Docs: Update for release FEX-2603

Merge pull request #5338 from cjacek/llvm22
Fix mingw builds with LLVM 22.
2026-10-06 21:00:17 +02:00 · 2026-03-04 12:44:21 -08:00 · 2026-02-27 12:05:41 -08:00 · 2026-02-27 15:30:46 +01:00 · 2026-02-27 15:27:36 +01:00 · 2026-02-26 18:11:20 -08:00
918 changed files with 52818 additions and 40980 deletions
@@ -2,8 +2,11 @@
 Source/Common/cpp-optparse/*

 # Files with human-indented tables for readability - don't mess with these
-FEXCore/Source/Interface/Core/X86Tables/*
+FEXCore/Source/Interface/Core/X86Tables/*.cpp

 # Inline headers with list-like content that can't be processed individually
 Source/Tools/LinuxEmulation/LinuxSyscalls/x*/SyscallsNames.inl
 Source/Tools/LinuxEmulation/LinuxSyscalls/x*/Ioctl/*.inl
+
+# Include files in unittests
+unittests/*ASM/Includes/*.inc
@@ -22,3 +22,6 @@

 # Minor reformat with clang-format-19
 9fdd96af61c969cb5732471223f00eda64b7a069
+
+# Reformat of X86Tables.h
+ba2b0ef809f66f1a6d334f000798fa2ceafab26f
@@ -24,238 +24,138 @@ jobs:
      fail-fast: false

    steps:
-    - uses: actions/checkout@v3
+    - uses: actions/checkout@v6
+      with:
+        fetch-depth: '0'
+        fetch-tags: 'true'

-    - name: Set runner label
-      run: echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
-
-    - name: Set rootfs paths
+    - name: Set runner info
      run: |
-        echo "FEX_ROOTFS_MOUNT=/mnt/AutoNFS/rootfs/" >> $GITHUB_ENV
-        echo "FEX_ROOTFS_PATH=$HOME/Rootfs/" >> $GITHUB_ENV
-        echo "FEX_ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
+        echo "runner_name=$(hostname)" >> $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: Setup Build Environment
+      uses: ./.github/workflows/setup-env

    - 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 -DBUILD_FEX_LINUX_TESTS=True -DBUILD_THUNKS=True -DCMAKE_INSTALL_PREFIX=${{runner.workspace}}/build/install
+      run: |
+        cmake -S . -B build -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=True \
+          -DENABLE_X86_HOST_DEBUG=True -DBUILD_FEX_LINUX_TESTS=True -DBUILD_THUNKS=True \
+          -DCMAKE_INSTALL_PREFIX="$PWD"/build/install
+
+    # These steps make a lot of noise but rarely fail.
+    # Put them in a separate step to make normal build logs easier to parse
+    - name: Noisy Build Targets
+      run: cmake --build build --target asm_files 32bit_asm_files JemallocLibs Catch2 vixl cephes_128bit

    - 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
+      run: cmake --build build

    - name: Install
-      working-directory: ${{runner.workspace}}/build
-      shell: bash
-      run: cmake --build . --config $BUILD_TYPE --target install
+      run: cmake --build build --target install

-    - 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
+    # GCC tests
+    - name: GCC64 Target Tests
      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
+      uses: ./.github/workflows/test
+      with:
+        target: gcc_target_tests_64

-    - 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
+    - name: GCC32 Target Tests
      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
+      uses: ./.github/workflows/test
+      with:
+        target: gcc_target_tests_32

-    - name: APITest tests
-      working-directory: ${{runner.workspace}}/build
-      shell: bash
-      run: cmake --build . --config $BUILD_TYPE --target api_tests
-
-    - name: APITest Test Results move
+    # API tests
+    - name: API Tests
      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
+      uses: ./.github/workflows/test
+      with:
+        target: api_tests

-    - 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
+    - name: FEXCore API Tests
      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
+      uses: ./.github/workflows/test
+      with:
+        target: fexcore_apitests

-    - name: ARMEmitter tests
-      working-directory: ${{runner.workspace}}/build
-      shell: bash
-      run: cmake --build . --config $BUILD_TYPE --target emitter_tests
-
-    - name: ARMEmitter Test Results move
+    # ARM emission tests
+    - name: ARM Emitter Tests
      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
+      uses: ./.github/workflows/test
+      with:
+        target: emitter_tests

-    - name: FEXLinuxTests
-      working-directory: ${{runner.workspace}}/build
-      shell: bash
+    # Linux  tests
+    - name: FEX Linux Tests
+      if: ${{ always() }}
+      uses: ./.github/workflows/test
+      with:
+        target: fex_linux_tests_all
      env:
-        # These tests require non-portable install due to thunks.
        FEX_PORTABLE: 0
-      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

+    # Thunking
    - 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
+      uses: ./.github/workflows/test
+      with:
+        target: thunkgen_tests

    - name: Test GL No-Thunks
-      if: matrix.arch[1] == 'x64'
-      working-directory: ${{runner.workspace}}/build
-      shell: bash
+      if: ${{ always() && matrix.arch[1] == 'x64' }}
+      uses: ./.github/workflows/test
+      with:
+        target: thunk_functional_tests_nothunks
      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
+        DISPLAY: ':0'

    - name: Test GL Thunks
-      if: matrix.arch[1] == 'x64'
-      working-directory: ${{runner.workspace}}/build
-      shell: bash
+      if: ${{ always() && matrix.arch[1] == 'x64' }}
+      uses: ./.github/workflows/test
+      with:
+        target: thunk_functional_tests_thunks
      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
+        DISPLAY: ':0'

+    # ASM tests
    - 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
+      uses: ./.github/workflows/test
+      with:
+        target: asm_tests

-    - 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
+    # POSIX tests
+    - name: POSIX Tests
      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
+      uses: ./.github/workflows/test
+      with:
+        target: posix_tests

-    - name: gvisor tests
-      working-directory: ${{runner.workspace}}/build
-      shell: bash
-      # Execute the gvisor tests
-      run: cmake --build . --config $BUILD_TYPE --target gvisor_tests
-
-    - name: GVisor Test Results move
+    # GVisor tests
+    - name: GVisor Tests
      if: ${{ always() }}
-      shell: bash
-      working-directory: ${{runner.workspace}}/build
-      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_GVisor.log || true
+      uses: ./.github/workflows/test
+      with:
+        target: gvisor_tests

+    # Struct verifier tests
    - name: Struct verifier tests
-      working-directory: ${{runner.workspace}}/build
-      shell: bash
-      run: cmake --build . --config $BUILD_TYPE --target struct_verifier
-
-    - name: Struct verifier 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_StructVerifier.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
+      uses: ./.github/workflows/test
+      with:
+        target: struct_verifier

    - 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
+      run: cmake --build build --target remove_old_shm_regions

    - name: Upload results
      if: ${{ always() }}
-      uses: 'actions/upload-artifact@v4'
+      uses: actions/upload-artifact@v6
      timeout-minutes: 1
      with:
-        name: Results-${{ env.runner_name }}
-        path: ${{runner.workspace}}/build/Testing/Temporary/LastTest_*.log
+        name: Results-${{ env.runner_name }}-${{ env.runner_label }}
+        path: results/*.log
        retention-days: 3
@@ -31,163 +31,94 @@ jobs:
      fail-fast: false

    steps:
-    - uses: actions/checkout@v3
+    - uses: actions/checkout@v6
+      with:
+        fetch-depth: '0'
+        fetch-tags: 'true'

-    - name: Set runner label
-      run: echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
-
-    - name: Set rootfs paths
+    - name: Set runner info
      run: |
-        echo "FEX_ROOTFS_MOUNT=/mnt/AutoNFS/rootfs/" >> $GITHUB_ENV
-        echo "FEX_ROOTFS_PATH=$HOME/Rootfs/" >> $GITHUB_ENV
-        echo "FEX_ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
+        echo "runner_name=$(hostname)" >> $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: Setup Build Environment
+      uses: ./.github/workflows/setup-env

    - 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 -DBUILD_FEX_LINUX_TESTS=True -DENABLE_GLIBC_ALLOCATOR_HOOK_FAULT=True -DENABLE_JEMALLOC_GLIBC_ALLOC=False -DCMAKE_INSTALL_PREFIX=${{runner.workspace}}/build/install
+      run: |
+        cmake -S . -B build -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DENABLE_LTO=False \
+          -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True -DBUILD_FEX_LINUX_TESTS=True \
+          -DENABLE_GLIBC_ALLOCATOR_HOOK_FAULT=True -DENABLE_JEMALLOC_GLIBC_ALLOC=False \
+          -DCMAKE_INSTALL_PREFIX="$PWD"/build/install
+
+    # These steps make a lot of noise but rarely fail.
+    # Put them in a separate step to make normal build logs easier to parse
+    - name: Noisy Build Targets
+      run: cmake --build build --target asm_files 32bit_asm_files JemallocLibs Catch2 vixl cephes_128bit

    - 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
+      run: cmake --build build

    - name: Install
-      working-directory: ${{runner.workspace}}/build
-      shell: bash
-      run: cmake --build . --config $BUILD_TYPE --target install
+      run: cmake --build build --target install

-    - 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
+    # GCC tests
+    - name: GCC64 Target Tests
      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
+      uses: ./.github/workflows/test
+      with:
+        target: gcc_target_tests_64

-    - 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
+    - name: GCC32 Target Tests
      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
+      uses: ./.github/workflows/test
+      with:
+        target: gcc_target_tests_32

-    - name: APITest tests
-      working-directory: ${{runner.workspace}}/build
-      shell: bash
-      run: cmake --build . --config $BUILD_TYPE --target api_tests
-
-    - name: APITest Test Results move
+    # API Tests
+    - name: API Tests
      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
+      uses: ./.github/workflows/test
+      with:
+        target: api_tests

-    - 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
+    - name: FEXCore API Tests
      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
+      uses: ./.github/workflows/test
+      with:
+        target: fexcore_apitests

-    - name: FEXLinuxTests
-      working-directory: ${{runner.workspace}}/build
-      shell: bash
-      run: cmake --build . --config $BUILD_TYPE --target fex_linux_tests_all
-
-    - name: FEXLinuxTests Results move
+    # Linux tests
+    - name: FEX Linux Tests
      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
+      uses: ./.github/workflows/test
+      with:
+        target: fex_linux_tests_all

+    # ASM Tests
    - 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
+      uses: ./.github/workflows/test
+      with:
+        target: asm_tests

-    - 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
+    # POSIX Tests
+    - name: POSIX Tests
      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: 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
+      uses: ./.github/workflows/test
+      with:
+        target: posix_tests

    - 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
+      run: cmake --build build --target remove_old_shm_regions

    - name: Upload results
      if: ${{ always() }}
-      uses: 'actions/upload-artifact@v4'
+      uses: actions/upload-artifact@v6
      timeout-minutes: 1
      with:
-        name: Results-${{ env.runner_name }}
-        path: ${{runner.workspace}}/build/Testing/Temporary/LastTest_*.log
+        name: Results-${{ env.runner_name }}-${{ env.runner_label }}
+        path: results/*.log
        retention-days: 3

@@ -24,83 +24,45 @@ jobs:
      fail-fast: false

    steps:
-    - uses: actions/checkout@v3
+    - uses: actions/checkout@v6
+      with:
+        fetch-depth: '0'
+        fetch-tags: 'true'

-    - name: Set runner label
-      run: echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
-
-    - name: Set rootfs paths
+    - name: Set runner info
      run: |
-        echo "FEX_ROOTFS_MOUNT=/mnt/AutoNFS/rootfs/" >> $GITHUB_ENV
-        echo "FEX_ROOTFS_PATH=$HOME/Rootfs/" >> $GITHUB_ENV
-        echo "FEX_ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
+        echo "runner_name=$(hostname)" >> $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: Setup Build Environment
+      uses: ./.github/workflows/setup-env

    - 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
+      run: |
+        cmake -S . -B build -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DENABLE_LTO=False \
+          -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True
+
+    # These steps make a lot of noise but rarely fail.
+    # Put them in a separate step to make normal build logs easier to parse
+    - name: Noisy Build Targets
+      run: cmake --build build --target asm_files 32bit_asm_files JemallocLibs Catch2 vixl cephes_128bit

    - 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
+      run: cmake --build build

+    # ASM tests
    - 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: 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
+      uses: ./.github/workflows/test
+      with:
+        target: asm_tests

    - name: Upload results
      if: ${{ always() }}
-      uses: 'actions/upload-artifact@v4'
+      uses: actions/upload-artifact@v6
      timeout-minutes: 1
      with:
-        name: Results-${{ env.runner_name }}
-        path: ${{runner.workspace}}/build/Testing/Temporary/LastTest_*.log
+        name: Results-${{ env.runner_name }}-${{ env.runner_label }}
+        path: results/*.log
        retention-days: 3

@@ -23,123 +23,56 @@ jobs:
      fail-fast: false

    steps:
-    - uses: actions/checkout@v3
+    - uses: actions/checkout@v6
+      with:
+        fetch-depth: '0'
+        fetch-tags: 'true'

-    - name: Set runner label
-      run: echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
-
-    - name: Set rootfs paths
+    - name: Set runner info
      run: |
-        echo "FEX_ROOTFS_MOUNT=/mnt/AutoNFS/rootfs/" >> $GITHUB_ENV
-        echo "FEX_ROOTFS_PATH=$HOME/Rootfs/" >> $GITHUB_ENV
-        echo "FEX_ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
+        echo "runner_name=$(hostname)" >> $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: Setup Build Environment
+      uses: ./.github/workflows/setup-env

-    - name : submodule checkout
-      # Need to update submodules
+    - name: Set VIXL_SIM_ENABLED
      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: Set vixl_sim x86
-      if: matrix.arch[1] == 'x64'
-      run: |
-        echo "VIXL_SIM_ENABLED=True" >> $GITHUB_ENV
-
-    - name: Set vixl_sim Arm64
-      if: matrix.arch[1] == 'ARM64'
-      run: |
-        echo "VIXL_SIM_ENABLED=False" >> $GITHUB_ENV
+        case '${{ matrix.arch[1] }}' in
+          x64) _sim=True ;;
+          ARM64) _sim=False ;;
+        esac
+        echo "VIXL_SIM_ENABLED=$_sim" >> $GITHUB_ENV

    - 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=$VIXL_SIM_ENABLED -DENABLE_VIXL_DISASSEMBLER=True -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True
+      run: |
+        cmake -S . -B build -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DENABLE_VIXL_SIMULATOR=$VIXL_SIM_ENABLED \
+          -DENABLE_VIXL_DISASSEMBLER=True -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True

    - name: Build
-      working-directory: ${{runner.workspace}}/build
-      shell: bash
      env:
        FEX_DISABLETELEMETRY: 1
-      # Execute the build.  You can specify a specific target with "--target <NAME>"
-      run: cmake --build . --config $BUILD_TYPE --target CodeSizeValidation instcountci_test_files
+      run: cmake --build build --target CodeSizeValidation instcountci_test_files

    - name: Instruction Count Tests
-      working-directory: ${{runner.workspace}}/build
-      shell: bash
-      # Execute the unit tests
-      run: cmake --build . --config $BUILD_TYPE --target instcountci_tests
-
-    - name: Instruction Count 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_InstCountCI.log || true
+      uses: ./.github/workflows/test
+      with:
+        target: instcountci_tests

    - name: Update local repo instcount
      if: ${{ always() }}
-      shell: bash
-      working-directory: ${{runner.workspace}}/build
-      run: cmake --build . --config $BUILD_TYPE --target instcountci_update_tests
+      run: cmake --build build --target instcountci_update_tests

-    - name: Get instcountCI diff
+    - name: Check InstCountCI diff
      if: ${{ always() }}
-      shell: bash
-      working-directory: ${{github.workspace}}/
-      run: git diff --output=${{runner.workspace}}/build/InstCountCI.diff
-
-    - name: Check if InstCountCI Diff exists
-      if: ${{ always() }}
-      shell: bash
-      working-directory: ${{github.workspace}}/
-      # Check if the file is empty
-      run: sh -c "! test -s ${{runner.workspace}}/build/InstCountCI.diff"
-
-    - 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
+      run: git --no-pager diff --exit-code HEAD

    - name: Upload results
      if: ${{ always() }}
-      uses: 'actions/upload-artifact@v4'
+      uses: actions/upload-artifact@v6
      timeout-minutes: 1
      with:
-        name: Results-${{ env.runner_name }}
-        path: ${{runner.workspace}}/build/Testing/Temporary/LastTest_*.log
+        name: Results-${{ env.runner_name }}-${{ env.runner_label }}
+        path: results/*.log
        retention-days: 3
-
-    - name: Upload results InstCountCI
-      if: ${{ always() }}
-      uses: 'actions/upload-artifact@v4'
-      timeout-minutes: 1
-      with:
-        name: Results-${{ env.runner_name }}-instcountci
-        path: ${{runner.workspace}}/build/InstCountCI.diff
-        retention-days: 3
-
@@ -20,7 +20,10 @@ jobs:
      fail-fast: false

    steps:
-    - uses: actions/checkout@v3
+    - uses: actions/checkout@v6
+      with:
+        fetch-depth: '0'
+        fetch-tags: 'true'

    - name: Set runner label
      run: echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
@@ -28,73 +31,23 @@ jobs:
    - name: Add MingGW to PATH
      run: echo "$HOME/llvm-mingw/build/bin/" >> $GITHUB_PATH

-    - name: Set CC x86
-      if: matrix.arch[1] == 'x64'
+    - name: Set CC
      run: |
-        echo "MINGW_TRIPLE=x86_64-w64-mingw32" >> $GITHUB_ENV
+        case '${{ matrix.arch[1] }}' in
+          x64) _cpu=x86_64 ;;
+          ARM64) _cpu=aarch64 ;;
+          ARM64EC) _cpu=arm64ec ;;
+        esac
+        echo "MINGW_TRIPLE=${_cpu}-w64-mingw32" >> $GITHUB_ENV

-    - name: Set CC Arm64
-      if: matrix.arch[1] == 'ARM64'
-      run: |
-        echo "MINGW_TRIPLE=aarch64-w64-mingw32" >> $GITHUB_ENV
-
-    - name: Set CC Arm64EC
-      if: matrix.arch[1] == 'ARM64EC'
-      run: |
-        echo "MINGW_TRIPLE=arm64ec-w64-mingw32" >> $GITHUB_ENV
-
-    - name: Set rootfs paths
-      run: |
-        echo "FEX_ROOTFS_MOUNT=/mnt/AutoNFS/rootfs/" >> $GITHUB_ENV
-        echo "FEX_ROOTFS_PATH=$HOME/Rootfs/" >> $GITHUB_ENV
-        echo "FEX_ROOTFS=$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: Setup Build Environment
+      uses: ./.github/workflows/setup-env

    - 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 -DCMAKE_TOOLCHAIN_FILE=$GITHUB_WORKSPACE/Data/CMake/toolchain_mingw.cmake -DMINGW_TRIPLE=$MINGW_TRIPLE -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True -DBUILD_TESTING=False -DCMAKE_INSTALL_PREFIX=${{runner.workspace}}/build/install
+      run: |
+        cmake -S . -B build -DCMAKE_BUILD_TYPE=$BUILD_TYPE -DCMAKE_TOOLCHAIN_FILE=$GITHUB_WORKSPACE/Data/CMake/toolchain_mingw.cmake \
+        -DMINGW_TRIPLE=$MINGW_TRIPLE -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True -DBUILD_TESTING=False \
+        -DCMAKE_INSTALL_PREFIX="$PWD"/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: ${{ always() }}
-      uses: 'actions/upload-artifact@v4'
-      timeout-minutes: 1
-      with:
-        name: Results-${{ env.runner_name }}
-        path: ${{runner.workspace}}/build/Testing/Temporary/LastTest_*.log
-        retention-days: 3
-
+      run: cmake --build build
@@ -1,7 +1,7 @@
-# Inspired by LLVM's pr-code-format.yml at 
+# Inspired by LLVM's pr-code-format.yml at
 # https://github.com/llvm/llvm-project/blob/main/.github/workflows/pr-code-format.yml

-name: "Check code formatting"
+name: Check code formatting
 on:
  pull_request:
    branches:
@@ -13,7 +13,7 @@ jobs:
    if: github.repository == 'FEX-Emu/FEX'

    steps:
-      - name: Fetch FEX sources
+      - name: Checkout
        uses: actions/checkout@v4
        with:
          ref: ${{ github.event.pull_request.head.sha }}
@@ -27,18 +27,13 @@ jobs:
          deepen_length: 500

      - name: Get changed files
-        id: changed-files
-        uses: step-security/changed-files@3dbe17c78367e7d60f00d78ae6781a35be47b4a1 # v45.0.1
-        with:
-          separator: ","
-          skip_initial_fetch: true
-
-      - name: "Listed files"
-        env:
-          CHANGED_FILES: ${{ steps.changed-files.outputs.all_changed_files }}
        run: |
-          echo "Formatting files:"
-          echo "$CHANGED_FILES"
+          BASE=$(git merge-base main HEAD)
+          FILES=$(git diff --name-only "$BASE" | tr '\n' ',' | sed 's/,$//')
+          echo "CHANGED_FILES=$FILES" >> $GITHUB_ENV
+
+          echo "Changed files:"
+          echo "$FILES"

      - name: Check git-clang-format-19 exists
        run: which git-clang-format-19
@@ -46,24 +41,23 @@ jobs:
      - name: Setup Python env
        uses: actions/setup-python@v4
        with:
-          python-version: '3.11'
-          cache: 'pip'
-          cache-dependency-path: './External/code-format-helper/requirements_formatting.txt'
+          python-version: 3.11
+          cache: pip
+          cache-dependency-path: ./External/code-format-helper/requirements_formatting.txt

      - name: Install python dependencies
        run: pip install -r ./External/code-format-helper/requirements_formatting.txt

      - name: Run code formatter
        env:
-          CLANG_FORMAT_PATH: 'git-clang-format-19'
+          CLANG_FORMAT_PATH: git-clang-format-19
          GITHUB_PR_NUMBER: ${{ github.event.pull_request.number }}
          START_REV: ${{ github.event.pull_request.base.sha }}
          END_REV: ${{ github.event.pull_request.head.sha }}
-          CHANGED_FILES: ${{ steps.changed-files.outputs.all_changed_files }}
        run: |
          python ./External/code-format-helper/code-format-helper.py \
-            --repo "FEX-emu/FEX" \
-            --issue-number $GITHUB_PR_NUMBER \
-            --start-rev $START_REV \
-            --end-rev $END_REV \
+            --repo "FEX-Emu/FEX" \
+            --issue-number "$GITHUB_PR_NUMBER" \
+            --start-rev "$START_REV" \
+            --end-rev "$END_REV" \
            --changed-files "$CHANGED_FILES"
@@ -0,0 +1,33 @@
+name: Setup Build Environment
+description: Setup RootFS and build environment
+
+inputs:
+  setup-rootfs:
+    description: 'Whether or not to set up the rootfs'
+    default: true
+
+runs:
+  using: composite
+  steps:
+    - name: Set rootfs paths
+      if: ${{ inputs.setup-rootfs == 'true' }}
+      shell: bash
+      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
+
+    - name: Update RootFS cache
+      if: ${{ inputs.setup-rootfs == 'true' }}
+      shell: bash
+      run: python3 Scripts/CI_FetchRootFS.py
+
+    - name: Checkout Submodules
+      shell: bash
+      run: |
+        git submodule sync --recursive
+        git submodule update --init --depth 1
+
+    - name: Clean Build Environment
+      shell: bash
+      run: rm -Rf build
@@ -0,0 +1,72 @@
+name: steamrt4 build
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - main
+
+env:
+  DEBIAN_FRONTEND: noninteractive
+  BUILD_TYPE: Release
+  CC: clang
+  CXX: clang++
+
+jobs:
+  steamrt4_build:
+    runs-on: ${{ matrix.arch }}
+    strategy:
+      matrix:
+        arch: [[self-hosted, ARM64, distrobox]]
+      fail-fast: false
+
+    steps:
+    - uses: actions/checkout@v6
+      with:
+        fetch-depth: '0'
+        fetch-tags: 'true'
+
+    - name: Set runner label
+      run: echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
+
+    - name: Setup Build Environment
+      uses: ./.github/workflows/setup-env
+      with:
+        setup-rootfs: false
+
+    # Setup everything required.
+    - name : distrobox setup
+      run: |
+        distrobox create -Y -i registry.gitlab.steamos.cloud/steamrt/steamrt4/sdk/arm64:4.0.20251117.183306 steamrt4 || true
+        distrobox upgrade steamrt4
+        distrobox enter --name steamrt4 -- sudo apt-get install -y \
+          git cmake ninja-build ccache \
+          lld clang \
+          libclang-dev llvm-dev \
+          libstdc++-14-dev-i386-cross libgcc-14-dev-i386-cross \
+          libstdc++-14-dev-amd64-cross libgcc-14-dev-amd64-cross
+
+    - name: Configure CMake
+      run: |
+        distrobox enter --name steamrt4 -- cmake -S . -B build -DCMAKE_BUILD_TYPE=$BUILD_TYPE \
+          -G Ninja -DBUILD_STEAM_SUPPORT=True -DENABLE_LTO=True -DENABLE_ASSERTIONS=False -DBUILD_THUNKS=True \
+          -DBUILD_FEXCONFIG=False -DBUILD_TESTING=False -DENABLE_CLANG_THUNKS=True -DUSE_LINKER=lld \
+          -DCMAKE_INSTALL_PREFIX=/usr
+
+    - name: Build
+      run: distrobox enter --name steamrt4 -- cmake --build build
+
+    - name: install
+      run: DESTDIR="$PWD"/install distrobox enter --name steamrt4 -- cmake --build build -t install
+
+    - name: Upload libraries
+      uses: actions/upload-artifact@v6
+      timeout-minutes: 1
+      with:
+        overwrite: true
+        name: steamrt4_steampipe_depot
+        path: ${{ github.workspace }}/install/*
+        retention-days: 60
+        compression-level: 9
@@ -0,0 +1,21 @@
+name: Run Test and Store Logs
+description: Run a test and store the log.
+inputs:
+  target:
+    description: 'The test target to run'
+    required: true
+
+runs:
+  using: composite
+  steps:
+    - name: Run Tests
+      shell: bash
+      run: cmake --build build --target ${{ inputs.target }}
+
+    - name: Move and Truncate Results
+      if: ${{ always() }}
+      shell: bash
+      run: |
+        mkdir -p results
+        mv build/Testing/Temporary/LastTest.log results/${{ inputs.target }}.log || true
+        truncate --size="<20M" results/${{ inputs.target }}.log || true
@@ -25,111 +25,59 @@ jobs:
      fail-fast: false

    steps:
-    - uses: actions/checkout@v3
+    - uses: actions/checkout@v6
+      with:
+        fetch-depth: '0'
+        fetch-tags: 'true'

-    - name: Set runner label
-      run: echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
-
-    - name: Set rootfs paths
+    - name: Set runner info
      run: |
-        echo "FEX_ROOTFS_MOUNT=/mnt/AutoNFS/rootfs/" >> $GITHUB_ENV
-        echo "FEX_ROOTFS_PATH=$HOME/Rootfs/" >> $GITHUB_ENV
-        echo "FEX_ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
+        echo "runner_name=$(hostname)" >> $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: Setup Build Environment
+      uses: ./.github/workflows/setup-env

    - 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_VIXL_DISASSEMBLER=True -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True
+      run: |
+        cmake -S . -B build -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DENABLE_VIXL_SIMULATOR=True -DENABLE_LTO=False \
+          -DENABLE_VIXL_DISASSEMBLER=True -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True
+
+    # These steps make a lot of noise but rarely fail.
+    # Put them in a separate step to make normal build logs easier to parse
+    - name: Noisy Build Targets
+      run: cmake --build build --target asm_files 32bit_asm_files JemallocLibs Catch2 vixl cephes_128bit

    - 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
+      run: cmake --build build

    - name: ASM Tests - SVE256
-      working-directory: ${{runner.workspace}}/build
-      shell: bash
-      # Execute the unit tests
-      run: cmake --build . --config $BUILD_TYPE --target asm_tests
-
-    - name: ASM Test SVE256 Results move
      if: ${{ always() }}
-      shell: bash
-      working-directory: ${{runner.workspace}}/build
-      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_ASM_SVE256Bit.log || true
+      uses: ./.github/workflows/test
+      with:
+        target: asm_tests

    - name: ASM Tests - SVE128
-      working-directory: ${{runner.workspace}}/build
-      shell: bash
+      if: ${{ always() }}
+      uses: ./.github/workflows/test
      env:
        FEX_FORCESVEWIDTH: "128"
-      # Execute the unit tests
-      run: cmake --build . --config $BUILD_TYPE --target asm_tests
-
-    - name: ASM Test 128-bit 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_SVE128Bit.log || true
+      with:
+        target: asm_tests

    - name: ASM Tests - ASIMD
-      working-directory: ${{runner.workspace}}/build
-      shell: bash
+      if: ${{ always() }}
+      uses: ./.github/workflows/test
      env:
        FEX_HOSTFEATURES: "disablesve"
-      # Execute the unit tests
-      run: cmake --build . --config $BUILD_TYPE --target asm_tests
-
-    - name: ASM Test ASIMD Results move
-      if: ${{ always() }}
-      shell: bash
-      working-directory: ${{runner.workspace}}/build
-      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_ASM_ASIMD.log || true
-
-    - name: Truncate test results
-      if: ${{ always() }}
-      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
+      with:
+        target: asm_tests

    - name: Upload results
      if: ${{ always() }}
-      uses: 'actions/upload-artifact@v4'
+      uses: actions/upload-artifact@v6
      timeout-minutes: 1
      with:
-        name: Results-${{ env.runner_name }}
-        path: ${{runner.workspace}}/build/Testing/Temporary/LastTest_*.log
+        name: Results-${{ env.runner_name }}-${{ env.runner_label }}
+        path: results/*.log
        retention-days: 3
-
@@ -0,0 +1,35 @@
+name: Wine DLL Build
+description: Build a wow64 or arm64ec Wine DLL
+
+inputs:
+  target:
+    description: 'The target (arm64ec or wow64)'
+    required: true
+
+runs:
+  using: composite
+  steps:
+    - name: Clean Build Environment
+      shell: bash
+      run: rm -Rf build_${{ inputs.target }}
+
+    - name: Configure CMake
+      shell: bash
+      run: |
+        case "${{ inputs.target }}" in
+          wow64) _cc=aarch64 ;;
+          arm64ec) _cc=arm64ec ;;
+        esac
+
+        cmake -S . -B build_${{ inputs.target }} -DCMAKE_BUILD_TYPE=$BUILD_TYPE -DCMAKE_TOOLCHAIN_FILE=Data/CMake/toolchain_mingw.cmake \
+          -DMINGW_TRIPLE=${_cc}-w64-mingw32 -DCMAKE_INSTALL_LIBDIR=/usr/lib/wine/aarch64-windows -G Ninja \
+          -DENABLE_LTO=False -DENABLE_ASSERTIONS=False -DENABLE_JEMALLOC_GLIBC_ALLOC=False \
+          -DBUILD_TESTING=False -DCMAKE_INSTALL_PREFIX=/usr -DTUNE_ARCH=generic -DTUNE_CPU=none
+
+    - name: Build
+      shell: bash
+      run: cmake --build build_${{ inputs.target }}
+
+    - name: Install
+      shell: bash
+      run: DESTDIR="$PWD"/install cmake --build build_${{ inputs.target }} -t install
@@ -17,72 +17,39 @@ jobs:
      fail-fast: false

    steps:
-    - uses: actions/checkout@v3
+    - uses: actions/checkout@v6
+      with:
+        fetch-depth: '0'
+        fetch-tags: 'true'

    - name: Add MingGW to PATH
      run: echo "$HOME/llvm-mingw/build/bin/" >> $GITHUB_PATH

-    - name : submodule checkout
+    - name: Checkout Submodules
      # Need to update submodules
      run: |
        git submodule sync --recursive
        git submodule update --init --depth 1

    - name: Clean install directory
-      run: |
-        rm -Rf ${{runner.workspace}}/build_install
-        mkdir ${{runner.workspace}}/build_install
+      run: rm -Rf install

-    - name: Clean Build Environment
-      run: |
-        rm -Rf ${{runner.workspace}}/build_arm64ec
-        rm -Rf ${{runner.workspace}}/build_wow64
+    - name: Build (wow64)
+      uses: ./.github/workflows/wine_build
+      with:
+        target: wow64

-    - name: Create Build Environment arm64ec
-      run: |
-        cmake -E make_directory ${{runner.workspace}}/build_arm64ec
-        cmake -E make_directory ${{runner.workspace}}/build_wow64
-
-    - name: Configure CMake arm64ec
-      shell: bash
-      working-directory: ${{runner.workspace}}/build_arm64ec
-      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -DCMAKE_TOOLCHAIN_FILE=$GITHUB_WORKSPACE/Data/CMake/toolchain_mingw.cmake -DMINGW_TRIPLE=arm64ec-w64-mingw32 -DCMAKE_INSTALL_LIBDIR=/usr/lib/wine/aarch64-windows -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=False -DENABLE_JEMALLOC_GLIBC_ALLOC=False -DCMAKE_INSTALL_PREFIX=/usr -DBUILD_TESTING=False -DCMAKE_INSTALL_PREFIX=/usr
-
-    - name: Configure CMake wow64
-      shell: bash
-      working-directory: ${{runner.workspace}}/build_wow64
-      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -DCMAKE_TOOLCHAIN_FILE=$GITHUB_WORKSPACE/Data/CMake/toolchain_mingw.cmake -DMINGW_TRIPLE=aarch64-w64-mingw32 -DCMAKE_INSTALL_LIBDIR=/usr/lib/wine/aarch64-windows -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=False -DENABLE_JEMALLOC_GLIBC_ALLOC=False -DCMAKE_INSTALL_PREFIX=/usr -DBUILD_TESTING=False -DCMAKE_INSTALL_PREFIX=/usr
-
-    - name: Build arm64ec
-      working-directory: ${{runner.workspace}}/build_arm64ec
-      shell: bash
-      run: cmake --build . --config $BUILD_TYPE
-
-    - name: install arm64ec
-      working-directory: ${{runner.workspace}}/build_arm64ec
-      shell: bash
-      env:
-        DESTDIR: ${{runner.workspace}}/build_install
-      run: cmake --build . --config $BUILD_TYPE -t install
-
-    - name: Build wow64
-      working-directory: ${{runner.workspace}}/build_wow64
-      shell: bash
-      run: cmake --build . --config $BUILD_TYPE
-
-    - name: install wow64
-      working-directory: ${{runner.workspace}}/build_wow64
-      shell: bash
-      env:
-        DESTDIR: ${{runner.workspace}}/build_install
-      run: cmake --build . --config $BUILD_TYPE -t install
+    - name: Build (arm64ec)
+      uses: ./.github/workflows/wine_build
+      with:
+        target: arm64ec

    - name: Upload libraries
-      uses: 'actions/upload-artifact@v4'
+      uses: actions/upload-artifact@v6
      timeout-minutes: 1
      with:
        overwrite: true
        name: wine_dll_artifacts
-        path: ${{runner.workspace}}/build_install/usr/lib/wine/aarch64-windows/lib*.dll
+        path: ${{ github.workspace }}/install/usr/lib/wine/aarch64-windows/lib*.dll
        retention-days: 60
        compression-level: 9
@@ -11,3 +11,5 @@ out/
 .vs/
 *.pyc
 .cache
+.idea/
+CMakeLists.txt.user
@@ -0,0 +1,32 @@
+variables:
+    DEBIAN_FRONTEND: noninteractive
+    GIT_SUBMODULE_STRATEGY: recursive
+    GIT_DEPTH: 0
+    CC: clang
+    CXX: clang++
+
+aarch64:
+  image: registry.gitlab.steamos.cloud/steamrt/steamrt4/sdk/arm64:4.0.20251117.183306
+  tags:
+    - docker
+    - linux
+    - arm64
+    - aarch64
+  script:
+    - apt-get -y update
+    - apt-get install -y
+        git cmake ninja-build ccache
+        lld clang
+        libclang-dev llvm-dev
+        libstdc++-14-dev-i386-cross libgcc-14-dev-i386-cross
+        libstdc++-14-dev-amd64-cross libgcc-14-dev-amd64-cross
+    - cmake -E make_directory build/
+    - cmake -DCMAKE_BUILD_TYPE=Release -G Ninja -DBUILD_STEAM_SUPPORT=True -DENABLE_LTO=True -DENABLE_ASSERTIONS=False -DBUILD_THUNKS=True -DBUILD_FEXCONFIG=False -DBUILD_TESTING=False -DENABLE_CLANG_THUNKS=True -DUSE_LINKER=lld -DCMAKE_INSTALL_PREFIX=/usr -DTUNE_ARCH=armv8.2-a -DTUNE_CPU=none . -B build/
+    - cmake --build build/ --config Release
+    - DESTDIR=$(pwd)/install/ cmake --build build/ --config Release -t install
+
+  artifacts:
+    name: "steamrt artifacts"
+    untracked: false
+    paths:
+      - install/
@@ -17,9 +17,6 @@
  shallow = true
 	path = External/fex-gcc-target-tests-bins
 	url = https://github.com/FEX-Emu/fex-gcc-target-tests-bins.git
-[submodule "External/jemalloc"]
-	path = External/jemalloc
-	url = https://github.com/FEX-Emu/jemalloc.git
 [submodule "External/fmt"]
 	path = External/fmt
 	url = https://github.com/fmtlib/fmt.git
@@ -32,10 +29,6 @@
 [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
@@ -49,3 +42,13 @@
 [submodule "External/range-v3"]
 	path = External/range-v3
 	url = https://github.com/ericniebler/range-v3.git
+[submodule "External/zydis"]
+	shallow = true
+	path = External/zydis
+	url = https://github.com/zyantific/zydis.git
+[submodule "External/unordered_dense"]
+	path = External/unordered_dense
+	url = https://github.com/martinus/unordered_dense.git
+[submodule "External/rpmalloc"]
+	path = External/rpmalloc
+	url = https://github.com/FEX-Emu/rpmalloc.git
@@ -1,46 +1,50 @@
 cmake_minimum_required(VERSION 3.14)
 project(FEX C CXX ASM)

-INCLUDE (CheckIncludeFiles)
-CHECK_INCLUDE_FILES ("gdb/jit-reader.h" HAVE_GDB_JIT_READER_H)
+include(CheckIncludeFiles)
+check_include_files("gdb/jit-reader.h" HAVE_GDB_JIT_READER_H)

-option(BUILD_FEX_LINUX_TESTS "Build FEXLinuxTests, requires x86 compiler" FALSE)
+option(BUILD_FEX_LINUX_TESTS "Build FEXLinuxTests (requires x86 compiler)" FALSE)
 option(BUILD_THUNKS "Build thunks" FALSE)
 option(BUILD_FEXCONFIG "Build FEXConfig" TRUE)
 option(ENABLE_CLANG_THUNKS "Build thunks with clang" TRUE)
-option(ENABLE_IWYU "Enables include what you use program" FALSE)
+option(ENABLE_IWYU "Enable the Include What You Use sanitizer" FALSE)
 option(ENABLE_LTO "Enable LTO with compilation" TRUE)
 option(ENABLE_XRAY "Enable building with LLVM X-Ray" FALSE)
-set(USE_LINKER "" CACHE STRING "Allow overriding the linker path directly")
-option(ENABLE_UBSAN "Enables Clang UBSAN" FALSE)
-option(ENABLE_ASAN "Enables Clang ASAN" FALSE)
-option(ENABLE_TSAN "Enables Clang TSAN" FALSE)
-option(ENABLE_COVERAGE "Enables Coverage" FALSE)
-option(ENABLE_ASSERTIONS "Enables assertions in build" FALSE)
-option(ENABLE_GDB_SYMBOLS "Enables GDBSymbols integration support" ${HAVE_GDB_JIT_READER_H})
-option(ENABLE_STRICT_WERROR "Enables stricter -Werror for CI" FALSE)
-option(ENABLE_WERROR "Enables -Werror" FALSE)
-option(ENABLE_JEMALLOC "Enables jemalloc allocator" TRUE)
-option(ENABLE_JEMALLOC_GLIBC_ALLOC "Enables jemalloc glibc allocator" TRUE)
-option(ENABLE_OFFLINE_TELEMETRY "Enables FEX offline telemetry" TRUE)
-option(ENABLE_COMPILE_TIME_TRACE "Enables time trace compile option" FALSE)
-option(ENABLE_LIBCXX "Enables LLVM libc++" FALSE)
-option(ENABLE_CCACHE "Enables ccache for compile caching" TRUE)
-option(ENABLE_VIXL_SIMULATOR "Enable use of VIXL simulator for emulation (only useful for CI testing)" FALSE)
-option(ENABLE_VIXL_DISASSEMBLER "Enables debug disassembler output with VIXL" FALSE)
-option(USE_LEGACY_BINFMTMISC "Uses legacy method of setting up binfmt_misc" FALSE)
-option(ENABLE_FEXCORE_PROFILER "Enables use of the FEXCore timeline profiling capabilities" FALSE)
-set (FEXCORE_PROFILER_BACKEND "gpuvis" CACHE STRING "Set which backend to use for the FEXCore profiler (gpuvis, tracy)")
+set(USE_LINKER "" CACHE STRING "Path to a custom linker program")
+option(ENABLE_UBSAN "Enable the Clang Undefined Behavior Sanitizer" FALSE)
+option(ENABLE_ASAN "Enable the Clang Address Sanitizer" FALSE)
+option(ENABLE_TSAN "Enable the Clang Thread Sanitizer" FALSE)
+option(ENABLE_COVERAGE "Enable Code Coverage" FALSE)
+option(ENABLE_ASSERTIONS "Enable debug assertions" FALSE)
+option(ENABLE_GDB_SYMBOLS "Enable GDBSymbols integration support" ${HAVE_GDB_JIT_READER_H})
+option(ENABLE_STRICT_WERROR "Enable stricter -Werror" FALSE)
+option(ENABLE_WERROR "Enable -Werror" FALSE)
+option(ENABLE_FEX_ALLOCATOR "Enable allocator for FEX" TRUE)
+option(ENABLE_JEMALLOC_GLIBC_ALLOC "Enable jemalloc glibc allocator" TRUE)
+option(ENABLE_OFFLINE_TELEMETRY "Enable FEX offline telemetry" TRUE)
+option(ENABLE_COMPILE_TIME_TRACE "Enable time trace compile option" FALSE)
+option(ENABLE_LIBCXX "Use LLVM's libc++ instead of the GNU libstdc++" FALSE)
+option(ENABLE_CCACHE "Enable ccache for build caching" TRUE)
+option(ENABLE_VIXL_SIMULATOR "Use the VIXL simulator for emulation (only useful for CI testing)" FALSE)
+option(ENABLE_VIXL_DISASSEMBLER "Enable debug disassembler output with VIXL" FALSE)
+option(ENABLE_ZYDIS "Enable x86/x86-64 guest disassembler output with Zydis" FALSE)
+option(USE_LEGACY_BINFMTMISC "Use legacy method of setting up binfmt_misc" FALSE)
+option(ENABLE_FEXCORE_PROFILER "Enable FEXCore's timeline profiling capabilities" FALSE)
+set(FEXCORE_PROFILER_BACKEND "gpuvis" CACHE STRING "Set which backend to use for FEXCore's profiler")
+set_property(CACHE FEXCORE_PROFILER_BACKEND PROPERTY STRINGS gpuvis tracy)
 option(ENABLE_GLIBC_ALLOCATOR_HOOK_FAULT "Enables glibc memory allocation hooking with fault for CI testing")
-option(USE_PDB_DEBUGINFO "Builds debug info in PDB format" FALSE)
+option(USE_PDB_DEBUGINFO "Build debug info in PDB format" FALSE)
+option(BUILD_STEAM_SUPPORT "Enable Steam integration" FALSE)
+
+set(X86_32_TOOLCHAIN_FILE "${CMAKE_CURRENT_SOURCE_DIR}/Data/CMake/toolchain_x86_32.cmake" CACHE FILEPATH "Toolchain file for the (cross-)compiler targeting i686")
+set(X86_64_TOOLCHAIN_FILE "${CMAKE_CURRENT_SOURCE_DIR}/Data/CMake/toolchain_x86_64.cmake" CACHE FILEPATH "Toolchain file for the (cross-)compiler targeting x86_64")
+set(X86_DEV_ROOTFS "/" CACHE FILEPATH "Path to the sysroot used for cross-compiling for i686 and x86_64")
+set(DATA_DIRECTORY "" CACHE PATH "Global data directory (override)")
+set(HOSTLIBS_DATA_DIRECTORY "" CACHE PATH "Global data directory (override)")

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

 include(GNUInstallDirs)
@@ -48,43 +52,95 @@ if (NOT HOSTLIBS_DATA_DIRECTORY)
  set(HOSTLIBS_DATA_DIRECTORY "${CMAKE_INSTALL_FULL_LIBDIR}/fex-emu")
 endif()

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

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

+## Architecture Handling ##
+string(TOLOWER ${CMAKE_SYSTEM_PROCESSOR} processor)
+if (processor MATCHES "x86|amd64")
+  option(ENABLE_X86_HOST_DEBUG "Enables compiling on x86_64 host" FALSE)
+  if (NOT ENABLE_X86_HOST_DEBUG)
+    message(FATAL_ERROR
+      " FEX doesn't support compiling for x86-64 hosts!"
+      " This is /only/ a supported configuration for FEX CI and nothing else!")
+  else()
+    message(STATUS "x86_64 debug build")
+  endif()
+
+  set(ARCHITECTURE_x86_64 1)
+  add_compile_definitions(ARCHITECTURE_x86_64=1)
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mcx16")
+elseif (processor MATCHES "^aarch64|^arm64|^armv8\.*")
+  set(ARCHITECTURE_arm64 1)
+  add_compile_definitions(ARCHITECTURE_arm64=1)
+
+  # arm64ec needs to define both arm64 and arm64ec
+  if (processor MATCHES "^arm64ec")
+    set(ARCHITECTURE_arm64ec 1)
+    add_compile_definitions(ARCHITECTURE_arm64ec=1)
+  endif()
+endif()
+
+if (NOT (ARCHITECTURE_arm64 OR ARCHITECTURE_arm64ec OR ARCHITECTURE_x86_64))
+  message(FATAL_ERROR "Unsupported processor type ${processor}."
+    " If you believe this is in error, file an issue.")
+endif()
+
+if (BUILD_STEAM_SUPPORT)
+  add_compile_definitions(FEX_STEAM_SUPPORT=1)
+endif()
+
 if (ENABLE_FEXCORE_PROFILER)
-  add_definitions(-DENABLE_FEXCORE_PROFILER=1)
+  add_compile_definitions(ENABLE_FEXCORE_PROFILER=1)
  string(TOUPPER "${FEXCORE_PROFILER_BACKEND}" FEXCORE_PROFILER_BACKEND)

  if (FEXCORE_PROFILER_BACKEND STREQUAL "GPUVIS")
-    add_definitions(-DFEXCORE_PROFILER_BACKEND=1)
+    add_compile_definitions(FEXCORE_PROFILER_BACKEND=1)
  elseif (FEXCORE_PROFILER_BACKEND STREQUAL "TRACY")
-    add_definitions(-DFEXCORE_PROFILER_BACKEND=2)
-    add_definitions(-DTRACY_ENABLE=1)
+    add_compile_definitions(FEXCORE_PROFILER_BACKEND=2)
+    add_compile_definitions(TRACY_ENABLE=1)
    # Required so that Tracy will only start in the selected guest application
-    add_definitions(-DTRACY_MANUAL_LIFETIME=1)
-    add_definitions(-DTRACY_DELAYED_INIT=1)
+    add_compile_definitions(TRACY_MANUAL_LIFETIME=1)
+    add_compile_definitions(TRACY_DELAYED_INIT=1)
    # This interferes with FEX's signal handling
-    add_definitions(-DTRACY_NO_CRASH_HANDLER=1)
+    add_compile_definitions(TRACY_NO_CRASH_HANDLER=1)
    # Tracy can gather call stack samples in regular intervals, but this
    # isn't useful for us since it would usually sample opaque JIT code
-    add_definitions(-DTRACY_NO_SAMPLING=1)
+    add_compile_definitions(TRACY_NO_SAMPLING=1)
    # This pulls in libbacktrace which allocators in global constructors (before FEX can set up its allocator hooks)
-    add_definitions(-DTRACY_NO_CALLSTACK=1)
-    if (MINGW_BUILD)
-      message(FATAL_ERROR "Tracy profiler not supported")
+    add_compile_definitions(TRACY_NO_CALLSTACK=1)
+    if (MINGW)
+      message(FATAL_ERROR "Tracy profiler not supported on MinGW")
    endif()
  else()
    message(FATAL_ERROR "Unknown FEXCore profiler backend ${FEXCORE_PROFILER_BACKEND}")
@@ -96,7 +152,7 @@ if (ENABLE_JEMALLOC_GLIBC_ALLOC AND ENABLE_GLIBC_ALLOCATOR_HOOK_FAULT)
 endif()

 if (ENABLE_GLIBC_ALLOCATOR_HOOK_FAULT)
-  add_definitions(-DGLIBC_ALLOCATOR_FAULT=1)
+  add_compile_definitions(GLIBC_ALLOCATOR_FAULT=1)
 endif()

 # uninstall target
@@ -111,9 +167,10 @@ if(NOT TARGET uninstall)
 endif()

 # These options are meant for package management
-set (TUNE_CPU "native" CACHE STRING "Override the CPU the build is tuned for")
-set (TUNE_ARCH "generic" CACHE STRING "Override the Arch the build is tuned for")
-set (OVERRIDE_VERSION "detect" CACHE STRING "Override the FEX version in the format of <MMYY>{.<REV>}")
+set(TUNE_CPU "native" CACHE STRING "Override the CPU the build is tuned for")
+set(TUNE_ARCH "generic" CACHE STRING "Override the Arch the build is tuned for")
+set(OVERRIDE_VERSION "detect" CACHE STRING "Override the FEX version")
+set(OVERRIDE_HASH "detect" CACHE STRING "Override the FEX git hash")

 string(TOUPPER "${CMAKE_BUILD_TYPE}" CMAKE_BUILD_TYPE)
 if (CMAKE_BUILD_TYPE MATCHES "DEBUG")
@@ -122,15 +179,14 @@ endif()

 if (ENABLE_ASSERTIONS)
  message(STATUS "Assertions enabled")
-  add_definitions(-DASSERTIONS_ENABLED=1)
+  add_compile_definitions(ASSERTIONS_ENABLED=1)
 endif()

 if (ENABLE_GDB_SYMBOLS)
  message(STATUS "GDBSymbols support enabled")
-  add_definitions(-DGDB_SYMBOLS_ENABLED=1)
+  add_compile_definitions(GDB_SYMBOLS_ENABLED=1)
 endif()

-
 set(CMAKE_CXX_STANDARD 20)
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/Bin)
@@ -140,33 +196,7 @@ cmake_policy(SET CMP0083 NEW) # Follow new PIE policy
 include(CheckPIESupported)
 check_pie_supported()

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

 include(CheckCXXSourceCompiles)
 set(CMAKE_REQUIRED_FLAGS "-std=c++11 -Wattributes -Werror=attributes")
@@ -182,23 +212,33 @@ check_cxx_source_compiles(
  HAS_CLANG_PRESERVE_ALL)
 unset(CMAKE_REQUIRED_FLAGS)
 if (HAS_CLANG_PRESERVE_ALL)
-  if (MINGW_BUILD)
+  if (MINGW)
    message(STATUS "Ignoring broken clang::preserve_all support")
    set(HAS_CLANG_PRESERVE_ALL FALSE)
  else()
    message(STATUS "Has clang::preserve_all")
  endif()
-endif ()
-
-if (_M_ARM_64 AND HAS_CLANG_PRESERVE_ALL)
-  add_definitions("-DFEX_PRESERVE_ALL_ATTR=__attribute__((preserve_all))" "-DFEX_HAS_PRESERVE_ALL_ATTR=1")
-else()
-  add_definitions("-DFEX_PRESERVE_ALL_ATTR=" "-DFEX_HAS_PRESERVE_ALL_ATTR=0")
 endif()

+if (ARCHITECTURE_arm64 AND HAS_CLANG_PRESERVE_ALL)
+  add_compile_definitions("FEX_PRESERVE_ALL_ATTR=__attribute__((preserve_all))" "FEX_HAS_PRESERVE_ALL_ATTR=1")
+else()
+  add_compile_definitions("FEX_PRESERVE_ALL_ATTR=" "FEX_HAS_PRESERVE_ALL_ATTR=0")
+endif()
+
+check_cxx_source_compiles(
+  "
+  #define _GNU_SOURCE
+  #include <errno.h>
+  int main() {
+  return program_invocation_name == nullptr;
+  }"
+  HAS_PROGRAM_INVOCATION_NAME)
+add_compile_definitions("HAS_PROGRAM_INVOCATION_NAME=${HAS_PROGRAM_INVOCATION_NAME}")
+
 if (ENABLE_VIXL_SIMULATOR)
  # We can run the simulator on both x86-64 or AArch64 hosts
-  add_definitions(-DVIXL_SIMULATOR=1 -DVIXL_INCLUDE_SIMULATOR_AARCH64=1)
+  add_compile_definitions(VIXL_SIMULATOR=1 VIXL_INCLUDE_SIMULATOR_AARCH64=1)
 endif()

 if (ENABLE_CCACHE)
@@ -219,7 +259,7 @@ if (ENABLE_COMPILE_TIME_TRACE)
  link_libraries(-ftime-trace)
 endif()

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

 if (USE_LINKER)
  message(STATUS "Overriding linker to: ${USE_LINKER}")
@@ -234,7 +274,7 @@ endif()

 if (NOT ENABLE_OFFLINE_TELEMETRY)
  # Disable FEX offline telemetry entirely if asked
-  add_definitions(-DFEX_DISABLE_TELEMETRY=1)
+  add_compile_definitions(FEX_DISABLE_TELEMETRY=1)
 endif()

 if (ENABLE_UBSAN)
@@ -245,13 +285,13 @@ if (ENABLE_UBSAN)
  # that are regularly access unaligned.
  # function: syscalls cast function pointers to void (*)(unsigned long...), causing warnings
  # related to this access.
-  add_definitions(-DENABLE_UBSAN=1)
+  add_compile_definitions(ENABLE_UBSAN=1)
  add_compile_options(-fno-omit-frame-pointer -fsanitize=undefined -fno-sanitize=alignment -fno-sanitize=function -fno-sanitize-recover=undefined)
  link_libraries(-fno-omit-frame-pointer -fsanitize=undefined -fno-sanitize=alignment -fno-sanitize=function -fno-sanitize-recover=undefined)
 endif()

 if (ENABLE_ASAN)
-  add_definitions(-DENABLE_ASAN=1)
+  add_compile_definitions(ENABLE_ASAN=1)
  add_compile_options(-fno-omit-frame-pointer -fsanitize=address -fsanitize-address-use-after-scope)
  link_libraries(-fno-omit-frame-pointer -fsanitize=address -fsanitize-address-use-after-scope)
 endif()
@@ -271,20 +311,20 @@ if (ENABLE_JEMALLOC_GLIBC_ALLOC)
  # Required for thunks to work.
  # All host native libraries will use this allocator, while *most* other FEX internal allocations will use the other jemalloc allocator.
  add_subdirectory(External/jemalloc_glibc/)
-elseif (NOT MINGW_BUILD)
-  message (STATUS
+elseif (NOT MINGW)
+  message(STATUS
    " jemalloc glibc allocator disabled!\n"
    " This is not a recommended configuration!\n"
    " This will very explicitly break thunk execution!\n"
    " Use at your own risk!")
 endif()

-if (ENABLE_JEMALLOC)
-  # The jemalloc subproject that all FEXCore fextl objects allocate through.
-  add_subdirectory(External/jemalloc/)
-elseif (NOT MINGW_BUILD)
+if (ENABLE_FEX_ALLOCATOR)
+  # The rpmalloc subproject that all FEXCore fextl objects allocate through.
+  add_subdirectory(External/rpmalloc/)
+elseif (NOT MINGW)
  message (STATUS
-    " jemalloc disabled!\n"
+    " FEX allocator is disabled!\n"
    " This is not a recommended configuration!\n"
    " This will very explicitly break 32-bit application execution!\n"
    " Use at your own risk!")
@@ -295,45 +335,64 @@ if (USE_PDB_DEBUGINFO)
  add_link_options(-g -Wl,--pdb=)
 endif()

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

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

-include_directories(External/robin-map/include/)
+## Modules ##
+list(APPEND CMAKE_MODULE_PATH ${CMAKE_SOURCE_DIR}/Data/CMake/)
+
+include(LinkerGC)
+
+## Externals ##
+
+find_package(unordered_dense QUIET CONFIG)
+if (NOT unordered_dense_FOUND)
+  add_subdirectory(External/unordered_dense)
+endif()

 include(CTest)
 if (BUILD_TESTING OR ENABLE_VIXL_DISASSEMBLER OR ENABLE_VIXL_SIMULATOR)
  add_subdirectory(External/vixl/)
-  include_directories(SYSTEM External/vixl/src/)
+endif()
+
+if (ENABLE_ZYDIS)
+  find_package(Zycore 1.5 MODULE QUIET)
+  find_package(Zydis 4.0 MODULE QUIET)
+
+  if (TARGET Zydis::Zydis AND TARGET Zycore::Zycore)
+    message(STATUS "Using system Zydis")
+  else()
+    set(ZYDIS_BUILD_TOOLS OFF CACHE BOOL "" FORCE)
+    set(ZYDIS_BUILD_EXAMPLES OFF CACHE BOOL "" FORCE)
+
+    message(STATUS "Using bundled Zydis")
+    add_subdirectory(External/zydis/)
+  endif()
 endif()

 if (ENABLE_FEXCORE_PROFILER AND FEXCORE_PROFILER_BACKEND STREQUAL "TRACY")
  add_subdirectory(External/tracy)
 endif()

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

 set(BUILD_SHARED_LIBS OFF)

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

-add_definitions(-Wno-trigraphs)
-add_definitions(-DGLOBAL_DATA_DIRECTORY="${DATA_DIRECTORY}/")
+add_compile_options(-Wno-trigraphs)
+add_compile_definitions(GLOBAL_DATA_DIRECTORY="${DATA_DIRECTORY}/")

 if (BUILD_TESTING)
  find_package(Catch2 3 QUIET)
@@ -364,7 +423,6 @@ if (NOT range-v3_FOUND)
 endif()

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

 include_directories(Source/)
 include_directories("${CMAKE_BINARY_DIR}/Source/")
@@ -407,7 +465,7 @@ if (NOT TUNE_ARCH STREQUAL "generic")
 endif()

 if (TUNE_CPU STREQUAL "native")
-  if(_M_ARM_64)
+  if(ARCHITECTURE_arm64)
    if (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 999999.0)
      # Clang 12.0 fixed the -mcpu=native bug with mixed big.little implementers
      # Clang can not currently check for native Apple M1 type in hypervisor. Currently disabled
@@ -448,8 +506,54 @@ elseif (NOT TUNE_CPU STREQUAL "none")
  endif()
 endif()

+set(GIT_DESCRIBE_STRING "FEX-Unknown")
+
+if (OVERRIDE_VERSION STREQUAL "detect")
+  find_package(Git)
+
+  if (GIT_FOUND)
+    execute_process(
+      COMMAND ${GIT_EXECUTABLE} describe --abbrev=7
+      WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
+      OUTPUT_VARIABLE GIT_DESCRIBE_STRING
+      ERROR_QUIET
+      OUTPUT_STRIP_TRAILING_WHITESPACE)
+  endif()
+else()
+  set(GIT_DESCRIBE_STRING "${OVERRIDE_VERSION}")
+endif()
+
+set(GIT_HASH "Unknown")
+
+if (OVERRIDE_HASH STREQUAL "detect")
+  find_package(Git)
+
+  if (GIT_FOUND)
+    execute_process(
+      COMMAND ${GIT_EXECUTABLE} rev-parse HEAD
+      WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
+      OUTPUT_VARIABLE GIT_HASH
+      ERROR_QUIET
+      OUTPUT_STRIP_TRAILING_WHITESPACE)
+  endif()
+else()
+  set(GIT_HASH "${OVERRIDE_HASH}")
+endif()
+
+message(STATUS "FEX version: ${GIT_DESCRIBE_STRING}")
+message(STATUS "FEX commit: ${GIT_HASH}")
+
+# Prepends 0x to every two-character sequence in the hash,
+# OR the final character of the hash, to plumb it for C++ usage. e.g.:
+# -DOVERRIDE_HASH=123456aa => 0x12, 0x34, 0x56, 0xaa,
+# -DOVERRIDE_HASH=12345678a => 0x12, 0x34, 0x56, 0x78, 0xa,
+string(REGEX
+  REPLACE "(..|.$)" "0x\\1, "
+  GIT_HASH_ARRAY "${GIT_HASH}")
+
 if (ENABLE_IWYU)
-  find_program(IWYU_EXE "iwyu")
+  find_program(IWYU_EXE
+    NAMES iwyu include-what-you-use)
  if (IWYU_EXE)
    message(STATUS "IWYU enabled")
    set(CMAKE_CXX_INCLUDE_WHAT_YOU_USE "${IWYU_EXE}")
@@ -461,7 +565,7 @@ add_compile_options(-Wall)
 if (BUILD_TESTING)
  message(STATUS "Unit tests are enabled")

-  set (TEST_JOB_COUNT "" CACHE STRING "Override number of parallel jobs to use while running tests")
+  set(TEST_JOB_COUNT "" CACHE STRING "Override number of parallel jobs to use while running tests")
  if (TEST_JOB_COUNT)
    message(STATUS "Running tests with ${TEST_JOB_COUNT} jobs")
  elseif(CMAKE_VERSION VERSION_LESS "3.29")
@@ -476,13 +580,16 @@ add_subdirectory(FEXHeaderUtils/)
 add_subdirectory(CodeEmitter/)
 add_subdirectory(FEXCore/)

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

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

 # Install the ThunksDB file
 file(GLOB CONFIG_SOURCES CONFIGURE_DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/Data/*.json)
@@ -499,7 +606,7 @@ if (BUILD_TESTING)
 endif()

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

  # Thunk targets for both host libraries and IDE integration
@@ -526,8 +633,7 @@ if (BUILD_THUNKS)
      "-DX86_DEV_ROOTFS=${X86_DEV_ROOTFS}"
    INSTALL_COMMAND ""
    BUILD_ALWAYS ON
-    DEPENDS thunkgen
-  )
+    DEPENDS thunkgen)

  ExternalProject_Add(guest-libs-32
    PREFIX guest-libs-32
@@ -545,65 +651,36 @@ if (BUILD_THUNKS)
      "-DX86_DEV_ROOTFS=${X86_DEV_ROOTFS}"
    INSTALL_COMMAND ""
    BUILD_ALWAYS ON
-    DEPENDS thunkgen
-  )
+    DEPENDS thunkgen)

