (Src1 & DestMask) | ((Src2 & SourceMask) << Op->lsb);
+  }
+  const uint64_t DestMask = ~(SourceMask << Op->lsb);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Dest);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src);
+  const uint64_t Res = (Src1 & DestMask) | ((Src2 & SourceMask) << Op->lsb);
  GD = Res;
 }

 DEF_OP(Bfe) {
  auto Op = IROp->C<IR::IROp_Bfe>();
-  uint8_t OpSize = IROp->Size;

-  LOGMAN_THROW_A_FMT(OpSize <= 8, "OpSize is too large for BFE: {}", OpSize);
+  LOGMAN_THROW_AA_FMT(IROp->Size <= 8, "OpSize is too large for BFE: {}", IROp->Size);
  uint64_t SourceMask = (1ULL << Op->Width) - 1;
-  if (Op->Width == 64)
+  if (Op->Width == 64) {
    SourceMask = ~0ULL;
+  }
  SourceMask <<= Op->lsb;
-  uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
+  const uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Src);
  GD = (Src & SourceMask) >> Op->lsb;
 }

 DEF_OP(Sbfe) {
  auto Op = IROp->C<IR::IROp_Sbfe>();
-  uint8_t OpSize = IROp->Size;

-  LOGMAN_THROW_A_FMT(OpSize <= 8, "OpSize is too large for SBFE: {}", OpSize);
-  int64_t Src = *GetSrc<int64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t ShiftLeftAmount = (64 - (Op->Width + Op->lsb));
-  uint64_t ShiftRightAmount = ShiftLeftAmount + Op->lsb;
+  LOGMAN_THROW_AA_FMT(IROp->Size <= 8, "OpSize is too large for SBFE: {}", IROp->Size);
+  int64_t Src = *GetSrc<int64_t*>(Data->SSAData, Op->Src);
+  const uint64_t ShiftLeftAmount = (64 - (Op->Width + Op->lsb));
+  const uint64_t ShiftRightAmount = ShiftLeftAmount + Op->lsb;
  Src <<= ShiftLeftAmount;
  Src >>= ShiftRightAmount;
  GD = Src;
@@ -861,85 +883,113 @@ DEF_OP(Sbfe) {

 DEF_OP(Select) {
  auto Op = IROp->C<IR::IROp_Select>();
-  uint8_t OpSize = IROp->Size;
-
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint8_t OpSize = IROp->Size;

  uint64_t ArgTrue;
  uint64_t ArgFalse;
-
  if (OpSize == 4) {
-    ArgTrue = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[2]);
-    ArgFalse = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[3]);
+    ArgTrue = *GetSrc<uint32_t*>(Data->SSAData, Op->TrueVal);
+    ArgFalse = *GetSrc<uint32_t*>(Data->SSAData, Op->FalseVal);
  } else {
-    ArgTrue = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[2]);
-    ArgFalse = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[3]);
+    ArgTrue = *GetSrc<uint64_t*>(Data->SSAData, Op->TrueVal);
+    ArgFalse = *GetSrc<uint64_t*>(Data->SSAData, Op->FalseVal);
  }

  bool CompResult;

-  if (Op->CompareSize == 4)
+  if (Op->CompareSize == 4) {
+    const auto Src1 = *GetSrc<uint32_t*>(Data->SSAData, Op->Cmp1);
+    const auto Src2 = *GetSrc<uint32_t*>(Data->SSAData, Op->Cmp2);
    CompResult = IsConditionTrue<uint32_t, int32_t, float>(Op->Cond.Val, Src1, Src2);
-  else
+  }
+  else if (Op->CompareSize == 8) {
+    const auto Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Cmp1);
+    const auto Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Cmp2);
    CompResult = IsConditionTrue<uint64_t, int64_t, double>(Op->Cond.Val, Src1, Src2);
+  }
+  else if (Op->CompareSize == 16) {
+    const auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Cmp1);
+    const auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Cmp2);
+    CompResult = IsConditionTrue<__uint128_t, __int128_t, double>(Op->Cond.Val, Src1, Src2);
+  }
+  else {
+    LOGMAN_MSG_A_FMT("Unknown select size: {}", Op->CompareSize);
+    FEX_UNREACHABLE;
+  }

  GD = CompResult ? ArgTrue : ArgFalse;
 }

 DEF_OP(VExtractToGPR) {
-  auto Op = IROp->C<IR::IROp_VExtractToGPR>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_VExtractToGPR>();
+  const auto OpSize = IROp->Size;

-  uint32_t SourceSize = GetOpSize(Data->CurrentIR, Op->Header.Args[0]);
+  constexpr auto AVXRegSize = Core::CPUState::XMM_AVX_REG_SIZE;
+  constexpr auto SSERegSize = Core::CPUState::XMM_SSE_REG_SIZE;
+  constexpr auto SSEBitSize = SSERegSize * 8;

-  LOGMAN_THROW_A_FMT(OpSize <= 16, "OpSize is too large for VExtractToGPR: {}", OpSize);
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto ElementSizeBits = ElementSize * 8;
+  const auto Shift = ElementSizeBits * Op->Index;

-  if (SourceSize == 16) {
-    __uint128_t SourceMask = (1ULL << (Op->Header.ElementSize * 8)) - 1;
-    uint64_t Shift = Op->Header.ElementSize * Op->Idx * 8;
-    if (Op->Header.ElementSize == 8)
+  const uint32_t SourceSize = GetOpSize(Data->CurrentIR, Op->Vector);
+
+  LOGMAN_THROW_AA_FMT(OpSize <= AVXRegSize,
+                      "OpSize is too large for VExtractToGPR: {}", OpSize);
+
+  if (SourceSize >= SSERegSize) {
+    __uint128_t SourceMask = (1ULL << ElementSizeBits) - 1;
+    if (ElementSize == 8) {
      SourceMask = ~0ULL;
+    }

-    __uint128_t Src = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
-    Src >>= Shift;
-    Src &= SourceMask;
-    memcpy(GDP, &Src, Op->Header.ElementSize);
+    const auto Src = *GetSrc<InterpVector256*>(Data->SSAData, Op->Vector);
+
+    const auto GetResult = [&] {
+      if (Shift >= SSEBitSize) {
+        const auto NormalizedShift = Shift - SSEBitSize;
+        return (Src.Upper >> NormalizedShift) & SourceMask;
+      } else {
+        return (Src.Lower >> Shift) & SourceMask;
+      }
+    };
+
+    const auto Result = GetResult();
+    memcpy(GDP, &Result, ElementSize);
  }
  else {
-    uint64_t SourceMask = (1ULL << (Op->Header.ElementSize * 8)) - 1;
-    uint64_t Shift = Op->Header.ElementSize * Op->Idx * 8;
-    if (Op->Header.ElementSize == 8)
+    uint64_t SourceMask = (1ULL << ElementSizeBits) - 1;
+    if (ElementSize == 8) {
      SourceMask = ~0ULL;
+    }

-    uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-    Src >>= Shift;
-    Src &= SourceMask;
-    GD = Src;
+    const uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Vector);
+    const uint64_t Result = (Src >> Shift) & SourceMask;
+    GD = Result;
  }
 }

 DEF_OP(Float_ToGPR_ZS) {
  auto Op = IROp->C<IR::IROp_Float_ToGPR_ZS>();
-  uint16_t Conv = (IROp->Size << 8) | Op->SrcElementSize;
+  const uint16_t Conv = (IROp->Size << 8) | Op->SrcElementSize;
  switch (Conv) {
    case 0x0804: { // int64_t <- float
-      int64_t Dst = (int64_t)std::trunc(*GetSrc<float*>(Data->SSAData, Op->Header.Args[0]));
+      const int64_t Dst = (int64_t)std::trunc(*GetSrc<float*>(Data->SSAData, Op->Scalar));
      memcpy(GDP, &Dst, IROp->Size);
      break;
    }
    case 0x0808: { // int64_t <- double
-      int64_t Dst = (int64_t)std::trunc(*GetSrc<double*>(Data->SSAData, Op->Header.Args[0]));
+      const int64_t Dst = (int64_t)std::trunc(*GetSrc<double*>(Data->SSAData, Op->Scalar));
      memcpy(GDP, &Dst, IROp->Size);
      break;
    }
    case 0x0404: { // int32_t <- float
-      int32_t Dst = (int32_t)std::trunc(*GetSrc<float*>(Data->SSAData, Op->Header.Args[0]));
+      const int32_t Dst = (int32_t)std::trunc(*GetSrc<float*>(Data->SSAData, Op->Scalar));
      memcpy(GDP, &Dst, IROp->Size);
      break;
    }
    case 0x0408: { // int32_t <- double
-      int32_t Dst = (int32_t)std::trunc(*GetSrc<double*>(Data->SSAData, Op->Header.Args[0]));
+      const int32_t Dst = (int32_t)std::trunc(*GetSrc<double*>(Data->SSAData, Op->Scalar));
      memcpy(GDP, &Dst, IROp->Size);
      break;
    }
@@ -948,25 +998,25 @@ DEF_OP(Float_ToGPR_ZS) {

 DEF_OP(Float_ToGPR_S) {
  auto Op = IROp->C<IR::IROp_Float_ToGPR_S>();
-  uint16_t Conv = (IROp->Size << 8) | Op->SrcElementSize;
+  const uint16_t Conv = (IROp->Size << 8) | Op->SrcElementSize;
  switch (Conv) {
    case 0x0804: { // int64_t <- float
-      int64_t Dst = (int64_t)std::nearbyint(*GetSrc<float*>(Data->SSAData, Op->Header.Args[0]));
+      const int64_t Dst = (int64_t)std::nearbyint(*GetSrc<float*>(Data->SSAData, Op->Scalar));
      memcpy(GDP, &Dst, IROp->Size);
      break;
    }
    case 0x0808: { // int64_t <- double
-      int64_t Dst = (int64_t)std::nearbyint(*GetSrc<double*>(Data->SSAData, Op->Header.Args[0]));
+      const int64_t Dst = (int64_t)std::nearbyint(*GetSrc<double*>(Data->SSAData, Op->Scalar));
      memcpy(GDP, &Dst, IROp->Size);
      break;
    }
    case 0x0404: { // int32_t <- float
-      int32_t Dst = (int32_t)std::nearbyint(*GetSrc<float*>(Data->SSAData, Op->Header.Args[0]));
+      const int32_t Dst = (int32_t)std::nearbyint(*GetSrc<float*>(Data->SSAData, Op->Scalar));
      memcpy(GDP, &Dst, IROp->Size);
      break;
    }
    case 0x0408: { // int32_t <- double
-      int32_t Dst = (int32_t)std::nearbyint(*GetSrc<double*>(Data->SSAData, Op->Header.Args[0]));
+      const int32_t Dst = (int32_t)std::nearbyint(*GetSrc<double*>(Data->SSAData, Op->Scalar));
      memcpy(GDP, &Dst, IROp->Size);
      break;
    }
@@ -977,9 +1027,9 @@ DEF_OP(FCmp) {
  auto Op = IROp->C<IR::IROp_FCmp>();
  uint32_t ResultFlags{};
  if (Op->ElementSize == 4) {
-    float Src1 = *GetSrc<float*>(Data->SSAData, Op->Header.Args[0]);
-    float Src2 = *GetSrc<float*>(Data->SSAData, Op->Header.Args[1]);
-    bool Unordered = std::isnan(Src1) || std::isnan(Src2);
+    const float Src1 = *GetSrc<float*>(Data->SSAData, Op->Scalar1);
+    const float Src2 = *GetSrc<float*>(Data->SSAData, Op->Scalar2);
+    const bool Unordered = std::isnan(Src1) || std::isnan(Src2);
    if (Op->Flags & (1 << IR::FCMP_FLAG_LT)) {
      if (Unordered || (Src1 < Src2)) {
        ResultFlags |= (1 << IR::FCMP_FLAG_LT);
@@ -997,9 +1047,9 @@ DEF_OP(FCmp) {
    }
  }
  else {
-    double Src1 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
-    double Src2 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[1]);
-    bool Unordered = std::isnan(Src1) || std::isnan(Src2);
+    const double Src1 = *GetSrc<double*>(Data->SSAData, Op->Scalar1);
+    const double Src2 = *GetSrc<double*>(Data->SSAData, Op->Scalar2);
+    const bool Unordered = std::isnan(Src1) || std::isnan(Src2);
    if (Op->Flags & (1 << IR::FCMP_FLAG_LT)) {
      if (Unordered || (Src1 < Src2)) {
        ResultFlags |= (1 << IR::FCMP_FLAG_LT);
@@ -313,30 +313,29 @@ uint64_t AtomicCompareAndSwap(uint64_t expected, uint64_t desired, uint64_t *add
 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
 DEF_OP(CASPair) {
  auto Op = IROp->C<IR::IROp_CASPair>();
-  uint8_t OpSize = IROp->Size;

  // Size is the size of each pair element
-  switch (OpSize) {
+  switch (IROp->ElementSize) {
    case 4: {
      GD = AtomicCompareAndSwap(
-        *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]),
-        *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]),
-        *GetSrc<uint64_t**>(Data->SSAData, Op->Header.Args[2])
+        *GetSrc<uint64_t*>(Data->SSAData, Op->Expected),
+        *GetSrc<uint64_t*>(Data->SSAData, Op->Desired),
+        *GetSrc<uint64_t**>(Data->SSAData, Op->Addr)
      );
      break;
    }
    case 8: {
-      std::atomic<__uint128_t> *MemData = *GetSrc<std::atomic<__uint128_t> **>(Data->SSAData, Op->Header.Args[2]);
+      std::atomic<__uint128_t> *MemData = *GetSrc<std::atomic<__uint128_t> **>(Data->SSAData, Op->Addr);

-      __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
-      __uint128_t Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
+      __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Expected);
+      __uint128_t Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Desired);

      __uint128_t Expected = Src1;
      bool Result = MemData->compare_exchange_strong(Expected, Src2);
      memcpy(GDP, Result ? &Src1 : &Expected, 16);
      break;
    }
-    default: LOGMAN_MSG_A_FMT("Unknown CAS size: {}", OpSize); break;
+    default: LOGMAN_MSG_A_FMT("Unknown CAS size: {}", IROp->ElementSize); break;
  }
 }

@@ -347,33 +346,33 @@ DEF_OP(CAS) {
  switch (OpSize) {
    case 1: {
      GD = AtomicCompareAndSwap(
-        *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[0]),
-        *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]),
-        *GetSrc<uint8_t**>(Data->SSAData, Op->Header.Args[2])
+        *GetSrc<uint8_t*>(Data->SSAData, Op->Expected),
+        *GetSrc<uint8_t*>(Data->SSAData, Op->Desired),
+        *GetSrc<uint8_t**>(Data->SSAData, Op->Addr)
      );
      break;
    }
    case 2: {
      GD = AtomicCompareAndSwap(
-        *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0]),
-        *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]),
-        *GetSrc<uint16_t**>(Data->SSAData, Op->Header.Args[2])
+        *GetSrc<uint16_t*>(Data->SSAData, Op->Expected),
+        *GetSrc<uint16_t*>(Data->SSAData, Op->Desired),
+        *GetSrc<uint16_t**>(Data->SSAData, Op->Addr)
      );
      break;
    }
    case 4: {
      GD = AtomicCompareAndSwap(
-        *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0]),
-        *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]),
-        *GetSrc<uint32_t**>(Data->SSAData, Op->Header.Args[2])
+        *GetSrc<uint32_t*>(Data->SSAData, Op->Expected),
+        *GetSrc<uint32_t*>(Data->SSAData, Op->Desired),
+        *GetSrc<uint32_t**>(Data->SSAData, Op->Addr)
      );
      break;
    }
    case 8: {
      GD = AtomicCompareAndSwap(
-        *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]),
-        *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]),
-        *GetSrc<uint64_t**>(Data->SSAData, Op->Header.Args[2])
+        *GetSrc<uint64_t*>(Data->SSAData, Op->Expected),
+        *GetSrc<uint64_t*>(Data->SSAData, Op->Desired),
+        *GetSrc<uint64_t**>(Data->SSAData, Op->Addr)
      );
      break;
    }
@@ -385,26 +384,26 @@ DEF_OP(AtomicAdd) {
  auto Op = IROp->C<IR::IROp_AtomicAdd>();
  switch (IROp->Size) {
    case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
      *MemData += Src;
      break;
    }
    case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
      *MemData += Src;
      break;
    }
    case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
      *MemData += Src;
      break;
    }
    case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
      *MemData += Src;
      break;
    }
@@ -416,26 +415,26 @@ DEF_OP(AtomicSub) {
  auto Op = IROp->C<IR::IROp_AtomicSub>();
  switch (IROp->Size) {
    case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
      *MemData -= Src;
      break;
    }
    case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
      *MemData -= Src;
      break;
    }
    case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
      *MemData -= Src;
      break;
    }
    case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
      *MemData -= Src;
      break;
    }
@@ -447,26 +446,26 @@ DEF_OP(AtomicAnd) {
  auto Op = IROp->C<IR::IROp_AtomicAnd>();
  switch (IROp->Size) {
    case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
      *MemData &= Src;
      break;
    }
    case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
      *MemData &= Src;
      break;
    }
    case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
      *MemData &= Src;
      break;
    }
    case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
      *MemData &= Src;
      break;
    }
@@ -478,26 +477,26 @@ DEF_OP(AtomicOr) {
  auto Op = IROp->C<IR::IROp_AtomicOr>();
  switch (IROp->Size) {
    case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
      *MemData |= Src;
      break;
    }
    case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
      *MemData |= Src;
      break;
    }
    case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
      *MemData |= Src;
      break;
    }
    case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
      *MemData |= Src;
      break;
    }
@@ -509,26 +508,26 @@ DEF_OP(AtomicXor) {
  auto Op = IROp->C<IR::IROp_AtomicXor>();
  switch (IROp->Size) {
    case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
      *MemData ^= Src;
      break;
    }
    case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
      *MemData ^= Src;
      break;
    }
    case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
      *MemData ^= Src;
      break;
    }
    case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
      *MemData ^= Src;
      break;
    }
@@ -540,29 +539,29 @@ DEF_OP(AtomicSwap) {
  auto Op = IROp->C<IR::IROp_AtomicSwap>();
  switch (IROp->Size) {
    case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
      uint8_t Previous = MemData->exchange(Src);
      GD = Previous;
      break;
    }
    case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
      uint16_t Previous = MemData->exchange(Src);
      GD = Previous;
      break;
    }
    case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
      uint32_t Previous = MemData->exchange(Src);
      GD = Previous;
      break;
    }
    case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
      uint64_t Previous = MemData->exchange(Src);
      GD = Previous;
      break;
@@ -575,29 +574,29 @@ DEF_OP(AtomicFetchAdd) {
  auto Op = IROp->C<IR::IROp_AtomicFetchAdd>();
  switch (IROp->Size) {
    case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
      uint8_t Previous = MemData->fetch_add(Src);
      GD = Previous;
      break;
    }
    case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
      uint16_t Previous = MemData->fetch_add(Src);
      GD = Previous;
      break;
    }
    case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
      uint32_t Previous = MemData->fetch_add(Src);
      GD = Previous;
      break;
    }
    case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
      uint64_t Previous = MemData->fetch_add(Src);
      GD = Previous;
      break;
@@ -610,29 +609,29 @@ DEF_OP(AtomicFetchSub) {
  auto Op = IROp->C<IR::IROp_AtomicFetchSub>();
  switch (IROp->Size) {
    case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
      uint8_t Previous = MemData->fetch_sub(Src);
      GD = Previous;
      break;
    }
    case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
      uint16_t Previous = MemData->fetch_sub(Src);
      GD = Previous;
      break;
    }
    case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
      uint32_t Previous = MemData->fetch_sub(Src);
      GD = Previous;
      break;
    }
    case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
      uint64_t Previous = MemData->fetch_sub(Src);
      GD = Previous;
      break;
@@ -645,29 +644,29 @@ DEF_OP(AtomicFetchAnd) {
  auto Op = IROp->C<IR::IROp_AtomicFetchAnd>();
  switch (IROp->Size) {
    case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
      uint8_t Previous = MemData->fetch_and(Src);
      GD = Previous;
      break;
    }
    case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
      uint16_t Previous = MemData->fetch_and(Src);
      GD = Previous;
      break;
    }
    case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
      uint32_t Previous = MemData->fetch_and(Src);
      GD = Previous;
      break;
    }
    case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
      uint64_t Previous = MemData->fetch_and(Src);
      GD = Previous;
      break;
@@ -680,29 +679,29 @@ DEF_OP(AtomicFetchOr) {
  auto Op = IROp->C<IR::IROp_AtomicFetchOr>();
  switch (IROp->Size) {
    case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
      uint8_t Previous = MemData->fetch_or(Src);
      GD = Previous;
      break;
    }
    case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
      uint16_t Previous = MemData->fetch_or(Src);
      GD = Previous;
      break;
    }
    case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
      uint32_t Previous = MemData->fetch_or(Src);
      GD = Previous;
      break;
    }
    case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
      uint64_t Previous = MemData->fetch_or(Src);
      GD = Previous;
      break;
@@ -715,29 +714,29 @@ DEF_OP(AtomicFetchXor) {
  auto Op = IROp->C<IR::IROp_AtomicFetchXor>();
  switch (IROp->Size) {
    case 1: {
-      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
      uint8_t Previous = MemData->fetch_xor(Src);
      GD = Previous;
      break;
    }
    case 2: {
-      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
      uint16_t Previous = MemData->fetch_xor(Src);
      GD = Previous;
      break;
    }
    case 4: {
-      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
      uint32_t Previous = MemData->fetch_xor(Src);
      GD = Previous;
      break;
    }
    case 8: {
-      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
      uint64_t Previous = MemData->fetch_xor(Src);
      GD = Previous;
      break;
@@ -751,22 +750,22 @@ DEF_OP(AtomicFetchNeg) {
  switch (IROp->Size) {
    case 1: {
      using Type = uint8_t;
-      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Header.Args[0]));
+      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Addr));
      break;
    }
    case 2: {
      using Type = uint16_t;
-      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Header.Args[0]));
+      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Addr));
      break;
    }
    case 4: {
      using Type = uint32_t;
-      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Header.Args[0]));
+      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Addr));
      break;
    }
    case 8: {
      using Type = uint64_t;
-      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Header.Args[0]));
+      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Addr));
      break;
    }
    default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
@@ -4,6 +4,7 @@ tags: backend|interpreter
 $end_info$
 */

+#include "Interface/Context/Context.h"
 #include "Interface/Core/Interpreter/InterpreterClass.h"
 #include "Interface/Core/Interpreter/InterpreterOps.h"
 #include "Interface/Core/Interpreter/InterpreterDefines.h"
@@ -16,33 +17,10 @@ $end_info$
 #include <unistd.h>

 namespace FEXCore::CPU {
-[[noreturn]]
-static void SignalReturn(FEXCore::Core::InternalThreadState *Thread) {
-  Thread->CTX->SignalThread(Thread, FEXCore::Core::SignalEvent::Return);
-
-  LOGMAN_MSG_A_FMT("unreachable");
-  FEX_UNREACHABLE;
-}
-
 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
-DEF_OP(GuestCallDirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
-DEF_OP(GuestCallIndirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
-DEF_OP(GuestReturn) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
-DEF_OP(SignalReturn) {
-  SignalReturn(Data->State);
-}

 DEF_OP(CallbackReturn) {
-  Data->State->CTX->InterpreterCallbackReturn(Data->State, Data->StackEntry);
+  Data->State->CurrentFrame->Pointers.Interpreter.CallbackReturn(Data->State, Data->StackEntry);
 }

 DEF_OP(ExitFunction) {
@@ -52,7 +30,7 @@ DEF_OP(ExitFunction) {
  uintptr_t* ContextPtr = reinterpret_cast<uintptr_t*>(Data->State->CurrentFrame);

  void *ContextData = reinterpret_cast<void*>(ContextPtr);
-  void *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
+  void *Src = GetSrc<void*>(Data->SSAData, Op->NewRIP);

  memcpy(ContextData, Src, OpSize);

@@ -61,22 +39,22 @@ DEF_OP(ExitFunction) {

 DEF_OP(Jump) {
  auto Op = IROp->C<IR::IROp_Jump>();
-  uintptr_t ListBegin = Data->CurrentIR->GetListData();
-  uintptr_t DataBegin = Data->CurrentIR->GetData();
+  const uintptr_t ListBegin = Data->CurrentIR->GetListData();
+  const uintptr_t DataBegin = Data->CurrentIR->GetData();

-  Data->BlockIterator = IR::NodeIterator(ListBegin, DataBegin, Op->Header.Args[0]);
+  Data->BlockIterator = IR::NodeIterator(ListBegin, DataBegin, Op->TargetBlock);
  Data->BlockResults.Redo = true;
 }

 DEF_OP(CondJump) {
  auto Op = IROp->C<IR::IROp_CondJump>();
-  uintptr_t ListBegin = Data->CurrentIR->GetListData();
-  uintptr_t DataBegin = Data->CurrentIR->GetData();
+  const uintptr_t ListBegin = Data->CurrentIR->GetListData();
+  const uintptr_t DataBegin = Data->CurrentIR->GetData();

  bool CompResult;

-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Cmp1);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Cmp2);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Cmp1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Cmp2);

  if (Op->CompareSize == 4)
    CompResult = IsConditionTrue<uint32_t, int32_t, float>(Op->Cond.Val, Src1, Src2);
@@ -101,7 +79,7 @@ DEF_OP(Syscall) {
    Args.Argument[j] = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[j]);
  }

-  uint64_t Res = FEXCore::Context::HandleSyscall(Data->State->CTX->SyscallHandler, Data->State->CurrentFrame, &Args);
+  uint64_t Res = FEXCore::Context::HandleSyscall(static_cast<Context::ContextImpl*>(Data->State->CTX)->SyscallHandler, Data->State->CurrentFrame, &Args);
  GD = Res;
 }

@@ -136,8 +114,8 @@ DEF_OP(InlineSyscall) {
 DEF_OP(Thunk) {
  auto Op = IROp->C<IR::IROp_Thunk>();

-  auto thunkFn = Data->State->CTX->ThunkHandler->LookupThunk(Op->ThunkNameHash);
-  thunkFn(*GetSrc<void**>(Data->SSAData, Op->Header.Args[0]));
+  auto thunkFn = static_cast<Context::ContextImpl*>(Data->State->CTX)->ThunkHandler->LookupThunk(Op->ThunkNameHash);
+  thunkFn(*GetSrc<void**>(Data->SSAData, Op->ArgPtr));
 }

 DEF_OP(ValidateCode) {
@@ -151,20 +129,29 @@ DEF_OP(ValidateCode) {
  }
 }

-DEF_OP(RemoveCodeEntry) {
-  Data->State->CTX->RemoveCodeEntry(Data->State, Data->CurrentEntry);
+DEF_OP(ThreadRemoveCodeEntry) {
+  static_cast<Context::ContextImpl*>(Data->State->CTX)->ThreadRemoveCodeEntryFromJit(Data->State->CurrentFrame, Data->CurrentEntry);
 }

 DEF_OP(CPUID) {
  auto Op = IROp->C<IR::IROp_CPUID>();
  uint64_t *DstPtr = GetDest<uint64_t*>(Data->SSAData, Node);
-  uint64_t Arg = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Leaf = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Arg = *GetSrc<uint64_t*>(Data->SSAData, Op->Function);
+  const uint64_t Leaf = *GetSrc<uint64_t*>(Data->SSAData, Op->Leaf);

-  auto Results = Data->State->CTX->CPUID.RunFunction(Arg, Leaf);
+  auto Results = Data->State->CTX->RunCPUIDFunction(Arg, Leaf);
  memcpy(DstPtr, &Results, sizeof(uint32_t) * 4);
 }

+DEF_OP(XGETBV) {
+  auto Op = IROp->C<IR::IROp_XGetBV>();
+  uint32_t *DstPtr = GetDest<uint32_t*>(Data->SSAData, Node);
+  const uint32_t Function = *GetSrc<uint32_t*>(Data->SSAData, Op->Function);
+
+  auto Results = Data->State->CTX->RunXCRFunction(Function);
+  memcpy(DstPtr, &Results, sizeof(uint32_t) * 2);
+}
+
 #undef DEF_OP

 } // namespace FEXCore::CPU
@@ -13,53 +13,94 @@ $end_info$
 namespace FEXCore::CPU {
 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
 DEF_OP(VInsGPR) {
-  auto Op = IROp->C<IR::IROp_VInsGPR>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_VInsGPR>();
+  const auto OpSize = IROp->Size;

-  __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
-  __uint128_t Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto ElementSizeBits = ElementSize * 8;
+  constexpr auto SSEBitSize = Core::CPUState::XMM_SSE_REG_SIZE * 8;

-  uint64_t Offset = Op->Index * Op->Header.ElementSize * 8;
-  __uint128_t Mask = (1ULL << (Op->Header.ElementSize * 8)) - 1;
-  if (Op->Header.ElementSize == 8) {
+  const uint64_t Offset = Op->DestIdx * ElementSizeBits;
+  const auto InUpperLane = Offset >= SSEBitSize;
+
+  __uint128_t Mask = (1ULL << ElementSizeBits) - 1;
+  if (ElementSize == 8) {
    Mask = ~0ULL;
  }
-  Src2 = Src2 & Mask;
-  Mask <<= Offset;
+
+  const auto Src1 = *GetSrc<InterpVector256*>(Data->SSAData, Op->DestVector);
+  const auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Src);
+
+  const auto Scalar = Src2 & Mask;
+  const auto ScaledOffset = InUpperLane ? Offset - SSEBitSize
+                                        : Offset;
+
+  // Now shift into place and set all bits but
+  // the ones where we're going to insert our value.
+  Mask <<= ScaledOffset;
  Mask = ~Mask;
-  __uint128_t Dst = Src1 & Mask;
-  Dst |= Src2 << Offset;
+
+  const auto Dst = [&] {
+    if (InUpperLane) {
+      return InterpVector256{
+        .Lower = Src1.Lower,
+        .Upper = (Src1.Upper & Mask) | (Scalar << ScaledOffset),
+      };
+    } else {
+      return InterpVector256{
+        .Lower = (Src1.Lower & Mask) | (Scalar << ScaledOffset),
+        .Upper = Src1.Upper,
+      };
+    }
+  }();

  memcpy(GDP, &Dst, OpSize);
 }

 DEF_OP(VCastFromGPR) {
  auto Op = IROp->C<IR::IROp_VCastFromGPR>();
-  memcpy(GDP, GetSrc<void*>(Data->SSAData, Op->Header.Args[0]), Op->Header.ElementSize);
+  memcpy(GDP, GetSrc<void*>(Data->SSAData, Op->Src), Op->Header.ElementSize);
+}
+
+DEF_OP(VDupFromGPR) {
+  const auto Op = IROp->C<IR::IROp_VDupFromGPR>();
+  const auto OpSize = IROp->Size;
+
+  const auto ElementSize = IROp->ElementSize;
+  const auto NumElements = OpSize / IROp->ElementSize;
+
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
+
+  const auto *Src = GetSrc<void*>(Data->SSAData, Op->Src);
+  for (size_t i = 0; i < NumElements; i++) {
+    memcpy(Tmp + (i * ElementSize), Src, ElementSize);
+  }
+
+  memcpy(GDP, Tmp, sizeof(Tmp));
 }

 DEF_OP(Float_FromGPR_S) {
  auto Op = IROp->C<IR::IROp_Float_FromGPR_S>();

-  uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
  switch (Conv) {
    case 0x0404: { // Float <- int32_t
-      float Dst = (float)*GetSrc<int32_t*>(Data->SSAData, Op->Header.Args[0]);
+      const float Dst = (float)*GetSrc<int32_t*>(Data->SSAData, Op->Src);
      memcpy(GDP, &Dst, Op->Header.ElementSize);
      break;
    }
    case 0x0408: { // Float <- int64_t
-      float Dst = (float)*GetSrc<int64_t*>(Data->SSAData, Op->Header.Args[0]);
+      const float Dst = (float)*GetSrc<int64_t*>(Data->SSAData, Op->Src);
      memcpy(GDP, &Dst, Op->Header.ElementSize);
      break;
    }
    case 0x0804: { // Double <- int32_t
-      double Dst = (double)*GetSrc<int32_t*>(Data->SSAData, Op->Header.Args[0]);
+      const double Dst = (double)*GetSrc<int32_t*>(Data->SSAData, Op->Src);
      memcpy(GDP, &Dst, Op->Header.ElementSize);
      break;
    }
    case 0x0808: { // Double <- int64_t
-      double Dst = (double)*GetSrc<int64_t*>(Data->SSAData, Op->Header.Args[0]);
+      const double Dst = (double)*GetSrc<int64_t*>(Data->SSAData, Op->Src);
      memcpy(GDP, &Dst, Op->Header.ElementSize);
      break;
    }
@@ -68,15 +109,15 @@ DEF_OP(Float_FromGPR_S) {

 DEF_OP(Float_FToF) {
  auto Op = IROp->C<IR::IROp_Float_FToF>();
-  uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
  switch (Conv) {
    case 0x0804: { // Double <- Float
-      double Dst = (double)*GetSrc<float*>(Data->SSAData, Op->Header.Args[0]);
+      const double Dst = (double)*GetSrc<float*>(Data->SSAData, Op->Scalar);
      memcpy(GDP, &Dst, 8);
      break;
    }
    case 0x0408: { // Float <- Double
-      float Dst = (float)*GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
+      const float Dst = (float)*GetSrc<double*>(Data->SSAData, Op->Scalar);
      memcpy(GDP, &Dst, 4);
      break;
    }
@@ -86,68 +127,78 @@ DEF_OP(Float_FToF) {

 DEF_OP(Vector_SToF) {
  auto Op = IROp->C<IR::IROp_Vector_SToF>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  void *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  uint8_t Tmp[16]{};
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};

-  uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;

-  auto Func = [](auto a, auto min, auto max) { return a; };
-  switch (Op->Header.ElementSize) {
+  const auto Func = [](auto a, auto min, auto max) { return a; };
+  switch (ElementSize) {
    DO_VECTOR_1SRC_2TYPE_OP(4, float, int32_t, Func, 0, 0)
    DO_VECTOR_1SRC_2TYPE_OP(8, double, int64_t, Func, 0, 0)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
  }
  memcpy(GDP, Tmp, OpSize);
 }

 DEF_OP(Vector_FToZS) {
-  auto Op = IROp->C<IR::IROp_Vector_FToZS>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_Vector_FToZS>();
+  const uint8_t OpSize = IROp->Size;

-  void *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  uint8_t Tmp[16]{};
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};

-  uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;

-  auto Func = [](auto a, auto min, auto max) { return std::trunc(a); };
-  switch (Op->Header.ElementSize) {
+  const auto Func = [](auto a, auto min, auto max) { return std::trunc(a); };
+  switch (ElementSize) {
    DO_VECTOR_1SRC_2TYPE_OP(4, int32_t, float, Func, 0, 0)
    DO_VECTOR_1SRC_2TYPE_OP(8, int64_t, double, Func, 0, 0)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
  }
  memcpy(GDP, Tmp, OpSize);
 }

 DEF_OP(Vector_FToS) {
-  auto Op = IROp->C<IR::IROp_Vector_FToS>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_Vector_FToS>();
+  const uint8_t OpSize = IROp->Size;

-  void *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  uint8_t Tmp[16]{};
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};

-  uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;

-  auto Func = [](auto a, auto min, auto max) { return std::nearbyint(a); };
-  switch (Op->Header.ElementSize) {
+  const auto Func = [](auto a, auto min, auto max) { return std::nearbyint(a); };
+  switch (ElementSize) {
    DO_VECTOR_1SRC_2TYPE_OP(4, int32_t, float, Func, 0, 0)
    DO_VECTOR_1SRC_2TYPE_OP(8, int64_t, double, Func, 0, 0)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
  }
  memcpy(GDP, Tmp, OpSize);
 }

 DEF_OP(Vector_FToF) {
-  auto Op = IROp->C<IR::IROp_Vector_FToF>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_Vector_FToF>();
+  const uint8_t OpSize = IROp->Size;

-  void *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  uint8_t Tmp[16]{};
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};

-  uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const uint16_t ElementSize = Op->Header.ElementSize;
+  const uint16_t Conv = (ElementSize << 8) | Op->SrcElementSize;

-  auto Func = [](auto a, auto min, auto max) { return a; };
+  const auto Func = [](auto a, auto min, auto max) { return a; };
  switch (Conv) {
    case 0x0804: { // Double <- float
      // Only the lower elements from the source
@@ -160,57 +211,60 @@ DEF_OP(Vector_FToF) {
      // Little bit tricky here
      // Sometimes is used to convert from a 128bit vector register
      // in to a 64bit vector register with different sized elements
-      // eg: %ssa5 i32v2 = Vector_FToF %ssa4 i128, #0x8
-      uint8_t Elements = (OpSize << 1) / Op->SrcElementSize;
+      // eg: %5 i32v2 = Vector_FToF %4 i128, #0x8
+      uint8_t Elements = OpSize == 8 ? 2 : OpSize / Op->SrcElementSize;
      DO_VECTOR_1SRC_2TYPE_OP_NOSIZE(float, double, Func, 0, 0)
      break;
    }
-    default: LOGMAN_MSG_A_FMT("Unknown Conversion Type : 0x{:04x}", Conv); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Conversion Type : 0x{:04x}", Conv);
+      break;
  }
  memcpy(GDP, Tmp, OpSize);
 }

 DEF_OP(Vector_FToI) {
-  auto Op = IROp->C<IR::IROp_Vector_FToI>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_Vector_FToI>();
+  const uint8_t OpSize = IROp->Size;

-  void *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  uint8_t Tmp[16]{};
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};

-  uint8_t Elements = OpSize / Op->Header.ElementSize;
-  auto Func_Nearest = [](auto a) { return std::rint(a); };
-  auto Func_Neg = [](auto a) { return std::floor(a); };
-  auto Func_Pos = [](auto a) { return std::ceil(a); };
-  auto Func_Trunc = [](auto a) { return std::trunc(a); };
-  auto Func_Host = [](auto a) { return std::rint(a); };
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
+  const auto Func_Nearest = [](auto a) { return std::rint(a); };
+  const auto Func_Neg = [](auto a) { return std::floor(a); };
+  const auto Func_Pos = [](auto a) { return std::ceil(a); };
+  const auto Func_Trunc = [](auto a) { return std::trunc(a); };
+  const auto Func_Host = [](auto a) { return std::rint(a); };

  switch (Op->Round) {
    case FEXCore::IR::Round_Nearest.Val:
-      switch (Op->Header.ElementSize) {
+      switch (ElementSize) {
        DO_VECTOR_1SRC_OP(4, float, Func_Nearest)
        DO_VECTOR_1SRC_OP(8, double, Func_Nearest)
      }
    break;
    case FEXCore::IR::Round_Negative_Infinity.Val:
-      switch (Op->Header.ElementSize) {
+      switch (ElementSize) {
        DO_VECTOR_1SRC_OP(4, float, Func_Neg)
        DO_VECTOR_1SRC_OP(8, double, Func_Neg)
      }
    break;
    case FEXCore::IR::Round_Positive_Infinity.Val:
-      switch (Op->Header.ElementSize) {
+      switch (ElementSize) {
        DO_VECTOR_1SRC_OP(4, float, Func_Pos)
        DO_VECTOR_1SRC_OP(8, double, Func_Pos)
      }
    break;
    case FEXCore::IR::Round_Towards_Zero.Val:
-      switch (Op->Header.ElementSize) {
+      switch (ElementSize) {
        DO_VECTOR_1SRC_OP(4, float, Func_Trunc)
        DO_VECTOR_1SRC_OP(8, double, Func_Trunc)
      }
    break;
    case FEXCore::IR::Round_Host.Val:
-      switch (Op->Header.ElementSize) {
+      switch (ElementSize) {
        DO_VECTOR_1SRC_OP(4, float, Func_Host)
        DO_VECTOR_1SRC_OP(8, double, Func_Host)
      }
@@ -298,12 +298,69 @@ namespace AES {
  }
 }

+namespace CRC32 {
+  // CRC32 per byte lookup table.
+  constexpr std::array<uint32_t, 256> CRC32CTable = []() consteval {
+    std::array<uint32_t, 256> Table{};
+
+    // Clang 11.x doesn't support bitreverse as a consteval
+    // constexpr uint32_t Polynomial = 0x1EDC6F41;
+    constexpr uint32_t PolynomialRev = 0x82F63B78; //__builtin_bitreverse32(Polynomial);
+
+    for (size_t Char = 0; Char < std::size(Table); ++Char) {
+      uint32_t CurrentChar = Char;
+      for (size_t i = 0; i < 8; ++i) {
+        if (CurrentChar & 1) {
+          CurrentChar = (CurrentChar >> 1) ^ PolynomialRev;
+        }
+        else {
+          CurrentChar >>= 1;
+        }
+      }
+      Table[Char] = CurrentChar;
+    }
+
+    return Table;
+  }();
+
+  uint32_t crc32cb(uint32_t Accumulator, uint8_t data) {
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ data] ^ Accumulator >> 8;
+    return Accumulator;
+  }
+
+  uint32_t crc32ch(uint32_t Accumulator, uint16_t data) {
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 0) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 8) & 0xFF)] ^ Accumulator >> 8;
+    return Accumulator;
+  }
+
+  uint32_t crc32cw(uint32_t Accumulator, uint32_t data) {
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >>  0) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >>  8) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 16) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 24) & 0xFF)] ^ Accumulator >> 8;
+    return Accumulator;
+  }
+
+  uint32_t crc32cx(uint32_t Accumulator, uint64_t data) {
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >>  0) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >>  8) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 16) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 24) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 32) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 40) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 48) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 56) & 0xFF)] ^ Accumulator >> 8;
+    return Accumulator;
+  }
+}
+
 namespace FEXCore::CPU {
 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)

 DEF_OP(AESImc) {
  auto Op = IROp->C<IR::IROp_VAESImc>();
-  __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
+  auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Vector);

  // Pseudo-code
  // Dst = InvMixColumns(STATE)
@@ -314,8 +371,8 @@ DEF_OP(AESImc) {

 DEF_OP(AESEnc) {
  auto Op = IROp->C<IR::IROp_VAESEnc>();
-  __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
-  __uint128_t Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
+  auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->State);
+  auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Key);

  // Pseudo-code
  // STATE = Src1
@@ -334,8 +391,8 @@ DEF_OP(AESEnc) {

 DEF_OP(AESEncLast) {
  auto Op = IROp->C<IR::IROp_VAESEncLast>();
-  __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
-  __uint128_t Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
+  auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->State);
+  auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Key);

  // Pseudo-code
  // STATE = Src1
@@ -352,8 +409,8 @@ DEF_OP(AESEncLast) {

 DEF_OP(AESDec) {
  auto Op = IROp->C<IR::IROp_VAESDec>();
-  __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
-  __uint128_t Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
+  auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->State);
+  auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Key);

  // Pseudo-code
  // STATE = Src1
@@ -372,8 +429,8 @@ DEF_OP(AESDec) {

 DEF_OP(AESDecLast) {
  auto Op = IROp->C<IR::IROp_VAESDecLast>();
-  __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
-  __uint128_t Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
+  auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->State);
+  auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Key);

  // Pseudo-code
  // STATE = Src1
@@ -390,7 +447,7 @@ DEF_OP(AESDecLast) {

 DEF_OP(AESKeyGenAssist) {
  auto Op = IROp->C<IR::IROp_VAESKeyGenAssist>();
-  uint8_t *Src1 = GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[0]);
+  const uint8_t *Src1 = GetSrc<uint8_t*>(Data->SSAData, Op->Src);

  // Pseudo-code
  // X3 = Src1[127:96]
@@ -429,6 +486,71 @@ DEF_OP(AESKeyGenAssist) {
  memcpy(GDP, &Tmp, sizeof(Tmp));
 }

+DEF_OP(CRC32) {
+  auto Op = IROp->C<IR::IROp_CRC32>();
+  uint32_t Src1 = *GetSrc<uint32_t*>(Data->SSAData, Op->Src1);
+  uint8_t *Src2 = GetSrc<uint8_t*>(Data->SSAData, Op->Src2);
+  uint32_t Tmp{};
+
+  switch (Op->SrcSize) {
+    case 1:
+      Tmp = CRC32::crc32cb(Src1, *(uint8_t*)Src2);
+      break;
+    case 2:
+      Tmp = CRC32::crc32ch(Src1, *(uint16_t*)Src2);
+      break;
+    case 4:
+      Tmp = CRC32::crc32cw(Src1, *(uint32_t*)Src2);
+      break;
+    case 8:
+      Tmp = CRC32::crc32cx(Src1, *(uint64_t*)Src2);
+      break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown CRC32C size: {}", Op->SrcSize);
+      break;
+
+  }
+  memcpy(GDP, &Tmp, sizeof(Tmp));
+}
+
+DEF_OP(PCLMUL) {
+  auto Op = IROp->C<IR::IROp_PCLMUL>();
+
+  const auto Selector = Op->Selector;
+  auto* Dst  = GetDest<uint64_t*>(Data->SSAData, Node);
+  auto* Src1 = GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  auto* Src2 = GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
+
+  const uint64_t TMP1 = (Selector & 0x01) == 0 ? Src1[0] : Src1[1];
+  const uint64_t TMP2 = (Selector & 0x10) == 0 ? Src2[0] : Src2[1];
+  
+  const auto make_lo = [](uint64_t lhs, uint64_t rhs) {
+    uint64_t result = 0;
+
+    for (size_t i = 0; i < 64; i++) {
+      if ((lhs & (1ULL << i)) != 0) {
+        result ^= rhs << i;
+      }
+    }
+
+    return result;
+  };
+  const auto make_hi = [](uint64_t lhs, uint64_t rhs) {
+    uint64_t result = 0;
+
+    for (size_t i = 1; i < 64; i++) {
+      if ((lhs & (1ULL << i)) != 0) {
+        result ^= rhs >> (64 - i);
+      }
+    }
+
+    return result;
+  };
+
+  Dst[0] = make_lo(TMP1, TMP2);
+  Dst[1] = make_hi(TMP1, TMP2);
+}
+
 #undef DEF_OP

 } // namespace FEXCore::CPU
@@ -8,7 +8,7 @@ $end_info$
 #include "Interface/Core/Interpreter/InterpreterOps.h"
 #include "Interface/Core/Interpreter/InterpreterDefines.h"

-#include "F80Ops.h"
+#include "Interface/Core/Interpreter/Fallbacks/F80Fallbacks.h"

 #include <cstdint>

@@ -20,99 +20,90 @@ DEF_OP(F80LOADFCW) {

 DEF_OP(F80ADD) {
  auto Op = IROp->C<IR::IROp_F80Add>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FADD(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FADD(Src1, Src2);

  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80SUB) {
  auto Op = IROp->C<IR::IROp_F80Sub>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FSUB(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FSUB(Src1, Src2);

  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80MUL) {
  auto Op = IROp->C<IR::IROp_F80Mul>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FMUL(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FMUL(Src1, Src2);

  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80DIV) {
  auto Op = IROp->C<IR::IROp_F80Div>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FDIV(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FDIV(Src1, Src2);

  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80FYL2X) {
  auto Op = IROp->C<IR::IROp_F80FYL2X>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FYL2X(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FYL2X(Src1, Src2);

  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80ATAN) {
  auto Op = IROp->C<IR::IROp_F80ATAN>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FATAN(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FATAN(Src1, Src2);

  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80FPREM1) {
  auto Op = IROp->C<IR::IROp_F80FPREM1>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FREM1(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FREM1(Src1, Src2);

  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80FPREM) {
  auto Op = IROp->C<IR::IROp_F80FPREM>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FREM(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FREM(Src1, Src2);

  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80SCALE) {
  auto Op = IROp->C<IR::IROp_F80SCALE>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FSCALE(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FSCALE(Src1, Src2);

  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80CVT) {
  auto Op = IROp->C<IR::IROp_F80CVT>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);

  switch (OpSize) {
    case 4: {
@@ -131,9 +122,9 @@ DEF_OP(F80CVT) {

 DEF_OP(F80CVTINT) {
  auto Op = IROp->C<IR::IROp_F80CVTInt>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);

  switch (OpSize) {
    case 2: {
@@ -158,116 +149,112 @@ DEF_OP(F80CVTINT) {
 DEF_OP(F80CVTTO) {
  auto Op = IROp->C<IR::IROp_F80CVTTo>();

-  switch (Op->Size) {
+  switch (Op->SrcSize) {
    case 4: {
-      float Src = *GetSrc<float *>(Data->SSAData, Op->Header.Args[0]);
+      float Src = *GetSrc<float *>(Data->SSAData, Op->X80Src);
      X80SoftFloat Tmp = Src;
      memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
      break;
    }
    case 8: {
-      double Src = *GetSrc<double *>(Data->SSAData, Op->Header.Args[0]);
+      double Src = *GetSrc<double *>(Data->SSAData, Op->X80Src);
      X80SoftFloat Tmp = Src;
      memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
      break;
    }
-    default: LogMan::Msg::DFmt("Unhandled size: {}", Op->Size);
+    default: LogMan::Msg::DFmt("Unhandled size: {}", Op->SrcSize);
  }
 }

 DEF_OP(F80CVTTOINT) {
  auto Op = IROp->C<IR::IROp_F80CVTToInt>();

-  switch (Op->Size) {
+  switch (Op->SrcSize) {
    case 2: {
-      int16_t Src = *GetSrc<int16_t*>(Data->SSAData, Op->Header.Args[0]);
+      int16_t Src = *GetSrc<int16_t*>(Data->SSAData, Op->Src);
      X80SoftFloat Tmp = Src;
      memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
      break;
    }
    case 4: {
-      int32_t Src = *GetSrc<int32_t*>(Data->SSAData, Op->Header.Args[0]);
+      int32_t Src = *GetSrc<int32_t*>(Data->SSAData, Op->Src);
      X80SoftFloat Tmp = Src;
      memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
      break;
    }
-    default: LogMan::Msg::DFmt("Unhandled size: {}", Op->Size);
+    default: LogMan::Msg::DFmt("Unhandled size: {}", Op->SrcSize);
  }
 }

 DEF_OP(F80ROUND) {
  auto Op = IROp->C<IR::IROp_F80Round>();
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FRNDINT(Src);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+  const auto Tmp = X80SoftFloat::FRNDINT(Src);

  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80F2XM1) {
  auto Op = IROp->C<IR::IROp_F80F2XM1>();
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::F2XM1(Src);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+  const auto Tmp = X80SoftFloat::F2XM1(Src);

  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80TAN) {
  auto Op = IROp->C<IR::IROp_F80TAN>();
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FTAN(Src);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+  const auto Tmp = X80SoftFloat::FTAN(Src);

  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80SQRT) {
  auto Op = IROp->C<IR::IROp_F80SQRT>();
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FSQRT(Src);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+  const auto Tmp = X80SoftFloat::FSQRT(Src);

  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80SIN) {
  auto Op = IROp->C<IR::IROp_F80SIN>();
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FSIN(Src);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+  const auto Tmp = X80SoftFloat::FSIN(Src);
+
  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80COS) {
  auto Op = IROp->C<IR::IROp_F80COS>();
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FCOS(Src);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+  const auto Tmp = X80SoftFloat::FCOS(Src);
+
  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80XTRACT_EXP) {
  auto Op = IROp->C<IR::IROp_F80XTRACT_EXP>();
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FXTRACT_EXP(Src);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+  const auto Tmp = X80SoftFloat::FXTRACT_EXP(Src);
+
  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80XTRACT_SIG) {
  auto Op = IROp->C<IR::IROp_F80XTRACT_SIG>();
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FXTRACT_SIG(Src);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+  const auto Tmp = X80SoftFloat::FXTRACT_SIG(Src);
+
  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }

 DEF_OP(F80CMP) {
  auto Op = IROp->C<IR::IROp_F80Cmp>();
  uint32_t ResultFlags{};
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
  bool eq, lt, nan;
  X80SoftFloat::FCMP(Src1, Src2, &eq, &lt, &nan);
  if (Op->Flags & (1 << IR::FCMP_FLAG_LT) &&
@@ -288,7 +275,7 @@ DEF_OP(F80CMP) {

 DEF_OP(F80BCDLOAD) {
  auto Op = IROp->C<IR::IROp_F80BCDLoad>();
-  uint8_t *Src1 = GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[0]);
+  const uint8_t *Src1 = GetSrc<uint8_t*>(Data->SSAData, Op->X80Src);
  uint64_t BCD{};
  // We walk through each uint8_t and pull out the BCD encoding
  // Each 4bit split is a digit
@@ -323,7 +310,7 @@ DEF_OP(F80BCDLOAD) {

 DEF_OP(F80BCDSTORE) {
  auto Op = IROp->C<IR::IROp_F80BCDStore>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
+  X80SoftFloat Src1 = X80SoftFloat::FRNDINT(*GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src));
  bool Negative = Src1.Sign;

  // Clear the Sign bit
@@ -356,6 +343,80 @@ DEF_OP(F80BCDSTORE) {
  memcpy(GDP, BCD, 10);
 }

+DEF_OP(F64SIN) {
+  auto Op = IROp->C<IR::IROp_F64SIN>();
+  const double Src = *GetSrc<double*>(Data->SSAData, Op->Src);
+  const double Tmp = sin(Src);
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+DEF_OP(F64COS) {
+  auto Op = IROp->C<IR::IROp_F64COS>();
+  const double Src = *GetSrc<double*>(Data->SSAData, Op->Src);
+  const double Tmp = cos(Src);
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+DEF_OP(F64TAN) {
+  auto Op = IROp->C<IR::IROp_F64TAN>();
+  const double Src = *GetSrc<double*>(Data->SSAData, Op->Src);
+  const double Tmp = tan(Src);
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+DEF_OP(F64F2XM1) {
+  auto Op = IROp->C<IR::IROp_F64F2XM1>();
+  const double Src = *GetSrc<double*>(Data->SSAData, Op->Src);
+  const double Tmp = exp2(Src) - 1.0;
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+DEF_OP(F64ATAN) {
+  auto Op = IROp->C<IR::IROp_F64ATAN>();
+  const double Src1 = *GetSrc<double*>(Data->SSAData, Op->Src1);
+  const double Src2 = *GetSrc<double*>(Data->SSAData, Op->Src2);
+  const double Tmp = atan2(Src1, Src2);
+
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+DEF_OP(F64FPREM) {
+  auto Op = IROp->C<IR::IROp_F64FPREM>();
+  const double Src1 = *GetSrc<double*>(Data->SSAData, Op->Src1);
+  const double Src2 = *GetSrc<double*>(Data->SSAData, Op->Src2);
+  const double Tmp = fmod(Src1, Src2);
+
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+DEF_OP(F64FPREM1) {
+  auto Op = IROp->C<IR::IROp_F64FPREM1>();
+  const double Src1 = *GetSrc<double*>(Data->SSAData, Op->Src1);
+  const double Src2 = *GetSrc<double*>(Data->SSAData, Op->Src2);
+  const double Tmp = remainder(Src1, Src2);
+
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+DEF_OP(F64FYL2X) {
+  auto Op = IROp->C<IR::IROp_F64FYL2X>();
+  const double Src1 = *GetSrc<double*>(Data->SSAData, Op->Src);
+  const double Src2 = *GetSrc<double*>(Data->SSAData, Op->Src2);
+  const double Tmp = Src2 * log2(Src1);
+
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
+DEF_OP(F64SCALE) {
+  auto Op = IROp->C<IR::IROp_F64SCALE>();
+  const double Src1 = *GetSrc<double*>(Data->SSAData, Op->Src1);
+  const double Src2 = *GetSrc<double*>(Data->SSAData, Op->Src2);
+  const double trunc = (double)(int64_t)(Src2); //truncate
+  const double Tmp = Src1 * exp2(trunc);
+
+  memcpy(GDP, &Tmp, sizeof(double));
+}
+
 #undef DEF_OP

 } // namespace FEXCore::CPU
@@ -4,12 +4,9 @@

 #include <FEXCore/IR/IR.h>

+#include "Interface/Core/Interpreter/Fallbacks/FallbackOpHandler.h"
+
 namespace FEXCore::CPU {
-template<IR::IROps Op>
-struct OpHandlers {
-
-};
-
 template<>
 struct OpHandlers<IR::OP_F80CVTTO> {
  static X80SoftFloat handle4(float src) {
@@ -222,11 +219,80 @@ struct OpHandlers<IR::OP_F80SCALE> {
  }
 };

+template<>
+struct OpHandlers<IR::OP_F64SIN> {
+  static double handle(double src) {
+    return sin(src);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64COS> {
+  static double handle(double src) {
+    return cos(src);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64TAN> {
+  static double handle(double src) {
+    return tan(src);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64F2XM1> {
+  static double handle(double src) {
+    return exp2(src) - 1.0;
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64ATAN> {
+  static double handle(double src1, double src2) {
+    return atan2(src1, src2);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64FPREM> {
+  static double handle(double src1, double src2) {
+    return fmod(src1, src2);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64FPREM1> {
+  static double handle(double src1, double src2) {
+    return remainder(src1, src2);
+  }
+};
+
+
+template<>
+struct OpHandlers<IR::OP_F64FYL2X> {
+  static double handle(double src1, double src2) {
+    return src2 * log2(src1);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64SCALE> {
+  static double handle(double src1, double src2) {
+    double trunc = (double)(int64_t)(src2); //truncate
+    return src1 * exp2(trunc);
+  }
+};
+
+
+
 template<>
 struct OpHandlers<IR::OP_F80BCDSTORE> {
  static X80SoftFloat handle(X80SoftFloat Src1) {
    bool Negative = Src1.Sign;

+    Src1 = X80SoftFloat::FRNDINT(Src1);
+
    // Clear the Sign bit
    Src1.Sign = 0;

@@ -326,5 +392,4 @@ struct OpHandlers<IR::OP_F80LOADFCW> {
  }
 };