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

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

  add_custom_target(uninstall_guest-libs
    COMMAND ${CMAKE_COMMAND} "--build" "." "--target" "uninstall"
-    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest
-  )
+    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest)

  add_custom_target(uninstall_guest-libs-32
    COMMAND ${CMAKE_COMMAND} "--build" "." "--target" "uninstall"
-    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest_32
-  )
+    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest_32)

  add_dependencies(uninstall uninstall_guest-libs)
  add_dependencies(uninstall uninstall_guest-libs-32)
 endif()

-set(FEX_VERSION_MAJOR "0")
-set(FEX_VERSION_MINOR "0")
-set(FEX_VERSION_PATCH "0")
-
-if (OVERRIDE_VERSION STREQUAL "detect")
-  find_package(Git)
-  if (GIT_FOUND)
-    execute_process(
-      COMMAND ${GIT_EXECUTABLE} describe --abbrev=0
-      WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
-      OUTPUT_VARIABLE GIT_DESCRIBE_STRING
-      RESULT_VARIABLE GIT_ERROR
-      ERROR_QUIET
-      OUTPUT_STRIP_TRAILING_WHITESPACE
-    )
-
-    if (NOT ${GIT_ERROR} EQUAL 0)
-      # Likely built in a way that doesn't have tags
-      # Setup a version tag that is unknown
-      set(GIT_DESCRIBE_STRING "FEX-0000")
-    endif()
-  endif()
-else()
-  set(GIT_DESCRIBE_STRING "FEX-${OVERRIDE_VERSION}")
+if (BUILD_STEAM_SUPPORT)
+  add_subdirectory(Source/Steam/)
 endif()
@@ -129,4 +129,4 @@
 			"variables": []
 		}
 	]
-}
+}
@@ -38,9 +38,7 @@ public:

  [[nodiscard]] BranchEncodeSucceeded adr(ARMEmitter::Register rd, const BackwardLabel* Label) {
    int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    LOGMAN_THROW_A_FMT(IsADRRange(Imm), "Unscaled offset too large");
-
-    if (IsADRRange(Imm)) [[likely]] {
+    if (IsADRRange(Imm)) {
      constexpr uint32_t Op = 0b0001'0000 << 24;
      DataProcessing_PCRel_Imm(Op, rd, Imm);
      return BranchEncodeSucceeded::Success;
@@ -73,9 +71,8 @@ public:

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

-    if (IsADRPRange(Imm) && IsADRPAligned(Imm)) [[likely]] {
+    if (IsADRPRange(Imm) && IsADRPAligned(Imm)) {
      constexpr uint32_t Op = 0b1001'0000 << 24;
      DataProcessing_PCRel_Imm(Op, rd, Imm);
      return BranchEncodeSucceeded::Success;
@@ -103,16 +100,22 @@ public:
  }

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

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

      // First emit ADRP
      adrp(rd, ADRPImm >> 12);
@@ -125,14 +128,19 @@ public:
      return BranchEncodeSucceeded::Success;
    }

-    // Can't encode.
-    return BranchEncodeSucceeded::Failure;
+    // Stinky path, we need to load the address as a sequence of movz+movk+movk
+    movz(ARMEmitter::Size::i64Bit, rd, (UImm >> 32) & 0xFFFF, 32);
+    movk(ARMEmitter::Size::i64Bit, rd, (UImm >> 16) & 0xFFFF, 16);
+    movk(ARMEmitter::Size::i64Bit, rd, UImm & 0xFFFF);
+
+    return BranchEncodeSucceeded::Success;
  }
  [[nodiscard]] BranchEncodeSucceeded LongAddressGen(ARMEmitter::Register rd, ForwardLabel* Label) {
    AddLocationToLabel(Label, ForwardLabel::Reference {.Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::InstType::LONG_ADDRESS_GEN});
-    // Emit a register index and a nop. These will be backpatched.
+    // Emit a register index and two nops. These will be backpatched.
    dc32(rd.Idx());
    nop();
+    nop();

    // Forward label doesn't know if it can encode until Bind.
    return BranchEncodeSucceeded::Success;
@@ -22,7 +22,7 @@ public:
  }
  [[nodiscard]] BranchEncodeSucceeded b(ARMEmitter::Condition Cond, const BackwardLabel* Label) {
    int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) [[likely]] {
+    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) {
      constexpr uint32_t Op = 0b0101'010 << 25;
      Branch_Conditional(Op, 0, 0, Cond, Imm >> 2);
      return BranchEncodeSucceeded::Success;
@@ -55,7 +55,7 @@ public:
  }
  [[nodiscard]] BranchEncodeSucceeded bc(ARMEmitter::Condition Cond, const BackwardLabel* Label) {
    int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) [[likely]] {
+    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) {
      constexpr uint32_t Op = 0b0101'010 << 25;
      Branch_Conditional(Op, 0, 1, Cond, Imm >> 2);
      return BranchEncodeSucceeded::Success;
@@ -116,7 +116,7 @@ public:
  }
  [[nodiscard]] BranchEncodeSucceeded b(const BackwardLabel* Label) {
    int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    if (Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0)) [[likely]] {
+    if (Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0)) {
      constexpr uint32_t Op = 0b0001'01 << 26;
      UnconditionalBranch(Op, Imm >> 2);
      return BranchEncodeSucceeded::Success;
@@ -151,7 +151,7 @@ public:

  [[nodiscard]] BranchEncodeSucceeded bl(const BackwardLabel* Label) {
    int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    if (Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0)) [[likely]] {
+    if (Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0)) {
      constexpr uint32_t Op = 0b1001'01 << 26;
      UnconditionalBranch(Op, Imm >> 2);

@@ -189,7 +189,7 @@ public:
  [[nodiscard]] BranchEncodeSucceeded cbz(ARMEmitter::Size s, ARMEmitter::Register rt, const BackwardLabel* Label) {
    int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());

-    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) [[likely]] {
+    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) {
      constexpr uint32_t Op = 0b0011'0100 << 24;
      CompareAndBranch(Op, s, rt, Imm >> 2);
      return BranchEncodeSucceeded::Success;
@@ -227,7 +227,7 @@ public:
  [[nodiscard]] BranchEncodeSucceeded cbnz(ARMEmitter::Size s, ARMEmitter::Register rt, const BackwardLabel* Label) {
    int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());

-    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) [[likely]] {
+    if (Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0)) {
      constexpr uint32_t Op = 0b0011'0101 << 24;
      CompareAndBranch(Op, s, rt, Imm >> 2);
      return BranchEncodeSucceeded::Success;
@@ -265,7 +265,7 @@ public:
  [[nodiscard]] BranchEncodeSucceeded tbz(ARMEmitter::Register rt, uint32_t Bit, const BackwardLabel* Label) {
    int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());

-    if (Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0)) [[likely]] {
+    if (Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0)) {
      constexpr uint32_t Op = 0b0011'0110 << 24;
      TestAndBranch(Op, rt, Bit, Imm >> 2);
      return BranchEncodeSucceeded::Success;
@@ -301,9 +301,8 @@ public:
  }
  [[nodiscard]] BranchEncodeSucceeded tbnz(ARMEmitter::Register rt, uint32_t Bit, const BackwardLabel* Label) {
    int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
-    LOGMAN_THROW_A_FMT(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0), "Unscaled offset too large");

-    if (Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0)) [[likely]] {
+    if (Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0)) {
      constexpr uint32_t Op = 0b0011'0111 << 24;
      TestAndBranch(Op, rt, Bit, Imm >> 2);
      return BranchEncodeSucceeded::Success;
@@ -53,6 +53,7 @@ public:
    if (!CurrentAlignment) {
      return;
    }
+    std::memset(CurrentOffset, 0, Size - CurrentAlignment);
    CurrentOffset += Size - CurrentAlignment;
  }

@@ -586,6 +586,10 @@ concept IsXOrWRegister = std::is_same_v<T, XRegister> || std::is_same_v<T, WRegi
 template<typename T>
 concept IsQOrDRegister = std::is_same_v<T, QRegister> || std::is_same_v<T, DRegister>;

+template<typename T>
+concept IsLabel = std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
+                  std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>;
+
 enum class BranchEncodeSucceeded {
  Success,
  Failure,
@@ -658,7 +662,7 @@ public:
    case ForwardLabel::InstType::ADR: {
      uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
      int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
-      if (!IsADRRange(Imm)) [[unlikely]] {
+      if (!IsADRRange(Imm)) {
        // Can't bind.
        return false;
      }
@@ -674,7 +678,7 @@ public:
      uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
      int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);

-      if (!(IsADRPRange(Imm) && IsADRPAligned(Imm))) [[unlikely]] {
+      if (!(IsADRPRange(Imm) && IsADRPAligned(Imm))) {
        // Can't bind.
        return false;
      }
@@ -691,7 +695,7 @@ public:
    case ForwardLabel::InstType::B: {
      uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
      int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
-      if (!(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0))) [[unlikely]] {
+      if (!(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0))) {
        // Can't bind.
        return false;
      }
@@ -707,7 +711,7 @@ public:
    case ForwardLabel::InstType::TEST_BRANCH: {
      uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
      int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
-      if (!(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0))) [[unlikely]] {
+      if (!(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0))) {
        // Can't bind.
        return false;
      }
@@ -724,7 +728,7 @@ public:
    case ForwardLabel::InstType::RELATIVE_LOAD: {
      uint32_t* Instruction = reinterpret_cast<uint32_t*>(Label->Location);
      int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
-      if (!(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0))) [[unlikely]] {
+      if (!(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0))) {
        // Can't bind.
        return false;
      }
@@ -737,38 +741,44 @@ public:
      break;
    }
    case ForwardLabel::InstType::LONG_ADDRESS_GEN: {
-      uint32_t* Instructions = reinterpret_cast<uint32_t*>(Label->Location);
-      int64_t ImmInstOne = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[0]);
-      int64_t ImmInstTwo = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[1]);
-      auto OriginalOffset = GetCursorOffset();
+      const auto* Instructions = reinterpret_cast<uint32_t*>(Label->Location);
+      const auto ImmInstOne = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[0]);
+      const auto ImmInstTwo = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[1]);
+      const auto ImmInstThree = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(&Instructions[2]);
+      const auto OriginalOffset = GetCursorOffset();

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

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

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

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

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

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

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

-    const auto bits = FEXCore::BitCast<uint64_t>(value);
+    const auto bits = std::bit_cast<uint64_t>(value);
    const auto sign = (bits & 0x80000000'00000000) >> 56;
    const auto expb2 = (bits & 0x20000000'00000000) >> 55;
    const auto b5_to_0 = (bits >> 48) & 0x3F;
@@ -15,13 +15,10 @@ foreach(GEN_CONFIG_SRC ${GEN_CONFIG_SOURCES})
  get_filename_component(CONFIG_NAME ${GEN_CONFIG_SRC} NAME_WLE)

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

  # Then install the configured json
-  install(
-    FILES ${CMAKE_BINARY_DIR}/Data/AppConfig/${CONFIG_NAME}
+  install(FILES ${CMAKE_BINARY_DIR}/Data/AppConfig/${CONFIG_NAME}
    DESTINATION ${DATA_DIRECTORY}/AppConfig/
    COMPONENT Runtime)
 endforeach()
@@ -0,0 +1,23 @@
+# SPDX-License-Identifier: MIT
+
+if (CMAKE_CROSSCOMPILING)
+    return()
+endif()
+
+include(FindPackageHandleStandardArgs)
+
+find_package(Zycore QUIET CONFIG)
+
+if (Zycore_CONSIDERED_CONFIGS)
+    find_package_handle_standard_args(Zycore CONFIG_MODE)
+else()
+    find_package(PkgConfig QUIET)
+    pkg_search_module(Zycore QUIET IMPORTED_TARGET zycore)
+    find_package_handle_standard_args(Zycore
+        REQUIRED_VARS zycore_LINK_LIBRARIES
+        VERSION_VAR zycore_VERSION)
+
+    if (TARGET PkgConfig::zycore)
+      add_library(Zycore::Zycore ALIAS PkgConfig::zycore)
+    endif()
+endif()
@@ -0,0 +1,23 @@
+# SPDX-License-Identifier: MIT
+
+if (CMAKE_CROSSCOMPILING)
+    return()
+endif()
+
+include(FindPackageHandleStandardArgs)
+
+find_package(Zydis QUIET CONFIG)
+
+if (Zydis_CONSIDERED_CONFIGS)
+    find_package_handle_standard_args(Zydis CONFIG_MODE)
+else()
+    find_package(PkgConfig QUIET)
+    pkg_search_module(Zydis QUIET IMPORTED_TARGET zydis)
+    find_package_handle_standard_args(Zydis
+        REQUIRED_VARS zydis_LINK_LIBRARIES
+        VERSION_VAR zydis_VERSION)
+
+    if (TARGET PkgConfig::zydis)
+      add_library(Zydis::Zydis ALIAS PkgConfig::zydis)
+    endif()
+endif()
@@ -0,0 +1,18 @@
+# SPDX-License-Identifier: MIT
+
+include(FindPackageHandleStandardArgs)
+
+find_package(PkgConfig QUIET)
+pkg_search_module(xxhash QUIET IMPORTED_TARGET xxhash libxxhash)
+find_package_handle_standard_args(xxhash
+    REQUIRED_VARS xxhash_LINK_LIBRARIES
+    VERSION_VAR xxhash_VERSION
+)
+
+if (xxhash_FOUND AND NOT TARGET xxHash::xxhash)
+    if (TARGET PkgConfig::xxhash)
+        add_library(xxHash::xxhash ALIAS PkgConfig::xxhash)
+    else()
+        add_library(xxHash::xxhash ALIAS xxhash)
+    endif()
+endif()
@@ -0,0 +1,15 @@
+# SPDX-License-Identifier: MIT
+
+# This applies some common linker options that reduce code size and linking time in Release mode. Namely:
+# --gc-sections: Linktime garbage collection, discards unused sections from the final output
+# --strip-all  : Similar to running `strip`, discards the symbol table from the final output
+# --as-needed  : Only includes libraries that are actually needed in the final output.
+
+macro(LinkerGC target)
+  if (CMAKE_BUILD_TYPE MATCHES "RELEASE")
+    target_link_options(${target} PRIVATE
+      "LINKER:--gc-sections"
+      "LINKER:--strip-all"
+      "LINKER:--as-needed")
+  endif()
+endmacro()
@@ -3,13 +3,10 @@ function(GenBinFmt Name)
  get_filename_component(FMT_NAME ${Name} NAME_WE)

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

  # Then install the configured binfmt
-  install(
-    FILES ${CMAKE_BINARY_DIR}/Data/binfmts/${FMT_NAME}
+  install(FILES ${CMAKE_BINARY_DIR}/Data/binfmts/${FMT_NAME}
    DESTINATION ${CMAKE_INSTALL_PREFIX}/share/binfmts/
    COMPONENT Runtime)
 endfunction()
@@ -18,8 +15,7 @@ if (NOT USE_LEGACY_BINFMTMISC)
  configure_file(FEX-x86.conf.in ${CMAKE_BINARY_DIR}/Data/binfmts/FEX-x86.conf)
  configure_file(FEX-x86_64.conf.in ${CMAKE_BINARY_DIR}/Data/binfmts/FEX-x86_64.conf)

-  install(
-    FILES ${CMAKE_BINARY_DIR}/Data/binfmts/FEX-x86.conf ${CMAKE_BINARY_DIR}/Data/binfmts/FEX-x86_64.conf
+  install(FILES ${CMAKE_BINARY_DIR}/Data/binfmts/FEX-x86.conf ${CMAKE_BINARY_DIR}/Data/binfmts/FEX-x86_64.conf
    DESTINATION ${CMAKE_INSTALL_PREFIX}/lib/binfmt.d/
    COMPONENT Runtime)
 else()
@@ -2,8 +2,8 @@

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

  cmakeToolchainFile = pkgs.substitute {
@@ -1,5 +1,5 @@

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

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

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

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

 # 128-bit library
-add_library(cephes_128bit STATIC ${SRCS_128BIT})
 target_link_libraries(cephes_128bit softfloat_3e)
 target_include_directories(cephes_128bit PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}/include/)
 target_compile_options(cephes_128bit PRIVATE -fno-builtin)
@@ -36,71 +36,91 @@ certifi==2025.7.14 \
    # via
    #   -r requirements_formatting.txt.in
    #   requests
-cffi==1.15.1 \
-    --hash=sha256:00a9ed42e88df81ffae7a8ab6d9356b371399b91dbdf0c3cb1e84c03a13aceb5 \
-    --hash=sha256:03425bdae262c76aad70202debd780501fabeaca237cdfddc008987c0e0f59ef \
-    --hash=sha256:04ed324bda3cda42b9b695d51bb7d54b680b9719cfab04227cdd1e04e5de3104 \
-    --hash=sha256:0e2642fe3142e4cc4af0799748233ad6da94c62a8bec3a6648bf8ee68b1c7426 \
-    --hash=sha256:173379135477dc8cac4bc58f45db08ab45d228b3363adb7af79436135d028405 \
-    --hash=sha256:198caafb44239b60e252492445da556afafc7d1e3ab7a1fb3f0584ef6d742375 \
-    --hash=sha256:1e74c6b51a9ed6589199c787bf5f9875612ca4a8a0785fb2d4a84429badaf22a \
-    --hash=sha256:2012c72d854c2d03e45d06ae57f40d78e5770d252f195b93f581acf3ba44496e \
-    --hash=sha256:21157295583fe8943475029ed5abdcf71eb3911894724e360acff1d61c1d54bc \
-    --hash=sha256:2470043b93ff09bf8fb1d46d1cb756ce6132c54826661a32d4e4d132e1977adf \
-    --hash=sha256:285d29981935eb726a4399badae8f0ffdff4f5050eaa6d0cfc3f64b857b77185 \
-    --hash=sha256:30d78fbc8ebf9c92c9b7823ee18eb92f2e6ef79b45ac84db507f52fbe3ec4497 \
-    --hash=sha256:320dab6e7cb2eacdf0e658569d2575c4dad258c0fcc794f46215e1e39f90f2c3 \
-    --hash=sha256:33ab79603146aace82c2427da5ca6e58f2b3f2fb5da893ceac0c42218a40be35 \
-    --hash=sha256:3548db281cd7d2561c9ad9984681c95f7b0e38881201e157833a2342c30d5e8c \
-    --hash=sha256:3799aecf2e17cf585d977b780ce79ff0dc9b78d799fc694221ce814c2c19db83 \
-    --hash=sha256:39d39875251ca8f612b6f33e6b1195af86d1b3e60086068be9cc053aa4376e21 \
-    --hash=sha256:3b926aa83d1edb5aa5b427b4053dc420ec295a08e40911296b9eb1b6170f6cca \
-    --hash=sha256:3bcde07039e586f91b45c88f8583ea7cf7a0770df3a1649627bf598332cb6984 \
-    --hash=sha256:3d08afd128ddaa624a48cf2b859afef385b720bb4b43df214f85616922e6a5ac \
-    --hash=sha256:3eb6971dcff08619f8d91607cfc726518b6fa2a9eba42856be181c6d0d9515fd \
-    --hash=sha256:40f4774f5a9d4f5e344f31a32b5096977b5d48560c5592e2f3d2c4374bd543ee \
-    --hash=sha256:4289fc34b2f5316fbb762d75362931e351941fa95fa18789191b33fc4cf9504a \
-    --hash=sha256:470c103ae716238bbe698d67ad020e1db9d9dba34fa5a899b5e21577e6d52ed2 \
-    --hash=sha256:4f2c9f67e9821cad2e5f480bc8d83b8742896f1242dba247911072d4fa94c192 \
-    --hash=sha256:50a74364d85fd319352182ef59c5c790484a336f6db772c1a9231f1c3ed0cbd7 \
-    --hash=sha256:54a2db7b78338edd780e7ef7f9f6c442500fb0d41a5a4ea24fff1c929d5af585 \
-    --hash=sha256:5635bd9cb9731e6d4a1132a498dd34f764034a8ce60cef4f5319c0541159392f \
-    --hash=sha256:59c0b02d0a6c384d453fece7566d1c7e6b7bae4fc5874ef2ef46d56776d61c9e \
-    --hash=sha256:5d598b938678ebf3c67377cdd45e09d431369c3b1a5b331058c338e201f12b27 \
-    --hash=sha256:5df2768244d19ab7f60546d0c7c63ce1581f7af8b5de3eb3004b9b6fc8a9f84b \
-    --hash=sha256:5ef34d190326c3b1f822a5b7a45f6c4535e2f47ed06fec77d3d799c450b2651e \
-    --hash=sha256:6975a3fac6bc83c4a65c9f9fcab9e47019a11d3d2cf7f3c0d03431bf145a941e \
-    --hash=sha256:6c9a799e985904922a4d207a94eae35c78ebae90e128f0c4e521ce339396be9d \
-    --hash=sha256:70df4e3b545a17496c9b3f41f5115e69a4f2e77e94e1d2a8e1070bc0c38c8a3c \
-    --hash=sha256:7473e861101c9e72452f9bf8acb984947aa1661a7704553a9f6e4baa5ba64415 \
-    --hash=sha256:8102eaf27e1e448db915d08afa8b41d6c7ca7a04b7d73af6514df10a3e74bd82 \
-    --hash=sha256:87c450779d0914f2861b8526e035c5e6da0a3199d8f1add1a665e1cbc6fc6d02 \
-    --hash=sha256:8b7ee99e510d7b66cdb6c593f21c043c248537a32e0bedf02e01e9553a172314 \
-    --hash=sha256:91fc98adde3d7881af9b59ed0294046f3806221863722ba7d8d120c575314325 \
-    --hash=sha256:94411f22c3985acaec6f83c6df553f2dbe17b698cc7f8ae751ff2237d96b9e3c \
-    --hash=sha256:98d85c6a2bef81588d9227dde12db8a7f47f639f4a17c9ae08e773aa9c697bf3 \
-    --hash=sha256:9ad5db27f9cabae298d151c85cf2bad1d359a1b9c686a275df03385758e2f914 \
-    --hash=sha256:a0b71b1b8fbf2b96e41c4d990244165e2c9be83d54962a9a1d118fd8657d2045 \
-    --hash=sha256:a0f100c8912c114ff53e1202d0078b425bee3649ae34d7b070e9697f93c5d52d \
-    --hash=sha256:a591fe9e525846e4d154205572a029f653ada1a78b93697f3b5a8f1f2bc055b9 \
-    --hash=sha256:a5c84c68147988265e60416b57fc83425a78058853509c1b0629c180094904a5 \
-    --hash=sha256:a66d3508133af6e8548451b25058d5812812ec3798c886bf38ed24a98216fab2 \
-    --hash=sha256:a8c4917bd7ad33e8eb21e9a5bbba979b49d9a97acb3a803092cbc1133e20343c \
-    --hash=sha256:b3bbeb01c2b273cca1e1e0c5df57f12dce9a4dd331b4fa1635b8bec26350bde3 \
-    --hash=sha256:cba9d6b9a7d64d4bd46167096fc9d2f835e25d7e4c121fb2ddfc6528fb0413b2 \
-    --hash=sha256:cc4d65aeeaa04136a12677d3dd0b1c0c94dc43abac5860ab33cceb42b801c1e8 \
-    --hash=sha256:ce4bcc037df4fc5e3d184794f27bdaab018943698f4ca31630bc7f84a7b69c6d \
-    --hash=sha256:cec7d9412a9102bdc577382c3929b337320c4c4c4849f2c5cdd14d7368c5562d \
-    --hash=sha256:d400bfb9a37b1351253cb402671cea7e89bdecc294e8016a707f6d1d8ac934f9 \
-    --hash=sha256:d61f4695e6c866a23a21acab0509af1cdfd2c013cf256bbf5b6b5e2695827162 \
-    --hash=sha256:db0fbb9c62743ce59a9ff687eb5f4afbe77e5e8403d6697f7446e5f609976f76 \
-    --hash=sha256:dd86c085fae2efd48ac91dd7ccffcfc0571387fe1193d33b6394db7ef31fe2a4 \
-    --hash=sha256:e00b098126fd45523dd056d2efba6c5a63b71ffe9f2bbe1a4fe1716e1d0c331e \
-    --hash=sha256:e229a521186c75c8ad9490854fd8bbdd9a0c9aa3a524326b55be83b54d4e0ad9 \
-    --hash=sha256:e263d77ee3dd201c3a142934a086a4450861778baaeeb45db4591ef65550b0a6 \
-    --hash=sha256:ed9cb427ba5504c1dc15ede7d516b84757c3e3d7868ccc85121d9310d27eed0b \
-    --hash=sha256:fa6693661a4c91757f4412306191b6dc88c1703f780c8234035eac011922bc01 \
-    --hash=sha256:fcd131dd944808b5bdb38e6f5b53013c5aa4f334c5cad0c72742f6eba4b73db0
+cffi==2.0.0 \
+    --hash=sha256:00bdf7acc5f795150faa6957054fbbca2439db2f775ce831222b66f192f03beb \
+    --hash=sha256:07b271772c100085dd28b74fa0cd81c8fb1a3ba18b21e03d7c27f3436a10606b \
+    --hash=sha256:087067fa8953339c723661eda6b54bc98c5625757ea62e95eb4898ad5e776e9f \
+    --hash=sha256:0a1527a803f0a659de1af2e1fd700213caba79377e27e4693648c2923da066f9 \
+    --hash=sha256:0cf2d91ecc3fcc0625c2c530fe004f82c110405f101548512cce44322fa8ac44 \
+    --hash=sha256:0f6084a0ea23d05d20c3edcda20c3d006f9b6f3fefeac38f59262e10cef47ee2 \
+    --hash=sha256:12873ca6cb9b0f0d3a0da705d6086fe911591737a59f28b7936bdfed27c0d47c \
+    --hash=sha256:19f705ada2530c1167abacb171925dd886168931e0a7b78f5bffcae5c6b5be75 \
+    --hash=sha256:1cd13c99ce269b3ed80b417dcd591415d3372bcac067009b6e0f59c7d4015e65 \
+    --hash=sha256:1e3a615586f05fc4065a8b22b8152f0c1b00cdbc60596d187c2a74f9e3036e4e \
+    --hash=sha256:1f72fb8906754ac8a2cc3f9f5aaa298070652a0ffae577e0ea9bd480dc3c931a \
+    --hash=sha256:1fc9ea04857caf665289b7a75923f2c6ed559b8298a1b8c49e59f7dd95c8481e \
+    --hash=sha256:203a48d1fb583fc7d78a4c6655692963b860a417c0528492a6bc21f1aaefab25 \
+    --hash=sha256:2081580ebb843f759b9f617314a24ed5738c51d2aee65d31e02f6f7a2b97707a \
+    --hash=sha256:21d1152871b019407d8ac3985f6775c079416c282e431a4da6afe7aefd2bccbe \
+    --hash=sha256:24b6f81f1983e6df8db3adc38562c83f7d4a0c36162885ec7f7b77c7dcbec97b \
+    --hash=sha256:256f80b80ca3853f90c21b23ee78cd008713787b1b1e93eae9f3d6a7134abd91 \
+    --hash=sha256:28a3a209b96630bca57cce802da70c266eb08c6e97e5afd61a75611ee6c64592 \
+    --hash=sha256:2c8f814d84194c9ea681642fd164267891702542f028a15fc97d4674b6206187 \
+    --hash=sha256:2de9a304e27f7596cd03d16f1b7c72219bd944e99cc52b84d0145aefb07cbd3c \
+    --hash=sha256:38100abb9d1b1435bc4cc340bb4489635dc2f0da7456590877030c9b3d40b0c1 \
+    --hash=sha256:3925dd22fa2b7699ed2617149842d2e6adde22b262fcbfada50e3d195e4b3a94 \
+    --hash=sha256:3e17ed538242334bf70832644a32a7aae3d83b57567f9fd60a26257e992b79ba \
+    --hash=sha256:3e837e369566884707ddaf85fc1744b47575005c0a229de3327f8f9a20f4efeb \
+    --hash=sha256:3f4d46d8b35698056ec29bca21546e1551a205058ae1a181d871e278b0b28165 \
+    --hash=sha256:44d1b5909021139fe36001ae048dbdde8214afa20200eda0f64c068cac5d5529 \
+    --hash=sha256:45d5e886156860dc35862657e1494b9bae8dfa63bf56796f2fb56e1679fc0bca \
+    --hash=sha256:4647afc2f90d1ddd33441e5b0e85b16b12ddec4fca55f0d9671fef036ecca27c \
+    --hash=sha256:4671d9dd5ec934cb9a73e7ee9676f9362aba54f7f34910956b84d727b0d73fb6 \
+    --hash=sha256:53f77cbe57044e88bbd5ed26ac1d0514d2acf0591dd6bb02a3ae37f76811b80c \
+    --hash=sha256:5eda85d6d1879e692d546a078b44251cdd08dd1cfb98dfb77b670c97cee49ea0 \
+    --hash=sha256:5fed36fccc0612a53f1d4d9a816b50a36702c28a2aa880cb8a122b3466638743 \
+    --hash=sha256:61d028e90346df14fedc3d1e5441df818d095f3b87d286825dfcbd6459b7ef63 \
+    --hash=sha256:66f011380d0e49ed280c789fbd08ff0d40968ee7b665575489afa95c98196ab5 \
+    --hash=sha256:6824f87845e3396029f3820c206e459ccc91760e8fa24422f8b0c3d1731cbec5 \
+    --hash=sha256:6c6c373cfc5c83a975506110d17457138c8c63016b563cc9ed6e056a82f13ce4 \
+    --hash=sha256:6d02d6655b0e54f54c4ef0b94eb6be0607b70853c45ce98bd278dc7de718be5d \
+    --hash=sha256:6d50360be4546678fc1b79ffe7a66265e28667840010348dd69a314145807a1b \
+    --hash=sha256:730cacb21e1bdff3ce90babf007d0a0917cc3e6492f336c2f0134101e0944f93 \
+    --hash=sha256:737fe7d37e1a1bffe70bd5754ea763a62a066dc5913ca57e957824b72a85e205 \
+    --hash=sha256:74a03b9698e198d47562765773b4a8309919089150a0bb17d829ad7b44b60d27 \
+    --hash=sha256:7553fb2090d71822f02c629afe6042c299edf91ba1bf94951165613553984512 \
+    --hash=sha256:7a66c7204d8869299919db4d5069a82f1561581af12b11b3c9f48c584eb8743d \
+    --hash=sha256:7cc09976e8b56f8cebd752f7113ad07752461f48a58cbba644139015ac24954c \
+    --hash=sha256:81afed14892743bbe14dacb9e36d9e0e504cd204e0b165062c488942b9718037 \
+    --hash=sha256:8941aaadaf67246224cee8c3803777eed332a19d909b47e29c9842ef1e79ac26 \
+    --hash=sha256:89472c9762729b5ae1ad974b777416bfda4ac5642423fa93bd57a09204712322 \
+    --hash=sha256:8ea985900c5c95ce9db1745f7933eeef5d314f0565b27625d9a10ec9881e1bfb \
+    --hash=sha256:8eca2a813c1cb7ad4fb74d368c2ffbbb4789d377ee5bb8df98373c2cc0dee76c \
+    --hash=sha256:92b68146a71df78564e4ef48af17551a5ddd142e5190cdf2c5624d0c3ff5b2e8 \
+    --hash=sha256:9332088d75dc3241c702d852d4671613136d90fa6881da7d770a483fd05248b4 \
+    --hash=sha256:94698a9c5f91f9d138526b48fe26a199609544591f859c870d477351dc7b2414 \
+    --hash=sha256:9a67fc9e8eb39039280526379fb3a70023d77caec1852002b4da7e8b270c4dd9 \
+    --hash=sha256:9de40a7b0323d889cf8d23d1ef214f565ab154443c42737dfe52ff82cf857664 \
+    --hash=sha256:a05d0c237b3349096d3981b727493e22147f934b20f6f125a3eba8f994bec4a9 \
+    --hash=sha256:afb8db5439b81cf9c9d0c80404b60c3cc9c3add93e114dcae767f1477cb53775 \
+    --hash=sha256:b18a3ed7d5b3bd8d9ef7a8cb226502c6bf8308df1525e1cc676c3680e7176739 \
+    --hash=sha256:b1e74d11748e7e98e2f426ab176d4ed720a64412b6a15054378afdb71e0f37dc \
+    --hash=sha256:b21e08af67b8a103c71a250401c78d5e0893beff75e28c53c98f4de42f774062 \
+    --hash=sha256:b4c854ef3adc177950a8dfc81a86f5115d2abd545751a304c5bcf2c2c7283cfe \
+    --hash=sha256:b882b3df248017dba09d6b16defe9b5c407fe32fc7c65a9c69798e6175601be9 \
+    --hash=sha256:baf5215e0ab74c16e2dd324e8ec067ef59e41125d3eade2b863d294fd5035c92 \
+    --hash=sha256:c649e3a33450ec82378822b3dad03cc228b8f5963c0c12fc3b1e0ab940f768a5 \
+    --hash=sha256:c654de545946e0db659b3400168c9ad31b5d29593291482c43e3564effbcee13 \
+    --hash=sha256:c6638687455baf640e37344fe26d37c404db8b80d037c3d29f58fe8d1c3b194d \
+    --hash=sha256:c8d3b5532fc71b7a77c09192b4a5a200ea992702734a2e9279a37f2478236f26 \
+    --hash=sha256:cb527a79772e5ef98fb1d700678fe031e353e765d1ca2d409c92263c6d43e09f \
+    --hash=sha256:cf364028c016c03078a23b503f02058f1814320a56ad535686f90565636a9495 \
+    --hash=sha256:d48a880098c96020b02d5a1f7d9251308510ce8858940e6fa99ece33f610838b \
+    --hash=sha256:d68b6cef7827e8641e8ef16f4494edda8b36104d79773a334beaa1e3521430f6 \
+    --hash=sha256:d9b29c1f0ae438d5ee9acb31cadee00a58c46cc9c0b2f9038c6b0b3470877a8c \
+    --hash=sha256:d9b97165e8aed9272a6bb17c01e3cc5871a594a446ebedc996e2397a1c1ea8ef \
+    --hash=sha256:da68248800ad6320861f129cd9c1bf96ca849a2771a59e0344e88681905916f5 \
+    --hash=sha256:da902562c3e9c550df360bfa53c035b2f241fed6d9aef119048073680ace4a18 \
+    --hash=sha256:dbd5c7a25a7cb98f5ca55d258b103a2054f859a46ae11aaf23134f9cc0d356ad \
+    --hash=sha256:dd4f05f54a52fb558f1ba9f528228066954fee3ebe629fc1660d874d040ae5a3 \
+    --hash=sha256:de8dad4425a6ca6e4e5e297b27b5c824ecc7581910bf9aee86cb6835e6812aa7 \
+    --hash=sha256:e11e82b744887154b182fd3e7e8512418446501191994dbf9c9fc1f32cc8efd5 \
+    --hash=sha256:e6e73b9e02893c764e7e8d5bb5ce277f1a009cd5243f8228f75f842bf937c534 \
+    --hash=sha256:f73b96c41e3b2adedc34a7356e64c8eb96e03a3782b535e043a986276ce12a49 \
+    --hash=sha256:f93fd8e5c8c0a4aa1f424d6173f14a892044054871c771f8566e4008eaa359d2 \
+    --hash=sha256:fc33c5141b55ed366cfaad382df24fe7dcbc686de5be719b207bb248e3053dc5 \
+    --hash=sha256:fc7de24befaeae77ba923797c7c87834c73648a05a4bde34b3b7e5588973a453 \
+    --hash=sha256:fe562eb1a64e67dd297ccc4f5addea2501664954f2692b69a76449ec7913ecbf
    # via
    #   cryptography
    #   pynacl
@@ -185,44 +205,56 @@ click==8.1.7 \
    --hash=sha256:ae74fb96c20a0277a1d615f1e4d73c8414f5a98db8b799a7931d1582f3390c28 \
    --hash=sha256:ca9853ad459e787e2192211578cc907e7594e294c7ccc834310722b41b9ca6de
    # via black
-cryptography==45.0.5 \
-    --hash=sha256:0027d566d65a38497bc37e0dd7c2f8ceda73597d2ac9ba93810204f56f52ebc7 \
-    --hash=sha256:101ee65078f6dd3e5a028d4f19c07ffa4dd22cce6a20eaa160f8b5219911e7d8 \
-    --hash=sha256:12e55281d993a793b0e883066f590c1ae1e802e3acb67f8b442e721e475e6463 \
-    --hash=sha256:14d96584701a887763384f3c47f0ca7c1cce322aa1c31172680eb596b890ec30 \
-    --hash=sha256:1e1da5accc0c750056c556a93c3e9cb828970206c68867712ca5805e46dc806f \
-    --hash=sha256:206210d03c1193f4e1ff681d22885181d47efa1ab3018766a7b32a7b3d6e6afd \
-    --hash=sha256:2089cc8f70a6e454601525e5bf2779e665d7865af002a5dec8d14e561002e135 \
-    --hash=sha256:3a264aae5f7fbb089dbc01e0242d3b67dffe3e6292e1f5182122bdf58e65215d \
-    --hash=sha256:3af26738f2db354aafe492fb3869e955b12b2ef2e16908c8b9cb928128d42c57 \
-    --hash=sha256:3fcfbefc4a7f332dece7272a88e410f611e79458fab97b5efe14e54fe476f4fd \
-    --hash=sha256:460f8c39ba66af7db0545a8c6f2eabcbc5a5528fc1cf6c3fa9a1e44cec33385e \
-    --hash=sha256:57c816dfbd1659a367831baca4b775b2a5b43c003daf52e9d57e1d30bc2e1b0e \
-    --hash=sha256:5aa1e32983d4443e310f726ee4b071ab7569f58eedfdd65e9675484a4eb67bd1 \
-    --hash=sha256:6ff8728d8d890b3dda5765276d1bc6fb099252915a2cd3aff960c4c195745dd0 \
-    --hash=sha256:7259038202a47fdecee7e62e0fd0b0738b6daa335354396c6ddebdbe1206af2a \
-    --hash=sha256:72e76caa004ab63accdf26023fccd1d087f6d90ec6048ff33ad0445abf7f605a \
-    --hash=sha256:7760c1c2e1a7084153a0f68fab76e754083b126a47d0117c9ed15e69e2103492 \
-    --hash=sha256:8c4a6ff8a30e9e3d38ac0539e9a9e02540ab3f827a3394f8852432f6b0ea152e \
-    --hash=sha256:9024beb59aca9d31d36fcdc1604dd9bbeed0a55bface9f1908df19178e2f116e \
-    --hash=sha256:90cb0a7bb35959f37e23303b7eed0a32280510030daba3f7fdfbb65defde6a97 \
-    --hash=sha256:91098f02ca81579c85f66df8a588c78f331ca19089763d733e34ad359f474174 \
-    --hash=sha256:926c3ea71a6043921050eaa639137e13dbe7b4ab25800932a8498364fc1abec9 \
-    --hash=sha256:982518cd64c54fcada9d7e5cf28eabd3ee76bd03ab18e08a48cad7e8b6f31b18 \
-    --hash=sha256:9b4cf6318915dccfe218e69bbec417fdd7c7185aa7aab139a2c0beb7468c89f0 \
-    --hash=sha256:ad0caded895a00261a5b4aa9af828baede54638754b51955a0ac75576b831b27 \
-    --hash=sha256:b85980d1e345fe769cfc57c57db2b59cff5464ee0c045d52c0df087e926fbe63 \
-    --hash=sha256:b8fa8b0a35a9982a3c60ec79905ba5bb090fc0b9addcfd3dc2dd04267e45f25e \
-    --hash=sha256:b9e38e0a83cd51e07f5a48ff9691cae95a79bea28fe4ded168a8e5c6c77e819d \
-    --hash=sha256:bd4c45986472694e5121084c6ebbd112aa919a25e783b87eb95953c9573906d6 \
-    --hash=sha256:be97d3a19c16a9be00edf79dca949c8fa7eff621763666a145f9f9535a5d7f42 \
-    --hash=sha256:c648025b6840fe62e57107e0a25f604db740e728bd67da4f6f060f03017d5097 \
-    --hash=sha256:d05a38884db2ba215218745f0781775806bde4f32e07b135348355fe8e4991d9 \
-    --hash=sha256:dd420e577921c8c2d31289536c386aaa30140b473835e97f83bc71ea9d2baf2d \
-    --hash=sha256:e357286c1b76403dd384d938f93c46b2b058ed4dfcdce64a770f0537ed3feb6f \
-    --hash=sha256:e6c00130ed423201c5bc5544c23359141660b07999ad82e34e7bb8f882bb78e0 \
-    --hash=sha256:e74d30ec9c7cb2f404af331d5b4099a9b322a8a6b25c4632755c8757345baac5 \
-    --hash=sha256:f3562c2f23c612f2e4a6964a61d942f891d29ee320edb62ff48ffb99f3de9ae8
+cryptography==46.0.5 \
+    --hash=sha256:02f547fce831f5096c9a567fd41bc12ca8f11df260959ecc7c3202555cc47a72 \
+    --hash=sha256:039917b0dc418bb9f6edce8a906572d69e74bd330b0b3fea4f79dab7f8ddd235 \
+    --hash=sha256:1abfdb89b41c3be0365328a410baa9df3ff8a9110fb75e7b52e66803ddabc9a9 \
+    --hash=sha256:2ae6971afd6246710480e3f15824ed3029a60fc16991db250034efd0b9fb4356 \
+    --hash=sha256:2b7a67c9cd56372f3249b39699f2ad479f6991e62ea15800973b956f4b73e257 \
+    --hash=sha256:351695ada9ea9618b3500b490ad54c739860883df6c1f555e088eaf25b1bbaad \
+    --hash=sha256:38946c54b16c885c72c4f59846be9743d699eee2b69b6988e0a00a01f46a61a4 \
+    --hash=sha256:3b4995dc971c9fb83c25aa44cf45f02ba86f71ee600d81091c2f0cbae116b06c \
+    --hash=sha256:3ce58ba46e1bc2aac4f7d9290223cead56743fa6ab94a5d53292ffaac6a91614 \
+    --hash=sha256:3ee190460e2fbe447175cda91b88b84ae8322a104fc27766ad09428754a618ed \
+    --hash=sha256:4108d4c09fbbf2789d0c926eb4152ae1760d5a2d97612b92d508d96c861e4d31 \
+    --hash=sha256:420d0e909050490d04359e7fdb5ed7e667ca5c3c402b809ae2563d7e66a92229 \
+    --hash=sha256:47fb8a66058b80e509c47118ef8a75d14c455e81ac369050f20ba0d23e77fee0 \
+    --hash=sha256:4c3341037c136030cb46e4b1e17b7418ea4cbd9dd207e4a6f3b2b24e0d4ac731 \
+    --hash=sha256:4d7e3d356b8cd4ea5aff04f129d5f66ebdc7b6f8eae802b93739ed520c47c79b \
+    --hash=sha256:4d8ae8659ab18c65ced284993c2265910f6c9e650189d4e3f68445ef82a810e4 \
+    --hash=sha256:4e817a8920bfbcff8940ecfd60f23d01836408242b30f1a708d93198393a80b4 \
+    --hash=sha256:50bfb6925eff619c9c023b967d5b77a54e04256c4281b0e21336a130cd7fc263 \
+    --hash=sha256:556e106ee01aa13484ce9b0239bca667be5004efb0aabbed28d353df86445595 \
+    --hash=sha256:582f5fcd2afa31622f317f80426a027f30dc792e9c80ffee87b993200ea115f1 \
+    --hash=sha256:5be7bf2fb40769e05739dd0046e7b26f9d4670badc7b032d6ce4db64dddc0678 \
+    --hash=sha256:60ee7e19e95104d4c03871d7d7dfb3d22ef8a9b9c6778c94e1c8fcc8365afd48 \
+    --hash=sha256:61aa400dce22cb001a98014f647dc21cda08f7915ceb95df0c9eaf84b4b6af76 \
+    --hash=sha256:68f68d13f2e1cb95163fa3b4db4bf9a159a418f5f6e7242564fc75fcae667fd0 \
+    --hash=sha256:7d1f30a86d2757199cb2d56e48cce14deddf1f9c95f1ef1b64ee91ea43fe2e18 \
+    --hash=sha256:7d731d4b107030987fd61a7f8ab512b25b53cef8f233a97379ede116f30eb67d \
+    --hash=sha256:803812e111e75d1aa73690d2facc295eaefd4439be1023fefc4995eaea2af90d \
+    --hash=sha256:80a8d7bfdf38f87ca30a5391c0c9ce4ed2926918e017c29ddf643d0ed2778ea1 \
+    --hash=sha256:8293f3dea7fc929ef7240796ba231413afa7b68ce38fd21da2995549f5961981 \
+    --hash=sha256:8456928655f856c6e1533ff59d5be76578a7157224dbd9ce6872f25055ab9ab7 \
+    --hash=sha256:890bcb4abd5a2d3f852196437129eb3667d62630333aacc13dfd470fad3aaa82 \
+    --hash=sha256:94a76daa32eb78d61339aff7952ea819b1734b46f73646a07decb40e5b3448e2 \
+    --hash=sha256:9f16fbdf4da055efb21c22d81b89f155f02ba420558db21288b3d0035bafd5f4 \
+    --hash=sha256:a3d1fae9863299076f05cb8a778c467578262fae09f9dc0ee9b12eb4268ce663 \
+    --hash=sha256:a3d507bb6a513ca96ba84443226af944b0f7f47dcc9a399d110cd6146481d24c \
+    --hash=sha256:abace499247268e3757271b2f1e244b36b06f8515cf27c4d49468fc9eb16e93d \
+    --hash=sha256:ba2a27ff02f48193fc4daeadf8ad2590516fa3d0adeeb34336b96f7fa64c1e3a \
+    --hash=sha256:bc84e875994c3b445871ea7181d424588171efec3e185dced958dad9e001950a \
+    --hash=sha256:bfd56bb4b37ed4f330b82402f6f435845a5f5648edf1ad497da51a8452d5d62d \
+    --hash=sha256:c18ff11e86df2e28854939acde2d003f7984f721eba450b56a200ad90eeb0e6b \
+    --hash=sha256:c3bcce8521d785d510b2aad26ae2c966092b7daa8f45dd8f44734a104dc0bc1a \
+    --hash=sha256:c4143987a42a2397f2fc3b4d7e3a7d313fbe684f67ff443999e803dd75a76826 \
+    --hash=sha256:c69fd885df7d089548a42d5ec05be26050ebcd2283d89b3d30676eb32ff87dee \
+    --hash=sha256:ced80795227d70549a411a4ab66e8ce307899fad2220ce5ab2f296e687eacde9 \
+    --hash=sha256:d66e421495fdb797610a08f43b05269e0a5ea7f5e652a89bfd5a7d3c1dee3648 \
+    --hash=sha256:d861ee9e76ace6cf36a6a89b959ec08e7bc2493ee39d07ffe5acb23ef46d27da \
+    --hash=sha256:e9251e3be159d1020c4030bd2e5f84d6a43fe54b6c19c12f51cde9542a2817b2 \
+    --hash=sha256:f145bba11b878005c496e93e257c1e88f154d278d2638e6450d17e0f31e558d2 \
+    --hash=sha256:fe346b143ff9685e40192a4960938545c699054ba11d4f9029f94751e3f71d87
    # via
    #   -r requirements_formatting.txt.in
    #   pyjwt
@@ -278,18 +310,35 @@ pyjwt==2.8.0 \
    --hash=sha256:57e28d156e3d5c10088e0c68abb90bfac3df82b40a71bd0daa20c65ccd5c23de \
    --hash=sha256:59127c392cc44c2da5bb3192169a91f429924e17aff6534d70fdc02ab3e04320
    # via pygithub
-pynacl==1.5.0 \
-    --hash=sha256:06b8f6fa7f5de8d5d2f7573fe8c863c051225a27b61e6860fd047b1775807858 \
-    --hash=sha256:0c84947a22519e013607c9be43706dd42513f9e6ae5d39d3613ca1e142fba44d \
-    --hash=sha256:20f42270d27e1b6a29f54032090b972d97f0a1b0948cc52392041ef7831fee93 \
-    --hash=sha256:401002a4aaa07c9414132aaed7f6836ff98f59277a234704ff66878c2ee4a0d1 \
-    --hash=sha256:52cb72a79269189d4e0dc537556f4740f7f0a9ec41c1322598799b0bdad4ef92 \
-    --hash=sha256:61f642bf2378713e2c2e1de73444a3778e5f0a38be6fee0fe532fe30060282ff \
-    --hash=sha256:8ac7448f09ab85811607bdd21ec2464495ac8b7c66d146bf545b0f08fb9220ba \
-    --hash=sha256:a36d4a9dda1f19ce6e03c9a784a2921a4b726b02e1c736600ca9c22029474394 \
-    --hash=sha256:a422368fc821589c228f4c49438a368831cb5bbc0eab5ebe1d7fac9dded6567b \
-    --hash=sha256:e46dae94e34b085175f8abb3b0aaa7da40767865ac82c928eeb9e57e1ea8a543
-    # via pygithub
+pynacl==1.6.2 \
+    --hash=sha256:018494d6d696ae03c7e656e5e74cdfd8ea1326962cc401bcf018f1ed8436811c \
+    --hash=sha256:04316d1fc625d860b6c162fff704eb8426b1a8bcd3abacea11142cbd99a6b574 \
+    --hash=sha256:22de65bb9010a725b0dac248f353bb072969c94fa8d6b1f34b87d7953cf7bbe4 \
+    --hash=sha256:26bfcd00dcf2cf160f122186af731ae30ab120c18e8375684ec2670dccd28130 \
+    --hash=sha256:2fef529ef3ee487ad8113d287a593fa26f48ee3620d92ecc6f1d09ea38e0709b \
+    --hash=sha256:320ef68a41c87547c91a8b58903c9caa641ab01e8512ce291085b5fe2fcb7590 \
+    --hash=sha256:3bffb6d0f6becacb6526f8f42adfb5efb26337056ee0831fb9a7044d1a964444 \
+    --hash=sha256:44081faff368d6c5553ccf55322ef2819abb40e25afaec7e740f159f74813634 \
+    --hash=sha256:46065496ab748469cdd999246d17e301b2c24ae2fdf739132e580a0e94c94a87 \
+    --hash=sha256:5811c72b473b2f38f7e2a3dc4f8642e3a3e9b5e7317266e4ced1fba85cae41aa \
+    --hash=sha256:622d7b07cc5c02c666795792931b50c91f3ce3c2649762efb1ef0d5684c81594 \
+    --hash=sha256:62985f233210dee6548c223301b6c25440852e13d59a8b81490203c3227c5ba0 \
+    --hash=sha256:68be3a09455743ff9505491220b64440ced8973fe930f270c8e07ccfa25b1f9e \
+    --hash=sha256:834a43af110f743a754448463e8fd61259cd4ab5bbedcf70f9dabad1d28a394c \
+    --hash=sha256:8845c0631c0be43abdd865511c41eab235e0be69c81dc66a50911594198679b0 \
+    --hash=sha256:8a66d6fb6ae7661c58995f9c6435bda2b1e68b54b598a6a10247bfcdadac996c \
+    --hash=sha256:8b097553b380236d51ed11356c953bf8ce36a29a3e596e934ecabe76c985a577 \
+    --hash=sha256:a84bf1c20339d06dc0c85d9aea9637a24f718f375d861b2668b2f9f96fa51145 \
+    --hash=sha256:a9f9932d8d2811ce1a8ffa79dcbdf3970e7355b5c8eb0c1a881a57e7f7d96e88 \
+    --hash=sha256:bc4a36b28dd72fb4845e5d8f9760610588a96d5a51f01d84d8c6ff9849968c14 \
+    --hash=sha256:c8a231e36ec2cab018c4ad4358c386e36eede0319a0c41fed24f840b1dac59f6 \
+    --hash=sha256:c949ea47e4206af7c8f604b8278093b674f7c79ed0d4719cc836902bf4517465 \
+    --hash=sha256:d071c6a9a4c94d79eb665db4ce5cedc537faf74f2355e4d502591d850d3913c0 \
+    --hash=sha256:d29bfe37e20e015a7d8b23cfc8bd6aa7909c92a1b8f41ee416bbb3e79ef182b2 \
+    --hash=sha256:fe9847ca47d287af41e82be1dd5e23023d3c31a951da134121ab02e42ac218c9
+    # via
+    #   -r requirements_formatting.txt.in
+    #   pygithub
 requests==2.32.4 \
    --hash=sha256:27babd3cda2a6d50b30443204ee89830707d396671944c998b5975b031ac2b2c \
    --hash=sha256:27d0316682c8a29834d3264820024b62a36942083d52caf2f14c0591336d3422
@@ -306,9 +355,9 @@ typing-extensions==4.14.1 \
    --hash=sha256:38b39f4aeeab64884ce9f74c94263ef78f3c22467c8724005483154c26648d36 \
    --hash=sha256:d1e1e3b58374dc93031d6eda2420a48ea44a36c2b4766a4fdeb3710755731d76
    # via pygithub
-urllib3==2.5.0 \
-    --hash=sha256:3fc47733c7e419d4bc3f6b3dc2b4f890bb743906a30d56ba4a5bfa4bbff92760 \
-    --hash=sha256:e6b01673c0fa6a13e374b50871808eb3bf7046c4b125b216f6bf1cc604cff0dc
+urllib3==2.6.3 \
+    --hash=sha256:1b62b6884944a57dbe321509ab94fd4d3b307075e0c2eae991ac71ee15ad38ed \
+    --hash=sha256:bf272323e553dfb2e87d9bfd225ca7b0f467b919d7bbd355436d3fd37cb0acd4
    # via
    #   -r requirements_formatting.txt.in
    #   pygithub
@@ -1,8 +1,9 @@
 black~=25.1
 darker==2.1.1
 PyGithub==2.6.1
-cryptography>=43.0.1
-urllib3>=2.5.0
+cryptography>=46.0.5
+urllib3>=2.6.3
 requests>=2.32.4
 idna>=3.7
 certifi>=2024.7.4
+PyNaCl>=1.6.2
@@ -1,3 +1,6 @@
 set(NAME tiny-json)
 set(SRCS tiny-json.c)
 add_library(${NAME} STATIC ${SRCS})
+
+target_include_directories(${NAME} PUBLIC ${CMAKE_CURRENT_LIST_DIR})
+add_library(${NAME}::${NAME} ALIAS ${NAME})
@@ -1,16 +1,16 @@
 cmake_minimum_required(VERSION 3.14)
-set (PROJECT_NAME FEXCore)
+set(PROJECT_NAME FEXCore)
 project(${PROJECT_NAME}
  VERSION 0.01
  LANGUAGES CXX)

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

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

 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
@@ -24,45 +24,10 @@ include(CheckCXXCompilerFlag)
 include(CheckIncludeFileCXX)
 include(CheckCXXSourceCompiles)

-if (EXISTS ${CMAKE_CURRENT_DIR}/External/vixl/)
-    # Useful to have for freestanding libFEXCore
-    add_subdirectory(External/vixl/)
-    include_directories(External/vixl/src/)
-endif()
-
 set(CMAKE_CXX_STANDARD 20)
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)

-set(GIT_SHORT_HASH "Unknown")
-set(GIT_DESCRIBE_STRING "FEX-Unknown")
-
-if (OVERRIDE_VERSION STREQUAL "detect")
-# Find our git hash
-  find_package(Git)
-
-  if (GIT_FOUND)
-    execute_process(
-      COMMAND ${GIT_EXECUTABLE} rev-parse --short=7 HEAD
-      WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
-      OUTPUT_VARIABLE GIT_SHORT_HASH
-      ERROR_QUIET
-      OUTPUT_STRIP_TRAILING_WHITESPACE
-    )
-    execute_process(
-      COMMAND ${GIT_EXECUTABLE} describe --abbrev=7
-      WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
-      OUTPUT_VARIABLE GIT_DESCRIBE_STRING
-      ERROR_QUIET
-      OUTPUT_STRIP_TRAILING_WHITESPACE
-    )
-  endif()
-else()
-  set(GIT_SHORT_HASH "${OVERRIDE_VERSION}")
-  set(GIT_DESCRIBE_STRING "FEX-${OVERRIDE_VERSION}")
-endif()
-
-configure_file(
-  ${CMAKE_CURRENT_SOURCE_DIR}/include/git_version.h.in
+configure_file(${CMAKE_CURRENT_SOURCE_DIR}/include/git_version.h.in
  ${CMAKE_BINARY_DIR}/generated/git_version.h)

 include_directories(${CMAKE_BINARY_DIR}/generated)
@@ -74,9 +39,11 @@ add_compile_options($<$<COMPILE_LANGUAGE:CXX>:-fno-strict-aliasing> $<$<COMPILE_

 add_subdirectory(Source/)

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

 if (BUILD_TESTING)
  add_subdirectory(unittests/)
@@ -156,7 +156,7 @@ def print_man_environment_tail():
    "APP_CONFIG_LOCATION",
    [
    "Allows the user to override where FEX looks for configuration files",
-    "By default FEX will look in {$HOME, $XDG_CONFIG_HOME}/.fex-emu/",
+    "By default FEX will look in ${XDG_CONFIG_HOME, $HOME/.config}/fex-emu/",
    "This will override the full path",
    "If FEX_PORTABLE is declared then relative paths are also supported",
    "For FEX: Relative to the FEX binary",
@@ -168,7 +168,7 @@ def print_man_environment_tail():
    "APP_CONFIG",
    [
    "Allows the user to override where FEX looks for only the application config file",
-    "By default FEX will look in {$HOME, $XDG_CONFIG_HOME}/.fex-emu/Config.json",
+    "By default FEX will look in ${XDG_CONFIG_HOME, $HOME/.config}/fex-emu/Config.json",
    "This will override this file location",
    "One must be careful with this option as it will override any applications that load with execve as well"
    "If you need to support applications that execve then use FEX_APP_CONFIG_LOCATION instead"
@@ -182,7 +182,7 @@ def print_man_environment_tail():
    "APP_DATA_LOCATION",
    [
    "Allows the user to override where FEX looks for data files",
-    "By default FEX will look in {$HOME, $XDG_DATA_HOME}/.fex-emu/",
+    "By default FEX will look in {$XDG_DATA_HOME, $HOME/.local/share}/fex-emu/",
    "This will override the full path",
    "This is the folder where FEX stores generated files like IR cache"
    ],
@@ -200,6 +200,15 @@ def print_man_environment_tail():
    ],
    "''", True)

+    print_man_env_option(
+    "APP_CACHE_LOCATION",
+    [
+    "Allows the user to override where FEX stores and loads cache files",
+    "By default FEX will look in ${XDG_CACHE_HOME, $HOME/.cache}/fex-emu/",
+    "This will override the full path, trailing forward-slash is expected to exist",
+    ],
+    "''", True)
+
 def print_man_header():
    header ='''.Dd {0}
 .Dt FEX
@@ -225,7 +234,7 @@ FEX is very much work in progress, so expect things to change.
 def print_man_tail():
    tail ='''.Sh FILES
 .Bl -tag -width "$prefix/share/fex-emu/GuestThunks" -compact
-.It Pa $XDG_HOME_DIR/.fex-emu
+.It Pa $XDG_CONFIG_DIR/fex-emu
 Default FEX user configuration directory
 .It Pa $prefix/share/fex-emu/AppConfig
 System level application configuration files
@@ -58,10 +58,10 @@ class OpDefinition:
    JITDispatch: bool
    JITDispatchOverride: str
    TiedSource: int
-    Inline: list
-    Arguments: list
-    EmitValidation: list
-    Desc: list
+    Inline: list[str]
+    Arguments: list[OpArgument]
+    EmitValidation: list[str]
+    Desc: list[str]

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

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

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

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

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

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

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

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

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

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

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

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

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

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


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

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

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

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

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

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

@@ -682,25 +678,25 @@ def print_ir_allocator_helpers():
    # Generate helpers with operands
    for op in IROps:
        if op.Name != "Last":
+            output_file.write("\t///\n".join(["\t/// {}\n" .format(comment) for comment in op.Desc]))
            output_file.write("\tIRPair<IROp_{}> _{}(" .format(op.Name, op.Name))

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

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

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

                if not LastArg:
@@ -756,22 +752,22 @@ def print_ir_allocator_helpers():

            # Now do the OrderedNode * version if necessary
            if op.SSAArgNum:
+                output_file.write("\t///\n".join(["\t/// {}\n" .format(comment) for comment in op.Desc]))
                output_file.write("\tIRPair<IROp_{}> _{}(" .format(op.Name, op.Name))

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

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

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

                    if not LastArg:
@@ -812,16 +808,15 @@ def print_ir_allocator_helpers():
                print_validation(op)

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

    output_file.write("#undef IROP_ALLOCATE_HELPERS\n")
    output_file.write("#endif\n")
@@ -852,8 +847,8 @@ def print_ir_dispatcher_dispatch():
    output_dispatch_file.write("#endif\n")


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

 output_filename = sys.argv[2]
 output_dispatcher_filename = sys.argv[3]
@@ -865,6 +860,7 @@ json_file.close()
 json_object = json.loads(json_text)
 json_object = {k.upper(): v for k, v in json_object.items()}

+enums = json_object["ENUMS"]
 ops = json_object["OPS"]
 irtypes = json_object["IRTYPES"]
 defines = json_object["DEFINES"]
@@ -874,7 +870,7 @@ parse_ops(ops)

 output_file = open(output_filename, "w")

-print_enums()
+print_enums(enums)
 print_ir_structs(defines)
 print_ir_sizes()
 print_ir_reg_classes()
@@ -1,20 +1,19 @@
-set (MAN_DIR share/man CACHE PATH "MAN_DIR")
+set(MAN_DIR share/man CACHE PATH "MAN_DIR")

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

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

-set (SRCS
+set(SRCS
  Common/JitSymbols.cpp
  Interface/Context/Context.cpp
  Interface/Core/LookupCache.cpp
@@ -31,7 +30,6 @@ set (SRCS
  Interface/Core/OpcodeDispatcher/X87.cpp
  Interface/Core/OpcodeDispatcher/X87F64.cpp
  Interface/Core/OpcodeDispatcher.cpp
-  Interface/Core/X86HelperGen.cpp
  Interface/Core/ArchHelpers/Arm64Emitter.cpp
  Interface/Core/Dispatcher/Dispatcher.cpp
  Interface/Core/Interpreter/Fallbacks/InterpreterFallbacks.cpp
@@ -69,10 +67,9 @@ set (SRCS
  Utils/LongJump.cpp
  Utils/Telemetry.cpp
  Utils/Threads.cpp
-  Utils/Profiler.cpp
-  )
+  Utils/Profiler.cpp)

-if (_M_ARM_64)
+if (ARCHITECTURE_arm64)
  list(APPEND SRCS Utils/ArchHelpers/Arm64.cpp)
 else()
  list(APPEND SRCS Utils/ArchHelpers/Arm64_stubs.cpp)
@@ -85,42 +82,49 @@ endif()

 set(DEFINES -DJIT_ARM64)