-
-}
+} // namespace FEXCore::CPU
@@ -0,0 +1,75 @@
+#pragma once
+
+#include <cstdint>
+
+namespace FEXCore::IR {
+enum IROps : uint16_t;
+}
+
+namespace FEXCore::CPU {
+
+// Base template for fallback handling.
+//
+// Registering and hooking up fallback is currently like so:
+//
+// 1. Go to InterpreterFallbacks.cpp and create a template specialization of
+//    the GetFallbackInfo member function.
+//
+//    This member function should reasonably define what the fallback you're
+//    going to create will take as parameters and return as a result. For example:
+//
+//    template<>
+//    FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(double), Core::FallbackHandlerIndex Index) {
+//      return {FABI_F80_F64, (void*)fn, Index};
+//    }
+//
+//    Defines info about a fallback that takes a double as an argument and
+//    returns a X80SoftFloat instance.
+//
+//    You will also want to define a new FallbackHandlerIndex enum member and use it
+//    to set up the new info handler into the Info array in FillFallbackIndexPointers.
+//
+// 1.1. (potentially optional). Define a new ABI element in the FallbackAPI enum.
+//      This ABI enum value will be used to tell the JITs how to handle the fallback
+//      properly. These enum values specify the return type followed by its argument types.
+//
+//      So, FABI_I64_F80_F80, for example indicates that the function will behave like a
+//      function as if were defined as:
+//
+//      uint64_t fn(X80SoftFloat, X80SoftFloat)
+//
+// 1.2. (potentially optional). If you needed to define a new enum ABI type like in 1.1, then
+//      you need to add the handling for it in the JITs, which can be found in the respective
+//      JIT's JIT.cpp file in a function called Op_Unhandled
+//
+//      You need to add a new case to the ABI switch statement using the new ABI type
+//      and do the necessary moving of data from register-allocated JIT parameters
+//      into that platform's registers that respects the calling convention. After this is
+//      done, most of the necessary background boilerplate is finished.
+//
+// 2. Now, make a specialization of this class with a member function named 'handle()'
+//    that takes the same parameters as the ones described in the fallback info function
+//    specialization.
+//
+//    For example, if you have the fallback info from the example in step 1, it would be:
+//
+//    template <>
+//    struct OpHandlers<IR::CoolNewIROpcode> {
+//      static X80SoftFloat handle(double src) {
+//        return ...;
+//      }
+//    };
+//
+// 3. Fill out the behavior of the OpHandler specialization to perform what you would like
+//    the fallback to do.
+//
+// 4. Add an implementation of the IR op to the Interpreter that passes through to the
+//    OpHandler implementation.
+//
+// 5. Done.
+//
+template <IR::IROps Op>
+struct OpHandlers {
+};
+
+} // namespace FEXCore::CPU
@@ -0,0 +1,336 @@
+#include <FEXCore/Core/CoreState.h>
+
+#include "Interface/Core/Interpreter/InterpreterOps.h"
+#include "Interface/Core/Interpreter/Fallbacks/F80Fallbacks.h"
+#include "Interface/Core/Interpreter/Fallbacks/VectorFallbacks.h"
+
+#include <cstddef>
+#include <cstdint>
+
+namespace FEXCore::CPU {
+
+template<typename R, typename... Args>
+static FallbackInfo GetFallbackInfo(R(*fn)(Args...), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_UNKNOWN, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(float), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_F32, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(double), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_F64, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(int16_t), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_I16, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(void(*fn)(uint16_t), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_VOID_U16, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(int32_t), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_I32, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(float(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F32_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(double(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F64_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(double(*fn)(double), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F64_F64, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(double(*fn)(double,double), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F64_F64_F64, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(int16_t(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_I16_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(int32_t(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_I32_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(int64_t(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_I64_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(uint64_t(*fn)(X80SoftFloat, X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_I64_F80_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(X80SoftFloat, X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_F80_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(uint32_t(*fn)(uint64_t, uint64_t, __uint128_t, __uint128_t, uint16_t), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_I32_I64_I64_I128_I128_I16, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(uint32_t(*fn)(__uint128_t, __uint128_t, uint16_t), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_I32_I128_I128_I16, (void*)fn, HandlerIndex};
+}
+
+void InterpreterOps::FillFallbackIndexPointers(uint64_t *Info) {
+  Info[Core::OPINDEX_F80LOADFCW] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80LOADFCW>::handle, Core::OPINDEX_F80LOADFCW).fn);
+  Info[Core::OPINDEX_F80CVTTO_4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTO>::handle4, Core::OPINDEX_F80CVTTO_4).fn);
+  Info[Core::OPINDEX_F80CVTTO_8] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTO>::handle8, Core::OPINDEX_F80CVTTO_8).fn);
+  Info[Core::OPINDEX_F80CVT_4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVT>::handle4, Core::OPINDEX_F80CVT_4).fn);
+  Info[Core::OPINDEX_F80CVT_8] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVT>::handle8, Core::OPINDEX_F80CVT_8).fn);
+  Info[Core::OPINDEX_F80CVTINT_2] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2, Core::OPINDEX_F80CVTINT_2).fn);
+  Info[Core::OPINDEX_F80CVTINT_4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4, Core::OPINDEX_F80CVTINT_4).fn);
+  Info[Core::OPINDEX_F80CVTINT_8] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8, Core::OPINDEX_F80CVTINT_8).fn);
+  Info[Core::OPINDEX_F80CVTINT_TRUNC2] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2t, Core::OPINDEX_F80CVTINT_TRUNC2).fn);
+  Info[Core::OPINDEX_F80CVTINT_TRUNC4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4t, Core::OPINDEX_F80CVTINT_TRUNC4).fn);
+  Info[Core::OPINDEX_F80CVTINT_TRUNC8] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8t, Core::OPINDEX_F80CVTINT_TRUNC8).fn);
+  Info[Core::OPINDEX_F80CMP_0] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<0>, Core::OPINDEX_F80CMP_0).fn);
+  Info[Core::OPINDEX_F80CMP_1] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<1>, Core::OPINDEX_F80CMP_1).fn);
+  Info[Core::OPINDEX_F80CMP_2] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<2>, Core::OPINDEX_F80CMP_2).fn);
+  Info[Core::OPINDEX_F80CMP_3] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<3>, Core::OPINDEX_F80CMP_3).fn);
+  Info[Core::OPINDEX_F80CMP_4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<4>, Core::OPINDEX_F80CMP_4).fn);
+  Info[Core::OPINDEX_F80CMP_5] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<5>, Core::OPINDEX_F80CMP_5).fn);
+  Info[Core::OPINDEX_F80CMP_6] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<6>, Core::OPINDEX_F80CMP_6).fn);
+  Info[Core::OPINDEX_F80CMP_7] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<7>, Core::OPINDEX_F80CMP_7).fn);
+  Info[Core::OPINDEX_F80CVTTOINT_2] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTOINT>::handle2, Core::OPINDEX_F80CVTTOINT_2).fn);
+  Info[Core::OPINDEX_F80CVTTOINT_4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTOINT>::handle4, Core::OPINDEX_F80CVTTOINT_4).fn);
+
+  // Unary
+  Info[Core::OPINDEX_F80ROUND] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80ROUND>::handle, Core::OPINDEX_F80ROUND).fn);
+  Info[Core::OPINDEX_F80F2XM1] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80F2XM1>::handle, Core::OPINDEX_F80F2XM1).fn);
+  Info[Core::OPINDEX_F80TAN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80TAN>::handle, Core::OPINDEX_F80TAN).fn);
+  Info[Core::OPINDEX_F80SQRT] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80SQRT>::handle, Core::OPINDEX_F80SQRT).fn);
+  Info[Core::OPINDEX_F80SIN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80SIN>::handle, Core::OPINDEX_F80SIN).fn);
+  Info[Core::OPINDEX_F80COS] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80COS>::handle, Core::OPINDEX_F80COS).fn);
+  Info[Core::OPINDEX_F80XTRACT_EXP] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80XTRACT_EXP>::handle, Core::OPINDEX_F80XTRACT_EXP).fn);
+  Info[Core::OPINDEX_F80XTRACT_SIG] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80XTRACT_SIG>::handle, Core::OPINDEX_F80XTRACT_SIG).fn);
+  Info[Core::OPINDEX_F80BCDSTORE] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80BCDSTORE>::handle, Core::OPINDEX_F80BCDSTORE).fn);
+  Info[Core::OPINDEX_F80BCDLOAD] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80BCDLOAD>::handle, Core::OPINDEX_F80BCDLOAD).fn);
+
+  // Binary
+  Info[Core::OPINDEX_F80ADD] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80ADD>::handle, Core::OPINDEX_F80ADD).fn);
+  Info[Core::OPINDEX_F80SUB] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80SUB>::handle, Core::OPINDEX_F80SUB).fn);
+  Info[Core::OPINDEX_F80MUL] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80MUL>::handle, Core::OPINDEX_F80MUL).fn);
+  Info[Core::OPINDEX_F80DIV] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80DIV>::handle, Core::OPINDEX_F80DIV).fn);
+  Info[Core::OPINDEX_F80FYL2X] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80FYL2X>::handle, Core::OPINDEX_F80FYL2X).fn);
+  Info[Core::OPINDEX_F80ATAN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80ATAN>::handle, Core::OPINDEX_F80ATAN).fn);
+  Info[Core::OPINDEX_F80FPREM1] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80FPREM1>::handle, Core::OPINDEX_F80FPREM1).fn);
+  Info[Core::OPINDEX_F80FPREM] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80FPREM>::handle, Core::OPINDEX_F80FPREM).fn);
+  Info[Core::OPINDEX_F80SCALE] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80SCALE>::handle, Core::OPINDEX_F80SCALE).fn);
+
+  // Double Precision
+  Info[Core::OPINDEX_F64SIN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64SIN>::handle, Core::OPINDEX_F64SIN).fn);
+  Info[Core::OPINDEX_F64COS] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64COS>::handle, Core::OPINDEX_F64COS).fn);
+  Info[Core::OPINDEX_F64TAN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64TAN>::handle, Core::OPINDEX_F64TAN).fn);
+  Info[Core::OPINDEX_F64ATAN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64ATAN>::handle, Core::OPINDEX_F64ATAN).fn);
+  Info[Core::OPINDEX_F64F2XM1] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64F2XM1>::handle, Core::OPINDEX_F64F2XM1).fn);
+  Info[Core::OPINDEX_F64FYL2X] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64FYL2X>::handle, Core::OPINDEX_F64FYL2X).fn);
+  Info[Core::OPINDEX_F64FPREM] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64FPREM>::handle, Core::OPINDEX_F64FPREM).fn);
+  Info[Core::OPINDEX_F64FPREM1] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64FPREM1>::handle, Core::OPINDEX_F64FPREM1).fn);
+  Info[Core::OPINDEX_F64SCALE] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64SCALE>::handle, Core::OPINDEX_F64SCALE).fn);
+
+  // SSE4.2 string instructions
+  Info[Core::OPINDEX_VPCMPESTRX] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_VPCMPESTRX>::handle, Core::OPINDEX_VPCMPESTRX).fn);
+  Info[Core::OPINDEX_VPCMPISTRX] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_VPCMPISTRX>::handle, Core::OPINDEX_VPCMPISTRX).fn);
+}
+
+bool InterpreterOps::GetFallbackHandler(IR::IROp_Header const *IROp, FallbackInfo *Info) {
+  uint8_t OpSize = IROp->Size;
+  switch(IROp->Op) {
+    case IR::OP_F80LOADFCW: {
+      *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80LOADFCW>::handle, Core::OPINDEX_F80LOADFCW);
+      return true;
+    }
+
+    case IR::OP_F80CVTTO: {
+      auto Op = IROp->C<IR::IROp_F80CVTTo>();
+
+      switch (Op->SrcSize) {
+        case 4: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTO>::handle4, Core::OPINDEX_F80CVTTO_4);
+          return true;
+        }
+        case 8: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTO>::handle8, Core::OPINDEX_F80CVTTO_8);
+          return true;
+        }
+      default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
+      }
+      break;
+    }
+    case IR::OP_F80CVT: {
+      switch (OpSize) {
+        case 4: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVT>::handle4, Core::OPINDEX_F80CVT_4);
+          return true;
+        }
+        case 8: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVT>::handle8, Core::OPINDEX_F80CVT_8);
+          return true;
+        }
+        default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
+      }
+      break;
+    }
+    case IR::OP_F80CVTINT: {
+      auto Op = IROp->C<IR::IROp_F80CVTInt>();
+
+      switch (OpSize) {
+        case 2: {
+          if (Op->Truncate) {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2t, Core::OPINDEX_F80CVTINT_TRUNC2);
+          }
+          else {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2, Core::OPINDEX_F80CVTINT_2);
+          }
+          return true;
+        }
+        case 4: {
+          if (Op->Truncate) {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4t, Core::OPINDEX_F80CVTINT_TRUNC4);
+          }
+          else {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4, Core::OPINDEX_F80CVTINT_4);
+          }
+          return true;
+        }
+        case 8: {
+          if (Op->Truncate) {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8t, Core::OPINDEX_F80CVTINT_TRUNC8);
+          }
+          else {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8, Core::OPINDEX_F80CVTINT_8);
+          }
+          return true;
+        }
+        default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
+      }
+      break;
+    }
+    case IR::OP_F80CMP: {
+      auto Op = IROp->C<IR::IROp_F80Cmp>();
+
+      static constexpr std::array handlers{
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<0>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<1>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<2>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<3>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<4>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<5>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<6>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<7>,
+      };
+
+      *Info = GetFallbackInfo(handlers[Op->Flags], (Core::FallbackHandlerIndex)(Core::OPINDEX_F80CMP_0 + Op->Flags));
+      return true;
+    }
+
+    case IR::OP_F80CVTTOINT: {
+      auto Op = IROp->C<IR::IROp_F80CVTToInt>();
+
+      switch (Op->SrcSize) {
+        case 2: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTOINT>::handle2, Core::OPINDEX_F80CVTTOINT_2);
+          return true;
+        }
+        case 4: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTOINT>::handle4, Core::OPINDEX_F80CVTTOINT_4);
+          return true;
+        }
+        default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
+      }
+      break;
+    }
+
+#define COMMON_X87_OP(OP) \
+    case IR::OP_F80##OP: { \
+      *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80##OP>::handle, Core::OPINDEX_F80##OP); \
+      return true; \
+    }
+
+#define COMMON_F64_OP(OP) \
+    case IR::OP_F64##OP: { \
+      *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64##OP>::handle, Core::OPINDEX_F64##OP); \
+      return true; \
+    }
+
+    // Unary
+    COMMON_X87_OP(ROUND)
+    COMMON_X87_OP(F2XM1)
+    COMMON_X87_OP(TAN)
+    COMMON_X87_OP(SQRT)
+    COMMON_X87_OP(SIN)
+    COMMON_X87_OP(COS)
+    COMMON_X87_OP(XTRACT_EXP)
+    COMMON_X87_OP(XTRACT_SIG)
+    COMMON_X87_OP(BCDSTORE)
+    COMMON_X87_OP(BCDLOAD)
+
+    // Binary
+    COMMON_X87_OP(ADD)
+    COMMON_X87_OP(SUB)
+    COMMON_X87_OP(MUL)
+    COMMON_X87_OP(DIV)
+    COMMON_X87_OP(FYL2X)
+    COMMON_X87_OP(ATAN)
+    COMMON_X87_OP(FPREM1)
+    COMMON_X87_OP(FPREM)
+    COMMON_X87_OP(SCALE)
+
+    // Double Precision Unary
+    COMMON_F64_OP(F2XM1)
+    COMMON_F64_OP(TAN)
+    COMMON_F64_OP(SIN)
+    COMMON_F64_OP(COS)
+
+    // Double Precision Binary
+    COMMON_F64_OP(FYL2X)
+    COMMON_F64_OP(ATAN)
+    COMMON_F64_OP(FPREM1)
+    COMMON_F64_OP(FPREM)
+    COMMON_F64_OP(SCALE)
+
+    // SSE4.2 Fallbacks
+    case IR::OP_VPCMPESTRX:
+      *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_VPCMPESTRX>::handle, Core::OPINDEX_VPCMPESTRX);
+      return true;
+    case IR::OP_VPCMPISTRX:
+      *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_VPCMPISTRX>::handle, Core::OPINDEX_VPCMPISTRX);
+      return true;
+
+    default:
+      break;
+  }
+
+  return false;
+}
+
+
+}
@@ -0,0 +1,408 @@
+#pragma once
+
+#include <algorithm>
+#include <cstddef>
+#include <cstdint>
+#include <cstdlib>
+#include <cstring>
+
+#include <FEXCore/IR/IR.h>
+
+#include "Interface/Core/Interpreter/Fallbacks/FallbackOpHandler.h"
+
+namespace FEXCore::CPU {
+
+template<>
+struct OpHandlers<IR::OP_VPCMPESTRX> {
+  enum class AggregationOp {
+    EqualAny     = 0b00,
+    Ranges       = 0b01,
+    EqualEach    = 0b10,
+    EqualOrdered = 0b11,
+  };
+
+  enum class SourceData {
+    U8,
+    U16,
+    S8,
+    S16,
+  };
+
+  enum class Polarity {
+    Positive,
+    Negative,
+    PositiveMasked,
+    NegativeMasked,
+  };
+
+  static uint32_t handle(uint64_t RAX, uint64_t RDX, __uint128_t lhs, __uint128_t rhs, uint16_t control) {
+    // Subtract by 1 in order to make validity limits 0-based
+    const auto valid_lhs = GetExplicitLength(RAX, control) - 1;
+    const auto valid_rhs = GetExplicitLength(RDX, control) - 1;
+
+    return MainBody(lhs, valid_lhs, rhs, valid_rhs, control);
+  }
+
+  // Main PCMPXSTRX algorithm body. Allows for reuse with both implicit and explicit length variants.
+  static uint32_t MainBody(const __uint128_t& lhs, int valid_lhs, const __uint128_t& rhs, int valid_rhs, uint16_t control) {
+    const uint32_t aggregation = PerformAggregation(lhs, valid_lhs, rhs, valid_rhs, control);
+    const int32_t upper_limit = (16 >> (control & 1)) - 1;
+
+    // Bits are arranged as:
+    // Bit #:   3    2    1    0
+    //         [OF | CF | SF | ZF]
+    uint32_t flags = 0;
+    flags |= (valid_rhs < upper_limit) ? 0b01 : 0b00;
+    flags |= (valid_lhs < upper_limit) ? 0b10 : 0b00;
+
+    const uint32_t result = HandlePolarity(aggregation, control, upper_limit, valid_rhs);
+    if (result != 0) {
+      flags |= 0b0100;
+    }
+    if ((result & 1) != 0) {
+      flags |= 0b1000;
+    }
+
+    // We tack the flags on top of the result to avoid needing to handle
+    // multiple return values in the JITs.
+    return result | (flags << 16);
+  }
+
+  static int32_t GetExplicitLength(uint64_t reg, uint16_t control) {
+    // Bit 8 controls whether or not the reg value is 64-bit or 32-bit.
+    int64_t value = 0;
+    if (((control >> 8) & 1) != 0) {
+      value = static_cast<int64_t>(reg);
+    } else {
+      // We need a sign extend in this case.
+      value = static_cast<int32_t>(reg);
+    }
+
+    // If control[0] is set, then we're dealing with words instead of bytes
+    const int64_t limit = (control & 1) != 0 ? 8 : 16;
+
+    // Length needs to saturate to 16 (if bytes) or 8 (if words)
+    // when the length value is greater than 16 (if bytes)/8 (if words)
+    // or if the length value is less than -16 (if bytes)/-8 (if words).
+    if (value < -limit || value > limit) {
+      return limit;
+    }
+
+    return std::abs(static_cast<int>(value));
+  }
+
+  static int32_t GetElement(const __uint128_t& vec, int32_t index, uint16_t control) {
+    const auto* vec_ptr = reinterpret_cast<const uint8_t*>(&vec);
+
+    // Control bits [1:0] define the data type being dealt with.
+    switch (static_cast<SourceData>(control & 0b11)) {
+    case SourceData::U8:
+      return static_cast<int32_t>(vec_ptr[index]);
+    case SourceData::U16: {
+      uint16_t value{};
+      std::memcpy(&value, vec_ptr + (sizeof(uint16_t) * static_cast<size_t>(index)), sizeof(value));
+      return value;
+    }
+    case SourceData::S8:
+      return static_cast<int8_t>(vec_ptr[index]);
+    case SourceData::S16:
+    default: {
+      int16_t value{};
+      std::memcpy(&value, vec_ptr + (sizeof(int16_t) * static_cast<size_t>(index)), sizeof(value));
+      return value;
+    }
+    }
+  }
+
+  static uint32_t PerformAggregation(const __uint128_t& lhs, int32_t valid_lhs,
+                                     const __uint128_t& rhs, int32_t valid_rhs,
+                                     uint16_t control) {
+    switch (static_cast<AggregationOp>((control >> 2) & 0b11)) {
+    case AggregationOp::EqualAny:
+      return HandleEqualAny(lhs, valid_lhs, rhs, valid_rhs, control);
+    case AggregationOp::Ranges:
+      return HandleRanges(lhs, valid_lhs, rhs, valid_rhs, control);
+    case AggregationOp::EqualEach:
+      return HandleEqualEach(lhs, valid_lhs, rhs, valid_rhs, control);
+    case AggregationOp::EqualOrdered:
+    default:
+      return HandleEqualOrdered(lhs, valid_lhs, rhs, valid_rhs, control);
+    }
+  }
+
+  static uint32_t HandlePolarity(uint32_t value, uint16_t control, int upper_limit, int valid_rhs) {
+    switch (static_cast<Polarity>((control >> 4) & 0b11)) {
+      case Polarity::Negative:
+        return value ^ ((2U << upper_limit) - 1);
+      case Polarity::NegativeMasked:
+        return value ^ ((1U << (valid_rhs + 1)) - 1);
+      case Polarity::Positive:
+      case Polarity::PositiveMasked:
+      default:
+        // Both positive masking and positive polarity are documented
+        // as both being equivalent to "IntRes2 = IntRes1", where IntRes1
+        // is our 'value' parameter, so we don't need to do anything in
+        // these cases except return the same value.
+        return value;
+    }
+  }
+
+  // Finds characters from an overall character set.
+  //
+  // Scans through RHS trying to find any characters contained in LHS.
+  // Think of this as a sort of vectorized version of strspn (kind of).
+  //
+  // e.g. Assume operating on two character vectors as unsigned words
+  //
+  //         0  1  2  3  4  5  6  7
+  // LHS -> [a, b, c, d, e, f, g, n]
+  // RHS -> [z, k, v, c, d, o, p, n]
+  //
+  // With both explicit lengths for each string being 8 (the max length for words),
+  // this would result in an intermediate result like:
+  //
+  //            0b1001'1000
+  //              │  │ │
+  // 'n' match ───┘  │ │
+  //                 │ │
+  // 'd' match ──────┘ │
+  //                   │
+  // 'c' match ────────┘
+  //
+  static uint32_t HandleEqualAny(const __uint128_t& lhs, int32_t valid_lhs,
+                                 const __uint128_t& rhs, int32_t valid_rhs,
+                                 uint16_t control) {
+    uint32_t result = 0;
+
+    for (int j = valid_rhs; j >= 0; j--) {
+      result <<= 1;
+
+      const int rhs_value = GetElement(rhs, j, control);
+      for (int i = valid_lhs; i >= 0; i--) {
+        const int lhs_value = GetElement(lhs, i, control);
+        result |= static_cast<uint32_t>(rhs_value == lhs_value);
+      }
+    }
+
+    return result;
+  }
+
+  // Determines if a character falls within a limited range
+  //
+  // Scans through rhs using a range denoted by two elements
+  // in lhs and determines if the respective character in rhs
+  // falls within its range.
+  //
+  // i.e.
+  //      lhs_upper_bound >= rhs_value && lhs_lower_bound <= rhs_value
+  //
+  // e.g. Assume operating on two character vectors as unsigned words
+  //
+  //         0  1  2  3  4  5  6  7
+  // LHS -> [a, z, A, Z, 0, 0, 0, 0]
+  // RHS -> [z, k, ., C, M, ;, \, ']
+  //
+  // With LHS's length being 4 and RHS's lenth being 8,
+  // this would result in an intermediate result like:
+  //
+  //                          0b0001'1011
+  //                               │ │ ││
+  // 'z' >= 'M' && 'a' <= 'M' ─────┘ │ ││
+  //                                 │ ││
+  // 'z' >= 'C' && 'a' <= 'C' ───────┘ ││
+  //                                   ││
+  // 'Z' >= 'k' && 'A' <= 'k' ─────────┘│
+  //                                    │
+  // 'Z' >= 'z' && 'A' <= 'z' ──────────┘
+  //
+  static uint32_t HandleRanges(const __uint128_t& lhs, int32_t valid_lhs,
+                               const __uint128_t& rhs, int32_t valid_rhs,
+                               uint16_t control) {
+    uint32_t result = 0;
+
+    for (int j = valid_rhs; j >= 0; j--) {
+      result <<= 1;
+
+      const int element = GetElement(rhs, j, control);
+      for (int i = (valid_lhs - 1) | 1; i >= 0; i -= 2) {
+        const int upper_bound = GetElement(lhs, i - 0, control);
+        const int lower_bound = GetElement(lhs, i - 1, control);
+
+        const bool ge = upper_bound >= element;
+        const bool le = lower_bound <= element;
+
+        result |= static_cast<uint32_t>(ge && le);
+      }
+    }
+
+    return result;
+  }
+
+  // Determines if each character is equal to one another (string compare)
+  //
+  // Essentially the PCMPXSTRX variant of memcmp/strcmp. Sets the bit of the
+  // resulting mask if both elements are equal to one another. Otherwise
+  // sets it to false.
+  //
+  // e.g. Assume operating on two character vectors as unsigned words
+  //
+  //         0  1  2  3  4  5  6  7
+  // LHS -> [a, b, c, d, e, f, g, n]
+  // RHS -> [a, b, c, d, e, f, e, x]
+  //
+  // With both explicit lengths for each string being 8 (the max length for words),
+  // this would result in an intermediate result like:
+  //
+  //            0b0011'1111
+  //                ││ ││││
+  // 'f' == 'f' ────┘│ ││││
+  //                 │ ││││
+  // 'e' == 'e' ─────┘ ││││
+  //                   ││││
+  // 'd' == 'd' ───────┘│││
+  //                    │││
+  // 'c' == 'c' ────────┘││
+  //                     ││
+  // 'b' == 'b' ─────────┘│
+  //                      │
+  // 'a' == 'a' ──────────┘
+  //
+  static uint32_t HandleEqualEach(const __uint128_t& lhs, int32_t valid_lhs,
+                                  const __uint128_t& rhs, int32_t valid_rhs,
+                                  uint16_t control) {
+    const auto upper_limit = (16 >> (control & 1)) - 1;
+    const auto max_valid = std::max(valid_lhs, valid_rhs);
+    const auto min_valid = std::min(valid_lhs, valid_rhs);
+
+    // All values past the end of string must be forced to true.
+    // (See 4.1.6 Valid/Invalid Override of Comparisons in the Intel Software Development Manual)
+    // So we can calculate this part of the mask ahead of time and set all those to-be bits to true
+    // and then progressively shift them into place over the course of execution.
+    uint32_t result = (1U << (upper_limit - max_valid)) - 1;
+    result <<= (max_valid - min_valid);
+
+    for (int i = min_valid; i >= 0; i--) {
+      const int lhs_element = GetElement(lhs, i, control);
+      const int rhs_element = GetElement(rhs, i, control);
+
+      result <<= 1;
+      result |= static_cast<uint32_t>(lhs_element == rhs_element);
+    }
+
+    return result;
+  }
+
+  // Determines if a substring exists within an overall string
+  //
+  // Somewhat equivalent to the behavior of strstr.
+  //
+  // Sets the corresponding index in the result where a substring is found.
+  //
+  // e.g. Assume operating on two character vectors as unsigned words
+  //
+  //         0  1  2  3  4  5  6  7
+  // LHS -> [b, a, x, z, y, v, o, m]
+  // RHS -> [b, a, d, b, a, n, k, s]
+  //
+  // With the length of LHS being 2 and the length of RHS being 8, we have a composition like:
+  //
+  //      Substring to look for
+  //       ┌──┴──┐
+  // LHS -> [b, a, x, z, y, v, o, m]
+  // RHS -> [b, a, d, b, a, n, k, s]
+  //       └───────────┬────────────┘
+  //         Entire string to search
+  //
+  // And we end up with a result like:
+  //
+  //            0b0000'1001
+  //                   │  │
+  // At index 3 ───────┘  │
+  //                      │
+  // At index 0 ──────────┘
+  //
+  static uint32_t HandleEqualOrdered(const __uint128_t& lhs, int32_t valid_lhs,
+                                     const __uint128_t& rhs, int32_t valid_rhs,
+                                     uint16_t control) {
+    const auto upper_limit = (16 >> (control & 1)) - 1;
+
+    // Edge case!
+    // If we have *no* valid characters in our inner string, then
+    // we need to return the intermediate result as
+    // 0xFF (if operating on words) or 0xFFFF (if operating on bytes)
+    if (valid_lhs == -1) {
+      return (2U << upper_limit) - 1;
+    }
+
+    uint32_t result = 0;
+    const int initial = valid_rhs == upper_limit ? valid_rhs
+                                                 : valid_rhs - valid_lhs;
+    for (int j = initial; j >= 0; j--) {
+      result <<= 1;
+
+      uint32_t value = 1;
+      const int start = std::min(valid_rhs - j, valid_lhs);
+      for (int i = start; i >= 0; i--) {
+        const int lhs_value = GetElement(lhs, i + 0, control);
+        const int rhs_value = GetElement(rhs, i + j, control);
+
+        value &= static_cast<uint32_t>(lhs_value == rhs_value);
+      }
+
+      result |= value;
+    }
+
+    return result;
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_VPCMPISTRX> {
+  // Essentially the same in terms of behavior with VPCMPESTRX instructions,
+  // with the only difference being that the length of the string is encoded
+  // as part of the data vectors passed in.
+  //
+  // i.e. Length is determined by the presence of a NUL (all-zero) character
+  //      within the data.
+  //
+  //      If no NUL character exists, then the length of the strings are assumed
+  //      to be the max length possible for the given character size specified
+  //      in the control flags (16 characters for 8-bit, and 8 characters for 16-bit).
+  //
+  static uint32_t handle(__uint128_t lhs, __uint128_t rhs, uint16_t control) {
+    // Subtract by 1 in order to make validity limits 0-based
+    const auto valid_lhs = GetImplicitLength(lhs, control) - 1;
+    const auto valid_rhs = GetImplicitLength(rhs, control) - 1;
+
+    return OpHandlers<IR::OP_VPCMPESTRX>::MainBody(lhs, valid_lhs, rhs, valid_rhs, control);
+  }
+
+  static int32_t GetImplicitLength(const __uint128_t& data, uint16_t control) {
+    const auto* data_u8 = reinterpret_cast<const uint8_t*>(&data);
+    const auto is_using_words = (control & 1) != 0;
+
+    int32_t length = 0;
+
+    if (is_using_words) {
+      const auto get_word = [data_u8](int32_t index) {
+        const auto* src = data_u8 + (index * sizeof(uint16_t));
+
+        uint16_t element{};
+        std::memcpy(&element, src, sizeof(uint16_t));
+        return element;
+      };
+
+      while (length < 8 && get_word(length) != 0) {
+        length++;
+      }
+    } else {
+      while (length < 16 && data_u8[length] != 0) {
+        length++;
+      }
+    }
+
+    return length;
+  }
+};
+
+} // namespace FEXCore::CPU
@@ -14,7 +14,7 @@ namespace FEXCore::CPU {
 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
 DEF_OP(GetHostFlag) {
  auto Op = IROp->C<IR::IROp_GetHostFlag>();
-  GD = (*GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]) >> Op->Flag) & 1;
+  GD = (*GetSrc<uint64_t*>(Data->SSAData, Op->Value) >> Op->Flag) & 1;
 }
 #undef DEF_OP

@@ -6,42 +6,39 @@
 #include <FEXCore/Core/CPUBackend.h>
 #include <FEXCore/IR/IR.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
+#include <FEXCore/fextl/string.h>
+#include <FEXCore/fextl/vector.h>

 namespace FEXCore::CPU {
+class Dispatcher;
 class X86DispatchGenerator;
 class Arm64DispatchGenerator;

-#define DESTMAP_AS_MAP 0
-#if DESTMAP_AS_MAP
-using DestMapType = std::unordered_map<uint32_t, uint32_t>;
-#else
-using DestMapType = std::vector<uint32_t>;
-#endif
+using DestMapType = fextl::vector<uint32_t>;

 class InterpreterCore final : public CPUBackend {
 public:
-  explicit InterpreterCore(FEXCore::Context::Context *ctx,
-                           FEXCore::Core::InternalThreadState *Thread,
-                           bool CompileThread);
+  explicit InterpreterCore(Dispatcher *Dispatch,
+                           FEXCore::Core::InternalThreadState *Thread);

-  [[nodiscard]] std::string GetName() override { return "Interpreter"; }
+  [[nodiscard]] fextl::string GetName() override { return "Interpreter"; }

-  [[nodiscard]] void *CompileCode(uint64_t Entry,
+  [[nodiscard]] CPUBackend::CompiledCode CompileCode(uint64_t Entry,
                                  FEXCore::IR::IRListView const *IR,
                                  FEXCore::Core::DebugData *DebugData,
-                                  FEXCore::IR::RegisterAllocationData *RAData) override;
+                                  FEXCore::IR::RegisterAllocationData *RAData, bool GDBEnabled) override;

  [[nodiscard]] void *MapRegion(void* HostPtr, uint64_t, uint64_t) override { return HostPtr; }

  [[nodiscard]] bool NeedsOpDispatch() override { return true; }

-  void CreateAsmDispatch(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread);
+  static void InitializeSignalHandlers(FEXCore::Context::ContextImpl *CTX);
+
+  void ClearCache() override;

 private:
-  FEXCore::Context::Context *CTX;
-  FEXCore::Core::InternalThreadState *State;
-
-  std::unique_ptr<Dispatcher> Dispatcher{};
+  size_t BufferUsed;
+  Dispatcher *Dispatch;
 };

 template<typename T>
@@ -1,75 +1,98 @@
 #include "Interface/Context/Context.h"

-#include "Interface/Core/ArchHelpers/Arm64.h"
-#include "Interface/Core/ArchHelpers/MContext.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Core/Interpreter/InterpreterClass.h"
 #include <FEXCore/Config/Config.h>
 #include <FEXCore/Core/CoreState.h>
-#include <FEXCore/Core/SignalDelegator.h>
 #include <FEXCore/Debug/InternalThreadState.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/MathUtils.h>
+#include <FEXCore/fextl/memory.h>

-#include <memory>
-#include <bits/types/stack_t.h>
 #include <signal.h>
 #include <stdint.h>
-#include <unordered_map>
 #include <utility>

 #include "InterpreterOps.h"

+#if defined(_M_X86_64)
+  #include "Interface/Core/Dispatcher/X86Dispatcher.h"
+#elif defined(_M_ARM_64)
+  #include "Interface/Core/Dispatcher/Arm64Dispatcher.h"
+#else
+  #error missing arch
+#endif
+
+static constexpr size_t INITIAL_CODE_SIZE = 1024 * 1024 * 16;
+static constexpr size_t MAX_CODE_SIZE = 1024 * 1024 * 128;
+
 namespace FEXCore::IR {
  class IRListView;
  class RegisterAllocationData;
 }

+
 namespace FEXCore::CPU {
-class CPUBackend;

-static void InterpreterExecution(FEXCore::Core::CpuStateFrame *Frame) {
-  auto Thread = Frame->Thread;
+InterpreterCore::InterpreterCore(Dispatcher *Dispatcher, FEXCore::Core::InternalThreadState *Thread)
+  : CPUBackend(Thread, INITIAL_CODE_SIZE, MAX_CODE_SIZE)
+  , Dispatch(Dispatcher)
+  {

-  auto LocalEntry = Thread->LocalIRCache.find(Thread->CurrentFrame->State.rip);
+  auto &Interpreter = Thread->CurrentFrame->Pointers.Interpreter;

-  InterpreterOps::InterpretIR(Thread, Thread->CurrentFrame->State.rip, LocalEntry->second.IR.get(), LocalEntry->second.DebugData.get());
+  Interpreter.FragmentExecuter = reinterpret_cast<uint64_t>(&InterpreterOps::InterpretIR);
+
+  ClearCache();
 }

-InterpreterCore::InterpreterCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, bool CompileThread)
-  : CTX {ctx}
-  , State {Thread} {
+CPUBackend::CompiledCode InterpreterCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRListView const *IR, [[maybe_unused]] FEXCore::Core::DebugData *DebugData, FEXCore::IR::RegisterAllocationData *RAData, bool GDBEnabled) {

-  if (!CompileThread &&
-      CTX->Config.Core == FEXCore::Config::CONFIG_INTERPRETER) {
-    CreateAsmDispatch(ctx, Thread);
-    CTX->SignalDelegation->RegisterHostSignalHandler(SignalDelegator::SIGNAL_FOR_PAUSE, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
-      InterpreterCore *Core = reinterpret_cast<InterpreterCore*>(Thread->CPUBackend.get());
-      return Core->Dispatcher->HandleSignalPause(Signal, info, ucontext);
-    }, true);
+  const auto IRSize = AlignUp(IR->GetInlineSize(), 16);
+  const auto MaxSize = IRSize + Dispatcher::MaxInterpreterTrampolineSize + GDBEnabled * Dispatcher::MaxGDBPauseCheckSize;

-#ifdef _M_ARM_64
-    CTX->SignalDelegation->RegisterHostSignalHandler(SIGBUS, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
-      return FEXCore::ArchHelpers::Arm64::HandleSIGBUS(true, Signal, info, ucontext);
-    }, true);
-#endif
-
-    auto GuestSignalHandler = [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack) -> bool {
-      InterpreterCore *Core = reinterpret_cast<InterpreterCore*>(Thread->CPUBackend.get());
-      return Core->Dispatcher->HandleGuestSignal(Signal, info, ucontext, GuestAction, GuestStack);
-    };
-
-    for (uint32_t Signal = 0; Signal <= SignalDelegator::MAX_SIGNALS; ++Signal) {
-      CTX->SignalDelegation->RegisterHostSignalHandlerForGuest(Signal, GuestSignalHandler);
-    }
+  if ((BufferUsed + MaxSize) > CurrentCodeBuffer->Size) {
+    static_cast<Context::ContextImpl*>(ThreadState->CTX)->ClearCodeCache(ThreadState);
  }
+
+  CPUBackend::CompiledCode CodeData{};
+
+  const auto BufferStartOffset = BufferUsed;
+  CodeData.BlockBegin = CodeData.BlockEntry = CurrentCodeBuffer->Ptr + BufferStartOffset;
+
+  auto DestBuffer = CodeData.BlockBegin;
+
+  if (GDBEnabled) {
+    const auto GDBSize = Dispatch->GenerateGDBPauseCheck(DestBuffer, Entry);
+    DestBuffer += GDBSize;
+    BufferUsed += GDBSize;
+  }
+
+  const auto TrampolineSize = Dispatch->GenerateInterpreterTrampoline(DestBuffer);
+  DestBuffer += TrampolineSize;
+  BufferUsed += TrampolineSize;
+
+
+  IR->Serialize(DestBuffer);
+  DestBuffer += IRSize;
+  BufferUsed += IRSize;
+
+  CodeData.Size = BufferUsed - BufferStartOffset;
+
+  return CodeData;
 }

-void *InterpreterCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRListView const *IR, [[maybe_unused]] FEXCore::Core::DebugData *DebugData, FEXCore::IR::RegisterAllocationData *RAData) {
-  return reinterpret_cast<void*>(InterpreterExecution);
+void InterpreterCore::ClearCache() {
+  // Calling this one is needed to setup the initial CurrentCodeBuffer
+  [[maybe_unused]] auto CodeBuffer = GetEmptyCodeBuffer();
+  BufferUsed = 0;
 }

-std::unique_ptr<CPUBackend> CreateInterpreterCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, bool CompileThread) {
-  return std::make_unique<InterpreterCore>(ctx, Thread, CompileThread);
+fextl::unique_ptr<CPUBackend> CreateInterpreterCore(FEXCore::Context::ContextImpl *ctx, FEXCore::Core::InternalThreadState *Thread) {
+  return fextl::make_unique<InterpreterCore>(ctx->Dispatcher.get(), Thread);
 }

+CPUBackendFeatures GetInterpreterBackendFeatures() {
+  return CPUBackendFeatures { };
+}
 }
@@ -1,9 +1,10 @@
 #pragma once

-#include <memory>
+#include <FEXCore/Core/CPUBackend.h>
+#include <FEXCore/fextl/memory.h>

 namespace FEXCore::Context {
-struct Context;
+class ContextImpl;
 }

 namespace FEXCore::Core {
@@ -12,9 +13,11 @@ namespace FEXCore::Core {

 namespace FEXCore::CPU {
 class CPUBackend;
+struct DispatcherConfig;

-[[nodiscard]] std::unique_ptr<CPUBackend> CreateInterpreterCore(FEXCore::Context::Context *ctx,
-                                                                FEXCore::Core::InternalThreadState *Thread,
-                                                                bool CompileThread);
+[[nodiscard]] fextl::unique_ptr<CPUBackend> CreateInterpreterCore(FEXCore::Context::ContextImpl *ctx,
+                                                                FEXCore::Core::InternalThreadState *Thread);
+void InitializeInterpreterSignalHandlers(FEXCore::Context::ContextImpl *CTX);
+CPUBackendFeatures GetInterpreterBackendFeatures();

 } // namespace FEXCore::CPU
@@ -159,21 +159,26 @@
  break;                                            \
  }

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

 template<typename Res>
 Res GetDest(void* SSAData, FEXCore::IR::NodeID Op) {
-  auto DstPtr = &reinterpret_cast<__uint128_t*>(SSAData)[Op.Value];
+  auto DstPtr = &reinterpret_cast<InterpVector256*>(SSAData)[Op.Value];
  return reinterpret_cast<Res>(DstPtr);
 }


 template<typename Res>
 Res GetSrc(void* SSAData, FEXCore::IR::OrderedNodeWrapper Src) {
-  auto DstPtr = &reinterpret_cast<__uint128_t*>(SSAData)[Src.ID().Value];
+  auto DstPtr = &reinterpret_cast<InterpVector256*>(SSAData)[Src.ID().Value];
  return reinterpret_cast<Res>(DstPtr);
 }
@@ -1,11 +1,7 @@
 #include "Interface/Context/Context.h"
 #include "Interface/Core/CPUID.h"
+#include "InterpreterDefines.h"
 #include "InterpreterOps.h"
-#include "F80Ops.h"
-
-#ifdef _M_ARM_64
-#include "Interface/Core/ArchHelpers/Arm64.h"
-#endif

 #include <FEXCore/Core/CPUBackend.h>
 #include <FEXCore/Core/CoreState.h>
@@ -56,6 +52,7 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
  REGISTER_OP(ADD,                    Add);
  REGISTER_OP(SUB,                    Sub);
  REGISTER_OP(NEG,                    Neg);
+  REGISTER_OP(ABS,                    Abs);
  REGISTER_OP(MUL,                    Mul);
  REGISTER_OP(UMUL,                   UMul);
  REGISTER_OP(DIV,                    Div);
@@ -112,10 +109,6 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
  REGISTER_OP(ATOMICFETCHNEG,         AtomicFetchNeg);

  // Branch ops
-  REGISTER_OP(GUESTCALLDIRECT,        GuestCallDirect);
-  REGISTER_OP(GUESTCALLINDIRECT,      GuestCallIndirect);
-  REGISTER_OP(GUESTRETURN,            GuestReturn);
-  REGISTER_OP(SIGNALRETURN,           SignalReturn);
  REGISTER_OP(CALLBACKRETURN,         CallbackReturn);
  REGISTER_OP(EXITFUNCTION,           ExitFunction);
  REGISTER_OP(JUMP,                   Jump);
@@ -124,12 +117,14 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
  REGISTER_OP(INLINESYSCALL,          InlineSyscall);
  REGISTER_OP(THUNK,                  Thunk);
  REGISTER_OP(VALIDATECODE,           ValidateCode);
-  REGISTER_OP(REMOVECODEENTRY,        RemoveCodeEntry);
+  REGISTER_OP(THREADREMOVECODEENTRY,        ThreadRemoveCodeEntry);
  REGISTER_OP(CPUID,                  CPUID);
+  REGISTER_OP(XGETBV,                 XGETBV);

  // Conversion ops
  REGISTER_OP(VINSGPR,                VInsGPR);
  REGISTER_OP(VCASTFROMGPR,           VCastFromGPR);
+  REGISTER_OP(VDUPFROMGPR,            VDupFromGPR);
  REGISTER_OP(FLOAT_FROMGPR_S,        Float_FromGPR_S);
  REGISTER_OP(FLOAT_FTOF,             Float_FToF);
  REGISTER_OP(VECTOR_STOF,            Vector_SToF);
@@ -156,9 +151,12 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
  REGISTER_OP(STOREMEM,               StoreMem);
  REGISTER_OP(LOADMEMTSO,             LoadMem);
  REGISTER_OP(STOREMEMTSO,            StoreMem);
-  REGISTER_OP(VLOADMEMELEMENT,        VLoadMemElement);
-  REGISTER_OP(VSTOREMEMELEMENT,       VStoreMemElement);
+  REGISTER_OP(VLOADVECTORMASKED,      VLoadVectorMasked);
+  REGISTER_OP(VSTOREVECTORMASKED,     VStoreVectorMasked);
+  REGISTER_OP(MEMSET,                 MemSet);
+  REGISTER_OP(MEMCPY,                 MemCpy);
  REGISTER_OP(CACHELINECLEAR,         CacheLineClear);
+  REGISTER_OP(CACHELINECLEAN,         CacheLineClean);
  REGISTER_OP(CACHELINEZERO,          CacheLineZero);

  // Misc ops
@@ -167,6 +165,7 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
  REGISTER_OP(CODEBLOCK,              NoOp);
  REGISTER_OP(BEGINBLOCK,             NoOp);
  REGISTER_OP(ENDBLOCK,               NoOp);
+  REGISTER_OP(GUESTOPCODE,            NoOp);
  REGISTER_OP(FENCE,                  Fence);
  REGISTER_OP(BREAK,                  Break);
  REGISTER_OP(PHI,                    NoOp);
@@ -176,19 +175,16 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
  REGISTER_OP(SETROUNDINGMODE,        SetRoundingMode);
  REGISTER_OP(INVALIDATEFLAGS,        NoOp);
  REGISTER_OP(PROCESSORID,            ProcessorID);
+  REGISTER_OP(RDRAND,                 RDRAND);
+  REGISTER_OP(YIELD,                  Yield);

  // Move ops
  REGISTER_OP(EXTRACTELEMENTPAIR,     ExtractElementPair);
  REGISTER_OP(CREATEELEMENTPAIR,      CreateElementPair);
-  REGISTER_OP(MOV,                    Mov);

  // Vector ops
  REGISTER_OP(VECTORZERO,             VectorZero);
  REGISTER_OP(VECTORIMM,              VectorImm);
-  REGISTER_OP(CREATEVECTOR2,          CreateVector2);
-  REGISTER_OP(CREATEVECTOR4,          CreateVector4);
-  REGISTER_OP(SPLATVECTOR2,           SplatVector);
-  REGISTER_OP(SPLATVECTOR4,           SplatVector);
  REGISTER_OP(VMOV,                   VMov);
  REGISTER_OP(VAND,                   VAnd);
  REGISTER_OP(VBIC,                   VBic);
@@ -227,6 +223,8 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
  REGISTER_OP(VZIP2,                  VZip);
  REGISTER_OP(VUNZIP,                 VUnZip);
  REGISTER_OP(VUNZIP2,                VUnZip);
+  REGISTER_OP(VTRN,                   VTrn);
+  REGISTER_OP(VTRN2,                  VTrn);
  REGISTER_OP(VBSL,                   VBSL);
  REGISTER_OP(VCMPEQ,                 VCMPEQ);
  REGISTER_OP(VCMPEQZ,                VCMPEQZ);
@@ -247,18 +245,13 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
  REGISTER_OP(VUSHRS,                 VUShrS);
  REGISTER_OP(VSSHRS,                 VSShrS);
  REGISTER_OP(VINSELEMENT,            VInsElement);
-  REGISTER_OP(VINSSCALARELEMENT,      VInsScalarElement);
-  REGISTER_OP(VEXTRACTELEMENT,        VExtractElement);
  REGISTER_OP(VDUPELEMENT,            VDupElement);
  REGISTER_OP(VEXTR,                  VExtr);
-  REGISTER_OP(VSLI,                   VSLI);
-  REGISTER_OP(VSRI,                   VSRI);
  REGISTER_OP(VUSHRI,                 VUShrI);
  REGISTER_OP(VSSHRI,                 VSShrI);
  REGISTER_OP(VSHLI,                  VShlI);
  REGISTER_OP(VUSHRNI,                VUShrNI);
  REGISTER_OP(VUSHRNI2,               VUShrNI2);
-  REGISTER_OP(VBITCAST,               VBitcast);
  REGISTER_OP(VSXTL,                  VSXTL);
  REGISTER_OP(VSXTL2,                 VSXTL2);
  REGISTER_OP(VUXTL,                  VUXTL);
@@ -274,7 +267,11 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
  REGISTER_OP(VUMULL2,                VUMull2);
  REGISTER_OP(VSMULL2,                VSMull2);
  REGISTER_OP(VUABDL,                 VUABDL);
+  REGISTER_OP(VUABDL2,                VUABDL2);
  REGISTER_OP(VTBL1,                  VTBL1);
+  REGISTER_OP(VREV64,                 VRev64);
+  REGISTER_OP(VPCMPESTRX,             VPCMPESTRX);
+  REGISTER_OP(VPCMPISTRX,             VPCMPISTRX);

  // Encryption ops
  REGISTER_OP(VAESIMC,                AESImc);
@@ -283,6 +280,8 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
  REGISTER_OP(VAESDEC,                AESDec);
  REGISTER_OP(VAESDECLAST,            AESDecLast);
  REGISTER_OP(VAESKEYGENASSIST,       AESKeyGenAssist);
+  REGISTER_OP(CRC32,                  CRC32);
+  REGISTER_OP(PCLMUL,                 PCLMUL);

  // F80 ops
  REGISTER_OP(F80LOADFCW,             F80LOADFCW);
@@ -311,6 +310,17 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
  REGISTER_OP(F80BCDLOAD,             F80BCDLOAD);
  REGISTER_OP(F80BCDSTORE,            F80BCDSTORE);

+  // F64 ops
+  REGISTER_OP(F64SIN,                 F64SIN);
+  REGISTER_OP(F64COS,                 F64COS);
+  REGISTER_OP(F64TAN,                 F64TAN);
+  REGISTER_OP(F64F2XM1,               F64F2XM1);
+  REGISTER_OP(F64ATAN,                F64ATAN);
+  REGISTER_OP(F64FPREM,               F64FPREM);
+  REGISTER_OP(F64FPREM1,              F64FPREM1);
+  REGISTER_OP(F64FYL2X,               F64FYL2X);
+  REGISTER_OP(F64SCALE,               F64SCALE);
+
  return Handlers;
 }();

@@ -321,237 +331,36 @@ void InterpreterOps::Op_Unhandled(FEXCore::IR::IROp_Header *IROp, IROpData *Data
 void InterpreterOps::Op_NoOp(FEXCore::IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node) {
 }

-template<typename R, typename... Args>
-static FallbackInfo GetFallbackInfo(R(*fn)(Args...)) {
-  return {FABI_UNKNOWN, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(float)) {
-  return {FABI_F80_F32, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(double)) {
-  return {FABI_F80_F64, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(int16_t)) {
-  return {FABI_F80_I16, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(void(*fn)(uint16_t)) {
-  return {FABI_VOID_U16, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(int32_t)) {
-  return {FABI_F80_I32, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(float(*fn)(X80SoftFloat)) {
-  return {FABI_F32_F80, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(double(*fn)(X80SoftFloat)) {
-  return {FABI_F64_F80, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(int16_t(*fn)(X80SoftFloat)) {
-  return {FABI_I16_F80, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(int32_t(*fn)(X80SoftFloat)) {
-  return {FABI_I32_F80, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(int64_t(*fn)(X80SoftFloat)) {
-  return {FABI_I64_F80, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(uint64_t(*fn)(X80SoftFloat, X80SoftFloat)) {
-  return {FABI_I64_F80_F80, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(X80SoftFloat)) {
-  return {FABI_F80_F80, (void*)fn};
-}
-
-template<>
-FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(X80SoftFloat, X80SoftFloat)) {
-  return {FABI_F80_F80_F80, (void*)fn};
-}
-
-bool InterpreterOps::GetFallbackHandler(IR::IROp_Header *IROp, FallbackInfo *Info) {
-  uint8_t OpSize = IROp->Size;
-  switch(IROp->Op) {
-    case IR::OP_F80LOADFCW: {
-      *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80LOADFCW>::handle);
-      return true;
-    }
-
-    case IR::OP_F80CVTTO: {
-      auto Op = IROp->C<IR::IROp_F80CVTTo>();
-
-      switch (Op->Size) {
-        case 4: {
-          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTO>::handle4);
-          return true;
-        }
-        case 8: {
-          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTO>::handle8);
-          return true;
-        }
-      default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
-      }
-      break;
-    }
-    case IR::OP_F80CVT: {
-      switch (OpSize) {
-        case 4: {
-          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVT>::handle4);
-          return true;
-        }
-        case 8: {
-          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVT>::handle8);
-          return true;
-        }
-        default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
-      }
-      break;
-    }
-    case IR::OP_F80CVTINT: {
-      auto Op = IROp->C<IR::IROp_F80CVTInt>();
-
-      switch (OpSize) {
-        case 2: {
-          *Info = GetFallbackInfo(Op->Truncate ? &FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2t : &FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2);
-          return true;
-        }
-        case 4: {
-          *Info = GetFallbackInfo(Op->Truncate ? &FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4t : &FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4);
-          return true;
-        }
-        case 8: {
-          *Info = GetFallbackInfo(Op->Truncate ? &FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8t : &FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8);
-          return true;
-        }
-        default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
-      }
-      break;
-    }
-    case IR::OP_F80CMP: {
-      auto Op = IROp->C<IR::IROp_F80Cmp>();
-
-      static constexpr std::array handlers{
-        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<0>,
-        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<1>,
-        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<2>,
-        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<3>,
-        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<4>,
-        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<5>,
-        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<6>,
-        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<7>,
-      };
-
-      *Info = GetFallbackInfo(handlers[Op->Flags]);
-      return true;
-    }
-
-    case IR::OP_F80CVTTOINT: {
-      auto Op = IROp->C<IR::IROp_F80CVTToInt>();
-
-      switch (Op->Size) {
-        case 2: {
-          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTOINT>::handle2);
-          return true;
-        }
-        case 4: {
-          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTOINT>::handle4);
-          return true;
-        }
-        default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
-      }
-      break;
-    }
-
-#define COMMON_X87_OP(OP) \
-    case IR::OP_F80##OP: { \
-      *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80##OP>::handle); \
-      return true; \
-    }
-
-    // Unary
-    COMMON_X87_OP(ROUND)
-    COMMON_X87_OP(F2XM1)
-    COMMON_X87_OP(TAN)
-    COMMON_X87_OP(SQRT)
-    COMMON_X87_OP(SIN)
-    COMMON_X87_OP(COS)
-    COMMON_X87_OP(XTRACT_EXP)
-    COMMON_X87_OP(XTRACT_SIG)
-    COMMON_X87_OP(BCDSTORE)
-    COMMON_X87_OP(BCDLOAD)
-
-    // Binary
-    COMMON_X87_OP(ADD)
-    COMMON_X87_OP(SUB)
-    COMMON_X87_OP(MUL)
-    COMMON_X87_OP(DIV)
-    COMMON_X87_OP(FYL2X)
-    COMMON_X87_OP(ATAN)
-    COMMON_X87_OP(FPREM1)
-    COMMON_X87_OP(FPREM)
-    COMMON_X87_OP(SCALE)
-
-    default:
-      break;
-  }
-
-  return false;
-}
-
-void InterpreterOps::InterpretIR(FEXCore::Core::InternalThreadState *Thread, uint64_t Entry, FEXCore::IR::IRListView *CurrentIR, FEXCore::Core::DebugData *DebugData) {
+void InterpreterOps::InterpretIR(FEXCore::Core::CpuStateFrame *Frame, FEXCore::IR::IRListView const *CurrentIR) {
  volatile void *StackEntry = alloca(0);

-  // Debug data is only passed in debug builds
-  #ifndef NDEBUG
-  // TODO: should be moved to an IR Op
-  Thread->Stats.InstructionsExecuted.fetch_add(DebugData->GuestInstructionCount);
-  #endif
+  const uintptr_t ListSize = CurrentIR->GetSSACount();

-  uintptr_t ListSize = CurrentIR->GetSSACount();
-
-  static_assert(sizeof(FEXCore::IR::IROp_Header) == 4);
  static_assert(sizeof(FEXCore::IR::OrderedNode) == 16);

  auto BlockEnd = CurrentIR->GetBlocks().end();

-  InterpreterOps::IROpData OpData{};
-  OpData.State = Thread;
-  OpData.SSAData = alloca(ListSize * 16);
-  OpData.CurrentEntry = Entry;
-  OpData.CurrentIR = CurrentIR;
-  OpData.StackEntry = StackEntry;
-  OpData.BlockIterator = CurrentIR->GetBlocks().begin();
+  constexpr size_t ListEntrySizeInBytes = sizeof(InterpVector256);
+  const size_t SSADataSize = ListSize * ListEntrySizeInBytes;

-  // Clear them all to zero. Required for Zero-extend semantics
-  memset(OpData.SSAData, 0, ListSize * 16);
+  InterpreterOps::IROpData OpData{
+    .State = Frame->Thread,
+    .CurrentEntry = Frame->State.rip,
+    .CurrentIR = CurrentIR,
+    .StackEntry = StackEntry,
+    .SSAData = alloca(SSADataSize),
+    .BlockResults = {},
+    .BlockIterator = CurrentIR->GetBlocks().begin(),
+  };
+
+  // Clear all SSAData entries to zero. Required for Zero-extend semantics
+  memset(OpData.SSAData, 0, SSADataSize);

  while (1) {
    using namespace FEXCore::IR;
    auto [BlockNode, BlockHeader] = OpData.BlockIterator();
    auto BlockIROp = BlockHeader->CW<IROp_CodeBlock>();
-    LOGMAN_THROW_A_FMT(BlockIROp->Header.Op == IR::OP_CODEBLOCK, "IR type failed to be a code block");
+    LOGMAN_THROW_AA_FMT(BlockIROp->Header.Op == IR::OP_CODEBLOCK, "IR type failed to be a code block");

    // Reset the block results per block
    memset(&OpData.BlockResults, 0, sizeof(OpData.BlockResults));
@@ -1,6 +1,7 @@
 #pragma once
 #include <stdint.h>

+#include <FEXCore/Core/CoreState.h>
 #include <FEXCore/IR/IR.h>
 #include <FEXCore/IR/IntrusiveIRList.h>

@@ -27,29 +28,35 @@ namespace FEXCore::CPU {
    FABI_F80_I32,
    FABI_F32_F80,
    FABI_F64_F80,
+    FABI_F64_F64,
+    FABI_F64_F64_F64,
    FABI_I16_F80,
    FABI_I32_F80,
    FABI_I64_F80,
    FABI_I64_F80_F80,
    FABI_F80_F80,
    FABI_F80_F80_F80,
+    FABI_I32_I64_I64_I128_I128_I16,
+    FABI_I32_I128_I128_I16,
  };

  struct FallbackInfo {
    FallbackABI ABI;
    void *fn;
+    FEXCore::Core::FallbackHandlerIndex HandlerIndex;
  };

  class InterpreterOps {

    public:
-      static void InterpretIR(FEXCore::Core::InternalThreadState *Thread, uint64_t Entry, FEXCore::IR::IRListView *CurrentIR, FEXCore::Core::DebugData *DebugData);
-      static bool GetFallbackHandler(IR::IROp_Header *IROp, FallbackInfo *Info);
+      static void InterpretIR(FEXCore::Core::CpuStateFrame *Frame, FEXCore::IR::IRListView const *IR);
+      static void FillFallbackIndexPointers(uint64_t *Info);
+      static bool GetFallbackHandler(IR::IROp_Header const *IROp, FallbackInfo *Info);

      struct IROpData {
        FEXCore::Core::InternalThreadState *State{};
        uint64_t CurrentEntry{};
-        FEXCore::IR::IRListView *CurrentIR{};
+        FEXCore::IR::IRListView const *CurrentIR{};
        volatile void *StackEntry{};
        void *SSAData{};
        struct {
@@ -78,6 +85,7 @@ namespace FEXCore::CPU {
  DEF_OP(Add);
  DEF_OP(Sub);
  DEF_OP(Neg);
+  DEF_OP(Abs);
  DEF_OP(Mul);
  DEF_OP(UMul);
  DEF_OP(Div);
@@ -137,10 +145,6 @@ namespace FEXCore::CPU {
  DEF_OP(AtomicFetchNeg);

  ///< Branch ops
-  DEF_OP(GuestCallDirect);
-  DEF_OP(GuestCallIndirect);
-  DEF_OP(GuestReturn);
-  DEF_OP(SignalReturn);
  DEF_OP(CallbackReturn);
  DEF_OP(ExitFunction);
  DEF_OP(Jump);
@@ -149,12 +153,14 @@ namespace FEXCore::CPU {
  DEF_OP(InlineSyscall);
  DEF_OP(Thunk);
  DEF_OP(ValidateCode);
-  DEF_OP(RemoveCodeEntry);
+  DEF_OP(ThreadRemoveCodeEntry);
  DEF_OP(CPUID);
+  DEF_OP(XGETBV);

  ///< Conversion ops
  DEF_OP(VInsGPR);
  DEF_OP(VCastFromGPR);
+  DEF_OP(VDupFromGPR);
  DEF_OP(Float_FromGPR_S);
  DEF_OP(Float_FToF);
  DEF_OP(Vector_SToF);
@@ -179,9 +185,12 @@ namespace FEXCore::CPU {
  DEF_OP(StoreFlag);
  DEF_OP(LoadMem);
  DEF_OP(StoreMem);
-  DEF_OP(VLoadMemElement);
-  DEF_OP(VStoreMemElement);
+  DEF_OP(VLoadVectorMasked);
+  DEF_OP(VStoreVectorMasked);
+  DEF_OP(MemSet);
+  DEF_OP(MemCpy);
  DEF_OP(CacheLineClear);
+  DEF_OP(CacheLineClean);
  DEF_OP(CacheLineZero);

  ///< Misc ops
@@ -194,6 +203,8 @@ namespace FEXCore::CPU {
  DEF_OP(GetRoundingMode);
  DEF_OP(SetRoundingMode);
  DEF_OP(ProcessorID);
+  DEF_OP(RDRAND);
+  DEF_OP(Yield);

  ///< Move ops
  DEF_OP(ExtractElementPair);
@@ -203,9 +214,6 @@ namespace FEXCore::CPU {
  ///< Vector ops
  DEF_OP(VectorZero);
  DEF_OP(VectorImm);
-  DEF_OP(CreateVector2);
-  DEF_OP(CreateVector4);
-  DEF_OP(SplatVector);
  DEF_OP(VMov);
  DEF_OP(VAnd);
  DEF_OP(VBic);
@@ -242,6 +250,7 @@ namespace FEXCore::CPU {
  DEF_OP(VSMax);
  DEF_OP(VZip);
  DEF_OP(VUnZip);
+  DEF_OP(VTrn);
  DEF_OP(VBSL);
  DEF_OP(VCMPEQ);
  DEF_OP(VCMPEQZ);
@@ -262,18 +271,13 @@ namespace FEXCore::CPU {
  DEF_OP(VUShrS);
  DEF_OP(VSShrS);
  DEF_OP(VInsElement);
-  DEF_OP(VInsScalarElement);
-  DEF_OP(VExtractElement);
  DEF_OP(VDupElement);
  DEF_OP(VExtr);
-  DEF_OP(VSLI);
-  DEF_OP(VSRI);
  DEF_OP(VUShrI);
  DEF_OP(VSShrI);
  DEF_OP(VShlI);
  DEF_OP(VUShrNI);
  DEF_OP(VUShrNI2);
-  DEF_OP(VBitcast);
  DEF_OP(VSXTL);
  DEF_OP(VSXTL2);
  DEF_OP(VUXTL);
@@ -289,7 +293,11 @@ namespace FEXCore::CPU {
  DEF_OP(VUMull2);
  DEF_OP(VSMull2);
  DEF_OP(VUABDL);
+  DEF_OP(VUABDL2);
  DEF_OP(VTBL1);
+  DEF_OP(VRev64);
+  DEF_OP(VPCMPESTRX);
+  DEF_OP(VPCMPISTRX);

  ///< Encryption ops
  DEF_OP(AESImc);
@@ -298,6 +306,8 @@ namespace FEXCore::CPU {
  DEF_OP(AESDec);
  DEF_OP(AESDecLast);
  DEF_OP(AESKeyGenAssist);
+  DEF_OP(CRC32);
+  DEF_OP(PCLMUL);

  ///< F80 ops
  DEF_OP(F80LOADFCW);
@@ -325,10 +335,30 @@ namespace FEXCore::CPU {
  DEF_OP(F80CMP);
  DEF_OP(F80BCDLOAD);
  DEF_OP(F80BCDSTORE);
+
+  //< F64 ops
+  DEF_OP(F64SIN);
+  DEF_OP(F64COS);
+  DEF_OP(F64TAN);
+  DEF_OP(F64F2XM1);
+  DEF_OP(F64ATAN);
+  DEF_OP(F64FPREM);
+  DEF_OP(F64FPREM1);
+  DEF_OP(F64FYL2X);
+  DEF_OP(F64SCALE);
 #undef DEF_OP
  template<typename unsigned_type, typename signed_type, typename float_type>
  [[nodiscard]] static bool IsConditionTrue(uint8_t Cond, uint64_t Src1, uint64_t Src2) {
    bool CompResult = false;
+    if constexpr (sizeof(unsigned_type) == 16) {
+      LOGMAN_THROW_A_FMT(Cond != FEXCore::IR::COND_FLU &&
+        Cond != FEXCore::IR::COND_FGE &&
+        Cond != FEXCore::IR::COND_FLEU &&
+        Cond != FEXCore::IR::COND_FGT &&
+        Cond != FEXCore::IR::COND_FU &&
+        Cond != FEXCore::IR::COND_FNU, "Unsupported comparison for 128-bit floats");
+    }
+
    switch (Cond) {
      case FEXCore::IR::COND_EQ:
        CompResult = static_cast<unsigned_type>(Src1) == static_cast<unsigned_type>(Src2);
@@ -391,7 +421,7 @@ namespace FEXCore::CPU {
    return CompResult;
  }

-  static uint8_t GetOpSize(FEXCore::IR::IRListView *CurrentIR, IR::OrderedNodeWrapper Node) {
+  static uint8_t GetOpSize(FEXCore::IR::IRListView const *CurrentIR, IR::OrderedNodeWrapper Node) {
    auto IROp = CurrentIR->GetOp<FEXCore::IR::IROp_Header>(Node);
    return IROp->Size;
  }
@@ -4,6 +4,7 @@ tags: backend|interpreter
 $end_info$
 */

+#include "Interface/Core/CPUID.h"
 #include "Interface/Core/Interpreter/InterpreterClass.h"
 #include "Interface/Core/Interpreter/InterpreterOps.h"
 #include "Interface/Core/Interpreter/InterpreterDefines.h"
@@ -22,95 +23,157 @@ static inline void CacheLineFlush(char *Addr) {
 #endif
 }

+static inline void CacheLineClean(char *Addr) {
+#ifdef _M_X86_64
+  __asm volatile (
+    "clwb (%[Addr]);"
+    :: [Addr] "r" (Addr)
+    : "memory");
+#elif _M_ARM_64
+  __asm volatile (
+    "dc cvac, %[Addr]"
+    :: [Addr] "r" (Addr)
+    : "memory");
+#else
+    LOGMAN_THROW_A_FMT("Unsupported architecture with cacheline clean");
+#endif
+}
+
 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
 DEF_OP(LoadContext) {
-  auto Op = IROp->C<IR::IROp_LoadContext>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_LoadContext>();
+  const auto OpSize = IROp->Size;
+
+  const auto ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
+  const auto Src = ContextPtr + Op->Offset;

-  uintptr_t ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
-  ContextPtr += Op->Offset;
  #define LOAD_CTX(x, y) \
    case x: { \
-      y const *MemData = reinterpret_cast<y const*>(ContextPtr); \
+      y const *MemData = reinterpret_cast<y const*>(Src); \
      GD = *MemData; \
      break; \
    }
+
  switch (OpSize) {
    LOAD_CTX(1, uint8_t)
    LOAD_CTX(2, uint16_t)
    LOAD_CTX(4, uint32_t)
    LOAD_CTX(8, uint64_t)
-    case 16: {
-      void const *MemData = reinterpret_cast<void const*>(ContextPtr);
+    case 16:
+    case 32: {
+      void const *MemData = reinterpret_cast<void const*>(Src);
      memcpy(GDP, MemData, OpSize);
      break;
    }
-    default:  LOGMAN_MSG_A_FMT("Unhandled LoadContext size: {}", OpSize);
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled LoadContext size: {}", OpSize);
+      break;
  }
  #undef LOAD_CTX
 }

 DEF_OP(StoreContext) {
-  auto Op = IROp->C<IR::IROp_StoreContext>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_StoreContext>();
+  const auto OpSize = IROp->Size;

-  uintptr_t ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
-  ContextPtr += Op->Offset;
+  const auto ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
+  const auto Dst = ContextPtr + Op->Offset;

-  void *MemData = reinterpret_cast<void*>(ContextPtr);
-  void *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
+  void *MemData = reinterpret_cast<void*>(Dst);
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Value);
  memcpy(MemData, Src, OpSize);
 }

 DEF_OP(LoadRegister) {
-  LOGMAN_MSG_A_FMT("Unimplemented");
-}
+  const auto Op = IROp->C<IR::IROp_LoadRegister>();
+  const auto OpSize = IROp->Size;

-DEF_OP(StoreRegister) {
-  LOGMAN_MSG_A_FMT("Unimplemented");
-}
-
-DEF_OP(LoadContextIndexed) {
-  auto Op = IROp->C<IR::IROp_LoadContextIndexed>();
-  uint64_t Index = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-
-  uintptr_t ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
-
-  ContextPtr += Op->BaseOffset;
-  ContextPtr += Index * Op->Stride;
+  const auto ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
+  const auto Src = ContextPtr + Op->Offset;

  #define LOAD_CTX(x, y) \
    case x: { \
-      y const *MemData = reinterpret_cast<y const*>(ContextPtr); \
+      y const *MemData = reinterpret_cast<y const*>(Src); \
      GD = *MemData; \
      break; \
    }
-  switch (IROp->Size) {
+
+  switch (OpSize) {
    LOAD_CTX(1, uint8_t)
    LOAD_CTX(2, uint16_t)
    LOAD_CTX(4, uint32_t)
    LOAD_CTX(8, uint64_t)
-    case 16: {
-      void const *MemData = reinterpret_cast<void const*>(ContextPtr);
-      memcpy(GDP, MemData, IROp->Size);
+    case 16:
+    case 32: {
+      void const *MemData = reinterpret_cast<void const*>(Src);
+      memcpy(GDP, MemData, OpSize);
      break;
    }
-    default:  LOGMAN_MSG_A_FMT("Unhandled LoadContextIndexed size: {}", IROp->Size);
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled LoadContext size: {}", OpSize);
+      break;
+  }
+  #undef LOAD_CTX
+}
+
+DEF_OP(StoreRegister) {
+  const auto Op = IROp->C<IR::IROp_StoreRegister>();
+  const auto OpSize = IROp->Size;
+
+  const auto ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
+  const auto Dst = ContextPtr + Op->Offset;
+
+  void *MemData = reinterpret_cast<void*>(Dst);
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Value);
+  memcpy(MemData, Src, OpSize);
+}
+
+DEF_OP(LoadContextIndexed) {
+  const auto Op = IROp->C<IR::IROp_LoadContextIndexed>();
+  const auto OpSize = IROp->Size;
+
+  const auto Index = *GetSrc<uint64_t*>(Data->SSAData, Op->Index);
+
+  const auto ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
+  const auto Src = ContextPtr + Op->BaseOffset + (Index * Op->Stride);
+
+  #define LOAD_CTX(x, y) \
+    case x: { \
+      y const *MemData = reinterpret_cast<y const*>(Src); \
+      GD = *MemData; \
+      break; \
+    }
+
+  switch (OpSize) {
+    LOAD_CTX(1, uint8_t)
+    LOAD_CTX(2, uint16_t)
+    LOAD_CTX(4, uint32_t)
+    LOAD_CTX(8, uint64_t)
+    case 16:
+    case 32: {
+      void const *MemData = reinterpret_cast<void const*>(Src);
+      memcpy(GDP, MemData, OpSize);
+      break;
+    }
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled LoadContextIndexed size: {}", OpSize);
+      break;
  }
  #undef LOAD_CTX
 }

 DEF_OP(StoreContextIndexed) {
-  auto Op = IROp->C<IR::IROp_StoreContextIndexed>();
-  uint64_t Index = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const auto Op = IROp->C<IR::IROp_StoreContextIndexed>();
+  const auto OpSize = IROp->Size;

-  uintptr_t ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
-  ContextPtr += Op->BaseOffset;
-  ContextPtr += Index * Op->Stride;
+  const auto Index = *GetSrc<uint64_t*>(Data->SSAData, Op->Index);

-  void *MemData = reinterpret_cast<void*>(ContextPtr);
-  void *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  memcpy(MemData, Src, IROp->Size);
+  const auto ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
+  const auto Dst = ContextPtr + Op->BaseOffset + (Index * Op->Stride);
+
+  void *MemData = reinterpret_cast<void*>(Dst);
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Value);
+  memcpy(MemData, Src, OpSize);
 }

 DEF_OP(SpillRegister) {
@@ -127,24 +190,36 @@ DEF_OP(LoadFlag) {
  uintptr_t ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
  ContextPtr += offsetof(FEXCore::Core::CPUState, flags[0]);
  ContextPtr += Op->Flag;
-  uint8_t const *MemData = reinterpret_cast<uint8_t const*>(ContextPtr);
-  GD = *MemData;
+
+  if (Op->Flag == 24 /* NZCV */) {
+    uint32_t const *MemData = reinterpret_cast<uint32_t const*>(ContextPtr);
+    GD = *MemData;
+  } else {
+    uint8_t const *MemData = reinterpret_cast<uint8_t const*>(ContextPtr);
+    GD = *MemData;
+  }
 }

 DEF_OP(StoreFlag) {
  auto Op = IROp->C<IR::IROp_StoreFlag>();
-  uint8_t Arg = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[0]);
+  uint32_t Arg = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);

  uintptr_t ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
  ContextPtr += offsetof(FEXCore::Core::CPUState, flags[0]);
  ContextPtr += Op->Flag;
-  uint8_t *MemData = reinterpret_cast<uint8_t*>(ContextPtr);
-  *MemData = Arg;
+
+  if (Op->Flag == 24 /* NZCV */) {
+    uint32_t *MemData = reinterpret_cast<uint32_t*>(ContextPtr);
+    *MemData = Arg;
+  } else {
+    uint8_t *MemData = reinterpret_cast<uint8_t*>(ContextPtr);
+    *MemData = Arg;
+  }
 }

 DEF_OP(LoadMem) {
-  auto Op = IROp->C<IR::IROp_LoadMem>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_LoadMem>();
+  const auto OpSize = IROp->Size;

  uint8_t const *MemData = *GetSrc<uint8_t const**>(Data->SSAData, Op->Addr);

@@ -157,7 +232,8 @@ DEF_OP(LoadMem) {
      case IR::MEM_OFFSET_SXTW.Val: MemData += (int32_t)Offset; break;
    }
  }
-  memset(GDP, 0, 16);
+
+  memset(GDP, 0, Core::CPUState::XMM_AVX_REG_SIZE);
  switch (OpSize) {
    case 1: {
      auto D = reinterpret_cast<const std::atomic<uint8_t>*>(MemData);
@@ -179,16 +255,15 @@ DEF_OP(LoadMem) {
      GD = D->load();
      break;
    }
-
    default:
-      memcpy(GDP, MemData, IROp->Size);
+      memcpy(GDP, MemData, OpSize);
      break;
  }
 }

 DEF_OP(StoreMem) {
-  auto Op = IROp->C<IR::IROp_StoreMem>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_StoreMem>();
+  const auto OpSize = IROp->Size;

  uint8_t *MemData = *GetSrc<uint8_t **>(Data->SSAData, Op->Addr);

@@ -220,39 +295,369 @@ DEF_OP(StoreMem) {
    }

    default:
-      memcpy(MemData, GetSrc<void*>(Data->SSAData, Op->Value), IROp->Size);
+      memcpy(MemData, GetSrc<void*>(Data->SSAData, Op->Value), OpSize);
      break;
  }
 }

-DEF_OP(VLoadMemElement) {
-  auto Op = IROp->C<IR::IROp_VLoadMemElement>();
-  void const *MemData = *GetSrc<void const**>(Data->SSAData, Op->Header.Args[0]);
+DEF_OP(VLoadVectorMasked) {
+  const auto Op = IROp->C<IR::IROp_VLoadVectorMasked>();
+  const auto OpSize = IROp->Size;

-  memcpy(GDP, GetSrc<void*>(Data->SSAData, Op->Header.Args[1]), 16);
-  memcpy(reinterpret_cast<void*>(reinterpret_cast<uintptr_t>(GDP) + (Op->Header.ElementSize * Op->Index)),
-    MemData, Op->Header.ElementSize);
+  const auto ElementSize = IROp->ElementSize;
+  const auto NumElements = OpSize / ElementSize;
+
+  const auto *MemData = *GetSrc<uint8_t const**>(Data->SSAData, Op->Addr);
+  const auto *Mask = GetSrc<uint8_t const*>(Data->SSAData, Op->Mask);
+
+  const auto SetElements = [NumElements]<typename T>(void* Dst, const T* MaskValues, const T* MemoryData) {
+    const auto SignBit = 1ULL << ((sizeof(T) * 8) - 1);
+    for (size_t i = 0; i < NumElements; i++) {
+      if ((MaskValues[i] & SignBit) != 0) {
+        std::memcpy(static_cast<uint8_t*>(Dst) + (i * sizeof(T)), MemoryData + i, sizeof(T));
+      }
+    }
+  };
+
+  if (!Op->Offset.IsInvalid()) {
+    auto Offset = *GetSrc<uintptr_t const*>(Data->SSAData, Op->Offset) * Op->OffsetScale;
+
+    switch(Op->OffsetType.Val) {
+      case IR::MEM_OFFSET_SXTX.Val: MemData +=  Offset; break;
+      case IR::MEM_OFFSET_UXTW.Val: MemData += (uint32_t)Offset; break;
+      case IR::MEM_OFFSET_SXTW.Val: MemData += (int32_t)Offset; break;
+    }
+  }
+
+  memset(GDP, 0, Core::CPUState::XMM_AVX_REG_SIZE);
+  switch (ElementSize) {
+    case 1: {
+      SetElements(GDP, Mask, MemData);
+      return;
+    }
+    case 2: {
+      SetElements(GDP,
+                  reinterpret_cast<const uint16_t*>(Mask),
+                  reinterpret_cast<const uint16_t*>(MemData));
+      return;
+    }
+    case 4: {
+      SetElements(GDP,
+                  reinterpret_cast<const uint32_t*>(Mask),
+                  reinterpret_cast<const uint32_t*>(MemData));
+      return;
+    }
+    case 8: {
+      SetElements(GDP,
+                  reinterpret_cast<const uint64_t*>(Mask),
+                  reinterpret_cast<const uint64_t*>(MemData));
+      return;
+    }
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled VLoadVectorMasked element size: {}", ElementSize);
+      return;
+  }
 }