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

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

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

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

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

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

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

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

 # Generate IR include file
@@ -135,11 +139,10 @@ add_custom_command(
  OUTPUT "${OUTPUT_NAME}" "${OUTPUT_DISPATCHER_NAME}"
  DEPENDS "${INPUT_NAME}"
  DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_generator.py"
-  COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_generator.py" "${INPUT_NAME}" "${OUTPUT_NAME}" "${OUTPUT_DISPATCHER_NAME}"
-  )
+  COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_generator.py"
+    "${INPUT_NAME}" "${OUTPUT_NAME}" "${OUTPUT_DISPATCHER_NAME}")

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

 # Generate IR documentation
 set(OUTPUT_IR_DOC "${CMAKE_BINARY_DIR}/IR.md")
@@ -148,11 +151,10 @@ add_custom_command(
  OUTPUT "${OUTPUT_IR_DOC}"
  DEPENDS "${INPUT_NAME}"
  DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_doc_generator.py"
-  COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_doc_generator.py" "${INPUT_NAME}" "${OUTPUT_IR_DOC}"
-  )
+  COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_doc_generator.py"
+    "${INPUT_NAME}" "${OUTPUT_IR_DOC}")

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

 # Create the target
 add_custom_target(IR_INC
@@ -176,14 +178,12 @@ add_custom_command(
  DEPENDS "${INPUT_CONFIG_NAME}"
  DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/config_generator.py"
  COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/config_generator.py" "${INPUT_CONFIG_NAME}" "${OUTPUT_CONFIG_NAME}" "${OUTPUT_MAN_NAME}"
-  "${OUTPUT_CONFIG_OPTION_NAME}"
-  )
+  "${OUTPUT_CONFIG_OPTION_NAME}")

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

 set_source_files_properties(${OUTPUT_CONFIG_NAME} PROPERTIES
  GENERATED TRUE)
@@ -202,8 +202,10 @@ add_custom_target(CONFIG_INC
  DEPENDS "${OUTPUT_MAN_NAME}"
  DEPENDS "${OUTPUT_MAN_NAME_COMPRESS}")

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

 # Add in diagnostic colours if the option is available.
 # Ninja code generator will kill colours if this isn't here
@@ -225,8 +227,7 @@ function(AddDefaultOptionsToTarget Name)
  target_compile_definitions(${Name} PRIVATE ${DEFINES})
  add_dependencies(${Name} CONFIG_INC IR_INC)

-  target_compile_options(${Name}
-    PRIVATE
+  target_compile_options(${Name} PRIVATE
    -Wall
    -Werror=cast-qual
    -Werror=ignored-qualifiers
@@ -234,82 +235,72 @@ function(AddDefaultOptionsToTarget Name)

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

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

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

-# Build FEXCore_Config static library
+# Build FEXCore_Base static library
 add_library(FEXCore_Base STATIC ${FEXCORE_BASE_SRCS})
-target_link_libraries(FEXCore_Base ${LIBS})
+target_link_libraries(FEXCore_Base PUBLIC ${LIBS})
 AddDefaultOptionsToTarget(FEXCore_Base)

 if (ENABLE_FEXCORE_PROFILER AND FEXCORE_PROFILER_BACKEND STREQUAL "TRACY")
-  target_link_libraries(FEXCore_Base TracyClient)
+  target_link_libraries(FEXCore_Base PUBLIC TracyClient)
 endif()

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

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

 function(AddLibrary Name Type)
  add_library(${Name} ${Type} $<TARGET_OBJECTS:${PROJECT_NAME}_object>)
-  target_link_libraries(${Name} FEXCore_Base)
-  target_compile_options(${Name} PRIVATE ${FEX_TUNE_COMPILE_FLAGS})
  set_target_properties(${Name} PROPERTIES OUTPUT_NAME FEXCore)

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

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

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

 # Meta-library to link jemalloc libraries enabled in the build configuration.
 # Only needed for targets that run emulation. For others, use JemallocDummy.
 add_library(JemallocLibs STATIC Utils/AllocatorHooks.cpp)
-if (ENABLE_JEMALLOC)
-  target_compile_definitions(JemallocLibs PRIVATE ENABLE_JEMALLOC=1 JEMALLOC_NO_RENAME=1)
-  target_link_libraries(JemallocLibs PUBLIC FEX_jemalloc)
+if (ENABLE_FEX_ALLOCATOR)
+  target_compile_definitions(JemallocLibs PRIVATE ENABLE_FEX_ALLOCATOR=1)
+  target_link_libraries(JemallocLibs PUBLIC rpmalloc)
 endif()
 if (ENABLE_JEMALLOC_GLIBC_ALLOC)
  set_source_files_properties(Interface/HLE/Thunks/Thunks.cpp PROPERTIES COMPILE_DEFINITIONS ENABLE_JEMALLOC_GLIBC=1)
  target_link_libraries(JemallocLibs INTERFACE FEX_jemalloc_glibc)
 endif()

-if (NOT MINGW_BUILD)
+if (NOT MINGW)
  # Dummy project to use for host tools.
  # This overrides use of jemalloc in FEXCore with the normal glibc allocator.
  add_library(JemallocDummy STATIC Utils/AllocatorHooks.cpp)
@@ -317,4 +308,4 @@ if (NOT MINGW_BUILD)
 endif()

 # The shared library should always link enabled jemalloc libraries
-target_link_libraries(${PROJECT_NAME}_shared JemallocLibs)
+target_link_libraries(${PROJECT_NAME}_shared PRIVATE JemallocLibs)
@@ -1,5 +1,6 @@
 // SPDX-License-Identifier: MIT
 #pragma once
+#include <FEXCore/Utils/TypeDefines.h>

 #include <FEXCore/fextl/memory.h>

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

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

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

 struct FEX_PACKED X80SoftFloat {
-#ifdef _M_X86_64
+#ifdef ARCHITECTURE_x86_64
 // Define this to push some operations to x87
 // Only useful to see if precision loss is killing something
 // #define DEBUG_X86_FLOAT
@@ -30,29 +30,33 @@ struct FEX_PACKED X80SoftFloat {
 #define BIGFLOAT float128_t
 #define BIGFLOATSIZE 16
 #endif
-#elif defined(_M_ARM_64)
+#elif defined(ARCHITECTURE_arm64)
 #define BIGFLOAT float128_t
 #define BIGFLOATSIZE 16
 #else
 #error No 128bit float for this target!
 #endif

-  uint64_t Significand : 64;
-  uint16_t Exponent    : 15;
-  uint16_t Sign        : 1;
+  uint64_t Significand;
+  union {
+    uint16_t Raw;
+    struct {
+      uint16_t Exponent : 15;
+      uint16_t Sign     : 1;
+    };
+  } Top;

  X80SoftFloat() {
    memset(this, 0, sizeof(*this));
  }
  X80SoftFloat(uint16_t _Sign, uint16_t _Exponent, uint64_t _Significand)
    : Significand {_Significand}
-    , Exponent {_Exponent}
-    , Sign {_Sign} {}
+    , Top {.Raw = static_cast<uint16_t>((_Exponent & 0x7FFF) | (_Sign << 15))} {}

  fextl::string str() const {
    fextl::ostringstream string;
-    string << std::hex << Sign;
-    string << "_" << Exponent;
+    string << std::hex << Top.Sign;
+    string << "_" << Top.Exponent;
    string << "_" << (Significand >> 63);
    string << "_" << (Significand & ((1ULL << 63) - 1));
    return string.str();
@@ -163,18 +167,18 @@ struct FEX_PACKED X80SoftFloat {
    X80SoftFloat result = 0;
    if (HandleInfinityOp(state, lhs, result)) {
      return result;
-    } else if (lhs.Exponent == 0x7FFF && (lhs.Significand & 0x7FFFFFFFFFFFFFFFULL)) { // NaN
+    } else if (lhs.Top.Exponent == 0x7FFF && (lhs.Significand & 0x7FFFFFFFFFFFFFFFULL)) { // NaN
      // propagate NaN
      state->exceptionFlags |= softfloat_flag_invalid;
      return lhs;
    }

    // Check for zero divisor - fprem(x, 0) is invalid operation
-    if (rhs.Exponent == 0 && rhs.Significand == 0) {
+    if (rhs.Top.Exponent == 0 && rhs.Significand == 0) {
      state->exceptionFlags |= softfloat_flag_invalid;
      // Return QNaN
-      result.Sign = 0;
-      result.Exponent = 0x7FFF;
+      result.Top.Sign = 0;
+      result.Top.Exponent = 0x7FFF;
      result.Significand = 0xC000000000000000ULL;
      return result;
    }
@@ -253,12 +257,12 @@ struct FEX_PACKED X80SoftFloat {
    return Result;
 #else
    // Zero is a special case, the significand for +/- 0 is +/- zero.
-    if (lhs.Exponent == 0x0 && lhs.Significand == 0x0) {
+    if (lhs.Top.Exponent == 0x0 && lhs.Significand == 0x0) {
      return lhs;
    }
    X80SoftFloat Tmp = lhs;
-    Tmp.Exponent = 0x3FFF;
-    Tmp.Sign = lhs.Sign;
+    Tmp.Top.Exponent = 0x3FFF;
+    Tmp.Top.Sign = lhs.Top.Sign;
    return Tmp;
 #endif
  }
@@ -280,12 +284,12 @@ struct FEX_PACKED X80SoftFloat {
    return Result;
 #else
    // Zero is a special case, the exponent is always -inf
-    if (lhs.Exponent == 0x0 && lhs.Significand == 0x0) {
+    if (lhs.Top.Exponent == 0x0 && lhs.Significand == 0x0) {
      X80SoftFloat Result(1, 0x7FFFUL, 0x8000'0000'0000'0000UL);
      return Result;
    }

-    int32_t TrueExp = lhs.Exponent - ExponentBias;
+    int32_t TrueExp = lhs.Top.Exponent - ExponentBias;
    return i32_to_extF80(TrueExp);
 #endif
  }
@@ -501,12 +505,12 @@ struct FEX_PACKED X80SoftFloat {

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

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

  FEXCore::VectorRegType ToVector() const {
@@ -518,7 +522,7 @@ struct FEX_PACKED X80SoftFloat {
  BIGFLOAT ToFMax(softfloat_state* state) const {
 #if BIGFLOATSIZE == 16
    const float128_t Result = extF80_to_f128(state, *this);
-    return FEXCore::BitCast<BIGFLOAT>(Result);
+    return std::bit_cast<BIGFLOAT>(Result);
 #else
    BIGFLOAT result {};
    memcpy(&result, this, sizeof(result));
@@ -572,23 +576,22 @@ struct FEX_PACKED X80SoftFloat {

  X80SoftFloat(extFloat80_t rhs) {
    Significand = rhs.signif;
-    Exponent = rhs.signExp & 0x7FFF;
-    Sign = rhs.signExp >> 15;
+    Top.Raw = rhs.signExp;
  }

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

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

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

@@ -606,8 +609,7 @@ struct FEX_PACKED X80SoftFloat {

  void operator=(extFloat80_t rhs) {
    Significand = rhs.signif;
-    Exponent = rhs.signExp & 0x7FFF;
-    Sign = rhs.signExp >> 15;
+    Top.Raw = rhs.signExp;
  }

  operator FEXCore::VectorRegType() const {
@@ -617,16 +619,16 @@ struct FEX_PACKED X80SoftFloat {
  operator extFloat80_t() const {
    extFloat80_t Result {};
    Result.signif = Significand;
-    Result.signExp = Exponent | (Sign << 15);
+    Result.signExp = Top.Raw;
    return Result;
  }

  static bool IsNan(const X80SoftFloat& lhs) {
-    return (lhs.Exponent == 0x7FFF) && (lhs.Significand & IntegerBit) && (lhs.Significand & Bottom62Significand);
+    return (lhs.Top.Exponent == 0x7FFF) && (lhs.Significand & IntegerBit) && (lhs.Significand & Bottom62Significand);
  }

  static bool SignBit(const X80SoftFloat& lhs) {
-    return lhs.Sign;
+    return lhs.Top.Sign;
  }

 private:
@@ -637,11 +639,11 @@ private:
  // Helper function to check for infinity and set invalid operation flag.
  // Returns true if infinity is dealt with, false otherwise.
  FEXCORE_PRESERVE_ALL_ATTR static bool HandleInfinityOp(softfloat_state* state, const X80SoftFloat& arg, X80SoftFloat& result) {
-    if (arg.Exponent == 0x7FFF && arg.Significand == 0x8000000000000000ULL) {
+    if (arg.Top.Exponent == 0x7FFF && arg.Significand == 0x8000000000000000ULL) {
      state->exceptionFlags |= softfloat_flag_invalid;
      // Return QNaN.
-      result.Sign = 0;
-      result.Exponent = 0x7FFF;
+      result.Top.Sign = 0;
+      result.Top.Exponent = 0x7FFF;
      result.Significand = 0xC000000000000000ULL;
      return true;
    }
@@ -649,9 +651,4 @@ private:
  }
 };

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

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

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

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

@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: MIT
 #pragma once

-#ifdef _M_X86_64
+#ifdef ARCHITECTURE_x86_64
 #include <xmmintrin.h>
 #include <immintrin.h>
 #else
@@ -13,7 +13,7 @@ struct VectorScalarF64Pair {
  double val[2];
 };

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

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

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

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

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

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

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

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

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

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


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

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

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

  // Now with the two layers merged in the map
@@ -225,8 +225,8 @@ void MetaLayer::MergeConfigMap(const LayerOptions& Options) {
  // Insert this layer's options, overlaying previous options that exist here
  for (auto& it : Options) {
    if (it.first == FEXCore::Config::ConfigOption::CONFIG_ENV || it.first == FEXCore::Config::ConfigOption::CONFIG_HOSTENV) {
-      LOGMAN_THROW_A_FMT(std::holds_alternative<DefaultValues::Type::StringArrayType>(it.second), "Tried to get config of invalid type!");
-      MergeEnvironmentVariables(it.first, std::get<DefaultValues::Type::StringArrayType>(it.second));
+      LOGMAN_THROW_A_FMT(std::holds_alternative<StringArrayType>(it.second), "Tried to get config of invalid type!");
+      MergeEnvironmentVariables(it.first, std::get<StringArrayType>(it.second));
    } else {
      OptionMap.insert_or_assign(it.first, it.second);
    }
@@ -307,12 +307,10 @@ constexpr char ContainerManager[] = "/run/host/container-manager";
 fextl::string FindContainer() {
  // We only support pressure-vessel at the moment
  if (FHU::Filesystem::Exists(ContainerManager)) {
-    fextl::vector<char> Manager {};
+    fextl::string Manager {};
    if (FEXCore::FileLoading::LoadFile(Manager, ContainerManager)) {
      // Trim the whitespace, may contain a newline
-      fextl::string ManagerStr = Manager.data();
-      ManagerStr = FEXCore::StringUtils::Trim(ManagerStr);
-      return ManagerStr;
+      return FEXCore::StringUtils::Trim(Manager);
    }
  }
  return {};
@@ -321,12 +319,10 @@ fextl::string FindContainer() {
 fextl::string FindContainerPrefix() {
  // We only support pressure-vessel at the moment
  if (FHU::Filesystem::Exists(ContainerManager)) {
-    fextl::vector<char> Manager {};
+    fextl::string 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) {
+      if (FEXCore::StringUtils::Trim(Manager) == "pressure-vessel") {
        // We are running inside of pressure vessel
        // Our $CMAKE_INSTALL_PREFIX paths are now inside of /run/host/$CMAKE_INSTALL_PREFIX
        return "/run/host/";
@@ -423,7 +419,7 @@ bool Exists(ConfigOption Option) {
  return Meta->OptionExists(Option);
 }

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

@@ -436,6 +432,12 @@ std::optional<T> GetConv(ConfigOption Option) {
  return Meta->GetConv<T>(Option);
 }

+template std::optional<bool> GetConv(ConfigOption Option);
+template std::optional<uint8_t> GetConv(ConfigOption Option);
+template std::optional<int32_t> GetConv(ConfigOption Option);
+template std::optional<uint32_t> GetConv(ConfigOption Option);
+template std::optional<uint64_t> GetConv(ConfigOption Option);
+
 void Set(ConfigOption Option, std::string_view Data) {
  Meta->Set(Option, Data);
 }
@@ -491,13 +493,12 @@ template Value<uint8_t>::Value(FEXCore::Config::ConfigOption _Option, uint8_t De
 template Value<uint64_t>::Value(FEXCore::Config::ConfigOption _Option, uint64_t Default);

 template<typename T>
-void Value<T>::GetListIfExists(FEXCore::Config::ConfigOption Option, DefaultValues::Type::StringArrayType* List) {
+void Value<T>::GetListIfExists(FEXCore::Config::ConfigOption Option, StringArrayType* List) {
  auto Value = FEXCore::Config::All(Option);
  List->clear();
  if (Value) {
    *List = **Value;
  }
 }
-template void Value<DefaultValues::Type::StringArrayType>::GetListIfExists(FEXCore::Config::ConfigOption Option,
-                                                                           DefaultValues::Type::StringArrayType* List);
+template void Value<StringArrayType>::GetListIfExists(FEXCore::Config::ConfigOption Option, StringArrayType* List);
 } // namespace FEXCore::Config
@@ -16,6 +16,20 @@
          "Maximum number of instruction to store in a block"
        ]
      },
+      "EnableCodeCachingWIP": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Enable the code caching subsystem"
+        ]
+      },
+      "EnableCodeCacheValidation": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Enable expensive validation when loading code caches"
+        ]
+      },
      "HostFeatures": {
        "Type": "strenum",
        "Default": "FEXCore::Config::HostFeatures::OFF",
@@ -59,7 +73,9 @@
          "ENABLEWFXT": "enablewfxt",
          "DISABLEWFXT": "disablewfxt",
          "ENABLE3DNOW": "enable3dnow",
-          "DISABLE3DNOW": "disable3dnow"
+          "DISABLE3DNOW": "disable3dnow",
+          "ENABLESSE4A": "enablesse4a",
+          "DISABLESSE4A": "disablesse4a"
        },
        "Desc": [
          "Allows controlling of the CPU features in the JIT.",
@@ -82,7 +98,8 @@
          "\t{enable,disable}svebitperm: Will force enable or disable svebitperm even if the host doesn't support it",
          "\t{enable,disable}preserveallabi: Will force enable or disable preserve_all abi even if the host doesn't support it",
          "\t{enable,disable}wfxt: Will force enable or disable wfxt even if the host doesn't support it",
-          "\t{enable,disable}3dnow: Will force enable or disable 3DNow even if the host doesn't support it"
+          "\t{enable,disable}3dnow: Will force enable or disable 3DNow! even if the host doesn't support it",
+          "\t{enable,disable}sse4a: Will force enable or disable SSE4a even if the host doesn't support it"
        ]
      },
      "SmallTSCScale": {
@@ -91,6 +108,20 @@
        "Desc": [
          "Scales the cycle counter on systems that have low frequencies."
        ]
+      },
+      "HideHybrid": {
+        "Type": "bool",
+        "Default": "true",
+        "Desc": [
+          "Hides hybrid CPU core arrangement."
+        ]
+      },
+      "CPUFeatureRegisters": {
+        "Type": "str",
+        "Default": "",
+        "Desc": [
+          "Allows overriding cpu feature flags for manual testing"
+        ]
      }
    },
    "Emulation": {
@@ -103,9 +134,9 @@
          "\teg: ~/RootFS/Debian_x86_64",
          "Or this can be a name of a rootfs",
          "If the named rootfs exists in the FEX data folder then it will use that one",
-          "\teg: $HOME/.fex-emu/RootFS/<RootFS name>/",
-          "Or if you have XDG_DATA_HOME the config will search in that directory",
-          "\teg: $XDG_DATA_HOME/.fex-emu/RootFS/<RootFS name>/"
+          "\teg: $XDG_DATA_HOME/fex-emu/RootFS/<RootFS name>/",
+          "If XDG_DATA_HOME is unset, ~/.local/share will be used in its place.",
+          "\teg: $HOME/.local/share/fex-emu/RootFS/<RootFS name>/"
        ]
      },
      "ThunkHostLibs": {
@@ -131,9 +162,9 @@
          "\teg: ~/MyThunkConfig.json",
          "Or this can be a named of a Thunk config file",
          "If the named config file exists in the FEX data folder folder the it will use that one",
-          "\teg: $HOME/.fex-emu/ThunkConfigs/<ThunkConfig name>",
-          "Or if you have XDG_DATA_HOME the config will search in that directory",
-          "\teg: $XDG_DATA_HOME/.fex-emu/ThunkConfigs/<ThunkConfig name>"
+          "\teg: $XDG_DATA_HOME/fex-emu/ThunkConfigs/<ThunkConfig name>",
+          "If XDG_DATA_HOME is unset, ~/.local/share will be used in its place.",
+          "\teg: $HOME/.local/share/fex-emu/ThunkConfigs/<ThunkConfig name>"
        ]
      },
      "Env": {
@@ -158,6 +189,44 @@
        "Desc": [
          "Allows the user to pass additional arguments to the application"
        ]
+      },
+      "DisableL2Cache": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Disables FEXCore's JIT L2 cache lookup. Saving memory.",
+          "Can potentially introduce more stutters."
+        ]
+      },
+      "DynamicL1Cache": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Switches FEXCore's JIT L1 cache to be dynamically sized. Saving memory.",
+          "Can potentially introduce more stutters."
+        ]
+      },
+      "DynamicL1CacheIncreaseCountHeuristic": {
+        "Type": "uint64",
+        "Default": "250",
+        "Desc": [
+          "Threshold of lookups per second that the L1 dynamic cache should increase its size.",
+          "Lower numbers means more aggressive scaling upward to the maximum size.",
+          "Higher numbers means more conservative scaling, using less memory.",
+          "Can potentially introduce stutters, more likely the higher the number.",
+          "Don't have this number smaller than the decrease count!"
+        ]
+      },
+      "DynamicL1CacheDecreaseCountHeuristic": {
+        "Type": "uint64",
+        "Default": "50",
+        "Desc": [
+          "Threshold of lookups per second that the L1 dynamic cache should decrease its size.",
+          "The higher the number, the more aggressively it reduces the L1 cache size.",
+          "Lower numbers means more conservative memory savings.",
+          "Can potentially introduce more stutters, more likely the higher the number.",
+          "Don't have this number larger than the increase count!"
+        ]
      }
    },
    "Debug": {
@@ -272,13 +341,21 @@
          "STATS": "stats"
        },
        "Desc": [
-          "Allows controlling of the vixl disassembler.",
+          "Allows controlling of the vixl disassembler for generated ARM code.",
          "\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",
          "\tstats: Will print stats when disassembling the code"
        ]
      },
+      "X86Disassemble": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Enables x86/x86-64 guest disassembly output for compiled blocks.",
+          "Requires FEX to be built with -DENABLE_ZYDIS=TRUE"
+        ]
+      },
      "ForceSVEWidth": {
        "Type": "uint32",
        "Default": "0",
@@ -309,7 +386,7 @@
        "Default": "server",
        "Desc": [
          "File to write FEX output to.",
-          "[stdout, stderr, server, <Filename>]"
+          "[stderr, server, <Filename>]"
        ]
      },
      "TelemetryDirectory": {
@@ -317,7 +394,7 @@
        "Default": "",
        "Desc": [
          "Redirects the telemetry folder that FEX usually writes to.",
-          "By default telemetry data is stored in {$FEX_APP_DATA_LOCATION,{$XDG_DATA_HOME,$HOME}/.fex-emu/Telemetry/}"
+          "By default telemetry data is stored in {$FEX_APP_DATA_LOCATION,{$XDG_DATA_HOME,$HOME}/fex-emu/Telemetry/}"
        ]
      },
      "ProfileStats": {
@@ -328,11 +405,11 @@
          "Requires a supported version of Mangohud to see the results"
        ]
      },
-      "TraceProfiler": {
+      "EnableGpuvisProfiling": {
        "Type": "bool",
        "Default": "false",
        "Desc": [
-          "Enables FEX's trace profiler. Using gpuvis or tracy"
+          "Enables profiling when FEX was built with the gpuvis profiler backend."
        ]
      }
    },
@@ -388,12 +465,19 @@
          "This is required to ensure a split-lock doesn't tear inside the process"
        ]
      },
+      "KernelUnalignedAtomicBackpatching": {
+        "Type": "bool",
+        "Default": "true",
+        "Desc": [
+          "When the kernel unaligned atomic handler is enabled, use backpatching to reduce kernel context switches."
+        ]
+      },
      "VolatileMetadata": {
        "Type": "bool",
        "Default": "true",
        "Desc": [
          "Use volatile metadata in PE files to inform TSO instructions when available.",
-	  "When metadata is unavailable falls back to the currently enabled TSO options."
+          "When metadata is unavailable falls back to the currently enabled TSO options."
        ]
      },
      "X87ReducedPrecision": {
@@ -403,23 +487,6 @@
          "Emulates X87 floating point using 64-bit precision. This reduces emulation accuracy and may result in rendering bugs."
        ]
      },
-      "ABILocalFlags": {
-        "Type": "bool",
-        "Default": "false",
-        "Desc": [
-          "When enabled enables an optimization around flags.",
-          "Assumes flags are not used across cals.",
-          "Hand-written assembly can violate this assumption."
-        ]
-      },
-      "ParanoidTSO": {
-        "Type": "bool",
-        "Default": "false",
-        "Desc": [
-          "Makes TSO operations even more strict.",
-          "Forces vector loadstores to also become atomic."
-        ]
-      },
      "StallProcess": {
        "Type": "bool",
        "Default": "false",
@@ -4,7 +4,6 @@
 #include "Common/JitSymbols.h"
 #include "Interface/Core/CPUBackend.h"
 #include "Interface/Core/CPUID.h"
-#include "Interface/Core/X86HelperGen.h"
 #include <Interface/IR/IntrusiveIRList.h>
 #include <FEXCore/Config/Config.h>
 #include <FEXCore/Core/Context.h>
@@ -29,6 +28,7 @@
 namespace FEXCore {
 class SignalDelegator;
 class ThunkHandler;
+struct LookupCacheWriteLockToken;

 namespace Core {
  struct DebugData;
@@ -61,7 +61,7 @@ struct CustomIRResult {
    , Data(Data) {}
 };

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

 class CodeCache : public AbstractCodeCache {
@@ -70,14 +70,54 @@ public:
  ~CodeCache();

  ContextImpl& CTX;
+  fextl::unique_ptr<ContextImpl> ValidationCTX;
+  fextl::unique_ptr<Core::InternalThreadState> ValidationThread;
+  FEXCore::Core::CPUState::gdt_segment ValidationGDT[32] {};
  bool IsGeneratingCache = false;

-  void LoadData(Core::InternalThreadState&, std::byte* MappedCacheFile, const ExecutableFileSectionInfo&) override;
+  FEX_CONFIG_OPT(EnableCodeCaching, ENABLECODECACHINGWIP);
+  FEX_CONFIG_OPT(EnableCodeCacheValidation, ENABLECODECACHEVALIDATION);
+
+  uint64_t ComputeCodeMapId(std::string_view Filename, int FD) override;
  bool SaveData(Core::InternalThreadState&, int TargetFD, const ExecutableFileSectionInfo&, uint64_t SerializedBaseAddress) override;
+  bool LoadData(Core::InternalThreadState*, std::byte* MappedCacheFile, const ExecutableFileSectionInfo&) override;
+
+  /**
+   * Performs expensive extra validation on the loaded code cache data.
+   *
+   * This kicks off an in-process recompile of all cached blocks and compares
+   * them with the cached data. Differences will be reported as fatal errors,
+   * which can uncover bugs like for example:
+   * - mismatches of the JIT configuration used during cache generation
+   * - hidden position dependencies due to missing FEX relocations
+   * - incorrect instruction padding
+   */
+  void Validate(const ExecutableFileSectionInfo&, fextl::set<uint64_t> GuestBlocks, const fextl::set<uint64_t>& HostBlocks,
+                std::span<std::byte> CachedCode);

  void InitiateCacheGeneration() override {
    IsGeneratingCache = true;
  }
+
+  /**
+   * Applies a set of FEX relocations to the given code section.
+   *
+   * FEX relocations describe runtime-dependencies of FEX-generated code.
+   * When loading a code cache, they are used to move cached code to the
+   * dynamically chosen base address of the guest binary.
+   *
+   * Conversely, relocations are applied in reverse when writing code caches
+   * to ensure consistency across generation runs.
+   *
+   * Note that FEX relocations are unrelated to ELF/PE relocations.
+   *
+   * @param GuestDelta Guest address offset to apply to RIP-relative data
+   * @param ForStorage True for serializing data (producing deterministic output); false for de-serializing it (resolving dynamic symbols)
+   *
+   * @return Returns true on success
+   */
+  [[nodiscard]]
+  bool ApplyCodeRelocations(uint64_t GuestDelta, std::span<std::byte> Code, std::span<const CPU::Relocation> Relocations, bool ForStorage);
 };

 class ContextImpl final : public FEXCore::Context::Context, public CPU::CodeBufferManager {
@@ -154,10 +194,20 @@ public:
    return CodeCache;
  }

-  void OnCodeBufferAllocated(CPU::CodeBuffer&) override;
+  void SetCodeMapWriter(fextl::unique_ptr<CodeMapWriter> Writer) override {
+    CodeMapWriter = std::move(Writer);
+  }
+
+  void FlushAndCloseCodeMap() override {
+    if (CodeMapWriter) {
+      CodeMapWriter.reset();
+    }
+  }
+
+  void OnCodeBufferAllocated(const std::shared_ptr<CPU::CodeBuffer>&) override;
  void ClearCodeCache(FEXCore::Core::InternalThreadState* Thread, bool NewCodeBuffer = true) override;
-  void InvalidateGuestCodeRange(FEXCore::Core::InternalThreadState* Thread, InvalidatedEntryAccumulator& Accumulator, uint64_t Start,
-                                uint64_t Length) override;
+  void InvalidateCodeBuffersCodeRange(uint64_t Start, uint64_t Length) override;
+  void InvalidateThreadCachedCodeRange(FEXCore::Core::InternalThreadState* Thread, uint64_t Start, uint64_t Length) override;
  FEXCore::ForkableSharedMutex& GetCodeInvalidationMutex() override {
    return CodeInvalidationMutex;
  }
@@ -197,7 +247,6 @@ public:
    FEX_CONFIG_OPT(TSOEnabled, TSOENABLED);
    FEX_CONFIG_OPT(VectorTSOEnabled, VECTORTSOENABLED);
    FEX_CONFIG_OPT(MemcpySetTSOEnabled, MEMCPYSETTSOENABLED);
-    FEX_CONFIG_OPT(ABILocalFlags, ABILOCALFLAGS);
    FEX_CONFIG_OPT(SMCChecks, SMCCHECKS);
    FEX_CONFIG_OPT(MaxInstPerBlock, MAXINST);
    FEX_CONFIG_OPT(RootFSPath, ROOTFS);
@@ -205,7 +254,6 @@ public:
    FEX_CONFIG_OPT(LibraryJITNaming, LIBRARYJITNAMING);
    FEX_CONFIG_OPT(BlockJITNaming, BLOCKJITNAMING);
    FEX_CONFIG_OPT(GDBSymbols, GDBSYMBOLS);
-    FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
    FEX_CONFIG_OPT(x87ReducedPrecision, X87REDUCEDPRECISION);
    FEX_CONFIG_OPT(DisableTelemetry, DISABLETELEMETRY);
    FEX_CONFIG_OPT(DisableVixlIndirectCalls, DISABLE_VIXL_INDIRECT_RUNTIME_CALLS);
@@ -226,14 +274,12 @@ public:
  FEXCore::ThunkHandler* ThunkHandler {};
  fextl::unique_ptr<FEXCore::CPU::Dispatcher> Dispatcher;
  CodeCache CodeCache;
+  fextl::unique_ptr<CodeMapWriter> CodeMapWriter;

  SignalDelegator* SignalDelegation {};
-  X86GeneratedCode X86CodeGen;

  ContextImpl(const FEXCore::HostFeatures& Features);

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

  // This is used as a replacement for the SMC writes in the mono callsite backpatcher that avoids atomic operations
@@ -268,9 +314,9 @@ public:

  FEXCore::JITSymbols Symbols;

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

  // If Atomic-based TSO emulation is enabled or not.
  bool IsAtomicTSOEnabled() const {
@@ -311,10 +357,6 @@ protected:
      AtomicTSOEmulationEnabled = false;
      VectorAtomicTSOEmulationEnabled = false;
      MemcpyAtomicTSOEmulationEnabled = false;
-    } else if (Config.ParanoidTSO) {
-      AtomicTSOEmulationEnabled = true;
-      VectorAtomicTSOEmulationEnabled = true;
-      MemcpyAtomicTSOEmulationEnabled = true;
    } else {
      AtomicTSOEmulationEnabled = Config.TSOEnabled;
      VectorAtomicTSOEmulationEnabled = Config.TSOEnabled && Config.VectorTSOEnabled;
@@ -353,5 +395,8 @@ private:

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

 namespace FEXCore::IR {

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

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

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

-    return A;
+    return B;
  }

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

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

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

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

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

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

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

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

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

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

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

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

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

@@ -661,7 +677,7 @@ void Arm64Emitter::FillSpecialRegs(ARMEmitter::Register TmpReg, ARMEmitter::Regi
  }
 }

-void Arm64Emitter::SpillStaticRegs(ARMEmitter::Register TmpReg, bool FPRs, uint32_t GPRSpillMask, uint32_t FPRSpillMask) {
+void Arm64Emitter::SpillStaticRegs(ARMEmitter::Register TmpReg, bool FPRs, uint32_t GPRSpillMask, uint32_t FPRSpillMask, bool NZCV) {
 #ifndef VIXL_SIMULATOR
  if (EmitterCTX->HostFeatures.SupportsAFP) {
    // Disable AFP features when spilling registers.
@@ -682,13 +698,15 @@ void Arm64Emitter::SpillStaticRegs(ARMEmitter::Register TmpReg, bool FPRs, uint3
  }
 #endif

-  // Regardless of what GPRs/FPRs we're spilling, we need to spill NZCV since it
-  // is always static and almost certainly clobbered by the subsequent code.
-  //
-  // TODO: Can we prove that NZCV is not used across a call in some cases and
-  // omit this? Might help x87 perf? Future idea.
-  mrs(TmpReg, ARMEmitter::SystemRegister::NZCV);
-  str(TmpReg.W(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.flags[24]));
+  if (NZCV) {
+    // Regardless of what GPRs/FPRs we're spilling, we need to spill NZCV since it
+    // is always static and almost certainly clobbered by the subsequent code.
+    //
+    // TODO: Can we prove that NZCV is not used across a call in some cases and
+    // omit this? Might help x87 perf? Future idea.
+    mrs(TmpReg, ARMEmitter::SystemRegister::NZCV);
+    str(TmpReg.W(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.flags[24]));
+  }

  // PF/AF are special, remove them from the mask
  uint32_t PFAFMask = ((1u << REG_PF.Idx()) | ((1u << REG_AF.Idx())));
@@ -710,7 +728,7 @@ void Arm64Emitter::SpillStaticRegs(ARMEmitter::Register TmpReg, bool FPRs, uint3
  }

  // Now handle PF/AF
-  if (PFAFSpillMask) {
+  if (NZCV && PFAFSpillMask) {
    auto PFOffset = offsetof(FEXCore::Core::CpuStateFrame, State.pf_raw);
    auto AFOffset = offsetof(FEXCore::Core::CpuStateFrame, State.af_raw);
    LOGMAN_THROW_A_FMT(PFAFSpillMask == PFAFMask, "PF/AF not spilled together");
@@ -760,7 +778,7 @@ void Arm64Emitter::SpillStaticRegs(ARMEmitter::Register TmpReg, bool FPRs, uint3
 }

 void Arm64Emitter::FillStaticRegs(bool FPRs, uint32_t GPRFillMask, uint32_t FPRFillMask, std::optional<ARMEmitter::Register> OptionalReg,
-                                  std::optional<ARMEmitter::Register> OptionalReg2) {
+                                  std::optional<ARMEmitter::Register> OptionalReg2, bool NZCV) {
  auto FindTempReg = [this](uint32_t* GPRFillMask) -> std::optional<ARMEmitter::Register> {
    for (auto Reg : StaticRegisters) {
      if (((1U << Reg.Idx()) & *GPRFillMask)) {
@@ -786,7 +804,7 @@ void Arm64Emitter::FillStaticRegs(bool FPRs, uint32_t GPRFillMask, uint32_t FPRF
  auto TmpReg = *OptionalReg;
  auto TmpReg2 = *OptionalReg2;

-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
  // Load STATE in from the CPU area as x28 is not callee saved in the ARM64EC ABI.
  ldr(TmpReg.X(), ARMEmitter::Reg::r18, TEB_CPU_AREA_OFFSET);
  ldr(STATE, TmpReg, CPU_AREA_EMULATOR_DATA_OFFSET);
@@ -794,13 +812,15 @@ void Arm64Emitter::FillStaticRegs(bool FPRs, uint32_t GPRFillMask, uint32_t FPRF

  ldr(REG_CALLRET_SP, STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.callret_sp));

-  // Regardless of what GPRs/FPRs we're filling, we need to fill NZCV since it
-  // is always static and was almost certainly clobbered.
-  //
-  // TODO: Can we prove that NZCV is not used across a call in some cases and
-  // omit this? Might help x87 perf? Future idea.
-  ldr(TmpReg.W(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.flags[24]));
-  msr(ARMEmitter::SystemRegister::NZCV, TmpReg);
+  if (NZCV) {
+    // Regardless of what GPRs/FPRs we're filling, we need to fill NZCV since it
+    // is always static and was almost certainly clobbered.
+    //
+    // TODO: Can we prove that NZCV is not used across a call in some cases and
+    // omit this? Might help x87 perf? Future idea.
+    ldr(TmpReg.W(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.flags[24]));
+    msr(ARMEmitter::SystemRegister::NZCV, TmpReg);
+  }

  FillSpecialRegs(TmpReg, TmpReg2, true, FPRs);

@@ -861,7 +881,7 @@ void Arm64Emitter::FillStaticRegs(bool FPRs, uint32_t GPRFillMask, uint32_t FPRF
  }

  // Now handle PF/AF
-  if (PFAFFillMask) {
+  if (NZCV && PFAFFillMask) {
    LOGMAN_THROW_A_FMT(PFAFFillMask == PFAFMask, "PF/AF not filled together");

    ldp<ARMEmitter::IndexType::OFFSET>(REG_PF.W(), REG_AF.W(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.pf_raw));
@@ -1,36 +1,38 @@
 // SPDX-License-Identifier: MIT
 #pragma once

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

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

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

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

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

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

-#ifndef _M_ARM_64EC
+#ifndef ARCHITECTURE_arm64ec
 // GPR temporaries. Only x3 can be used across spill boundaries
 // so if these ever need to change, be very careful about that.
 constexpr auto TMP1 = ARMEmitter::XReg::x0;
@@ -104,9 +106,20 @@ constexpr ARMEmitter::PRegister PRED_TMP_32B = ARMEmitter::PReg::p7;
 // This class contains common emitter utility functions that can
 // be used by both Arm64 JIT and ARM64 Dispatcher
 class Arm64Emitter : public ARMEmitter::Emitter {
-protected:
+public:
  Arm64Emitter(FEXCore::Context::ContextImpl* ctx, void* EmissionPtr = nullptr, size_t size = 0);

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

  std::span<const ARMEmitter::Register> StaticRegisters {};
@@ -116,18 +129,16 @@ protected:
  std::span<const ARMEmitter::VRegister> GeneralFPRegisters {};
  uint32_t PairRegisters = 0;

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

  // Correlate an ARM register back to an x86 register index.
  // Returning REG_INVALID if there was no mapping.
  FEXCore::X86State::X86Reg GetX86RegRelationToARMReg(ARMEmitter::Register Reg);

-  void SpillStaticRegs(ARMEmitter::Register TmpReg, bool FPRs = true, uint32_t GPRSpillMask = ~0U, uint32_t FPRSpillMask = ~0U);
+  void SpillStaticRegs(ARMEmitter::Register TmpReg, bool FPRs = true, uint32_t GPRSpillMask = ~0U, uint32_t FPRSpillMask = ~0U, bool NZCV = true);
  void FillStaticRegs(bool FPRs = true, uint32_t GPRFillMask = ~0U, uint32_t FPRFillMask = ~0U,
                      std::optional<ARMEmitter::Register> OptionalReg = std::nullopt,
-                      std::optional<ARMEmitter::Register> OptionalReg2 = std::nullopt);
+                      std::optional<ARMEmitter::Register> OptionalReg2 = std::nullopt, bool NZCV = true);

  // Register 0-18 + 29 + 30 are caller saved
  static constexpr uint32_t CALLER_GPR_MASK = 0b0110'0000'0000'0111'1111'1111'1111'1111U;
@@ -270,6 +281,8 @@ protected:

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

 } // namespace FEXCore::CPU
@@ -1,14 +1,18 @@
 // SPDX-License-Identifier: MIT
-#include "FEXCore/IR/IR.h"
-#include "FEXCore/Utils/AllocatorHooks.h"
+#include "FEXCore/Config/Config.h"
 #include "Interface/Context/Context.h"
 #include "Interface/Core/CPUBackend.h"
+#include "Interface/Core/LookupCache.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
+
+#include <FEXCore/IR/IR.h>
+#include <FEXCore/Utils/AllocatorHooks.h>
+#include <FEXCore/Utils/PrctlUtils.h>
+
 #include <cstdint>

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

@@ -274,37 +278,37 @@ namespace CPU {
    : ThreadState(ThreadState)
    , CodeBuffers(CodeBuffers) {

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

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

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

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

 #ifndef FEX_DISABLE_TELEMETRY
    // Fill in telemetry values
    for (size_t i = 0; i < FEXCore::Telemetry::TYPE_LAST; ++i) {
      auto& Telem = FEXCore::Telemetry::GetTelemetryValue(static_cast<FEXCore::Telemetry::TelemetryType>(i));
-      Common.TelemetryValueAddresses[i] = reinterpret_cast<uint64_t>(&Telem);
+      Ptrs.TelemetryValueAddresses[i] = reinterpret_cast<uint64_t>(&Telem);
    }
 #endif
  }
@@ -346,7 +350,7 @@ namespace CPU {
  }

  CodeBuffer::CodeBuffer(size_t Size)
-    : Size(Size) {
+    : AllocatedSize(Size) {
    Ptr = static_cast<uint8_t*>(FEXCore::Allocator::VirtualAlloc(Size, true));
    LOGMAN_THROW_A_FMT(!!Ptr, "Couldn't allocate code buffer");

@@ -357,11 +361,13 @@ namespace CPU {
      LogMan::Msg::EFmt("Failed to mprotect last page of code buffer.");
    }

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

  CodeBuffer::~CodeBuffer() {
-    FEXCore::Allocator::VirtualFree(Ptr, Size);
+    FEXCore::Allocator::VirtualFree(Ptr, AllocatedSize);
  }

  auto CodeBufferManager::AllocateNew(size_t Size) -> fextl::shared_ptr<CodeBuffer> {
@@ -395,14 +401,20 @@ namespace CPU {
    Latest = Buffer;
    LatestOffset = 0;

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

    return Buffer;
  }

  fextl::shared_ptr<CodeBuffer> CodeBufferManager::GetLatest() {
    if (!Latest) {
-      AllocateNew(INITIAL_CODE_SIZE);
+      if (FEXCore::Config::Get_ENABLECODECACHINGWIP()) {
+        // Start with a larger code buffer to avoid resizes that would discard
+        // code loaded from caches
+        AllocateNew(MAX_CODE_SIZE);
+      } else {
+        AllocateNew(INITIAL_CODE_SIZE);
+      }
    }
    return Latest;
  }
@@ -413,7 +425,7 @@ namespace CPU {
      return GetLatest();
    }

-    auto NewCodeBufferSize = GetLatest()->Size;
+    auto NewCodeBufferSize = GetLatest()->AllocatedSize;
    NewCodeBufferSize = std::min<size_t>(NewCodeBufferSize * 2, MAX_CODE_SIZE);
    return AllocateNew(NewCodeBufferSize);
  }
@@ -423,7 +435,7 @@ namespace CPU {
    auto CheckCodeBuffer = [](CodeBuffer& Buffer, uintptr_t Address) {
      // The last page of the code buffer is protected, so we need to exclude it from the valid range
      // when checking if the address is in the code buffer.
-      uintptr_t LastPageAddr = AlignDown(reinterpret_cast<uintptr_t>(Buffer.Ptr) + Buffer.Size - 1, FEXCore::Utils::FEX_PAGE_SIZE);
+      uintptr_t LastPageAddr = AlignDown(reinterpret_cast<uintptr_t>(Buffer.Ptr) + Buffer.AllocatedSize - 1, FEXCore::Utils::FEX_PAGE_SIZE);
      return (Address >= reinterpret_cast<uintptr_t>(Buffer.Ptr) && Address < LastPageAddr);
    };

@@ -43,7 +43,7 @@ struct GuestToHostMap;
 namespace CPU {
  struct CodeBuffer {
    uint8_t* Ptr;
-    size_t Size;
+    size_t AllocatedSize; // including guard page; see UsableSize()

    fextl::unique_ptr<GuestToHostMap> LookupCache;

@@ -54,6 +54,11 @@ namespace CPU {
    CodeBuffer& operator=(CodeBuffer&&) = delete;

    ~CodeBuffer();
+
+    /// Returns the number of bytes available for storing code
+    size_t UsableSize() const {
+      return AllocatedSize - FEXCore::Utils::FEX_PAGE_SIZE;
+    }
  };

  /**
@@ -81,7 +86,7 @@ namespace CPU {
    // Protects writes to the latest CodeBuffer and changes to LatestOffset
    FEXCore::ForkableUniqueMutex CodeBufferWriteMutex;

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

  private:
    fextl::shared_ptr<CodeBuffer> Latest;
@@ -161,7 +166,7 @@ namespace CPU {
    virtual CompiledCode CompileCode(uint64_t Entry, uint64_t Size, bool SingleInst, const FEXCore::IR::IRListView* IR,
                                     FEXCore::Core::DebugData* DebugData, bool CheckTF) = 0;

-    virtual fextl::vector<FEXCore::CPU::Relocation> TakeRelocations() = 0;
+    virtual fextl::vector<FEXCore::CPU::Relocation> TakeRelocations(uint64_t GuestBaseAddress) = 0;

    virtual void ClearCache() {}

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

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

 namespace FEXCore {
 namespace ProductNames {
-#ifdef _M_ARM_64
+#ifdef ARCHITECTURE_arm64
  static const char ARM_UNKNOWN[] = "Unknown ARM CPU";
  static const char ARM_A57[] = "Cortex-A57";
  static const char ARM_A72[] = "Cortex-A72";
@@ -88,12 +89,13 @@ namespace ProductNames {
  static const char ARM_Blizzard_M2Pro[] = "Apple Blizzard (M2 Pro)";
  static const char ARM_Avalanche_M2Max[] = "Apple Avalanche (M2 Max)";
  static const char ARM_Blizzard_M2Max[] = "Apple Blizzard (M2 Max)";
+  static const char ARM_AppleSilicon[] = "Apple Silicon";

  static const char ARM_ORYON_1[] = "Oryon-1";
  static const char ARM_Ampere_1[] = "AmpereOne";
  static const char ARM_Ampere_1A[] = "AmpereOneA";
  static const char ARM_Ampere_1B[] = "AmpereOneB";
-#else
+  static const char ARM_Ampere_1C[] = "AmpereOneC";
 #endif
 } // namespace ProductNames

@@ -138,7 +140,7 @@ constexpr uint32_t FAMILY_IDENTIFIER = GenerateFamily(CPUFamily {
 });
 #endif

-#ifdef _M_ARM_64
+#ifdef ARCHITECTURE_arm64
 uint32_t GetCycleCounterFrequency() {
  uint64_t Result {};
  __asm("mrs %[Res], CNTFRQ_EL0" : [Res] "=r"(Result));
@@ -152,6 +154,7 @@ uint32_t GetCPUID_TPIDRRO() {
 }

 void CPUIDEmu::SetupHostHybridFlag() {
+  FEX_CONFIG_OPT(HideHybrid, HIDEHYBRID);
  PerCPUData.resize(Cores);

  uint64_t MIDR {};
@@ -168,6 +171,11 @@ void CPUIDEmu::SetupHostHybridFlag() {
    MIDR = NewMIDR;
  }

+  if (HideHybrid()) {
+    // Hide the hybrid flag.
+    Hybrid = false;
+  }
+
  struct CPUMIDR {
    uint8_t Implementer;
    uint16_t Part;
@@ -178,7 +186,7 @@ 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, 66> CPUMIDRs = {{
+  static constexpr std::array<CPUMIDR, 67> CPUMIDRs = {{
    // Typically big CPU cores
    {0x51, 0x001, 1, ProductNames::ARM_ORYON_1}, // Qualcomm Oryon-1

@@ -188,6 +196,7 @@ void CPUIDEmu::SetupHostHybridFlag() {
    {0x61, 0x029, 1, ProductNames::ARM_Firestorm_M1Max}, // Apple Firestorm (M1 Max)
    {0x61, 0x025, 1, ProductNames::ARM_Firestorm_M1Pro}, // Apple Firestorm (M1 Pro)
    {0x61, 0x023, 1, ProductNames::ARM_Firestorm_M1},    // Apple Firestorm (M1)
+    {0x61, 0, 1, ProductNames::ARM_AppleSilicon},        // QEmu Apple Silicon

    {0x41, 0xd8c, 1, ProductNames::ARM_C1Ultra},   // C1-Ultra
    {0x41, 0xd90, 1, ProductNames::ARM_C1Premium}, // C1-Premium
@@ -226,6 +235,7 @@ void CPUIDEmu::SetupHostHybridFlag() {
    {0xc0, 0xac3, 1, ProductNames::ARM_Ampere_1},  // AmpereOne
    {0xc0, 0xac4, 1, ProductNames::ARM_Ampere_1A}, // AmpereOneA
    {0xc0, 0xac5, 1, ProductNames::ARM_Ampere_1B}, // AmpereOneB
+    {0xc0, 0xac7, 1, ProductNames::ARM_Ampere_1C}, // AmpereOneC

    {0x4e, 0x010, 1, ProductNames::ARM_Olympus}, // Olympus
    {0x4e, 0x004, 1, ProductNames::ARM_Carmel},  // Carmel
@@ -383,7 +393,8 @@ void CPUIDEmu::SetupHostHybridFlag() {
  } else {
    // If we aren't hybrid then just claim everything is big
    for (size_t i = 0; i < Cores; ++i) {
-      uint32_t MIDR = PerCPUData[i].MIDR;
+      const auto MIDRIndex = HideHybrid() ? 0 : i;
+      uint32_t MIDR = PerCPUData[MIDRIndex].MIDR;
      auto MIDROption = FindDefinedMIDR(MIDR);

      PerCPUData[i].IsBig = true;
@@ -441,10 +452,10 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) const {

  Res.eax = FAMILY_IDENTIFIER;

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

  Res.ecx = (1 << 0) |                                      // SSE3
            (CTX->HostFeatures.SupportsPMULL_128Bit << 1) | // PCLMULQDQ
@@ -507,7 +518,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) const {
            (1 << 25) | // SSE
            (1 << 26) | // SSE2
            (0 << 27) | // Self Snoop
-            (1 << 28) | // Max APIC IDs reserved field is valid
+            (0 << 28) | // (HTT) Max APIC IDs reserved field is valid
            (1 << 29) | // Thermal monitor
            (0 << 30) | // Reserved
            (0 << 31);  // Pending break enable
@@ -854,10 +865,10 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_4000_0001h(uint32_t Leaf) con
  constexpr uint32_t MaximumSubLeafNumber = 2;
  if (Leaf == 0) {
    // EAX[3:0] Is the host architecture that FEX is running under
-#ifdef _M_X86_64
+#ifdef ARCHITECTURE_x86_64
    // EAX[3:0] = 1 = x86_64 host architecture
    Res.eax |= 0b0001;
-#elif defined(_M_ARM_64)
+#elif defined(ARCHITECTURE_arm64)
    // EAX[3:0] = 2 = AArch64 host architecture
    Res.eax |= 0b0010;
 #else
@@ -909,38 +920,38 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0001h(uint32_t Leaf) con

  Res.eax = FAMILY_IDENTIFIER;

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

  Res.edx = (1 << 0) |                                // FPU
            (1 << 1) |                                // Virtual mode extensions
@@ -1094,9 +1105,9 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0008h(uint32_t Leaf) con
            (CTX->HostFeatures.SupportsCLZERO << 0); // CLZERO support

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

  return Res;
 }
@@ -1227,7 +1238,7 @@ CPUIDEmu::CPUIDEmu(const FEXCore::Context::ContextImpl* ctx)

  SetupFeatures();

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

  struct CPUData {
    const char* ProductName {};
-#ifdef _M_ARM_64
+#ifdef ARCHITECTURE_arm64
    uint32_t MIDR {};
 #endif
    bool IsBig {};
@@ -277,7 +277,7 @@ private:
    // 0: Highest function parameter and ID
    {SupportsConstant::CONSTANT, NeedsLeafConstant::NOLEAFCONSTANT},
    // 1: Processor info
-    {SupportsConstant::CONSTANT, NeedsLeafConstant::NOLEAFCONSTANT},
+    {SupportsConstant::NONCONSTANT, NeedsLeafConstant::NOLEAFCONSTANT},
    // 2: Cache and TLB info
    {SupportsConstant::CONSTANT, NeedsLeafConstant::NOLEAFCONSTANT},
    // 3: Serial Number(previously), now reserved
@@ -1,12 +1,218 @@
 // SPDX-License-Identifier: MIT
-#include <Interface/Context/Context.h>
+#include "Utils/SpinWaitLock.h"

+#include <Interface/Context/Context.h>
+#include <Interface/Core/ArchHelpers/Arm64Emitter.h>
+#include <Interface/Core/Dispatcher/Dispatcher.h>
+#include <Interface/Core/JIT/DebugData.h>
+#include <Interface/Core/JIT/Relocations.h>
+#include <Interface/Core/LookupCache.h>
+#include <Interface/Core/OpcodeDispatcher.h>
+#include <Interface/IR/PassManager.h>
+
+#include <FEXCore/Core/Thunks.h>
 #include <FEXCore/HLE/SourcecodeResolver.h>
+#include <FEXCore/HLE/SyscallHandler.h>
+
+#include <FEXHeaderUtils/Filesystem.h>
+
+#include <git_version.h>
+
+#include <xxhash.h>
+
+#include <fstream>

 namespace FEXCore {

+#if __clang_major__ < 16
+ExecutableFileInfo::ExecutableFileInfo(fextl::unique_ptr<HLE::SourcecodeMap> Map, uint64_t FileId, fextl::string Filename)
+  : SourcecodeMap(std::move(Map))
+  , FileId(FileId)
+  , Filename(Filename) {}
+#endif
 ExecutableFileInfo::~ExecutableFileInfo() = default;

+fextl::string CodeMap::GetBaseFilename(const ExecutableFileInfo& MainExecutable, bool AddNombSuffix) {
+  auto FileId = MainExecutable.FileId;
+
+  std::string_view base_filename = FHU::Filesystem::GetFilename(std::string_view {MainExecutable.Filename});
+  if (FileId != 0xffff'ffff'ffff'ffff) {
+    return fextl::fmt::format("{}-{:016x}{}", base_filename, MainExecutable.FileId, AddNombSuffix ? "-nomb" : "");
+  }
+
+  return "";
+}
+
+fextl::map<CodeMapFileId, CodeMap::ParsedContents> CodeMap::ParseCodeMap(std::ifstream& File) {
+  fextl::map<CodeMapFileId, CodeMap::ParsedContents> Ret;
+  while (true) {
+    Entry Entry;
+    File.read(reinterpret_cast<char*>(&Entry), sizeof(Entry));
+    if (!File) {
+      break;
+    }
+
+    if (Entry.FileId == LoadExternalLibrary.FileId && Entry.BlockOffset == LoadExternalLibrary.BlockOffset) {
+      ExternalLibraryInfo Info;
+      File.read(reinterpret_cast<char*>(&Info), sizeof(Info));
+
+      fextl::string Filename;
+      std::getline(File, Filename, '\0');
+
+      // Align to 4-byte boundary
+      char Null[4];
+      File.read(Null, AlignUp(Filename.size() + 1, 4) - Filename.size() - 1);
+      if (!File) {
+        break;
+      }
+      Ret[Info.ExternalFileId].Filename = std::move(Filename);
+    } else if (Entry.FileId == SetExecutableFileId {}.Marker.FileId && Entry.BlockOffset == SetExecutableFileId {}.Marker.BlockOffset) {
+      CodeMapFileId ExecutableFileId;
+      File.read(reinterpret_cast<char*>(&ExecutableFileId), sizeof(ExecutableFileId));
+      if (!File) {
+        break;
+      }
+      Ret[ExecutableFileId].IsExecutable = true;
+    } else {
+      if (!Ret.contains(Entry.FileId)) {
+        LogMan::Msg::EFmt("Code map referenced unknown file id {:016x}", Entry.FileId);
+      } else {
+        Ret[Entry.FileId].Blocks.insert(Entry.BlockOffset);
+      }
+    }
+
+    if (!File) {
+      break;
+    }
+  }
+  return Ret;
+}
+
+CodeMapWriter::CodeMapWriter(CodeMapOpener& Opener, bool OpenEagerly)
+  : Buffer(4096)
+  , FileOpener(Opener) {
+  if (OpenEagerly) {
+    CodeMapFD = FileOpener.OpenCodeMapFile();
+  }
+}
+
+CodeMapWriter::~CodeMapWriter() {
+  if (CodeMapFD.value_or(-1) != -1) {
+    Flush(BufferOffset);
+    close(*CodeMapFD);
+  }
+}
+
+bool CodeMapWriter::IsWriteEnabled(const ExecutableFileSectionInfo& Section) {
+  if (CodeMapFD == -1) {
+    return false;
+  }
+
+  // PV libraries can't yet be read by FEXServer, so skip dumping them
+  if (Section.FileInfo.Filename.starts_with("/run/pressure-vessel")) {
+    return false;
+  }
+
+  if (CodeMapFD) {
+    return true;
+  }
+
+  // Acquire mutex and re-check CodeMapFD to avoid race conditions
+  auto lk = std::unique_lock {Mutex};
+  if (!CodeMapFD) {
+    CodeMapFD = FileOpener.OpenCodeMapFile();
+  }
+
+  return CodeMapFD != -1;
+}
+
+void CodeMapWriter::Flush(size_t Offset) {
+  // Acquire exclusive lock and flush circular buffer
+  std::unique_lock Lock {Mutex};
+  Flush(Offset, Lock);
+}
+
+void CodeMapWriter::Flush(size_t Offset, std::unique_lock<std::shared_mutex>&) {
+  write(*CodeMapFD, Buffer.data(), Offset);
+  BufferOffset = 0;
+}
+
+void CodeMapWriter::AppendBlock(const FEXCore::ExecutableFileSectionInfo& SectionInfo, uint64_t BlockEntry) {
+  if (!IsWriteEnabled(SectionInfo)) {
+    return;
+  }
+
+  BlockEntry -= SectionInfo.FileStartVA;
+  if (BlockEntry > std::numeric_limits<uint32_t>::max()) {
+    ERROR_AND_DIE_FMT("Cannot write code map");
+  }
+
+  // Register new library if not already known
+  bool NewLibraryLoad = false;
+  {
+    // Check prior registration with shared lock
+    std::shared_lock Lock {Mutex};
+    NewLibraryLoad = !KnownFileIds.contains(SectionInfo.FileInfo.FileId);
+  }
+  if (NewLibraryLoad) {
+    // Register to map with exclusive lock
+    std::unique_lock Lock {Mutex};
+    NewLibraryLoad &= KnownFileIds.insert(SectionInfo.FileInfo.FileId).second;
+  }
+  if (NewLibraryLoad) {
+    // Add entry to code map
+    AppendLibraryLoad(SectionInfo.FileInfo);
+  }
+
+  // Register the actual code block
+  CodeMap::Entry DataEntry {SectionInfo.FileInfo.FileId, static_cast<uint32_t>(BlockEntry)};
+  AppendData(std::as_bytes(std::span {&DataEntry, 1}));
+}
+
+void CodeMapWriter::AppendLibraryLoad(const FEXCore::ExecutableFileInfo& FileInfo) {
+  // See CodeMap::ExternalLibraryInfo
+  auto ExternalFileId = FileInfo.FileId;
+  auto TotalSize = AlignUp(sizeof(CodeMap::LoadExternalLibrary) + sizeof(ExternalFileId) + FileInfo.Filename.size() + 1, 4);
+  const auto Data = reinterpret_cast<char*>(alloca(TotalSize));
+  auto WritePtr = std::copy_n(reinterpret_cast<const char*>(&CodeMap::LoadExternalLibrary), sizeof(CodeMap::LoadExternalLibrary), Data);
+  WritePtr = std::copy_n(reinterpret_cast<const char*>(&ExternalFileId), sizeof(ExternalFileId), WritePtr);
+  WritePtr = std::copy(FileInfo.Filename.begin(), FileInfo.Filename.end(), WritePtr);
+  std::fill(WritePtr, Data + TotalSize, 0);
+  AppendData(std::as_bytes(std::span {Data, TotalSize}));
+}
+
+void CodeMapWriter::AppendSetMainExecutable(const FEXCore::ExecutableFileInfo& FileInfo) {
+  CodeMap::SetExecutableFileId Data {.ExecutableFileId = FileInfo.FileId};
+  AppendData(std::span {reinterpret_cast<const std::byte*>(&Data), sizeof(Data)});
+}
+
+void CodeMapWriter::AppendData(std::span<const std::byte> Data) {
+  std::shared_lock Lock {Mutex};
+  auto Offset = BufferOffset.fetch_add(Data.size_bytes());
+  if (Offset + Data.size_bytes() > Buffer.size()) {
+    // Acquire exclusive lock and flush the buffer.
+    // Under heavy pressure, multiple threads may observe an exhausted buffer simultaneously.
+    // The thread with the last in-bounds Offset is responsible for flushing the buffer.
+    Lock.unlock();
+    bool IsResponsibleForFlush = false;
+    {
+      std::unique_lock ExclusiveLock {Mutex};
+      IsResponsibleForFlush = (Offset <= Buffer.size());
+      if (IsResponsibleForFlush) {
+        Flush(Offset, ExclusiveLock);
+      }
+    }
+    if (!IsResponsibleForFlush) {
+      // Wait for the buffer to be flushed on the responsible thread
+      Utils::SpinWaitLock::WaitPred<std::less_equal<>, size_t>(reinterpret_cast<size_t*>(&BufferOffset), Buffer.size());
+    }
+    AppendData(Data);
+    return;
+  }
+
+  memcpy(&Buffer.at(Offset), Data.data(), Data.size_bytes());
+}
+
 } // namespace FEXCore

 namespace FEXCore::Context {
@@ -15,12 +221,445 @@ CodeCache::CodeCache(ContextImpl& CTX_)
  : CTX(CTX_) {}
 CodeCache::~CodeCache() = default;