-DEF_OP(VStoreMemElement) {
-  #define STORE_DATA(x, y) \
-    case x: { \
-      y *MemData = *GetSrc<y**>(Data->SSAData, Op->Header.Args[0]); \
-      memcpy(MemData, &GetSrc<y*>(Data->SSAData, Op->Header.Args[1])[Op->Index], sizeof(y)); \
-      break; \
+DEF_OP(VStoreVectorMasked) {
+  const auto Op = IROp->C<IR::IROp_VStoreVectorMasked>();
+  const auto OpSize = IROp->Size;
+
+  const auto ElementSize = IROp->ElementSize;
+  const auto NumElements = OpSize / ElementSize;
+
+  auto *Dst = *GetSrc<uint8_t**>(Data->SSAData, Op->Addr);
+  const auto *RegData = GetSrc<uint8_t const*>(Data->SSAData, Op->Data);
+  const auto *Mask = GetSrc<uint8_t const*>(Data->SSAData, Op->Mask);
+
+  const auto SetElements = [NumElements]<typename T>(void* Dst, const T* MaskValues, const T* DataVals) {
+    const auto SignBit = 1ULL << ((sizeof(T) * 8) - 1);
+    for (size_t i = 0; i < NumElements; i++) {
+      if ((MaskValues[i] & SignBit) != 0) {
+        std::memcpy(static_cast<uint8_t*>(Dst) + (i * sizeof(T)), DataVals + i, sizeof(T));
+      }
    }
+  };

-  auto Op = IROp->C<IR::IROp_VStoreMemElement>();
-  uint8_t OpSize = IROp->Size;
+  if (!Op->Offset.IsInvalid()) {
+    auto Offset = *GetSrc<uintptr_t const*>(Data->SSAData, Op->Offset) * Op->OffsetScale;

-  switch (OpSize) {
-    STORE_DATA(1, uint8_t)
-    STORE_DATA(2, uint16_t)
-    STORE_DATA(4, uint32_t)
-    STORE_DATA(8, uint64_t)
-    default: LOGMAN_MSG_A_FMT("Unhandled StoreMem size"); break;
+    switch(Op->OffsetType.Val) {
+      case IR::MEM_OFFSET_SXTX.Val: Dst +=  Offset; break;
+      case IR::MEM_OFFSET_UXTW.Val: Dst += (uint32_t)Offset; break;
+      case IR::MEM_OFFSET_SXTW.Val: Dst += (int32_t)Offset; break;
+    }
+  }
+
+  switch (ElementSize) {
+    case 1: {
+      SetElements(Dst, Mask, RegData);
+      return;
+    }
+    case 2: {
+      SetElements(Dst,
+                  reinterpret_cast<const uint16_t*>(Mask),
+                  reinterpret_cast<const uint16_t*>(RegData));
+      return;
+    }
+    case 4: {
+      SetElements(Dst,
+                  reinterpret_cast<const uint32_t*>(Mask),
+                  reinterpret_cast<const uint32_t*>(RegData));
+      return;
+    }
+    case 8: {
+      SetElements(Dst,
+                  reinterpret_cast<const uint64_t*>(Mask),
+                  reinterpret_cast<const uint64_t*>(RegData));
+      return;
+    }
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled VStoreVectorMasked element size: {}", ElementSize);
+      return;
+  }
+}
+
+DEF_OP(MemSet) {
+  const auto Op = IROp->C<IR::IROp_MemSet>();
+  const int32_t Size = Op->Size;
+
+  char *MemData = *GetSrc<char **>(Data->SSAData, Op->Addr);
+  uint64_t MemPrefix{};
+  if (!Op->Prefix.IsInvalid()) {
+    MemPrefix = *GetSrc<uint64_t*>(Data->SSAData, Op->Prefix);
+  }
+
+  const auto Value = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
+  const auto Length = *GetSrc<uint64_t*>(Data->SSAData, Op->Length);
+  const auto Direction = *GetSrc<uint8_t*>(Data->SSAData, Op->Direction);
+
+  auto MemSetElements = [](auto* Memory, uint64_t Value, size_t Length) {
+    for (size_t i = 0; i < Length; ++i) {
+      Memory[i] = Value;
+    }
+  };
+
+  auto MemSetElementsInverse = [](auto* Memory, uint64_t Value, size_t Length) {
+    for (size_t i = 0; i < Length; ++i) {
+      Memory[-i] = Value;
+    }
+  };
+
+  if (Direction == 0) { // Forward
+    if (Op->IsAtomic) {
+      switch (Size) {
+        case 1:
+          MemSetElements(reinterpret_cast<std::atomic<uint8_t>*>(MemData + MemPrefix), Value, Length);
+          break;
+        case 2:
+          MemSetElements(reinterpret_cast<std::atomic<uint16_t>*>(MemData + MemPrefix), Value, Length);
+          break;
+        case 4:
+          MemSetElements(reinterpret_cast<std::atomic<uint32_t>*>(MemData + MemPrefix), Value, Length);
+          break;
+        case 8:
+          MemSetElements(reinterpret_cast<std::atomic<uint64_t>*>(MemData + MemPrefix), Value, Length);
+          break;
+        default:
+          LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, Size);
+          break;
+      }
+    }
+    else {
+      switch (Size) {
+        case 1:
+          MemSetElements(reinterpret_cast<uint8_t*>(MemData + MemPrefix), Value, Length);
+          break;
+        case 2:
+          MemSetElements(reinterpret_cast<uint16_t*>(MemData + MemPrefix), Value, Length);
+          break;
+        case 4:
+          MemSetElements(reinterpret_cast<uint32_t*>(MemData + MemPrefix), Value, Length);
+          break;
+        case 8:
+          MemSetElements(reinterpret_cast<uint64_t*>(MemData + MemPrefix), Value, Length);
+          break;
+        default:
+          LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, Size);
+          break;
+      }
+    }
+    GD = reinterpret_cast<uint64_t>(MemData + (Length * Size));
+  }
+  else { // Backward
+    if (Op->IsAtomic) {
+      switch (Size) {
+        case 1:
+          MemSetElementsInverse(reinterpret_cast<std::atomic<uint8_t>*>(MemData + MemPrefix), Value, Length);
+          break;
+        case 2:
+          MemSetElementsInverse(reinterpret_cast<std::atomic<uint16_t>*>(MemData + MemPrefix), Value, Length);
+          break;
+        case 4:
+          MemSetElementsInverse(reinterpret_cast<std::atomic<uint32_t>*>(MemData + MemPrefix), Value, Length);
+          break;
+        case 8:
+          MemSetElementsInverse(reinterpret_cast<std::atomic<uint64_t>*>(MemData + MemPrefix), Value, Length);
+          break;
+        default:
+          LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, Size);
+          break;
+      }
+    }
+    else {
+      switch (Size) {
+        case 1:
+          MemSetElementsInverse(reinterpret_cast<uint8_t*>(MemData + MemPrefix), Value, Length);
+          break;
+        case 2:
+          MemSetElementsInverse(reinterpret_cast<uint16_t*>(MemData + MemPrefix), Value, Length);
+          break;
+        case 4:
+          MemSetElementsInverse(reinterpret_cast<uint32_t*>(MemData + MemPrefix), Value, Length);
+          break;
+        case 8:
+          MemSetElementsInverse(reinterpret_cast<uint64_t*>(MemData + MemPrefix), Value, Length);
+          break;
+        default:
+          LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, Size);
+          break;
+      }
+    }
+    GD = reinterpret_cast<uint64_t>(MemData - (Length * Size));
+  }
+}
+
+DEF_OP(MemCpy) {
+  const auto Op = IROp->C<IR::IROp_MemCpy>();
+  const int32_t Size = Op->Size;
+
+  uint64_t *DstPtr = GetDest<uint64_t*>(Data->SSAData, Node);
+
+  char *MemDataDest = *GetSrc<char **>(Data->SSAData, Op->AddrDest);
+  char *MemDataSrc = *GetSrc<char **>(Data->SSAData, Op->AddrSrc);
+
+  uint64_t DestPrefix{};
+  uint64_t SrcPrefix{};
+  if (!Op->PrefixDest.IsInvalid()) {
+    DestPrefix = *GetSrc<uint64_t*>(Data->SSAData, Op->PrefixDest);
+
+  }
+  if (!Op->PrefixSrc.IsInvalid()) {
+    SrcPrefix = *GetSrc<uint64_t*>(Data->SSAData, Op->PrefixSrc);
+  }
+
+  const auto Length = *GetSrc<uint64_t*>(Data->SSAData, Op->Length);
+  const auto Direction = *GetSrc<uint8_t*>(Data->SSAData, Op->Direction);
+
+  auto MemSetElementsAtomic = [](auto* MemDst, auto* MemSrc, size_t Length) {
+    for (size_t i = 0; i < Length; ++i) {
+      MemDst[i].store(MemSrc[i].load());
+    }
+  };
+
+  auto MemSetElementsAtomicInverse = [](auto* MemDst, auto* MemSrc, size_t Length) {
+    for (size_t i = 0; i < Length; ++i) {
+      MemDst[-i].store(MemSrc[-i].load());
+    }
+  };
+
+  auto MemSetElements = [](auto* MemDst, auto* MemSrc, size_t Length) {
+    for (size_t i = 0; i < Length; ++i) {
+      MemDst[i] = MemSrc[i];
+    }
+  };
+
+  auto MemSetElementsInverse = [](auto* MemDst, auto* MemSrc, size_t Length) {
+    for (size_t i = 0; i < Length; ++i) {
+      MemDst[-i] = MemSrc[-i];
+    }
+  };
+
+  if (Direction == 0) { // Forward
+    if (Op->IsAtomic) {
+      switch (Size) {
+        case 1:
+          MemSetElementsAtomic(reinterpret_cast<std::atomic<uint8_t>*>(MemDataDest + DestPrefix), reinterpret_cast<std::atomic<uint8_t>*>(MemDataSrc + SrcPrefix), Length);
+          break;
+        case 2:
+          MemSetElementsAtomic(reinterpret_cast<std::atomic<uint16_t>*>(MemDataDest + DestPrefix), reinterpret_cast<std::atomic<uint16_t>*>(MemDataSrc + SrcPrefix), Length);
+          break;
+        case 4:
+          MemSetElementsAtomic(reinterpret_cast<std::atomic<uint32_t>*>(MemDataDest + DestPrefix), reinterpret_cast<std::atomic<uint32_t>*>(MemDataSrc + SrcPrefix), Length);
+          break;
+        case 8:
+          MemSetElementsAtomic(reinterpret_cast<std::atomic<uint64_t>*>(MemDataDest + DestPrefix), reinterpret_cast<std::atomic<uint64_t>*>(MemDataSrc + SrcPrefix), Length);
+          break;
+        default:
+          LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, Size);
+          break;
+      }
+    }
+    else {
+      switch (Size) {
+        case 1:
+          MemSetElements(reinterpret_cast<uint8_t*>(MemDataDest + DestPrefix), reinterpret_cast<uint8_t*>(MemDataSrc + SrcPrefix), Length);
+          break;
+        case 2:
+          MemSetElements(reinterpret_cast<uint16_t*>(MemDataDest + DestPrefix), reinterpret_cast<uint16_t*>(MemDataSrc + SrcPrefix), Length);
+          break;
+        case 4:
+          MemSetElements(reinterpret_cast<uint32_t*>(MemDataDest + DestPrefix), reinterpret_cast<uint32_t*>(MemDataSrc + SrcPrefix), Length);
+          break;
+        case 8:
+          MemSetElements(reinterpret_cast<uint64_t*>(MemDataDest + DestPrefix), reinterpret_cast<uint64_t*>(MemDataSrc + SrcPrefix), Length);
+          break;
+        default:
+          LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, Size);
+          break;
+      }
+    }
+    DstPtr[0] = reinterpret_cast<uint64_t>(MemDataDest + (Length * Size));
+    DstPtr[1] = reinterpret_cast<uint64_t>(MemDataSrc + (Length * Size));
+  }
+  else { // Backward
+    if (Op->IsAtomic) {
+      switch (Size) {
+        case 1:
+          MemSetElementsAtomicInverse(reinterpret_cast<std::atomic<uint8_t>*>(MemDataDest + DestPrefix), reinterpret_cast<std::atomic<uint8_t>*>(MemDataSrc + SrcPrefix), Length);
+          break;
+        case 2:
+          MemSetElementsAtomicInverse(reinterpret_cast<std::atomic<uint16_t>*>(MemDataDest + DestPrefix), reinterpret_cast<std::atomic<uint16_t>*>(MemDataSrc + SrcPrefix), Length);
+          break;
+        case 4:
+          MemSetElementsAtomicInverse(reinterpret_cast<std::atomic<uint32_t>*>(MemDataDest + DestPrefix), reinterpret_cast<std::atomic<uint32_t>*>(MemDataSrc + SrcPrefix), Length);
+          break;
+        case 8:
+          MemSetElementsAtomicInverse(reinterpret_cast<std::atomic<uint64_t>*>(MemDataDest + DestPrefix), reinterpret_cast<std::atomic<uint64_t>*>(MemDataSrc + SrcPrefix), Length);
+          break;
+        default:
+          LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, Size);
+          break;
+      }
+    }
+    else {
+      switch (Size) {
+        case 1:
+          MemSetElementsInverse(reinterpret_cast<uint8_t*>(MemDataDest + DestPrefix), reinterpret_cast<uint8_t*>(MemDataSrc + SrcPrefix), Length);
+          break;
+        case 2:
+          MemSetElementsInverse(reinterpret_cast<uint16_t*>(MemDataDest + DestPrefix), reinterpret_cast<uint16_t*>(MemDataSrc + SrcPrefix), Length);
+          break;
+        case 4:
+          MemSetElementsInverse(reinterpret_cast<uint32_t*>(MemDataDest + DestPrefix), reinterpret_cast<uint32_t*>(MemDataSrc + SrcPrefix), Length);
+          break;
+        case 8:
+          MemSetElementsInverse(reinterpret_cast<uint64_t*>(MemDataDest + DestPrefix), reinterpret_cast<uint64_t*>(MemDataSrc + SrcPrefix), Length);
+          break;
+        default:
+          LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, Size);
+          break;
+      }
+    }
+    DstPtr[0] = reinterpret_cast<uint64_t>(MemDataDest - (Length * Size));
+    DstPtr[1] = reinterpret_cast<uint64_t>(MemDataSrc - (Length * Size));
  }
-  #undef STORE_DATA
 }

 DEF_OP(CacheLineClear) {
@@ -264,6 +669,15 @@ DEF_OP(CacheLineClear) {
  CacheLineFlush(MemData);
 }

+DEF_OP(CacheLineClean) {
+  auto Op = IROp->C<IR::IROp_CacheLineClean>();
+
+  char *MemData = *GetSrc<char **>(Data->SSAData, Op->Addr);
+
+  // 64-byte cache line clear
+  CacheLineClean(MemData);
+}
+
 DEF_OP(CacheLineZero) {
  auto Op = IROp->C<IR::IROp_CacheLineZero>();

@@ -4,6 +4,8 @@ tags: backend|interpreter
 $end_info$
 */

+#include "Interface/Context/Context.h"
+
 #include "Interface/Core/Interpreter/InterpreterClass.h"
 #include "Interface/Core/Interpreter/InterpreterOps.h"
 #include "Interface/Core/Interpreter/InterpreterDefines.h"
@@ -14,15 +16,9 @@ $end_info$
 #ifdef _M_X86_64
 #include <xmmintrin.h>
 #endif
+#include <sys/random.h>

 namespace FEXCore::CPU {
-[[noreturn]]
-static void StopThread(FEXCore::Core::InternalThreadState *Thread) {
-  Thread->CTX->StopThread(Thread);
-
-  LOGMAN_MSG_A_FMT("unreachable");
-  FEX_UNREACHABLE;
-}

 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
 DEF_OP(Fence) {
@@ -43,14 +39,26 @@ DEF_OP(Fence) {

 DEF_OP(Break) {
  auto Op = IROp->C<IR::IROp_Break>();
-  switch (Op->Reason) {
-    case FEXCore::IR::Break_Halt: // HLT
-      StopThread(Data->State);
+
+  Data->State->CurrentFrame->SynchronousFaultData.FaultToTopAndGeneratedException = 1;
+  Data->State->CurrentFrame->SynchronousFaultData.Signal = Op->Reason.Signal;
+  Data->State->CurrentFrame->SynchronousFaultData.TrapNo = Op->Reason.TrapNumber;
+  Data->State->CurrentFrame->SynchronousFaultData.err_code = Op->Reason.ErrorRegister;
+  Data->State->CurrentFrame->SynchronousFaultData.si_code = Op->Reason.si_code;
+
+  switch (Op->Reason.Signal) {
+  case SIGILL:
+    FHU::Syscalls::tgkill(Data->State->ThreadManager.PID, Data->State->ThreadManager.TID, SIGILL);
    break;
-    case FEXCore::IR::Break_InvalidInstruction:
-      FHU::Syscalls::tgkill(Data->State->ThreadManager.PID, Data->State->ThreadManager.TID, SIGILL);
+  case SIGTRAP:
+    FHU::Syscalls::tgkill(Data->State->ThreadManager.PID, Data->State->ThreadManager.TID, SIGTRAP);
+    break;
+  case SIGSEGV:
+    FHU::Syscalls::tgkill(Data->State->ThreadManager.PID, Data->State->ThreadManager.TID, SIGSEGV);
+    break;
+  default:
+    FHU::Syscalls::tgkill(Data->State->ThreadManager.PID, Data->State->ThreadManager.TID, SIGTRAP);
    break;
-  default: LOGMAN_MSG_A_FMT("Unknown Break Reason: {}", Op->Reason); break;
  }
 }

@@ -85,7 +93,7 @@ DEF_OP(GetRoundingMode) {

 DEF_OP(SetRoundingMode) {
  auto Op = IROp->C<IR::IROp_SetRoundingMode>();
-  uint8_t GuestRounding = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[0]);
+  const auto GuestRounding = *GetSrc<uint8_t*>(Data->SSAData, Op->RoundMode);
 #ifdef _M_ARM_64
  uint64_t HostRounding{};
  __asm volatile(R"(
@@ -125,16 +133,16 @@ DEF_OP(SetRoundingMode) {

 DEF_OP(Print) {
  auto Op = IROp->C<IR::IROp_Print>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;

  if (OpSize <= 8) {
-    uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
+    const auto Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
    LogMan::Msg::IFmt(">>>> Value in Arg: 0x{:x}, {}", Src, Src);
  }
  else if (OpSize == 16) {
-    __uint128_t Src = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
-    uint64_t Src0 = Src;
-    uint64_t Src1 = Src >> 64;
+    const auto Src = *GetSrc<__uint128_t*>(Data->SSAData, Op->Value);
+    const uint64_t Src0 = Src;
+    const uint64_t Src1 = Src >> 64;
    LogMan::Msg::IFmt(">>>> Value[0] in Arg: 0x{:x}, {}", Src0, Src0);
    LogMan::Msg::IFmt("     Value[1] in Arg: 0x{:x}, {}", Src1, Src1);
  }
@@ -148,6 +156,19 @@ DEF_OP(ProcessorID) {
  GD = (CPUNode << 12) | CPU;
 }

+DEF_OP(RDRAND) {
+  // We are ignoring Op->GetReseeded in the interpreter
+  uint64_t *DstPtr = GetDest<uint64_t*>(Data->SSAData, Node);
+  ssize_t Result = ::getrandom(&DstPtr[0], 8, 0);
+
+  // Second result is if we managed to read a valid random number or not
+  DstPtr[1] = Result == 8 ? 1 : 0;
+}
+
+DEF_OP(Yield) {
+  // Nop implementation
+}
+
 #undef DEF_OP

 } // namespace FEXCore::CPU
@@ -14,27 +14,20 @@ namespace FEXCore::CPU {
 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
 DEF_OP(ExtractElementPair) {
  auto Op = IROp->C<IR::IROp_ExtractElementPair>();
-  uintptr_t Src = GetSrc<uintptr_t>(Data->SSAData, Op->Header.Args[0]);
+  const auto Src = GetSrc<uintptr_t>(Data->SSAData, Op->Pair);
  memcpy(GDP,
    reinterpret_cast<void*>(Src + Op->Header.Size * Op->Element), Op->Header.Size);
 }

 DEF_OP(CreateElementPair) {
  auto Op = IROp->C<IR::IROp_CreateElementPair>();
-  void *Src_Lower = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  void *Src_Upper = GetSrc<void*>(Data->SSAData, Op->Header.Args[1]);
+  const void *Src_Lower = GetSrc<void*>(Data->SSAData, Op->Lower);
+  const void *Src_Upper = GetSrc<void*>(Data->SSAData, Op->Upper);

  uint8_t *Dst = GetDest<uint8_t*>(Data->SSAData, Node);

-  memcpy(Dst, Src_Lower, Op->Header.Size);
-  memcpy(Dst + Op->Header.Size, Src_Upper, Op->Header.Size);
-}
-
-DEF_OP(Mov) {
-  auto Op = IROp->C<IR::IROp_Mov>();
-  uint8_t OpSize = IROp->Size;
-
-  memcpy(GDP, GetSrc<void*>(Data->SSAData, Op->Header.Args[0]), OpSize);
+  memcpy(Dst, Src_Lower, IROp->ElementSize);
+  memcpy(Dst + IROp->ElementSize, Src_Upper, IROp->ElementSize);
 }

 #undef DEF_OP
@@ -0,0 +1,131 @@
+/*
+$info$
+tags: backend|arm64
+desc: relocation logic of the arm64 splatter backend
+$end_info$
+*/
+#include "Interface/Context/Context.h"
+#include "Interface/Core/JIT/Arm64/JITClass.h"
+#include "Interface/HLE/Thunks/Thunks.h"
+
+namespace FEXCore::CPU {
+
+uint64_t Arm64JITCore::GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
+  switch (Op) {
+    case FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol::SYMBOL_LITERAL_EXITFUNCTION_LINKER:
+      return ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker;
+    break;
+    default:
+      ERROR_AND_DIE_FMT("Unknown named symbol literal: {}", static_cast<uint32_t>(Op));
+    break;
+  }
+  return ~0ULL;
+}
+
+void Arm64JITCore::InsertNamedThunkRelocation(ARMEmitter::Register Reg, const IR::SHA256Sum &Sum) {
+  Relocation MoveABI{};
+  MoveABI.NamedThunkMove.Header.Type = FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE;
+  // Offset is the offset from the entrypoint of the block
+  auto CurrentCursor = GetCursorAddress<uint8_t *>();
+  MoveABI.NamedThunkMove.Offset = CurrentCursor - CodeData.BlockBegin;
+  MoveABI.NamedThunkMove.Symbol = Sum;
+  MoveABI.NamedThunkMove.RegisterIndex = Reg.Idx();
+
+  uint64_t Pointer = reinterpret_cast<uint64_t>(EmitterCTX->ThunkHandler->LookupThunk(Sum));
+
+  LoadConstant(ARMEmitter::Size::i64Bit, Reg, Pointer, EmitterCTX->Config.CacheObjectCodeCompilation());
+  Relocations.emplace_back(MoveABI);
+}
+
+Arm64JITCore::NamedSymbolLiteralPair Arm64JITCore::InsertNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
+  uint64_t Pointer = GetNamedSymbolLiteral(Op);
+
+  Arm64JITCore::NamedSymbolLiteralPair Lit {
+    .Lit = Pointer,
+    .MoveABI = {
+      .NamedSymbolLiteral = {
+        .Header = {
+          .Type = FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL,
+        },
+        .Symbol = Op,
+        .Offset = 0,
+      },
+    },
+  };
+  return Lit;
+}
+
+void Arm64JITCore::PlaceNamedSymbolLiteral(NamedSymbolLiteralPair &Lit) {
+  // Offset is the offset from the entrypoint of the block
+  auto CurrentCursor = GetCursorAddress<uint8_t *>();
+  Lit.MoveABI.NamedSymbolLiteral.Offset = CurrentCursor - CodeData.BlockBegin;
+
+  Bind(&Lit.Loc);
+  dc64(Lit.Lit);
+  Relocations.emplace_back(Lit.MoveABI);
+}
+
+void Arm64JITCore::InsertGuestRIPMove(ARMEmitter::Register Reg, uint64_t Constant) {
+  Relocation MoveABI{};
+  MoveABI.GuestRIPMove.Header.Type = FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE;
+  // Offset is the offset from the entrypoint of the block
+  auto CurrentCursor = GetCursorAddress<uint8_t *>();
+  MoveABI.GuestRIPMove.Offset = CurrentCursor - CodeData.BlockBegin;
+  MoveABI.GuestRIPMove.GuestRIP = Constant;
+  MoveABI.GuestRIPMove.RegisterIndex = Reg.Idx();
+
+  LoadConstant(ARMEmitter::Size::i64Bit, Reg, Constant, EmitterCTX->Config.CacheObjectCodeCompilation());
+  Relocations.emplace_back(MoveABI);
+}
+
+bool Arm64JITCore::ApplyRelocations(uint64_t GuestEntry, uint64_t CodeEntry, uint64_t CursorEntry, size_t NumRelocations, const char* EntryRelocations) {
+  size_t DataIndex{};
+  for (size_t j = 0; j < NumRelocations; ++j) {
+    const FEXCore::CPU::Relocation *Reloc = reinterpret_cast<const FEXCore::CPU::Relocation *>(&EntryRelocations[DataIndex]);
+    LOGMAN_THROW_AA_FMT((DataIndex % alignof(Relocation)) == 0, "Alignment of relocation wasn't adhered to");
+
+    switch (Reloc->Header.Type) {
+      case FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL: {
+        uint64_t Pointer = GetNamedSymbolLiteral(Reloc->NamedSymbolLiteral.Symbol);
+        // Relocation occurs at the cursorEntry + offset relative to that cursor
+        SetCursorOffset(CursorEntry + Reloc->NamedSymbolLiteral.Offset);
+
+        // Generate a literal so we can place it
+        dc64(Pointer);
+
+        DataIndex += sizeof(Reloc->NamedSymbolLiteral);
+        break;
+      }
+      case FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE: {
+        uint64_t Pointer = reinterpret_cast<uint64_t>(EmitterCTX->ThunkHandler->LookupThunk(Reloc->NamedThunkMove.Symbol));
+        if (Pointer == ~0ULL) {
+          return false;
+        }
+
+        // Relocation occurs at the cursorEntry + offset relative to that cursor.
+        SetCursorOffset(CursorEntry + Reloc->NamedThunkMove.Offset);
+        LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Register(Reloc->NamedThunkMove.RegisterIndex), Pointer, true);
+        DataIndex += sizeof(Reloc->NamedThunkMove);
+        break;
+      }
+      case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE: {
+        // XXX: Reenable once the JIT Object Cache is upstream
+        // XXX: Should spin the relocation list, create a list of guest RIP moves, and ask for them all once, reduces lock contention.
+        uint64_t Pointer = ~0ULL; // EmitterCTX->JITObjectCache->FindRelocatedRIP(Reloc->GuestRIPMove.GuestRIP);
+        if (Pointer == ~0ULL) {
+          return false;
+        }
+
+        // Relocation occurs at the cursorEntry + offset relative to that cursor.
+        SetCursorOffset(CursorEntry + Reloc->GuestRIPMove.Offset);
+        LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Register(Reloc->GuestRIPMove.RegisterIndex), Pointer, true);
+        DataIndex += sizeof(Reloc->GuestRIPMove);
+        break;
+      }
+    }
+  }
+
+  return true;
+}
+}
+
@@ -4,6 +4,9 @@ tags: backend|arm64
 $end_info$
 */

+#include "Interface/Context/Context.h"
+#include "FEXCore/IR/IR.h"
+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
 #include "Interface/Core/LookupCache.h"

 #include "Interface/Core/JIT/Arm64/JITClass.h"
@@ -15,49 +18,25 @@ $end_info$
 #include <Interface/HLE/Thunks/Thunks.h>

 namespace FEXCore::CPU {
-using namespace vixl;
-using namespace vixl::aarch64;
-#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
-DEF_OP(GuestCallDirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
-DEF_OP(GuestCallIndirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
-DEF_OP(GuestReturn) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
-DEF_OP(SignalReturn) {
-  // First we must reset the stack
-  ResetStack();
-
-  // Now branch to our signal return helper
-  // This can't be a direct branch since the code needs to live at a constant location
-  LoadConstant(x0, ThreadSharedData.SignalReturnInstruction);
-  br(x0);
-}
+#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header const *IROp, IR::NodeID Node)

 DEF_OP(CallbackReturn) {
-
  // spill back to CTX
-  SpillStaticRegs();
+  SpillStaticRegs(TMP1);

  // First we must reset the stack
  ResetStack();

  // We can now lower the ref counter again
-  LoadConstant(x0, reinterpret_cast<uint64_t>(ThreadSharedData.SignalHandlerRefCounterPtr));
-  ldr(w2, MemOperand(x0));
-  sub(w2, w2, 1);
-  str(w2, MemOperand(x0));
+
+  ldr(ARMEmitter::WReg::w2, STATE, offsetof(FEXCore::Core::CpuStateFrame, SignalHandlerRefCounter));
+  sub(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r2, 1);
+  str(ARMEmitter::WReg::w2, STATE, offsetof(FEXCore::Core::CpuStateFrame, SignalHandlerRefCounter));

  // We need to adjust an additional 8 bytes to get back to the original "misaligned" RSP state
-  ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])));
-  add(x2, x2, 8);
-  str(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])));
+  ldr(ARMEmitter::XReg::x2, STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP]));
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r2, 8);
+  str(ARMEmitter::XReg::x2, STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP]));

  PopCalleeSavedRegisters();

@@ -68,110 +47,111 @@ DEF_OP(CallbackReturn) {
 DEF_OP(ExitFunction) {
  auto Op = IROp->C<IR::IROp_ExitFunction>();

-  Label FullLookup;
-
  ResetStack();

-  aarch64::Register RipReg;
  uint64_t NewRIP;

  if (IsInlineConstant(Op->NewRIP, &NewRIP) || IsInlineEntrypointOffset(Op->NewRIP, &NewRIP)) {
-    Literal l_BranchHost{ThreadSharedData.Dispatcher->ExitFunctionLinkerAddress};
-    Literal l_BranchGuest{NewRIP};
+    ARMEmitter::ForwardLabel l_BranchHost;
+    ARMEmitter::ForwardLabel l_BranchGuest;

-    ldr(x0, &l_BranchHost);
-    blr(x0);
+    ldr(ARMEmitter::XReg::x0, &l_BranchHost);
+    blr(ARMEmitter::Reg::r0);
+
+    Bind(&l_BranchHost);
+    dc64(ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker);
+    Bind(&l_BranchGuest);
+    dc64(NewRIP);

-    place(&l_BranchHost);
-    place(&l_BranchGuest);
  } else {
-    RipReg = GetReg<RA_64>(Op->Header.Args[0].ID());
+
+    ARMEmitter::ForwardLabel FullLookup;
+    auto RipReg = GetReg(Op->NewRIP.ID());

    // L1 Cache
-    LoadConstant(x0, ThreadState->LookupCache->GetL1Pointer());
+    ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.L1Pointer));

-    and_(x3, RipReg, LookupCache::L1_ENTRIES_MASK);
-    add(x0, x0, Operand(x3, Shift::LSL, 4));
+    and_(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, RipReg, LookupCache::L1_ENTRIES_MASK);
+    add(ARMEmitter::XReg::x0, ARMEmitter::XReg::x0, ARMEmitter::XReg::x3, ARMEmitter::ShiftType::LSL, 4);

-    ldp(x1, x0, MemOperand(x0));
-    cmp(x0, RipReg);
-    b(&FullLookup, Condition::ne);
-    br(x1);
+    ldp<ARMEmitter::IndexType::OFFSET>(ARMEmitter::XReg::x1, ARMEmitter::XReg::x0, ARMEmitter::Reg::r0, 0);
+    cmp(ARMEmitter::XReg::x0, RipReg.X());
+    b(ARMEmitter::Condition::CC_NE, &FullLookup);
+    br(ARMEmitter::Reg::r1);

-    bind(&FullLookup);
-    LoadConstant(TMP1, ThreadSharedData.Dispatcher->AbsoluteLoopTopAddress);
-    str(RipReg, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip)));
+    Bind(&FullLookup);
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.DispatcherLoopTop));
+    str(RipReg.X(), STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip));
    br(TMP1);
  }
 }

 DEF_OP(Jump) {
  const auto Op = IROp->C<IR::IROp_Jump>();
-  const auto ArgID = Op->Args(0).ID();
+  const auto Target = Op->TargetBlock.ID();

-  PendingTargetLabel = &JumpTargets.try_emplace(ArgID).first->second;
+  PendingTargetLabel = &JumpTargets.try_emplace(Target).first->second;
 }

-#define GRCMP(Node) (Op->CompareSize == 4 ? GetReg<RA_32>(Node) : GetReg<RA_64>(Node))
-#define GRFCMP(Node) (Op->CompareSize == 4 ? GetDst(Node).S() : GetDst(Node).D())
-
-static Condition MapBranchCC(IR::CondClassType Cond) {
+static ARMEmitter::Condition MapBranchCC(IR::CondClassType Cond) {
  switch (Cond.Val) {
-  case FEXCore::IR::COND_EQ: return Condition::eq;
-  case FEXCore::IR::COND_NEQ: return Condition::ne;
-  case FEXCore::IR::COND_SGE: return Condition::ge;
-  case FEXCore::IR::COND_SLT: return Condition::lt;
-  case FEXCore::IR::COND_SGT: return Condition::gt;
-  case FEXCore::IR::COND_SLE: return Condition::le;
-  case FEXCore::IR::COND_UGE: return Condition::cs;
-  case FEXCore::IR::COND_ULT: return Condition::cc;
-  case FEXCore::IR::COND_UGT: return Condition::hi;
-  case FEXCore::IR::COND_ULE: return Condition::ls;
-  case FEXCore::IR::COND_FLU: return Condition::lt;
-  case FEXCore::IR::COND_FGE: return Condition::ge;
-  case FEXCore::IR::COND_FLEU:return Condition::le;
-  case FEXCore::IR::COND_FGT: return Condition::gt;
-  case FEXCore::IR::COND_FU:  return Condition::vs;
-  case FEXCore::IR::COND_FNU: return Condition::vc;
+  case FEXCore::IR::COND_EQ:  return ARMEmitter::Condition::CC_EQ;
+  case FEXCore::IR::COND_NEQ: return ARMEmitter::Condition::CC_NE;
+  case FEXCore::IR::COND_SGE: return ARMEmitter::Condition::CC_GE;
+  case FEXCore::IR::COND_SLT: return ARMEmitter::Condition::CC_LT;
+  case FEXCore::IR::COND_SGT: return ARMEmitter::Condition::CC_GT;
+  case FEXCore::IR::COND_SLE: return ARMEmitter::Condition::CC_LE;
+  case FEXCore::IR::COND_UGE: return ARMEmitter::Condition::CC_CS;
+  case FEXCore::IR::COND_ULT: return ARMEmitter::Condition::CC_CC;
+  case FEXCore::IR::COND_UGT: return ARMEmitter::Condition::CC_HI;
+  case FEXCore::IR::COND_ULE: return ARMEmitter::Condition::CC_LS;
+  case FEXCore::IR::COND_FLU: return ARMEmitter::Condition::CC_LT;
+  case FEXCore::IR::COND_FGE: return ARMEmitter::Condition::CC_GE;
+  case FEXCore::IR::COND_FLEU:return ARMEmitter::Condition::CC_LE;
+  case FEXCore::IR::COND_FGT: return ARMEmitter::Condition::CC_GT;
+  case FEXCore::IR::COND_FU:  return ARMEmitter::Condition::CC_VS;
+  case FEXCore::IR::COND_FNU: return ARMEmitter::Condition::CC_VC;
  case FEXCore::IR::COND_VS:
  case FEXCore::IR::COND_VC:
  case FEXCore::IR::COND_MI:
  case FEXCore::IR::COND_PL:
  default:
  LOGMAN_MSG_A_FMT("Unsupported compare type");
-  return Condition::nv;
+  return ARMEmitter::Condition::CC_NV;
  }
 }