-void CodeCache::LoadData(Core::InternalThreadState& Thread, std::byte* MappedCacheFile, const ExecutableFileSectionInfo& GuestRIPLookup) {
-  // TODO
+uint64_t CodeCache::ComputeCodeMapId(std::string_view Filename, int FD) {
+  if (Filename.empty()) {
+    return 0xffff'ffff'ffff'ffff;
+  }
+
+  // For now, we just use the file path as an identifier.
+  // TODO: Ensure the hash is unique enough to distinguish executables while remaining independent of the installation location
+  return XXH3_64bits(Filename.data(), Filename.size());
 }

+struct CodeCacheHeader {
+  std::array<char, 4> Magic = ExpectedMagic;
+  uint32_t FormatVersion = 1;
+  uint8_t FEXVersion[20] = {};
+  uint32_t NumBlocks;
+  uint32_t NumCodePages;
+  uint32_t CodeBufferSize;
+  uint32_t NumRelocations;
+  uint32_t padding;
+  uint64_t SerializedBaseAddress;
+  // TODO: Consider including information from LookupCache.BlockLinks
+
+  static constexpr std::array<char, 4> ExpectedMagic = {'F', 'X', 'C', 'C'};
+};
+
+template<typename T>
+concept OrderedContainer = requires { typename T::key_compare; };
+
 bool CodeCache::SaveData(Core::InternalThreadState& Thread, int fd, const ExecutableFileSectionInfo& SourceBinary, uint64_t SerializedBaseAddress) {
-  // TODO
+  auto CodeBuffer = CTX.GetLatest();
+  auto& LookupCache = *Thread.LookupCache->Shared;
+  auto Relocations = Thread.CPUBackend->TakeRelocations(SourceBinary.FileStartVA);
+
+  // Write file header
+  CodeCacheHeader header {};
+  static_assert(GIT_HASH.size() == sizeof(header.FEXVersion));
+  std::ranges::copy(GIT_HASH, header.FEXVersion);
+  header.NumBlocks = LookupCache.BlockList.size();
+  header.NumCodePages = LookupCache.CodePages.size();
+  header.CodeBufferSize = CTX.LatestOffset;
+  header.NumRelocations = Relocations.size();
+  header.SerializedBaseAddress = SerializedBaseAddress;
+  ::write(fd, &header, sizeof(header));
+
+  // Dump guest<->host block mappings
+  {
+    // Cache contents must be deterministic, so copy the unordered block list and then sort by key
+    static_assert(!OrderedContainer<decltype(LookupCache.BlockList)>, "Already deterministic; drop temporary container");
+    fextl::vector<std::pair<uint64_t, const GuestToHostMap::BlockEntry*>> BlockList;
+    BlockList.reserve(LookupCache.BlockList.size());
+    for (auto& [Guest, BlockEntry] : LookupCache.BlockList) {
+      static_assert(sizeof(Guest) == 8, "Breaking change in code cache data layout");
+      BlockList.emplace_back(Guest, &BlockEntry);
+    }
+    std::ranges::sort(BlockList);
+
+    for (auto [Guest, Host] : BlockList) {
+      static_assert(sizeof(Host->HostCode) == 8, "Breaking change in code cache data layout");
+      static_assert(sizeof(Host->CodePages[0]) == 8, "Breaking change in code cache data layout");
+
+      Guest -= SourceBinary.FileStartVA;
+      ::write(fd, &Guest, sizeof(Guest));
+      uint64_t HostCode = Host->HostCode - reinterpret_cast<uintptr_t>(CodeBuffer->Ptr);
+      ::write(fd, &HostCode, sizeof(HostCode));
+      uint64_t NumCodePages = Host->CodePages.size();
+      ::write(fd, &NumCodePages, sizeof(NumCodePages));
+      LOGMAN_THROW_A_FMT(std::ranges::is_sorted(Host->CodePages), "Code pages aren't sorted");
+      for (auto CodePage : Host->CodePages) {
+        CodePage -= SourceBinary.FileStartVA;
+        ::write(fd, &CodePage, sizeof(CodePage));
+      }
+    }
+  }
+
+  // Dump relocations
+  static_assert(sizeof(Relocations[0]) == 48, "Breaking change in code cache data layout");
+  ::write(fd, Relocations.data(), Relocations.size() * sizeof(Relocations[0]));
+
+  // Pad to next page in file so that the CodeBuffer can be mmap'ed into process on load
+  char Zero[64] {};
+  auto Off = lseek(fd, 0, SEEK_CUR);
+  while (Off != AlignUp(Off, Utils::FEX_PAGE_SIZE)) {
+    auto BytesToWrite = std::min(AlignUp(Off, Utils::FEX_PAGE_SIZE) - Off, sizeof(Zero));
+    ::write(fd, Zero, BytesToWrite);
+    Off += BytesToWrite;
+  }
+
+  // Dump the host code (relocated for position-independent serialization)
+  std::span CodeBufferData(reinterpret_cast<std::byte*>(CodeBuffer->Ptr), reinterpret_cast<std::byte*>(CodeBuffer->Ptr) + CTX.LatestOffset);
+  if (!ApplyCodeRelocations(SerializedBaseAddress, CodeBufferData, Relocations, true)) {
+    LOGMAN_THROW_A_FMT(false, "Failed to apply code relocations");
+    return false;
+  }
+  ::write(fd, CodeBufferData.data(), CodeBufferData.size());
+
+  // Dump code pages
+  static_assert(OrderedContainer<decltype(LookupCache.CodePages)>, "Non-deterministic data source");
+  for (const auto& [PageIndex, Entrypoints] : LookupCache.CodePages) {
+    uint64_t PageAddr = (PageIndex << 12) - SourceBinary.FileStartVA;
+    ::write(fd, &PageAddr, sizeof(PageAddr));
+    uint64_t NumEntrypoints = Entrypoints.size();
+    ::write(fd, &NumEntrypoints, sizeof(NumEntrypoints));
+    for (uint64_t Entrypoint : Entrypoints) {
+      Entrypoint -= SourceBinary.FileStartVA;
+      ::write(fd, &Entrypoint, sizeof(Entrypoint));
+    }
+  }
+
+  return true;
+}
+
+bool CodeCache::LoadData(Core::InternalThreadState* Thread, std::byte* MappedCacheFile, const ExecutableFileSectionInfo& BinarySection) {
+  if (!EnableCodeCaching) {
+    return true;
+  }
+
+  namespace ranges = std::ranges;
+
+  // Read file header
+  CodeCacheHeader header {};
+  ::memcpy(&header, MappedCacheFile, sizeof(header));
+  MappedCacheFile += sizeof(header);
+
+  LogMan::Msg::IFmt("Cache load: {:5} blocks; base={:#14x}; off={:#9x}-{:#09x}; {:016x} {}", header.NumBlocks, BinarySection.FileStartVA,
+                    BinarySection.BeginVA - BinarySection.FileStartVA, BinarySection.EndVA - BinarySection.FileStartVA,
+                    BinarySection.FileInfo.FileId, BinarySection.FileInfo.Filename);
+
+  if (!ranges::equal(header.Magic, header.ExpectedMagic)) {
+    LogMan::Msg::EFmt("Invalid cache file header");
+    return false;
+  }
+
+  if (!ranges::equal(header.FEXVersion, GIT_HASH)) {
+    LogMan::Msg::IFmt("Cache generated from old FEX version {:02x}, current is {:02x}; skipping", fmt::join(header.FEXVersion, ""),
+                      fmt::join(GIT_HASH, ""));
+    return false;
+  }
+
+  if (header.NumBlocks == 0) {
+    // Valid caches are never empty
+    LogMan::Msg::IFmt("Code cache empty, aborting");
+    return false;
+  }
+
+  // Read guest<->host block mappings
+  using BlockListEntry = decltype(GuestToHostMap::BlockList)::value_type;
+  fextl::vector<BlockListEntry> BlockList(header.NumBlocks);
+  {
+    for (auto& BlockPtr : BlockList) {
+      ::memcpy(&BlockPtr.first, MappedCacheFile, sizeof(BlockPtr.first));
+      MappedCacheFile += sizeof(BlockPtr.first);
+      ::memcpy(&BlockPtr.second.HostCode, MappedCacheFile, sizeof(BlockPtr.second.HostCode));
+      MappedCacheFile += sizeof(BlockPtr.second.HostCode);
+      uint64_t NumGuestPages;
+      ::memcpy(&NumGuestPages, MappedCacheFile, sizeof(NumGuestPages));
+      MappedCacheFile += sizeof(NumGuestPages);
+
+      BlockPtr.second.CodePages.resize(NumGuestPages);
+      ::memcpy(BlockPtr.second.CodePages.data(), MappedCacheFile, std::span {BlockPtr.second.CodePages}.size_bytes());
+      MappedCacheFile += std::span {BlockPtr.second.CodePages}.size_bytes();
+    }
+
+    // Consistency check: VMA regions at the top and end should belong to the same file
+    auto [min_val, max_val] = ranges::minmax_element(BlockList, std::less {}, &decltype(BlockList)::value_type::first);
+    auto MinBound = CTX.SyscallHandler->LookupExecutableFileSection(Thread, min_val->first + BinarySection.FileStartVA);
+    auto MaxBound = CTX.SyscallHandler->LookupExecutableFileSection(Thread, max_val->first + BinarySection.FileStartVA);
+    if (&MinBound->FileInfo != &BinarySection.FileInfo || &MaxBound->FileInfo != &BinarySection.FileInfo) {
+      ERROR_AND_DIE_FMT("Cached blocks offsets {:#x}-{:#x} out of bounds for guest library {} ({:016x} @ {:#x}) while trying to load "
+                        "section {:#x}-{:#x}!",
+                        min_val->first, max_val->first, BinarySection.FileInfo.Filename, BinarySection.FileInfo.FileId,
+                        BinarySection.FileStartVA, BinarySection.BeginVA, BinarySection.EndVA);
+    }
+
+    // Constrain BlockList to the given ExecutableFileSectionInfo
+    LOGMAN_THROW_A_FMT(ranges::is_sorted(BlockList, [](auto& a, auto& b) { return a.first < b.first; }), "Expected sorted block list");
+    auto begin = ranges::lower_bound(BlockList, BinarySection.BeginVA - BinarySection.FileStartVA, std::less {}, &BlockListEntry::first);
+    auto end =
+      ranges::upper_bound(begin, BlockList.end(), BinarySection.EndVA - BinarySection.FileStartVA - 1, std::less {}, &BlockListEntry::first);
+    BlockList.erase(end, BlockList.end());
+    BlockList.erase(BlockList.begin(), begin);
+    if (BlockList.empty()) {
+      // Not an error since there is just no data to load
+      LogMan::Msg::IFmt("No blocks cached in this range, aborting");
+      return true;
+    }
+  }
+
+  // Read relocations
+  fextl::vector<FEXCore::CPU::Relocation> Relocations(header.NumRelocations, FEXCore::CPU::Relocation::Default());
+  ::memcpy(Relocations.data(), MappedCacheFile, Relocations.size() * sizeof(Relocations[0]));
+  MappedCacheFile += Relocations.size() * sizeof(Relocations[0]);
+
+  // Pad to next page in file, which contains CodeBuffer data
+  MappedCacheFile = reinterpret_cast<std::byte*>(AlignUp(reinterpret_cast<uintptr_t>(MappedCacheFile), Utils::FEX_PAGE_SIZE));
+
+  // Prepare CodeBuffer: Page aligned and big enough to hold all cached data
+  auto Lock = std::unique_lock {CTX.CodeBufferWriteMutex};
+  if (Thread) {
+    if (auto Prev = Thread->CPUBackend->CheckCodeBufferUpdate()) {
+      Allocator::VirtualDontNeed(Thread->CallRetStackBase, FEXCore::Core::InternalThreadState::CALLRET_STACK_SIZE);
+      auto lk = Thread->LookupCache->AcquireWriteLock();
+      Thread->LookupCache->ChangeGuestToHostMapping(*Prev, *CTX.GetLatest()->LookupCache, lk);
+    }
+  }
+
+  auto CodeBuffer = CTX.GetLatest();
+  LOGMAN_THROW_A_FMT(reinterpret_cast<uintptr_t>(CodeBuffer->Ptr) % 0x1000 == 0, "Expected CodeBuffer base to be page-aligned");
+  const auto Delta = AlignUp(CTX.LatestOffset, 0x1000) - CTX.LatestOffset;
+  CTX.LatestOffset += Delta;
+
+  while (CTX.LatestOffset + header.CodeBufferSize > CodeBuffer->UsableSize()) {
+    if (Thread) {
+      CTX.ClearCodeCache(Thread);
+      CodeBuffer = CTX.GetLatest();
+      LogMan::Msg::IFmt("Increased code buffer size to {} MiB for cache load", CodeBuffer->AllocatedSize / 1024 / 1024);
+    } else {
+      ERROR_AND_DIE_FMT("Cannot extend codebuffer without thread!");
+    }
+  }
+
+  // Read CodeBuffer data from file. Make sure the destination is page-aligned.
+  // TODO: Only load the data needed for the selected section
+  auto CodeBufferRange =
+    std::as_writable_bytes(std::span {CodeBuffer->Ptr, CodeBuffer->UsableSize()}).subspan(CTX.LatestOffset, header.CodeBufferSize);
+  ::memcpy(CodeBufferRange.data(), MappedCacheFile, header.CodeBufferSize);
+  MappedCacheFile += header.CodeBufferSize;
+  CTX.LatestOffset += header.CodeBufferSize;
+
+  // Apply FEX relocations
+  auto Ret = ApplyCodeRelocations(BinarySection.FileStartVA, CodeBufferRange, Relocations, false);
+  LOGMAN_THROW_A_FMT(Ret == true, "Failed to apply code cache relocations");
+
+  {
+    auto& LookupCache = *CodeBuffer->LookupCache;
+    auto WriteLock = LookupCache.AcquireWriteLock();
+
+    // Register blocks to LookupCache
+    for (auto& [Guest, Host] : BlockList) {
+      for (auto& CodePage : Host.CodePages) {
+        CodePage += BinarySection.FileStartVA;
+      }
+      auto HostCode = reinterpret_cast<void*>(Host.HostCode + reinterpret_cast<uintptr_t>(CodeBufferRange.data()));
+      LookupCache.AddBlockMapping(Guest + BinarySection.FileStartVA, std::move(Host.CodePages), HostCode, WriteLock);
+    }
+
+    // Register loaded code ranges
+    fextl::vector<uint64_t> Entrypoints;
+    for (uint32_t i = 0; i < header.NumCodePages; ++i) {
+      uint64_t CodePage;
+      memcpy(&CodePage, MappedCacheFile, sizeof(CodePage));
+      CodePage += BinarySection.FileStartVA;
+      MappedCacheFile += sizeof(CodePage);
+
+      uint64_t NumEntrypoints;
+      memcpy(&NumEntrypoints, MappedCacheFile, sizeof(NumEntrypoints));
+      MappedCacheFile += sizeof(NumEntrypoints);
+
+      Entrypoints.resize(NumEntrypoints);
+      memcpy(Entrypoints.data(), MappedCacheFile, NumEntrypoints * sizeof(Entrypoints[0]));
+      MappedCacheFile += NumEntrypoints * sizeof(Entrypoints[0]);
+      for (auto& Entrypoint : Entrypoints) {
+        Entrypoint += BinarySection.FileStartVA;
+      }
+
+      if (LookupCache.AddBlockExecutableRange(Entrypoints, CodePage, FEXCore::Utils::FEX_PAGE_SIZE, WriteLock)) {
+        CTX.SyscallHandler->MarkGuestExecutableRange(Thread, CodePage, FEXCore::Utils::FEX_PAGE_SIZE);
+      }
+    }
+  }
+
+  if (EnableCodeCacheValidation) {
+    fextl::set<uint64_t> GuestBlocks, HostBlocks;
+    for (auto& [Guest, Host] : BlockList) {
+      GuestBlocks.insert(Guest + BinarySection.FileStartVA);
+      HostBlocks.insert(Host.HostCode);
+    }
+
+    Validate(BinarySection, std::move(GuestBlocks), HostBlocks, CodeBufferRange);
+  }
+
+  return true;
+}
+
+void CodeCache::Validate(const ExecutableFileSectionInfo& Section, fextl::set<uint64_t> GuestBlocks, const fextl::set<uint64_t>& HostBlocks,
+                         std::span<std::byte> CachedCode) {
+  LOGMAN_THROW_A_FMT(!HostBlocks.empty(), "Tried to validate without any host blocks");
+  // Skip any cached data before the first host block
+  CachedCode = CachedCode.subspan(*HostBlocks.begin() - sizeof(CPU::CPUBackend::JITCodeHeader));
+
+  if (!ValidationCTX) {
+    ValidationCTX.reset(static_cast<ContextImpl*>(FEXCore::Context::Context::CreateNewContext(CTX.HostFeatures).release()));
+    ValidationCTX->SetSignalDelegator(CTX.SignalDelegation);
+    ValidationCTX->SetSyscallHandler(CTX.SyscallHandler);
+    ValidationCTX->SetThunkHandler(CTX.ThunkHandler);
+    if (!ValidationCTX->InitCore()) {
+      ERROR_AND_DIE_FMT("Failed to create cache load validation context");
+    }
+
+    ValidationThread.reset(ValidationCTX->CreateThread(0, 0, nullptr));
+
+    auto Frame = ValidationThread->CurrentFrame;
+    Frame->State.segment_arrays[FEXCore::Core::CPUState::SEGMENT_ARRAY_INDEX_GDT] = &ValidationGDT[0];
+    Frame->State.segment_arrays[FEXCore::Core::CPUState::SEGMENT_ARRAY_INDEX_LDT] = &ValidationGDT[0];
+    Frame->State.cs_idx = 0;
+    Frame->State.cs_cached = 0;
+
+    if (ValidationCTX->Config.Is64BitMode()) {
+      ValidationGDT[0].L = 1; // L = Long Mode = 64-bit
+      ValidationGDT[0].D = 0; // D = Default Operand Size = Reserved
+    } else {
+      ValidationGDT[0].L = 0; // L = Long Mode = 32-bit
+      ValidationGDT[0].D = 1; // D = Default Operand Size = 32-bit
+    }
+  }
+
+  auto NewCodeBuffer = ValidationCTX->GetLatest();
+  while (CachedCode.size_bytes() > NewCodeBuffer->UsableSize()) {
+    ValidationCTX->ClearCodeCache(ValidationThread.get());
+    NewCodeBuffer = ValidationCTX->GetLatest();
+    LogMan::Msg::IFmt("Increased cache validation code buffer size to {} MiB", NewCodeBuffer->AllocatedSize / 1024 / 1024);
+  }
+
+  std::span<std::byte> CodeBufferRangeRef =
+    std::as_writable_bytes(std::span {NewCodeBuffer->Ptr, NewCodeBuffer->Ptr + NewCodeBuffer->UsableSize()}).subspan(0, CachedCode.size_bytes());
+
+  while (!GuestBlocks.empty()) {
+    auto [CompiledBlocks, _, _2, _3, _4] = ValidationCTX->CompileCode(ValidationThread.get(), *GuestBlocks.begin(), 0 /* TODO: Set MaxInst? */);
+    for (auto& Entry : CompiledBlocks.EntryPoints) {
+      GuestBlocks.erase(Entry.first);
+    }
+  }
+
+  // Patch FEX-internal function addresses with values from the main Context to ensure the code blocks are comparable
+  auto NewRelocations = ValidationThread->CPUBackend->TakeRelocations(Section.FileStartVA);
+  NewRelocations.erase(std::remove_if(NewRelocations.begin(), NewRelocations.end(), [](const CPU::Relocation& Reloc) {
+    return Reloc.Header.Type != CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL && Reloc.Header.Type != CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE;
+  }));
+  (void)ApplyCodeRelocations(Section.FileStartVA, CodeBufferRangeRef, NewRelocations, false);
+
+  if (ValidationCTX->LatestOffset <= CodeBufferRangeRef.size()) {
+    // Reference compilation produced fewer bytes than our cache, so validation is going to fail.
+    // Make sure we don't output any garbage bytes though.
+    CodeBufferRangeRef = CodeBufferRangeRef.subspan(0, ValidationCTX->LatestOffset);
+  }
+
+  auto [Mismatch, _] = std::mismatch(CodeBufferRangeRef.begin(), CodeBufferRangeRef.end(), CachedCode.begin());
+  if (Mismatch != CodeBufferRangeRef.end()) {
+    // Align down to instruction size
+    auto Idx = AlignDown(std::distance(CodeBufferRangeRef.begin(), Mismatch), 4);
+
+    auto BlockIt = std::prev(HostBlocks.lower_bound(*HostBlocks.begin() + Idx + 1));
+    std::optional<uint64_t> GuestBlockAddr;
+    std::optional<uint64_t> GuestBlockAddrRef;
+    if (BlockIt != HostBlocks.end()) {
+      for (int i : {0, 1}) {
+        std::span Buffer = (i == 0 ? CachedCode : CodeBufferRangeRef);
+
+        // Second instruction is always a constant load for relative offset to the (multi)block start
+        int32_t addr = (*reinterpret_cast<uint32_t*>(&Buffer[*BlockIt - *HostBlocks.begin() + 4]) & 0x3ff'ffe0) << 11;
+        addr >>= 14;
+        auto header = reinterpret_cast<CPU::CPUBackend::JITCodeHeader*>(&Buffer[*BlockIt - *HostBlocks.begin() + 4 + addr]);
+        auto tail = reinterpret_cast<CPU::CPUBackend::JITCodeTail*>(reinterpret_cast<uintptr_t>(header) + header->OffsetToBlockTail);
+        (i == 0 ? GuestBlockAddr : GuestBlockAddrRef) = tail->RIP - Section.FileStartVA;
+        LogMan::Msg::EFmt("Recorded rip {}: {:#x} (offset {:#x})", i, tail->RIP, tail->RIP - Section.FileStartVA);
+
+        if (i == 1) {
+          if (tail->RIP >= Section.BeginVA && tail->RIP < Section.EndVA) {
+            auto [IRView, TotalInstructions, TotalInstructionsLength, StartAddr, Length, _] =
+              ValidationCTX->GenerateIR(ValidationThread.get(), tail->RIP, false, FEXCore::Config::Get_MAXINST());
+            fextl::stringstream ss;
+            FEXCore::IR::Dump(&ss, &*IRView);
+            LogMan::Msg::EFmt("IR:\n{}", ss.str());
+          } else {
+            LogMan::Msg::EFmt("Can't dump IR for out-of-range RIP {:#x}", tail->RIP);
+          }
+        }
+      }
+    }
+
+    fextl::string GuestBlockInfo = "UNKNOWN";
+    if (GuestBlockAddr) {
+      GuestBlockInfo = fextl::fmt::format("{:#x}", GuestBlockAddr.value());
+    }
+    if (GuestBlockAddr != GuestBlockAddrRef) {
+      GuestBlockInfo += " (MISMATCH)";
+    }
+    ERROR_AND_DIE_FMT("Cache validation failed at offset {:#x}: {:02x} <-> {:02x} (at {} <-> {}, guest block {})", Idx,
+                      fmt::join(CachedCode.subspan(Idx, 4), ""), fmt::join(CodeBufferRangeRef.subspan(Idx, 4), ""),
+                      fmt::ptr(CachedCode.data()), fmt::ptr(CodeBufferRangeRef.data()), GuestBlockInfo);
+  }
+
+  // Reset Context state for next validation
+  ValidationThread->LookupCache->ClearCache(ValidationThread->LookupCache->AcquireWriteLock());
+  ValidationCTX->LatestOffset = 0;
+
+  LogMan::Msg::IFmt("\tSuccessfully validated cache");
+}
+
+bool CodeCache::ApplyCodeRelocations(uint64_t GuestEntry, std::span<std::byte> Code,
+                                     std::span<const FEXCore::CPU::Relocation> EntryRelocations, bool ForStorage) {
+  CPU::Arm64Emitter Emitter(&CTX, Code.data(), Code.size_bytes());
+  for (size_t j = 0; j < EntryRelocations.size(); ++j) {
+    const FEXCore::CPU::Relocation& Reloc = EntryRelocations[j];
+    Emitter.SetCursorOffset(Reloc.Header.Offset);
+
+    switch (Reloc.Header.Type) {
+    case FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL: {
+      // Generate a literal so we can place it
+      uint64_t Pointer = ForStorage ? 0 : GetNamedSymbolLiteral(CTX, Reloc.NamedSymbolLiteral.Symbol);
+      Emitter.dc64(Pointer);
+      break;
+    }
+    case FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE: {
+      uint64_t Pointer = ForStorage ? 0 : reinterpret_cast<uint64_t>(CTX.ThunkHandler->LookupThunk(Reloc.NamedThunkMove.Symbol));
+      if (Pointer == ~0ULL) {
+        return false;
+      }
+      // TODO: Pointers are required to fit within 48-bit VA space.
+      // But forcing 6-byte broke relocations.
+      Emitter.LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Register(Reloc.NamedThunkMove.RegisterIndex), Pointer,
+                           CPU::Arm64Emitter::PadType::DOPAD);
+      break;
+    }
+    case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_LITERAL: {
+      Emitter.dc64(GuestEntry + Reloc.GuestRIP.GuestRIP);
+      break;
+    }
+    case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE: {
+      uint64_t Pointer = Reloc.GuestRIP.GuestRIP + GuestEntry;
+      // TODO: Pointers are required to fit within 48-bit VA space.
+      // But forcing 6-byte broke relocations.
+      Emitter.LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Register(Reloc.GuestRIP.RegisterIndex), Pointer, CPU::Arm64Emitter::PadType::DOPAD);
+      break;
+    }
+
+    default: ERROR_AND_DIE_FMT("Unknown relocation type {}", ToUnderlying(Reloc.Header.Type));
+    }
+  }
+
  return true;
 }

@@ -9,6 +9,9 @@ $end_info$
 */

 #include <cstdint>
+#ifdef ZYDIS_DISASSEMBLER
+#include <Zydis/Zydis.h>
+#endif
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
 #include "Interface/Core/LookupCache.h"
 #include "Interface/Core/CPUBackend.h"
@@ -345,7 +348,7 @@ bool ContextImpl::InitCore() {
  // Set up the SignalDelegator config since core is initialized.
  SignalDelegation->SetConfig(Dispatcher->MakeSignalDelegatorConfig());

-#if defined(_WIN32) && !defined(_M_ARM_64EC)
+#if defined(_WIN32) && !defined(ARCHITECTURE_arm64ec)
  // WOW64 always needs the interrupt fault check to be enabled.
  Config.NeedsPendingInterruptFaultCheck = true;
 #endif
@@ -363,6 +366,9 @@ void ContextImpl::HandleCallback(FEXCore::Core::InternalThreadState* Thread, uin
 }

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

  // If it is the parent thread that died then just leave
@@ -376,8 +382,10 @@ void ContextImpl::InitializeCompiler(FEXCore::Core::InternalThreadState* Thread)
  Thread->FrontendDecoder = fextl::make_unique<FEXCore::Frontend::Decoder>(Thread);
  Thread->PassManager = fextl::make_unique<FEXCore::IR::PassManager>();

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

  Dispatcher->InitThreadPointers(Thread);

@@ -398,6 +406,7 @@ ContextImpl::CreateThread(uint64_t InitialRIP, uint64_t StackPointer, const FEXC
  FEXCore::Core::InternalThreadState* Thread = new FEXCore::Core::InternalThreadState {
    .CTX = this,
  };
+  FEXCore::Allocator::VirtualName("FEXMem_ThreadState", Thread, sizeof(*Thread));

  Thread->CurrentFrame->State.gregs[X86State::REG_RSP] = StackPointer;
  Thread->CurrentFrame->State.rip = InitialRIP;
@@ -434,6 +443,10 @@ void ContextImpl::UnlockAfterFork(FEXCore::Core::InternalThreadState* LiveThread

  Profiler::PostForkAction(Child);
  if (Child) {
+    if (CodeMapWriter) {
+      CodeMapWriter->ResetAfterFork();
+    }
+
    CodeInvalidationMutex.StealAndDropActiveLocks();
    if (Config.StrictInProcessSplitLocks) {
      StrictSplitLockMutex = 0;
@@ -456,9 +469,14 @@ void ContextImpl::LockBeforeFork(FEXCore::Core::InternalThreadState* Thread) {
 }
 #endif

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

@@ -470,7 +488,8 @@ void ContextImpl::ClearCodeCache(FEXCore::Core::InternalThreadState* Thread, boo
    Thread->CPUBackend->ClearCache();
  } else {
    // Clear L1+L2 cache of this thread, and clear L3 cache across any threads using it
-    Thread->LookupCache->ClearCache();
+    auto lk = Thread->LookupCache->AcquireWriteLock();
+    Thread->LookupCache->ClearCache(lk);
  }
  Allocator::VirtualDontNeed(Thread->CallRetStackBase, FEXCore::Core::InternalThreadState::CALLRET_STACK_SIZE);
 }
@@ -523,9 +542,24 @@ ContextImpl::GenerateIR(FEXCore::Core::InternalThreadState* Thread, uint64_t Gue

    const auto GPRSize = Thread->OpDispatcher->GetGPROpSize();

+#ifdef ZYDIS_DISASSEMBLER
+    const auto ZydisMachineMode = Config.Is64BitMode ? ZYDIS_MACHINE_MODE_LONG_64 : ZYDIS_MACHINE_MODE_LEGACY_32;
+    if (FEXCore::Config::Get_X86DISASSEMBLE()) {
+      const uint64_t DecodedMin = Thread->FrontendDecoder->DecodedMinAddress;
+      const uint64_t DecodedMax = Thread->FrontendDecoder->DecodedMaxAddress;
+      LogMan::Msg::IFmt("Guest x86 Begin (RIP={:#x}, {:#x}-{:#x})", GuestRIP, DecodedMin, DecodedMax);
+    }
+#endif
+
    for (size_t j = 0; j < CodeBlocks->size(); ++j) {
      const FEXCore::Frontend::Decoder::DecodedBlocks& Block = CodeBlocks->at(j);

+#ifdef ZYDIS_DISASSEMBLER
+      if (FEXCore::Config::Get_X86DISASSEMBLE() && CodeBlocks->size() > 1) {
+        LogMan::Msg::IFmt("  Block {} Entry={:#x} NumInsts={}", j, Block.Entry, Block.NumInstructions);
+      }
+#endif
+
      bool BlockInForceTSOValidRange = false;
      auto InstForceTSOIt = ForceTSOInstructions.end();
      if (ForceTSOValidRanges.Contains({Block.Entry, Block.Entry + Block.Size})) {
@@ -557,6 +591,19 @@ ContextImpl::GenerateIR(FEXCore::Core::InternalThreadState* Thread, uint64_t Gue

        TableInfo = Block.DecodedInstructions[i].TableInfo;
        DecodedInfo = &Block.DecodedInstructions[i];
+
+#ifdef ZYDIS_DISASSEMBLER
+        if (FEXCore::Config::Get_X86DISASSEMBLE()) {
+          const uint8_t* InstBytes = reinterpret_cast<const uint8_t*>(InstAddress);
+          ZydisDisassembledInstruction ZydisInst;
+          if (ZYAN_SUCCESS(ZydisDisassembleIntel(ZydisMachineMode, InstAddress, InstBytes, DecodedInfo->InstSize, &ZydisInst))) {
+            LogMan::Msg::IFmt("    {:#x}: {}", InstAddress, ZydisInst.text);
+          } else {
+            LogMan::Msg::IFmt("    {:#x}: (decode failed, {} bytes)", InstAddress, DecodedInfo->InstSize);
+          }
+        }
+#endif
+
        bool IsLocked = DecodedInfo->Flags & FEXCore::X86Tables::DecodeFlags::FLAG_LOCK;

        // Do a partial register cache flush before every instruction. This
@@ -642,10 +689,11 @@ ContextImpl::GenerateIR(FEXCore::Core::InternalThreadState* Thread, uint64_t Gue
              LogMan::Msg::EFmt("Invalid or Unknown instruction: {} 0x{:x}", TableInfo->Name ?: "UND", Block.Entry - GuestRIP);
            }

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

@@ -676,6 +724,12 @@ ContextImpl::GenerateIR(FEXCore::Core::InternalThreadState* Thread, uint64_t Gue
      }
    }

+#ifdef ZYDIS_DISASSEMBLER
+    if (FEXCore::Config::Get_X86DISASSEMBLE()) {
+      LogMan::Msg::IFmt("Guest x86 End");
+    }
+#endif
+
    Thread->OpDispatcher->Finalize();

    Thread->FrontendDecoder->DelayedDisownBuffer();
@@ -709,9 +763,10 @@ ContextImpl::GenerateIR(FEXCore::Core::InternalThreadState* Thread, uint64_t Gue

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

@@ -728,7 +783,7 @@ ContextImpl::CompileCodeResult ContextImpl::CompileCode(FEXCore::Core::InternalT
  // but this would increase lock contention. Redundant frontend runs aren't
  // as expensive and are easily reverted.
  if (MaxInst != 1) {
-    if (auto Block = Thread->LookupCache->FindBlock(GuestRIP)) {
+    if (auto Block = Thread->LookupCache->FindBlock(Thread, GuestRIP)) {
      Thread->OpDispatcher->DelayedDisownBuffer();
      return {.CompiledCode = {.BlockBegin = reinterpret_cast<uint8_t*>(Block), .EntryPoints = {{GuestRIP, reinterpret_cast<uint8_t*>(Block)}}},
              .DebugData = nullptr,
@@ -769,10 +824,13 @@ uintptr_t ContextImpl::CompileBlock(FEXCore::Core::CpuStateFrame* Frame, uint64_

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

+  // Accumulate a JIT count now, as even if another thread raced us, it should count as a compile.
+  FEXCORE_PROFILE_INSTANT_INCREMENT(Thread, AccumulatedJITCount, 1);
+
  auto [CompiledCode, DebugData, StartAddr, Length, NeedsAddGuestCodeRanges] = CompileCode(Thread, GuestRIP, MaxInst);
  auto CodePtr = CompiledCode.EntryPoints[GuestRIP];
  if (CodePtr == nullptr) {
@@ -786,7 +844,7 @@ uintptr_t ContextImpl::CompileBlock(FEXCore::Core::CpuStateFrame* Frame, uint64_
  if (Config.BlockJITNaming()) {
    auto FragmentBasePtr = CompiledCode.BlockBegin;

-    auto GuestRIPLookup = SyscallHandler->LookupExecutableFileSection(*Thread, GuestRIP);
+    auto GuestRIPLookup = SyscallHandler->LookupExecutableFileSection(Thread, GuestRIP);

    if (DebugData->Subblocks.size()) {
      for (auto& Subblock : DebugData->Subblocks) {
@@ -809,7 +867,7 @@ uintptr_t ContextImpl::CompileBlock(FEXCore::Core::CpuStateFrame* Frame, uint64_
  }

  if (Config.LibraryJITNaming() || Config.GDBSymbols()) {
-    auto MappedSection = SyscallHandler->LookupExecutableFileSection(*Thread, GuestRIP);
+    auto MappedSection = SyscallHandler->LookupExecutableFileSection(Thread, GuestRIP);
    if (MappedSection) {
      if (Config.LibraryJITNaming()) {
        Symbols.RegisterNamedRegion(Thread->SymbolBuffer.get(), CodePtr, DebugData->HostCodeSize, MappedSection->FileInfo.Filename);
@@ -826,20 +884,32 @@ uintptr_t ContextImpl::CompileBlock(FEXCore::Core::CpuStateFrame* Frame, uint64_
    Thread->CPUBackend->ClearRelocations();
  }

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

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

  return (uintptr_t)CodePtr;
@@ -866,49 +936,37 @@ uintptr_t ContextImpl::CompileSingleStep(FEXCore::Core::CpuStateFrame* Frame, ui
  return (uintptr_t)CodePtr;
 }

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

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

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

-void ContextImpl::InvalidateGuestCodeRange(FEXCore::Core::InternalThreadState* Thread, InvalidatedEntryAccumulator& Accumulator,
-                                           uint64_t Start, uint64_t Length) {
-  InvalidateGuestThreadCodeRange(Thread, Accumulator, Start, Length);
-}
-
-bool ContextImpl::ThreadRemoveCodeEntry(FEXCore::Core::InternalThreadState* Thread, uint64_t GuestRIP) {
-  LogMan::Throw::AFmt(static_cast<ContextImpl*>(Thread->CTX)->CodeInvalidationMutex.try_lock() == false, "CodeInvalidationMutex needs to "
-                                                                                                         "be unique_locked here");
-
-  return Thread->LookupCache->Erase(Thread->CurrentFrame, GuestRIP);
-}
-
 void ContextImpl::ThreadRemoveCodeEntryFromJit(FEXCore::Core::CpuStateFrame* Frame, uint64_t GuestRIP) {
  static_cast<ContextImpl*>(Frame->Thread->CTX)->SyscallHandler->InvalidateGuestCodeRange(Frame->Thread, GuestRIP, 1);
 }
@@ -950,12 +1008,13 @@ void ContextImpl::AddThunkTrampolineIRHandler(uintptr_t Entrypoint, uintptr_t Gu

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

+      // Thunk entry-points don't get cached, don't need to be padded.
      if (GPRSize == IR::OpSize::i64Bit) {
        IR::Ref R = emit->_StoreRegister(emit->Constant(Entrypoint), GPRSize);
-        R->Reg = IR::PhysicalRegister(IR::GPRFixedClass, X86State::REG_R11).Raw;
+        R->Reg = IR::PhysicalRegister(IR::RegClass::GPRFixed, X86State::REG_R11).Raw;
      } else {
-        emit->_StoreContext(GPRSize, IR::FPRClass, emit->_VCastFromGPR(IR::OpSize::i64Bit, IR::OpSize::i64Bit, emit->Constant(Entrypoint)),
-                            offsetof(Core::CPUState, mm[0][0]));
+        emit->_StoreContextFPR(GPRSize, emit->_VCastFromGPR(IR::OpSize::i64Bit, IR::OpSize::i64Bit, emit->Constant(Entrypoint)),
+                               offsetof(Core::CPUState, mm[0][0]));
      }
      emit->_ExitFunction(IR::OpSize::i64Bit, emit->Constant(GuestThunkEntrypoint), IR::BranchHint::None, emit->Invalid(), emit->Invalid());
    },
@@ -976,7 +1035,7 @@ void ContextImpl::AddThunkTrampolineIRHandler(uintptr_t Entrypoint, uintptr_t Gu
 void ContextImpl::AddForceTSOInformation(const IntervalList<uint64_t>& ValidRanges, fextl::set<uint64_t>&& Instructions) {
  LogMan::Throw::AFmt(CodeInvalidationMutex.try_lock() == false, "CodeInvalidationMutex needs to be unique_locked here");
  ForceTSOValidRanges.Insert(ValidRanges);
-  ForceTSOInstructions.merge(Instructions);
+  ForceTSOInstructions.merge(std::move(Instructions));
 }

 void ContextImpl::RemoveForceTSOInformation(uint64_t Address, uint64_t Size) {
@@ -1020,6 +1079,5 @@ void ContextImpl::MonoBackpatcherWrite(FEXCore::Core::CpuStateFrame* Frame, uint

 void ContextImpl::ConfigureAOTGen(FEXCore::Core::InternalThreadState* Thread, fextl::set<uint64_t>* ExternalBranches, uint64_t SectionMaxAddress) {
  Thread->FrontendDecoder->SetExternalBranches(ExternalBranches);
-  Thread->FrontendDecoder->SetSectionMaxAddress(SectionMaxAddress);
 }
 } // namespace FEXCore::Context
@@ -5,7 +5,6 @@
 #include "Interface/Core/CPUBackend.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Core/LookupCache.h"
-#include "Interface/Core/X86HelperGen.h"
 #include "Utils/MemberFunctionToPointer.h"

 #include <FEXCore/Config/Config.h>
@@ -27,7 +26,6 @@

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

 namespace FEXCore::CPU {
@@ -36,12 +34,14 @@ static void SleepThread(FEXCore::Context::ContextImpl* CTX, FEXCore::Core::CpuSt
  CTX->SyscallHandler->SleepThread(CTX, Frame);
 }

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

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

 Dispatcher::~Dispatcher() {
@@ -95,8 +95,8 @@ void Dispatcher::EmitDispatcher() {

  ARMEmitter::BiDirectionalLabel LoopTop {};

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

  AbsoluteLoopTopAddressEnterECFillSRA = GetCursorAddress<uint64_t>();
  ldr(STATE, EC_ENTRY_CPUAREA_REG, CPU_AREA_EMULATOR_DATA_OFFSET);
@@ -104,10 +104,10 @@ void Dispatcher::EmitDispatcher() {

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

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

  AbsoluteLoopTopAddressEnterEC = GetCursorAddress<uint64_t>();
  // Load ThreadState and write the target PC there
@@ -128,7 +128,7 @@ void Dispatcher::EmitDispatcher() {
  ldp<ARMEmitter::IndexType::OFFSET>(TMP1, TMP2, REG_CALLRET_SP);
  // EC_CALL_CHECKER_PC_REG is REG_PF which isn't touched by any of the above
  sub(ARMEmitter::Size::i64Bit, TMP1, EC_CALL_CHECKER_PC_REG, TMP1);
-  cbnz(ARMEmitter::Size::i64Bit, TMP1, &LoopTop);
+  (void)cbnz(ARMEmitter::Size::i64Bit, TMP1, &LoopTop);

  // If the entry at the TOS is for the target address, pop it and return to the JIT code
  add(ARMEmitter::Size::i64Bit, REG_CALLRET_SP, REG_CALLRET_SP, 0x10);
@@ -146,7 +146,7 @@ void Dispatcher::EmitDispatcher() {
  // Load in our RIP
  ldr(RipReg, STATE_PTR(CpuStateFrame, State.rip));

-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
  // Clobbers TMP1/2
  // Check the EC code bitmap incase we need to exit the JIT to call into native code.
  ARMEmitter::ForwardLabel l_NotECCode;
@@ -158,76 +158,82 @@ void Dispatcher::EmitDispatcher() {
  ldr(TMP1, TMP1, TMP2, ARMEmitter::ExtendedType::LSL_64, 0);
  lsr(ARMEmitter::Size::i64Bit, TMP2, RipReg, 12);
  lsrv(ARMEmitter::Size::i64Bit, TMP1, TMP1, TMP2);
-  tbz(TMP1, 0, &l_NotECCode);
+  (void)tbz(TMP1, 0, &l_NotECCode);

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

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

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

  ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
  (void)cbnz(ARMEmitter::Size::i32Bit, TMP1, &CompileSingleStep);

-  // This is the block cache lookup routine
-  // It matches what is going on it LookupCache.h::FindBlock
-  ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.L2Pointer));
-
-  // Mask the address by the virtual address size so we can check for aliases
-  uint64_t VirtualMemorySize = CTX->Config.VirtualMemSize;
-  if (std::popcount(VirtualMemorySize) == 1) {
-    and_(ARMEmitter::Size::i64Bit, TMP4, RipReg.R(), VirtualMemorySize - 1);
-  } else {
-    LoadConstant(ARMEmitter::Size::i64Bit, TMP4, VirtualMemorySize);
-    and_(ARMEmitter::Size::i64Bit, TMP4, RipReg.R(), TMP4);
-  }
-
  ARMEmitter::ForwardLabel NoBlock;

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

-    // Load the pointer from the offset
-    ldr(TMP1, TMP1, TMP2, ARMEmitter::ExtendedType::LSL_64, 3);
+    // Mask the address by the virtual address size so we can check for aliases
+    uint64_t VirtualMemorySize = CTX->Config.VirtualMemSize;
+    if (std::popcount(VirtualMemorySize) == 1) {
+      and_(ARMEmitter::Size::i64Bit, TMP4, RipReg.R(), VirtualMemorySize - 1);
+    } else {
+      LoadConstant(ARMEmitter::Size::i64Bit, TMP4, VirtualMemorySize);
+      and_(ARMEmitter::Size::i64Bit, TMP4, RipReg.R(), TMP4);
+    }

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

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

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

@@ -252,7 +258,7 @@ void Dispatcher::EmitDispatcher() {
    str(TMP2, STATE, offsetof(FEXCore::Core::CPUState, DeferredSignalRefCount));
 #endif

-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
    ldr(TMP2, ARMEmitter::XReg::x18, TEB_CPU_AREA_OFFSET);
    LoadConstant(ARMEmitter::Size::i32Bit, TMP1, 1);
    strb(TMP1.W(), TMP2, CPU_AREA_IN_SYSCALL_CALLBACK_OFFSET);
@@ -260,7 +266,7 @@ void Dispatcher::EmitDispatcher() {

    Body();

-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
    ldr(TMP2, ARMEmitter::XReg::x18, TEB_CPU_AREA_OFFSET);
    strb(ARMEmitter::WReg::zr, TMP2, CPU_AREA_IN_SYSCALL_CALLBACK_OFFSET);
 #endif
@@ -284,7 +290,7 @@ void Dispatcher::EmitDispatcher() {
      mov(ARMEmitter::XReg::x0, STATE);
      mov(ARMEmitter::XReg::x1, ARMEmitter::XReg::lr);

-      ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionLink));
+      ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, Pointers.ExitFunctionLink));
      if (!CTX->Config.DisableVixlIndirectCalls) [[unlikely]] {
        GenerateIndirectRuntimeCall<uintptr_t, void*, void*>(ARMEmitter::Reg::r2);
      } else {
@@ -481,7 +487,7 @@ void Dispatcher::EmitDispatcher() {

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

    ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]));
    sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r2, CTX->Config.Is64BitMode ? 16 : 12);
@@ -537,8 +543,8 @@ void Dispatcher::EmitDispatcher() {
    return Address;
  };

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

  // Interpreter fallbacks
  {
@@ -619,6 +625,226 @@ void Dispatcher::ExecuteJITCallback(FEXCore::Core::CpuStateFrame* Frame, uint64_

 #endif

+void Dispatcher::EmitI32ToExtF80() {
+  ARMEmitter::ForwardLabel ZeroCase;
+  ARMEmitter::ForwardLabel Done;
+
+  (void)cbz(ARMEmitter::Size::i32Bit, TMP2, &ZeroCase);
+
+  lsr(ARMEmitter::Size::i32Bit, TMP4, TMP2, 31);
+  tst(ARMEmitter::Size::i32Bit, TMP2, TMP2);
+  neg(ARMEmitter::Size::i32Bit, TMP3, TMP2);
+  csel(ARMEmitter::Size::i32Bit, TMP3, TMP3, TMP2, ARMEmitter::Condition::CC_MI);
+
+  clz(ARMEmitter::Size::i32Bit, TMP1, TMP3);
+
+  mov(ARMEmitter::Size::i32Bit, TMP2, 0x401E);
+  sub(ARMEmitter::Size::i32Bit, TMP2, TMP2, TMP1);
+  orr(ARMEmitter::Size::i32Bit, TMP2, TMP2, TMP4, ARMEmitter::ShiftType::LSL, 15);
+
+  lslv(ARMEmitter::Size::i32Bit, TMP3, TMP3, TMP1);
+
+  lsl(ARMEmitter::Size::i64Bit, TMP3, TMP3, 32);
+
+  fmov(ARMEmitter::Size::i64Bit, VTMP1.D(), TMP3);
+  ins(ARMEmitter::SubRegSize::i16Bit, VTMP1, 4, TMP2);
+
+  (void)b(&Done);
+
+  (void)Bind(&ZeroCase);
+  movi(ARMEmitter::SubRegSize::i64Bit, VTMP1.Q(), 0);
+
+  (void)Bind(&Done);
+}
+
+void Dispatcher::EmitI16ToExtF80() {
+  sxth(ARMEmitter::Size::i32Bit, TMP2, TMP2);
+
+  ARMEmitter::ForwardLabel ZeroCase;
+  ARMEmitter::ForwardLabel Done;
+
+  (void)cbz(ARMEmitter::Size::i32Bit, TMP2, &ZeroCase);
+
+  lsr(ARMEmitter::Size::i32Bit, TMP4, TMP2, 31);
+  tst(ARMEmitter::Size::i32Bit, TMP2, TMP2);
+  neg(ARMEmitter::Size::i32Bit, TMP3, TMP2);
+  csel(ARMEmitter::Size::i32Bit, TMP3, TMP3, TMP2, ARMEmitter::Condition::CC_MI);
+
+  clz(ARMEmitter::Size::i32Bit, TMP1, TMP3);
+
+  mov(ARMEmitter::Size::i32Bit, TMP2, 0x401E);
+  sub(ARMEmitter::Size::i32Bit, TMP2, TMP2, TMP1);
+  orr(ARMEmitter::Size::i32Bit, TMP2, TMP2, TMP4, ARMEmitter::ShiftType::LSL, 15);
+
+  lslv(ARMEmitter::Size::i32Bit, TMP3, TMP3, TMP1);
+
+  lsl(ARMEmitter::Size::i64Bit, TMP3, TMP3, 32);
+
+  fmov(ARMEmitter::Size::i64Bit, VTMP1.D(), TMP3);
+  ins(ARMEmitter::SubRegSize::i16Bit, VTMP1, 4, TMP2);
+
+  (void)b(&Done);
+
+  (void)Bind(&ZeroCase);
+  movi(ARMEmitter::SubRegSize::i64Bit, VTMP1.Q(), 0);
+
+  (void)Bind(&Done);
+}
+
+void Dispatcher::EmitF32ToExtF80() {
+  ARMEmitter::ForwardLabel InfNaN;
+  ARMEmitter::ForwardLabel ZeroDenormal;
+  ARMEmitter::ForwardLabel Denormal;
+  ARMEmitter::ForwardLabel NaN;
+  ARMEmitter::ForwardLabel Done;
+  ARMEmitter::BiDirectionalLabel NormalPath;
+  ARMEmitter::ForwardLabel ZeroResult;
+
+  fmov(ARMEmitter::Size::i32Bit, TMP1, VTMP1.S());
+
+  ubfx(ARMEmitter::Size::i32Bit, TMP2, TMP1, 23, 8);
+  and_(ARMEmitter::Size::i32Bit, TMP3, TMP1, 0x007FFFFF);
+  lsr(ARMEmitter::Size::i32Bit, TMP4, TMP1, 31);
+
+  cmp(ARMEmitter::Size::i32Bit, TMP2, 0xFF);
+  (void)b(ARMEmitter::Condition::CC_EQ, &InfNaN);
+
+  (void)cbz(ARMEmitter::Size::i32Bit, TMP2, &ZeroDenormal);
+
+  (void)Bind(&NormalPath);
+  // Exponent bias adjustment, where bias is 0x3F80
+  LoadConstant(ARMEmitter::Size::i32Bit, TMP1, 0x3F80);
+  add(ARMEmitter::Size::i32Bit, TMP2, TMP2, TMP1);
+  orr(ARMEmitter::Size::i32Bit, TMP2, TMP2, TMP4, ARMEmitter::ShiftType::LSL, 15);
+
+  // Set implicit bit and shift fraction to extF80 position
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP1, 0x00800000ULL);
+  orr(ARMEmitter::Size::i64Bit, TMP3, TMP3, TMP1);
+  lsl(ARMEmitter::Size::i64Bit, TMP3, TMP3, 40);
+
+  fmov(ARMEmitter::Size::i64Bit, VTMP1.D(), TMP3);
+  ins(ARMEmitter::SubRegSize::i16Bit, VTMP1, 4, TMP2);
+  (void)b(&Done);
+
+  (void)Bind(&ZeroDenormal);
+  (void)cbz(ARMEmitter::Size::i32Bit, TMP3, &ZeroResult);
+
+  (void)Bind(&Denormal);
+  clz(ARMEmitter::Size::i32Bit, TMP1, TMP3);
+  sub(ARMEmitter::Size::i32Bit, TMP1, TMP1, 8);
+  mov(ARMEmitter::Size::i32Bit, TMP2, 1);
+  sub(ARMEmitter::Size::i32Bit, TMP2, TMP2, TMP1);
+  lslv(ARMEmitter::Size::i32Bit, TMP3, TMP3, TMP1);
+  (void)b(&NormalPath);
+
+  (void)Bind(&ZeroResult);
+  lsl(ARMEmitter::Size::i32Bit, TMP2, TMP4, 15);
+  movi(ARMEmitter::SubRegSize::i64Bit, VTMP1.Q(), 0);
+  ins(ARMEmitter::SubRegSize::i16Bit, VTMP1, 4, TMP2);
+  (void)b(&Done);
+
+  (void)Bind(&InfNaN);
+  (void)cbnz(ARMEmitter::Size::i32Bit, TMP3, &NaN);
+
+  lsl(ARMEmitter::Size::i32Bit, TMP2, TMP4, 15);
+  orr(ARMEmitter::Size::i32Bit, TMP2, TMP2, 0x7FFF);
+
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP3, 0x8000000000000000ULL);
+  fmov(ARMEmitter::Size::i64Bit, VTMP1.D(), TMP3);
+  ins(ARMEmitter::SubRegSize::i16Bit, VTMP1, 4, TMP2);
+  (void)b(&Done);
+
+  (void)Bind(&NaN);
+  lsl(ARMEmitter::Size::i32Bit, TMP2, TMP4, 15);
+  orr(ARMEmitter::Size::i32Bit, TMP2, TMP2, 0x7FFF);
+
+  lsl(ARMEmitter::Size::i64Bit, TMP3, TMP3, 40);
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP1, 0xC000000000000000ULL);
+  orr(ARMEmitter::Size::i64Bit, TMP3, TMP3, TMP1);
+
+  fmov(ARMEmitter::Size::i64Bit, VTMP1.D(), TMP3);
+  ins(ARMEmitter::SubRegSize::i16Bit, VTMP1, 4, TMP2);
+
+  (void)Bind(&Done);
+}
+
+void Dispatcher::EmitF64ToExtF80() {
+  ARMEmitter::ForwardLabel InfNaN;
+  ARMEmitter::ForwardLabel ZeroDenormal;
+  ARMEmitter::ForwardLabel Denormal;
+  ARMEmitter::ForwardLabel NaN;
+  ARMEmitter::ForwardLabel Done;
+  ARMEmitter::BiDirectionalLabel NormalPath;
+  ARMEmitter::ForwardLabel ZeroResult;
+
+  fmov(ARMEmitter::Size::i64Bit, TMP1, VTMP1.D());
+
+  lsr(ARMEmitter::Size::i64Bit, TMP4, TMP1, 63);
+  ubfx(ARMEmitter::Size::i64Bit, TMP2, TMP1, 52, 11);
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP3, 0x000FFFFFFFFFFFFFULL);
+  and_(ARMEmitter::Size::i64Bit, TMP3, TMP1, TMP3);
+
+  cmp(ARMEmitter::Size::i64Bit, TMP2, 0x7FF);
+  (void)b(ARMEmitter::Condition::CC_EQ, &InfNaN);
+
+  (void)cbz(ARMEmitter::Size::i64Bit, TMP2, &ZeroDenormal);
+
+  (void)Bind(&NormalPath);
+  // Exponent bias adjustment where bias difference is 0x3C00
+  add(ARMEmitter::Size::i64Bit, TMP2, TMP2, 0x3000);
+  add(ARMEmitter::Size::i64Bit, TMP2, TMP2, 0xC00);
+  orr(ARMEmitter::Size::i64Bit, TMP2, TMP2, TMP4, ARMEmitter::ShiftType::LSL, 15);
+
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP1, 0x0010000000000000ULL);
+  orr(ARMEmitter::Size::i64Bit, TMP3, TMP3, TMP1);
+  lsl(ARMEmitter::Size::i64Bit, TMP3, TMP3, 11);
+
+  fmov(ARMEmitter::Size::i64Bit, VTMP1.D(), TMP3);
+  ins(ARMEmitter::SubRegSize::i16Bit, VTMP1, 4, TMP2);
+  (void)b(&Done);
+
+  (void)Bind(&ZeroDenormal);
+  (void)cbz(ARMEmitter::Size::i64Bit, TMP3, &ZeroResult);
+
+  (void)Bind(&Denormal);
+  clz(ARMEmitter::Size::i64Bit, TMP1, TMP3);
+  sub(ARMEmitter::Size::i64Bit, TMP1, TMP1, 11);
+  mov(ARMEmitter::Size::i64Bit, TMP2, 1);
+  sub(ARMEmitter::Size::i64Bit, TMP2, TMP2, TMP1);
+  lslv(ARMEmitter::Size::i64Bit, TMP3, TMP3, TMP1);
+  (void)b(&NormalPath);
+
+  (void)Bind(&ZeroResult);
+  lsl(ARMEmitter::Size::i64Bit, TMP2, TMP4, 15);
+  movi(ARMEmitter::SubRegSize::i64Bit, VTMP1.Q(), 0);
+  ins(ARMEmitter::SubRegSize::i16Bit, VTMP1, 4, TMP2);
+  (void)b(&Done);
+
+  (void)Bind(&InfNaN);
+  (void)cbnz(ARMEmitter::Size::i64Bit, TMP3, &NaN);
+
+  lsl(ARMEmitter::Size::i64Bit, TMP2, TMP4, 15);
+  orr(ARMEmitter::Size::i64Bit, TMP2, TMP2, 0x7FFF);
+
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP3, 0x8000000000000000ULL);
+  fmov(ARMEmitter::Size::i64Bit, VTMP1.D(), TMP3);
+  ins(ARMEmitter::SubRegSize::i16Bit, VTMP1, 4, TMP2);
+  (void)b(&Done);
+
+  (void)Bind(&NaN);
+  lsl(ARMEmitter::Size::i64Bit, TMP2, TMP4, 15);
+  orr(ARMEmitter::Size::i64Bit, TMP2, TMP2, 0x7FFF);
+
+  lsl(ARMEmitter::Size::i64Bit, TMP3, TMP3, 11);
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP1, 0xC000000000000000ULL);
+  orr(ARMEmitter::Size::i64Bit, TMP3, TMP3, TMP1);
+
+  fmov(ARMEmitter::Size::i64Bit, VTMP1.D(), TMP3);
+  ins(ARMEmitter::SubRegSize::i16Bit, VTMP1, 4, TMP2);
+
+  (void)Bind(&Done);
+}
+
 uint64_t Dispatcher::GenerateABICall(FallbackABI ABI) {
  auto Address = GetCursorAddress<uint64_t>();
  constexpr static auto FallbackPointerReg = TMP4;
@@ -689,65 +915,33 @@ uint64_t Dispatcher::GenerateABICall(FallbackABI ABI) {

  switch (ABI) {
  case FABI_F80_I16_F32_PTR: {
-    // Linux Reg/Win32 Reg:
-    // tmp4 (x4/x13): FallbackHandler
-    // x30: return
-    // vtmp1 (v0/v16): source
-    SpillForABICall(CTX->HostFeatures.SupportsPreserveAllABI, TMP3, true);
-
-    if (!TMP_ABIARGS) {
-      fmov(VABI1.S(), VTMP1.S());
-    }
-    ldrh(ARMEmitter::WReg::w0, STATE, offsetof(FEXCore::Core::CPUState, FCW));
-    mov(ARMEmitter::XReg::x1, STATE);
-    if (!CTX->Config.DisableVixlIndirectCalls) [[unlikely]] {
-      GenerateIndirectRuntimeCall<FEXCore::VectorRegType, uint16_t, float, uint64_t>(FallbackPointerReg);
-    } else {
-      blr(FallbackPointerReg);
-    }
-
-    FillF80Result();
+    // Save NZCV - it's a static register (guest x86 flags) and the inline code clobbers it
+    mrs(TMP1, ARMEmitter::SystemRegister::NZCV);
+    str(TMP1.W(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.flags[24]));
+    EmitF32ToExtF80();
+    ldr(TMP1.W(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.flags[24]));
+    msr(ARMEmitter::SystemRegister::NZCV, TMP1);
  } break;
  case FABI_F80_I16_F64_PTR: {
-    // Linux Reg/Win32 Reg:
-    // tmp4 (x4/x13): FallbackHandler
-    // x30: return
-    // vtmp1 (v0/v16): source
-    SpillForABICall(CTX->HostFeatures.SupportsPreserveAllABI, TMP3, true);
-
-    if (!TMP_ABIARGS) {
-      fmov(VABI1.D(), VTMP1.D());
-    }
-    ldrh(ARMEmitter::WReg::w0, STATE, offsetof(FEXCore::Core::CPUState, FCW));
-    mov(ARMEmitter::XReg::x1, STATE);
-    if (!CTX->Config.DisableVixlIndirectCalls) [[unlikely]] {
-      GenerateIndirectRuntimeCall<FEXCore::VectorRegType, uint16_t, double, uint64_t>(FallbackPointerReg);
-    } else {
-      blr(FallbackPointerReg);
-    }
-
-    FillF80Result();
+    mrs(TMP1, ARMEmitter::SystemRegister::NZCV);
+    str(TMP1.W(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.flags[24]));
+    EmitF64ToExtF80();
+    ldr(TMP1.W(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.flags[24]));
+    msr(ARMEmitter::SystemRegister::NZCV, TMP1);
+  } break;
+  case FABI_F80_I16_I16_PTR: {
+    mrs(TMP1, ARMEmitter::SystemRegister::NZCV);
+    str(TMP1.W(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.flags[24]));
+    EmitI16ToExtF80();
+    ldr(TMP1.W(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.flags[24]));
+    msr(ARMEmitter::SystemRegister::NZCV, TMP1);
  } break;
-  case FABI_F80_I16_I16_PTR:
  case FABI_F80_I16_I32_PTR: {
-    // Linux Reg/Win32 Reg:
-    // tmp4 (x4/x13): FallbackHandler
-    // x30: return
-    // tmp2 (x1/x11): source
-    SpillForABICall(CTX->HostFeatures.SupportsPreserveAllABI, TMP3, true);
-
-    if (!TMP_ABIARGS) {
-      mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, TMP2);
-    }
-    ldrh(ARMEmitter::WReg::w0, STATE, offsetof(FEXCore::Core::CPUState, FCW));
-    mov(ARMEmitter::XReg::x2, STATE);
-    if (!CTX->Config.DisableVixlIndirectCalls) [[unlikely]] {
-      GenerateIndirectRuntimeCall<FEXCore::VectorRegType, uint16_t, uint32_t, uint64_t>(FallbackPointerReg);
-    } else {
-      blr(FallbackPointerReg);
-    }
-
-    FillF80Result();
+    mrs(TMP1, ARMEmitter::SystemRegister::NZCV);
+    str(TMP1.W(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.flags[24]));
+    EmitI32ToExtF80();
+    ldr(TMP1.W(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.flags[24]));
+    msr(ARMEmitter::SystemRegister::NZCV, TMP1);
  } break;
  case FABI_F32_I16_F80_PTR: {
    // Linux Reg/Win32 Reg:
@@ -1079,27 +1273,25 @@ uint64_t Dispatcher::GenerateABICall(FallbackABI ABI) {
 void Dispatcher::InitThreadPointers(FEXCore::Core::InternalThreadState* Thread) {
  // Setup dispatcher specific pointers that need to be accessed from JIT code
  {
-    auto& Common = Thread->CurrentFrame->Pointers.Common;
+    auto& Ptrs = Thread->CurrentFrame->Pointers;

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

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

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

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

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

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

 protected:
@@ -92,6 +97,15 @@ private:

  void EmitDispatcher();
  uint64_t GenerateABICall(FallbackABI ABI);
+
+  // Inline softfloat conversion emitters - avoid FPCR save/restore overhead
+  // These emit ARM64 code that performs the conversion using only integer ops
+  void EmitI16ToExtF80();
+  void EmitI32ToExtF80();
+  void EmitF32ToExtF80();
+  void EmitF64ToExtF80();
+
+  FEX_CONFIG_OPT(DisableL2Cache, DISABLEL2CACHE);
 };

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

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

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

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

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

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

 void Decoder::DecodeModRM_16(X86Tables::DecodedOperand* Operand, X86Tables::ModRMDecoded ModRM) {
@@ -192,7 +200,9 @@ void Decoder::DecodeModRM_16(X86Tables::DecodedOperand* Operand, X86Tables::ModR
    DisplacementSize = 1;
  }
  if (DisplacementSize) {
-    Literal = ReadData(DisplacementSize);
+    bool IsRelocation = false;
+    std::tie(Literal, IsRelocation) = ReadData(DisplacementSize);
+    LOGMAN_THROW_A_FMT(!IsRelocation, "1/2 byte relocations unsupported");
    if (DisplacementSize == 1) {
      Literal = static_cast<int8_t>(Literal);
    }
@@ -298,7 +308,10 @@ void Decoder::DecodeModRM_64(X86Tables::DecodedOperand* Operand, X86Tables::ModR
    LOGMAN_THROW_A_FMT(Displacement <= 4, "Number of bytes should be <= 4 for literal src");

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

-    Operand->Type = DecodedOperand::OpType::GPRIndirect;
+    Operand->Type = IsRelocation ? DecodedOperand::OpType::GPRIndirectRelocation : DecodedOperand::OpType::GPRIndirect;
    Operand->Data.GPRIndirect.GPR = MapModRMToReg(DecodeInst->Flags & DecodeFlags::FLAG_REX_XGPR_B ? 1 : 0, ModRM.rm, false, false, false, false);
    Operand->Data.GPRIndirect.Displacement = Literal;
  }
@@ -620,31 +632,29 @@ bool Decoder::NormalOp(const FEXCore::X86Tables::X86InstInfo* Info, uint16_t Op,
  if (Bytes != 0) {
    LOGMAN_THROW_A_FMT(Bytes <= 8, "Number of bytes should be <= 8 for literal src");

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

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

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

-    DecodeInst->Src[CurrentSrc].Type = DecodedOperand::OpType::Literal;
-    DecodeInst->Src[CurrentSrc].Data.Literal.Value = Literal;
-    ++CurrentSrc;
-
-    if (Bytes == 8) [[unlikely]] {
-      DecodeInst->Src[CurrentSrc].Data.Literal.Size = 4;
      DecodeInst->Src[CurrentSrc].Type = DecodedOperand::OpType::Literal;
-      DecodeInst->Src[CurrentSrc].Data.Literal.Value = Literal >> 32;
+      DecodeInst->Src[CurrentSrc].Data.Literal.Value = Literal;
    }

    Bytes = 0;
@@ -759,13 +769,23 @@ bool Decoder::NormalOpHeader(const FEXCore::X86Tables::X86InstInfo* Info, uint16

    if (Op == 0xC5) { // Two byte VEX
      pp = Byte1 & 0b11;
-      options.vvvv = 15 - ((Byte1 & 0b01111000) >> 3);
+      const uint8_t vvvv = ((Byte1 & 0b01111000) >> 3);
+      if (!BlockInfo.Is64BitMode && vvvv <= 0b0111) {
+        // Invalid on 32-bit, can't use the high registers.
+        return false;
+      }
+      options.vvvv = 15 - vvvv;
      options.L = (Byte1 & 0b100) != 0;
    } else { // 0xC4 = Three byte VEX
      const uint8_t Byte2 = ReadByte();
      pp = Byte2 & 0b11;
      map_select = Byte1 & 0b11111;
-      options.vvvv = 15 - ((Byte2 & 0b01111000) >> 3);
+      const uint8_t vvvv = ((Byte2 & 0b01111000) >> 3);
+      if (!BlockInfo.Is64BitMode && vvvv <= 0b0111) {
+        // Invalid on 32-bit, can't use the high registers.
+        return false;
+      }
+      options.vvvv = 15 - vvvv;
      options.w = (Byte2 & 0b10000000) != 0;
      options.L = (Byte2 & 0b100) != 0;
      if ((Byte1 & 0b01000000) == 0) {
@@ -838,6 +858,7 @@ bool Decoder::DecodeInstructionImpl(uint64_t PC) {
      switch (EscapeOp) {
      case 0x0F:
        [[unlikely]] { // 3DNow!
+          DecodeREXIfValid(-2);
          // 3DNow! Instruction Encoding: 0F 0F [ModRM] [SIB] [Displacement] [Opcode]
          // Decode ModRM
          uint8_t ModRMByte = ReadByte();
@@ -862,6 +883,7 @@ bool Decoder::DecodeInstructionImpl(uint64_t PC) {
          break;
        }
      case 0x38: { // F38 Table!
+        DecodeREXIfValid(-2);
        constexpr uint16_t PF_38_NONE = 0;
        constexpr uint16_t PF_38_66 = (1U << 0);
        constexpr uint16_t PF_38_F2 = (1U << 1);
@@ -887,11 +909,11 @@ bool Decoder::DecodeInstructionImpl(uint64_t PC) {
          DecodeInst->Flags &= ~DecodeFlags::FLAG_OPERAND_SIZE;
          DecodeFlags::PopOpAddrIf(&DecodeInst->Flags, DecodeFlags::FLAG_OPERAND_SIZE_LAST);
        }
-
        return NormalOpHeader(&FEXCore::X86Tables::H0F38TableOps[LocalOp], LocalOp);
        break;
      }
      case 0x3A: { // F3A Table!
+        DecodeREXIfValid(-2);
        constexpr uint16_t PF_3A_NONE = 0;
        constexpr uint16_t PF_3A_66 = (1 << 0);
        constexpr uint16_t PF_3A_REX = (1 << 1);
@@ -921,6 +943,7 @@ bool Decoder::DecodeInstructionImpl(uint64_t PC) {
          bool NoOverlay = (FEXCore::X86Tables::SecondBaseOps[EscapeOp].Flags & InstFlags::FLAGS_NO_OVERLAY) != 0;
          bool NoOverlay66 = (FEXCore::X86Tables::SecondBaseOps[EscapeOp].Flags & InstFlags::FLAGS_NO_OVERLAY66) != 0;

+          DecodeREXIfValid(-2);
          if (NoOverlay) { // This section of the table ignores prefix extention
            return NormalOpHeader(&FEXCore::X86Tables::SecondBaseOps[EscapeOp], EscapeOp);
          } else if (LastEscapePrefix == 0xF3) { // REP
@@ -1000,29 +1023,9 @@ bool Decoder::DecodeInstructionImpl(uint64_t PC) {
        }

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

@@ -1038,17 +1041,53 @@ bool Decoder::DecodeInstructionImpl(uint64_t PC) {
  return true;
 }

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

-  if (ErrorDuringDecoding || HitNonExecutableRange) [[unlikely]] {
+  if (ErrorDuringDecoding || HitNonExecutableRange || HitBadRelocation) [[unlikely]] {
    // Put an invalid instruction in the stream so the core can raise SIGILL if hit
    // Error while decoding instruction. We don't know the table or instruction size
    DecodeInst->TableInfo = nullptr;
+    auto Result = ErrorDuringDecoding   ? DecodedBlockStatus::INVALID_INST :
+                  DecodeInst->InstSize  ? DecodedBlockStatus::PARTIAL_DECODE_INST :
+                  HitNonExecutableRange ? DecodedBlockStatus::NOEXEC_INST :
+                                          DecodedBlockStatus::BAD_RELOCATION;
    DecodeInst->InstSize = 0;
-    return ErrorDuringDecoding ? DecodedBlockStatus::INVALID_INST : DecodedBlockStatus::NOEXEC_INST;
+    return Result;
  } else if (!DecodeInst->TableInfo || (DecodeInst->TableInfo->Type == TYPE_INST && !DecodeInst->TableInfo->OpcodeDispatcher.OpDispatch)) {
    // If there wasn't an error during decoding but we have no dispatcher for the instruction then claim invalid instruction.
    return DecodedBlockStatus::INVALID_INST;
@@ -1132,9 +1171,9 @@ void Decoder::BranchTargetInMultiblockRange() {
  // Forbid distant branches to have the cost code better match the guest code layout, avoiding massive (range-wise) code
  // blocks in highly fragmented guest code. Such branches are often not-taken branches to garbage in obfuscated code.
  constexpr uint64_t MAX_FORWARD_BRANCH_DIST = FEXCore::Utils::FEX_PAGE_SIZE * 4;
-  bool ValidMultiblockMember = TargetRIP >= SymbolMinAddress && TargetRIP < std::min(InstEnd + MAX_FORWARD_BRANCH_DIST, SymbolMaxAddress);
+  bool ValidMultiblockMember = TargetRIP >= EntryPoint && TargetRIP < std::min(InstEnd + MAX_FORWARD_BRANCH_DIST, SectionMaxAddress);

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

@@ -1325,19 +1364,23 @@ void Decoder::DecodeInstructionsAtEntry(FEXCore::Core::InternalThreadState* Thre
  BlockInfo.Is64BitMode = CSSegment->L == 1;
  LOGMAN_THROW_A_FMT(BlockInfo.Is64BitMode == CTX->Config.Is64BitMode, "Expected operating mode to not change at runtime!");

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

  uint64_t TotalInstructions {};

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

  DecodedMinAddress = EntryPoint;
@@ -1450,7 +1493,11 @@ void Decoder::DecodeInstructionsAtEntry(FEXCore::Core::InternalThreadState* Thre
          EraseBlock = true;
        } else {
          LogMan::Msg::EFmt("{} instruction in entry block: {:X}",
-                            BlockIt->BlockStatus == DecodedBlockStatus::INVALID_INST ? "Invalid" : "NoExec", OpAddress);
+                            BlockIt->BlockStatus == DecodedBlockStatus::INVALID_INST   ? "Invalid" :
+                            BlockIt->BlockStatus == DecodedBlockStatus::NOEXEC_INST    ? "NoExec" :
+                            BlockIt->BlockStatus == DecodedBlockStatus::BAD_RELOCATION ? "BadRelocation" :
+                                                                                         "PartialDecode",
+                            OpAddress);
        }
        break;
      }
@@ -4,9 +4,12 @@
 #include "Interface/Core/X86Tables/X86Tables.h"
 #include "Interface/IR/IR.h"

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

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

  // New Frontend decoding
@@ -58,9 +63,6 @@ public:
  uint64_t DecodedMinAddress {};
  uint64_t DecodedMaxAddress {~0ULL};

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

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

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

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

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

  const uint8_t* InstStream {};
  IR::OpSize GetGPROpSize() const {
@@ -129,13 +137,11 @@ private:
  FEXCore::X86Tables::DecodedInst* DecodeInst;

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

  DecodedBlockInformation BlockInfo;
@@ -144,6 +150,8 @@ private:
  fextl::set<uint64_t> VisitedBlocks;
  fextl::set<uint64_t>* ExternalBranches {nullptr};

+  const fextl::robin_map<uint32_t, GuestRelocationType>* Relocations {nullptr};
+
  // ModRM rm decoding
  using DecodeModRMPtr = void (FEXCore::Frontend::Decoder::*)(X86Tables::DecodedOperand* Operand, X86Tables::ModRMDecoded ModRM);
  void DecodeModRM_16(X86Tables::DecodedOperand* Operand, X86Tables::ModRMDecoded ModRM);
@@ -435,12 +435,12 @@ struct OpHandlers<IR::OP_F80BCDSTORE> {
    FEXCORE_PROFILE_INSTANT_INCREMENT(Frame->Thread, AccumulatedFloatFallbackCount, 1);
    X80SoftFloat Src1 = Src1q;
    ScopedSoftFloatState State {FCW, Frame};
-    bool Negative = Src1.Sign;
+    bool Negative = Src1.Top.Sign;

    Src1 = X80SoftFloat::FRNDINT(&State.State, Src1);

    // Clear the Sign bit
-    Src1.Sign = 0;
+    Src1.Top.Sign = 0;

    uint64_t Tmp = Src1.ToI64(&State.State);
    X80SoftFloat Rv;
@@ -503,7 +503,7 @@ struct OpHandlers<IR::OP_F80BCDLOAD> {
    X80SoftFloat Tmp;

    Tmp = BCD;
-    Tmp.Sign = Negative;
+    Tmp.Top.Sign = Negative;
    return Tmp;
  }
 };
@@ -2,14 +2,14 @@
 #include "Interface/Core/Interpreter/Fallbacks/VectorFallbacks.h"
 #include "Interface/IR/IR.h"

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

 #include <cstring>

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

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

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

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

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

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

-  if (Op->Cond == FEXCore::IR::COND_AL) {
+  if (Op->Cond == IR::CondClass::AL) {
    neg(ConvertSize48(IROp), GetReg(Node), GetReg(Op->Src));
  } else {
    cneg(ConvertSize48(IROp), GetReg(Node), GetReg(Op->Src), MapCC(Op->Cond));
@@ -917,7 +924,7 @@ DEF_OP(Div) {
      mov(EmitSize, TMP2, Lower);
      mov(EmitSize, TMP3, Divisor);

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

      str<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, -16);
      blr(TMP4);
@@ -1000,7 +1007,7 @@ DEF_OP(UDiv) {
      mov(EmitSize, TMP2, Lower);
      mov(EmitSize, TMP3, Divisor);

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

      str<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, -16);
      blr(TMP4);
@@ -1190,6 +1197,19 @@ DEF_OP(Rev) {
  }
 }

+DEF_OP(Rbit) {
+  auto Op = IROp->C<IR::IROp_Rbit>();
+  const auto OpSize = IROp->Size;
+
+  LOGMAN_THROW_A_FMT(OpSize == IR::OpSize::i32Bit || OpSize == IR::OpSize::i64Bit, "Unsupported {} size: {}", __func__, OpSize);
+  const auto EmitSize = ConvertSize48(IROp);
+
+  const auto Dst = GetReg(Node);
+  const auto Src = GetReg(Op->Src);
+
+  rbit(EmitSize, Dst, Src);
+}
+
 DEF_OP(Bfi) {
  auto Op = IROp->C<IR::IROp_Bfi>();
  const auto EmitSize = ConvertSize(IROp);
@@ -1273,6 +1293,16 @@ DEF_OP(Sbfe) {
  sbfx(ConvertSize(IROp), Dst, Src, Op->lsb, Op->Width);
 }

+DEF_OP(MaskGenerateFromBitWidth) {
+  auto Op = IROp->C<IR::IROp_MaskGenerateFromBitWidth>();
+  auto BitWidth = GetReg(Op->BitWidth);
+
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP1, -1);
+  cmp(ARMEmitter::Size::i64Bit, BitWidth, 0);
+  lslv(ARMEmitter::Size::i64Bit, TMP2, TMP1, BitWidth);
+  csinv(ARMEmitter::Size::i64Bit, GetReg(Node), TMP1, TMP2, ARMEmitter::Condition::CC_EQ);
+}
+
 DEF_OP(Select) {
  auto Op = IROp->C<IR::IROp_Select>();
  const auto OpSize = IROp->Size;
@@ -11,36 +11,33 @@ $end_info$
 #include <FEXCore/Core/Thunks.h>

 namespace FEXCore::CPU {
-
-uint64_t Arm64JITCore::GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
+uint64_t GetNamedSymbolLiteral(FEXCore::Context::ContextImpl& CTX, FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
  switch (Op) {
  case FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol::SYMBOL_LITERAL_EXITFUNCTION_LINKER:
-    return ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker;
-    break;
-  default: ERROR_AND_DIE_FMT("Unknown named symbol literal: {}", static_cast<uint32_t>(Op)); break;
+    return CTX.Dispatcher->GetExitFunctionLinkerAddress();
+
+  default: ERROR_AND_DIE_FMT("Unknown named symbol literal: {}", static_cast<uint32_t>(Op));
  }
-  return ~0ULL;
 }

 void Arm64JITCore::InsertNamedThunkRelocation(ARMEmitter::Register Reg, const IR::SHA256Sum& Sum) {
  Relocation MoveABI {};
-  MoveABI.NamedThunkMove.Header.Type = FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE;
-  // Offset is the offset from the entrypoint of the block
-  auto CurrentCursor = GetCursorAddress<uint8_t*>();
-  MoveABI.NamedThunkMove.Offset = CurrentCursor - CodeData.BlockBegin;
+  MoveABI.NamedThunkMove.Header = {.Offset = GetCursorOffset(), .Type = FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE};
  MoveABI.NamedThunkMove.Symbol = Sum;
  MoveABI.NamedThunkMove.RegisterIndex = Reg.Idx();

  uint64_t Pointer = reinterpret_cast<uint64_t>(EmitterCTX->ThunkHandler->LookupThunk(Sum));

-  LoadConstant(ARMEmitter::Size::i64Bit, Reg, Pointer, false);
+  // Pointers are required to fit within 48-bit VA space.
+  // TODO: Force 6-byte `MaxSize`, with zext extension to 64-bit. Current code not smart enough to handle negatives.
+  LoadConstant(ARMEmitter::Size::i64Bit, Reg, Pointer, FEXCore::CPU::Arm64Emitter::PadType::AUTOPAD);
  Relocations.emplace_back(MoveABI);
 }

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

-  Arm64JITCore::NamedSymbolLiteralPair Lit {
+  NamedSymbolLiteralPair Lit {
    .Lit = Pointer,
    .MoveABI =
      {
@@ -48,92 +45,76 @@ Arm64JITCore::NamedSymbolLiteralPair Arm64JITCore::InsertNamedSymbolLiteral(FEXC
          {
            .Header =
              {
+                .Offset = 0, // Set by PlaceNamedSymbolLiteral
                .Type = FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL,
              },
            .Symbol = Op,
-            .Offset = 0,
          },
      },
  };
  return Lit;
 }

-void Arm64JITCore::PlaceNamedSymbolLiteral(NamedSymbolLiteralPair& Lit) {
-  // Offset is the offset from the entrypoint of the block
-  auto CurrentCursor = GetCursorAddress<uint8_t*>();
-  Lit.MoveABI.NamedSymbolLiteral.Offset = CurrentCursor - CodeData.BlockBegin;
+void Arm64JITCore::PlaceNamedSymbolLiteral(NamedSymbolLiteralPair Lit) {
+  switch (Lit.MoveABI.Header.Type) {
+  case RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL:
+  case RelocationTypes::RELOC_GUEST_RIP_LITERAL: {
+    Lit.MoveABI.Header.Offset = GetCursorOffset();
+    break;
+  }
+
+  default: ERROR_AND_DIE_FMT("Unknown relocation type for {}", __FUNCTION__);
+  }

  BindOrRestart(&Lit.Loc);
  dc64(Lit.Lit);
  Relocations.emplace_back(Lit.MoveABI);
 }

+auto Arm64JITCore::InsertGuestRIPLiteral(uint64_t GuestRIP) -> NamedSymbolLiteralPair {
+  return {
+    .Lit = GuestRIP,
+    .MoveABI =
+      {
+        .GuestRIP = {.Header =
+                       {
+                         .Offset = 0, // Set by PlaceNamedSymbolLiteral
+                         .Type = FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_LITERAL,
+                       },
+                     // NOTE: Cache serialization will subtract the guest binary base address later to produce consistency results
+                     .GuestRIP = GuestRIP},
+      },
+  };
+}
+
 void Arm64JITCore::InsertGuestRIPMove(ARMEmitter::Register Reg, uint64_t Constant) {
  Relocation MoveABI {};
-  MoveABI.GuestRIPMove.Header.Type = FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE;
-  // Offset is the offset from the entrypoint of the block
-  auto CurrentCursor = GetCursorAddress<uint8_t*>();
-  MoveABI.GuestRIPMove.Offset = CurrentCursor - CodeData.BlockBegin;
-  MoveABI.GuestRIPMove.GuestRIP = Constant;
-  MoveABI.GuestRIPMove.RegisterIndex = Reg.Idx();
+  MoveABI.GuestRIP.Header = {.Offset = GetCursorOffset(), .Type = FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE};
+  // NOTE: Cache serialization will subtract the guest binary base address later to produce consistency results
+  MoveABI.GuestRIP.GuestRIP = Constant;
+  MoveABI.GuestRIP.RegisterIndex = Reg.Idx();

-  LoadConstant(ARMEmitter::Size::i64Bit, Reg, Constant, false);
+  // Pointers are required to fit within 48-bit VA space.
+  // TODO: Force 6-byte `MaxSize`, with sign extension to 64-bit. Current code not smart enough to handle negatives.
+  // 48-bit sign extension works because x86-64 guests only receive 47-bit VA space, with 48-bit being reserved for kernel.
+  // Additional quirk, "canonical" 48-bit pointers on x86-64, sign extend the 48-bit as well (Which is why kernel pointers are negative).
+  LoadConstant(ARMEmitter::Size::i64Bit, Reg, Constant, FEXCore::CPU::Arm64Emitter::PadType::AUTOPAD);
  Relocations.emplace_back(MoveABI);
 }

-bool Arm64JITCore::ApplyRelocations(uint64_t GuestEntry, std::span<std::byte> Code, std::span<const FEXCore::CPU::Relocation> Relocations) {
-  const auto OrigBase = GetBufferBase();
-  const auto OrigSize = GetBufferSize();
-  const auto OrigOffset = GetCursorOffset();
-
-  SetBuffer(reinterpret_cast<std::uint8_t*>(Code.data()), Code.size_bytes());
-  for (auto& Reloc : Relocations) {
-    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(Reloc.NamedSymbolLiteral.Offset);
-
-      // Generate a literal so we can place it
-      dc64(Pointer);
-      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(Reloc.NamedThunkMove.Offset);
-      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Register(Reloc.NamedThunkMove.RegisterIndex), Pointer, true);
-      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) {
-        SetBuffer(OrigBase, OrigSize);
-        SetCursorOffset(OrigOffset);
-        return false;
-      }
-
-      // Relocation occurs at the cursorEntry + offset relative to that cursor.
-      SetCursorOffset(Reloc.GuestRIPMove.Offset);
-      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Register(Reloc.GuestRIPMove.RegisterIndex), Pointer, true);
+fextl::vector<FEXCore::CPU::Relocation> Arm64JITCore::TakeRelocations(uint64_t GuestBaseAddress) {
+  // Rebase relocations to library base address
+  for (auto& Relocation : Relocations) {
+    switch (Relocation.Header.Type) {
+    case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE:
+    case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_LITERAL: {
+      Relocation.GuestRIP.GuestRIP -= GuestBaseAddress;
      break;
    }
+    default:;
    }
  }

-  SetBuffer(OrigBase, OrigSize);
-  SetCursorOffset(OrigOffset);
-
-  return true;
-}
-
-fextl::vector<FEXCore::CPU::Relocation> Arm64JITCore::TakeRelocations() {
  return std::move(Relocations);
 }

@@ -138,26 +138,6 @@ DEF_OP(CAS) {
  }
 }

-DEF_OP(AtomicXor) {
-  auto Op = IROp->C<IR::IROp_AtomicXor>();
-  const auto EmitSize = ConvertSize(IROp);
-  const auto SubEmitSize = ConvertSubRegSize8(IROp->Size);
-
-  auto MemSrc = GetReg(Op->Addr);
-  auto Src = GetReg(Op->Value);
-
-  if (CTX->HostFeatures.SupportsAtomics) {
-    steorl(SubEmitSize, Src, MemSrc);
-  } else {
-    ARMEmitter::BackwardLabel LoopTop;
-    (void)Bind(&LoopTop);
-    ldaxr(SubEmitSize, TMP2, MemSrc);
-    eor(EmitSize, TMP2, TMP2, Src);
-    stlxr(SubEmitSize, TMP2, TMP2, MemSrc);
-    (void)cbnz(EmitSize, TMP2, &LoopTop);
-  }
-}
-
 DEF_OP(AtomicSwap) {
  auto Op = IROp->C<IR::IROp_AtomicSwap>();
  const auto OpSize = IROp->Size;
@@ -349,7 +329,7 @@ DEF_OP(TelemetrySetValue) {
  auto Op = IROp->C<IR::IROp_TelemetrySetValue>();
  auto Src = GetReg(Op->Value);

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

  // Cortex fuses cmp+cset.
  cmp(ARMEmitter::Size::i32Bit, Src, 0);
@@ -56,12 +56,12 @@ DEF_OP(ExitFunction) {
  uint64_t NewRIP;

  if (IsInlineConstant(Op->NewRIP, &NewRIP) || IsInlineEntrypointOffset(Op->NewRIP, &NewRIP)) {
-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
    if (NewRIP < EC_CODE_BITMAP_MAX_ADDRESS && RtlIsEcCode(NewRIP)) {
      str(REG_CALLRET_SP, STATE_PTR(CpuStateFrame, State.callret_sp));
      add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, StaticRegisters[X86State::REG_RSP], 0);
-      LoadConstant(ARMEmitter::Size::i64Bit, EC_CALL_CHECKER_PC_REG, NewRIP);
-      ldr(TMP2, STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionEC));
+      InsertGuestRIPMove(EC_CALL_CHECKER_PC_REG, NewRIP);
+      ldr(TMP2, STATE_PTR(CpuStateFrame, Pointers.ExitFunctionEC));
      br(TMP2);
    } else {
 #endif
@@ -150,16 +150,16 @@ DEF_OP(ExitFunction) {
        ARMEmitter::ForwardLabel TFUnset;
        ldrb(TMP1, STATE_PTR(CpuStateFrame, State.flags[X86State::RFLAG_TF_RAW_LOC]));
        (void)cbz(ARMEmitter::Size::i32Bit, TMP1, &TFUnset);
-        LoadConstant(ARMEmitter::Size::i64Bit, TMP1, NewRIP);
+        InsertGuestRIPMove(TMP1, NewRIP);
        str(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip));
-        ldr(TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.DispatcherLoopTop));
+        ldr(TMP2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.DispatcherLoopTop));
        blr(TMP2);
        (void)Bind(&TFUnset);
      }

      EmitLinkedBranch(NewRIP, Op->Hint == IR::BranchHint::Call);
      (void)Bind(&l_CallReturn);
-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
    }
 #endif
  } else {
@@ -174,21 +174,19 @@ DEF_OP(ExitFunction) {
    }

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

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

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

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

    (void)Bind(&SkipFullLookup);
@@ -235,16 +233,16 @@ DEF_OP(CondJump) {
    LOGMAN_THROW_A_FMT(IsGPR(Op->Cmp1), "CondJump: Expected GPR");
    LOGMAN_THROW_A_FMT(isConst, "CondJump: Expected constant source");

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

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

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

  SpillStaticRegs(TMP1, true, GPRSpillMask, FPRSpillMask);

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

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

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

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

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

-    // Now the registers we've spilled are back in their original host registers
-    // We can safely claim we are no longer in a syscall
-    str(ARMEmitter::XReg::zr, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));
+  // Now the registers we've spilled are back in their original host registers
+  // We can safely claim we are no longer in a syscall
+  str(ARMEmitter::XReg::zr, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));

-    PopDynamicRegs();
+  PopDynamicRegs();

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

-DEF_OP(InlineSyscall) {
-  auto Op = IROp->C<IR::IROp_InlineSyscall>();
-  // Arguments are passed as follows:
-  // X8: SyscallNumber - RA INTERSECT
-  // X0: Arg0 & Return
-  // X1: Arg1
-  // X2: Arg2
-  // X3: Arg3
-  // X4: Arg4 - RA INTERSECT
-  // X5: Arg5 - RA INTERSECT
-  // X6: Arg6 - Doesn't exist in x86-64 land. RA INTERSECT
-
-  // One argument is removed from the SyscallArguments::MAX_ARGS since the first argument was syscall number
-  const static std::array<ARMEmitter::XRegister, FEXCore::HLE::SyscallArguments::MAX_ARGS - 1> RegArgs = {
-    {ARMEmitter::XReg::x0, ARMEmitter::XReg::x1, ARMEmitter::XReg::x2, ARMEmitter::XReg::x3, ARMEmitter::XReg::x4, ARMEmitter::XReg::x5}};
-
-  bool Intersects {};
-  // We always need to spill x8 since we can't know if it is live at this SSA location
-  uint32_t SpillMask = 1U << 8;
-  for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS - 1; ++i) {
-    if (Op->Header.Args[i].IsInvalid()) {
-      break;
-    }
-
-    auto Reg = GetReg(Op->Header.Args[i]);
-    if (Reg == ARMEmitter::Reg::r8 || Reg == ARMEmitter::Reg::r4 || Reg == ARMEmitter::Reg::r5) {
-
-      SpillMask |= (1U << Reg.Idx());
-      Intersects = true;
-    }
-  }
-
-  // Ordering is incredibly important here
-  // We must spill any overlapping registers first THEN claim we are in a syscall without invalidating state at all
-  // Only spill the registers that intersect with our usage
-  SpillStaticRegs(TMP1, false, SpillMask);
-
-  // Now that we are spilled, store in the state that we are in a syscall
-  // Still without overwriting registers that matter
-  // 16bit LoadConstant to be a single instruction
-  // We must always spill at least one register (x8) so this value always has a bit set
-  // This gives the signal handler a value to check to see if we are in a syscall at all
-  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, SpillMask & 0xFFFF);
-  str(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));
-
-  // Now that we have claimed to be a syscall we can set up the arguments
-  const auto EmitSize = CTX->Config.Is64BitMode() ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
-  const auto EmitSubSize = CTX->Config.Is64BitMode() ? ARMEmitter::SubRegSize::i64Bit : ARMEmitter::SubRegSize::i32Bit;
-  if (Intersects) {
-    for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS - 1; ++i) {
-      if (Op->Header.Args[i].IsInvalid()) {
-        break;
-      }
-
-      auto Reg = GetReg(Op->Header.Args[i]);
-      if (SpillMask & (1U << Reg.Idx())) {
-        // In the case of intersection with x4, x5, or x8 then these are currently SRA
-        // for registers RAX, RDX, and RSP. Which have just been spilled
-        // Just load back from the context.
-        auto Correlation = GetX86RegRelationToARMReg(Reg);
-        LOGMAN_THROW_A_FMT(Correlation != X86State::REG_INVALID, "Invalid register mapping");
-        ldr(EmitSubSize, RegArgs[i].R(), STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[Correlation]));
-      } else {
-        mov(EmitSize, RegArgs[i].R(), Reg);
-      }
-    }
-  } else {
-    for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS - 1; ++i) {
-      if (Op->Header.Args[i].IsInvalid()) {
-        break;
-      }
-
-      mov(EmitSize, RegArgs[i].R(), GetReg(Op->Header.Args[i]));
-    }
-  }
-
-  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r8, Op->HostSyscallNumber);
-  svc(0);
-  // On updated signal mask we can receive a signal RIGHT HERE
-
-  if ((Op->Flags & FEXCore::IR::SyscallFlags::NORETURN) != FEXCore::IR::SyscallFlags::NORETURN) {
-    // Now that we are done in the syscall we need to carefully peel back the state
-    // First unspill the registers from before
-    FillStaticRegs(false, SpillMask, ~0U, ARMEmitter::Reg::r8, ARMEmitter::Reg::r1);
-
-    // Now the registers we've spilled are back in their original host registers
-    // We can safely claim we are no longer in a syscall
-    str(ARMEmitter::XReg::zr, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));
-
-    // Result is now in x0
-    // Move result to its destination register
-    mov(EmitSize, GetReg(Node), ARMEmitter::Reg::r0);
+  if (OSABI != FEXCore::HLE::SyscallOSABI::OS_GENERIC) {
+    // Move result to its destination register.
+    // Only if `NORETURNEDRESULT` wasn't set, otherwise we might overwrite the CPUState refilled with `FillStaticRegs`
+    mov(ARMEmitter::Size::i64Bit, GetReg(Node), ARMEmitter::Reg::r0);
  }
 }

@@ -425,14 +322,13 @@ DEF_OP(Thunk) {
  // X0: CTX
  // X1: Args (from guest stack)

-  SpillStaticRegs(TMP1); // spill to ctx before ra64 spill
+  SpillStaticRegs(TMP1, true, ~0U, ~0U, false); // spill to ctx before ra64 spill

  PushDynamicRegs(TMP1);

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

-  auto thunkFn = static_cast<Context::ContextImpl*>(ThreadState->CTX)->ThunkHandler->LookupThunk(Op->ThunkNameHash);
-  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, (uintptr_t)thunkFn);
+  InsertNamedThunkRelocation(ARMEmitter::Reg::r2, Op->ThunkNameHash);
  if (!CTX->Config.DisableVixlIndirectCalls) [[unlikely]] {
    GenerateIndirectRuntimeCall<void, void*, void*>(ARMEmitter::Reg::r2);
  } else {
@@ -441,7 +337,7 @@ DEF_OP(Thunk) {

  PopDynamicRegs();

-  FillStaticRegs(); // load from ctx after ra64 refill
+  FillStaticRegs(true, ~0U, ~0U, std::nullopt, std::nullopt, false); // load from ctx after ra64 refill
 }

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

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

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

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

  // x0 = CPUID Handler
  // x1 = XCR Function
-  ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDObj));
-  ldr(ARMEmitter::XReg::x2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.XCRFunction));
+  ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.CPUIDObj));
+  ldr(ARMEmitter::XReg::x2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.XCRFunction));
  if (!CTX->Config.DisableVixlIndirectCalls) [[unlikely]] {
    GenerateIndirectRuntimeCall<uint64_t, void*, uint32_t>(ARMEmitter::Reg::r2);
  } else {
@@ -423,11 +423,11 @@ DEF_OP(Vector_FToI) {
    const auto Mask = PRED_TMP_32B.Merging();

    switch (Op->Round) {
-    case FEXCore::IR::Round_Nearest.Val: frintn(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
-    case FEXCore::IR::Round_Negative_Infinity.Val: frintm(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
-    case FEXCore::IR::Round_Positive_Infinity.Val: frintp(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
-    case FEXCore::IR::Round_Towards_Zero.Val: frintz(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
-    case FEXCore::IR::Round_Host.Val: frinti(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::Nearest: frintn(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::NegInfinity: frintm(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::PosInfinity: frintp(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::TowardsZero: frintz(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
+    case IR::RoundMode::Host: frinti(SubEmitSize, Dst.Z(), Mask, Vector.Z()); break;
    }
  } else {
    const auto IsScalar = ElementSize == OpSize;
@@ -449,21 +449,21 @@ DEF_OP(Vector_FToI) {
  }

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

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

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

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

    // Now narrow from f64 to f32.
@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: MIT
 /*
 $info$
-glossary: Splatter ~ a code generator backend that concaternates configurable macros instead of doing isel
+glossary: Splatter ~ a code generator backend that concatenates configurable macros instead of doing isel
 glossary: IR ~ Intermediate Representation, our high-level opcode representation, loosely modeling arm64
 glossary: SSA ~ Single Static Assignment, a form of representing IR in memory
 glossary: Basic Block ~ A block of instructions with no control flow, terminated by control flow
@@ -133,8 +133,8 @@ void Arm64JITCore::Op_Unhandled(const IR::IROp_Header* IROp, IR::Ref Node) {
      const auto Src1 = GetVReg(IROp->Args[0]);
      fmov(VTMP1.S(), Src1.S());

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(JumpThunkStartAddress)).store(BranchInst, std::memory_order::relaxed);
  ARMEmitter::Emitter::ClearICache(reinterpret_cast<void*>(JumpThunkStartAddress), 4);
@@ -532,13 +533,13 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX
  if (TFSet) {
    // If TF is set, the cache must be skipped as different code needs to be generated.
    Frame->State.rip = GuestRip;
-    return Frame->Pointers.Common.DispatcherLoopTop;
+    return Frame->Pointers.DispatcherLoopTop;
  } else {
    {
      // Guard the LookupCache lock with the code invalidation mutex, to avoid issues with forking
      auto lk_inval =
        GuardSignalDeferringSection<std::shared_lock>(static_cast<Context::ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);
-      HostCode = Thread->LookupCache->FindBlock(GuestRip);
+      HostCode = Thread->LookupCache->FindBlock(Thread, GuestRip);
    }
    if (!HostCode) {
      // Hold a reference to the code buffer, to avoid linking unmapped code if compilation triggers a recreation.
@@ -563,7 +564,7 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX
  auto lk_inval = GuardSignalDeferringSection<std::shared_lock>(static_cast<Context::ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);

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

  // For non-calls, this would extend into the block's code, however that's fine as an out-of-range adr would never
  // be generated avoiding any false positives.
@@ -576,14 +577,12 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX

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

    std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(CallerAddress)).store(BranchInst, std::memory_order::relaxed);
@@ -591,7 +590,7 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX
  } else {
    // This case is common between calls and jumps as the thunk callsite can be left untouched.
    std::atomic_ref<uint64_t>(Record->HostCode).store(HostCode, std::memory_order::seq_cst);
-#ifdef _M_ARM_64
+#ifdef ARCHITECTURE_arm64
    // Make memory write visible to other threads reading the same location
    asm volatile("dc cvau, %0; dsb ish" : : "r"(Record->HostCode) :);
 #endif
@@ -602,7 +601,7 @@ uint64_t Arm64JITCore::ExitFunctionLink(FEXCore::Core::CpuStateFrame* Frame, FEX
    std::atomic_ref<uint32_t>(*reinterpret_cast<uint32_t*>(JumpThunkStartAddress)).store(LdrInst, std::memory_order::relaxed);
    ARMEmitter::Emitter::ClearICache(reinterpret_cast<void*>(JumpThunkStartAddress), 4);

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

  return HostCode;
@@ -623,46 +622,42 @@ Arm64JITCore::Arm64JITCore(FEXCore::Context::ContextImpl* ctx, FEXCore::Core::In

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

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

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

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

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

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

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

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

  CurrentCodeBuffer = CodeBuffers.GetLatest();
@@ -678,13 +673,13 @@ void Arm64JITCore::EmitDetectionString() {
 void Arm64JITCore::ClearCache() {
  // NOTE: Holding on to the reference here is required to ensure validity of the WriteLock mutex
  auto PrevCodeBuffer = CurrentCodeBuffer;
-  std::lock_guard lk(PrevCodeBuffer->LookupCache->WriteLock);
+  auto lk = PrevCodeBuffer->LookupCache->AcquireWriteLock();

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

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

 Arm64JITCore::~Arm64JITCore() {}
@@ -761,7 +756,7 @@ void Arm64JITCore::EmitTFCheck() {

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

  (void)Bind(&l_TFBlocked);
@@ -779,12 +774,12 @@ void Arm64JITCore::EmitSuspendInterruptCheck() {
         offsetof(FEXCore::Core::InternalThreadState, InterruptFaultPage) - offsetof(FEXCore::Core::InternalThreadState, BaseFrameState));
  }

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

  ldr(TMP2.W(), STATE_PTR(CpuStateFrame, SuspendDoorbell));
  ARMEmitter::ForwardLabel l_NoSuspend;
-  cbz(ARMEmitter::Size::i32Bit, TMP2, &l_NoSuspend);
+  (void)cbz(ARMEmitter::Size::i32Bit, TMP2, &l_NoSuspend);
  brk(SuspendMagic);
  (void)Bind(&l_NoSuspend);
 #endif
@@ -810,22 +805,35 @@ void Arm64JITCore::EmitEntryPoint(ARMEmitter::BackwardLabel& HeaderLabel, bool C
      sub(ARMEmitter::Size::i64Bit, ARMEmitter::XReg::rsp, ARMEmitter::XReg::rsp, TMP1, ARMEmitter::ExtendedType::LSL_64, 0);
    }
  }
+
+  EmitSuspendInterruptCheck();
 }

 CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size, bool SingleInst, const FEXCore::IR::IRListView* IR,
                                                   FEXCore::Core::DebugData* DebugData, bool CheckTF) {
  FEXCORE_PROFILE_SCOPED("Arm64::CompileCode");
+
+  const auto PrevNumAllocations = Relocations.size();
+
  this->Entry = Entry;
  this->DebugData = DebugData;
  this->IR = IR;
  RequiresFarARM64Jumps = false;
+  SSANodeMultiplier = 24;

-  switch (static_cast<RestartOptions::Control>(FEXCore::LongJump::SetJump(RestartControl.RestartJump))) {
+  // Prepare restart via long jump in case branch encoding fails.
+  // This uses UncheckedLongJump since we don't implement std::longjmp in WoA setups
+  switch (static_cast<RestartOptions::Control>(FEXCore::UncheckedLongJump::SetJump(ThreadState->RestartJump))) {
  case RestartOptions::Control::Incoming:
    // Nothing
    break;
  case RestartOptions::Control::EnableFarARM64Jumps: RequiresFarARM64Jumps = true; break;
-  default: ERROR_AND_DIE_FMT("Unhandled Arm64 restart condition!");
+  case RestartOptions::Control::NeedsLargerJITSpace:
+    // Get rid of the claimed buffer immediately, we can't fit in it at all.
+    TempAllocator.UnclaimBuffer();
+    SSANodeMultiplier *= 2;
+    break;
+  default: LOGMAN_MSG_A_FMT("Unhandled Arm64 restart condition!");
  }

  uint32_t SSACount = IR->GetSSACount();
@@ -837,12 +845,19 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
  CodeData.EntryPoints.clear();

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

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

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

@@ -932,8 +947,6 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size

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

@@ -974,22 +987,28 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size

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

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

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

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

  // Entries that live after the JITCodeTail.
  // These entries correlate JIT code regions with guest RIP regions.
@@ -1007,23 +1026,13 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
  //   FEXCore::Utils::vl64 GuestRIPOffset;
  // };

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

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

@@ -1039,14 +1048,20 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
    }
  }

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

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

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

  // Migrate the compile output from temporary storage to the actual CodeBuffer.
  // This can block progress in other compiling threads, so the duration of the lock should be as small as possible.
@@ -1055,7 +1070,6 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size

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

    // Bring CodeBuffer up to date
    {
@@ -1063,18 +1077,18 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
                                                                                                   "doesn't match up!\n");
      if (auto Prev = CheckCodeBufferUpdate()) {
        Allocator::VirtualDontNeed(ThreadState->CallRetStackBase, FEXCore::Core::InternalThreadState::CALLRET_STACK_SIZE);
-        ThreadState->LookupCache->ChangeGuestToHostMapping(*Prev, *CurrentCodeBuffer->LookupCache);
+        auto lk = ThreadState->LookupCache->AcquireWriteLock();
+        ThreadState->LookupCache->ChangeGuestToHostMapping(*Prev, *CurrentCodeBuffer->LookupCache, lk);
      }

      // NOTE: 16-byte alignment of the new cursor offset must be preserved for block linking records
-      SetBuffer(CurrentCodeBuffer->Ptr, CurrentCodeBuffer->Size);
-      SetCursorOffset(AlignUp(CodeBuffers.LatestOffset, 16));
-      if ((GetCursorOffset() + TempSize) > (CurrentCodeBuffer->Size - Utils::FEX_PAGE_SIZE)) {
+      SetBuffer(CurrentCodeBuffer->Ptr, CurrentCodeBuffer->AllocatedSize);
+      SetCursorOffset(CodeBuffers.LatestOffset);
+      Align16B();
+      if ((GetCursorOffset() + TempSize) > CurrentCodeBuffer->UsableSize()) {
        CTX->ClearCodeCache(ThreadState);
      }

-      Align16B();
-
      CodeBuffers.LatestOffset = GetCursorOffset();
    }

@@ -1086,6 +1100,10 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
    }
    CodeBegin += Delta;

+    for (std::size_t Idx = PrevNumAllocations; Idx != Relocations.size(); ++Idx) {
+      Relocations[Idx].Header.Offset += CodeBuffers.LatestOffset;
+    }
+
    // Copy over CodeBuffer contents
    memcpy(GetCursorAddress<uint8_t*>(), TempCodeBuffer, TempSize);
    SetCursorOffset(CodeBuffers.LatestOffset + TempSize);
@@ -10,13 +10,17 @@ $end_info$
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
 #include "Interface/Core/CPUBackend.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
+#include "Interface/Core/JIT/Relocations.h"
 #include "Interface/IR/IR.h"
 #include "Interface/IR/IntrusiveIRList.h"
 #include "Interface/IR/RegisterAllocationData.h"

+#include <FEXCore/Config/Config.h>
 #include <FEXCore/Core/CoreState.h>
 #include <FEXCore/IR/IR.h>
+#include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/fextl/map.h>
+#include <FEXCore/fextl/memory.h>
 #include <FEXCore/fextl/string.h>
 #include <FEXCore/fextl/vector.h>
 #include <FEXCore/Utils/LongJump.h>
@@ -26,16 +30,19 @@ $end_info$
 #include <array>
 #include <cstdint>
 #include <functional>
+#include <optional>
 #include <utility>
 #include <variant>

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

 namespace FEXCore::CPU {
 class Arm64JITCore final : public CPUBackend, public Arm64Emitter {
@@ -54,9 +61,6 @@ public:
  }

 private:
-  FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
-  FEX_CONFIG_OPT(HalfBarrierTSOEnabled, HALFBARRIERTSOENABLED);
-
  const bool HostSupportsSVE128 {};
  const bool HostSupportsSVE256 {};
  const bool HostSupportsAVX256 {};
@@ -64,10 +68,10 @@ private:
  const bool HostSupportsAFP {};

  struct RestartOptions {
-    FEXCore::LongJump::JumpBuf RestartJump;
    enum class Control : uint64_t {
      Incoming = 0,
      EnableFarARM64Jumps = 1,
+      NeedsLargerJITSpace = 2,
    };
  };

@@ -75,6 +79,8 @@ private:
  // In the rare case when those assumptions are broken, FEX needs to safely restart the JIT.
  RestartOptions RestartControl {};
  bool RequiresFarARM64Jumps {};
+  // Default to 6 instructions per SSA node.
+  uint32_t SSANodeMultiplier {24};

  ARMEmitter::BiDirectionalLabel* PendingTargetLabel {};
  ARMEmitter::BiDirectionalLabel* PendingCallReturnTargetLabel {};
@@ -105,11 +111,13 @@ private:

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

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

@@ -128,11 +136,13 @@ private:

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

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

@@ -150,8 +160,8 @@ private:
  }

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

  [[nodiscard]]
@@ -168,7 +178,7 @@ private:
  // Converts IR-base shift type to ARMEmitter shift type.
  // Will be a no-op, only a type conversion since the two definitions match.
  [[nodiscard]]
-  ARMEmitter::ShiftType ConvertIRShiftType(IR::ShiftType Shift) const {
+  static ARMEmitter::ShiftType ConvertIRShiftType(IR::ShiftType Shift) {
    return Shift == IR::ShiftType::LSL ? ARMEmitter::ShiftType::LSL :
           Shift == IR::ShiftType::LSR ? ARMEmitter::ShiftType::LSR :
           Shift == IR::ShiftType::ASR ? ARMEmitter::ShiftType::ASR :
@@ -176,18 +186,23 @@ private:
  }

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

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

  [[nodiscard]]
-  ARMEmitter::SubRegSize ConvertSubRegSize16(IR::OpSize ElementSize) {
+  static ARMEmitter::Size ConvertSize(IR::OpSize Size) {
+    return Size == IR::OpSize::i64Bit ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
+  }
+
+  [[nodiscard]]
+  static ARMEmitter::SubRegSize ConvertSubRegSize16(IR::OpSize ElementSize) {
    LOGMAN_THROW_A_FMT(ElementSize == IR::OpSize::i8Bit || ElementSize == IR::OpSize::i16Bit || ElementSize == IR::OpSize::i32Bit ||
                         ElementSize == IR::OpSize::i64Bit || ElementSize == IR::OpSize::i128Bit,
                       "Invalid size");
@@ -199,105 +214,105 @@ private:
  }

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  [[nodiscard]]
@@ -339,9 +354,7 @@ private:
  }

  // Restart helpers
-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
  void bl_OrRestart(T* Label) {
    if (bl(Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
      return;
@@ -349,12 +362,10 @@ private:

    // We can support this but currently unnecessary.
    ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
  }

-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
  void b_OrRestart(T* Label) {
    if (b(Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
      return;
@@ -362,12 +373,10 @@ private:

    // We can support this but currently unnecessary.
    ERROR_AND_DIE_FMT("Tried to branch larger than 128MB away!");
-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
  }

-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
  void b_OrRestart(ARMEmitter::Condition Cond, T* Label) {
    if (RequiresFarARM64Jumps) {
      ARMEmitter::ForwardLabel Skip {};
@@ -385,12 +394,10 @@ private:
      return;
    }

-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
  }

-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
  void cbz_OrRestart(ARMEmitter::Size s, ARMEmitter::Register rt, T* Label) {
    if (RequiresFarARM64Jumps) {
      ARMEmitter::ForwardLabel Skip {};
@@ -408,12 +415,10 @@ private:
      return;
    }

-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
  }

-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
  void cbnz_OrRestart(ARMEmitter::Size s, ARMEmitter::Register rt, T* Label) {
    if (RequiresFarARM64Jumps) {
      ARMEmitter::ForwardLabel Skip {};
@@ -431,12 +436,10 @@ private:
      return;
    }

-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
  }

-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
  void tbz_OrRestart(ARMEmitter::Register rt, uint32_t Bit, T* Label) {
    if (RequiresFarARM64Jumps) {
      ARMEmitter::ForwardLabel Skip {};
@@ -454,12 +457,10 @@ private:
      return;
    }

-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
  }

-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
  void tbnz_OrRestart(ARMEmitter::Register rt, uint32_t Bit, T* Label) {
    if (RequiresFarARM64Jumps) {
      ARMEmitter::ForwardLabel Skip {};
@@ -477,38 +478,40 @@ private:
      return;
    }

-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
  }

-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
  void adr_OrRestart(ARMEmitter::Register rd, T* Label) {
+    if (RequiresFarARM64Jumps) {
+      if (LongAddressGen(rd, Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
+        ERROR_AND_DIE_FMT("Unable to encode long ADR.");
+      }
+      return;
+    }
    if (adr(rd, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
      return;
    }

-    // We can support this but currently unnecessary.
-    ERROR_AND_DIE_FMT("Long ADR currently unsupported!");
-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
  }

-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
  void adrp_OrRestart(ARMEmitter::Register rd, T* Label) {
+    if (RequiresFarARM64Jumps) {
+      if (LongAddressGen(rd, Label) == ARMEmitter::BranchEncodeSucceeded::Failure) {
+        ERROR_AND_DIE_FMT("Unable to encode long ADRP.");
+      }
+      return;
+    }
    if (adrp(rd, Label) == ARMEmitter::BranchEncodeSucceeded::Success) {
      return;
    }

-    // We can support this but currently unnecessary.
-    ERROR_AND_DIE_FMT("Long ADRP currently unsupported!");
-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
  }

-  template<typename T>
-  requires (std::is_same_v<T, ARMEmitter::ForwardLabel> || std::is_same_v<T, ARMEmitter::BackwardLabel> ||
-            std::is_same_v<T, ARMEmitter::BiDirectionalLabel> || std::is_same_v<T, ARMEmitter::ForwardLabel::Reference>)
+  template<ARMEmitter::IsLabel T>
  void BindOrRestart(T* Label) {
    if (Bind(Label)) {
      return;
@@ -516,11 +519,11 @@ private:

    if (RequiresFarARM64Jumps) {
      // This should have been caught before this point.
-      ERROR_AND_DIE_FMT("Oops. Unhandled long bind.");
+      ERROR_AND_DIE_FMT("Unhandled long bind");
      return;
    }

-    FEXCore::LongJump::LongJump(RestartControl.RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
+    FEXCore::UncheckedLongJump::LongJump(ThreadState->RestartJump, FEXCore::ToUnderlying(RestartOptions::Control::EnableFarARM64Jumps));
  }

  // This is purely a debugging aid for developers to see if they are in JIT code space when inspecting raw memory
@@ -533,8 +536,6 @@ private:
   * @name Relocations
   * @{ */

-  uint64_t GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op);
-
  /**
   * @brief A literal pair relocation object for named symbol literals
   */
@@ -571,19 +572,30 @@ private:
   */
  NamedSymbolLiteralPair InsertNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op);

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

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

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

  /**  @} */

@@ -606,7 +618,7 @@ private:
  void Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize, ARMEmitter::VRegister Dst, ARMEmitter::VRegister IncomingDst,
                           std::optional<ARMEmitter::Register> BaseAddr, ARMEmitter::VRegister VectorIndexLow,
                           std::optional<ARMEmitter::VRegister> VectorIndexHigh, ARMEmitter::VRegister MaskReg, IR::OpSize VectorIndexSize,
-                           size_t DataElementOffsetStart, size_t IndexElementOffsetStart, uint8_t OffsetScale);
+                           size_t DataElementOffsetStart, size_t IndexElementOffsetStart, uint8_t OffsetScale, IR::OpSize AddrSize);

  void EmitTFCheck();

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

    switch (Op->Stride) {
@@ -372,7 +373,7 @@ DEF_OP(SpillRegister) {
  const auto OpSize = IROp->Size;
  const uint32_t SlotOffset = Op->Slot * MaxSpillSlotSize;

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    const auto Src = GetReg(Op->Value);
    switch (OpSize) {
    case IR::OpSize::i8Bit: {
@@ -413,7 +414,7 @@ DEF_OP(SpillRegister) {
    }
    default: LOGMAN_MSG_A_FMT("Unhandled SpillRegister size: {}", OpSize); break;
    }
-  } else if (Op->Class == FEXCore::IR::FPRClass) {
+  } else if (Op->Class == FEXCore::IR::RegClass::FPR) {
    const auto Src = GetVReg(Op->Value);

    switch (OpSize) {
@@ -452,7 +453,7 @@ DEF_OP(SpillRegister) {
    default: LOGMAN_MSG_A_FMT("Unhandled SpillRegister size: {}", OpSize); break;
    }
  } else {
-    LOGMAN_MSG_A_FMT("Unhandled SpillRegister class: {}", Op->Class.Val);
+    LOGMAN_MSG_A_FMT("Unhandled SpillRegister class: {}", Op->Class);
  }
 }

@@ -461,7 +462,7 @@ DEF_OP(FillRegister) {
  const auto OpSize = IROp->Size;
  const uint32_t SlotOffset = Op->Slot * MaxSpillSlotSize;

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    const auto Dst = GetReg(Node);
    switch (OpSize) {
    case IR::OpSize::i8Bit: {
@@ -502,7 +503,7 @@ DEF_OP(FillRegister) {
    }
    default: LOGMAN_MSG_A_FMT("Unhandled FillRegister size: {}", OpSize); break;
    }
-  } else if (Op->Class == FEXCore::IR::FPRClass) {
+  } else if (Op->Class == FEXCore::IR::RegClass::FPR) {
    const auto Dst = GetVReg(Node);

    switch (OpSize) {
@@ -541,7 +542,7 @@ DEF_OP(FillRegister) {
    default: LOGMAN_MSG_A_FMT("Unhandled FillRegister size: {}", OpSize); break;
    }
  } else {
-    LOGMAN_MSG_A_FMT("Unhandled FillRegister class: {}", Op->Class.Val);
+    LOGMAN_MSG_A_FMT("Unhandled FillRegister class: {}", Op->Class);
  }
 }

@@ -565,6 +566,24 @@ DEF_OP(LoadDF) {
  ldrsb(Dst.X(), STATE, offsetof(FEXCore::Core::CPUState, flags[Flag]));
 }

+DEF_OP(ContextClear) {
+  auto Op = IROp->C<IR::IROp_ContextClear>();
+  if (CTX->HostFeatures.SupportsCLZERO) {
+    // We can use CLZero directly when hardware supports it.
+    // Provides a fairly generous speed-up on Ampere1A hardware.
+    // TODO: When FEAT_MOPS hardware ships, test memset using MOPS.
+    for (size_t i = 0; i < Op->Size; i += 64) {
+      add(ARMEmitter::Size::i64Bit, TMP1, STATE.R(), Op->Offset + i);
+      dc(ARMEmitter::DataCacheOperation::ZVA, TMP1);
+    }
+  } else {
+    movi(ARMEmitter::SubRegSize::i64Bit, VTMP1.Q(), 0);
+    for (size_t i = 0; i < Op->Size; i += 32) {
+      stp<ARMEmitter::IndexType::OFFSET>(VTMP1.Q(), VTMP1.Q(), STATE.R(), Op->Offset + i);
+    }
+  }
+}
+
 ARMEmitter::ExtendedMemOperand Arm64JITCore::GenerateMemOperand(
  IR::OpSize AccessSize, ARMEmitter::Register Base, IR::OrderedNodeWrapper Offset, IR::MemOffsetType OffsetType, uint8_t OffsetScale) {
  if (Offset.IsInvalid()) {
@@ -578,14 +597,14 @@ ARMEmitter::ExtendedMemOperand Arm64JITCore::GenerateMemOperand(
      return ARMEmitter::ExtendedMemOperand(Base.X(), ARMEmitter::IndexType::OFFSET, Const);
    } else {
      auto RegOffset = GetReg(Offset);
-      switch (OffsetType.Val) {
-      case IR::MEM_OFFSET_SXTX.Val:
+      switch (OffsetType) {
+      case IR::MemOffsetType::SXTX:
        return ARMEmitter::ExtendedMemOperand(Base.X(), RegOffset.X(), ARMEmitter::ExtendedType::SXTX, FEXCore::ilog2(OffsetScale));
-      case IR::MEM_OFFSET_UXTW.Val:
+      case IR::MemOffsetType::UXTW:
        return ARMEmitter::ExtendedMemOperand(Base.X(), RegOffset.X(), ARMEmitter::ExtendedType::UXTW, FEXCore::ilog2(OffsetScale));
-      case IR::MEM_OFFSET_SXTW.Val:
+      case IR::MemOffsetType::SXTW:
        return ARMEmitter::ExtendedMemOperand(Base.X(), RegOffset.X(), ARMEmitter::ExtendedType::SXTW, FEXCore::ilog2(OffsetScale));
-      default: LOGMAN_MSG_A_FMT("Unhandled GenerateMemOperand OffsetType: {}", OffsetType.Val); break;
+      default: LOGMAN_MSG_A_FMT("Unhandled GenerateMemOperand OffsetType: {}", OffsetType); break;
      }
    }
  }
@@ -612,20 +631,20 @@ ARMEmitter::Register Arm64JITCore::ApplyMemOperand(IR::OpSize AccessSize, ARMEmi
    add(ARMEmitter::Size::i64Bit, Tmp, Base, Tmp, ARMEmitter::ShiftType::LSL, FEXCore::ilog2(OffsetScale));
  } else {
    auto RegOffset = GetReg(Offset);
-    switch (OffsetType.Val) {
-    case IR::MEM_OFFSET_SXTX.Val:
+    switch (OffsetType) {
+    case IR::MemOffsetType::SXTX:
      add(ARMEmitter::Size::i64Bit, Tmp, Base, RegOffset, ARMEmitter::ExtendedType::SXTX, FEXCore::ilog2(OffsetScale));
      break;

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

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

-    default: LOGMAN_MSG_A_FMT("Unhandled OffsetType: {}", OffsetType.Val); break;
+    default: LOGMAN_MSG_A_FMT("Unhandled OffsetType: {}", OffsetType); break;
    }
  }
  return Tmp;
@@ -676,7 +695,7 @@ ARMEmitter::SVEMemOperand Arm64JITCore::GenerateSVEMemOperand(IR::OpSize AccessS
  // Note that we do nothing with the offset type and offset scale,
  // since SVE loads and stores don't have the ability to perform an
  // optional extension or shift as part of their behavior.
-  LOGMAN_THROW_A_FMT(OffsetType.Val == IR::MEM_OFFSET_SXTX.Val, "Currently only the default offset type (SXTX) is supported.");
+  LOGMAN_THROW_A_FMT(OffsetType == IR::MemOffsetType::SXTX, "Currently only the default offset type (SXTX) is supported.");

  const auto RegOffset = GetReg(Offset);
  return ARMEmitter::SVEMemOperand(Base.X(), RegOffset.X());
@@ -689,7 +708,7 @@ DEF_OP(LoadMem) {
  const auto MemReg = GetReg(Op->Addr);
  const auto MemSrc = GenerateMemOperand(OpSize, MemReg, Op->Offset, Op->OffsetType, Op->OffsetScale);

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

    switch (OpSize) {
@@ -723,7 +742,7 @@ DEF_OP(LoadMemPair) {
  const auto Op = IROp->C<IR::IROp_LoadMemPair>();
  const auto Addr = GetReg(Op->Addr);