-
 DEF_OP(CondJump) {
  auto Op = IROp->C<IR::IROp_CondJump>();

-  Label *TrueTargetLabel = &JumpTargets.try_emplace(Op->TrueBlock.ID()).first->second;
+  auto TrueTargetLabel = &JumpTargets.try_emplace(Op->TrueBlock.ID()).first->second;

  uint64_t Const;
  const bool isConst = IsInlineConstant(Op->Cmp2, &Const);

+  const auto Size = Op->CompareSize == 4 ? ARMEmitter::Size::i32Bit : ARMEmitter::Size::i64Bit;
+  const auto SubSize = ARMEmitter::ToVectorSizePair(Op->CompareSize == 4 ? ARMEmitter::SubRegSize::i32Bit : ARMEmitter::SubRegSize::i64Bit);
+
  if (isConst && Const == 0 && Op->Cond.Val == FEXCore::IR::COND_EQ) {
    LOGMAN_THROW_A_FMT(IsGPR(Op->Cmp1.ID()), "CondJump: Expected GPR");
-    cbz(GRCMP(Op->Cmp1.ID()), TrueTargetLabel);
+    cbz(Size, GetReg(Op->Cmp1.ID()), TrueTargetLabel);
  } else if (isConst && Const == 0 && Op->Cond.Val == FEXCore::IR::COND_NEQ) {
    LOGMAN_THROW_A_FMT(IsGPR(Op->Cmp1.ID()), "CondJump: Expected GPR");
-    cbnz(GRCMP(Op->Cmp1.ID()), TrueTargetLabel);
+    cbnz(Size, GetReg(Op->Cmp1.ID()), TrueTargetLabel);
  } else {
    if (IsGPR(Op->Cmp1.ID())) {
      if (isConst) {
-        cmp(GRCMP(Op->Cmp1.ID()), Const);
+        cmp(Size, GetReg(Op->Cmp1.ID()), Const);
      } else {
-        cmp(GRCMP(Op->Cmp1.ID()), GRCMP(Op->Cmp2.ID()));
+        cmp(Size, GetReg(Op->Cmp1.ID()), GetReg(Op->Cmp2.ID()));
      }
    } else if (IsFPR(Op->Cmp1.ID())) {
-      fcmp(GRFCMP(Op->Cmp1.ID()), GRFCMP(Op->Cmp2.ID()));
+      fcmp(SubSize.Scalar, GetVReg(Op->Cmp1.ID()), GetVReg(Op->Cmp2.ID()));
    } else {
      LOGMAN_MSG_A_FMT("CondJump: Expected GPR or FPR");
    }

-    b(TrueTargetLabel, MapBranchCC(Op->Cond));
+    b(MapBranchCC(Op->Cond), TrueTargetLabel);
  }

  PendingTargetLabel = &JumpTargets.try_emplace(Op->FalseBlock.ID()).first->second;
@@ -184,32 +164,64 @@ DEF_OP(Syscall) {
  // X1: ThreadState
  // X2: Pointer to SyscallArguments

-  PushDynamicRegsAndLR();
-  SpillStaticRegs();
+  FEXCore::IR::SyscallFlags Flags = Op->Flags;
+  PushDynamicRegsAndLR(TMP1);

-  uint64_t SPOffset = AlignUp(FEXCore::HLE::SyscallArguments::MAX_ARGS * 8, 16);
-  sub(sp, sp, SPOffset);
-  for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS; ++i) {
-    if (Op->Header.Args[i].IsInvalid()) continue;
-    str(GetReg<RA_64>(Op->Header.Args[i].ID()), MemOperand(sp, i * 8));
+  uint32_t GPRSpillMask = ~0U;
+  uint32_t FPRSpillMask = ~0U;
+  if ((Flags & FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY) == FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY) {
+    // Need to spill all caller saved registers still
+    GPRSpillMask = CALLER_GPR_MASK;
+    FPRSpillMask = CALLER_FPR_MASK;
  }

-  LoadConstant(x0, reinterpret_cast<uint64_t>(CTX->SyscallHandler));
-  mov(x1, STATE);
-  mov(x2, sp);
+  SpillStaticRegs(TMP1, true, GPRSpillMask, FPRSpillMask);

-  LoadConstant(x3, reinterpret_cast<uint64_t>(FEXCore::Context::HandleSyscall));
-  blr(x3);
+  // Now that we are spilled, store in the state that we are in a syscall
+  // Still without overwriting registers that matter
+  // 16bit LoadConstant to be a single instruction
+  // This gives the signal handler a value to check to see if we are in a syscall at all
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, GPRSpillMask & 0xFFFF);
+  str(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));

-  add(sp, sp, SPOffset);
+  uint64_t SPOffset = AlignUp(FEXCore::HLE::SyscallArguments::MAX_ARGS * 8, 16);
+  sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, SPOffset);
+  for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS; ++i) {
+    if (Op->Header.Args[i].IsInvalid()) continue;
+    str(GetReg(Op->Header.Args[i].ID()).X(), ARMEmitter::Reg::rsp, i * 8);
+  }

-  // Result is now in x0
-  // Fix the stack and any values that were stepped on
-  FillStaticRegs();
-  PopDynamicRegsAndLR();
+  ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SyscallHandlerObj));
+  ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SyscallHandlerFunc));
+  mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, STATE.R());

-  // Move result to its destination register
-  mov(GetReg<RA_64>(Node), x0);
+  // SP supporting move
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::rsp, 0);
+#ifdef VIXL_SIMULATOR
+  GenerateIndirectRuntimeCall<uint64_t, void*, void*, void*>(ARMEmitter::Reg::r3);
+#else
+  blr(ARMEmitter::Reg::r3);
+#endif
+
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, SPOffset);
+
+  if ((Flags & FEXCore::IR::SyscallFlags::NORETURN) != FEXCore::IR::SyscallFlags::NORETURN) {
+    // Result is now in x0
+    // Fix the stack and any values that were stepped on
+    FillStaticRegs(true, GPRSpillMask, FPRSpillMask);
+
+    // Now the registers we've spilled are back in their original host registers
+    // We can safely claim we are no longer in a syscall
+    str(ARMEmitter::XReg::zr, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));
+
+    PopDynamicRegsAndLR();
+
+    if ((Flags & FEXCore::IR::SyscallFlags::NORETURNEDRESULT) != FEXCore::IR::SyscallFlags::NORETURNEDRESULT) {
+      // Move result to its destination register.
+      // Only if `NORETURNEDRESULT` wasn't set, otherwise we might overwrite the CPUState refilled with `FillStaticRegs`
+      mov(ARMEmitter::Size::i64Bit, GetReg(Node), ARMEmitter::Reg::r0);
+    }
+  }
 }

 DEF_OP(InlineSyscall) {
@@ -225,101 +237,61 @@ DEF_OP(InlineSyscall) {
  // X6: Arg6 - Doesn't exist in x86-64 land. RA INTERSECT

  // One argument is removed from the SyscallArguments::MAX_ARGS since the first argument was syscall number
-  const static std::array<vixl::aarch64::Register, FEXCore::HLE::SyscallArguments::MAX_ARGS-1> RegArgs = {{
-    x0, x1, x2, x3, x4, x5
+  const static std::array<ARMEmitter::XRegister, FEXCore::HLE::SyscallArguments::MAX_ARGS-1> RegArgs = {{
+    ARMEmitter::XReg::x0, ARMEmitter::XReg::x1, ARMEmitter::XReg::x2, ARMEmitter::XReg::x3, ARMEmitter::XReg::x4, ARMEmitter::XReg::x5
  }};

  bool Intersects{};
  // We always need to spill x8 since we can't know if it is live at this SSA location
  uint32_t SpillMask = 1U << 8;
-  std::vector<vixl::aarch64::Register> IntersectRegs(FEXCore::HLE::SyscallArguments::MAX_ARGS);
  for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS-1; ++i) {
    if (Op->Header.Args[i].IsInvalid()) break;

-    auto Reg = GetReg<RA_64>(Op->Header.Args[i].ID());
-    if (Reg.GetCode() == x8.GetCode() ||
-        Reg.GetCode() == x4.GetCode() ||
-        Reg.GetCode() == x5.GetCode()) {
+    auto Reg = GetReg(Op->Header.Args[i].ID());
+    if (Reg == ARMEmitter::Reg::r8 ||
+        Reg == ARMEmitter::Reg::r4 ||
+        Reg == ARMEmitter::Reg::r5) {

-      SpillMask |= (1U << Reg.GetCode());
+      SpillMask |= (1U << Reg.Idx());
      Intersects = true;
    }
  }
-  // XXX: For some reason spilling only the x4, x5, and x8 registers was causing issues
-  // Come back to this once investigation reveals why it fails the gvisor ioctl test
-  // For now override to all GPRs
-  SpillMask = ~0U;

  // Ordering is incredibly important here
  // We must spill any overlapping registers first THEN claim we are in a syscall without invalidating state at all
  // Only spill the registers that intersect with our usage
-  SpillStaticRegs(false, SpillMask);
+  SpillStaticRegs(TMP1, false, SpillMask);

  // Now that we are spilled, store in the state that we are in a syscall
  // Still without overwriting registers that matter
  // 16bit LoadConstant to be a single instruction
  // We must always spill at least one register (x8) so this value always has a bit set
  // This gives the signal handler a value to check to see if we are in a syscall at all
-  LoadConstant(x0, SpillMask & 0xFFFF);
-  str(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo)));
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, SpillMask & 0xFFFF);
+  str(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));

  // Now that we have claimed to be a syscall we can set up the arguments
+  const auto EmitSize = CTX->Config.Is64BitMode() ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
+  const auto EmitSubSize = CTX->Config.Is64BitMode() ? ARMEmitter::SubRegSize::i64Bit : ARMEmitter::SubRegSize::i32Bit;
  if (Intersects) {
    for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS-1; ++i) {
      if (Op->Header.Args[i].IsInvalid()) break;

-      if (CTX->Config.Is64BitMode()) {
-        auto Reg = GetReg<RA_64>(Op->Header.Args[i].ID());
-        // In the case of intersection with x4, x5, or x8 then these are currently SRA
-        // for registers RAX, RBX, and RSI. Which have just been spilled
-        // Just load back from the context. Could be slightly smarter but this is fairly uncommon
-        if (Reg.GetCode() == x8.GetCode()) {
-          ldr(RegArgs[i], MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSI])));
-        }
-        else if (Reg.GetCode() == x4.GetCode()) {
-          ldr(RegArgs[i], MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RAX])));
-        }
-        else if (Reg.GetCode() == x5.GetCode()) {
-          ldr(RegArgs[i], MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RBX])));
-        }
+      auto Reg = GetReg(Op->Header.Args[i].ID());
+      // In the case of intersection with x4, x5, or x8 then these are currently SRA
+      // for registers RAX, RBX, and RSI. Which have just been spilled
+      // Just load back from the context. Could be slightly smarter but this is fairly uncommon
+      if (Reg == ARMEmitter::Reg::r8) {
+        ldr(EmitSubSize, RegArgs[i].R(), STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSI]));
      }
-    }
-
-    for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS-1; ++i) {
-      if (Op->Header.Args[i].IsInvalid()) break;
-
-      if (CTX->Config.Is64BitMode()) {
-        auto Reg = GetReg<RA_64>(Op->Header.Args[i].ID());
-        // In the case of intersection with x4, x5, or x8 then these are currently SRA
-        // for registers RAX, RBX, and RSI. Which have just been spilled
-        // Just load back from the context. Could be slightly smarter but this is fairly uncommon
-        if (Reg.GetCode() == x8.GetCode()) {
-          ldr(RegArgs[i], MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSI])));
-        }
-        else if (Reg.GetCode() == x4.GetCode()) {
-          ldr(RegArgs[i], MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RAX])));
-        }
-        else if (Reg.GetCode() == x5.GetCode()) {
-          ldr(RegArgs[i], MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RBX])));
-        }
-        else {
-          mov(RegArgs[i], Reg);
-        }
+      else if (Reg == ARMEmitter::Reg::r4) {
+        ldr(EmitSubSize, RegArgs[i].R(), STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RAX]));
+      }
+      else if (Reg == ARMEmitter::Reg::r5) {
+        ldr(EmitSubSize, RegArgs[i].R(), STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RBX]));
      }
      else {
-        auto Reg = GetReg<RA_32>(Op->Header.Args[i].ID());
-        if (Reg.GetCode() == w8.GetCode()) {
-          ldr(RegArgs[i].W(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSI])));
-        }
-        else if (Reg.GetCode() == w4.GetCode()) {
-          ldr(RegArgs[i].W(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RAX])));
-        }
-        else if (Reg.GetCode() == w5.GetCode()) {
-          ldr(RegArgs[i].W(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RBX])));
-        }
-        else {
-          uxtw(RegArgs[i].W(), Reg);
-        }
+        mov(EmitSize, RegArgs[i].R(), Reg);
      }
    }
  }
@@ -327,34 +299,26 @@ DEF_OP(InlineSyscall) {
    for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS-1; ++i) {
      if (Op->Header.Args[i].IsInvalid()) break;

-      if (CTX->Config.Is64BitMode()) {
-        mov(RegArgs[i], GetReg<RA_64>(Op->Header.Args[i].ID()));
-      }
-      else {
-        uxtw(RegArgs[i], GetReg<RA_64>(Op->Header.Args[i].ID()));
-      }
+      mov(EmitSize, RegArgs[i].R(), GetReg(Op->Header.Args[i].ID()));
    }
  }

-  LoadConstant(x8, Op->HostSyscallNumber);
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r8, Op->HostSyscallNumber);
  svc(0);
  // On updated signal mask we can receive a signal RIGHT HERE

-  // Now that we are done in the syscall we need to carefully peel back the state
-  // First unspill the registers from before
-  FillStaticRegs(false, SpillMask);
+  if ((Op->Flags & FEXCore::IR::SyscallFlags::NORETURN) != FEXCore::IR::SyscallFlags::NORETURN) {
+    // Now that we are done in the syscall we need to carefully peel back the state
+    // First unspill the registers from before
+    FillStaticRegs(false, SpillMask);

-  // Now the registers we've spilled are back in their original host registers
-  // We can safely claim we are no longer in a syscall
-  str(xzr, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo)));
+    // Now the registers we've spilled are back in their original host registers
+    // We can safely claim we are no longer in a syscall
+    str(ARMEmitter::XReg::zr, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));

-  // Result is now in x0
-  // Move result to its destination register
-  if (CTX->Config.Is64BitMode()) {
-    mov(GetReg<RA_64>(Node), x0);
-  }
-  else {
-    uxtw(GetReg<RA_64>(Node), x0);
+    // Result is now in x0
+    // Move result to its destination register
+    mov(EmitSize, GetReg(Node), ARMEmitter::Reg::r0);
  }
 }

@@ -364,15 +328,19 @@ DEF_OP(Thunk) {
  // X0: CTX
  // X1: Args (from guest stack)

-  SpillStaticRegs(); // spill to ctx before ra64 spill
+  SpillStaticRegs(TMP1); // spill to ctx before ra64 spill

-  PushDynamicRegsAndLR();
+  PushDynamicRegsAndLR(TMP1);

-  mov(x0, GetReg<RA_64>(Op->Header.Args[0].ID()));
+  mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, GetReg(Op->ArgPtr.ID()));

-  auto thunkFn = ThreadState->CTX->ThunkHandler->LookupThunk(Op->ThunkNameHash);
-  LoadConstant(x2, (uintptr_t)thunkFn);
-  blr(x2);
+  auto thunkFn = static_cast<Context::ContextImpl*>(ThreadState->CTX)->ThunkHandler->LookupThunk(Op->ThunkNameHash);
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, (uintptr_t)thunkFn);
+#ifdef VIXL_SIMULATOR
+  GenerateIndirectRuntimeCall<void, void*, void*>(ARMEmitter::Reg::r2);
+#else
+  blr(ARMEmitter::Reg::r2);
+#endif

  PopDynamicRegsAndLR();

@@ -385,61 +353,67 @@ DEF_OP(ValidateCode) {
  int len = Op->CodeLength;
  int idx = 0;

-  LoadConstant(GetReg<RA_64>(Node), 0);
-  LoadConstant(x0, Entry + Op->Offset);
-  LoadConstant(x1, 1);
+  LoadConstant(ARMEmitter::Size::i64Bit, GetReg(Node), 0);
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, Entry + Op->Offset);
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, 1);
+
+  const auto Dst = GetReg(Node);

  while (len >= 8)
  {
-    ldr(x2, MemOperand(x0, idx));
-    LoadConstant(x3, *(const uint32_t *)(OldCode + idx));
-    cmp(x2, x3);
-    csel(GetReg<RA_64>(Node), GetReg<RA_64>(Node), x1, Condition::eq);
+    ldr(ARMEmitter::XReg::x2, ARMEmitter::Reg::r0, idx);
+    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, *(const uint32_t *)(OldCode + idx));
+    cmp(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r3);
+    csel(ARMEmitter::Size::i64Bit, Dst, Dst, ARMEmitter::Reg::r1, ARMEmitter::Condition::CC_EQ);
    len -= 8;
    idx += 8;
  }
  while (len >= 4)
  {
-    ldr(w2, MemOperand(x0, idx));
-    LoadConstant(w3, *(const uint32_t *)(OldCode + idx));
-    cmp(w2, w3);
-    csel(GetReg<RA_64>(Node), GetReg<RA_64>(Node), x1, Condition::eq);
+    ldr(ARMEmitter::WReg::w2, ARMEmitter::Reg::r0, idx);
+    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, *(const uint32_t *)(OldCode + idx));
+    cmp(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r3);
+    csel(ARMEmitter::Size::i64Bit, Dst, Dst, ARMEmitter::Reg::r1, ARMEmitter::Condition::CC_EQ);
    len -= 4;
    idx += 4;
  }
  while (len >= 2)
  {
-    ldrh(w2, MemOperand(x0, idx));
-    LoadConstant(w3, *(const uint16_t *)(OldCode + idx));
-    cmp(w2, w3);
-    csel(GetReg<RA_64>(Node), GetReg<RA_64>(Node), x1, Condition::eq);
+    ldrh(ARMEmitter::Reg::r2, ARMEmitter::Reg::r0, idx);
+    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, *(const uint16_t *)(OldCode + idx));
+    cmp(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r3);
+    csel(ARMEmitter::Size::i64Bit, Dst, Dst, ARMEmitter::Reg::r1, ARMEmitter::Condition::CC_EQ);
    len -= 2;
    idx += 2;
  }
  while (len >= 1)
  {
-    ldrb(w2, MemOperand(x0, idx));
-    LoadConstant(w3, *(const uint8_t *)(OldCode + idx));
-    cmp(w2, w3);
-    csel(GetReg<RA_64>(Node), GetReg<RA_64>(Node), x1, Condition::eq);
+    ldrb(ARMEmitter::Reg::r2, ARMEmitter::Reg::r0, idx);
+    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, *(const uint8_t *)(OldCode + idx));
+    cmp(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r3);
+    csel(ARMEmitter::Size::i64Bit, Dst, Dst, ARMEmitter::Reg::r1, ARMEmitter::Condition::CC_EQ);
    len -= 1;
    idx += 1;
  }
 }

-DEF_OP(RemoveCodeEntry) {
+DEF_OP(ThreadRemoveCodeEntry) {
  // Arguments are passed as follows:
  // X0: Thread
  // X1: RIP

-  PushDynamicRegsAndLR();
+  PushDynamicRegsAndLR(TMP1);
+  SpillStaticRegs(TMP1);

-  mov(x0, STATE);
-  LoadConstant(x1, Entry);
+  mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, STATE.R());
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, Entry);

-  LoadConstant(x2, reinterpret_cast<uintptr_t>(&Context::Context::RemoveCodeEntryFromJit));
-  SpillStaticRegs();
-  blr(x2);
+  ldr(ARMEmitter::XReg::x2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.ThreadRemoveCodeEntryFromJIT));
+#ifdef VIXL_SIMULATOR
+  GenerateIndirectRuntimeCall<void, void*, void*>(ARMEmitter::Reg::r2);
+#else
+  blr(ARMEmitter::Reg::r2);
+#endif
  FillStaticRegs();

  // Fix the stack and any values that were stepped on
@@ -449,55 +423,61 @@ DEF_OP(RemoveCodeEntry) {
 DEF_OP(CPUID) {
  auto Op = IROp->C<IR::IROp_CPUID>();

-  PushDynamicRegsAndLR();
+  PushDynamicRegsAndLR(TMP1);
+  SpillStaticRegs(TMP1);

  // x0 = CPUID Handler
  // x1 = CPUID Function
  // x2 = CPUID Leaf
-  LoadConstant(x0, reinterpret_cast<uint64_t>(&CTX->CPUID));
-  mov(x1, GetReg<RA_64>(Op->Header.Args[0].ID()));
-  mov(x2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+  ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDObj));
+  ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDFunction));
+  mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, GetReg(Op->Function.ID()));
+  mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, GetReg(Op->Leaf.ID()));
+#ifdef VIXL_SIMULATOR
+  GenerateIndirectRuntimeCall<__uint128_t, void*, uint64_t, uint64_t>(ARMEmitter::Reg::r3);
+#else
+  blr(ARMEmitter::Reg::r3);
+#endif

-  using ClassPtrType = FEXCore::CPUID::FunctionResults (FEXCore::CPUIDEmu::*)(uint32_t, uint32_t);
-  union PtrCast {
-    ClassPtrType ClassPtr;
-    uintptr_t Data;
-  };
-
-  PtrCast Ptr;
-  Ptr.ClassPtr = &FEXCore::CPUIDEmu::RunFunction;
-  LoadConstant(x3, Ptr.Data);
-  SpillStaticRegs();
-  blr(x3);
  FillStaticRegs();

  PopDynamicRegsAndLR();

  // Results are in x0, x1
  // Results want to be in a i64v2 vector
-  auto Dst = GetSrcPair<RA_64>(Node);
-  mov(Dst.first,  x0);
-  mov(Dst.second, x1);
+  auto Dst = GetRegPair(Node);
+  mov(ARMEmitter::Size::i64Bit, Dst.first,  ARMEmitter::Reg::r0);
+  mov(ARMEmitter::Size::i64Bit, Dst.second, ARMEmitter::Reg::r1);
+}
+
+DEF_OP(XGETBV) {
+  auto Op = IROp->C<IR::IROp_XGetBV>();
+
+  PushDynamicRegsAndLR(TMP1);
+  SpillStaticRegs(TMP1);
+
+  // x0 = CPUID Handler
+  // x1 = XCR Function
+  ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDObj));
+  ldr(ARMEmitter::XReg::x2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.XCRFunction));
+  mov(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r1, GetReg(Op->Function.ID()));
+#ifdef VIXL_SIMULATOR
+  GenerateIndirectRuntimeCall<uint64_t, void*, uint32_t>(ARMEmitter::Reg::r2);
+#else
+  blr(ARMEmitter::Reg::r2);
+#endif
+
+  FillStaticRegs();
+
+  PopDynamicRegsAndLR();
+
+  // Results are in x0
+  // Results want to be in a i32v2 vector
+  auto Dst = GetRegPair(Node);
+  mov(ARMEmitter::Size::i32Bit, Dst.first,  ARMEmitter::Reg::r0);
+  lsr(ARMEmitter::Size::i64Bit, Dst.second, ARMEmitter::Reg::r0, 32);
 }

 #undef DEF_OP
-void Arm64JITCore::RegisterBranchHandlers() {
-#define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &Arm64JITCore::Op_##x
-  REGISTER_OP(GUESTCALLDIRECT,   GuestCallDirect);
-  REGISTER_OP(GUESTCALLINDIRECT, GuestCallIndirect);
-  REGISTER_OP(GUESTRETURN,       GuestReturn);
-  REGISTER_OP(SIGNALRETURN,      SignalReturn);
-  REGISTER_OP(CALLBACKRETURN,    CallbackReturn);
-  REGISTER_OP(EXITFUNCTION,      ExitFunction);
-  REGISTER_OP(JUMP,              Jump);
-  REGISTER_OP(CONDJUMP,          CondJump);
-  REGISTER_OP(SYSCALL,           Syscall);
-  REGISTER_OP(INLINESYSCALL,     InlineSyscall);
-  REGISTER_OP(THUNK,             Thunk);
-  REGISTER_OP(VALIDATECODE,      ValidateCode);
-  REGISTER_OP(REMOVECODEENTRY,   RemoveCodeEntry);
-  REGISTER_OP(CPUID,             CPUID);
-#undef REGISTER_OP
-}
 }

@@ -4,91 +4,183 @@ tags: backend|arm64
 $end_info$
 */

+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
 #include "Interface/Core/JIT/Arm64/JITClass.h"

 namespace FEXCore::CPU {
-
-using namespace vixl;
-using namespace vixl::aarch64;
-#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
+#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header const *IROp, IR::NodeID Node)
 DEF_OP(VInsGPR) {
-  auto Op = IROp->C<IR::IROp_VInsGPR>();
-  mov(GetDst(Node), GetSrc(Op->Header.Args[0].ID()));
-  switch (Op->Header.ElementSize) {
-    case 1: {
-      ins(GetDst(Node).V16B(), Op->Index, GetReg<RA_32>(Op->Header.Args[1].ID()));
-    break;
+  const auto Op = IROp->C<IR::IROp_VInsGPR>();
+  const auto OpSize = IROp->Size;
+
+  const auto DestIdx = Op->DestIdx;
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+
+  LOGMAN_THROW_AA_FMT(ElementSize == 8 || ElementSize == 4 || ElementSize == 2 || ElementSize == 1, "Unexpected {} size", __func__);
+  const auto SubEmitSize = ElementSize == 8 ? ARMEmitter::SubRegSize::i64Bit :
+    ElementSize == 4 ? ARMEmitter::SubRegSize::i32Bit :
+    ElementSize == 2 ? ARMEmitter::SubRegSize::i16Bit :
+    ElementSize == 1 ? ARMEmitter::SubRegSize::i8Bit : ARMEmitter::SubRegSize::i8Bit;
+  const auto ElementsPer128Bit = 16 / ElementSize;
+
+  const auto Dst = GetVReg(Node);
+  const auto DestVector = GetVReg(Op->DestVector.ID());
+  const auto Src = GetReg(Op->Src.ID());
+
+  if (HostSupportsSVE && Is256Bit) {
+    const auto ElementSizeBits = ElementSize * 8;
+    const auto Offset = ElementSizeBits * DestIdx;
+
+    const auto SSEBitSize = Core::CPUState::XMM_SSE_REG_SIZE * 8;
+    const auto InUpperLane = Offset >= SSEBitSize;
+
+    // This is going to be a little gross. Pls forgive me.
+    // Since SVE has the whole vector length agnostic programming
+    // thing going on, we can't exactly freely insert entries into
+    // arbitrary locations in the vector.
+    //
+    // SVE *does* have INSR, however this only shifts the entire
+    // vector to the left by an element size and inserts a value
+    // at the beginning of the vector. Not *quite* what we need.
+    // (though INSR *is* very useful for other things).
+    //
+    // The idea is (in the case of the upper lane), move the upper
+    // lane down, insert into it and recombine with the lower lane.
+    //
+    // In the case of the lower lane, insert and then recombine with
+    // the upper lane.
+
+    if (InUpperLane) {
+      // Move the upper lane down for the insertion.
+      const auto CompactPred = ARMEmitter::PReg::p0;
+      not_(CompactPred, PRED_TMP_32B.Zeroing(), PRED_TMP_16B);
+      compact(ARMEmitter::SubRegSize::i64Bit, VTMP1.Z(), CompactPred, DestVector.Z());
    }
-    case 2: {
-      ins(GetDst(Node).V8H(), Op->Index, GetReg<RA_32>(Op->Header.Args[1].ID()));
-    break;
+
+    // Put data in place for destructive SPLICE below.
+    mov(Dst.Z(), DestVector.Z());
+
+    // Inserts the GPR value into the given V register.
+    // Also automatically adjusts the index in the case of using the
+    // moved upper lane.
+    const auto Insert = [&](const FEXCore::ARMEmitter::VRegister& reg, int index) {
+      if (InUpperLane) {
+        index -= ElementsPer128Bit;
+      }
+      ins(SubEmitSize, reg, index, Src);
+    };
+
+    if (InUpperLane) {
+      Insert(VTMP1, DestIdx);
+      splice<ARMEmitter::OpType::Destructive>(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), PRED_TMP_16B, Dst.Z(), VTMP1.Z());
+    } else {
+      Insert(Dst, DestIdx);
+      splice<ARMEmitter::OpType::Destructive>(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), PRED_TMP_16B, Dst.Z(), DestVector.Z());
    }
-    case 4: {
-      ins(GetDst(Node).V4S(), Op->Index, GetReg<RA_32>(Op->Header.Args[1].ID()));
-    break;
-    }
-    case 8: {
-      ins(GetDst(Node).V2D(), Op->Index, GetReg<RA_64>(Op->Header.Args[1].ID()));
-    break;
-    }
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+  } else {
+    mov(Dst.Q(), DestVector.Q());
+    ins(SubEmitSize, Dst, DestIdx, Src);
  }
 }

 DEF_OP(VCastFromGPR) {
  auto Op = IROp->C<IR::IROp_VCastFromGPR>();
+  auto Dst = GetVReg(Node);
+  auto Src = GetReg(Op->Src.ID());
+
  switch (Op->Header.ElementSize) {
    case 1:
-      uxtb(TMP1.W(), GetReg<RA_32>(Op->Header.Args[0].ID()));
-      fmov(GetDst(Node).S(), TMP1.W());
+      uxtb(ARMEmitter::Size::i32Bit, TMP1, Src);
+      fmov(ARMEmitter::Size::i32Bit, Dst.S(), TMP1);
      break;
    case 2:
-      uxth(TMP1.W(), GetReg<RA_32>(Op->Header.Args[0].ID()));
-      fmov(GetDst(Node).S(), TMP1.W());
+      uxth(ARMEmitter::Size::i32Bit, TMP1, Src);
+      fmov(ARMEmitter::Size::i32Bit, Dst.S(), TMP1);
      break;
    case 4:
-      fmov(GetDst(Node).S(), GetReg<RA_32>(Op->Header.Args[0].ID()).W());
+      fmov(ARMEmitter::Size::i32Bit, Dst.S(), Src);
      break;
    case 8:
-      fmov(GetDst(Node).D(), GetReg<RA_64>(Op->Header.Args[0].ID()).X());
+      fmov(ARMEmitter::Size::i64Bit, Dst.D(), Src);
      break;
    default: LOGMAN_MSG_A_FMT("Unknown castGPR element size: {}", Op->Header.ElementSize);
  }
 }

+DEF_OP(VDupFromGPR) {
+  const auto Op = IROp->C<IR::IROp_VDupFromGPR>();
+  const auto OpSize = IROp->Size;
+
+  const auto Dst = GetVReg(Node);
+  const auto Src = GetReg(Op->Src.ID());
+
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+  const auto ElementSize = IROp->ElementSize;
+
+  LOGMAN_THROW_AA_FMT(ElementSize == 8 || ElementSize == 4 || ElementSize == 2 || ElementSize == 1,
+                      "Unexpected {} element size: {}", __func__, ElementSize);
+
+  const auto SubEmitSize =
+    ElementSize == 8 ? ARMEmitter::SubRegSize::i64Bit :
+    ElementSize == 4 ? ARMEmitter::SubRegSize::i32Bit :
+    ElementSize == 2 ? ARMEmitter::SubRegSize::i16Bit :
+    ElementSize == 1 ? ARMEmitter::SubRegSize::i8Bit : ARMEmitter::SubRegSize::i8Bit;
+
+  if (HostSupportsSVE && Is256Bit) {
+    dup(SubEmitSize, Dst.Z(), Src);
+  } else {
+    dup(SubEmitSize, Dst.Q(), Src);
+  }
+}
+
 DEF_OP(Float_FromGPR_S) {
-  auto Op = IROp->C<IR::IROp_Float_FromGPR_S>();
-  uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const auto Op = IROp->C<IR::IROp_Float_FromGPR_S>();
+
+  const uint16_t ElementSize = Op->Header.ElementSize;
+  const uint16_t Conv = (ElementSize << 8) | Op->SrcElementSize;
+
+  auto Dst = GetVReg(Node);
+  auto Src = GetReg(Op->Src.ID());
+
  switch (Conv) {
    case 0x0404: { // Float <- int32_t
-      scvtf(GetDst(Node).S(), GetReg<RA_32>(Op->Header.Args[0].ID()));
+      scvtf(ARMEmitter::Size::i32Bit, Dst.S(), Src);
      break;
    }
    case 0x0408: { // Float <- int64_t
-      scvtf(GetDst(Node).S(), GetReg<RA_64>(Op->Header.Args[0].ID()));
+      scvtf(ARMEmitter::Size::i64Bit, Dst.S(), Src);
      break;
    }
    case 0x0804: { // Double <- int32_t
-      scvtf(GetDst(Node).D(), GetReg<RA_32>(Op->Header.Args[0].ID()));
+      scvtf(ARMEmitter::Size::i32Bit, Dst.D(), Src);
      break;
    }
    case 0x0808: { // Double <- int64_t
-      scvtf(GetDst(Node).D(), GetReg<RA_64>(Op->Header.Args[0].ID()));
+      scvtf(ARMEmitter::Size::i64Bit, Dst.D(), Src);
      break;
    }
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled conversion mask: Mask=0x{:04x}, ElementSize={}, SrcElementSize={}",
+                       Conv, ElementSize, Op->SrcElementSize);
+      break;
  }
 }