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

@@ -751,13 +770,13 @@ DEF_OP(LoadMemTSO) {

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

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

-  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == FEXCore::IR::GPRClass) {
+  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == IR::RegClass::GPR) {
    const auto Dst = GetReg(Node);
    uint64_t Offset = 0;
    if (!Op->Offset.IsInvalid()) {
@@ -776,12 +795,10 @@ DEF_OP(LoadMemTSO) {
      case IR::OpSize::i64Bit: ldapur(Dst.X(), MemReg, Offset); break;
      default: LOGMAN_MSG_A_FMT("Unhandled LoadMemTSO size: {}", OpSize); break;
      }
-      if (HalfBarrierTSOEnabled() && !ParanoidTSO()) {
-        // Half-barrier once back-patched.
-        nop();
-      }
+      // Half-barrier once back-patched.
+      nop();
    }
-  } else if (CTX->HostFeatures.SupportsRCPC && Op->Class == FEXCore::IR::GPRClass) {
+  } else if (CTX->HostFeatures.SupportsRCPC && Op->Class == IR::RegClass::GPR) {
    const auto Dst = GetReg(Node);
    if (OpSize == IR::OpSize::i8Bit) {
      // 8bit load is always aligned to natural alignment
@@ -793,12 +810,10 @@ DEF_OP(LoadMemTSO) {
      case IR::OpSize::i64Bit: ldapr(Dst.X(), MemReg); break;
      default: LOGMAN_MSG_A_FMT("Unhandled LoadMemTSO size: {}", OpSize); break;
      }
-      if (HalfBarrierTSOEnabled() && !ParanoidTSO()) {
-        // Half-barrier once back-patched.
-        nop();
-      }
+      // Half-barrier once back-patched.
+      nop();
    }
-  } else if (Op->Class == FEXCore::IR::GPRClass) {
+  } else if (Op->Class == IR::RegClass::GPR) {
    const auto Dst = GetReg(Node);
    if (OpSize == IR::OpSize::i8Bit) {
      // 8bit load is always aligned to natural alignment
@@ -810,10 +825,8 @@ DEF_OP(LoadMemTSO) {
      case IR::OpSize::i64Bit: ldar(Dst.X(), MemReg); break;
      default: LOGMAN_MSG_A_FMT("Unhandled LoadMemTSO size: {}", OpSize); break;
      }
-      if (HalfBarrierTSOEnabled() && !ParanoidTSO()) {
-        // Half-barrier once back-patched.
-        nop();
-      }
+      // Half-barrier once back-patched.
+      nop();
    }
  } else {
    const auto Dst = GetVReg(Node);
@@ -1045,7 +1058,7 @@ void Arm64JITCore::Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize,
                                       ARMEmitter::VRegister IncomingDst, std::optional<ARMEmitter::Register> BaseAddr,
                                       ARMEmitter::VRegister VectorIndexLow, std::optional<ARMEmitter::VRegister> VectorIndexHigh,
                                       ARMEmitter::VRegister MaskReg, IR::OpSize VectorIndexSize, size_t DataElementOffsetStart,
-                                       size_t IndexElementOffsetStart, uint8_t OffsetScale) {
+                                       size_t IndexElementOffsetStart, uint8_t OffsetScale, IR::OpSize AddrSize) {
  LOGMAN_THROW_A_FMT(ElementSize >= IR::OpSize::i8Bit && ElementSize <= IR::OpSize::i64Bit, "Invalid element size");

  const auto PerformSMove = [this](IR::OpSize ElementSize, const ARMEmitter::Register Dst, const ARMEmitter::VRegister Vector, int index) {
@@ -1121,17 +1134,17 @@ void Arm64JITCore::Emulate128BitGather(IR::OpSize Size, IR::OpSize ElementSize,
    // Calculate memory position for this gather load
    if (BaseAddr.has_value()) {
      if (VectorIndexSize == IR::OpSize::i32Bit) {
-        add(ARMEmitter::Size::i64Bit, TempMemReg, *BaseAddr, WorkingReg, ARMEmitter::ExtendedType::SXTW, FEXCore::ilog2(OffsetScale));
+        add(ConvertSize(AddrSize), TempMemReg, *BaseAddr, WorkingReg, ARMEmitter::ExtendedType::SXTW, FEXCore::ilog2(OffsetScale));
      } else {
-        add(ARMEmitter::Size::i64Bit, TempMemReg, *BaseAddr, WorkingReg, ARMEmitter::ShiftType::LSL, FEXCore::ilog2(OffsetScale));
+        add(ConvertSize(AddrSize), TempMemReg, *BaseAddr, WorkingReg, ARMEmitter::ShiftType::LSL, FEXCore::ilog2(OffsetScale));
      }
    } else {
      ///< In this case we have no base address, All addresses come from the vector register itself
      if (VectorIndexSize == IR::OpSize::i32Bit) {
        // Sign extend and shift in to the 64-bit register
-        sbfiz(ARMEmitter::Size::i64Bit, TempMemReg, WorkingReg, FEXCore::ilog2(OffsetScale), 32);
+        sbfiz(ConvertSize(AddrSize), TempMemReg, WorkingReg, FEXCore::ilog2(OffsetScale), 32);
      } else {
-        lsl(ARMEmitter::Size::i64Bit, TempMemReg, WorkingReg, FEXCore::ilog2(OffsetScale));
+        lsl(ConvertSize(AddrSize), TempMemReg, WorkingReg, FEXCore::ilog2(OffsetScale));
      }
    }

@@ -1189,7 +1202,8 @@ DEF_OP(VLoadVectorGatherMasked) {

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

  if (SupportsSVELoad) {
    uint8_t SVEScale = FEXCore::ilog2(OffsetScale);
@@ -1247,7 +1261,7 @@ DEF_OP(VLoadVectorGatherMasked) {
  } else {
    LOGMAN_THROW_A_FMT(!Is256Bit, "Can't emulate this gather load in the backend! Programming error!");
    Emulate128BitGather(IROp->Size, IROp->ElementSize, Dst, IncomingDst, BaseAddr, VectorIndexLow, VectorIndexHigh, MaskReg,
-                        VectorIndexSize, DataElementOffsetStart, IndexElementOffsetStart, OffsetScale);
+                        VectorIndexSize, DataElementOffsetStart, IndexElementOffsetStart, OffsetScale, Op->AddrSize);
  }
 }

@@ -1272,7 +1286,9 @@ DEF_OP(VLoadVectorGatherMaskedQPS) {
    !Op->VectorIndexHigh.IsInvalid() ? std::make_optional(GetVReg(Op->VectorIndexHigh)) : std::nullopt;

  ///< If the host supports SVE and the offset scale matches SVE limitations then it can do an SVE style load.
-  if (HostSupportsSVE128 && (OffsetScale == 1 || OffsetScale == 4)) {
+  const bool SupportsSVELoad = HostSupportsSVE128 && (OffsetScale == 1 || OffsetScale == 4) && Op->AddrSize == IR::OpSize::i64Bit;
+
+  if (SupportsSVELoad) {
    ARMEmitter::SVEModType ModType = ARMEmitter::SVEModType::MOD_NONE;
    if (OffsetScale != 1) {
      ModType = ARMEmitter::SVEModType::MOD_LSL;
@@ -1326,7 +1342,7 @@ DEF_OP(VLoadVectorGatherMaskedQPS) {
    }
  } else {
    Emulate128BitGather(IR::OpSize::i128Bit, IR::OpSize::i32Bit, Dst, IncomingDst, BaseAddr, VectorIndexLow, VectorIndexHigh, MaskReg,
-                        IR::OpSize::i64Bit, 0, 0, OffsetScale);
+                        IR::OpSize::i64Bit, 0, 0, OffsetScale, Op->AddrSize);
  }
 }

@@ -1627,7 +1643,7 @@ DEF_OP(StoreMem) {
  const auto MemReg = GetReg(Op->Addr);
  const auto MemSrc = GenerateMemOperand(OpSize, MemReg, Op->Offset, Op->OffsetType, Op->OffsetScale);

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    const auto Src = GetZeroableReg(Op->Value);
    switch (OpSize) {
    case IR::OpSize::i8Bit: strb(Src, MemSrc); break;
@@ -1738,7 +1754,7 @@ DEF_OP(StoreMemPair) {
  const auto OpSize = IROp->Size;
  const auto Addr = GetReg(Op->Addr);

-  if (Op->Class == FEXCore::IR::GPRClass) {
+  if (Op->Class == IR::RegClass::GPR) {
    const auto Src1 = GetZeroableReg(Op->Value1);
    const auto Src2 = GetZeroableReg(Op->Value2);
    switch (OpSize) {
@@ -1765,13 +1781,13 @@ DEF_OP(StoreMemTSO) {

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

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

-  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == FEXCore::IR::GPRClass) {
+  if (CTX->HostFeatures.SupportsTSOImm9 && Op->Class == IR::RegClass::GPR) {
    const auto Src = GetZeroableReg(Op->Value);
    uint64_t Offset = 0;
    if (!Op->Offset.IsInvalid()) {
@@ -1783,10 +1799,8 @@ DEF_OP(StoreMemTSO) {
      // 8bit load is always aligned to natural alignment
      stlurb(Src, MemReg, Offset);
    } else {
-      if (HalfBarrierTSOEnabled() && !ParanoidTSO()) {
-        // Half-barrier once back-patched.
-        nop();
-      }
+      // Half-barrier once back-patched.
+      nop();
      switch (OpSize) {
      case IR::OpSize::i16Bit: stlurh(Src, MemReg, Offset); break;
      case IR::OpSize::i32Bit: stlur(Src.W(), MemReg, Offset); break;
@@ -1794,17 +1808,15 @@ DEF_OP(StoreMemTSO) {
      default: LOGMAN_MSG_A_FMT("Unhandled StoreMemTSO size: {}", OpSize); break;
      }
    }
-  } else if (Op->Class == FEXCore::IR::GPRClass) {
+  } else if (Op->Class == IR::RegClass::GPR) {
    const auto Src = GetZeroableReg(Op->Value);

    if (OpSize == IR::OpSize::i8Bit) {
      // 8bit load is always aligned to natural alignment
      stlrb(Src, MemReg);
    } else {
-      if (HalfBarrierTSOEnabled() && !ParanoidTSO()) {
-        // Half-barrier once back-patched.
-        nop();
-      }
+      // Half-barrier once back-patched.
+      nop();
      switch (OpSize) {
      case IR::OpSize::i16Bit: stlrh(Src, MemReg); break;
      case IR::OpSize::i32Bit: stlr(Src.W(), MemReg); break;
@@ -1898,9 +1910,7 @@ DEF_OP(MemSet) {
      // 8bit load is always aligned to natural alignment
      stlrb(Value.W(), TMP2);
    } else {
-      if (HalfBarrierTSOEnabled() && !ParanoidTSO()) {
-        nop();
-      }
+      nop();
      switch (OpSize) {
      case 2: stlrh(Value.W(), TMP2); break;
      case 4: stlr(Value.W(), TMP2); break;
@@ -2118,11 +2128,9 @@ DEF_OP(MemCpy) {
        default: LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, Size); break;
        }

-        if (HalfBarrierTSOEnabled() && !ParanoidTSO()) {
-          // Placeholders for backpatching barriers (one per load/store)
-          nop();
-          nop();
-        }
+        // Placeholders for backpatching barriers (one per load/store)
+        nop();
+        nop();

        switch (OpSize) {
        case 2: stlrh(TMP4.W(), TMP2); break;
@@ -2144,11 +2152,9 @@ DEF_OP(MemCpy) {
        default: LOGMAN_MSG_A_FMT("Unhandled {} size: {}", __func__, Size); break;
        }

-        if (HalfBarrierTSOEnabled() && !ParanoidTSO()) {
-          // Placeholders for backpatching barriers (one per load/store)
-          nop();
-          nop();
-        }
+        // Placeholders for backpatching barriers (one per load/store)
+        nop();
+        nop();

        switch (OpSize) {
        case 2: stlrh(TMP4.W(), TMP2); break;
@@ -15,6 +15,7 @@ $end_info$

 #include <FEXCore/Core/SignalDelegator.h>
 #include <FEXCore/Debug/InternalThreadState.h>
+#include <FEXCore/Utils/EnumUtils.h>

 namespace FEXCore::CPU {

@@ -47,10 +48,10 @@ DEF_OP(GuestOpcode) {
 DEF_OP(Fence) {
  auto Op = IROp->C<IR::IROp_Fence>();
  switch (Op->Fence) {
-  case IR::Fence_Load.Val: dmb(ARMEmitter::BarrierScope::LD); break;
-  case IR::Fence_LoadStore.Val: dmb(ARMEmitter::BarrierScope::SY); break;
-  case IR::Fence_Store.Val: dmb(ARMEmitter::BarrierScope::ST); break;
-  case IR::Fence_Inst.Val: isb(); break;
+  case IR::FenceType::Load: dmb(ARMEmitter::BarrierScope::LD); break;
+  case IR::FenceType::LoadStore: dmb(ARMEmitter::BarrierScope::SY); break;
+  case IR::FenceType::Store: dmb(ARMEmitter::BarrierScope::ST); break;
+  case IR::FenceType::Inst: isb(); break;
  default: LOGMAN_MSG_A_FMT("Unknown Fence: {}", Op->Fence); break;
  }
 }
@@ -77,19 +78,19 @@ DEF_OP(Break) {

  switch (Op->Reason.Signal) {
  case Core::FAULT_SIGILL:
-    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGILL));
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.GuestSignal_SIGILL));
    br(TMP1);
    break;
  case Core::FAULT_SIGTRAP:
-    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP));
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.GuestSignal_SIGTRAP));
    br(TMP1);
    break;
  case Core::FAULT_SIGSEGV:
-    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGSEGV));
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.GuestSignal_SIGSEGV));
    br(TMP1);
    break;
  default:
-    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP));
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.GuestSignal_SIGTRAP));
    br(TMP1);
    break;
  }
@@ -107,10 +108,10 @@ DEF_OP(GetRoundingMode) {
  // zero. Just swapping 01 and 10. That's a bitfield reverse. Round mode is in
  // bottom two bits. After reversing as a 32-bit operation, it'll be in [31:30]
  // and ripe for reinsertion back at 0.
-  static_assert(IR::ROUND_MODE_NEAREST == 0);
-  static_assert(IR::ROUND_MODE_NEGATIVE_INFINITY == 1);
-  static_assert(IR::ROUND_MODE_POSITIVE_INFINITY == 2);
-  static_assert(IR::ROUND_MODE_TOWARDS_ZERO == 3);
+  static_assert(FEXCore::ToUnderlying(IR::RoundMode::Nearest) == 0);
+  static_assert(FEXCore::ToUnderlying(IR::RoundMode::NegInfinity) == 1);
+  static_assert(FEXCore::ToUnderlying(IR::RoundMode::PosInfinity) == 2);
+  static_assert(FEXCore::ToUnderlying(IR::RoundMode::TowardsZero) == 3);

  rbit(ARMEmitter::Size::i32Bit, TMP1, Dst);
  bfi(ARMEmitter::Size::i64Bit, Dst, TMP1, 30, 2);
@@ -188,11 +189,11 @@ DEF_OP(Print) {

  if (IsGPR(Op->Value)) {
    mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, GetReg(Op->Value));
-    ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.PrintValue));
+    ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.PrintValue));
  } else {
    fmov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, GetVReg(Op->Value), false);
    fmov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, GetVReg(Op->Value), true);
-    ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.PrintVectorValue));
+    ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.PrintVectorValue));
  }

  if (!CTX->Config.DisableVixlIndirectCalls) [[unlikely]] {
@@ -240,7 +241,7 @@ DEF_OP(ProcessorID) {
  sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, 16);

  // Load the getcpu syscall number
-#if defined(_M_X86_64)
+#if defined(ARCHITECTURE_x86_64)
  // Just to ensure the syscall number doesn't change if compiled for an x86_64 host.
  constexpr auto GetCPUSyscallNum = 0xa8;
 #else
@@ -304,20 +305,20 @@ DEF_OP(MonoBackpatcherWrite) {
    mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, TMP4);
  }

-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
  ldr(TMP2, ARMEmitter::XReg::x18, TEB_CPU_AREA_OFFSET);
  LoadConstant(ARMEmitter::Size::i32Bit, TMP1, 1);
  strb(TMP1.W(), TMP2, CPU_AREA_IN_SYSCALL_CALLBACK_OFFSET);
 #endif

-  ldr(ARMEmitter::XReg::x4, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.MonoBackpatcherWrite));
+  ldr(ARMEmitter::XReg::x4, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.MonoBackpatcherWrite));
  if (!CTX->Config.DisableVixlIndirectCalls) [[unlikely]] {
    GenerateIndirectRuntimeCall<void, void*, uint8_t, uint64_t, uint64_t>(ARMEmitter::Reg::r4);
  } else {
    blr(ARMEmitter::Reg::r4);
  }

-#ifdef _M_ARM_64EC
+#ifdef ARCHITECTURE_arm64ec
  ldr(TMP2, ARMEmitter::XReg::x18, TEB_CPU_AREA_OFFSET);
  strb(ARMEmitter::WReg::zr, TMP2, CPU_AREA_IN_SYSCALL_CALLBACK_OFFSET);
 #endif
@@ -18,18 +18,4 @@ DEF_OP(RMWHandle) {
  mov(ARMEmitter::Size::i64Bit, GetReg(Node), GetReg(IROp->Args[0]));
 }

-DEF_OP(Swap1) {
-  auto Op = IROp->C<IR::IROp_Swap1>();
-  auto A = GetReg(Op->A), B = GetReg(Op->B);
-  LOGMAN_THROW_A_FMT(B == GetReg(Node), "Invariant");
-
-  mov(ARMEmitter::Size::i64Bit, TMP1, A);
-  mov(ARMEmitter::Size::i64Bit, A, B);
-  mov(ARMEmitter::Size::i64Bit, B, TMP1);
-}
-
-DEF_OP(Swap2) {
-  // Implemented above
-}
-
 } // namespace FEXCore::CPU
@@ -1,79 +1,100 @@
 // SPDX-License-Identifier: MIT
 #pragma once
 #include <FEXCore/IR/IR.h>
+#include <FEXCore/Utils/CompilerDefs.h>
+
+namespace FEXCore::Context {
+class ContextImpl;
+}

 namespace FEXCore::CPU {
-enum class RelocationTypes : uint8_t {
+enum class RelocationTypes : uint32_t {
  // 8 byte literal in memory for symbol
  // Aligned to struct RelocNamedSymbolLiteral
  RELOC_NAMED_SYMBOL_LITERAL,

  // Fixed size named thunk move
-  // 4 instruction constant generation on AArch64
-  // 64-bit mov on x86-64
+  // 4 instruction constant generation
  // Aligned to struct RelocNamedThunkMove
  RELOC_NAMED_THUNK_MOVE,

+  // 8 byte literal (relative to binary base address)
+  RELOC_GUEST_RIP_LITERAL,
+
  // Fixed size guest RIP move
-  // 4 instruction constant generation on AArch64
-  // 64-bit mov on x86-64
-  // Aligned to struct RelocGuestRIPMove
+  // 4 instruction constant generation
+  // Aligned to struct RelocGuestRIP
  RELOC_GUEST_RIP_MOVE,
 };

-struct RelocationTypeHeader final {
+struct FEX_PACKED RelocationHeader final {
+  // Offset to the relocated host code data
+  uint64_t Offset {};
+
  RelocationTypes Type;
 };

 struct RelocNamedSymbolLiteral final {
-  enum class NamedSymbol : uint8_t {
+  enum class NamedSymbol : uint32_t {
    ///< Thread specific relocations
    // JIT Literal pointers
    SYMBOL_LITERAL_EXITFUNCTION_LINKER,
  };

-  RelocationTypeHeader Header {};
+  RelocationHeader Header {};

  NamedSymbol Symbol;

-  // Offset in to the code section to begin the relocation
-  uint64_t Offset {};
+  uint32_t Pad[8];
 };

 struct RelocNamedThunkMove final {
-  RelocationTypeHeader Header {};
+  RelocationHeader Header {};

  // GPR index the constant is being moved to
-  uint8_t RegisterIndex;
+  uint32_t RegisterIndex;

  // The thunk SHA256 hash
  IR::SHA256Sum Symbol;
-
-  // Offset in to the code section to begin the relocation
-  uint64_t Offset {};
 };

-struct RelocGuestRIPMove final {
-  RelocationTypeHeader Header {};
+struct RelocGuestRIP final {
+  RelocationHeader Header {};

-  // GPR index the constant is being moved to
+  // GPR index the constant is being moved to (for non-literal relocations)
  uint8_t RegisterIndex;

-  // Offset in to the code section to begin the relocation
-  uint64_t Offset {};
+  char Pad[3];

-  // The unrelocated RIP that is being moved
+  // The base RIP (to be moved by the register for non-literal relocations).
+  // In a serialized code cache, this is relative to the binary base address.
  uint64_t GuestRIP;
+
+  uint32_t pad2[6] {};
 };

 union Relocation {
-  RelocationTypeHeader Header {};
+  // Clang 16 Can't default-initialize this union
+  static Relocation Default() {
+#if __clang_major__ < 17
+    Relocation Ret {.Header {}};
+    memset(&Ret, 0, sizeof(Ret));
+    return Ret;
+#else
+    return {};
+#endif
+  }
+
+  RelocationHeader Header {};

  RelocNamedSymbolLiteral NamedSymbolLiteral;
  // This makes our union of relocations at least 48 bytes
  // It might be more efficient to not use a union
  RelocNamedThunkMove NamedThunkMove;

-  RelocGuestRIPMove GuestRIPMove;
+  RelocGuestRIP GuestRIP;
 };
+
+uint64_t GetNamedSymbolLiteral(FEXCore::Context::ContextImpl&, RelocNamedSymbolLiteral::NamedSymbol);
+
 } // namespace FEXCore::CPU
@@ -41,6 +41,7 @@ namespace FEXCore::CPU {
    const auto Op = IROp->C<IR::IROp_##FEXOp>();                                                                                    \
    const auto OpSize = IROp->Size;                                                                                                 \
    const auto Is256Bit = OpSize == IR::OpSize::i256Bit;                                                                            \
+    const auto Is128Bit = OpSize == IR::OpSize::i128Bit;                                                                            \
    LOGMAN_THROW_A_FMT(!Is256Bit || HostSupportsSVE256, "Need SVE256 support in order to use {} with 256-bit operation", __func__); \
                                                                                                                                    \
    const auto Dst = GetVReg(Node);                                                                                                 \
@@ -49,8 +50,10 @@ namespace FEXCore::CPU {
                                                                                                                                    \
    if (HostSupportsSVE256 && Is256Bit) {                                                                                           \
      ARMOp(Dst.Z(), Vector1.Z(), Vector2.Z());                                                                                     \
-    } else {                                                                                                                        \
+    } else if (Is128Bit) {                                                                                                          \
      ARMOp(Dst.Q(), Vector1.Q(), Vector2.Q());                                                                                     \
+    } else {                                                                                                                        \
+      ARMOp(Dst.D(), Vector1.D(), Vector2.D());                                                                                     \
    }                                                                                                                               \
  }

@@ -744,11 +747,11 @@ DEF_OP(VFToIScalarInsert) {
    auto Src = *std::get_if<ARMEmitter::VRegister>(&SrcVar);

    switch (RoundMode) {
-    case IR::Round_Nearest: frintn(SubRegSize.Scalar, Dst, Src); break;
-    case IR::Round_Negative_Infinity: frintm(SubRegSize.Scalar, Dst, Src); break;
-    case IR::Round_Positive_Infinity: frintp(SubRegSize.Scalar, Dst, Src); break;
-    case IR::Round_Towards_Zero: frintz(SubRegSize.Scalar, Dst, Src); break;
-    case IR::Round_Host: frinti(SubRegSize.Scalar, Dst, Src); break;
+    case IR::RoundMode::Nearest: frintn(SubRegSize.Scalar, Dst, Src); break;
+    case IR::RoundMode::NegInfinity: frintm(SubRegSize.Scalar, Dst, Src); break;
+    case IR::RoundMode::PosInfinity: frintp(SubRegSize.Scalar, Dst, Src); break;
+    case IR::RoundMode::TowardsZero: frintz(SubRegSize.Scalar, Dst, Src); break;
+    case IR::RoundMode::Host: frinti(SubRegSize.Scalar, Dst, Src); break;
    }
  };

@@ -974,7 +977,7 @@ DEF_OP(LoadNamedVectorConstant) {
  }
  // Load the pointer.
  auto GenerateMemOperand = [this](IR::OpSize OpSize, uint32_t NamedConstant, ARMEmitter::Register Base) {
-    const auto ConstantOffset = offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.NamedVectorConstants[NamedConstant]);
+    const auto ConstantOffset = offsetof(FEXCore::Core::CpuStateFrame, Pointers.NamedVectorConstants[NamedConstant]);

    if (ConstantOffset <= 255 || // Unscaled 9-bit signed
        ((ConstantOffset & (IR::OpSizeToSize(OpSize) - 1)) == 0 &&
@@ -982,13 +985,13 @@ DEF_OP(LoadNamedVectorConstant) {
      return ARMEmitter::ExtendedMemOperand(Base.X(), ARMEmitter::IndexType::OFFSET, ConstantOffset);
    }

-    ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.NamedVectorConstantPointers[NamedConstant]));
+    ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.NamedVectorConstantPointers[NamedConstant]));
    return ARMEmitter::ExtendedMemOperand(TMP1, ARMEmitter::IndexType::OFFSET, 0);
  };

  if (OpSize == IR::OpSize::i256Bit) {
    // Handle SVE 32-byte variant upfront.
-    ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.NamedVectorConstantPointers[Op->Constant]));
+    ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.NamedVectorConstantPointers[Op->Constant]));
    ld1b<ARMEmitter::SubRegSize::i8Bit>(Dst.Z(), PRED_TMP_32B.Zeroing(), TMP1, 0);
    return;
  }
@@ -1010,7 +1013,7 @@ DEF_OP(LoadNamedVectorIndexedConstant) {
  const auto Dst = GetVReg(Node);

  // Load the pointer.
-  ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.Common.IndexedNamedVectorConstantPointers[Op->Constant]));
+  ldr(TMP1, STATE_PTR(CpuStateFrame, Pointers.IndexedNamedVectorConstantPointers[Op->Constant]));

  switch (OpSize) {
  case IR::OpSize::i8Bit: ldrb(Dst, TMP1, Op->Index); break;
@@ -1114,6 +1117,29 @@ DEF_OP(VAddP) {
  }
 }

+DEF_OP(VOrn) {
+  const auto Op = IROp->C<IR::IROp_VOrn>();
+  const auto OpSize = IROp->Size;
+
+  const auto Is256Bit = OpSize == IR::OpSize::i256Bit;
+  const auto Is128Bit = OpSize == IR::OpSize::i128Bit;
+  LOGMAN_THROW_A_FMT(!Is256Bit || HostSupportsSVE256, "Need SVE256 support in order to use {} with 256-bit operation", __func__);
+
+  const auto Dst = GetVReg(Node);
+  const auto Vector1 = GetVReg(Op->Vector1);
+  const auto Vector2 = GetVReg(Op->Vector2);
+
+  if (HostSupportsSVE256 && Is256Bit) {
+    const auto Pred = PRED_TMP_32B.Merging();
+    not_(ARMEmitter::SubRegSize::i8Bit, VTMP1.Z(), Pred, Vector2.Z());
+    orr(Dst.Z(), Vector1.Z(), VTMP1.Z());
+  } else if (Is128Bit) {
+    orn(Dst.Q(), Vector1.Q(), Vector2.Q());
+  } else {
+    orn(Dst.D(), Vector1.D(), Vector2.D());
+  }
+}
+
 DEF_OP(VFAddV) {
  const auto Op = IROp->C<IR::IROp_VFAddV>();
  const auto OpSize = IROp->Size;
@@ -15,7 +15,7 @@ $end_info$

 namespace FEXCore {
 GuestToHostMap::GuestToHostMap()
-  : BlockLinks_mbr {fextl::pmr::get_default_resource()} {
+  : BlockLinks_mbr {"FEXMem_BlockLinks"} {
  BlockLinks_pma = fextl::make_unique<std::pmr::polymorphic_allocator<std::byte>>(&BlockLinks_mbr);
  // Setup our PMR map.
  BlockLinks = BlockLinks_pma->new_object<BlockLinksMapType>();
@@ -24,7 +24,7 @@ GuestToHostMap::GuestToHostMap()
 LookupCache::LookupCache(FEXCore::Context::ContextImpl* CTX)
  : ctx {CTX} {

-  TotalCacheSize = ctx->Config.VirtualMemSize / 4096 * 8 + CODE_SIZE + L1_SIZE;
+  TotalCacheSize = ctx->Config.VirtualMemSize / FEXCore::Utils::FEX_PAGE_SIZE * 8 + CODE_SIZE + MAX_L1_SIZE;

  // Block cache ends up looking like this
  // PageMemoryMap[VirtualMemoryRegion >> 12]
@@ -39,6 +39,10 @@ LookupCache::LookupCache(FEXCore::Context::ContextImpl* CTX)
  // We need one pointer per page of virtual memory
  // At 64GB of virtual memory this will allocate 128MB of virtual memory space
  PagePointer = reinterpret_cast<uintptr_t>(FEXCore::Allocator::VirtualAlloc(TotalCacheSize, false, false));
+  LOGMAN_THROW_A_FMT(PagePointer != -1ULL, "Failed to allocate PagePointer");
+
+  FEXCore::Allocator::VirtualName("FEXMem_Lookup", reinterpret_cast<void*>(PagePointer),
+                                  ctx->Config.VirtualMemSize / FEXCore::Utils::FEX_PAGE_SIZE * 8 + CODE_SIZE);
  CTX->SyscallHandler->MarkOvercommitRange(PagePointer, TotalCacheSize);

  // Allocate our memory backing our pages
@@ -46,14 +50,21 @@ LookupCache::LookupCache(FEXCore::Context::ContextImpl* CTX)
  // XXX: We can drop down to 16KB if we store 4byte offsets from the code base
  // We currently limit to 128MB of real memory for caching for the total cache size.
  // Can end up being inefficient if we compile a small number of blocks per page
-  PageMemory = PagePointer + ctx->Config.VirtualMemSize / 4096 * 8;
-  LOGMAN_THROW_A_FMT(PageMemory != -1ULL, "Failed to allocate page memory");
+  PageMemory = PagePointer + ctx->Config.VirtualMemSize / FEXCore::Utils::FEX_PAGE_SIZE * 8;

  // L1 Cache
  L1Pointer = PageMemory + CODE_SIZE;
-  LOGMAN_THROW_A_FMT(L1Pointer != -1ULL, "Failed to allocate L1Pointer");
+  FEXCore::Allocator::VirtualName("FEXMem_Lookup_L1", reinterpret_cast<void*>(L1Pointer), MAX_L1_SIZE);

  VirtualMemSize = ctx->Config.VirtualMemSize;
+
+  if (DynamicL1Cache()) {
+    // Start at minimum size when dynamic.
+    L1PointerMask = MIN_L1_ENTRIES - 1;
+  } else {
+    // Start at maximum instead.
+    L1PointerMask = MAX_L1_ENTRIES - 1;
+  }
 }

 LookupCache::~LookupCache() {
@@ -64,31 +75,27 @@ LookupCache::~LookupCache() {
  // These will get freed when their memory allocators are deallocated.
 }

-void LookupCache::ClearL2Cache() {
-  auto lk = Shared->AcquireLock();
+void LookupCache::ClearL2Cache(const FEXCore::LookupCacheBaseLockToken& lk) {
  // Clear out the page memory
  // PagePointer and PageMemory are sequential with each other. Clear both at once.
-  FEXCore::Allocator::VirtualDontNeed(reinterpret_cast<void*>(PagePointer), ctx->Config.VirtualMemSize / 4096 * 8 + CODE_SIZE, false);
+  FEXCore::Allocator::VirtualDontNeed(reinterpret_cast<void*>(PagePointer),
+                                      ctx->Config.VirtualMemSize / FEXCore::Utils::FEX_PAGE_SIZE * 8 + CODE_SIZE, false);
  AllocateOffset = 0;
 }

-void LookupCache::ClearThreadLocalCaches() {
-  auto lk = Shared->AcquireLock();
-
+void LookupCache::ClearThreadLocalCaches(const LookupCacheWriteLockToken&) {
  // Clear L1 and L2 by clearing the full cache.
  FEXCore::Allocator::VirtualDontNeed(reinterpret_cast<void*>(PagePointer), TotalCacheSize, false);
+  CachedCodePages.clear();
 }

-void LookupCache::ClearCache() {
-  auto lk = Shared->AcquireLock();
-
+void LookupCache::ClearCache(const LookupCacheWriteLockToken& lk) {
  // Clear L1 and L2 by clearing the full cache.
-  FEXCore::Allocator::VirtualDontNeed(reinterpret_cast<void*>(PagePointer), TotalCacheSize, false);
-
+  ClearThreadLocalCaches(lk);
  Shared->ClearCache(lk);
 }

-void GuestToHostMap::ClearCache(const LockToken&) {
+void GuestToHostMap::ClearCache(const LookupCacheWriteLockToken&) {
  // Allocate a new pointer from the BlockLinks pma again.
  BlockLinks = BlockLinks_pma->new_object<BlockLinksMapType>();
  // All code is gone, clear the block list
@@ -2,30 +2,57 @@
 #pragma once
 #include "Interface/Context/Context.h"
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/SHMStats.h>
+#include "Utils/WritePriorityMutex.h"
+
 #include <FEXCore/fextl/map.h>
 #include <FEXCore/fextl/memory_resource.h>
 #include <FEXCore/fextl/robin_map.h>
+#include <FEXCore/fextl/robin_set.h>
 #include <FEXCore/fextl/vector.h>
 #include <FEXCore/fextl/memory_resource.h>

 #include <cstdint>
-#include <functional>
 #include <stddef.h>
 #include <utility>
 #include <mutex>

 namespace FEXCore {
+struct LookupCacheBaseLockToken {
+protected:
+  // Protected constructor - only derived classes can construct
+  LookupCacheBaseLockToken() = default;
+};
+
+struct LookupCacheWriteLockToken : public LookupCacheBaseLockToken {
+private:
+  // Only constructible by GuestToHostMap
+  friend struct GuestToHostMap;
+  LookupCacheWriteLockToken(FEXCore::Utils::WritePriorityMutex::Mutex& Mutex)
+    : Lock {Mutex} {}
+  std::lock_guard<FEXCore::Utils::WritePriorityMutex::Mutex> Lock;
+};
+
+struct LookupCacheReadLockToken : public LookupCacheBaseLockToken {
+private:
+  // Only constructible by GuestToHostMap
+  friend struct GuestToHostMap;
+  LookupCacheReadLockToken(FEXCore::Utils::WritePriorityMutex::Mutex& Mutex)
+    : Lock {Mutex} {}
+  std::shared_lock<FEXCore::Utils::WritePriorityMutex::Mutex> Lock;
+};

 struct GuestToHostMap {
-  std::recursive_mutex WriteLock;
-
-  struct LockToken {
-    std::lock_guard<std::recursive_mutex> Lock;
-  };
+  FEXCore::Utils::WritePriorityMutex::Mutex Lock {};

  [[nodiscard]]
-  LockToken AcquireLock() {
-    return LockToken {std::lock_guard {WriteLock}};
+  LookupCacheWriteLockToken AcquireWriteLock() {
+    return LookupCacheWriteLockToken {Lock};
+  }
+
+  [[nodiscard]]
+  LookupCacheReadLockToken AcquireReadLock() {
+    return LookupCacheReadLockToken {Lock};
  }

  struct BlockLinkTag {
@@ -49,53 +76,72 @@ struct GuestToHostMap {
  // walking each block member and destructing objects.
  //
  // This makes `BlockLinks` look like a raw pointer that could memory leak, but since it is backed by the MBR, it won't.
-  std::pmr::monotonic_buffer_resource BlockLinks_mbr;
+  fextl::pmr::named_monotonic_page_buffer_resource BlockLinks_mbr;
  using BlockLinksMapType = std::pmr::map<BlockLinkTag, FEXCore::Context::BlockDelinkerFunc>;
  fextl::unique_ptr<std::pmr::polymorphic_allocator<std::byte>> BlockLinks_pma;
  BlockLinksMapType* BlockLinks;

-  fextl::robin_map<uint64_t, uint64_t> BlockList;
+  struct BlockEntry {
+    uint64_t HostCode;
+    fextl::vector<uint64_t> CodePages;
+  };
+
+  fextl::robin_map<uint64_t, BlockEntry> BlockList;

  fextl::map<uint64_t, fextl::vector<uint64_t>> CodePages;

  GuestToHostMap();

  // Adds to Guest -> Host code mapping
-  void AddBlockMapping(uint64_t Address, void* HostCode, const LockToken&) {
+  const BlockEntry& AddBlockMapping(uint64_t Address, const fextl::vector<uint64_t>& CodePages, void* HostCode, const LookupCacheWriteLockToken&) {
    // This may replace an existing mapping
    // NOTE: Generally no previous entry should exist, however there is one exception:
    //       If the backend updates the active thread's CodeBuffer, the new associated LookupCache
    //       may already contain the block address. Since is comparatively rare, we'll just leak
    //       one of the two blocks in this case.
-    BlockList[Address] = (uintptr_t)HostCode;
+    return BlockList.insert_or_assign(Address, BlockEntry {(uintptr_t)HostCode, CodePages}).first->second;
  }

-  std::optional<uintptr_t> FindBlock(uint64_t Address, const LockToken&) {
+  const BlockEntry* FindBlock(uint64_t Address, const LookupCacheReadLockToken&) {
    auto HostCode = BlockList.find(Address);
    if (HostCode == BlockList.end()) {
-      return std::nullopt;
+      return nullptr;
    }
-    return HostCode->second;
+    return &HostCode->second;
  }

-  bool Erase(FEXCore::Core::CpuStateFrame* Frame, uint64_t Address, const LockToken&) {
+  bool Erase(uint64_t Address, const LookupCacheWriteLockToken&) {
    // Sever any links to this block
    auto lower = BlockLinks->lower_bound({Address, nullptr});
    auto upper = BlockLinks->upper_bound({Address, reinterpret_cast<FEXCore::Context::ExitFunctionLinkData*>(UINTPTR_MAX)});
    for (auto it = lower; it != upper; it = BlockLinks->erase(it)) {
-      it->second(Frame, it->first.HostLink);
+      it->second(it->first.HostLink);
    }

    // Remove from BlockList
    return BlockList.erase(Address) != 0;
  }

+  void InvalidateRange(uint64_t Start, uint64_t Length) {
+    auto lk = AcquireWriteLock();
+
+    auto lower = CodePages.lower_bound(Start >> 12);
+    auto upper = CodePages.upper_bound((Start + Length - 1) >> 12);
+
+    for (auto it = lower; it != upper; it++) {
+      for (const auto& Entry : it->second) {
+        Erase(Entry, lk);
+      }
+    }
+    CodePages.erase(lower, upper);
+  }
+
  void AddBlockLink(uint64_t GuestDestination, FEXCore::Context::ExitFunctionLinkData* HostLink,
-                    const FEXCore::Context::BlockDelinkerFunc& delinker, const LockToken&) {
+                    const FEXCore::Context::BlockDelinkerFunc& delinker, const LookupCacheWriteLockToken&) {
    BlockLinks->insert({{GuestDestination, HostLink}, delinker});
  }

-  bool AddBlockExecutableRange(const fextl::set<uint64_t>& Addresses, uint64_t Start, uint64_t Length, const LockToken&) {
+  bool AddBlockExecutableRange(const std::ranges::input_range auto& Addresses, uint64_t Start, uint64_t Length, const LookupCacheWriteLockToken&) {
    bool rv = false;

    for (auto CurrentPage = Start >> 12, EndPage = (Start + Length - 1) >> 12; CurrentPage <= EndPage; CurrentPage++) {
@@ -107,7 +153,7 @@ struct GuestToHostMap {
    return rv;
  }

-  void ClearCache(const LockToken&);
+  void ClearCache(const LookupCacheWriteLockToken&);
 };

 class LookupCache {
@@ -122,122 +168,199 @@ public:

  // Swaps out the underlying GuestToHostMap and clears all associated caches.
  // This interface requires the previous CodeBuffer to be provided despite not using it. This ensures the shared write lock is still valid.
-  void ChangeGuestToHostMapping([[maybe_unused]] CPU::CodeBuffer& Prev, GuestToHostMap& NewMap) {
-    ClearThreadLocalCaches();
+  void ChangeGuestToHostMapping([[maybe_unused]] CPU::CodeBuffer& Prev, GuestToHostMap& NewMap, const LookupCacheWriteLockToken& lk) {
+    ClearThreadLocalCaches(lk);
    Shared = &NewMap;
  }

-  uintptr_t FindBlock(uint64_t Address) {
+  uintptr_t FindBlock(FEXCore::Core::InternalThreadState* Thread, uint64_t Address) {
    // Try L1, no lock needed
-    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
+    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1PointerMask];
    if (L1Entry.GuestCode == Address) {
      return L1Entry.HostCode;
    }

    // L2 and L3 need to be locked
-    auto lk = Shared->AcquireLock();
+    uintptr_t HostPtr {};
+    {
+      std::optional<FEXCore::SHMStats::AccumulationBlock<uint64_t>> LockTime(
+        Thread->ThreadStats ? &Thread->ThreadStats->AccumulatedCacheReadLockTime : nullptr);
+      auto lk = Shared->AcquireReadLock();
+      LockTime.reset();

-    // Try L2
-    const auto PageIndex = (Address & (VirtualMemSize - 1)) >> 12;
-    const auto PageOffset = Address & (0x0FFF);
+      if (!DisableL2Cache()) {
+        // Try L2
+        const auto PageIndex = (Address & (VirtualMemSize - 1)) >> 12;
+        const auto PageOffset = Address & (0x0FFF);

-    const auto Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
-    auto LocalPagePointer = Pointers[PageIndex];
+        const auto Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
+        auto LocalPagePointer = Pointers[PageIndex];

-    // Do we a page pointer for this address?
-    if (LocalPagePointer) {
-      // Find there pointer for the address in the blocks
-      auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
+        // Do we a page pointer for this address?
+        if (LocalPagePointer) {
+          // Find there pointer for the address in the blocks
+          auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);

-      if (BlockPointers[PageOffset].GuestCode == Address) {
-        L1Entry.GuestCode = Address;
-        L1Entry.HostCode = BlockPointers[PageOffset].HostCode;
-        return L1Entry.HostCode;
+          if (BlockPointers[PageOffset].GuestCode == Address) {
+            L1Entry.GuestCode = Address;
+            L1Entry.HostCode = BlockPointers[PageOffset].HostCode;
+            HostPtr = L1Entry.HostCode;
+          }
+        }
+      }
+
+      if (!HostPtr) {
+        // Try L3
+        auto Entry = Shared->FindBlock(Address, lk);
+        if (Entry) {
+          CacheBlockMapping(Address, *Entry, false, lk);
+          HostPtr = Entry->HostCode;
+        }
      }
    }

-    // Try L3
-    auto HostCode = Shared->FindBlock(Address, lk);
-    if (HostCode) {
-      CacheBlockMapping(Address, HostCode.value());
-      return HostCode.value();
+    if (HostPtr && DynamicL1Cache()) {
+      UpdateDynamicL1Stats(Thread);
    }

-    // Failed to find
-    return 0;
+    FEXCORE_PROFILE_INSTANT_INCREMENT(Thread, AccumulatedCacheMissCount, 1);
+
+    return HostPtr;
+  }
+
+  void UpdateDynamicL1Stats(FEXCore::Core::InternalThreadState* Thread) {
+    // If host pointer was found in L2 or L3, then add it to the counter.
+    // Keeping track not L1 misses, but specifically L2/L3 hits.
+    ++L2L3CacheHits;
+
+    const auto CurrentTime = std::chrono::system_clock::now();
+    const auto Period = CurrentTime - LastPeriod;
+    if (Period >= SamplePeriod) {
+      // If larger than the sample period then check if we need to increase L1 cache size.
+      const double AveragePerSecond = static_cast<double>(L2L3CacheHits) /
+                                      static_cast<double>(std::chrono::duration_cast<std::chrono::milliseconds>(Period).count()) * 1000.0;
+
+      if (AveragePerSecond >= DynamicL1CacheIncreaseCountHeuristic()) {
+        if (CurrentL1Entries < MAX_L1_ENTRIES) {
+          CurrentL1Entries <<= 1;
+          L1PointerMask = CurrentL1Entries - 1;
+
+          // Update the thread's L1 pointer mask to increase how much cache it uses.
+          // Since we're in C-code, this is safe to update here.
+          Thread->CurrentFrame->State.L1Mask = GetScaledL1PointerMask();
+        }
+      } else if (AveragePerSecond < DynamicL1CacheDecreaseCountHeuristic()) {
+        if (CurrentL1Entries > MIN_L1_ENTRIES) {
+          CurrentL1Entries >>= 1;
+          L1PointerMask = CurrentL1Entries - 1;
+
+          // Madvise the entries that we are dropping. Gives the memory back to the OS.
+          LookupCacheEntry* FirstZeroL1Entry = &reinterpret_cast<LookupCacheEntry*>(L1Pointer)[CurrentL1Entries];
+          size_t ZeroMemorySize = (MAX_L1_ENTRIES - CurrentL1Entries) * sizeof(LookupCacheEntry);
+          FEXCore::Allocator::VirtualDontNeed(FirstZeroL1Entry, ZeroMemorySize, false);
+
+          // Update the thread's L1 pointer mask to increase how much cache it uses.
+          // Since we're in C-code, this is safe to update here.
+          Thread->CurrentFrame->State.L1Mask = GetScaledL1PointerMask();
+        }
+      }
+
+      // Update Last period to start again.
+      LastPeriod = CurrentTime;
+      L2L3CacheHits = 0;
+    }
  }

  GuestToHostMap* Shared = nullptr;

  // Appends a list of Block {Address} to CodePages [Start, Start + Length)
  // Returns true if new pages are marked as containing code
-  bool AddBlockExecutableRange(const fextl::set<uint64_t>& Addresses, uint64_t Start, uint64_t Length) {
-    auto lk = Shared->AcquireLock();
+  bool AddBlockExecutableRange(FEXCore::Core::InternalThreadState* Thread, const fextl::set<uint64_t>& Addresses, uint64_t Start, uint64_t Length) {
+    std::optional<FEXCore::SHMStats::AccumulationBlock<uint64_t>> LockTime(
+      Thread->ThreadStats ? &Thread->ThreadStats->AccumulatedCacheWriteLockTime : nullptr);
+    auto lk = Shared->AcquireWriteLock();
+    LockTime.reset();
+
    return Shared->AddBlockExecutableRange(Addresses, Start, Length, lk);
  }

  // Adds to Guest -> Host code mapping
-  void AddBlockMapping(uint64_t Address, void* HostCode) {
-    auto lk = Shared->AcquireLock();
+  void AddBlockMapping(FEXCore::Core::InternalThreadState* Thread, uint64_t Address, const fextl::vector<uint64_t>& CodePages, void* HostCode) {
+    std::optional<FEXCore::SHMStats::AccumulationBlock<uint64_t>> LockTime(
+      Thread->ThreadStats ? &Thread->ThreadStats->AccumulatedCacheWriteLockTime : nullptr);
+    auto lk = Shared->AcquireWriteLock();
+    LockTime.reset();

-    Shared->AddBlockMapping(Address, HostCode, lk);
+    const auto& Entry = Shared->AddBlockMapping(Address, CodePages, HostCode, lk);

    // There is no need to update L1 or L2, they will get updated on first lookup
    // However, adding to L1 here increases performance
-    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
-    L1Entry.GuestCode = Address;
-    L1Entry.HostCode = (uintptr_t)HostCode;
+    CacheBlockMapping(Address, Entry, true, lk);
  }

-  // NOTE: It's the caller's responsibility to call Erase() for all other
-  //       GuestToHostMaps that share the same LookupCache. Otherwise, the
-  //       L1/L2 caches will contain stale references to deallocated memory.
-  bool Erase(FEXCore::Core::CpuStateFrame* Frame, uint64_t Address) {
-    auto lk = Shared->AcquireLock();
-
-    bool ErasedAny = Shared->Erase(Frame, Address, lk);
-
+  // Invalidates L1/L2 for a given guest block
+  void InvalidateCache(uint64_t Address, const LookupCacheWriteLockToken& lk) {
    // Do L1
-    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
+    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1PointerMask];
    if (L1Entry.GuestCode == Address) {
      L1Entry.GuestCode = 0;
-      ErasedAny = true;
      // Leave L1Entry.HostCode as is, so that concurrent lookups won't read a null pointer
      // This is a soft guarantee for cross thread invalidation, as atomics are not used
      // and it hasn't been thoroughly tested
    }

-    // Do full map
-    Address = Address & (VirtualMemSize - 1);
-    uint64_t PageOffset = Address & (0x0FFF);
-    Address >>= 12;
+    if (!DisableL2Cache()) {
+      // Do full map
+      Address = Address & (VirtualMemSize - 1);
+      uint64_t PageOffset = Address & (0x0FFF);
+      Address >>= 12;

-    uintptr_t* Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
-    uint64_t LocalPagePointer = Pointers[Address];
-    if (!LocalPagePointer) {
-      // Page for this code didn't even exist, nothing to do
-      return ErasedAny;
+      uintptr_t* Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
+      uint64_t LocalPagePointer = Pointers[Address];
+      if (!LocalPagePointer) {
+        // Page for this code didn't even exist, nothing to do
+        return;
+      }
+
+      // Page exists, just set the offset to zero
+      auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
+      BlockPointers[PageOffset].GuestCode = 0;
+      BlockPointers[PageOffset].HostCode = 0;
    }
-
-    // Page exists, just set the offset to zero
-    auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
-    BlockPointers[PageOffset].GuestCode = 0;
-    BlockPointers[PageOffset].HostCode = 0;
-    return true;
  }

-  void AddBlockLink(uint64_t GuestDestination, FEXCore::Context::ExitFunctionLinkData* HostLink, const FEXCore::Context::BlockDelinkerFunc& delinker) {
-    auto lk = Shared->AcquireLock();
+  // Invalidates all L1/L2 entries for all guest block that intersect the given range
+  bool InvalidateCacheRange(uint64_t Start, uint64_t Length) {
+    auto lk = Shared->AcquireWriteLock();
+
+    auto lower = CachedCodePages.lower_bound(Start >> 12);
+    auto upper = CachedCodePages.upper_bound((Start + Length - 1) >> 12);
+
+    for (auto it = lower; it != upper; it++) {
+      for (const auto& Entry : it->second) {
+        InvalidateCache(Entry, lk);
+      }
+    }
+    bool ret = upper != lower;
+    CachedCodePages.erase(lower, upper);
+    return ret;
+  }
+
+  void AddBlockLink(uint64_t GuestDestination, FEXCore::Context::ExitFunctionLinkData* HostLink,
+                    const FEXCore::Context::BlockDelinkerFunc& delinker, const LookupCacheWriteLockToken& lk) {
    Shared->AddBlockLink(GuestDestination, HostLink, delinker, lk);
  }

-  void ClearCache();
-  void ClearL2Cache();
-  void ClearThreadLocalCaches();
+  void ClearCache(const LookupCacheWriteLockToken&);
+  void ClearL2Cache(const LookupCacheBaseLockToken&);
+  void ClearThreadLocalCaches(const LookupCacheWriteLockToken&);

  uintptr_t GetL1Pointer() const {
    return L1Pointer;
  }
+  uintptr_t GetScaledL1PointerMask() const {
+    return L1PointerMask << FEXCore::ilog2(sizeof(LookupCache::LookupCacheEntry));
+  }
  uintptr_t GetPagePointer() const {
    return PagePointer;
  }
@@ -245,9 +368,6 @@ public:
    return VirtualMemSize;
  }

-  constexpr static size_t L1_ENTRIES = 1 * 1024 * 1024; // Must be a power of 2
-  constexpr static size_t L1_ENTRIES_MASK = L1_ENTRIES - 1;
-
  // This needs to be taken before reads or writes to L2, L3, CodePages,
  // and before writes to L1. Concurrent access from a thread that this LookupCache doesn't belong to
  // may only happen during cross thread invalidation (::Erase).
@@ -255,45 +375,52 @@ public:
  // Some care is taken so that L1 lookups can be done without locks, and even tearing is unlikely to lead to a crash.
  // This approach has not been fully vetted yet.
  // Also note that L1 lookups might be inlined in the JIT Dispatcher and/or block ends.
-  auto AcquireLock() {
-    return Shared->AcquireLock();
+  auto AcquireWriteLock() {
+    return Shared->AcquireWriteLock();
  }

 private:
-  void CacheBlockMapping(uint64_t Address, uintptr_t HostCode) {
-    // Do L1
-    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
-    L1Entry.GuestCode = Address;
-    L1Entry.HostCode = HostCode;
-
-    // Do ful map
-    auto FullAddress = Address;
-    Address = Address & (VirtualMemSize - 1);
-
-    uint64_t PageOffset = Address & (0x0FFF);
-    Address >>= 12;
-    uintptr_t* Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
-    uint64_t LocalPagePointer = Pointers[Address];
-    if (!LocalPagePointer) {
-      // We don't have a page pointer for this address
-      // Allocate one now if we can
-      uintptr_t NewPageBacking = AllocateBackingForPage();
-      if (!NewPageBacking) {
-        // Couldn't allocate, clear L2 and retry
-        ClearL2Cache();
-        CacheBlockMapping(Address, HostCode);
-        return;
-      }
-      Pointers[Address] = NewPageBacking;
-      LocalPagePointer = NewPageBacking;
+  void CacheBlockMapping(uint64_t Address, const GuestToHostMap::BlockEntry& Entry, bool L1Only, const LookupCacheBaseLockToken& lk) {
+    for (const auto& CodePage : Entry.CodePages) {
+      CachedCodePages[CodePage >> 12].insert(Address);
    }

-    // Add the new pointer to the page block
-    auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
+    // Do L1
+    auto& L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1PointerMask];
+    L1Entry.GuestCode = Address;
+    L1Entry.HostCode = Entry.HostCode;

-    // This silently replaces existing mappings
-    BlockPointers[PageOffset].GuestCode = FullAddress;
-    BlockPointers[PageOffset].HostCode = HostCode;
+    if (!DisableL2Cache() && !L1Only) {
+      // Do ful map
+      auto FullAddress = Address;
+      Address = Address & (VirtualMemSize - 1);
+
+      uint64_t PageOffset = Address & (0x0FFF);
+      Address >>= 12;
+
+      uintptr_t* Pointers = reinterpret_cast<uintptr_t*>(PagePointer);
+      uint64_t LocalPagePointer = Pointers[Address];
+      if (!LocalPagePointer) {
+        // We don't have a page pointer for this address
+        // Allocate one now if we can
+        uintptr_t NewPageBacking = AllocateBackingForPage();
+        if (!NewPageBacking) {
+          // Couldn't allocate, clear L2 and retry
+          ClearL2Cache(lk);
+          CacheBlockMapping(FullAddress, Entry, false, lk);
+          return;
+        }
+        Pointers[Address] = NewPageBacking;
+        LocalPagePointer = NewPageBacking;
+      }
+
+      // Add the new pointer to the page block
+      auto BlockPointers = reinterpret_cast<LookupCacheEntry*>(LocalPagePointer);
+
+      // This silently replaces existing mappings
+      BlockPointers[PageOffset].GuestCode = FullAddress;
+      BlockPointers[PageOffset].HostCode = Entry.HostCode;
+    }
  }

  uintptr_t AllocateBackingForPage() {
@@ -310,19 +437,38 @@ private:
    return PageMemory + NewBase;
  }

+  // Maps from a page index to all blocks in the page that have at some point been fetched into L1/L2
+  fextl::map<uint64_t, fextl::robin_set<uint64_t>> CachedCodePages;
+
  uintptr_t PagePointer;
  uintptr_t PageMemory;
  uintptr_t L1Pointer;
+  uintptr_t L1PointerMask;

  size_t TotalCacheSize;

+  // Start with 8k entries in L1 to give 128KB of L1 cache to each thread.
+  // Max out at 1 million entries to give each thread 16MB of L1 cache maximum.
+  constexpr static size_t MIN_L1_ENTRIES = 8 * 1024;        // Must be a power of 2
+  constexpr static size_t MAX_L1_ENTRIES = 1 * 1024 * 1024; // Must be a power of 2
+
  constexpr static size_t CODE_SIZE = 128 * 1024 * 1024;
-  constexpr static size_t SIZE_PER_PAGE = 4096 * sizeof(LookupCacheEntry);
-  constexpr static size_t L1_SIZE = L1_ENTRIES * sizeof(LookupCacheEntry);
+  constexpr static size_t SIZE_PER_PAGE = FEXCore::Utils::FEX_PAGE_SIZE * sizeof(LookupCacheEntry);
+  constexpr static size_t MAX_L1_SIZE = MAX_L1_ENTRIES * sizeof(LookupCacheEntry);

  size_t AllocateOffset {};

  FEXCore::Context::ContextImpl* ctx;
  uint64_t VirtualMemSize {};
+
+  size_t CurrentL1Entries = MIN_L1_ENTRIES;
+  uint64_t L2L3CacheHits {};
+  std::chrono::time_point<std::chrono::system_clock> LastPeriod {};
+  constexpr static std::chrono::seconds SamplePeriod {1};
+  FEX_CONFIG_OPT(DynamicL1CacheIncreaseCountHeuristic, DYNAMICL1CACHEINCREASECOUNTHEURISTIC);
+  FEX_CONFIG_OPT(DynamicL1CacheDecreaseCountHeuristic, DYNAMICL1CACHEDECREASECOUNTHEURISTIC);
+
+  FEX_CONFIG_OPT(DynamicL1Cache, DYNAMICL1CACHE);
+  FEX_CONFIG_OPT(DisableL2Cache, DISABLEL2CACHE);
 };
 } // namespace FEXCore
@@ -27,6 +27,42 @@
 #include <xxhash.h>

 namespace FEXCore::IR {
+enum class VectorCompareType {
+  // SSE comparisons.
+  EQ_OQ = 0,
+  LT_OS = 1,
+  LE_OS = 2,
+  UNORD_Q = 3,
+  NEQ_UQ = 4,
+  NLT_US = 5,
+  NLE_US = 6,
+  ORD_Q = 7,
+  // AVX-only comparisons.
+  EQ_UQ = 8,
+  NGE_US = 9,
+  NGT_US = 10,
+  FALSE_OQ = 11,
+  NEQ_OQ = 12,
+  GE_OS = 13,
+  GT_OS = 14,
+  TRUE_UQ = 15,
+  EQ_OS = 16,
+  LT_OQ = 17,
+  LE_OQ = 18,
+  UNORD_S = 19,
+  NEQ_US = 20,
+  NLT_UQ = 21,
+  NLE_UQ = 22,
+  ORD_S = 23,
+  EQ_US = 24,
+  NGE_UQ = 25,
+  NGT_UQ = 26,
+  FALSE_OS = 27,
+  NEQ_OS = 28,
+  GE_OQ = 29,
+  GT_OQ = 30,
+  TRUE_US = 31,
+};

 enum class MemoryAccessType {
  // Choose TSO or Non-TSO depending on access type
@@ -139,27 +175,28 @@ public:
    FlushRegisterCache();
    return _Jump(_TargetBlock);
  }
-  IRPair<IROp_CondJump> CondJump(Ref _Cmp1, Ref _Cmp2, Ref _TrueBlock, Ref _FalseBlock, CondClassType _Cond = {COND_NEQ},
+  IRPair<IROp_CondJump> CondJump(Ref _Cmp1, Ref _Cmp2, Ref _TrueBlock, Ref _FalseBlock, CondClass _Cond = CondClass::NEQ,
                                 IR::OpSize _CompareSize = OpSize::iInvalid) {
    FlushRegisterCache();
    return _CondJump(_Cmp1, _Cmp2, _TrueBlock, _FalseBlock, _Cond, _CompareSize);
  }
-  IRPair<IROp_CondJump> CondJump(Ref ssa0, CondClassType cond = {COND_NEQ}) {
+  IRPair<IROp_CondJump> CondJump(Ref ssa0, CondClass cond = CondClass::NEQ) {
    FlushRegisterCache();
    return _CondJump(ssa0, cond);
  }
-  IRPair<IROp_CondJump> CondJump(Ref ssa0, Ref ssa1, Ref ssa2, CondClassType cond = {COND_NEQ}) {
+  IRPair<IROp_CondJump> CondJump(Ref ssa0, Ref ssa1, Ref ssa2, CondClass cond = CondClass::NEQ) {
    FlushRegisterCache();
    return _CondJump(ssa0, ssa1, ssa2, cond);
  }
-  IRPair<IROp_CondJump> CondJumpNZCV(CondClassType Cond) {
+  IRPair<IROp_CondJump> CondJumpNZCV(CondClass Cond) {
    FlushRegisterCache();
    return _CondJump(InvalidNode, InvalidNode, InvalidNode, InvalidNode, Cond, OpSize::iInvalid, true);
  }
  IRPair<IROp_CondJump> CondJumpBit(Ref Src, unsigned Bit, bool Set) {
    FlushRegisterCache();
    auto InlineConst = _InlineConstant(Bit);
-    return _CondJump(Src, InlineConst, InvalidNode, InvalidNode, {Set ? COND_TSTNZ : COND_TSTZ}, OpSize::iInvalid, false);
+    auto Cond = Set ? CondClass::TSTNZ : CondClass::TSTZ;
+    return _CondJump(Src, InlineConst, InvalidNode, InvalidNode, Cond, OpSize::iInvalid, false);
  }
  IRPair<IROp_ExitFunction> ExitFunction(Ref NewRIP, BranchHint Hint = BranchHint::None) {
    FlushRegisterCache();
@@ -251,7 +288,7 @@ public:
    auto ExitBlock = CreateNewCodeBlockAfter(BackwardBlock);

    auto DF = GetRFLAG(X86State::RFLAG_DF_RAW_LOC);
-    CondJump(DF, Zero, ForwardBlock, BackwardBlock, {COND_EQ});
+    CondJump(DF, Zero, ForwardBlock, BackwardBlock, CondClass::EQ);

    for (auto D = 0; D < 2; ++D) {
      SetCurrentCodeBlock(D ? BackwardBlock : ForwardBlock);
@@ -318,7 +355,6 @@ public:

  void UnhandledOp(OpcodeArgs);
  void MOVGPROp(OpcodeArgs, uint32_t SrcIndex);
-  void MOVGPRImmediate(OpcodeArgs);
  void MOVGPRNTOp(OpcodeArgs);
  void MOVVectorAlignedOp(OpcodeArgs);
  void MOVVectorUnalignedOp(OpcodeArgs);
@@ -356,6 +392,7 @@ public:
  void CALLFARIndirectOp(OpcodeArgs);
  void RETFARIndirectOp(OpcodeArgs);
  void TESTOp(OpcodeArgs, uint32_t SrcIndex);
+  void ARPLOp(OpcodeArgs);
  void MOVSXDOp(OpcodeArgs);
  void MOVSXOp(OpcodeArgs);
  void MOVZXOp(OpcodeArgs);
@@ -372,7 +409,7 @@ public:
  void CMOVOp(OpcodeArgs);
  void CPUIDOp(OpcodeArgs);
  void XGetBVOp(OpcodeArgs);
-  uint32_t LoadConstantShift(X86Tables::DecodedOp Op, bool Is1Bit);
+  uint32_t GetConstantShift(X86Tables::DecodedOp Op, bool Is1Bit);
  void SHLOp(OpcodeArgs);
  void SHLImmediateOp(OpcodeArgs, bool SHL1Bit);
  void SHROp(OpcodeArgs);
@@ -564,7 +601,7 @@ public:
  template<IR::OpSize DstElementSize, IR::OpSize SrcElementSize>
  void AVXInsertScalar_CVT_Float_To_Float(OpcodeArgs);

-  RoundType TranslateRoundType(uint8_t Mode);
+  RoundMode TranslateRoundType(uint8_t Mode);

  template<IR::OpSize ElementSize>
  void InsertScalarRound(OpcodeArgs);
@@ -758,7 +795,6 @@ public:
  void X87FXTRACT(OpcodeArgs);
  void X87FYL2X(OpcodeArgs, bool IsFYL2XP1);
  void X87LDENV(OpcodeArgs);
-  void X87LDSW(OpcodeArgs);
  void X87ModifySTP(OpcodeArgs, bool Inc);
  void X87OpHelper(OpcodeArgs, FEXCore::IR::IROps IROp, bool ZeroC2);

@@ -907,6 +943,10 @@ public:
  void VPCLMULQDQOp(OpcodeArgs);

  void CRC32(OpcodeArgs);
+  void Extrq_imm(OpcodeArgs);
+  void Insertq_imm(OpcodeArgs);
+  void Extrq(OpcodeArgs);
+  void Insertq(OpcodeArgs);

  void BreakOp(OpcodeArgs, FEXCore::IR::BreakDefinition BreakDefinition);
  void UnimplementedOp(OpcodeArgs);
@@ -985,7 +1025,6 @@ public:
  void AVX128_VPSIGN(OpcodeArgs, IR::OpSize ElementSize);
  void AVX128_UCOMISx(OpcodeArgs, IR::OpSize ElementSize);
  void AVX128_VectorScalarInsertALU(OpcodeArgs, FEXCore::IR::IROps IROp, IR::OpSize ElementSize);
-  Ref AVX128_VFCMPImpl(IR::OpSize ElementSize, Ref Src1, Ref Src2, uint8_t CompType);
  void AVX128_VFCMP(OpcodeArgs, IR::OpSize ElementSize);
  void AVX128_InsertScalarFCMP(OpcodeArgs, IR::OpSize ElementSize);
  void AVX128_MOVBetweenGPR_FPR(OpcodeArgs);
@@ -1004,9 +1043,7 @@ public:
  void AVX128_VINSERT(OpcodeArgs);
  void AVX128_VINSERTPS(OpcodeArgs);