 DEF_OP(Float_FToF) {
  auto Op = IROp->C<IR::IROp_Float_FToF>();
-  uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+
+  auto Dst = GetVReg(Node);
+  auto Src = GetVReg(Op->Scalar.ID());
+
  switch (Conv) {
    case 0x0804: { // Double <- Float
-      fcvt(GetDst(Node).D(), GetSrc(Op->Header.Args[0].ID()).S());
+      fcvt(Dst.D(), Src.S());
      break;
    }
    case 0x0408: { // Float <- Double
-      fcvt(GetDst(Node).S(), GetSrc(Op->Header.Args[0].ID()).D());
+      fcvt(Dst.S(), Src.D());
      break;
    }
    default: LOGMAN_MSG_A_FMT("Unknown FCVT sizes: 0x{:x}", Conv);
@@ -96,132 +188,208 @@ DEF_OP(Float_FToF) {
 }

 DEF_OP(Vector_SToF) {
-  auto Op = IROp->C<IR::IROp_Vector_SToF>();
-  switch (Op->Header.ElementSize) {
-    case 4:
-      scvtf(GetDst(Node).V4S(), GetSrc(Op->Header.Args[0].ID()).V4S());
-    break;
-    case 8:
-      scvtf(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2D());
-    break;
-    default: LOGMAN_MSG_A_FMT("Unknown Vector_SToF element size: {}", Op->Header.ElementSize);
+  const auto Op = IROp->C<IR::IROp_Vector_SToF>();
+  const auto OpSize = IROp->Size;
+
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+
+  LOGMAN_THROW_AA_FMT(ElementSize == 8 || ElementSize == 4 || ElementSize == 2, "Unexpected {} size", __func__);
+  const auto SubEmitSize = ElementSize == 8 ? ARMEmitter::SubRegSize::i64Bit :
+    ElementSize == 4 ? ARMEmitter::SubRegSize::i32Bit :
+    ElementSize == 2 ? ARMEmitter::SubRegSize::i16Bit : ARMEmitter::SubRegSize::i16Bit;
+
+  const auto Dst = GetVReg(Node);
+  const auto Vector = GetVReg(Op->Vector.ID());
+  if (HostSupportsSVE && Is256Bit) {
+    const auto Mask = PRED_TMP_32B;
+    scvtf(Dst.Z(), SubEmitSize, Mask.Merging(), Vector.Z(), SubEmitSize);
+  } else {
+    scvtf(SubEmitSize, Dst.Q(), Vector.Q());
  }
 }

 DEF_OP(Vector_FToZS) {
-  auto Op = IROp->C<IR::IROp_Vector_FToZS>();
-  switch (Op->Header.ElementSize) {
-    case 4:
-      fcvtzs(GetDst(Node).V4S(), GetSrc(Op->Header.Args[0].ID()).V4S());
-    break;
-    case 8:
-      fcvtzs(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2D());
-    break;
-    default: LOGMAN_MSG_A_FMT("Unknown Vector_FToZS element size: {}", Op->Header.ElementSize);
+  const auto Op = IROp->C<IR::IROp_Vector_FToZS>();
+  const auto OpSize = IROp->Size;
+
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+
+  LOGMAN_THROW_AA_FMT(ElementSize == 8 || ElementSize == 4 || ElementSize == 2, "Unexpected {} size", __func__);
+  const auto SubEmitSize = ElementSize == 8 ? ARMEmitter::SubRegSize::i64Bit :
+    ElementSize == 4 ? ARMEmitter::SubRegSize::i32Bit :
+    ElementSize == 2 ? ARMEmitter::SubRegSize::i16Bit : ARMEmitter::SubRegSize::i16Bit;
+
+  const auto Dst = GetVReg(Node);
+  const auto Vector = GetVReg(Op->Vector.ID());
+  if (HostSupportsSVE && Is256Bit) {
+    const auto Mask = PRED_TMP_32B;
+    fcvtzs(Dst.Z(), SubEmitSize, Mask.Merging(), Vector.Z(), SubEmitSize);
+  } else {
+    fcvtzs(SubEmitSize, Dst.Q(), Vector.Q());
  }
 }

 DEF_OP(Vector_FToS) {
-  auto Op = IROp->C<IR::IROp_Vector_FToS>();
-  switch (Op->Header.ElementSize) {
-    case 4:
-      frinti(GetDst(Node).V4S(), GetSrc(Op->Header.Args[0].ID()).V4S());
-      fcvtzs(GetDst(Node).V4S(), GetDst(Node).V4S());
-    break;
-    case 8:
-      frinti(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2D());
-      fcvtzs(GetDst(Node).V2D(), GetDst(Node).V2D());
-    break;
-    default: LOGMAN_MSG_A_FMT("Unknown Vector_FToS element size: {}", Op->Header.ElementSize);
+  const auto Op = IROp->C<IR::IROp_Vector_FToS>();
+  const auto OpSize = IROp->Size;
+
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+
+  LOGMAN_THROW_AA_FMT(ElementSize == 8 || ElementSize == 4 || ElementSize == 2, "Unexpected {} size", __func__);
+  const auto SubEmitSize = ElementSize == 8 ? ARMEmitter::SubRegSize::i64Bit :
+    ElementSize == 4 ? ARMEmitter::SubRegSize::i32Bit :
+    ElementSize == 2 ? ARMEmitter::SubRegSize::i16Bit : ARMEmitter::SubRegSize::i16Bit;
+
+  const auto Dst = GetVReg(Node);
+  const auto Vector = GetVReg(Op->Vector.ID());
+
+  if (HostSupportsSVE && Is256Bit) {
+    const auto Mask = PRED_TMP_32B;
+    frinti(SubEmitSize, Dst.Z(), Mask.Merging(), Vector.Z());
+    fcvtzs(Dst.Z(), SubEmitSize, Mask.Merging(), Dst.Z(), SubEmitSize);
+  } else {
+    const auto Dst = GetVReg(Node);
+    const auto Vector = GetVReg(Op->Vector.ID());
+    frinti(SubEmitSize, Dst.Q(), Vector.Q());
+    fcvtzs(SubEmitSize, Dst.Q(), Dst.Q());
  }
 }

 DEF_OP(Vector_FToF) {
-  auto Op = IROp->C<IR::IROp_Vector_FToF>();
-  uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const auto Op = IROp->C<IR::IROp_Vector_FToF>();
+  const auto OpSize = IROp->Size;

-  switch (Conv) {
-    case 0x0804: { // Double <- Float
-      fcvtl(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2S());
-      break;
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+  const auto Conv = (ElementSize << 8) | Op->SrcElementSize;
+
+  LOGMAN_THROW_AA_FMT(ElementSize == 8 || ElementSize == 4 || ElementSize == 2, "Unexpected {} size", __func__);
+  const auto SubEmitSize = ElementSize == 8 ? ARMEmitter::SubRegSize::i64Bit :
+    ElementSize == 4 ? ARMEmitter::SubRegSize::i32Bit :
+    ElementSize == 2 ? ARMEmitter::SubRegSize::i16Bit : ARMEmitter::SubRegSize::i16Bit;
+
+  const auto Dst = GetVReg(Node);
+  const auto Vector = GetVReg(Op->Vector.ID());
+
+  if (HostSupportsSVE && Is256Bit) {
+    // Curiously, FCVTLT and FCVTNT have no bottom variants,
+    // and also interesting is that FCVTLT will iterate the
+    // source vector by accessing each odd element and storing
+    // them consecutively in the destination.
+    //
+    // FCVTNT is somewhat like the opposite. It will read each
+    // consecutive element, but store each result into every odd
+    // element in the destination vector.
+    //
+    // We need to undo the behavior of FCVTNT with UZP2. In the case
+    // of FCVTLT, we instead need to set the vector up with ZIP1, so
+    // that the elements will be processed correctly.
+
+    const auto Mask = PRED_TMP_32B.Merging();
+
+    switch (Conv) {
+      case 0x0402: { // Float <- Half
+        zip1(FEXCore::ARMEmitter::SubRegSize::i16Bit, Dst.Z(), Vector.Z(), Vector.Z());
+        fcvtlt(FEXCore::ARMEmitter::SubRegSize::i32Bit, Dst.Z(), Mask, Dst.Z());
+        break;
+      }
+      case 0x0804: { // Double <- Float
+        zip1(FEXCore::ARMEmitter::SubRegSize::i32Bit, Dst.Z(), Vector.Z(), Vector.Z());
+        fcvtlt(FEXCore::ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Dst.Z());
+        break;
+      }
+      case 0x0204: { // Half <- Float
+        fcvtnt(FEXCore::ARMEmitter::SubRegSize::i16Bit, Dst.Z(), Mask, Vector.Z());
+        uzp2(FEXCore::ARMEmitter::SubRegSize::i16Bit, Dst.Z(), Dst.Z(), Dst.Z());
+        break;
+      }
+      case 0x0408: { // Float <- Double
+        fcvtnt(FEXCore::ARMEmitter::SubRegSize::i32Bit, Dst.Z(), Mask, Vector.Z());
+        uzp2(FEXCore::ARMEmitter::SubRegSize::i32Bit, Dst.Z(), Dst.Z(), Dst.Z());
+        break;
+      }
+      default:
+        LOGMAN_MSG_A_FMT("Unknown Vector_FToF Type : 0x{:04x}", Conv);
+        break;
    }
-    case 0x0408: { // Float <- Double
-      fcvtn(GetDst(Node).V2S(), GetSrc(Op->Header.Args[0].ID()).V2D());
-      break;
+  } else {
+    switch (Conv) {
+      case 0x0402: // Float <- Half
+      case 0x0804: { // Double <- Float
+        fcvtl(SubEmitSize, Dst.D(), Vector.D());
+        break;
+      }
+      case 0x0204: // Half <- Float
+      case 0x0408: { // Float <- Double
+        fcvtn(SubEmitSize, Dst.D(), Vector.D());
+        break;
+      }
+      default:
+        LOGMAN_MSG_A_FMT("Unknown Vector_FToF Type : 0x{:04x}", Conv);
+        break;
    }
-    default: LOGMAN_MSG_A_FMT("Unknown Vector_FToF Type : 0x{:04x}", Conv); break;
  }
 }

 DEF_OP(Vector_FToI) {
-  auto Op = IROp->C<IR::IROp_Vector_FToI>();
-  switch (Op->Round) {
-    case FEXCore::IR::Round_Nearest.Val:
-      switch (Op->Header.ElementSize) {
-        case 4:
-          frintn(GetDst(Node).V4S(), GetSrc(Op->Header.Args[0].ID()).V4S());
+  const auto Op = IROp->C<IR::IROp_Vector_FToI>();
+  const auto OpSize = IROp->Size;
+
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+  LOGMAN_THROW_AA_FMT(ElementSize == 8 || ElementSize == 4 || ElementSize == 2, "Unexpected {} size", __func__);
+
+  const auto SubEmitSize = ElementSize == 8 ? ARMEmitter::SubRegSize::i64Bit :
+    ElementSize == 4 ? ARMEmitter::SubRegSize::i32Bit :
+    ElementSize == 2 ? ARMEmitter::SubRegSize::i16Bit : ARMEmitter::SubRegSize::i16Bit;
+
+  const auto Dst = GetVReg(Node);
+  const auto Vector = GetVReg(Op->Vector.ID());
+
+  if (HostSupportsSVE && Is256Bit) {
+    const auto Mask = PRED_TMP_32B.Merging();
+
+    switch (Op->Round) {
+      case FEXCore::IR::Round_Nearest.Val:
+        frintn(SubEmitSize, Dst.Z(), Mask, Vector.Z());
        break;
-        case 8:
-          frintn(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2D());
+      case FEXCore::IR::Round_Negative_Infinity.Val:
+        frintm(SubEmitSize, Dst.Z(), Mask, Vector.Z());
        break;
-      }
-    break;
-    case FEXCore::IR::Round_Negative_Infinity.Val:
-      switch (Op->Header.ElementSize) {
-        case 4:
-          frintm(GetDst(Node).V4S(), GetSrc(Op->Header.Args[0].ID()).V4S());
+      case FEXCore::IR::Round_Positive_Infinity.Val:
+        frintp(SubEmitSize, Dst.Z(), Mask, Vector.Z());
        break;
-        case 8:
-          frintm(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2D());
+      case FEXCore::IR::Round_Towards_Zero.Val:
+        frintz(SubEmitSize, Dst.Z(), Mask, Vector.Z());
        break;
-      }
-    break;
-    case FEXCore::IR::Round_Positive_Infinity.Val:
-      switch (Op->Header.ElementSize) {
-        case 4:
-          frintp(GetDst(Node).V4S(), GetSrc(Op->Header.Args[0].ID()).V4S());
+      case FEXCore::IR::Round_Host.Val:
+        frinti(SubEmitSize, Dst.Z(), Mask, Vector.Z());
        break;
-        case 8:
-          frintp(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2D());
+    }
+  } else {
+    switch (Op->Round) {
+      case FEXCore::IR::Round_Nearest.Val:
+        frinti(SubEmitSize, Dst.Q(), Vector.Q());
        break;
-      }
-    break;
-    case FEXCore::IR::Round_Towards_Zero.Val:
-      switch (Op->Header.ElementSize) {
-        case 4:
-          frintz(GetDst(Node).V4S(), GetSrc(Op->Header.Args[0].ID()).V4S());
+      case FEXCore::IR::Round_Negative_Infinity.Val:
+        frintm(SubEmitSize, Dst.Q(), Vector.Q());
        break;
-        case 8:
-          frintz(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2D());
+      case FEXCore::IR::Round_Positive_Infinity.Val:
+        frintp(SubEmitSize, Dst.Q(), Vector.Q());
        break;
-      }
-    break;
-    case FEXCore::IR::Round_Host.Val:
-      switch (Op->Header.ElementSize) {
-        case 4:
-          frinti(GetDst(Node).V4S(), GetSrc(Op->Header.Args[0].ID()).V4S());
+      case FEXCore::IR::Round_Towards_Zero.Val:
+        frintz(SubEmitSize, Dst.Q(), Vector.Q());
        break;
-        case 8:
-          frinti(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].ID()).V2D());
+      case FEXCore::IR::Round_Host.Val:
+        frinti(SubEmitSize, Dst.Q(), Vector.Q());
        break;
-      }
-    break;
+    }
  }
 }

 #undef DEF_OP
-void Arm64JITCore::RegisterConversionHandlers() {
-#define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &Arm64JITCore::Op_##x
-  REGISTER_OP(VINSGPR,         VInsGPR);
-  REGISTER_OP(VCASTFROMGPR,    VCastFromGPR);
-  REGISTER_OP(FLOAT_FROMGPR_S, Float_FromGPR_S);
-  REGISTER_OP(FLOAT_FTOF,      Float_FToF);
-  REGISTER_OP(VECTOR_STOF,     Vector_SToF);
-  REGISTER_OP(VECTOR_FTOZS,    Vector_FToZS);
-  REGISTER_OP(VECTOR_FTOS,     Vector_FToS);
-  REGISTER_OP(VECTOR_FTOF,     Vector_FToF);
-  REGISTER_OP(VECTOR_FTOI,     Vector_FToI);
-#undef REGISTER_OP
-}
 }

@@ -4,100 +4,176 @@ tags: backend|arm64
 $end_info$
 */

+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
 #include "Interface/Core/JIT/Arm64/JITClass.h"
 #include "Interface/IR/Passes/RegisterAllocationPass.h"

 namespace FEXCore::CPU {
-using namespace vixl;
-using namespace vixl::aarch64;
-#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
+#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header const *IROp, IR::NodeID Node)

 DEF_OP(AESImc) {
  auto Op = IROp->C<IR::IROp_VAESImc>();
-  aesimc(GetDst(Node).V16B(), GetSrc(Op->Header.Args[0].ID()).V16B());
+  aesimc(GetVReg(Node), GetVReg(Op->Vector.ID()));
 }

 DEF_OP(AESEnc) {
-	auto Op = IROp->C<IR::IROp_VAESEnc>();
-  eor(VTMP2.V16B(), VTMP2.V16B(), VTMP2.V16B());
-  mov(VTMP1.V16B(), GetSrc(Op->Header.Args[0].ID()).V16B());
-  aese(VTMP1.V16B(), VTMP2.V16B());
-  aesmc(VTMP1.V16B(), VTMP1.V16B());
-  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Header.Args[1].ID()).V16B());
+  const auto Op = IROp->C<IR::IROp_VAESEnc>();
+  const auto OpSize = IROp->Size;
+
+  const auto Dst = GetVReg(Node);
+  const auto Key = GetVReg(Op->Key.ID());
+  const auto State = GetVReg(Op->State.ID());
+
+  LOGMAN_THROW_AA_FMT(OpSize == Core::CPUState::XMM_SSE_REG_SIZE,
+                      "Currently only supports 128-bit operations.");
+
+  eor(VTMP2.Q(), VTMP2.Q(), VTMP2.Q());
+  mov(VTMP1.Q(), State.Q());
+  aese(VTMP1, VTMP2);
+  aesmc(VTMP1, VTMP1);
+  eor(Dst.Q(), VTMP1.Q(), Key.Q());
 }

 DEF_OP(AESEncLast) {
-	auto Op = IROp->C<IR::IROp_VAESEncLast>();
-  eor(VTMP2.V16B(), VTMP2.V16B(), VTMP2.V16B());
-  mov(VTMP1.V16B(), GetSrc(Op->Header.Args[0].ID()).V16B());
-  aese(VTMP1.V16B(), VTMP2.V16B());
-  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Header.Args[1].ID()).V16B());
+  const auto Op = IROp->C<IR::IROp_VAESEncLast>();
+  const auto OpSize = IROp->Size;
+
+  const auto Dst = GetVReg(Node);
+  const auto Key = GetVReg(Op->Key.ID());
+  const auto State = GetVReg(Op->State.ID());
+
+  LOGMAN_THROW_AA_FMT(OpSize == Core::CPUState::XMM_SSE_REG_SIZE,
+                      "Currently only supports 128-bit operations.");
+
+  eor(VTMP2.Q(), VTMP2.Q(), VTMP2.Q());
+  mov(VTMP1.Q(), State.Q());
+  aese(VTMP1, VTMP2);
+  eor(Dst.Q(), VTMP1.Q(), Key.Q());
 }

 DEF_OP(AESDec) {
-	auto Op = IROp->C<IR::IROp_VAESDec>();
-  eor(VTMP2.V16B(), VTMP2.V16B(), VTMP2.V16B());
-  mov(VTMP1.V16B(), GetSrc(Op->Header.Args[0].ID()).V16B());
-  aesd(VTMP1.V16B(), VTMP2.V16B());
-  aesimc(VTMP1.V16B(), VTMP1.V16B());
-  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Header.Args[1].ID()).V16B());
+  const auto Op = IROp->C<IR::IROp_VAESDec>();
+  const auto OpSize = IROp->Size;
+
+  const auto Dst = GetVReg(Node);
+  const auto Key = GetVReg(Op->Key.ID());
+  const auto State = GetVReg(Op->State.ID());
+
+  LOGMAN_THROW_AA_FMT(OpSize == Core::CPUState::XMM_SSE_REG_SIZE,
+                      "Currently only supports 128-bit operations.");
+
+  eor(VTMP2.Q(), VTMP2.Q(), VTMP2.Q());
+  mov(VTMP1.Q(), State.Q());
+  aesd(VTMP1, VTMP2);
+  aesimc(VTMP1, VTMP1);
+  eor(Dst.Q(), VTMP1.Q(), Key.Q());
 }

 DEF_OP(AESDecLast) {
-	auto Op = IROp->C<IR::IROp_VAESDecLast>();
-  eor(VTMP2.V16B(), VTMP2.V16B(), VTMP2.V16B());
-  mov(VTMP1.V16B(), GetSrc(Op->Header.Args[0].ID()).V16B());
-  aesd(VTMP1.V16B(), VTMP2.V16B());
-  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Header.Args[1].ID()).V16B());
+  const auto Op = IROp->C<IR::IROp_VAESDecLast>();
+  const auto OpSize = IROp->Size;
+
+  const auto Dst = GetVReg(Node);
+  const auto Key = GetVReg(Op->Key.ID());
+  const auto State = GetVReg(Op->State.ID());
+
+  LOGMAN_THROW_AA_FMT(OpSize == Core::CPUState::XMM_SSE_REG_SIZE,
+                      "Currently only supports 128-bit operations.");
+
+  eor(VTMP2.Q(), VTMP2.Q(), VTMP2.Q());
+  mov(VTMP1.Q(), State.Q());
+  aesd(VTMP1, VTMP2);
+  eor(Dst.Q(), VTMP1.Q(), Key.Q());
 }

 DEF_OP(AESKeyGenAssist) {
-	auto Op = IROp->C<IR::IROp_VAESKeyGenAssist>();
+  auto Op = IROp->C<IR::IROp_VAESKeyGenAssist>();

-  aarch64::Label Constant;
-  aarch64::Label PastConstant;
+  ARMEmitter::ForwardLabel Constant;
+  ARMEmitter::ForwardLabel PastConstant;

  // Do a "regular" AESE step
-  eor(VTMP2.V16B(), VTMP2.V16B(), VTMP2.V16B());
-  mov(VTMP1.V16B(), GetSrc(Op->Header.Args[0].ID()).V16B());
-  aese(VTMP1.V16B(), VTMP2.V16B());
+  eor(VTMP2.Q(), VTMP2.Q(), VTMP2.Q());
+  mov(VTMP1.Q(), GetVReg(Op->Src.ID()).Q());
+  aese(VTMP1, VTMP2);

  // Do a table shuffle to undo ShiftRows
-  adr(TMP1.X(), &Constant);
-  ldr(VTMP3, MemOperand(TMP1.X()));
+  ldr(VTMP3.Q(), &Constant);

  // Now EOR in the RCON
  if (Op->RCON) {
-    tbl(VTMP1.V16B(), VTMP1.V16B(), VTMP3.V16B());
+    tbl(VTMP1.Q(), VTMP1.Q(), VTMP3.Q());

-    LoadConstant(TMP1.W(), Op->RCON);
-    ins(VTMP2.V4S(), 1, TMP1.W());
-    ins(VTMP2.V4S(), 3, TMP1.W());
-    eor(GetDst(Node).V16B(), VTMP1.V16B(), VTMP2.V16B());
+    LoadConstant(ARMEmitter::Size::i64Bit, TMP1, static_cast<uint64_t>(Op->RCON) << 32);
+    dup(ARMEmitter::SubRegSize::i64Bit, VTMP2.Q(), TMP1);
+    eor(GetVReg(Node).Q(), VTMP1.Q(), VTMP2.Q());
  }
  else {
-    tbl(GetDst(Node).V16B(), VTMP1.V16B(), VTMP3.V16B());
+    tbl(GetVReg(Node).Q(), VTMP1.Q(), VTMP3.Q());
  }

  b(&PastConstant);
-  bind(&Constant);
-  dc32(0x0B0E0104);
-  dc32(0x040B0E01);
-  dc32(0x0306090C);
-  dc32(0x0C030609);
-  bind(&PastConstant);
+  Bind(&Constant);
+  dc64(0x040B0E01'0B0E0104ULL);
+  dc64(0x0C030609'0306090CULL);
+  Bind(&PastConstant);
+}
+
+DEF_OP(CRC32) {
+  auto Op = IROp->C<IR::IROp_CRC32>();
+
+  const auto Dst = GetReg(Node);
+  const auto Src1 = GetReg(Op->Src1.ID());
+  const auto Src2 = GetReg(Op->Src2.ID());
+
+  switch (Op->SrcSize) {
+    case 1:
+      crc32cb(Dst.W(), Src1.W(), Src2.W());
+      break;
+    case 2:
+      crc32ch(Dst.W(), Src1.W(), Src2.W());
+      break;
+    case 4:
+      crc32cw(Dst.W(), Src1.W(), Src2.W());
+      break;
+    case 8:
+      crc32cx(Dst.X(), Src1.X(), Src2.X());
+      break;
+    default: LOGMAN_MSG_A_FMT("Unknown CRC32 size: {}", Op->SrcSize);
+  }
+}
+
+DEF_OP(PCLMUL) {
+  const auto Op = IROp->C<IR::IROp_PCLMUL>();
+  const auto OpSize = IROp->Size;
+
+  const auto Dst  = GetVReg(Node);
+  const auto Src1 = GetVReg(Op->Src1.ID());
+  const auto Src2 = GetVReg(Op->Src2.ID());
+
+  LOGMAN_THROW_AA_FMT(OpSize == Core::CPUState::XMM_SSE_REG_SIZE,
+                      "Currently only supports 128-bit operations.");
+
+  switch (Op->Selector) {
+  case 0b00000000:
+    pmull(ARMEmitter::SubRegSize::i128Bit, Dst.D(), Src1.D(), Src2.D());
+    break;
+  case 0b00000001:
+    dup(ARMEmitter::SubRegSize::i64Bit, VTMP1.Q(), Src1.Q(), 1);
+    pmull(ARMEmitter::SubRegSize::i128Bit, Dst.D(), VTMP1.D(), Src2.D());
+    break;
+  case 0b00010000:
+    dup(ARMEmitter::SubRegSize::i64Bit, VTMP1.Q(), Src2.Q(), 1);
+    pmull(ARMEmitter::SubRegSize::i128Bit, Dst.D(), VTMP1.D(), Src1.D());
+    break;
+  case 0b00010001:
+    pmull2(ARMEmitter::SubRegSize::i128Bit, Dst.Q(), Src1.Q(), Src2.Q());
+    break;
+  default:
+    LOGMAN_MSG_A_FMT("Unknown PCLMUL selector: {}", Op->Selector);
+    break;
+  }
 }

 #undef DEF_OP
-void Arm64JITCore::RegisterEncryptionHandlers() {
-#define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &Arm64JITCore::Op_##x
-  REGISTER_OP(VAESIMC,     AESImc);
-  REGISTER_OP(VAESENC,     AESEnc);
-  REGISTER_OP(VAESENCLAST, AESEncLast);
-  REGISTER_OP(VAESDEC,     AESDec);
-  REGISTER_OP(VAESDECLAST, AESDecLast);
-  REGISTER_OP(VAESKEYGENASSIST, AESKeyGenAssist);
-
-#undef REGISTER_OP
-}
 }
@@ -7,20 +7,12 @@ $end_info$
 #include "Interface/Core/JIT/Arm64/JITClass.h"

 namespace FEXCore::CPU {
-
-using namespace vixl;
-using namespace vixl::aarch64;
-#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
+#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header const *IROp, IR::NodeID Node)
 DEF_OP(GetHostFlag) {
  auto Op = IROp->C<IR::IROp_GetHostFlag>();
-  ubfx(GetReg<RA_64>(Node), GetReg<RA_64>(Op->Header.Args[0].ID()), Op->Flag, 1);
+  ubfx(ARMEmitter::Size::i64Bit, GetReg(Node), GetReg(Op->Value.ID()), Op->Flag, 1);
 }

 #undef DEF_OP
-void Arm64JITCore::RegisterFlagHandlers() {
-#define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &Arm64JITCore::Op_##x
-  REGISTER_OP(GETHOSTFLAG, GetHostFlag);
-#undef REGISTER_OP
-}
 }

@@ -7,52 +7,42 @@ $end_info$
 #pragma once

 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"

-#include "aarch64/assembler-aarch64.h"
-#include "aarch64/disasm-aarch64.h"
-#include "aarch64/assembler-aarch64.h"
+#include <aarch64/assembler-aarch64.h>
+#include <aarch64/disasm-aarch64.h>

+#include <FEXCore/Core/CoreState.h>
 #include <FEXCore/Core/CPUBackend.h>
 #include <FEXCore/IR/IR.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
+#include <FEXCore/IR/RegisterAllocationData.h>
+#include <FEXCore/fextl/map.h>
+#include <FEXCore/fextl/string.h>
+#include <FEXCore/fextl/vector.h>

-#define STATE x28
-#define TMP1 x0
-#define TMP2 x1
-#define TMP3 x2
-#define TMP4 x3
-
-#define VTMP1 v1
-#define VTMP2 v2
-#define VTMP3 v3
+#include <array>
+#include <cstdint>
+#include <utility>

 namespace FEXCore::Core {
  struct InternalThreadState;
 }

 namespace FEXCore::CPU {
-using namespace vixl;
-using namespace vixl::aarch64;
-
 class Arm64JITCore final : public CPUBackend, public Arm64Emitter  {
 public:
-  struct CodeBuffer {
-    uint8_t *Ptr;
-    size_t Size;
-  };
-
-  explicit Arm64JITCore(FEXCore::Context::Context *ctx,
-                        FEXCore::Core::InternalThreadState *Thread,
-                        bool CompileThread);
+  explicit Arm64JITCore(FEXCore::Context::ContextImpl *ctx,
+                        FEXCore::Core::InternalThreadState *Thread);
  ~Arm64JITCore() override;

-  [[nodiscard]] std::string GetName() override { return "JIT"; }
+  [[nodiscard]] fextl::string GetName() override { return "JIT"; }

-  [[nodiscard]] void *CompileCode(uint64_t Entry,
+  [[nodiscard]] CPUBackend::CompiledCode CompileCode(uint64_t Entry,
                                  FEXCore::IR::IRListView const *IR,
                                  FEXCore::Core::DebugData *DebugData,
-                                  FEXCore::IR::RegisterAllocationData *RAData) override;
+                                  FEXCore::IR::RegisterAllocationData *RAData, bool GDBEnabled) override;

  [[nodiscard]] void *MapRegion(void* HostPtr, uint64_t, uint64_t) override { return HostPtr; }

@@ -60,79 +50,87 @@ public:

  void ClearCache() override;

-  static constexpr size_t INITIAL_CODE_SIZE = 1024 * 1024 * 16;
-  [[nodiscard]] CodeBuffer AllocateNewCodeBuffer(size_t Size);
-
-  void CopyNecessaryDataForCompileThread(CPUBackend *Original) override;
-  bool IsAddressInJITCode(uint64_t Address, bool IncludeDispatcher = true, bool IncludeCompileService = true) const override {
-    return Dispatcher->IsAddressInJITCode(Address, IncludeDispatcher, IncludeCompileService);
-  }
+  void ClearRelocations() override { Relocations.clear(); }

 private:
  FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
+  const bool HostSupportsSVE{};

-  std::unique_ptr<FEXCore::CPU::Dispatcher> Dispatcher;
-  Label *PendingTargetLabel;
-  FEXCore::Context::Context *CTX;
-  FEXCore::Core::InternalThreadState *ThreadState;
+  ARMEmitter::BiDirectionalLabel *PendingTargetLabel;
+  FEXCore::Context::ContextImpl *CTX;
  FEXCore::IR::IRListView const *IR;
  uint64_t Entry;
+  CPUBackend::CompiledCode CodeData{};

-  std::map<IR::NodeID, aarch64::Label> JumpTargets;
+  fextl::map<IR::NodeID, ARMEmitter::BiDirectionalLabel> JumpTargets;

-  /**
-   * @name Register Allocation
-   * @{ */
-  constexpr static uint32_t NumGPRs = RA64.size();
-  constexpr static uint32_t NumFPRs = RAFPR.size();
-  constexpr static uint32_t NumGPRPairs = RA64Pair.size();
-  constexpr static uint32_t NumCalleeGPRs = 10;
-  constexpr static uint32_t NumCalleeGPRPairs = 5;
-  constexpr static uint32_t RegisterCount = NumGPRs + NumFPRs + NumGPRPairs;
-  constexpr static uint32_t RegisterClasses = 6;
+  [[nodiscard]] FEXCore::ARMEmitter::Register GetReg(IR::NodeID Node) const {
+    const auto Reg = GetPhys(Node);

-  constexpr static uint64_t GPRBase = (0ULL << 32);
-  constexpr static uint64_t FPRBase = (1ULL << 32);
-  constexpr static uint64_t GPRPairBase = (2ULL << 32);
+    LOGMAN_THROW_AA_FMT(Reg.Class == IR::GPRFixedClass.Val || Reg.Class == IR::GPRClass.Val, "Unexpected Class: {}", Reg.Class);

-  /**  @} */
+    if (Reg.Class == IR::GPRFixedClass.Val) {
+      return StaticRegisters[Reg.Reg];
+    } else if (Reg.Class == IR::GPRClass.Val) {
+      return GeneralRegisters[Reg.Reg];
+    }

-  constexpr static uint8_t RA_32 = 0;
-  constexpr static uint8_t RA_64 = 1;
-  constexpr static uint8_t RA_FPR = 2;
+    FEX_UNREACHABLE;
+  }

-  template<uint8_t RAType>
-  [[nodiscard]] aarch64::Register GetReg(IR::NodeID Node) const;
+  [[nodiscard]] FEXCore::ARMEmitter::VRegister GetVReg(IR::NodeID Node) const {
+    const auto Reg = GetPhys(Node);

-  template<>
-  [[nodiscard]] aarch64::Register GetReg<RA_32>(IR::NodeID Node) const;
-  template<>
-  [[nodiscard]] aarch64::Register GetReg<RA_64>(IR::NodeID Node) const;
+    LOGMAN_THROW_AA_FMT(Reg.Class == IR::FPRFixedClass.Val || Reg.Class == IR::FPRClass.Val, "Unexpected Class: {}", Reg.Class);

-  template<uint8_t RAType>
-  [[nodiscard]] std::pair<aarch64::Register, aarch64::Register> GetSrcPair(IR::NodeID Node) const;
+    if (Reg.Class == IR::FPRFixedClass.Val) {
+      return StaticFPRegisters[Reg.Reg];
+    } else if (Reg.Class == IR::FPRClass.Val) {
+      return GeneralFPRegisters[Reg.Reg];
+    }

-  template<>
-  [[nodiscard]] std::pair<aarch64::Register, aarch64::Register> GetSrcPair<RA_32>(IR::NodeID Node) const;
-  template<>
-  [[nodiscard]] std::pair<aarch64::Register, aarch64::Register> GetSrcPair<RA_64>(IR::NodeID Node) const;
+    FEX_UNREACHABLE;
+  }

-  [[nodiscard]] aarch64::VRegister GetSrc(IR::NodeID Node) const;
-  [[nodiscard]] aarch64::VRegister GetDst(IR::NodeID Node) const;
+  [[nodiscard]] std::pair<FEXCore::ARMEmitter::Register, FEXCore::ARMEmitter::Register> GetRegPair(IR::NodeID Node) const {
+    const auto Reg = GetPhys(Node);
+
+    LOGMAN_THROW_AA_FMT(Reg.Class == IR::GPRPairClass.Val, "Unexpected Class: {}", Reg.Class);
+
+    return GeneralPairRegisters[Reg.Reg];
+  }

  [[nodiscard]] FEXCore::IR::RegisterClassType GetRegClass(IR::NodeID Node) const;

-  [[nodiscard]] IR::PhysicalRegister GetPhys(IR::NodeID Node) const;
+  [[nodiscard]] IR::PhysicalRegister GetPhys(IR::NodeID Node) const {
+    auto PhyReg = RAData->GetNodeRegister(Node);
+
+    LOGMAN_THROW_A_FMT(!PhyReg.IsInvalid(), "Couldn't Allocate register for node: ssa{}. Class: {}", Node, PhyReg.Class);
+
+    return PhyReg;
+  }

  [[nodiscard]] bool IsFPR(IR::NodeID Node) const;
  [[nodiscard]] bool IsGPR(IR::NodeID Node) const;
+  [[nodiscard]] bool IsGPRPair(IR::NodeID Node) const;

-  [[nodiscard]] MemOperand GenerateMemOperand(uint8_t AccessSize,
-                                              aarch64::Register Base,
+  [[nodiscard]] FEXCore::ARMEmitter::ExtendedMemOperand GenerateMemOperand(uint8_t AccessSize,
+                                              FEXCore::ARMEmitter::Register Base,
                                              IR::OrderedNodeWrapper Offset,
                                              IR::MemOffsetType OffsetType,
                                              uint8_t OffsetScale);

+  // NOTE: Will use TMP1 as a way to encode immediates that happen to fall outside
+  //       the limits of the scalar plus immediate variant of SVE load/stores.
+  //
+  //       TMP1 is safe to use again once this memory operand is used with its
+  //       equivalent loads or stores that this was called for.
+  [[nodiscard]] FEXCore::ARMEmitter::SVEMemOperand GenerateSVEMemOperand(uint8_t AccessSize,
+                                                    FEXCore::ARMEmitter::Register Base,
+                                                    IR::OrderedNodeWrapper Offset,
+                                                    IR::MemOffsetType OffsetType,
+                                                    uint8_t OffsetScale);
+
  [[nodiscard]] bool IsInlineConstant(const IR::OrderedNodeWrapper& Node, uint64_t* Value = nullptr) const;
  [[nodiscard]] bool IsInlineEntrypointOffset(const IR::OrderedNodeWrapper& WNode, uint64_t* Value) const;

@@ -145,60 +143,85 @@ private:
  vixl::aarch64::Decoder Decoder;
 #endif

-  void EmplaceNewCodeBuffer(CodeBuffer Buffer) {
-    CurrentCodeBuffer = &CodeBuffers.emplace_back(Buffer);
-  }
-
-  void FreeCodeBuffer(CodeBuffer Buffer);
-
-  // This is the initial code buffer that we will fall back to
-  // In a program without signals and code clearing, we will typically
-  // only have this code buffer
-  CodeBuffer InitialCodeBuffer{};
-  // This is the array of /additional/ code buffers that we may need to allocate
-  // Allocation only occurs when we've hit signals and need to clear code cache
-  // For code safety we can't delete code buffers until outside of all signals
-  std::vector<CodeBuffer> CodeBuffers{};
-
-  // This is the current code buffer that we are tracking
-  CodeBuffer *CurrentCodeBuffer{};
-
-  // We don't want to mvoe above 128MB atm because that means we will have to encode longer jumps
-  static constexpr size_t MAX_CODE_SIZE = 1024 * 1024 * 128;
-  static constexpr size_t MAX_DISPATCHER_CODE_SIZE = 4096 * 2;
-
 #if DEBUG
  vixl::aarch64::Disassembler Disasm;
 #endif

-  static uint64_t ExitFunctionLink(Arm64JITCore *core, FEXCore::Core::CpuStateFrame *Frame, uint64_t *record);
-
-  struct CompilerSharedData {
-    uint64_t SignalReturnInstruction{};
-    uint64_t UnimplementedInstructionAddress{};
-    uint64_t OverflowExceptionInstructionAddress{};
-
-    uint32_t *SignalHandlerRefCounterPtr{};
-    FEXCore::CPU::Dispatcher *Dispatcher{};
-  };
-
-  CompilerSharedData ThreadSharedData;
+  // This is purely a debugging aid for developers to see if they are in JIT code space when inspecting raw memory
+  void EmitDetectionString();
  IR::RegisterAllocationPass *RAPass;
  IR::RegisterAllocationData *RAData;
+  FEXCore::Core::DebugData *DebugData;

-  using OpHandler = void (Arm64JITCore::*)(IR::IROp_Header *IROp, IR::NodeID Node);
-  std::array<OpHandler, IR::IROps::OP_LAST + 1> OpHandlers {};
-  void RegisterALUHandlers();
-  void RegisterAtomicHandlers();
-  void RegisterBranchHandlers();
-  void RegisterConversionHandlers();
-  void RegisterFlagHandlers();
-  void RegisterMemoryHandlers();
-  void RegisterMiscHandlers();
-  void RegisterMoveHandlers();
-  void RegisterVectorHandlers();
-  void RegisterEncryptionHandlers();
-#define DEF_OP(x) void Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
+  void ResetStack();
+  /**
+   * @name Relocations
+   * @{ */
+
+    uint64_t GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op);
+
+    /**
+     * @brief A literal pair relocation object for named symbol literals
+     */
+    struct NamedSymbolLiteralPair {
+      ARMEmitter::ForwardLabel Loc;
+      uint64_t Lit;
+      Relocation MoveABI{};
+    };
+
+    /**
+     * @brief Inserts a thunk relocation
+     *
+     * @param Reg - The GPR to move the thunk handler in to
+     * @param Sum - The hash of the thunk
+     */
+    void InsertNamedThunkRelocation(ARMEmitter::Register Reg, const IR::SHA256Sum &Sum);
+
+    /**
+     * @brief Inserts a guest GPR move relocation
+     *
+     * @param Reg - The GPR to move the guest RIP in to
+     * @param Constant - The guest RIP that will be relocated
+     */
+    void InsertGuestRIPMove(ARMEmitter::Register Reg, uint64_t Constant);
+
+    /**
+     * @brief Inserts a named symbol as a literal in memory
+     *
+     * Need to use `PlaceNamedSymbolLiteral` with the return value to place the literal in the desired location
+     *
+     * @param Op The named symbol to place
+     *
+     * @return A temporary `NamedSymbolLiteralPair`
+     */
+    NamedSymbolLiteralPair InsertNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op);
+
+    /**
+     * @brief Place the named symbol literal relocation in memory
+     *
+     * @param Lit - Which literal to place
+     */
+    void PlaceNamedSymbolLiteral(NamedSymbolLiteralPair &Lit);
+
+    fextl::vector<FEXCore::CPU::Relocation> Relocations;
+
+    ///< Relocation code loading
+    bool ApplyRelocations(uint64_t GuestEntry, uint64_t CodeEntry, uint64_t CursorEntry, size_t NumRelocations, const char* EntryRelocations);
+
+  /**  @} */
+
+  uint32_t SpillSlots{};
+  using OpType = void (Arm64JITCore::*)(IR::IROp_Header const *IROp, IR::NodeID Node);
+
+  // Runtime selection;
+  // Load and store register style.
+  OpType RT_LoadRegister;
+  OpType RT_StoreRegister;
+  // Load and store TSO memory style
+  OpType RT_LoadMemTSO;
+  OpType RT_StoreMemTSO;
+
+#define DEF_OP(x) void Op_##x(IR::IROp_Header const *IROp, IR::NodeID Node)

  ///< Unhandled handler
  DEF_OP(Unhandled);
@@ -214,8 +237,10 @@ private:
  DEF_OP(InlineEntrypointOffset);
  DEF_OP(CycleCounter);
  DEF_OP(Add);
+  DEF_OP(TestNZ);
  DEF_OP(Sub);
  DEF_OP(Neg);
+  DEF_OP(Abs);
  DEF_OP(Mul);
  DEF_OP(UMul);
  DEF_OP(Div);
@@ -225,6 +250,8 @@ private:
  DEF_OP(MulH);
  DEF_OP(UMulH);
  DEF_OP(Or);
+  DEF_OP(Orlshl);
+  DEF_OP(Orlshr);
  DEF_OP(And);
  DEF_OP(Andn);
  DEF_OP(Xor);
@@ -275,10 +302,6 @@ private:
  DEF_OP(AtomicFetchNeg);

  ///< Branch ops
-  DEF_OP(GuestCallDirect);
-  DEF_OP(GuestCallIndirect);
-  DEF_OP(GuestReturn);
-  DEF_OP(SignalReturn);
  DEF_OP(CallbackReturn);
  DEF_OP(ExitFunction);
  DEF_OP(Jump);
@@ -287,12 +310,14 @@ private:
  DEF_OP(InlineSyscall);
  DEF_OP(Thunk);
  DEF_OP(ValidateCode);
-  DEF_OP(RemoveCodeEntry);
+  DEF_OP(ThreadRemoveCodeEntry);
  DEF_OP(CPUID);
+  DEF_OP(XGETBV);

  ///< Conversion ops
  DEF_OP(VInsGPR);
  DEF_OP(VCastFromGPR);
+  DEF_OP(VDupFromGPR);
  DEF_OP(Float_FromGPR_S);
  DEF_OP(Float_FToF);
  DEF_OP(Vector_SToF);
@@ -305,10 +330,12 @@ private:
  DEF_OP(GetHostFlag);

  ///< Memory ops
-    DEF_OP(LoadContext);
+  DEF_OP(LoadContext);
  DEF_OP(StoreContext);
  DEF_OP(LoadRegister);
  DEF_OP(StoreRegister);
+  DEF_OP(LoadRegisterSRA);
+  DEF_OP(StoreRegisterSRA);
  DEF_OP(LoadContextIndexed);
  DEF_OP(StoreContextIndexed);
  DEF_OP(SpillRegister);
@@ -319,15 +346,18 @@ private:
  DEF_OP(StoreMem);
  DEF_OP(LoadMemTSO);
  DEF_OP(StoreMemTSO);
+  DEF_OP(VLoadVectorMasked);
+  DEF_OP(VStoreVectorMasked);
+  DEF_OP(MemSet);
+  DEF_OP(MemCpy);
  DEF_OP(ParanoidLoadMemTSO);
  DEF_OP(ParanoidStoreMemTSO);
-  DEF_OP(VLoadMemElement);
-  DEF_OP(VStoreMemElement);
  DEF_OP(CacheLineClear);
+  DEF_OP(CacheLineClean);
  DEF_OP(CacheLineZero);

  ///< Misc ops
-  DEF_OP(EndBlock);
+  DEF_OP(GuestOpcode);
  DEF_OP(Fence);
  DEF_OP(Break);
  DEF_OP(Phi);
@@ -336,19 +366,16 @@ private:
  DEF_OP(GetRoundingMode);
  DEF_OP(SetRoundingMode);
  DEF_OP(ProcessorID);
+  DEF_OP(RDRAND);
+  DEF_OP(Yield);

  ///< Move ops
  DEF_OP(ExtractElementPair);
  DEF_OP(CreateElementPair);
-  DEF_OP(Mov);

  ///< Vector ops
  DEF_OP(VectorZero);
  DEF_OP(VectorImm);
-  DEF_OP(CreateVector2);
-  DEF_OP(CreateVector4);
-  DEF_OP(SplatVector2);
-  DEF_OP(SplatVector4);
  DEF_OP(VMov);
  DEF_OP(VAnd);
  DEF_OP(VBic);
@@ -387,6 +414,8 @@ private:
  DEF_OP(VZip2);
  DEF_OP(VUnZip);
  DEF_OP(VUnZip2);
+  DEF_OP(VTrn);
+  DEF_OP(VTrn2);
  DEF_OP(VBSL);
  DEF_OP(VCMPEQ);
  DEF_OP(VCMPEQZ);
@@ -407,18 +436,13 @@ private:
  DEF_OP(VUShrS);
  DEF_OP(VSShrS);
  DEF_OP(VInsElement);
-  DEF_OP(VInsScalarElement);
-  DEF_OP(VExtractElement);
  DEF_OP(VDupElement);
  DEF_OP(VExtr);
-  DEF_OP(VSLI);
-  DEF_OP(VSRI);
  DEF_OP(VUShrI);
  DEF_OP(VSShrI);
  DEF_OP(VShlI);
  DEF_OP(VUShrNI);
  DEF_OP(VUShrNI2);
-  DEF_OP(VBitcast);
  DEF_OP(VSXTL);
  DEF_OP(VSXTL2);
  DEF_OP(VUXTL);
@@ -433,7 +457,9 @@ private:
  DEF_OP(VUMull2);
  DEF_OP(VSMull2);
  DEF_OP(VUABDL);
+  DEF_OP(VUABDL2);
  DEF_OP(VTBL1);
+  DEF_OP(VRev64);

  ///< Encryption ops
  DEF_OP(AESImc);
@@ -442,9 +468,9 @@ private:
  DEF_OP(AESDec);
  DEF_OP(AESDecLast);
  DEF_OP(AESKeyGenAssist);
+  DEF_OP(CRC32);
+  DEF_OP(PCLMUL);
 #undef DEF_OP
 };

-
-}
-
+} // namespace FEXCore::CPU
@@ -4,32 +4,36 @@ tags: backend|arm64
 $end_info$
 */

+#ifndef _WIN32
+#include <syscall.h>
+#endif
+
+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
 #include "Interface/Core/JIT/Arm64/JITClass.h"
+#include "FEXCore/Debug/InternalThreadState.h"
+
+#include <FEXCore/Core/SignalDelegator.h>

 namespace FEXCore::CPU {
-static void PrintValue(uint64_t Value) {
-  LogMan::Msg::DFmt("Value: 0x{:x}", Value);
-}
+#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header const *IROp, IR::NodeID Node)

-static void PrintVectorValue(uint64_t Value, uint64_t ValueUpper) {
-  LogMan::Msg::DFmt("Value: 0x{:016x}'{:016x}", ValueUpper, Value);
+DEF_OP(GuestOpcode) {
+  auto Op = IROp->C<IR::IROp_GuestOpcode>();
+  // metadata
+  DebugData->GuestOpcodes.push_back({Op->GuestEntryOffset, GetCursorAddress<uint8_t*>() - CodeData.BlockBegin});
 }

-using namespace vixl;
-using namespace vixl::aarch64;
-#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
-
 DEF_OP(Fence) {
  auto Op = IROp->C<IR::IROp_Fence>();
  switch (Op->Fence) {
    case IR::Fence_Load.Val:
-      dmb(FullSystem, BarrierReads);
+      dmb(FEXCore::ARMEmitter::BarrierScope::LD);
      break;
    case IR::Fence_LoadStore.Val:
-      dmb(FullSystem, BarrierAll);
+      dmb(FEXCore::ARMEmitter::BarrierScope::SY);
      break;
    case IR::Fence_Store.Val:
-      dmb(FullSystem, BarrierWrites);
+      dmb(FEXCore::ARMEmitter::BarrierScope::ST);
      break;
    default: LOGMAN_MSG_A_FMT("Unknown Fence: {}", Op->Fence); break;
  }
@@ -37,127 +41,128 @@ DEF_OP(Fence) {

 DEF_OP(Break) {
  auto Op = IROp->C<IR::IROp_Break>();
-  switch (Op->Reason) {
-    case FEXCore::IR::Break_Unimplemented: // Hard fault
-    case FEXCore::IR::Break_Interrupt: // Guest ud2
-      hlt(4);
-      break;
-    case FEXCore::IR::Break_Overflow: // overflow
-      ResetStack();
-      LoadConstant(TMP1, ThreadSharedData.Dispatcher->OverflowExceptionInstructionAddress);
-      br(TMP1);
-      break;
-    case FEXCore::IR::Break_Halt: { // HLT
-      // Time to quit
-      // Set our stack to the starting stack location
-      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, ReturningStackLocation)));
-      add(sp, TMP1, 0);

-      // Now we need to jump to the thread stop handler
-      LoadConstant(TMP1, ThreadSharedData.Dispatcher->ThreadStopHandlerAddressSpillSRA);
-      br(TMP1);
-      break;
-    }
-    case FEXCore::IR::Break_Interrupt3: { // INT3
-      ResetStack();
+  // First we must reset the stack
+  ResetStack();

-      LoadConstant(TMP1, ThreadSharedData.Dispatcher->ThreadPauseHandlerAddressSpillSRA);
-      br(TMP1);
-      break;
-    }
-    case FEXCore::IR::Break_InvalidInstruction:
-    {
-      ResetStack();
+  Core::CpuStateFrame::SynchronousFaultDataStruct State = {
+    .FaultToTopAndGeneratedException = 1,
+    .Signal = Op->Reason.Signal,
+    .TrapNo = Op->Reason.TrapNumber,
+    .si_code = Op->Reason.si_code,
+    .err_code = Op->Reason.ErrorRegister,
+  };

-      LoadConstant(TMP1, ThreadSharedData.Dispatcher->UnimplementedInstructionAddress);
-      br(TMP1);
+  uint64_t Constant{};
+  memcpy(&Constant, &State, sizeof(State));

-      break;
-    }
-    default: LOGMAN_MSG_A_FMT("Unknown Break reason: {}", Op->Reason);
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, Constant);
+  str(ARMEmitter::XReg::x1, STATE, offsetof(FEXCore::Core::CpuStateFrame, SynchronousFaultData));
+
+  switch (Op->Reason.Signal) {
+  case Core::FAULT_SIGILL:
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGILL));
+    br(TMP1);
+    break;
+  case Core::FAULT_SIGTRAP:
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP));
+    br(TMP1);
+    break;
+  case Core::FAULT_SIGSEGV:
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGSEGV));
+    br(TMP1);
+    break;
+  default:
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP));
+    br(TMP1);
+    break;
  }
 }

 DEF_OP(GetRoundingMode) {
-  auto Dst = GetReg<RA_64>(Node);
-  mrs(Dst, FPCR);
-  lsr(Dst, Dst,  22);
+  auto Dst = GetReg(Node);
+  mrs(Dst, ARMEmitter::SystemRegister::FPCR);
+  lsr(ARMEmitter::Size::i64Bit, Dst, Dst, 22);

  // FTZ is already in the correct location
  // Rounding mode is different
-  and_(TMP1, Dst, 0b11);
+  and_(ARMEmitter::Size::i64Bit, TMP1, Dst, 0b11);

-  cmp(TMP1, 1);
-  LoadConstant(TMP3, IR::ROUND_MODE_POSITIVE_INFINITY);
-  csel(TMP2, TMP3, xzr, vixl::aarch64::Condition::eq);
+  cmp(ARMEmitter::Size::i64Bit, TMP1, 1);
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP3, IR::ROUND_MODE_POSITIVE_INFINITY);
+  csel(ARMEmitter::Size::i64Bit, TMP2, TMP3, ARMEmitter::Reg::zr, ARMEmitter::Condition::CC_EQ);

-  cmp(TMP1, 2);
-  LoadConstant(TMP3, IR::ROUND_MODE_NEGATIVE_INFINITY);
-  csel(TMP2, TMP3, TMP2, vixl::aarch64::Condition::eq);
+  cmp(ARMEmitter::Size::i64Bit, TMP1, 2);
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP3, IR::ROUND_MODE_NEGATIVE_INFINITY);
+  csel(ARMEmitter::Size::i64Bit, TMP2, TMP3, TMP2, ARMEmitter::Condition::CC_EQ);

-  cmp(TMP1, 3);
-  LoadConstant(TMP3, IR::ROUND_MODE_TOWARDS_ZERO);
-  csel(TMP2, TMP3, TMP2, vixl::aarch64::Condition::eq);
+  cmp(ARMEmitter::Size::i64Bit, TMP1, 3);
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP3, IR::ROUND_MODE_TOWARDS_ZERO);
+  csel(ARMEmitter::Size::i64Bit, TMP2, TMP3, TMP2, ARMEmitter::Condition::CC_EQ);

-  orr(Dst, Dst, TMP2);
+  orr(ARMEmitter::Size::i64Bit, Dst, Dst, TMP2.R());

-  bfi(Dst, TMP2, 0, 2);
+  bfi(ARMEmitter::Size::i64Bit, Dst, TMP2, 0, 2);
 }

 DEF_OP(SetRoundingMode) {
  auto Op = IROp->C<IR::IROp_SetRoundingMode>();
-  auto Src = GetReg<RA_64>(Op->Header.Args[0].ID());
+  auto Src = GetReg(Op->RoundMode.ID());

  // Setup the rounding flags correctly
-  and_(TMP1, Src, 0b11);
+  and_(ARMEmitter::Size::i64Bit, TMP1, Src, 0b11);

-  cmp(TMP1, IR::ROUND_MODE_POSITIVE_INFINITY);
-  LoadConstant(TMP3, 1);
-  csel(TMP2, TMP3, xzr, vixl::aarch64::Condition::eq);
+  cmp(ARMEmitter::Size::i64Bit, TMP1, IR::ROUND_MODE_POSITIVE_INFINITY);
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP3, 1);
+  csel(ARMEmitter::Size::i64Bit, TMP2, TMP3, ARMEmitter::Reg::zr, ARMEmitter::Condition::CC_EQ);

-  cmp(TMP1, IR::ROUND_MODE_NEGATIVE_INFINITY);
-  LoadConstant(TMP3, 2);
-  csel(TMP2, TMP3, TMP2, vixl::aarch64::Condition::eq);
+  cmp(ARMEmitter::Size::i64Bit, TMP1, IR::ROUND_MODE_NEGATIVE_INFINITY);
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP3, 2);
+  csel(ARMEmitter::Size::i64Bit, TMP2, TMP3, TMP2, ARMEmitter::Condition::CC_EQ);

-  cmp(TMP1, IR::ROUND_MODE_TOWARDS_ZERO);
-  LoadConstant(TMP3, 3);
-  csel(TMP2, TMP3, TMP2, vixl::aarch64::Condition::eq);
+  cmp(ARMEmitter::Size::i64Bit, TMP1, IR::ROUND_MODE_TOWARDS_ZERO);
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP3, 3);
+  csel(ARMEmitter::Size::i64Bit, TMP2, TMP3, TMP2, ARMEmitter::Condition::CC_EQ);

-  mrs(TMP1, FPCR);
+  mrs(TMP1, ARMEmitter::SystemRegister::FPCR);

+  // vixl simulator doesn't support anything beyond ties-to-even rounding
+#ifndef VIXL_SIMULATOR
  // Insert the rounding flags
-  bfi(TMP1, TMP2, 22, 2);
+  bfi(ARMEmitter::Size::i64Bit, TMP1, TMP2, 22, 2);
+#endif

  // Insert the FTZ flag
-  lsr(TMP2, Src, 2);
-  bfi(TMP1, TMP2, 24, 1);
+  lsr(ARMEmitter::Size::i64Bit, TMP2, Src, 2);
+  bfi(ARMEmitter::Size::i64Bit, TMP1, TMP2, 24, 1);

  // Now save the new FPCR
-  msr(FPCR, TMP1);
+  msr(ARMEmitter::SystemRegister::FPCR, TMP1);
 }

 DEF_OP(Print) {
  auto Op = IROp->C<IR::IROp_Print>();

-  PushDynamicRegsAndLR();
+  PushDynamicRegsAndLR(TMP1);
+  SpillStaticRegs(TMP1);

-  if (IsGPR(Op->Header.Args[0].ID())) {
-    mov(x0, GetReg<RA_64>(Op->Header.Args[0].ID()));
-    LoadConstant(x3, reinterpret_cast<uint64_t>(PrintValue));
+  if (IsGPR(Op->Value.ID())) {
+    mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, GetReg(Op->Value.ID()));
+    ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.PrintValue));
  }
  else {
-    fmov(x0, GetSrc(Op->Header.Args[0].ID()).V1D());
-    // Bug in vixl that source vector needs to b V1D rather than V2D?
-    fmov(x1, GetSrc(Op->Header.Args[0].ID()).V1D(), 1);
-    LoadConstant(x3, reinterpret_cast<uint64_t>(PrintVectorValue));
+    fmov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, GetVReg(Op->Value.ID()), false);
+    fmov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, GetVReg(Op->Value.ID()), true);
+    ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.PrintVectorValue));
  }
-  SpillStaticRegs();
-  blr(x3);
-  FillStaticRegs();

+  blr(ARMEmitter::Reg::r3);
+
+  FillStaticRegs();
  PopDynamicRegsAndLR();
 }

+#ifndef _WIN32
 DEF_OP(ProcessorID) {
  // We always need to spill x8 since we can't know if it is live at this SSA location
  uint32_t SpillMask = 1U << 8;
@@ -165,34 +170,34 @@ DEF_OP(ProcessorID) {
  // Ordering is incredibly important here
  // We must spill any overlapping registers first THEN claim we are in a syscall without invalidating state at all
  // Only spill the registers that intersect with our usage
-  SpillStaticRegs(false, SpillMask);
+  SpillStaticRegs(TMP1, false, SpillMask);

  // Now that we are spilled, store in the state that we are in a syscall
  // Still without overwriting registers that matter
  // 16bit LoadConstant to be a single instruction
  // We must always spill at least one register (x8) so this value always has a bit set
  // This gives the signal handler a value to check to see if we are in a syscall at all
-  LoadConstant(x0, SpillMask & 0xFFFF);
-  str(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo)));
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, SpillMask & 0xFFFF);
+  str(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));

  // Allocate some temporary space for storing the uint32_t CPU and Node IDs
-  sub(sp, sp, 16);
+  sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, 16);

  // Load the getcpu syscall number
-  LoadConstant(x8, SYS_getcpu);
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r8, SYS_getcpu);

  // CPU pointer in x0
-  add(x0, sp, 0);
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, ARMEmitter::Reg::rsp, 0);
  // Node in x1
-  add(x1, sp, 4);
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, ARMEmitter::Reg::rsp, 4);

  svc(0);
  // On updated signal mask we can receive a signal RIGHT HERE

  // Load the values returned by the kernel
-  ldp(w0, w1, MemOperand(sp));
+  ldp<ARMEmitter::IndexType::OFFSET>(ARMEmitter::WReg::w0, ARMEmitter::WReg::w1, ARMEmitter::Reg::rsp);
  // Deallocate stack space
-  sub(sp, sp, 16);
+  sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, 16);

  // Now that we are done in the syscall we need to carefully peel back the state
  // First unspill the registers from before
@@ -200,34 +205,42 @@ DEF_OP(ProcessorID) {

  // Now the registers we've spilled are back in their original host registers
  // We can safely claim we are no longer in a syscall
-  str(xzr, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo)));
-
+  str(ARMEmitter::XReg::zr, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));

  // Now store the result in the destination in the expected format
  // uint32_t Res = (node << 12) | cpu;
  // CPU is in w0
  // Node is in w1
-  orr(GetReg<RA_64>(Node), x0, Operand(x1, LSL, 12));
+  orr(ARMEmitter::Size::i64Bit, GetReg(Node), ARMEmitter::Reg::r0, ARMEmitter::Reg::r1, ARMEmitter::ShiftType::LSL, 12);
+}
+#else
+DEF_OP(ProcessorID) {
+  ERROR_AND_DIE_FMT("Unsupported");
+}
+#endif
+
+DEF_OP(RDRAND) {
+  auto Op = IROp->C<IR::IROp_RDRAND>();
+
+  // Results are in x0, x1
+  // Results want to be in a i64v2 vector
+  auto Dst = GetRegPair(Node);
+
+  if (Op->GetReseeded) {
+    mrs(Dst.first, ARMEmitter::SystemRegister::RNDRRS);
+  }
+  else {
+    mrs(Dst.first, ARMEmitter::SystemRegister::RNDR);
+  }
+
+  // If the rng number is valid then NZCV is 0b0000, otherwise NZCV is 0b0100
+  cset(ARMEmitter::Size::i64Bit, Dst.second, ARMEmitter::Condition::CC_NE);
+}
+
+DEF_OP(Yield) {
+  yield();
 }

 #undef DEF_OP
-void Arm64JITCore::RegisterMiscHandlers() {
-#define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &Arm64JITCore::Op_##x
-  REGISTER_OP(DUMMY,      NoOp);
-  REGISTER_OP(IRHEADER,   NoOp);
-  REGISTER_OP(CODEBLOCK,  NoOp);
-  REGISTER_OP(BEGINBLOCK, NoOp);
-  REGISTER_OP(ENDBLOCK,   NoOp);
-  REGISTER_OP(FENCE,      Fence);
-  REGISTER_OP(BREAK,      Break);
-  REGISTER_OP(PHI,        NoOp);
-  REGISTER_OP(PHIVALUE,   NoOp);
-  REGISTER_OP(PRINT,      Print);
-  REGISTER_OP(GETROUNDINGMODE, GetRoundingMode);
-  REGISTER_OP(SETROUNDINGMODE, SetRoundingMode);
-  REGISTER_OP(INVALIDATEFLAGS,   NoOp);
-  REGISTER_OP(PROCESSORID,   ProcessorID);
-#undef REGISTER_OP
-}
 }

@@ -7,79 +7,40 @@ $end_info$
 #include "Interface/Core/JIT/Arm64/JITClass.h"

 namespace FEXCore::CPU {
-
-using namespace vixl;
-using namespace vixl::aarch64;
-#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
+#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header const *IROp, IR::NodeID Node)
 DEF_OP(ExtractElementPair) {
  auto Op = IROp->C<IR::IROp_ExtractElementPair>();
-  switch (Op->Header.Size) {
-    case 4: {
-      auto Src = GetSrcPair<RA_32>(Op->Header.Args[0].ID());
-      std::array<aarch64::Register, 2> Regs = {Src.first, Src.second};
-      mov (GetReg<RA_32>(Node), Regs[Op->Element]);
-      break;
-    }
-    case 8: {
-      auto Src = GetSrcPair<RA_64>(Op->Header.Args[0].ID());
-      std::array<aarch64::Register, 2> Regs = {Src.first, Src.second};
-      mov (GetReg<RA_64>(Node), Regs[Op->Element]);
-      break;
-    }
-    default: LOGMAN_MSG_A_FMT("Unknown Size"); break;
-  }
+  LOGMAN_THROW_AA_FMT(Op->Header.Size == 4 || Op->Header.Size == 8, "Invalid size");
+  const auto EmitSize = Op->Header.Size == 8 ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
+
+  const auto Src = GetRegPair(Op->Pair.ID());
+  const std::array<ARMEmitter::Register, 2> Regs = {Src.first, Src.second};
+  mov(EmitSize, GetReg(Node), Regs[Op->Element]);
 }

 DEF_OP(CreateElementPair) {
  auto Op = IROp->C<IR::IROp_CreateElementPair>();
-  std::pair<aarch64::Register, aarch64::Register> Dst;
-  aarch64::Register RegFirst;
-  aarch64::Register RegSecond;
-  aarch64::Register RegTmp;
+  LOGMAN_THROW_AA_FMT(IROp->ElementSize == 4 || IROp->ElementSize == 8, "Invalid size");
+  std::pair<ARMEmitter::Register, ARMEmitter::Register> Dst = GetRegPair(Node);
+  ARMEmitter::Register RegFirst = GetReg(Op->Lower.ID());
+  ARMEmitter::Register RegSecond = GetReg(Op->Upper.ID());
+  ARMEmitter::Register RegTmp = TMP1.R();

-  switch (Op->Header.Size) {
-    case 4: {
-      Dst = GetSrcPair<RA_32>(Node);
-      RegFirst = GetReg<RA_32>(Op->Header.Args[0].ID());
-      RegSecond = GetReg<RA_32>(Op->Header.Args[1].ID());
-      RegTmp = w0;
-      break;
-    }
-    case 8: {
-      Dst = GetSrcPair<RA_64>(Node);
-      RegFirst = GetReg<RA_64>(Op->Header.Args[0].ID());
-      RegSecond = GetReg<RA_64>(Op->Header.Args[1].ID());
-      RegTmp = x0;
-      break;
-    }
-    default: LOGMAN_MSG_A_FMT("Unknown Size"); break;
-  }
+  const auto EmitSize = IROp->ElementSize == 8 ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;

-  if (Dst.first.GetCode() != RegSecond.GetCode()) {
-    mov(Dst.first, RegFirst);
-    mov(Dst.second, RegSecond);
-  } else if (Dst.second.GetCode() != RegFirst.GetCode()) {
-    mov(Dst.second, RegSecond);
-    mov(Dst.first, RegFirst);
+  if (Dst.first.Idx() != RegSecond.Idx()) {
+    mov(EmitSize, Dst.first, RegFirst);
+    mov(EmitSize, Dst.second, RegSecond);
+  } else if (Dst.second.Idx() != RegFirst.Idx()) {
+    mov(EmitSize, Dst.second, RegSecond);
+    mov(EmitSize, Dst.first, RegFirst);
  } else {
-    mov(RegTmp, RegFirst);
-    mov(Dst.second, RegSecond);
-    mov(Dst.first, RegTmp);
+    mov(EmitSize, RegTmp, RegFirst);
+    mov(EmitSize, Dst.second, RegSecond);
+    mov(EmitSize, Dst.first, RegTmp);
  }
 }

-DEF_OP(Mov) {
-  auto Op = IROp->C<IR::IROp_Mov>();
-  mov(GetReg<RA_64>(Node), GetReg<RA_64>(Op->Header.Args[0].ID()));
-}
-
 #undef DEF_OP
-void Arm64JITCore::RegisterMoveHandlers() {
-#define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &Arm64JITCore::Op_##x
-  REGISTER_OP(EXTRACTELEMENTPAIR, ExtractElementPair);
-  REGISTER_OP(CREATEELEMENTPAIR,  CreateElementPair);
-  REGISTER_OP(MOV,                Mov);
-#undef REGISTER_OP
-}
 }

@@ -1,9 +1,10 @@
 #pragma once

-#include <memory>
+#include <FEXCore/Core/CPUBackend.h>
+#include <FEXCore/fextl/memory.h>

 namespace FEXCore::Context {
-struct Context;
+class ContextImpl;
 }

 namespace FEXCore::Core {
@@ -13,11 +14,12 @@ struct InternalThreadState;
 namespace FEXCore::CPU {
 class CPUBackend;

-[[nodiscard]] std::unique_ptr<CPUBackend> CreateX86JITCore(FEXCore::Context::Context *ctx,
-                                                           FEXCore::Core::InternalThreadState *Thread,
-                                                           bool CompileThread);
+[[nodiscard]] fextl::unique_ptr<CPUBackend> CreateX86JITCore(FEXCore::Context::ContextImpl *ctx,
+                                                           FEXCore::Core::InternalThreadState *Thread);
+CPUBackendFeatures GetX86JITBackendFeatures();
+
+[[nodiscard]] fextl::unique_ptr<CPUBackend> CreateArm64JITCore(FEXCore::Context::ContextImpl *ctx,
+                                                             FEXCore::Core::InternalThreadState *Thread);
+CPUBackendFeatures GetArm64JITBackendFeatures();

-[[nodiscard]] std::unique_ptr<CPUBackend> CreateArm64JITCore(FEXCore::Context::Context *ctx,
-                                                             FEXCore::Core::InternalThreadState *Thread,
-                                                             bool CompileThread);
 } // namespace FEXCore::CPU