-  Ref AVX128_PHSUBImpl(Ref Src1, Ref Src2, size_t ElementSize);
  void AVX128_VPHSUB(OpcodeArgs, IR::OpSize ElementSize);
-
  void AVX128_VPHSUBSW(OpcodeArgs);

  void AVX128_VADDSUBP(OpcodeArgs, IR::OpSize ElementSize);
@@ -1098,8 +1135,8 @@ public:
  void AVX128_VFMAScalarImpl(OpcodeArgs, IROps IROp, uint8_t Src1Idx, uint8_t Src2Idx, uint8_t AddendIdx);
  void AVX128_VFMAddSubImpl(OpcodeArgs, bool AddSub, uint8_t Src1Idx, uint8_t Src2Idx, uint8_t AddendIdx);

-  RefPair AVX128_VPGatherQPSImpl(Ref Dest, Ref Mask, RefVSIB VSIB);
-  RefPair AVX128_VPGatherImpl(OpSize Size, OpSize ElementLoadSize, OpSize AddrElementSize, RefPair Dest, RefPair Mask, RefVSIB VSIB);
+  RefPair AVX128_VPGatherQPSImpl(OpcodeArgs, Ref Dest, Ref Mask, RefVSIB VSIB);
+  RefPair AVX128_VPGatherImpl(OpcodeArgs, OpSize Size, OpSize ElementLoadSize, OpSize AddrElementSize, RefPair Dest, RefPair Mask, RefVSIB VSIB);

  void AVX128_VPGATHER(OpcodeArgs, OpSize AddrElementSize);

@@ -1109,8 +1146,8 @@ public:
  // End of AVX 128-bit implementation

  // AVX 256-bit operations
-  void StoreResult_WithAVXInsert(VectorOpType Type, FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, Ref Value,
-                                 IR::OpSize Align, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+  void StoreResult_WithAVXInsert(VectorOpType Type, RegClass Class, FEXCore::X86Tables::DecodedOp Op, Ref Value,
+                                 IR::OpSize Align = IR::OpSize::iInvalid, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
    if (Op->Dest.IsGPR() && Op->Dest.Data.GPR.GPR >= X86State::REG_XMM_0 && Op->Dest.Data.GPR.GPR <= X86State::REG_XMM_15 &&
        GetGuestVectorLength() == OpSize::i256Bit && Type == VectorOpType::SSE) {
      const auto gpr = Op->Dest.Data.GPR.GPR;
@@ -1161,7 +1198,7 @@ public:
    }
  }

-  void StoreContextHelper(IR::OpSize Size, RegisterClassType Class, Ref Value, uint32_t Offset) {
+  void StoreContextHelper(IR::OpSize Size, RegClass Class, Ref Value, uint32_t Offset) {
    // For i128Bit, we won't see a normal Constant to inline, but as a special
    // case we can replace with a 2x64-bit store which can use inline zeroes.
    if (Size == OpSize::i128Bit) {
@@ -1173,7 +1210,7 @@ public:

        if (Const->Constant == IR::NamedVectorConstant::NAMED_VECTOR_ZERO) {
          Ref Zero = _Constant(0);
-          Ref STP = _StoreContextPair(IR::OpSize::i64Bit, GPRClass, Zero, Zero, Offset);
+          Ref STP = _StoreContextPair(IR::OpSize::i64Bit, RegClass::GPR, Zero, Zero, Offset);

          // XXX: This works around InlineConstant not having an associated
          // register class, else we'd just do InlineConstant above.
@@ -1229,16 +1266,16 @@ public:

      if (Index >= GPR0Index && Index <= GPR15Index) {
        Ref R = _StoreRegister(Value, GPRSize);
-        R->Reg = PhysicalRegister(GPRFixedClass, Index - GPR0Index).Raw;
+        R->Reg = PhysicalRegister(RegClass::GPRFixed, Index - GPR0Index).Raw;
      } else if (Index == PFIndex) {
        _StorePF(Value, GPRSize);
      } else if (Index == AFIndex) {
        _StoreAF(Value, GPRSize);
      } else if (Index >= FPR0Index && Index <= FPR15Index) {
        Ref R = _StoreRegister(Value, VectorSize);
-        R->Reg = PhysicalRegister(FPRFixedClass, Index - FPR0Index).Raw;
+        R->Reg = PhysicalRegister(RegClass::FPRFixed, Index - FPR0Index).Raw;
      } else if (Index == DFIndex) {
-        _StoreContext(OpSize::i8Bit, GPRClass, Value, offsetof(Core::CPUState, flags[X86State::RFLAG_DF_RAW_LOC]));
+        _StoreContextGPR(OpSize::i8Bit, Value, offsetof(Core::CPUState, flags[X86State::RFLAG_DF_RAW_LOC]));
      } else {
        bool Partial = RegCache.Partial & (1ull << Index);
        auto Size = Partial ? OpSize::i64Bit : CacheIndexToOpSize(Index);
@@ -1263,7 +1300,7 @@ public:
          StoreContextHelper(Size, Class, Value, Offset);
          // If Partial and MMX register, then we need to store all 1s in bits 64-80
          if (Partial && Index >= MM0Index && Index <= MM7Index) {
-            _StoreContext(OpSize::i16Bit, IR::GPRClass, Constant(0xFFFF), Offset + 8);
+            _StoreContextGPR(OpSize::i16Bit, Constant(0xFFFF), Offset + 8);
          }
        }
      }
@@ -1543,7 +1580,7 @@ private:
  [[nodiscard]]
  static bool IsOperandMem(const X86Tables::DecodedOperand& Operand, bool Load) {
    // Literals are immediates as sources but memory addresses as destinations.
-    return !(Load && Operand.IsLiteral()) && !Operand.IsGPR();
+    return !(Load && (Operand.IsLiteral() || Operand.IsLiteralRelocation())) && !Operand.IsGPR();
  }

  [[nodiscard]]
@@ -1553,23 +1590,63 @@ private:

  AddressMode DecodeAddress(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, MemoryAccessType AccessType, bool IsLoad);

-  Ref LoadSource(RegisterClassType Class, const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, uint32_t Flags,
+  Ref LoadSource(RegClass Class, const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, uint32_t Flags,
                 const LoadSourceOptions& Options = {});
-  Ref LoadSource_WithOpSize(RegisterClassType Class, const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand,
-                            IR::OpSize OpSize, uint32_t Flags, const LoadSourceOptions& Options = {});
-  void StoreResult_WithOpSize(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op,
-                              const FEXCore::X86Tables::DecodedOperand& Operand, const Ref Src, IR::OpSize OpSize, IR::OpSize Align,
-                              MemoryAccessType AccessType = MemoryAccessType::DEFAULT);
-  void StoreResult(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, const FEXCore::X86Tables::DecodedOperand& Operand,
-                   const Ref Src, IR::OpSize Align, MemoryAccessType AccessType = MemoryAccessType::DEFAULT);
-  void StoreResult(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, const Ref Src, IR::OpSize Align,
+  Ref LoadSourceGPR(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, uint32_t Flags,
+                    const LoadSourceOptions& Options = {}) {
+    return LoadSource(RegClass::GPR, Op, Operand, Flags, Options);
+  }
+  Ref LoadSourceFPR(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, uint32_t Flags,
+                    const LoadSourceOptions& Options = {}) {
+    return LoadSource(RegClass::FPR, Op, Operand, Flags, Options);
+  }
+
+  Ref LoadSource_WithOpSize(RegClass Class, const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, IR::OpSize OpSize,
+                            uint32_t Flags, const LoadSourceOptions& Options = {});
+  Ref LoadSourceGPR_WithOpSize(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, IR::OpSize OpSize, uint32_t Flags,
+                               const LoadSourceOptions& Options = {}) {
+    return LoadSource_WithOpSize(RegClass::GPR, Op, Operand, OpSize, Flags, Options);
+  }
+  Ref LoadSourceFPR_WithOpSize(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, IR::OpSize OpSize, uint32_t Flags,
+                               const LoadSourceOptions& Options = {}) {
+    return LoadSource_WithOpSize(RegClass::FPR, Op, Operand, OpSize, Flags, Options);
+  }
+
+  void StoreResult_WithOpSize(RegClass Class, X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, IR::OpSize OpSize,
+                              IR::OpSize Align, MemoryAccessType AccessType = MemoryAccessType::DEFAULT);
+  void StoreResultGPR_WithOpSize(X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, IR::OpSize OpSize,
+                                 IR::OpSize Align = IR::OpSize::iInvalid, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult_WithOpSize(RegClass::GPR, Op, Operand, Src, OpSize, Align, AccessType);
+  }
+  void StoreResultFPR_WithOpSize(X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, IR::OpSize OpSize,
+                                 IR::OpSize Align = IR::OpSize::iInvalid, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult_WithOpSize(RegClass::FPR, Op, Operand, Src, OpSize, Align, AccessType);
+  }
+
+  void StoreResult(RegClass Class, X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, OpSize Align,
                   MemoryAccessType AccessType = MemoryAccessType::DEFAULT);
+  void StoreResultGPR(X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, OpSize Align = OpSize::iInvalid,
+                      MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult(RegClass::GPR, Op, Operand, Src, Align, AccessType);
+  }
+  void StoreResultFPR(X86Tables::DecodedOp Op, const X86Tables::DecodedOperand& Operand, Ref Src, OpSize Align = OpSize::iInvalid,
+                      MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult(RegClass::FPR, Op, Operand, Src, Align, AccessType);
+  }
+
+  void StoreResult(RegClass Class, X86Tables::DecodedOp Op, Ref Src, OpSize Align, MemoryAccessType AccessType = MemoryAccessType::DEFAULT);
+  void StoreResultGPR(X86Tables::DecodedOp Op, Ref Src, OpSize Align = OpSize::iInvalid, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult(RegClass::GPR, Op, Src, Align, AccessType);
+  }
+  void StoreResultFPR(X86Tables::DecodedOp Op, Ref Src, OpSize Align = OpSize::iInvalid, MemoryAccessType AccessType = MemoryAccessType::DEFAULT) {
+    StoreResult(RegClass::FPR, Op, Src, Align, AccessType);
+  }

  // In several instances, it's desirable to get a base address with the segment offset
  // applied to it. This pulls all the common-case appending into a single set of functions.
  [[nodiscard]]
  Ref MakeSegmentAddress(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, IR::OpSize OpSize) {
-    Ref Mem = LoadSource_WithOpSize(GPRClass, Op, Operand, OpSize, Op->Flags, {.LoadData = false});
+    Ref Mem = LoadSourceGPR_WithOpSize(Op, Operand, OpSize, Op->Flags, {.LoadData = false});
    return AppendSegmentOffset(Mem, Op->Flags);
  }
  [[nodiscard]]
@@ -1614,6 +1691,9 @@ private:
    return IR::SizeToOpSize(GetSrcSize(Op));
  }

+  [[nodiscard]]
+  IR::OpSize GetStringOpSize(X86Tables::DecodedOp Op) const;
+
  // Set flag tracking to prepare for an operation that directly writes NZCV.
  void HandleNZCVWrite() {
    CachedNZCV = nullptr;
@@ -1805,14 +1885,15 @@ private:
      // For DF, we need to transform 0/1 into 1/-1
      StoreDF(_SubShift(OpSize::i64Bit, Constant(1), Value, ShiftType::LSL, 1));
    } else if (BitOffset == FEXCore::X86State::RFLAG_TF_RAW_LOC) {
-      auto PackedTF = _LoadContext(OpSize::i8Bit, GPRClass, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
+      auto PackedTF = _LoadContextGPR(OpSize::i8Bit, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
      // An exception should still be raised after an instruction that unsets TF, leave the unblocked bit set but unset
      // the TF bit to cause such behaviour. The handling code at the start of the next block will then unset the
      // unblocked bit before raising the exception.
-      auto NewPackedTF = _Select(FEXCore::IR::COND_EQ, Value, Constant(0), _And(OpSize::i32Bit, PackedTF, Constant(~1)), Constant(1));
-      _StoreContext(OpSize::i8Bit, GPRClass, NewPackedTF, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
+      auto NewPackedTF =
+        _Select(OpSize::i64Bit, OpSize::i64Bit, CondClass::EQ, Value, Constant(0), _And(OpSize::i32Bit, PackedTF, Constant(~1)), Constant(1));
+      _StoreContextGPR(OpSize::i8Bit, NewPackedTF, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
    } else {
-      _StoreContext(OpSize::i8Bit, GPRClass, Value, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
+      _StoreContextGPR(OpSize::i8Bit, Value, offsetof(FEXCore::Core::CPUState, flags[BitOffset]));
    }
  }

@@ -1838,12 +1919,12 @@ private:
  }

  [[nodiscard]]
-  static CondClassType CondForNZCVBit(unsigned BitOffset, bool Invert) {
+  static CondClass CondForNZCVBit(unsigned BitOffset, bool Invert) {
    switch (BitOffset) {
-    case X86State::RFLAG_SF_RAW_LOC: return {Invert ? COND_PL : COND_MI};
-    case X86State::RFLAG_ZF_RAW_LOC: return {Invert ? COND_NEQ : COND_EQ};
-    case X86State::RFLAG_CF_RAW_LOC: return {Invert ? COND_ULT : COND_UGE};
-    case X86State::RFLAG_OF_RAW_LOC: return {Invert ? COND_FNU : COND_FU};
+    case X86State::RFLAG_SF_RAW_LOC: return Invert ? CondClass::PL : CondClass::MI;
+    case X86State::RFLAG_ZF_RAW_LOC: return Invert ? CondClass::NEQ : CondClass::EQ;
+    case X86State::RFLAG_CF_RAW_LOC: return Invert ? CondClass::ULT : CondClass::UGE;
+    case X86State::RFLAG_OF_RAW_LOC: return Invert ? CondClass::FNU : CondClass::FU;
    default: FEX_UNREACHABLE;
    }
  }
@@ -1874,11 +1955,11 @@ private:
  }

  [[nodiscard]]
-  static RegisterClassType CacheIndexClass(int Index) {
+  static RegClass CacheIndexClass(int Index) {
    if ((Index >= MM0Index && Index <= MM7Index) || Index >= FPR0Index) {
-      return FPRClass;
+      return RegClass::FPR;
    } else {
-      return GPRClass;
+      return RegClass::GPR;
    }
  }

@@ -1910,14 +1991,14 @@ private:
    RegCache.Written &= ~Bit;
  }

-  Ref LoadRegCache(uint64_t Offset, uint8_t Index, RegisterClassType RegClass, IR::OpSize Size) {
+  Ref LoadRegCache(uint64_t Offset, uint8_t Index, RegClass Class, IR::OpSize Size) {
    LOGMAN_THROW_A_FMT(Index < 64, "valid index");
    uint64_t Bit = (1ull << (uint64_t)Index);

    if (Size == OpSize::i128Bit && (RegCache.Partial & Bit)) {
      // We need to load the full register extend if we previously did a partial access.
      Ref Value = RegCache.Value[Index];
-      Ref Full = _LoadContext(Size, RegClass, Offset);
+      Ref Full = _LoadContext(Size, Class, Offset);

      // If we did a partial store, we're inserting into the full register
      if (RegCache.Written & Bit) {
@@ -1931,7 +2012,7 @@ private:
      if (Index == DFIndex) {
        RegCache.Value[Index] = _LoadDF();
      } else if ((Index >= MM0Index && Index <= MM7Index) || Index >= AVXHigh0Index) {
-        RegCache.Value[Index] = _LoadContext(Size, RegClass, Offset);
+        RegCache.Value[Index] = _LoadContext(Size, Class, Offset);

        // We may have done a partial load, this requires special handling.
        if (Size == OpSize::i64Bit) {
@@ -1942,7 +2023,7 @@ private:
      } else if (Index == AFIndex) {
        RegCache.Value[Index] = _LoadAF(Size);
      } else {
-        RegCache.Value[Index] = _LoadRegister(Offset, RegClass, Size);
+        RegCache.Value[Index] = _LoadRegister(Offset, Class, Size);
      }

      RegCache.Cached |= Bit;
@@ -1951,21 +2032,21 @@ private:
    return RegCache.Value[Index];
  }

-  RefPair AllocatePair(FEXCore::IR::RegisterClassType Class, IR::OpSize Size) {
-    if (Class == FPRClass) {
+  RefPair AllocatePair(RegClass Class, IR::OpSize Size) {
+    if (Class == RegClass::FPR) {
      return {_AllocateFPR(Size, Size), _AllocateFPR(Size, Size)};
    } else {
      return {_AllocateGPR(false), _AllocateGPR(false)};
    }
  }

-  RefPair LoadContextPair_Uncached(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, unsigned Offset) {
+  RefPair LoadContextPair_Uncached(RegClass Class, IR::OpSize Size, unsigned Offset) {
    RefPair Values = AllocatePair(Class, Size);
    _LoadContextPair(Size, Class, Offset, Values.Low, Values.High);
    return Values;
  }

-  RefPair LoadRegCachePair(uint64_t Offset, uint8_t Index, RegisterClassType RegClass, IR::OpSize Size) {
+  RefPair LoadRegCachePair(uint64_t Offset, uint8_t Index, RegClass Class, IR::OpSize Size) {
    LOGMAN_THROW_A_FMT(Index != DFIndex, "must be pairable");
    LOGMAN_THROW_A_FMT(Size != IR::OpSize::iUnsized, "Invalid size!");

@@ -1973,7 +2054,7 @@ private:
    uint64_t Bits = (3ull << (uint64_t)Index);
    const auto SizeInt = IR::OpSizeToSize(Size);
    if (((RegCache.Partial | RegCache.Cached) & Bits) == 0 && ((Offset / SizeInt) < 64)) {
-      auto Values = LoadContextPair_Uncached(RegClass, Size, Offset);
+      auto Values = LoadContextPair_Uncached(Class, Size, Offset);
      RegCache.Value[Index] = Values.Low;
      RegCache.Value[Index + 1] = Values.High;
      RegCache.Cached |= Bits;
@@ -1985,13 +2066,13 @@ private:

    // Fallback on a pair of loads
    return {
-      .Low = LoadRegCache(Offset, Index, RegClass, Size),
-      .High = LoadRegCache(Offset + SizeInt, Index + 1, RegClass, Size),
+      .Low = LoadRegCache(Offset, Index, Class, Size),
+      .High = LoadRegCache(Offset + SizeInt, Index + 1, Class, Size),
    };
  }

  Ref LoadGPR(uint8_t Reg) {
-    return LoadRegCache(Reg, GPR0Index + Reg, GPRClass, GetGPROpSize());
+    return LoadRegCache(Reg, GPR0Index + Reg, RegClass::GPR, GetGPROpSize());
  }

  Ref LoadContext(IR::OpSize Size, uint8_t Index) {
@@ -2007,7 +2088,7 @@ private:
  }

  Ref LoadXMMRegister(uint8_t Reg) {
-    return LoadRegCache(Reg, FPR0Index + Reg, FPRClass, GetGuestVectorLength());
+    return LoadRegCache(Reg, FPR0Index + Reg, RegClass::FPR, GetGuestVectorLength());
  }

  Ref LoadDF() {
@@ -2025,6 +2106,12 @@ private:
    RegCache.Written |= Bit;
  }

+  void InvalidateHighAVXRegisters() {
+    for (size_t i = 0; i < 16; ++i) {
+      InvalidateReg(AVXHigh0Index + i);
+    }
+  }
+
  void StoreRegister(uint8_t Reg, bool FPR, Ref Value) {
    StoreContext(Reg + (FPR ? FPR0Index : GPR0Index), Value);
  }
@@ -2062,7 +2149,7 @@ private:
      // Recover the sign bit, it is the logical DF value
      return _Lshr(OpSize::i64Bit, LoadDF(), Constant(63));
    } else {
-      return _LoadContext(OpSize::i8Bit, GPRClass, offsetof(Core::CPUState, flags[BitOffset]));
+      return _LoadContextGPR(OpSize::i8Bit, offsetof(Core::CPUState, flags[BitOffset]));
    }
  }

@@ -2079,18 +2166,18 @@ private:
  }

  // Safe version of NZCVSelect that handles inverted carries automatically.
-  Ref NZCVSelect(OpSize OpSize, CondClassType Cond, Ref TrueV, Ref FalseV, bool CarryIsInverted = false) {
+  Ref NZCVSelect(OpSize OpSize, CondClass Cond, Ref TrueV, Ref FalseV, bool CarryIsInverted = false) {
    switch (Cond) {
-    case IR::COND_UGE: /* cs */
-    case IR::COND_ULT: /* cc */
+    case CondClass::UGE: /* cs */
+    case CondClass::ULT: /* cc */
      // Invert the condition to match our expectations.
      if (CarryIsInverted != CFInverted) {
-        Cond = {Cond == COND_UGE ? COND_ULT : COND_UGE};
+        Cond = (Cond == CondClass::UGE) ? CondClass::ULT : CondClass::UGE;
      }
      break;

-    case IR::COND_UGT: /* hi */
-    case IR::COND_ULE: /* ls */
+    case CondClass::UGT: /* hi */
+    case CondClass::ULE: /* ls */
      // No clever optimization we can do here, rectify carry itself.
      RectifyCarryInvert(CarryIsInverted);
      break;
@@ -2179,7 +2266,7 @@ private:

    HandleNZCV_RMW();
    CalculatePF(_ShiftFlags(OpSizeFromSrc(Op), Result, Dest, Shift, Src, OldPF, CFInverted));
-    StoreResult(GPRClass, Op, Result, OpSize::iInvalid);
+    StoreResultGPR(Op, Result);
  }

  // Helper to derive Dest by a given builder-using Expression with the opcode
@@ -2252,8 +2339,7 @@ private:
    CachedIndexedNamedVectorConstants.clear();
  }

-  std::optional<CondClassType> DecodeNZCVCondition(uint8_t OP);
-  Ref SelectBit(Ref Cmp, IR::OpSize ResultSize, Ref TrueValue, Ref FalseValue);
+  std::optional<CondClass> DecodeNZCVCondition(uint8_t OP);
  Ref SelectCC0All1(uint8_t OP);

  /**
@@ -2267,8 +2353,8 @@ private:
    if (Size != OpSize::i32Bit) {
      return;
    }
-    auto Dest = LoadSource(GPRClass, Op, Op->Dest, Op->Flags);
-    StoreResult(GPRClass, Op, Dest, OpSize::iInvalid);
+    auto Dest = LoadSourceGPR(Op, Op->Dest, Op->Flags);
+    StoreResultGPR(Op, Dest);
  }

  using ZeroShiftFunctionPtr = void (OpDispatchBuilder::*)(FEXCore::X86Tables::DecodedOp Op);
@@ -2303,7 +2389,7 @@ private:

    ///< Jump to zeroshift block or end block depending on if it was provided.
    IRPair<IROp_CodeBlock> TailHandling = ZeroShiftResult ? ZeroShiftBlock : EndBlock;
-    CondJump(Shift, Zero, TailHandling, SetBlock, {COND_EQ});
+    CondJump(Shift, Zero, TailHandling, SetBlock, CondClass::EQ);

    SetCurrentCodeBlock(SetBlock);
    StartNewBlock();
@@ -2346,9 +2432,7 @@ private:
  void CalculateFlags_MUL(IR::OpSize SrcSize, Ref Res, Ref High);
  void CalculateFlags_UMUL(Ref High);
  void CalculateFlags_Logical(IR::OpSize SrcSize, Ref Res);
-  void CalculateFlags_ShiftLeft(IR::OpSize SrcSize, Ref Res, Ref Src1, Ref Src2);
  void CalculateFlags_ShiftLeftImmediate(IR::OpSize SrcSize, Ref Res, Ref Src1, uint64_t Shift);
-  void CalculateFlags_ShiftRight(IR::OpSize SrcSize, Ref Res, Ref Src1, Ref Src2);
  void CalculateFlags_ShiftRightImmediate(IR::OpSize SrcSize, Ref Res, Ref Src1, uint64_t Shift);
  void CalculateFlags_ShiftRightDoubleImmediate(IR::OpSize SrcSize, Ref Res, Ref Src1, uint64_t Shift);
  void CalculateFlags_ShiftRightImmediateCommon(IR::OpSize SrcSize, Ref Res, Ref Src1, uint64_t Shift);
@@ -2365,7 +2449,7 @@ private:
    LOGMAN_THROW_A_FMT(MMXState == MMXState_X87, "Expected state to be x87");
    _StackForceSlow();
    SetX87Top(Constant(0)); // top reset to zero
-    _StoreContext(OpSize::i8Bit, GPRClass, Constant(0xFFFFUL), offsetof(FEXCore::Core::CPUState, AbridgedFTW));
+    _StoreContextGPR(OpSize::i8Bit, Constant(0xFFFFUL), offsetof(FEXCore::Core::CPUState, AbridgedFTW));
    MMXState = MMXState_MMX;
  }

@@ -2402,44 +2486,62 @@ private:
  IROp_IRHeader* CurrentHeader {};

  [[nodiscard]]
-  bool IsTSOEnabled(FEXCore::IR::RegisterClassType Class) const {
+  bool IsTSOEnabled(RegClass Class) const {
    if (ForceTSO == ForceTSOMode::ForceEnabled) {
      return true;
    } else if (ForceTSO == ForceTSOMode::ForceDisabled) {
      return false;
-    } else if (Class == FPRClass) {
+    } else if (Class == RegClass::FPR) {
      return CTX->IsVectorAtomicTSOEnabled();
    } else {
      return CTX->IsAtomicTSOEnabled();
    }
  }

-  Ref _StoreMemAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, Ref Addr, Ref Value, IR::OpSize Align = IR::OpSize::i8Bit) {
+  Ref _StoreMemAutoTSO(RegClass Class, OpSize Size, Ref Addr, Ref Value, OpSize Align = OpSize::i8Bit) {
    if (IsTSOEnabled(Class)) {
-      return _StoreMemTSO(Class, Size, Value, Addr, Invalid(), Align, MEM_OFFSET_SXTX, 1);
+      return _StoreMemTSO(Class, Size, Value, Addr, Invalid(), Align, MemOffsetType::SXTX, 1);
    } else {
-      return _StoreMem(Class, Size, Value, Addr, Invalid(), Align, MEM_OFFSET_SXTX, 1);
+      return _StoreMem(Class, Size, Value, Addr, Invalid(), Align, MemOffsetType::SXTX, 1);
    }
  }
-
-  Ref _LoadMemAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, Ref ssa0, IR::OpSize Align = IR::OpSize::i8Bit) {
-    if (IsTSOEnabled(Class)) {
-      return _LoadMemTSO(Class, Size, ssa0, Invalid(), Align, MEM_OFFSET_SXTX, 1);
-    } else {
-      return _LoadMem(Class, Size, ssa0, Invalid(), Align, MEM_OFFSET_SXTX, 1);
-    }
+  Ref _StoreMemGPRAutoTSO(OpSize Size, Ref Addr, Ref Value, OpSize Align = OpSize::i8Bit) {
+    return _StoreMemAutoTSO(RegClass::GPR, Size, Addr, Value, Align);
+  }
+  Ref _StoreMemFPRAutoTSO(OpSize Size, Ref Addr, Ref Value, OpSize Align = OpSize::i8Bit) {
+    return _StoreMemAutoTSO(RegClass::FPR, Size, Addr, Value, Align);
  }

-  Ref _LoadMemAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, AddressMode A, IR::OpSize Align = IR::OpSize::i8Bit) {
-    bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
-    A = SelectAddressMode(this, A, GetGPROpSize(), CTX->HostFeatures.SupportsTSOImm9, AtomicTSO, Class != GPRClass, Size);
+  Ref _LoadMemAutoTSO(RegClass Class, OpSize Size, Ref ssa0, OpSize Align = OpSize::i8Bit) {
+    if (IsTSOEnabled(Class)) {
+      return _LoadMemTSO(Class, Size, ssa0, Invalid(), Align, MemOffsetType::SXTX, 1);
+    } else {
+      return _LoadMem(Class, Size, ssa0, Invalid(), Align, MemOffsetType::SXTX, 1);
+    }
+  }
+  Ref _LoadMemGPRAutoTSO(OpSize Size, Ref ssa0, OpSize Align = OpSize::i8Bit) {
+    return _LoadMemAutoTSO(RegClass::GPR, Size, ssa0, Align);
+  }
+  Ref _LoadMemFPRAutoTSO(OpSize Size, Ref ssa0, OpSize Align = OpSize::i8Bit) {
+    return _LoadMemAutoTSO(RegClass::FPR, Size, ssa0, Align);
+  }
+
+  Ref _LoadMemAutoTSO(RegClass Class, OpSize Size, const AddressMode& A, OpSize Align = OpSize::i8Bit) {
+    const bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
+    const auto B = SelectAddressMode(this, A, GetGPROpSize(), CTX->HostFeatures.SupportsTSOImm9, AtomicTSO, Class != RegClass::GPR, Size);

    if (AtomicTSO) {
-      return _LoadMemTSO(Class, Size, A.Base, A.Index, Align, A.IndexType, A.IndexScale);
+      return _LoadMemTSO(Class, Size, B.Base, B.Index, Align, B.IndexType, B.IndexScale);
    } else {
-      return _LoadMem(Class, Size, A.Base, A.Index, Align, A.IndexType, A.IndexScale);
+      return _LoadMem(Class, Size, B.Base, B.Index, Align, B.IndexType, B.IndexScale);
    }
  }
+  Ref _LoadMemGPRAutoTSO(OpSize Size, const AddressMode& A, OpSize Align = OpSize::i8Bit) {
+    return _LoadMemAutoTSO(RegClass::GPR, Size, A, Align);
+  }
+  Ref _LoadMemFPRAutoTSO(OpSize Size, const AddressMode& A, OpSize Align = OpSize::i8Bit) {
+    return _LoadMemAutoTSO(RegClass::FPR, Size, A, Align);
+  }

  AddressMode SelectPairAddressMode(AddressMode A, IR::OpSize Size) {
    LOGMAN_THROW_A_FMT(Size != IR::OpSize::iUnsized, "Invalid size!");
@@ -2457,56 +2559,72 @@ private:
  }


-  RefPair LoadMemPair(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, Ref Base, unsigned Offset) {
+  RefPair LoadMemPair(RegClass Class, OpSize Size, Ref Base, uint32_t Offset) {
    RefPair Values = AllocatePair(Class, Size);
    _LoadMemPair(Class, Size, Base, Offset, Values.Low, Values.High);
    return Values;
  }
+  RefPair LoadMemPairFPR(OpSize Size, Ref Base, uint32_t Offset) {
+    return LoadMemPair(RegClass::FPR, Size, Base, Offset);
+  }

-  RefPair _LoadMemPairAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, AddressMode A, IR::OpSize Align = IR::OpSize::i8Bit) {
-    bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
+  RefPair _LoadMemPairAutoTSO(RegClass Class, OpSize Size, const AddressMode& A, OpSize Align = OpSize::i8Bit) {
+    const bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;

    // Use ldp if possible, otherwise fallback on two loads.
-    if (!AtomicTSO && !A.Segment && Size >= OpSize::i32Bit & Size <= OpSize::i128Bit) {
-      A = SelectPairAddressMode(A, Size);
-      return LoadMemPair(Class, Size, A.Base, A.Offset);
-    } else {
-      AddressMode HighA = A;
-      HighA.Offset += 16;
-
-      return {
-        .Low = _LoadMemAutoTSO(Class, Size, A, Align),
-        .High = _LoadMemAutoTSO(Class, Size, HighA, Align),
-      };
+    if (!AtomicTSO && !A.Segment && Size >= OpSize::i32Bit && Size <= OpSize::i128Bit) {
+      const auto B = SelectPairAddressMode(A, Size);
+      return LoadMemPair(Class, Size, B.Base, B.Offset);
    }
+
+    AddressMode HighA = A;
+    HighA.Offset += 16;
+
+    return {
+      .Low = _LoadMemAutoTSO(Class, Size, A, Align),
+      .High = _LoadMemAutoTSO(Class, Size, HighA, Align),
+    };
+  }
+  RefPair _LoadMemPairFPRAutoTSO(OpSize Size, const AddressMode& A, OpSize Align = OpSize::i8Bit) {
+    return _LoadMemPairAutoTSO(RegClass::FPR, Size, A, Align);
  }

-  Ref _StoreMemAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, AddressMode A, Ref Value, IR::OpSize Align = IR::OpSize::i8Bit) {
-    bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
-    A = SelectAddressMode(this, A, GetGPROpSize(), CTX->HostFeatures.SupportsTSOImm9, AtomicTSO, Class != GPRClass, Size);
+  Ref _StoreMemAutoTSO(RegClass Class, OpSize Size, const AddressMode& A, Ref Value, OpSize Align = OpSize::i8Bit) {
+    const bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
+    const auto B = SelectAddressMode(this, A, GetGPROpSize(), CTX->HostFeatures.SupportsTSOImm9, AtomicTSO, Class != RegClass::GPR, Size);

    if (AtomicTSO) {
-      return _StoreMemTSO(Class, Size, Value, A.Base, A.Index, Align, A.IndexType, A.IndexScale);
+      return _StoreMemTSO(Class, Size, Value, B.Base, B.Index, Align, B.IndexType, B.IndexScale);
    } else {
-      return _StoreMem(Class, Size, Value, A.Base, A.Index, Align, A.IndexType, A.IndexScale);
+      return _StoreMem(Class, Size, Value, B.Base, B.Index, Align, B.IndexType, B.IndexScale);
    }
  }
+  Ref _StoreMemGPRAutoTSO(OpSize Size, const AddressMode& A, Ref Value, OpSize Align = OpSize::i8Bit) {
+    return _StoreMemAutoTSO(RegClass::GPR, Size, A, Value, Align);
+  }
+  Ref _StoreMemFPRAutoTSO(OpSize Size, const AddressMode& A, Ref Value, OpSize Align = OpSize::i8Bit) {
+    return _StoreMemAutoTSO(RegClass::FPR, Size, A, Value, Align);
+  }

-  void _StoreMemPairAutoTSO(FEXCore::IR::RegisterClassType Class, IR::OpSize Size, AddressMode A, Ref Value1, Ref Value2,
-                            IR::OpSize Align = IR::OpSize::i8Bit) {
+  void _StoreMemPairAutoTSO(RegClass Class, OpSize Size, const AddressMode& A, Ref Value1, Ref Value2, OpSize Align = OpSize::i8Bit) {
    const auto SizeInt = IR::OpSizeToSize(Size);
-    bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;
+    const bool AtomicTSO = IsTSOEnabled(Class) && !A.NonTSO;

    // Use stp if possible, otherwise fallback on two stores.
-    if (!AtomicTSO && !A.Segment && Size >= OpSize::i32Bit & Size <= OpSize::i128Bit) {
-      A = SelectPairAddressMode(A, Size);
-      _StoreMemPair(Class, Size, Value1, Value2, A.Base, A.Offset);
+    if (!AtomicTSO && !A.Segment && Size >= OpSize::i32Bit && Size <= OpSize::i128Bit) {
+      const auto B = SelectPairAddressMode(A, Size);
+      _StoreMemPair(Class, Size, Value1, Value2, B.Base, B.Offset);
    } else {
-      _StoreMemAutoTSO(Class, Size, A, Value1, OpSize::i8Bit);
-      A.Offset += SizeInt;
-      _StoreMemAutoTSO(Class, Size, A, Value2, OpSize::i8Bit);
+      auto B = A;
+
+      _StoreMemAutoTSO(Class, Size, B, Value1, OpSize::i8Bit);
+      B.Offset += SizeInt;
+      _StoreMemAutoTSO(Class, Size, B, Value2, OpSize::i8Bit);
    }
  }
+  void _StoreMemPairFPRAutoTSO(OpSize Size, const AddressMode& A, Ref Value1, Ref Value2, OpSize Align = OpSize::i8Bit) {
+    return _StoreMemPairAutoTSO(RegClass::FPR, Size, A, Value1, Value2, Align);
+  }

  Ref Pop(IR::OpSize Size, Ref SP_RMW) {
    Ref Value = _AllocateGPR(false);
@@ -35,20 +35,16 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_LoadSource_WithOpSize(
  } else {
    LOGMAN_THROW_A_FMT(IsOperandMem(Operand, true), "only memory sources");

-    AddressMode A = DecodeAddress(Op, Operand, AccessType, true /* IsLoad */);
-
-    AddressMode HighA = A;
-    HighA.Offset += 16;
-
    if (Operand.IsSIB()) {
      const bool IsVSIB = (Op->Flags & X86Tables::DecodeFlags::FLAG_VSIB_BYTE) != 0;
      LOGMAN_THROW_A_FMT(!IsVSIB, "VSIB uses LoadVSIB instead");
    }

+    const AddressMode A = DecodeAddress(Op, Operand, AccessType, true /* IsLoad */);
    if (NeedsHigh) {
-      return _LoadMemPairAutoTSO(FPRClass, OpSize::i128Bit, A, OpSize::i8Bit);
+      return _LoadMemPairFPRAutoTSO(OpSize::i128Bit, A, OpSize::i8Bit);
    } else {
-      return {.Low = _LoadMemAutoTSO(FPRClass, OpSize::i128Bit, A, OpSize::i8Bit)};
+      return {.Low = _LoadMemFPRAutoTSO(OpSize::i128Bit, A, OpSize::i8Bit)};
    }
  }
 }
@@ -56,7 +52,8 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_LoadSource_WithOpSize(
 OpDispatchBuilder::RefVSIB
 OpDispatchBuilder::AVX128_LoadVSIB(const X86Tables::DecodedOp& Op, const X86Tables::DecodedOperand& Operand, uint32_t Flags, bool NeedsHigh) {
  const bool IsVSIB = (Op->Flags & X86Tables::DecodeFlags::FLAG_VSIB_BYTE) != 0;
-  LOGMAN_THROW_A_FMT(Operand.IsSIB() && IsVSIB, "Trying to load VSIB for something that isn't the correct type!");
+  LOGMAN_THROW_A_FMT((Operand.IsSIB() || Operand.IsSIBRelocation()) && IsVSIB, "Trying to load VSIB for something that isn't the correct "
+                                                                               "type!");

  // VSIB is a very special case which has a ton of encoded data.
  // Get it in a format we can reason about.
@@ -68,13 +65,25 @@ OpDispatchBuilder::AVX128_LoadVSIB(const X86Tables::DecodedOp& Op, const X86Tabl
                     "Base must be a GPR.");
  const auto Index_XMM_gpr = Index_gpr - X86State::REG_XMM_0;

-  return {
+  OpDispatchBuilder::RefVSIB A {
    .Low = AVX128_LoadXMMRegister(Index_XMM_gpr, false),
    .High = NeedsHigh ? AVX128_LoadXMMRegister(Index_XMM_gpr, true) : Invalid(),
    .BaseAddr = Base_gpr != FEXCore::X86State::REG_INVALID ? LoadGPRRegister(Base_gpr, OpSize::i64Bit, 0, false) : nullptr,
-    .Displacement = Operand.Data.SIB.Offset,
    .Scale = Operand.Data.SIB.Scale,
  };
+
+  if (Operand.IsSIBRelocation()) {
+    auto EPOffset = _EntrypointOffset(OpSize::i64Bit, Operand.Data.SIB.Offset);
+    if (A.BaseAddr) {
+      A.BaseAddr = Add(OpSize::i64Bit, EPOffset, A.BaseAddr);
+    } else {
+      A.BaseAddr = EPOffset;
+    }
+  } else {
+    A.Displacement = static_cast<int32_t>(Operand.Data.SIB.Offset);
+  }
+
+  return A;
 }

 void OpDispatchBuilder::AVX128_StoreResult_WithOpSize(FEXCore::X86Tables::DecodedOp Op, const FEXCore::X86Tables::DecodedOperand& Operand,
@@ -95,9 +104,9 @@ void OpDispatchBuilder::AVX128_StoreResult_WithOpSize(FEXCore::X86Tables::Decode
    AddressMode A = DecodeAddress(Op, Operand, AccessType, false /* IsLoad */);

    if (Src.High) {
-      _StoreMemPairAutoTSO(FPRClass, OpSize::i128Bit, A, Src.Low, Src.High, OpSize::i8Bit);
+      _StoreMemPairFPRAutoTSO(OpSize::i128Bit, A, Src.Low, Src.High, OpSize::i8Bit);
    } else {
-      _StoreMemAutoTSO(FPRClass, OpSize::i128Bit, A, Src.Low, OpSize::i8Bit);
+      _StoreMemFPRAutoTSO(OpSize::i128Bit, A, Src.Low, OpSize::i8Bit);
    }
  }
 }
@@ -151,13 +160,13 @@ void OpDispatchBuilder::AVX128_VMOVScalarImpl(OpcodeArgs, IR::OpSize ElementSize
    AVX128_StoreResult_WithOpSize(Op, Op->Dest, RefPair {.Low = Result, .High = High});
  } else if (Op->Dest.IsGPR()) {
    // VMOVSS/SD xmm1, mem32/mem64
-    Ref Src = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], ElementSize, Op->Flags);
+    Ref Src = LoadSourceFPR_WithOpSize(Op, Op->Src[1], ElementSize, Op->Flags);
    auto High = LoadZeroVector(OpSize::i128Bit);
    AVX128_StoreResult_WithOpSize(Op, Op->Dest, RefPair {.Low = Src, .High = High});
  } else {
    // VMOVSS/SD mem32/mem64, xmm1
    auto Src = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, false);
-    StoreResult_WithOpSize(FPRClass, Op, Op->Dest, Src.Low, ElementSize, OpSize::iInvalid);
+    StoreResultFPR_WithOpSize(Op, Op->Dest, Src.Low, ElementSize);
  }
 }

@@ -331,16 +340,12 @@ void OpDispatchBuilder::AVX128_VZERO(OpcodeArgs) {
      AVX128_StoreXMMRegister(i, ZeroVector, false);
    }

-    // More efficient for non-SRA upper-halves to use a cached constant and store directly.
-    for (uint32_t i = 0; i < NumRegs; i++) {
-      AVX128_StoreXMMRegister(i, ZeroVector, true);
-    }
+    InvalidateHighAVXRegisters();
+    _ContextClear(offsetof(FEXCore::Core::CPUState, avx_high), sizeof(FEXCore::Core::CPUState::avx_high[0]) * NumRegs);
  } else {
    // Likewise, VZEROUPPER will only ever zero only up to the first 16 registers
-    const auto ZeroVector = LoadZeroVector(OpSize::i128Bit);
-    for (uint32_t i = 0; i < NumRegs; i++) {
-      AVX128_StoreXMMRegister(i, ZeroVector, true);
-    }
+    InvalidateHighAVXRegisters();
+    _ContextClear(offsetof(FEXCore::Core::CPUState, avx_high), sizeof(FEXCore::Core::CPUState::avx_high[0]) * NumRegs);
  }
 }

@@ -351,7 +356,7 @@ void OpDispatchBuilder::AVX128_MOVVectorNT(OpcodeArgs) {
  if (Op->Dest.IsGPR()) {
    ///< MOVNTDQA load non-temporal comes from SSE4.1 and is extended by AVX/AVX2.
    RefPair Src {};
-    Ref SrcAddr = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags, {.LoadData = false});
+    Ref SrcAddr = LoadSourceGPR(Op, Op->Src[0], Op->Flags, {.LoadData = false});
    Src.Low = _VLoadNonTemporal(OpSize::i128Bit, SrcAddr, 0);

    if (Is128Bit) {
@@ -362,7 +367,7 @@ void OpDispatchBuilder::AVX128_MOVVectorNT(OpcodeArgs) {
    AVX128_StoreResult_WithOpSize(Op, Op->Dest, Src);
  } else {
    auto Src = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, !Is128Bit, MemoryAccessType::STREAM);
-    Ref Dest = LoadSource(GPRClass, Op, Op->Dest, Op->Flags, {.LoadData = false});
+    Ref Dest = LoadSourceGPR(Op, Op->Dest, Op->Flags, {.LoadData = false});

    if (Is128Bit) {
      // Single store non-temporal for 128-bit operations.
@@ -379,7 +384,7 @@ void OpDispatchBuilder::AVX128_MOVQ(OpcodeArgs) {
  if (Op->Src[0].IsGPR()) {
    Src = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, false);
  } else {
-    Src.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], OpSize::i64Bit, Op->Flags);
+    Src.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], OpSize::i64Bit, Op->Flags);
  }

  // This instruction is a bit special that if the destination is a register then it'll ZEXT the 64bit source to 256bit
@@ -390,7 +395,7 @@ void OpDispatchBuilder::AVX128_MOVQ(OpcodeArgs) {
    Src.High = ZeroVector;
    AVX128_StoreResult_WithOpSize(Op, Op->Dest, Src);
  } else {
-    StoreResult_WithOpSize(FPRClass, Op, Op->Dest, Src.Low, OpSize::i64Bit, OpSize::i64Bit);
+    StoreResultFPR_WithOpSize(Op, Op->Dest, Src.Low, OpSize::i64Bit, OpSize::i64Bit);
  }
 }

@@ -399,7 +404,7 @@ void OpDispatchBuilder::AVX128_VMOVLP(OpcodeArgs) {

  if (!Op->Dest.IsGPR()) {
    ///< VMOVLPS/PD mem64, xmm1
-    StoreResult_WithOpSize(FPRClass, Op, Op->Dest, Src1.Low, OpSize::i64Bit, OpSize::i64Bit);
+    StoreResultFPR_WithOpSize(Op, Op->Dest, Src1.Low, OpSize::i64Bit, OpSize::i64Bit);
  } else if (!Op->Src[1].IsGPR()) {
    ///< VMOVLPS/PD xmm1, xmm2, mem64
    // Bits[63:0] come from Src2[63:0]
@@ -463,7 +468,7 @@ void OpDispatchBuilder::AVX128_VMOVDDUP(OpcodeArgs) {
    // 128-bit operation only loads 8-bytes.
    // 256-bit operation loads a full 32-bytes.
    if (Is128Bit) {
-      Src.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], OpSize::i64Bit, Op->Flags);
+      Src.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], OpSize::i64Bit, Op->Flags);
    } else {
      Src = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, true);
    }
@@ -558,18 +563,18 @@ void OpDispatchBuilder::AVX128_InsertCVTGPR_To_FPR(OpcodeArgs, IR::OpSize DstEle

  if (Op->Src[1].IsGPR()) {
    // If the source is a GPR then convert directly from the GPR.
-    auto Src2 = LoadSource_WithOpSize(GPRClass, Op, Op->Src[1], GetGPROpSize(), Op->Flags);
+    auto Src2 = LoadSourceGPR_WithOpSize(Op, Op->Src[1], GetGPROpSize(), Op->Flags);
    Result.Low = _VSToFGPRInsert(OpSize::i128Bit, DstElementSize, SrcSize, Src1.Low, Src2, false);
  } else if (SrcSize != DstElementSize) {
    // If the source is from memory but the Source size and destination size aren't the same,
    // then it is more optimal to load in to a GPR and convert between GPR->FPR.
    // ARM GPR->FPR conversion supports different size source and destinations while FPR->FPR doesn't.
-    auto Src2 = LoadSource(GPRClass, Op, Op->Src[1], Op->Flags);
+    auto Src2 = LoadSourceGPR(Op, Op->Src[1], Op->Flags);
    Result.Low = _VSToFGPRInsert(DstSize, DstElementSize, SrcSize, Src1.Low, Src2, false);
  } else {
    // In the case of cvtsi2s{s,d} where the source and destination are the same size,
    // then it is more optimal to load in to the FPR register directly and convert there.
-    auto Src2 = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+    auto Src2 = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
    // Always signed
    Result.Low = _VSToFVectorInsert(DstSize, DstElementSize, DstElementSize, Src1.Low, Src2, false, false);
  }
@@ -589,11 +594,11 @@ void OpDispatchBuilder::AVX128_CVTFPR_To_GPR(OpcodeArgs, IR::OpSize SrcElementSi
  if (Op->Src[0].IsGPR()) {
    Src = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, false);
  } else {
-    Src.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], SrcElementSize, Op->Flags);
+    Src.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], SrcElementSize, Op->Flags);
  }

  Ref Result = CVTFPR_To_GPRImpl(Op, Src.Low, SrcElementSize, HostRoundingMode);
-  StoreResult(GPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultGPR(Op, Result);
 }

 void OpDispatchBuilder::AVX128_VANDN(OpcodeArgs) {
@@ -636,7 +641,7 @@ void OpDispatchBuilder::AVX128_UCOMISx(OpcodeArgs, IR::OpSize ElementSize) {
  if (Op->Src[0].IsGPR()) {
    Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, false);
  } else {
-    Src2.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], SrcSize, Op->Flags);
+    Src2.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], SrcSize, Op->Flags);
  }

  Comiss(ElementSize, Src1.Low, Src2.Low);
@@ -653,7 +658,7 @@ void OpDispatchBuilder::AVX128_VectorScalarInsertALU(OpcodeArgs, FEXCore::IR::IR
  if (Op->Src[1].IsGPR()) {
    Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, false);
  } else {
-    Src2.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], SrcSize, Op->Flags);
+    Src2.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[1], SrcSize, Op->Flags);
  }

  // If OpSize == ElementSize then it only does the lower scalar op
@@ -667,10 +672,10 @@ void OpDispatchBuilder::AVX128_VFCMP(OpcodeArgs, IR::OpSize ElementSize) {

  struct {
    FEXCore::X86Tables::DecodedOp Op;
-    uint8_t CompType {};
+    uint32_t CompType {};
  } Capture {
    .Op = Op,
-    .CompType = CompType,
+    .CompType = CompType & 0b11111u,
  };

  AVX128_VectorBinaryImpl(Op, OpSizeFromSrc(Op), ElementSize, [this, &Capture](IR::OpSize _ElementSize, Ref Src1, Ref Src2) {
@@ -690,13 +695,13 @@ void OpDispatchBuilder::AVX128_InsertScalarFCMP(OpcodeArgs, IR::OpSize ElementSi
  if (Op->Src[1].IsGPR()) {
    Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, false);
  } else {
-    Src2.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], SrcSize, Op->Flags);
+    Src2.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[1], SrcSize, Op->Flags);
  }

  const uint8_t CompType = Op->Src[2].Literal();

  RefPair Result {};
-  Result.Low = InsertScalarFCMPOpImpl(OpSize::i128Bit, OpSize::i128Bit, ElementSize, Src1.Low, Src2.Low, CompType, false);
+  Result.Low = InsertScalarFCMPOpImpl(OpSize::i128Bit, OpSize::i128Bit, ElementSize, Src1.Low, Src2.Low, CompType & 0b11111, false);
  Result.High = LoadZeroVector(OpSize::i128Bit);
  AVX128_StoreResult_WithOpSize(Op, Op->Dest, Result);
 }
@@ -708,12 +713,12 @@ void OpDispatchBuilder::AVX128_MOVBetweenGPR_FPR(OpcodeArgs) {
    RefPair Result {};
    if (Op->Src[0].IsGPR()) {
      // Loading from GPR and moving to Vector.
-      Ref Src = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], GetGPROpSize(), Op->Flags);
+      Ref Src = LoadSourceFPR_WithOpSize(Op, Op->Src[0], GetGPROpSize(), Op->Flags);
      // zext to 128bit
      Result.Low = _VCastFromGPR(OpSize::i128Bit, OpSizeFromSrc(Op), Src);
    } else {
      // Loading from Memory as a scalar. Zero extend
-      Result.Low = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      Result.Low = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    }

    Result.High = LoadZeroVector(OpSize::i128Bit);
@@ -726,11 +731,11 @@ void OpDispatchBuilder::AVX128_MOVBetweenGPR_FPR(OpcodeArgs) {
      auto ElementSize = OpSizeFromDst(Op);
      // Extract element from GPR. Zero extending in the process.
      Src.Low = _VExtractToGPR(OpSizeFromSrc(Op), ElementSize, Src.Low, 0);
-      StoreResult(GPRClass, Op, Op->Dest, Src.Low, OpSize::iInvalid);
+      StoreResultGPR(Op, Op->Dest, Src.Low);
    } else {
      // Storing first element to memory.
-      Ref Dest = LoadSource(GPRClass, Op, Op->Dest, Op->Flags, {.LoadData = false});
-      _StoreMem(FPRClass, OpSizeFromDst(Op), Dest, Src.Low, OpSize::i8Bit);
+      Ref Dest = LoadSourceGPR(Op, Op->Dest, Op->Flags, {.LoadData = false});
+      _StoreMemFPR(OpSizeFromDst(Op), Dest, Src.Low, OpSize::i8Bit);
    }
  }
 }
@@ -758,7 +763,7 @@ void OpDispatchBuilder::AVX128_PExtr(OpcodeArgs, IR::OpSize ElementSize) {
    const auto GPRSize = GetGPROpSize();
    // Extract already zero extends the result.
    Ref Result = _VExtractToGPR(OpSize::i128Bit, OverridenElementSize, Src.Low, Index);
-    StoreResult_WithOpSize(GPRClass, Op, Op->Dest, Result, GPRSize, OpSize::iInvalid);
+    StoreResultGPR_WithOpSize(Op, Op->Dest, Result, GPRSize);
    return;
  }

@@ -779,7 +784,7 @@ void OpDispatchBuilder::AVX128_ExtendVectorElements(OpcodeArgs, IR::OpSize Eleme
      const auto SrcSize = OpSizeFromSrc(Op);
      const auto LoadSize = Is256Bit ? IR::SizeToOpSize(IR::OpSizeToSize(SrcSize) * 2) : SrcSize;

-      return LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], LoadSize, Op->Flags);
+      return LoadSourceFPR_WithOpSize(Op, Op->Src[0], LoadSize, Op->Flags);
    }
  };

@@ -868,7 +873,7 @@ void OpDispatchBuilder::AVX128_MOVMSK(OpcodeArgs, IR::OpSize ElementSize) {
    auto GPRHigh = Mask8Byte(Src.High);
    GPR = _Orlshl(OpSize::i64Bit, GPRLow, GPRHigh, 2);
  }
-  StoreResult_WithOpSize(GPRClass, Op, Op->Dest, GPR, GetGPROpSize(), OpSize::iInvalid);
+  StoreResultGPR_WithOpSize(Op, Op->Dest, GPR, GetGPROpSize());
 }

 void OpDispatchBuilder::AVX128_MOVMSKB(OpcodeArgs) {
@@ -897,7 +902,7 @@ void OpDispatchBuilder::AVX128_MOVMSKB(OpcodeArgs) {
    Result = _Orlshl(OpSize::i64Bit, Result, ResultHigh, 16);
  }

-  StoreResult(GPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultGPR(Op, Result);
 }

 void OpDispatchBuilder::AVX128_PINSRImpl(OpcodeArgs, IR::OpSize ElementSize, const X86Tables::DecodedOperand& Src1Op,
@@ -910,7 +915,7 @@ void OpDispatchBuilder::AVX128_PINSRImpl(OpcodeArgs, IR::OpSize ElementSize, con

  if (Src2Op.IsGPR()) {
    // If the source is a GPR then convert directly from the GPR.
-    auto Src2 = LoadSource_WithOpSize(GPRClass, Op, Src2Op, GetGPROpSize(), Op->Flags);
+    auto Src2 = LoadSourceGPR_WithOpSize(Op, Src2Op, GetGPROpSize(), Op->Flags);
    Result.Low = _VInsGPR(OpSize::i128Bit, ElementSize, Index, Src1.Low, Src2);
  } else {
    // If loading from memory then we only load the element size
@@ -1047,7 +1052,7 @@ void OpDispatchBuilder::AVX128_InsertScalar_CVT_Float_To_Float(OpcodeArgs, IR::O
  // Then zero extends the top 128-bit.
  const auto SrcSize = Op->Src[1].IsGPR() ? OpSize::i128Bit : SrcElementSize;
  auto Src1 = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, false);
-  Ref Src2 = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], SrcSize, Op->Flags, {.AllowUpperGarbage = true});
+  Ref Src2 = LoadSourceFPR_WithOpSize(Op, Op->Src[1], SrcSize, Op->Flags, {.AllowUpperGarbage = true});

  Ref Result = _VFToFScalarInsert(OpSize::i128Bit, DstElementSize, SrcElementSize, Src1.Low, Src2, false);
  AVX128_StoreResult_WithOpSize(Op, Op->Dest, AVX128_Zext(Result));
@@ -1076,7 +1081,7 @@ void OpDispatchBuilder::AVX128_Vector_CVT_Float_To_Float(OpcodeArgs, IR::OpSize
  } else {
    // Handle 64-bit memory source.
    // In the case of cvtps2pd xmm, m64.
-    Src.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], LoadSize, Op->Flags);
+    Src.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], LoadSize, Op->Flags);
  }

  RefPair Result {};
@@ -1154,7 +1159,7 @@ void OpDispatchBuilder::AVX128_Vector_CVT_Int_To_Float(OpcodeArgs, IR::OpSize Sr
      // unnecessarily zero extend the vector. Otherwise, if
      // memory, then we want to load the element size exactly.
      const auto LoadSize = IR::SizeToOpSize(8 * (IR::OpSizeToSize(Size) / 16));
-      return RefPair {.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], LoadSize, Op->Flags)};
+      return RefPair {.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], LoadSize, Op->Flags)};
    } else {
      return AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, !Is128Bit);
    }
@@ -1304,7 +1309,7 @@ void OpDispatchBuilder::AVX128_InsertScalarRound(OpcodeArgs, IR::OpSize ElementS
  if (Op->Src[1].IsGPR()) {
    Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, false);
  } else {
-    Src2.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], SrcSize, Op->Flags);
+    Src2.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[1], SrcSize, Op->Flags);
  }

  // If OpSize == ElementSize then it only does the lower scalar op
@@ -1573,20 +1578,20 @@ void OpDispatchBuilder::AVX128_VMASKMOVImpl(OpcodeArgs, IR::OpSize ElementSize,
    auto Address = MakeAddress(Op->Dest);

    auto Data = AVX128_LoadSource_WithOpSize(Op, DataOp, Op->Flags, !Is128Bit);
-    _VStoreVectorMasked(OpSize::i128Bit, ElementSize, Mask.Low, Data.Low, Address, Invalid(), MEM_OFFSET_SXTX, 1);
+    _VStoreVectorMasked(OpSize::i128Bit, ElementSize, Mask.Low, Data.Low, Address, Invalid(), MemOffsetType::SXTX, 1);
    if (!Is128Bit) {
-      _VStoreVectorMasked(OpSize::i128Bit, ElementSize, Mask.High, Data.High, Address, _InlineConstant(16), MEM_OFFSET_SXTX, 1);
+      _VStoreVectorMasked(OpSize::i128Bit, ElementSize, Mask.High, Data.High, Address, _InlineConstant(16), MemOffsetType::SXTX, 1);
    }
  } else {
    auto Address = MakeAddress(DataOp);

    RefPair Result {};
-    Result.Low = _VLoadVectorMasked(OpSize::i128Bit, ElementSize, Mask.Low, Address, Invalid(), MEM_OFFSET_SXTX, 1);
+    Result.Low = _VLoadVectorMasked(OpSize::i128Bit, ElementSize, Mask.Low, Address, Invalid(), MemOffsetType::SXTX, 1);

    if (Is128Bit) {
      Result.High = LoadZeroVector(OpSize::i128Bit);
    } else {
-      Result.High = _VLoadVectorMasked(OpSize::i128Bit, ElementSize, Mask.High, Address, _InlineConstant(16), MEM_OFFSET_SXTX, 1);
+      Result.High = _VLoadVectorMasked(OpSize::i128Bit, ElementSize, Mask.High, Address, _InlineConstant(16), MemOffsetType::SXTX, 1);
    }
    AVX128_StoreResult_WithOpSize(Op, Op->Dest, Result);
  }
@@ -1616,11 +1621,11 @@ void OpDispatchBuilder::AVX128_MASKMOV(OpcodeArgs) {
  // RDI source (DS prefix by default)
  auto MemDest = MakeSegmentAddress(X86State::REG_RDI, Op->Flags, X86Tables::DecodeFlags::FLAG_DS_PREFIX);

-  Ref XMMReg = _LoadMem(FPRClass, Size, MemDest, OpSize::i8Bit);
+  Ref XMMReg = _LoadMemFPR(Size, MemDest, OpSize::i8Bit);

  // If the Mask element high bit is set then overwrite the element with the source, else keep the memory variant
  XMMReg = _VBSL(Size, MaskSrc.Low, VectorSrc.Low, XMMReg);
-  _StoreMem(FPRClass, Size, MemDest, XMMReg, OpSize::i8Bit);
+  _StoreMemFPR(Size, MemDest, XMMReg, OpSize::i8Bit);
 }

 void OpDispatchBuilder::AVX128_VectorVariableBlend(OpcodeArgs, IR::OpSize ElementSize) {
@@ -1660,7 +1665,7 @@ void OpDispatchBuilder::AVX128_SaveAVXState(Ref MemBase) {

  for (uint32_t i = 0; i < NumRegs; i += 2) {
    RefPair Pair = LoadContextPair(OpSize::i128Bit, AVXHigh0Index + i);
-    _StoreMemPair(FPRClass, OpSize::i128Bit, Pair.Low, Pair.High, MemBase, i * 16 + 576);
+    _StoreMemPairFPR(OpSize::i128Bit, Pair.Low, Pair.High, MemBase, i * 16 + 576);
  }
 }

@@ -1668,7 +1673,7 @@ void OpDispatchBuilder::AVX128_RestoreAVXState(Ref MemBase) {
  const auto NumRegs = Is64BitMode ? 16U : 8U;

  for (uint32_t i = 0; i < NumRegs; i += 2) {
-    auto YMMHRegs = LoadMemPair(FPRClass, OpSize::i128Bit, MemBase, i * 16 + 576);
+    auto YMMHRegs = LoadMemPairFPR(OpSize::i128Bit, MemBase, i * 16 + 576);

    AVX128_StoreXMMRegister(i, YMMHRegs.Low, true);
    AVX128_StoreXMMRegister(i + 1, YMMHRegs.High, true);
@@ -1960,7 +1965,7 @@ void OpDispatchBuilder::AVX128_VFMAImpl(OpcodeArgs, IROps IROp, uint8_t Src1Idx,
 }

 void OpDispatchBuilder::AVX128_VFMAScalarImpl(OpcodeArgs, IROps IROp, uint8_t Src1Idx, uint8_t Src2Idx, uint8_t AddendIdx) {
-  const auto SrcSize = OpSizeFromSrc(Op);
+  const OpSize ElementSize = Op->Flags & X86Tables::DecodeFlags::FLAG_OPTION_AVX_W ? OpSize::i64Bit : OpSize::i32Bit;

  auto Dest = AVX128_LoadSource_WithOpSize(Op, Op->Dest, Op->Flags, false).Low;
  auto Src1 = AVX128_LoadSource_WithOpSize(Op, Op->Src[0], Op->Flags, false).Low;
@@ -1968,13 +1973,13 @@ void OpDispatchBuilder::AVX128_VFMAScalarImpl(OpcodeArgs, IROps IROp, uint8_t Sr
  if (Op->Src[1].IsGPR()) {
    Src2 = AVX128_LoadSource_WithOpSize(Op, Op->Src[1], Op->Flags, false).Low;
  } else {
-    Src2 = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], SrcSize, Op->Flags);
+    Src2 = LoadSourceFPR_WithOpSize(Op, Op->Src[1], ElementSize, Op->Flags);
  }

  Ref Sources[3] = {Dest, Src1, Src2};

  DeriveOp(Result_Low, IROp,
-           _VFMLAScalarInsert(OpSize::i128Bit, SrcSize, Dest, Sources[Src1Idx - 1], Sources[Src2Idx - 1], Sources[AddendIdx - 1]));
+           _VFMLAScalarInsert(OpSize::i128Bit, ElementSize, Dest, Sources[Src1Idx - 1], Sources[Src2Idx - 1], Sources[AddendIdx - 1]));
  AVX128_StoreResult_WithOpSize(Op, Op->Dest, AVX128_Zext(Result_Low));
 }

@@ -2016,8 +2021,8 @@ void OpDispatchBuilder::AVX128_VFMAddSubImpl(OpcodeArgs, bool AddSub, uint8_t Sr
  AVX128_StoreResult_WithOpSize(Op, Op->Dest, Result);
 }

-OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherImpl(OpSize Size, OpSize ElementLoadSize, OpSize AddrElementSize, RefPair Dest,
-                                                                  RefPair Mask, RefVSIB VSIB) {
+OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherImpl(OpcodeArgs, OpSize Size, OpSize ElementLoadSize, OpSize AddrElementSize,
+                                                                  RefPair Dest, RefPair Mask, RefVSIB VSIB) {
  LOGMAN_THROW_A_FMT(AddrElementSize == OpSize::i32Bit || AddrElementSize == OpSize::i64Bit, "Unknown address element size");
  const auto Is128Bit = Size == OpSize::i128Bit;

@@ -2061,10 +2066,13 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherImpl(OpSize Size, O
    }
  }

+  const auto GPRSize = GetGPROpSize();
+  auto AddrSize = (Op->Flags & X86Tables::DecodeFlags::FLAG_ADDRESS_SIZE) != 0 ? (GPRSize >> 1) : GPRSize;
+
  RefPair Result {};
  ///< Calculate the low-half.
  Result.Low = _VLoadVectorGatherMasked(OpSize::i128Bit, ElementLoadSize, Dest.Low, Mask.Low, BaseAddr, VSIB.Low, VSIB.High,
-                                        AddrElementSize, VSIB.Scale, 0, 0);
+                                        AddrElementSize, VSIB.Scale, 0, 0, AddrSize);

  if (Is128Bit) {
    Result.High = LoadZeroVector(OpSize::i128Bit);
@@ -2101,7 +2109,7 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherImpl(OpSize Size, O

    ///< Calculate the high-half.
    auto ResultHigh = _VLoadVectorGatherMasked(OpSize::i128Bit, ElementLoadSize, DestReg, MaskReg, BaseAddr, AddrAddressing.Low,
-                                               AddrAddressing.High, AddrElementSize, VSIB.Scale, DataElementOffset, IndexElementOffset);
+                                               AddrAddressing.High, AddrElementSize, VSIB.Scale, DataElementOffset, IndexElementOffset, AddrSize);

    if (AddrElementSize == OpSize::i64Bit && ElementLoadSize == OpSize::i32Bit) {
      // If we only fetched 128-bits worth of data then the upper-result is all zero.
@@ -2114,7 +2122,7 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherImpl(OpSize Size, O
  return Result;
 }

-OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherQPSImpl(Ref Dest, Ref Mask, RefVSIB VSIB) {
+OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherQPSImpl(OpcodeArgs, Ref Dest, Ref Mask, RefVSIB VSIB) {

  ///< BaseAddr doesn't need to exist, calculate that here.
  Ref BaseAddr = VSIB.BaseAddr;
@@ -2142,8 +2150,11 @@ OpDispatchBuilder::RefPair OpDispatchBuilder::AVX128_VPGatherQPSImpl(Ref Dest, R

  RefPair Result {};

+  const auto GPRSize = GetGPROpSize();
+  auto AddrSize = (Op->Flags & X86Tables::DecodeFlags::FLAG_ADDRESS_SIZE) != 0 ? (GPRSize >> 1) : GPRSize;
+
  ///< Calculate the low-half.
-  Result.Low = _VLoadVectorGatherMaskedQPS(OpSize::i128Bit, OpSize::i32Bit, Dest, Mask, BaseAddr, VSIB.Low, VSIB.High, VSIB.Scale);
+  Result.Low = _VLoadVectorGatherMaskedQPS(OpSize::i128Bit, OpSize::i32Bit, Dest, Mask, BaseAddr, VSIB.Low, VSIB.High, VSIB.Scale, AddrSize);
  Result.High = LoadZeroVector(OpSize::i128Bit);
  if (VSIB.High == Invalid()) {
    // Special case for only loading two floats.
@@ -2202,15 +2213,15 @@ void OpDispatchBuilder::AVX128_VPGATHER(OpcodeArgs, OpSize AddrElementSize) {
    }

    ///< AddressElementSize is now OpSize::i64Bit
-    Result = AVX128_VPGatherQPSImpl(Dest.Low, Mask.Low, VSIBLow);
+    Result = AVX128_VPGatherQPSImpl(Op, Dest.Low, Mask.Low, VSIBLow);
    if (NeedsHighAddrBytes) {
-      auto Res = AVX128_VPGatherQPSImpl(Dest.High, Mask.High, VSIBHigh);
+      auto Res = AVX128_VPGatherQPSImpl(Op, Dest.High, Mask.High, VSIBHigh);
      Result.High = Res.Low;
    }
  } else if (AddrElementSize == OpSize::i64Bit && ElementLoadSize == OpSize::i32Bit) {
-    Result = AVX128_VPGatherQPSImpl(Dest.Low, Mask.Low, VSIB);
+    Result = AVX128_VPGatherQPSImpl(Op, Dest.Low, Mask.Low, VSIB);
  } else {
-    Result = AVX128_VPGatherImpl(Size, ElementLoadSize, AddrElementSize, Dest, Mask, VSIB);
+    Result = AVX128_VPGatherImpl(Op, Size, ElementLoadSize, AddrElementSize, Dest, Mask, VSIB);
  }
  AVX128_StoreResult_WithOpSize(Op, Op->Dest, Result);

@@ -2234,7 +2245,7 @@ void OpDispatchBuilder::AVX128_VCVTPH2PS(OpcodeArgs) {
    // In the event that a memory operand is used as the source operand,
    // the access width will always be half the size of the destination vector width
    // (i.e. 128-bit vector -> 64-bit mem, 256-bit vector -> 128-bit mem)
-    Src.Low = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], SrcSize, Op->Flags);
+    Src.Low = LoadSourceFPR_WithOpSize(Op, Op->Src[0], SrcSize, Op->Flags);
  }

  RefPair Result {};
@@ -2289,7 +2300,7 @@ void OpDispatchBuilder::AVX128_VCVTPS2PH(OpcodeArgs) {
  }

  if (!Op->Dest.IsGPR()) {
-    StoreResult_WithOpSize(FPRClass, Op, Op->Dest, Result.Low, StoreSize, OpSize::iInvalid);
+    StoreResultFPR_WithOpSize(Op, Op->Dest, Result.Low, StoreSize);
  } else {
    AVX128_StoreResult_WithOpSize(Op, Op->Dest, Result);
  }
@@ -53,7 +53,7 @@ constexpr inline DispatchTableEntry OpDispatch_BaseOpTable[] = {
  {0xAA, 2, &OpDispatchBuilder::STOSOp},
  {0xAC, 2, &OpDispatchBuilder::LODSOp},
  {0xAE, 2, &OpDispatchBuilder::SCASOp},
-  {0xB0, 16, &OpDispatchBuilder::Bind<&OpDispatchBuilder::MOVGPRImmediate>},
+  {0xB0, 16, &OpDispatchBuilder::Bind<&OpDispatchBuilder::MOVGPROp, 0>},
  {0xC2, 2, &OpDispatchBuilder::RETOp},
  {0xC8, 1, &OpDispatchBuilder::EnterOp},
  {0xC9, 1, &OpDispatchBuilder::LEAVEOp},
@@ -23,8 +23,8 @@ void OpDispatchBuilder::SHA1NEXTEOp(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);

  // ARMv8 SHA1 extension provides a `SHA1H` instruction which does a fixed rotate by 30.
  // This only operates on element 0 rather than element 3. We don't have the luxury of rewriting the x86 SHA algorithm to take advantage of this.
@@ -36,7 +36,7 @@ void OpDispatchBuilder::SHA1NEXTEOp(OpcodeArgs) {
  auto Tmp = _VAdd(OpSize::i128Bit, OpSize::i32Bit, Src, RotatedNode);
  auto Result = _VInsElement(OpSize::i128Bit, OpSize::i32Bit, 3, 3, Src, Tmp);

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::SHA1MSG1Op(OpcodeArgs) {
@@ -44,15 +44,15 @@ void OpDispatchBuilder::SHA1MSG1Op(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);

  Ref NewVec = _VExtr(OpSize::i128Bit, OpSize::i64Bit, Dest, Src, 1);

  // [W0, W1, W2, W3] ^ [W2, W3, W4, W5]
  Ref Result = _VXor(OpSize::i128Bit, OpSize::i8Bit, Dest, NewVec);

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::SHA1MSG2Op(OpcodeArgs) {
@@ -60,8 +60,8 @@ void OpDispatchBuilder::SHA1MSG2Op(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);

  // ARM SHA1 mostly matches x86 semantics, except the input and outputs are both flipped from elements 0,1,2,3 to 3,2,1,0.
  auto Src1 = SHADataShuffle(Dest);
@@ -70,7 +70,7 @@ void OpDispatchBuilder::SHA1MSG2Op(OpcodeArgs) {
  // The result is swizzled differently than expected
  auto Result = SHADataShuffle(_VSha1SU1(Src1, Src2));

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::SHA1RNDS4Op(OpcodeArgs) {
@@ -79,8 +79,8 @@ void OpDispatchBuilder::SHA1RNDS4Op(OpcodeArgs) {
    return;
  }
  const uint64_t Imm8 = Op->Src[1].Literal() & 0b11;
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);

  Ref Result {};
  Ref ConstantVector {};
@@ -112,7 +112,7 @@ void OpDispatchBuilder::SHA1RNDS4Op(OpcodeArgs) {
  case 3: Result = SHADataShuffle(_VSha1P(Src1, ZeroRegister, Src2)); break;
  }

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::SHA256MSG1Op(OpcodeArgs) {
@@ -120,12 +120,12 @@ void OpDispatchBuilder::SHA256MSG1Op(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);

  auto Result = _VSha256U0(Dest, Src);

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::SHA256MSG2Op(OpcodeArgs) {
@@ -133,8 +133,8 @@ void OpDispatchBuilder::SHA256MSG2Op(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);

  auto Src1 = _VExtr(OpSize::i128Bit, OpSize::i32Bit, Dest, Dest, 3);
  auto DupDst = _VDupElement(OpSize::i128Bit, OpSize::i32Bit, Dest, 3);
@@ -142,7 +142,7 @@ void OpDispatchBuilder::SHA256MSG2Op(OpcodeArgs) {

  auto Result = _VSha256U1(Src1, Src2);

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::SHA256RNDS2Op(OpcodeArgs) {
@@ -150,8 +150,8 @@ void OpDispatchBuilder::SHA256RNDS2Op(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  // Hardcoded to XMM0
  auto XMM0 = LoadXMMRegister(0);

@@ -177,7 +177,7 @@ void OpDispatchBuilder::SHA256RNDS2Op(OpcodeArgs) {
  auto B = _VSha256H2(EFGH, ABCD, Key);
  auto Result = shuffle_abcd(A, B);

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::AESImcOp(OpcodeArgs) {
@@ -185,9 +185,9 @@ void OpDispatchBuilder::AESImcOp(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  Ref Result = _VAESImc(Src);
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::AESEncOp(OpcodeArgs) {
@@ -195,10 +195,10 @@ void OpDispatchBuilder::AESEncOp(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  Ref Result = _VAESEnc(OpSize::i128Bit, Dest, Src, LoadZeroVector(OpSize::i128Bit));
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::VAESEncOp(OpcodeArgs) {
@@ -208,11 +208,11 @@ void OpDispatchBuilder::VAESEncOp(OpcodeArgs) {
  // TODO: Handle 256-bit VAESENC.
  LOGMAN_THROW_A_FMT(Is128Bit, "256-bit VAESENC unimplemented");

-  Ref State = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
-  Ref Key = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+  Ref State = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
+  Ref Key = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
  Ref Result = _VAESEnc(DstSize, State, Key, LoadZeroVector(DstSize));

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::AESEncLastOp(OpcodeArgs) {
@@ -220,10 +220,10 @@ void OpDispatchBuilder::AESEncLastOp(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  Ref Result = _VAESEncLast(OpSize::i128Bit, Dest, Src, LoadZeroVector(OpSize::i128Bit));
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::VAESEncLastOp(OpcodeArgs) {
@@ -233,11 +233,11 @@ void OpDispatchBuilder::VAESEncLastOp(OpcodeArgs) {
  // TODO: Handle 256-bit VAESENCLAST.
  LOGMAN_THROW_A_FMT(Is128Bit, "256-bit VAESENCLAST unimplemented");

-  Ref State = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
-  Ref Key = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+  Ref State = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
+  Ref Key = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
  Ref Result = _VAESEncLast(DstSize, State, Key, LoadZeroVector(DstSize));

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::AESDecOp(OpcodeArgs) {
@@ -245,10 +245,10 @@ void OpDispatchBuilder::AESDecOp(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  Ref Result = _VAESDec(OpSize::i128Bit, Dest, Src, LoadZeroVector(OpSize::i128Bit));
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::VAESDecOp(OpcodeArgs) {
@@ -258,11 +258,11 @@ void OpDispatchBuilder::VAESDecOp(OpcodeArgs) {
  // TODO: Handle 256-bit VAESDEC.
  LOGMAN_THROW_A_FMT(Is128Bit, "256-bit VAESDEC unimplemented");

-  Ref State = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
-  Ref Key = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+  Ref State = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
+  Ref Key = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
  Ref Result = _VAESDec(DstSize, State, Key, LoadZeroVector(DstSize));

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::AESDecLastOp(OpcodeArgs) {
@@ -270,10 +270,10 @@ void OpDispatchBuilder::AESDecLastOp(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  Ref Result = _VAESDecLast(OpSize::i128Bit, Dest, Src, LoadZeroVector(OpSize::i128Bit));
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::VAESDecLastOp(OpcodeArgs) {
@@ -283,15 +283,15 @@ void OpDispatchBuilder::VAESDecLastOp(OpcodeArgs) {
  // TODO: Handle 256-bit VAESDECLAST.
  LOGMAN_THROW_A_FMT(Is128Bit, "256-bit VAESDECLAST unimplemented");

-  Ref State = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
-  Ref Key = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+  Ref State = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
+  Ref Key = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
  Ref Result = _VAESDecLast(DstSize, State, Key, LoadZeroVector(DstSize));

-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 Ref OpDispatchBuilder::AESKeyGenAssistImpl(OpcodeArgs) {
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  const uint64_t RCON = Op->Src[1].Literal();

  auto KeyGenSwizzle = LoadAndCacheNamedVectorConstant(OpSize::i128Bit, NAMED_VECTOR_AESKEYGENASSIST_SWIZZLE);
@@ -305,7 +305,7 @@ void OpDispatchBuilder::AESKeyGenAssist(OpcodeArgs) {
  }

  Ref Result = AESKeyGenAssistImpl(Op);
-  StoreResult(FPRClass, Op, Result, OpSize::iInvalid);
+  StoreResultFPR(Op, Result);
 }

 void OpDispatchBuilder::PCLMULQDQOp(OpcodeArgs) {
@@ -313,12 +313,12 @@ void OpDispatchBuilder::PCLMULQDQOp(OpcodeArgs) {
    UnimplementedOp(Op);
    return;
  }
-  Ref Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
-  Ref Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+  Ref Dest = LoadSourceFPR(Op, Op->Dest, Op->Flags);
+  Ref Src = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  const auto Selector = static_cast<uint8_t>(Op->Src[1].Literal());

  auto Res = _PCLMUL(OpSize::i128Bit, Dest, Src, Selector & 0b1'0001);
-  StoreResult(FPRClass, Op, Res, OpSize::iInvalid);
+  StoreResultFPR(Op, Res);
 }

 void OpDispatchBuilder::VPCLMULQDQOp(OpcodeArgs) {
@@ -328,12 +328,12 @@ void OpDispatchBuilder::VPCLMULQDQOp(OpcodeArgs) {
  }
  const auto DstSize = OpSizeFromDst(Op);

-  Ref Src1 = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
-  Ref Src2 = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
+  Ref Src1 = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
+  Ref Src2 = LoadSourceFPR(Op, Op->Src[1], Op->Flags);
  const auto Selector = static_cast<uint8_t>(Op->Src[2].Literal());

  Ref Res = _PCLMUL(DstSize, Src1, Src2, Selector & 0b1'0001);
-  StoreResult(FPRClass, Op, Res, OpSize::iInvalid);
+  StoreResultFPR(Op, Res);
 }

 } // namespace FEXCore::IR
@@ -263,7 +263,7 @@ void OpDispatchBuilder::CalculateDeferredFlags() {
 Ref OpDispatchBuilder::IncrementByCarry(OpSize OpSize, Ref Src) {
  // If CF not inverted, we use .cc since the increment happens when the
  // condition is false. If CF inverted, invert to use .cs. A bit mindbendy.
-  return _NZCVSelectIncrement(OpSize, {CFInverted ? COND_UGE : COND_ULT}, Src, Src);
+  return _NZCVSelectIncrement(OpSize, CFInverted ? CondClass::UGE : CondClass::ULT, Src, Src);
 }

 Ref OpDispatchBuilder::CalculateFlags_ADC(IR::OpSize SrcSize, Ref Src1, Ref Src2) {
@@ -290,7 +290,7 @@ Ref OpDispatchBuilder::CalculateFlags_ADC(IR::OpSize SrcSize, Ref Src1, Ref Src2
    Res = _Bfe(OpSize, IR::OpSizeAsBits(SrcSize), 0, Res);

    // TODO: We can fold that second Bfe in (cmp uxth).
-    auto SelectCFInv = Select01(OpSize, CondClassType {COND_UGE}, Res, Src2PlusCF);
+    auto SelectCFInv = Select01(OpSize, CondClass::UGE, Res, Src2PlusCF);

    SetNZ_ZeroCV(SrcSize, Res);
    SetCFInverted(SelectCFInv);
@@ -324,7 +324,7 @@ Ref OpDispatchBuilder::CalculateFlags_SBB(IR::OpSize SrcSize, Ref Src1, Ref Src2
    Res = Sub(OpSize, Src1, Src2PlusCF);
    Res = _Bfe(OpSize, IR::OpSizeAsBits(SrcSize), 0, Res);

-    auto SelectCFInv = Select01(OpSize, CondClassType {COND_UGE}, Src1, Src2PlusCF);
+    auto SelectCFInv = Select01(OpSize, CondClass::UGE, Src1, Src2PlusCF);

    SetNZ_ZeroCV(SrcSize, Res);
    SetCFInverted(SelectCFInv);
@@ -406,7 +406,7 @@ void OpDispatchBuilder::CalculateFlags_MUL(IR::OpSize SrcSize, Ref Res, Ref High
  // If High = SignBit, then sets to nZCv. Else sets to nzcV. Since SF/ZF
  // undefined, this does what we need after inverting carry.
  auto Zero = _InlineConstant(0);
-  _CondSubNZCV(OpSize::i64Bit, Zero, Zero, CondClassType {COND_EQ}, 0x1 /* nzcV */);
+  _CondSubNZCV(OpSize::i64Bit, Zero, Zero, CondClass::EQ, 0x1 /* nzcV */);
  CFInverted = true;
 }

@@ -423,7 +423,7 @@ void OpDispatchBuilder::CalculateFlags_UMUL(Ref High) {

  // If High = 0, then sets to nZCv. Else sets to nzcV. Since SF/ZF undefined,
  // this does what we need.
-  _CondSubNZCV(Size, Zero, Zero, CondClassType {COND_EQ}, 0x1 /* nzcV */);
+  _CondSubNZCV(Size, Zero, Zero, CondClass::EQ, 0x1 /* nzcV */);
  CFInverted = true;
 }

@@ -151,6 +151,9 @@ constexpr DispatchTableEntry OpDispatch_SecondaryGroupTables[] = {
  {OPD(FEXCore::X86Tables::TYPE_GROUP_16, PF_F2, 3), 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::Prefetch, false, false, 3>},
  {OPD(FEXCore::X86Tables::TYPE_GROUP_16, PF_F2, 4), 4, &OpDispatchBuilder::NOPOp},

+  // GROUP 17
+  {OPD(FEXCore::X86Tables::TYPE_GROUP_17, PF_66, 0), 1, &OpDispatchBuilder::Extrq_imm},
+
  // GROUP P
  {OPD(FEXCore::X86Tables::TYPE_GROUP_P, PF_NONE, 0), 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::Prefetch, false, false, 1>},
  {OPD(FEXCore::X86Tables::TYPE_GROUP_P, PF_NONE, 1), 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::Prefetch, true, false, 1>},
@@ -145,7 +145,7 @@ constexpr DispatchTableEntry OpDispatch_TwoByteOpTable[] = {

 #ifndef _WIN32
  // FEX reserved instructions
-  {0x37, 1, &OpDispatchBuilder::CallbackReturnOp},
+  {0x3E, 1, &OpDispatchBuilder::CallbackReturnOp},
  {0x3F, 1, &OpDispatchBuilder::ThunkOp},
 #endif
 };
@@ -198,6 +198,8 @@ constexpr DispatchTableEntry OpDispatch_SecondaryRepNEModTables[] = {
  {0x5E, 1, &OpDispatchBuilder::VectorScalarInsertALUOp<IR::OP_VFDIVSCALARINSERT, OpSize::i64Bit>},
  {0x5F, 1, &OpDispatchBuilder::VectorScalarInsertALUOp<IR::OP_VFMAXSCALARINSERT, OpSize::i64Bit>},
  {0x70, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::PSHUFWOp, true>},
+  {0x78, 1, &OpDispatchBuilder::Insertq_imm},
+  {0x79, 1, &OpDispatchBuilder::Insertq},
  {0x7C, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::VectorALUOp, IR::OP_VFADDP, OpSize::i32Bit>},
  {0x7D, 1, &OpDispatchBuilder::HSUBP<OpSize::i32Bit>},
  {0xD0, 1, &OpDispatchBuilder::ADDSUBPOp<OpSize::i32Bit>},
@@ -256,6 +258,7 @@ constexpr DispatchTableEntry OpDispatch_SecondaryOpSizeModTables[] = {
  {0x75, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::VectorALUOp, IR::OP_VCMPEQ, OpSize::i16Bit>},
  {0x76, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::VectorALUOp, IR::OP_VCMPEQ, OpSize::i32Bit>},
  {0x78, 1, nullptr}, // GROUP 17
+  {0x79, 1, &OpDispatchBuilder::Extrq},
  {0x7C, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::VectorALUOp, IR::OP_VFADDP, OpSize::i64Bit>},
  {0x7D, 1, &OpDispatchBuilder::HSUBP<OpSize::i64Bit>},
  {0x7E, 1, &OpDispatchBuilder::Bind<&OpDispatchBuilder::MOVBetweenGPR_FPR, OpDispatchBuilder::VectorOpType::SSE>},
@@ -17,7 +17,6 @@ $end_info$
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/FPState.h>

-#include <cmath>
 #include <stddef.h>
 #include <stdint.h>

@@ -28,7 +27,7 @@ class OrderedNode;
 Ref OpDispatchBuilder::GetX87Top() {
  // Yes, we are storing 3 bits in a single flag register.
  // Deal with it
-  return _LoadContext(OpSize::i8Bit, GPRClass, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
+  return _LoadContextGPR(OpSize::i8Bit, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
 }

 void OpDispatchBuilder::SetX87FTW(Ref FTW) {
@@ -52,24 +51,22 @@ void OpDispatchBuilder::SetX87FTW(Ref FTW) {
  FTW = _Orlshr(OpSize::i32Bit, FTW, FTW, 4);

  // ...and that's it. StoreContext implicitly does the final masking.
-  _StoreContext(OpSize::i8Bit, GPRClass, FTW, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
+  _StoreContextGPR(OpSize::i8Bit, FTW, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
 }

 void OpDispatchBuilder::SetX87Top(Ref Value) {
-  _StoreContext(OpSize::i8Bit, GPRClass, Value, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
+  _StoreContextGPR(OpSize::i8Bit, Value, offsetof(FEXCore::Core::CPUState, flags) + FEXCore::X86State::X87FLAG_TOP_LOC);
 }

 // Float LoaD operation with memory operand
 void OpDispatchBuilder::FLD(OpcodeArgs, IR::OpSize Width) {
-  const auto ReadWidth = (Width == OpSize::f80Bit) ? OpSize::i128Bit : Width;
-
-  Ref Data = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], Width, Op->Flags);
+  Ref Data = LoadSourceFPR_WithOpSize(Op, Op->Src[0], Width, Op->Flags);
  Ref ConvertedData = Data;
  // Convert to 80bit float
  if (Width == OpSize::i32Bit || Width == OpSize::i64Bit) {
-    ConvertedData = _F80CVTTo(Data, ReadWidth);
+    ConvertedData = _F80CVTTo(Data, Width);
  }
-  _PushStack(ConvertedData, Data, ReadWidth, true);
+  _PushStack(ConvertedData, Data, Width);
 }

 // Float LoaD operation with memory operand
@@ -79,27 +76,27 @@ void OpDispatchBuilder::FLDFromStack(OpcodeArgs) {

 void OpDispatchBuilder::FBLD(OpcodeArgs) {
  // Read from memory
-  Ref Data = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], OpSize::f80Bit, Op->Flags);
+  Ref Data = LoadSourceFPR_WithOpSize(Op, Op->Src[0], OpSize::f80Bit, Op->Flags);
  Ref ConvertedData = _F80BCDLoad(Data);
-  _PushStack(ConvertedData, Data, OpSize::i128Bit, true);
+  _PushStack(ConvertedData, Invalid(), OpSize::iInvalid);
 }

 void OpDispatchBuilder::FBSTP(OpcodeArgs) {
  Ref converted = _F80BCDStore(_ReadStackValue(0));
-  StoreResult_WithOpSize(FPRClass, Op, Op->Dest, converted, OpSize::f80Bit, OpSize::i8Bit);
+  StoreResultFPR_WithOpSize(Op, Op->Dest, converted, OpSize::f80Bit, OpSize::i8Bit);
  _PopStackDestroy();
 }

 void OpDispatchBuilder::FLD_Const(OpcodeArgs, NamedVectorConstant K) {
  // Update TOP
  Ref Data = LoadAndCacheNamedVectorConstant(OpSize::i128Bit, K);
-  _PushStack(Data, Data, OpSize::i128Bit, true);
+  _PushStack(Data, Data, OpSize::f80Bit);
 }

 void OpDispatchBuilder::FILD(OpcodeArgs) {
  const auto ReadWidth = OpSizeFromSrc(Op);
  // Read from memory
-  Ref Data = LoadSource_WithOpSize(GPRClass, Op, Op->Src[0], ReadWidth, Op->Flags);
+  Ref Data = LoadSourceGPR_WithOpSize(Op, Op->Src[0], ReadWidth, Op->Flags);

  // Sign extend to 64bits
  if (ReadWidth != OpSize::i64Bit) {
@@ -112,27 +109,28 @@ void OpDispatchBuilder::FILD(OpcodeArgs) {
  // Extract sign and make integer absolute
  auto zero = Constant(0);
  _SubNZCV(OpSize::i64Bit, Data, zero);
-  auto sign = _NZCVSelect(OpSize::i64Bit, CondClassType {COND_SLT}, Constant(0x8000), zero);
-  auto absolute = _Neg(OpSize::i64Bit, Data, CondClassType {COND_MI});
+  auto sign = _NZCVSelect(OpSize::i64Bit, CondClass::SLT, Constant(0x8000), zero);
+  auto absolute = _Neg(OpSize::i64Bit, Data, CondClass::MI);

  // left justify the absolute integer
  auto shift = Sub(OpSize::i64Bit, Constant(63), _FindMSB(IR::OpSize::i64Bit, absolute));
  auto shifted = _Lshl(OpSize::i64Bit, absolute, shift);

  auto adjusted_exponent = Sub(OpSize::i64Bit, Constant(0x3fff + 63), shift);
-  auto zeroed_exponent = _Select(COND_EQ, absolute, zero, zero, adjusted_exponent);
+  auto zeroed_exponent = _Select(OpSize::i64Bit, OpSize::i64Bit, CondClass::EQ, absolute, zero, zero, adjusted_exponent);
  auto upper = _Or(OpSize::i64Bit, sign, zeroed_exponent);

  Ref ConvertedData = _VLoadTwoGPRs(shifted, upper);
-  _PushStack(ConvertedData, Data, ReadWidth, false);
+  _PushStack(ConvertedData, Invalid(), OpSize::iInvalid);
 }

 void OpDispatchBuilder::FST(OpcodeArgs, IR::OpSize Width) {
-  const auto SourceSize = ReducedPrecisionMode ? OpSize::i64Bit : OpSize::i128Bit;
+  LOGMAN_THROW_A_FMT(Width == OpSize::i32Bit || Width == OpSize::i64Bit || Width == OpSize::f80Bit, "Invalid store width for FST");
+  const auto SourceSize = ReducedPrecisionMode ? OpSize::i64Bit : OpSize::f80Bit;
  AddressMode A = DecodeAddress(Op, Op->Dest, MemoryAccessType::DEFAULT, false);

  A = SelectAddressMode(this, A, GetGPROpSize(), CTX->HostFeatures.SupportsTSOImm9, false, false, Width);
-  _StoreStackMem(SourceSize, Width, A.Base, A.Index, OpSize::iInvalid, A.IndexType, A.IndexScale, /*Float=*/true);
+  _StoreStackMem(SourceSize, Width, A.Base, A.Index, OpSize::iInvalid, A.IndexType, A.IndexScale);

  if (Op->TableInfo->Flags & X86Tables::InstFlags::FLAGS_POP) {
    _PopStackDestroy();
@@ -166,12 +164,12 @@ void OpDispatchBuilder::FIST(OpcodeArgs, bool Truncate) {
    // Check for NaN/Infinity: exponent = 0x7fff
    SaveNZCV();
    _TestNZ(OpSize::i64Bit, Exponent, Constant(0x7fff));
-    Ref IsSpecial = _NZCVSelect01({COND_EQ});
+    Ref IsSpecial = _NZCVSelect01(CondClass::EQ);

    // For overflow detection, check if exponent indicates a value >= 2^15
    // Biased exponent for 2^15 is 0x3fff + 15 = 0x400e
    SubWithFlags(OpSize::i64Bit, Exponent, 0x400e);
-    Ref IsOverflow = _NZCVSelect01({COND_UGE});
+    Ref IsOverflow = _NZCVSelect01(CondClass::UGE);

    // Set Invalid Operation flag if overflow or special value
    Ref InvalidFlag = _Or(OpSize::i64Bit, IsSpecial, IsOverflow);
@@ -180,7 +178,7 @@ void OpDispatchBuilder::FIST(OpcodeArgs, bool Truncate) {

  Data = _F80CVTInt(Size, Data, Truncate);

-  StoreResult_WithOpSize(GPRClass, Op, Op->Dest, Data, Size, OpSize::i8Bit);
+  StoreResultGPR_WithOpSize(Op, Op->Dest, Data, Size, OpSize::i8Bit);

  if ((Op->TableInfo->Flags & X86Tables::InstFlags::FLAGS_POP) != 0) {
    _PopStackDestroy();
@@ -206,10 +204,10 @@ void OpDispatchBuilder::FADD(OpcodeArgs, IR::OpSize Width, bool Integer, OpDispa
  // We have one memory argument
  Ref Arg {};
  if (Integer) {
-    Arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    Arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
    Arg = _F80CVTToInt(Arg, Width);
  } else {
-    Arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    Arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    Arg = _F80CVTTo(Arg, Width);
  }

@@ -236,10 +234,10 @@ void OpDispatchBuilder::FMUL(OpcodeArgs, IR::OpSize Width, bool Integer, OpDispa
  // We have one memory argument
  Ref arg {};
  if (Integer) {
-    arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
    arg = _F80CVTToInt(arg, Width);
  } else {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    arg = _F80CVTTo(arg, Width);
  }

@@ -273,10 +271,10 @@ void OpDispatchBuilder::FDIV(OpcodeArgs, IR::OpSize Width, bool Integer, bool Re
  // We have one memory argument
  Ref arg {};
  if (Integer) {
-    arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
    arg = _F80CVTToInt(arg, Width);
  } else {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    arg = _F80CVTTo(arg, Width);
  }

@@ -314,10 +312,10 @@ void OpDispatchBuilder::FSUB(OpcodeArgs, IR::OpSize Width, bool Integer, bool Re
  // We have one memory argument
  Ref Arg {};
  if (Integer) {
-    Arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    Arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
    Arg = _F80CVTToInt(Arg, Width);
  } else {
-    Arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    Arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    Arg = _F80CVTTo(Arg, Width);
  }

@@ -340,7 +338,7 @@ Ref OpDispatchBuilder::GetX87FTW_Helper() {
  // bytes, we use the well-known bit twiddling algorithm:
  //
  // https://graphics.stanford.edu/~seander/bithacks.html#InterleaveBMN
-  Ref X = _LoadContext(OpSize::i8Bit, GPRClass, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
+  Ref X = _LoadContextGPR(OpSize::i8Bit, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
  X = _Orlshl(OpSize::i32Bit, X, X, 4);
  X = _And(OpSize::i32Bit, X, Constant(0x0f0f0f0f));
  X = _Orlshl(OpSize::i32Bit, X, X, 2);
@@ -381,41 +379,41 @@ void OpDispatchBuilder::X87FNSTENV(OpcodeArgs) {
  _SyncStackToSlow();

  const auto Size = OpSizeFromSrc(Op);
-  Ref Mem = LoadSource(GPRClass, Op, Op->Dest, Op->Flags, {.LoadData = false});
+  Ref Mem = LoadSourceGPR(Op, Op->Dest, Op->Flags, {.LoadData = false});
  Mem = AppendSegmentOffset(Mem, Op->Flags);

  {
-    auto FCW = _LoadContext(OpSize::i16Bit, GPRClass, offsetof(FEXCore::Core::CPUState, FCW));
-    _StoreMem(GPRClass, Size, Mem, FCW, Size);
+    auto FCW = _LoadContextGPR(OpSize::i16Bit, offsetof(FEXCore::Core::CPUState, FCW));
+    _StoreMemGPR(Size, Mem, FCW, Size);
  }

-  { _StoreMem(GPRClass, Size, ReconstructFSW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MEM_OFFSET_SXTX, 1); }
+  { _StoreMemGPR(Size, ReconstructFSW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MemOffsetType::SXTX, 1); }

  auto ZeroConst = Constant(0);

  {
    // FTW
-    _StoreMem(GPRClass, Size, GetX87FTW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, GetX87FTW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MemOffsetType::SXTX, 1);
  }

  {
    // Instruction Offset
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 3), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 3), Size, MemOffsetType::SXTX, 1);
  }

  {
    // Instruction CS selector (+ Opcode)
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 4), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 4), Size, MemOffsetType::SXTX, 1);
  }

  {
    // Data pointer offset
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 5), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 5), Size, MemOffsetType::SXTX, 1);
  }

  {
    // Data pointer selector
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 6), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 6), Size, MemOffsetType::SXTX, 1);
  }
 }

@@ -441,20 +439,20 @@ void OpDispatchBuilder::X87LDENV(OpcodeArgs) {
  _StackForceSlow();

  const auto Size = OpSizeFromSrc(Op);
-  Ref Mem = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags, {.LoadData = false});
+  Ref Mem = LoadSourceGPR(Op, Op->Src[0], Op->Flags, {.LoadData = false});
  Mem = AppendSegmentOffset(Mem, Op->Flags);

-  auto NewFCW = _LoadMem(GPRClass, OpSize::i16Bit, Mem, OpSize::i16Bit);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  auto NewFCW = _LoadMemGPR(OpSize::i16Bit, Mem, OpSize::i16Bit);
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));

  Ref MemLocation = Add(OpSize::i64Bit, Mem, IR::OpSizeToSize(Size) * 1);
-  auto NewFSW = _LoadMem(GPRClass, Size, MemLocation, Size);
+  auto NewFSW = _LoadMemGPR(Size, MemLocation, Size);
  ReconstructX87StateFromFSW_Helper(NewFSW);

  {
    // FTW
    Ref MemLocation = Add(OpSize::i64Bit, Mem, IR::OpSizeToSize(Size) * 2);
-    SetX87FTW(_LoadMem(GPRClass, Size, MemLocation, Size));
+    SetX87FTW(_LoadMemGPR(Size, MemLocation, Size));
  }
 }

@@ -483,61 +481,61 @@ void OpDispatchBuilder::X87FNSAVE(OpcodeArgs) {
  Ref Mem = MakeSegmentAddress(Op, Op->Dest);
  Ref Top = GetX87Top();
  {
-    auto FCW = _LoadContext(OpSize::i16Bit, GPRClass, offsetof(FEXCore::Core::CPUState, FCW));
-    _StoreMem(GPRClass, Size, Mem, FCW, Size);
+    auto FCW = _LoadContextGPR(OpSize::i16Bit, offsetof(FEXCore::Core::CPUState, FCW));
+    _StoreMemGPR(Size, Mem, FCW, Size);
  }

-  { _StoreMem(GPRClass, Size, ReconstructFSW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MEM_OFFSET_SXTX, 1); }
+  { _StoreMemGPR(Size, ReconstructFSW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MemOffsetType::SXTX, 1); }

  auto ZeroConst = Constant(0);

  {
    // FTW
-    _StoreMem(GPRClass, Size, GetX87FTW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, GetX87FTW_Helper(), Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MemOffsetType::SXTX, 1);
  }

  {
    // Instruction Offset
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 3), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 3), Size, MemOffsetType::SXTX, 1);
  }

  {
    // Instruction CS selector (+ Opcode)
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 4), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 4), Size, MemOffsetType::SXTX, 1);
  }

  {
    // Data pointer offset
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 5), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 5), Size, MemOffsetType::SXTX, 1);
  }

  {
    // Data pointer selector
-    _StoreMem(GPRClass, Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 6), Size, MEM_OFFSET_SXTX, 1);
+    _StoreMemGPR(Size, ZeroConst, Mem, Constant(IR::OpSizeToSize(Size) * 6), Size, MemOffsetType::SXTX, 1);
  }

  auto SevenConst = Constant(7);
  const auto LoadSize = ReducedPrecisionMode ? OpSize::i64Bit : OpSize::i128Bit;
  for (int i = 0; i < 7; ++i) {
-    Ref data = _LoadContextIndexed(Top, LoadSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit), FPRClass);
+    Ref data = _LoadContextFPRIndexed(Top, LoadSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit));
    if (ReducedPrecisionMode) {
      data = _F80CVTTo(data, OpSize::i64Bit);
    }
-    _StoreMem(FPRClass, OpSize::i128Bit, data, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * i)), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
+    _StoreMemFPR(OpSize::i128Bit, data, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * i)), OpSize::i8Bit, MemOffsetType::SXTX, 1);
    Top = _And(OpSize::i32Bit, Add(OpSize::i32Bit, Top, 1), SevenConst);
  }

  // The final st(7) needs a bit of special handling here
-  Ref data = _LoadContextIndexed(Top, LoadSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit), FPRClass);
+  Ref data = _LoadContextFPRIndexed(Top, LoadSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit));
  if (ReducedPrecisionMode) {
    data = _F80CVTTo(data, OpSize::i64Bit);
  }
  // ST7 broken in to two parts
  // Lower 64bits [63:0]
  // upper 16 bits [79:64]
-  _StoreMem(FPRClass, OpSize::i64Bit, data, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (7 * 10)), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
+  _StoreMemFPR(OpSize::i64Bit, data, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (7 * 10)), OpSize::i8Bit, MemOffsetType::SXTX, 1);
  auto topBytes = _VDupElement(OpSize::i128Bit, OpSize::i16Bit, data, 4);
-  _StoreMem(FPRClass, OpSize::i16Bit, topBytes, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (7 * 10) + 8), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
+  _StoreMemFPR(OpSize::i16Bit, topBytes, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (7 * 10) + 8), OpSize::i8Bit, MemOffsetType::SXTX, 1);

  // reset to default
  FNINIT(Op);
@@ -548,8 +546,8 @@ void OpDispatchBuilder::X87FRSTOR(OpcodeArgs) {
  const auto Size = OpSizeFromSrc(Op);
  Ref Mem = MakeSegmentAddress(Op, Op->Src[0]);

-  auto NewFCW = _LoadMem(GPRClass, OpSize::i16Bit, Mem, OpSize::i16Bit);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  auto NewFCW = _LoadMemGPR(OpSize::i16Bit, Mem, OpSize::i16Bit);
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
  if (ReducedPrecisionMode) {
    // ignore the rounding precision, we're always 64-bit in F64.
    // extract rounding mode
@@ -561,11 +559,11 @@ void OpDispatchBuilder::X87FRSTOR(OpcodeArgs) {
    _SetRoundingMode(roundingMode, false, roundingMode);
  }

-  auto NewFSW = _LoadMem(GPRClass, Size, Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MEM_OFFSET_SXTX, 1);
+  auto NewFSW = _LoadMemGPR(Size, Mem, Constant(IR::OpSizeToSize(Size) * 1), Size, MemOffsetType::SXTX, 1);
  Ref Top = ReconstructX87StateFromFSW_Helper(NewFSW);
  {
    // FTW
-    SetX87FTW(_LoadMem(GPRClass, Size, Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MEM_OFFSET_SXTX, 1));
+    SetX87FTW(_LoadMemGPR(Size, Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MemOffsetType::SXTX, 1));
  }

  auto SevenConst = Constant(7);
@@ -574,14 +572,14 @@ void OpDispatchBuilder::X87FRSTOR(OpcodeArgs) {
  Ref Mask = _VLoadTwoGPRs(low, high);
  const auto StoreSize = ReducedPrecisionMode ? OpSize::i64Bit : OpSize::i128Bit;
  for (int i = 0; i < 7; ++i) {
-    Ref Reg = _LoadMem(FPRClass, OpSize::i128Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * i)), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
+    Ref Reg = _LoadMemFPR(OpSize::i128Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * i)), OpSize::i8Bit, MemOffsetType::SXTX, 1);
    // Mask off the top bits
    Reg = _VAnd(OpSize::i128Bit, OpSize::i128Bit, Reg, Mask);
    if (ReducedPrecisionMode) {
      // Convert to double precision
      Reg = _F80CVT(OpSize::i64Bit, Reg);
    }
-    _StoreContextIndexed(Reg, Top, StoreSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit), FPRClass);
+    _StoreContextFPRIndexed(Reg, Top, StoreSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit));

    Top = _And(OpSize::i32Bit, Add(OpSize::i32Bit, Top, 1), SevenConst);
  }
@@ -590,19 +588,19 @@ void OpDispatchBuilder::X87FRSTOR(OpcodeArgs) {
  // ST7 broken in to two parts
  // Lower 64bits [63:0]
  // upper 16 bits [79:64]
-  Ref Reg = _LoadMem(FPRClass, OpSize::i64Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * 7)), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
-  Ref RegHigh = _LoadMem(FPRClass, OpSize::i16Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * 7) + 8), OpSize::i8Bit, MEM_OFFSET_SXTX, 1);
+  Ref Reg = _LoadMemFPR(OpSize::i64Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * 7)), OpSize::i8Bit, MemOffsetType::SXTX, 1);
+  Ref RegHigh = _LoadMemFPR(OpSize::i16Bit, Mem, Constant((IR::OpSizeToSize(Size) * 7) + (10 * 7) + 8), OpSize::i8Bit, MemOffsetType::SXTX, 1);
  Reg = _VInsElement(OpSize::i128Bit, OpSize::i16Bit, 4, 0, Reg, RegHigh);
  if (ReducedPrecisionMode) {
    Reg = _F80CVT(OpSize::i64Bit, Reg); // Convert to double precision
  }
-  _StoreContextIndexed(Reg, Top, StoreSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit), FPRClass);
+  _StoreContextFPRIndexed(Reg, Top, StoreSize, MMBaseOffset(), IR::OpSizeToSize(OpSize::i128Bit));
 }

 // Load / Store Control Word
 void OpDispatchBuilder::X87FSTCW(OpcodeArgs) {
-  auto FCW = _LoadContext(OpSize::i16Bit, GPRClass, offsetof(FEXCore::Core::CPUState, FCW));
-  StoreResult(GPRClass, Op, FCW, OpSize::iInvalid);
+  auto FCW = _LoadContextGPR(OpSize::i16Bit, offsetof(FEXCore::Core::CPUState, FCW));
+  StoreResultGPR(Op, FCW);
 }

 void OpDispatchBuilder::X87FLDCW(OpcodeArgs) {
@@ -610,8 +608,8 @@ void OpDispatchBuilder::X87FLDCW(OpcodeArgs) {
  // to switch for now to slow mode whenever these are manually changed.
  // Remove the next line and try DF_04.asm in fast path.
  _StackForceSlow();
-  Ref NewFCW = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  Ref NewFCW = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
 }

 void OpDispatchBuilder::FXCH(OpcodeArgs) {
@@ -647,10 +645,10 @@ void OpDispatchBuilder::FCOMI(OpcodeArgs, IR::OpSize Width, bool Integer, OpDisp
    if (Width == OpSize::i16Bit || Width == OpSize::i32Bit || Width == OpSize::i64Bit) {
      // Memory arg
      if (Integer) {
-        arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+        arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
        b = _F80CVTToInt(arg, Width);
      } else {
-        arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+        arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
        b = _F80CVTTo(arg, Width);
      }
    } else {
@@ -766,7 +764,7 @@ Ref OpDispatchBuilder::ReconstructFSW_Helper(Ref T) {
 void OpDispatchBuilder::X87FNSTSW(OpcodeArgs) {
  Ref TopValue = _SyncStackToSlow();
  Ref StatusWord = ReconstructFSW_Helper(TopValue);
-  StoreResult(GPRClass, Op, StatusWord, OpSize::iInvalid);
+  StoreResultGPR(Op, StatusWord);
 }

 void OpDispatchBuilder::FNCLEX(OpcodeArgs) {
@@ -785,12 +783,12 @@ void OpDispatchBuilder::FNINIT(OpcodeArgs) {

  // Init FCW to 0x037F
  auto NewFCW = Constant(0x037F);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));

  // Set top to zero
  SetX87Top(Zero);
  // Tags all get marked as invalid
-  _StoreContext(OpSize::i8Bit, GPRClass, Zero, offsetof(FEXCore::Core::CPUState, AbridgedFTW));
+  _StoreContextGPR(OpSize::i8Bit, Zero, offsetof(FEXCore::Core::CPUState, AbridgedFTW));

  // Reinits the simulated stack
  _InitStack();
@@ -863,7 +861,7 @@ void OpDispatchBuilder::X87FXAM(OpcodeArgs) {
  auto TopValid = _StackValidTag(0);

  // In the case of top being invalid then C3:C2:C0 is 0b101
-  auto C3 = Select01(OpSize::i32Bit, CondClassType {COND_NEQ}, TopValid, Constant(1));
+  auto C3 = Select01(OpSize::i32Bit, CondClass::NEQ, TopValid, Constant(1));

  auto C2 = TopValid;
  auto C0 = C3; // Mirror C3 until something other than zero is supported
@@ -878,8 +876,8 @@ void OpDispatchBuilder::X87FXTRACT(OpcodeArgs) {
  _PopStackDestroy();
  auto Exp = _F80XTRACT_EXP(Top);
  auto Sig = _F80XTRACT_SIG(Top);
-  _PushStack(Exp, Exp, OpSize::f80Bit, true);
-  _PushStack(Sig, Sig, OpSize::f80Bit, true);
+  _PushStack(Exp, Invalid(), OpSize::iInvalid);
+  _PushStack(Sig, Invalid(), OpSize::iInvalid);
 }

 } // namespace FEXCore::IR
@@ -29,38 +29,37 @@ void OpDispatchBuilder::X87LDENVF64(OpcodeArgs) {
  const auto Size = OpSizeFromSrc(Op);
  Ref Mem = MakeSegmentAddress(Op, Op->Src[0]);

-  auto NewFCW = _LoadMem(GPRClass, OpSize::i16Bit, Mem, OpSize::i16Bit);
+  auto NewFCW = _LoadMemGPR(OpSize::i16Bit, Mem, OpSize::i16Bit);
  // ignore the rounding precision, we're always 64-bit in F64.
  // extract rounding mode
  Ref roundingMode = _Bfe(OpSize::i32Bit, 3, 10, NewFCW);
  _SetRoundingMode(roundingMode, false, roundingMode);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));

-  auto NewFSW = _LoadMem(GPRClass, Size, Mem, Constant(IR::OpSizeToSize(Size)), Size, MEM_OFFSET_SXTX, 1);
+  auto NewFSW = _LoadMemGPR(Size, Mem, Constant(IR::OpSizeToSize(Size)), Size, MemOffsetType::SXTX, 1);
  ReconstructX87StateFromFSW_Helper(NewFSW);

  {
    // FTW
-    SetX87FTW(_LoadMem(GPRClass, Size, Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MEM_OFFSET_SXTX, 1));
+    SetX87FTW(_LoadMemGPR(Size, Mem, Constant(IR::OpSizeToSize(Size) * 2), Size, MemOffsetType::SXTX, 1));
  }
 }

 void OpDispatchBuilder::X87FLDCWF64(OpcodeArgs) {
  _StackForceSlow();

-  Ref NewFCW = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+  Ref NewFCW = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
  // ignore the rounding precision, we're always 64-bit in F64.
  // extract rounding mode
  Ref roundingMode = _Bfe(OpSize::i32Bit, 3, 10, NewFCW);
  _SetRoundingMode(roundingMode, false, roundingMode);
-  _StoreContext(OpSize::i16Bit, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
+  _StoreContextGPR(OpSize::i16Bit, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));
 }

 // F64 ops
 // Float load op with memory operand
 void OpDispatchBuilder::FLDF64(OpcodeArgs, IR::OpSize Width) {
-  const auto ReadWidth = (Width == OpSize::f80Bit) ? OpSize::i128Bit : Width;
-  Ref Data = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], Width, Op->Flags);
+  Ref Data = LoadSourceFPR_WithOpSize(Op, Op->Src[0], Width, Op->Flags);
  // Convert to 64bit float
  Ref ConvertedData = Data;
  if (Width == OpSize::i32Bit) {
@@ -68,39 +67,39 @@ void OpDispatchBuilder::FLDF64(OpcodeArgs, IR::OpSize Width) {
  } else if (Width == OpSize::f80Bit) {
    ConvertedData = _F80CVT(OpSize::i64Bit, Data);
  }
-  _PushStack(ConvertedData, Data, ReadWidth, true);
+  _PushStack(ConvertedData, Data, Width);
 }

 void OpDispatchBuilder::FBLDF64(OpcodeArgs) {
  // Read from memory
-  Ref Data = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], OpSize::f80Bit, Op->Flags);
+  Ref Data = LoadSourceFPR_WithOpSize(Op, Op->Src[0], OpSize::f80Bit, Op->Flags);
  Ref ConvertedData = _F80BCDLoad(Data);
  ConvertedData = _F80CVT(OpSize::i64Bit, ConvertedData);
-  _PushStack(ConvertedData, Data, OpSize::i64Bit, true);
+  _PushStack(ConvertedData, Invalid(), OpSize::iInvalid);
 }

 void OpDispatchBuilder::FBSTPF64(OpcodeArgs) {
  Ref converted = _F80CVTTo(_ReadStackValue(0), OpSize::i64Bit);
  converted = _F80BCDStore(converted);
-  StoreResult_WithOpSize(FPRClass, Op, Op->Dest, converted, OpSize::f80Bit, OpSize::i8Bit);
+  StoreResultFPR_WithOpSize(Op, Op->Dest, converted, OpSize::f80Bit, OpSize::i8Bit);
  _PopStackDestroy();
 }

 void OpDispatchBuilder::FLDF64_Const(OpcodeArgs, uint64_t Num) {
  auto Data = _VCastFromGPR(OpSize::i64Bit, OpSize::i64Bit, Constant(Num));
-  _PushStack(Data, Data, OpSize::i64Bit, true);
+  _PushStack(Data, Data, OpSize::i64Bit);
 }

 void OpDispatchBuilder::FILDF64(OpcodeArgs) {
  const auto ReadWidth = OpSizeFromSrc(Op);

  // Read from memory
-  Ref Data = LoadSource_WithOpSize(GPRClass, Op, Op->Src[0], ReadWidth, Op->Flags);
+  Ref Data = LoadSourceGPR_WithOpSize(Op, Op->Src[0], ReadWidth, Op->Flags);
  if (ReadWidth == OpSize::i16Bit) {
    Data = _Sbfe(OpSize::i64Bit, IR::OpSizeAsBits(ReadWidth), 0, Data);
  }
  auto ConvertedData = _Float_FromGPR_S(OpSize::i64Bit, ReadWidth == OpSize::i32Bit ? OpSize::i32Bit : OpSize::i64Bit, Data);
-  _PushStack(ConvertedData, Data, ReadWidth, false);
+  _PushStack(ConvertedData, Invalid(), OpSize::iInvalid);
 }

 void OpDispatchBuilder::FISTF64(OpcodeArgs, bool Truncate) {
@@ -112,7 +111,7 @@ void OpDispatchBuilder::FISTF64(OpcodeArgs, bool Truncate) {
  } else {
    data = _Float_ToGPR_S(Size == OpSize::i32Bit ? OpSize::i32Bit : OpSize::i64Bit, OpSize::i64Bit, data);
  }
-  StoreResult_WithOpSize(GPRClass, Op, Op->Dest, data, Size, OpSize::i8Bit);
+  StoreResultGPR_WithOpSize(Op, Op->Dest, data, Size, OpSize::i8Bit);

  if ((Op->TableInfo->Flags & X86Tables::InstFlags::FLAGS_POP) != 0) {
    _PopStackDestroy();
@@ -138,16 +137,16 @@ void OpDispatchBuilder::FADDF64(OpcodeArgs, IR::OpSize Width, bool Integer, OpDi
  Ref arg {};

  if (Integer) {
-    arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
    if (Width == OpSize::i16Bit) {
      arg = _Sbfe(OpSize::i64Bit, 16, 0, arg);
    }
    arg = _Float_FromGPR_S(OpSize::i64Bit, Width == OpSize::i64Bit ? OpSize::i64Bit : OpSize::i32Bit, arg);
  } else if (Width == OpSize::i32Bit) {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    arg = _Float_FToF(OpSize::i64Bit, OpSize::i32Bit, arg);
  } else if (Width == OpSize::i64Bit) {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  } else {
    FEX_UNREACHABLE;
  }
@@ -176,16 +175,16 @@ void OpDispatchBuilder::FMULF64(OpcodeArgs, IR::OpSize Width, bool Integer, OpDi
  Ref arg {};

  if (Integer) {
-    arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
    if (Width == OpSize::i16Bit) {
      arg = _Sbfe(OpSize::i64Bit, 16, 0, arg);
    }
    arg = _Float_FromGPR_S(OpSize::i64Bit, Width == OpSize::i64Bit ? OpSize::i64Bit : OpSize::i32Bit, arg);
  } else if (Width == OpSize::i32Bit) {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    arg = _Float_FToF(OpSize::i64Bit, OpSize::i32Bit, arg);
  } else if (Width == OpSize::i64Bit) {
-    arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+    arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
  } else {
    FEX_UNREACHABLE;
  }
@@ -228,16 +227,16 @@ void OpDispatchBuilder::FDIVF64(OpcodeArgs, IR::OpSize Width, bool Integer, bool

  if (Width == OpSize::i16Bit || Width == OpSize::i32Bit || Width == OpSize::i64Bit) {
    if (Integer) {
-      Arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+      Arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
      if (Width == OpSize::i16Bit) {
        Arg = _Sbfe(OpSize::i64Bit, 16, 0, Arg);
      }
      Arg = _Float_FromGPR_S(OpSize::i64Bit, Width == OpSize::i64Bit ? OpSize::i64Bit : OpSize::i32Bit, Arg);
    } else if (Width == OpSize::i32Bit) {
-      Arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      Arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
      Arg = _Float_FToF(OpSize::i64Bit, OpSize::i32Bit, Arg);
    } else if (Width == OpSize::i64Bit) {
-      Arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      Arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    }
  } else {
    FEX_UNREACHABLE;
@@ -285,16 +284,16 @@ void OpDispatchBuilder::FSUBF64(OpcodeArgs, IR::OpSize Width, bool Integer, bool

  if (Width == OpSize::i16Bit || Width == OpSize::i32Bit || Width == OpSize::i64Bit) {
    if (Integer) {
-      arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+      arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
      if (Width == OpSize::i16Bit) {
        arg = _Sbfe(OpSize::i64Bit, 16, 0, arg);
      }
      arg = _Float_FromGPR_S(OpSize::i64Bit, Width == OpSize::i64Bit ? OpSize::i64Bit : OpSize::i32Bit, arg);
    } else if (Width == OpSize::i32Bit) {
-      arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
      arg = _Float_FToF(OpSize::i64Bit, OpSize::i32Bit, arg);
    } else if (Width == OpSize::i64Bit) {
-      arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    }
  } else {
    FEX_UNREACHABLE;
@@ -332,16 +331,16 @@ void OpDispatchBuilder::FCOMIF64(OpcodeArgs, IR::OpSize Width, bool Integer, OpD
  } else if (Width == OpSize::i16Bit || Width == OpSize::i32Bit || Width == OpSize::i64Bit) {
    // Memory arg
    if (Integer) {
-      arg = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
+      arg = LoadSourceGPR(Op, Op->Src[0], Op->Flags);
      if (Width == OpSize::i16Bit) {
        arg = _Sbfe(OpSize::i64Bit, 16, 0, arg);
      }
      b = _Float_FromGPR_S(OpSize::i64Bit, Width == OpSize::i64Bit ? OpSize::i64Bit : OpSize::i32Bit, arg);
    } else if (Width == OpSize::i32Bit) {
-      arg = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      arg = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
      b = _Float_FToF(OpSize::i64Bit, OpSize::i32Bit, arg);
    } else if (Width == OpSize::i64Bit) {
-      b = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
+      b = LoadSourceFPR(Op, Op->Src[0], Op->Flags);
    }
  } else {
    FEX_UNREACHABLE;
@@ -393,11 +392,11 @@ void OpDispatchBuilder::X87FXTRACTF64(OpcodeArgs) {
  SaveNZCV();
  _TestNZ(OpSize::i64Bit, Gpr, Constant(0x7fff'ffff'ffff'ffffUL));

-  Ref Sig = _NZCVSelectV(OpSize::i64Bit, {COND_EQ}, SigZV, SigNZV);
-  Ref Exp = _NZCVSelectV(OpSize::i64Bit, {COND_EQ}, ExpZV, ExpNZV);
+  Ref Sig = _NZCVSelectV(OpSize::i64Bit, CondClass::EQ, SigZV, SigNZV);
+  Ref Exp = _NZCVSelectV(OpSize::i64Bit, CondClass::EQ, ExpZV, ExpNZV);

  _PopStackDestroy();
-  _PushStack(Exp, Exp, OpSize::i64Bit, true);
-  _PushStack(Sig, Sig, OpSize::i64Bit, true);
+  _PushStack(Exp, Invalid(), OpSize::iInvalid);
+  _PushStack(Sig, Invalid(), OpSize::iInvalid);
 }
 } // namespace FEXCore::IR
@@ -1,87 +0,0 @@
-// SPDX-License-Identifier: MIT
-/*
-$info$
-tags: glue|x86-guest-code
-desc: Guest-side assembly helpers used by the backends
-$end_info$
-*/
-
-#include "Interface/Core/X86HelperGen.h"
-#include "FEXCore/Utils/AllocatorHooks.h"
-
-#include <FEXCore/Config/Config.h>
-#include <FEXCore/Utils/Allocator.h>
-#include <FEXHeaderUtils/Syscalls.h>
-
-#include <cstdint>
-#include <cstring>
-
-namespace FEXCore {
-constexpr size_t CODE_SIZE = 0x1000;
-
-X86GeneratedCode::X86GeneratedCode() {
-#ifdef _WIN32
-  // No need to allocate anything in this config.
-#else
-
-  // Allocate a page for our emulated guest
-  CodePtr = AllocateGuestCodeSpace(CODE_SIZE);
-
-  constexpr std::array<uint8_t, 2> SignalReturnCode = {
-    0x0F, 0x37, // CALLBACKRET FEX Instruction
-  };
-
-  CallbackReturn = reinterpret_cast<uint64_t>(CodePtr);
-
-  memcpy(reinterpret_cast<void*>(CallbackReturn), SignalReturnCode.data(), SignalReturnCode.size());
-
-  mprotect(CodePtr, CODE_SIZE, PROT_READ);
-#endif
-}
-
-X86GeneratedCode::~X86GeneratedCode() {
-#ifndef _WIN32
-  FEXCore::Allocator::VirtualFree(CodePtr, CODE_SIZE);
-#endif
-}
-
-void* X86GeneratedCode::AllocateGuestCodeSpace(size_t Size) {
-#ifndef _WIN32
-  FEX_CONFIG_OPT(Is64BitMode, IS64BIT_MODE);
-
-  if (Is64BitMode()) {
-    // 64bit mode can have its sigret handler anywhere
-    return FEXCore::Allocator::VirtualAlloc(Size);
-  }
-
-  // First 64bit page
-  constexpr uintptr_t LOCATION_MAX = 0x1'0000'0000;
-
-  // 32bit mode
-  // We need to have the sigret handler in the lower 32bits of memory space
-  // Scan top down and try to allocate a location
-  for (size_t Location = 0xFFFF'E000; Location != 0x0; Location -= 0x1000) {
-    void* Ptr = ::mmap(reinterpret_cast<void*>(Location), Size, PROT_READ | PROT_WRITE, MAP_FIXED_NOREPLACE | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
-
-    if (Ptr != MAP_FAILED && reinterpret_cast<uintptr_t>(Ptr) >= LOCATION_MAX) {
-      // Failed to map in the lower 32bits
-      // Try again
-      // Can happen in the case that host kernel ignores MAP_FIXED_NOREPLACE
-      ::munmap(Ptr, Size);
-      continue;
-    }
-
-    if (Ptr != MAP_FAILED) {
-      return Ptr;
-    }
-  }
-
-  // Can't do anything about this
-  // Here's hoping the application doesn't use signals
-  return MAP_FAILED;
-#else
-  return nullptr;
-#endif
-}
-
-} // namespace FEXCore