Introduce BOX64_NOVULKANOVERLAY to disabled x64 Vulkan Overlay

cpuid: Enable Invariant TSC bit for Intel CPU type (#3579 )
* cpuid: Enable Invariant TSC bit for Intel CPU type The Invariant TSC bit (CPUID leaf 0x80000007, EDX bit 8) was already enabled for AMD CPU type but was commented out for Intel CPU type. ARM64 provides an invariant system counter via CNTPCT_EL0, so advertising Invariant TSC capability is correct on ARM64 hardware regardless of the emulated CPU vendor. Some applications (notably World of Warcraft and VanillaFixes) check this bit and refuse to run or crash when it is not set. Fixes #2485 * Gate Invariant TSC behind BOX64_RDTSC_INV env var Software often reports the timer clock as the CPU clock, which is incorrect under Box64. Make the Invariant TSC bit in CPUID leaf 0x80000007 opt-in for the Intel CPU type path via BOX64_RDTSC_INV=1 (disabled by default). The AMD CPU type path already reports this bit unconditionally. * cpuid: fix BOX64_RDTSC_INV availability in WowBox64.dll and sanitize comments - Set wine flag to 1 so BOX64_RDTSC_INV is parseable and initialised correctly when compiled as WowBox64.dll (_WIN32 context) - Remove verbose comment annotation; env var name is evident from code Co-Authored-By: Claude Sonnet 4.6 <[email protected]> --------- Co-authored-by: Joe <[email protected]> Co-authored-by: Claude Sonnet 4.6 <[email protected]>
2026-11-12 18:57:06 -10:00 · 2026-03-03 22:08:01 +01:00 · 2026-03-03 20:23:00 +01:00 · 2026-03-03 17:51:22 +01:00 · 2026-03-03 14:04:47 +01:00 · 2026-03-03 14:02:11 +01:00
762 changed files with 102823 additions and 34067 deletions
@@ -1,6 +1,9 @@
 ### Auto build Box64 and release its binary with Github Action
 name: Build and Release Box64

+permissions:
+  actions: 'write'
+
 on:
  workflow_dispatch:
  release:
@@ -22,14 +25,18 @@ on:
      - "CMakeLists.txt"
      - "**/*.yml"

+defaults:
+  run:
+    # needed on container images which could default to sh
+    shell: bash
+
 jobs:
  build:
    strategy:
      fail-fast: false
      matrix:
-        platform: [X64, RISCV, RK3588, ARM64, ANDROID, TERMUX, LARCH64, ANDROID_GLIBC, WOW64]
+        platform: [X64, RISCV, RK3588, ARM64, ANDROID, TERMUX, LARCH64, PPC64LE, ANDROID_GLIBC, WOW64, ARM64-GCC-8]
        type: [Release, Trace, StaticBuild, Box32]
-        os: [ubuntu-latest, ubuntu-22.04-arm]
        exclude:
          - platform: ANDROID
            type: StaticBuild
@@ -41,41 +48,151 @@ jobs:
            type: Box32
          - platform: TERMUX
            type: Box32
-          - platform: X64
-            os: ubuntu-22.04-arm
-          - platform: RISCV
-            os: ubuntu-22.04-arm
-          - platform: RK3588
-            os: ubuntu-latest
-          - platform: ARM64
-            os: ubuntu-latest
-          - platform: ANDROID
-            os: ubuntu-22.04-arm
-          - platform: TERMUX
-            os: ubuntu-22.04-arm
-          - platform: LARCH64
-            os: ubuntu-22.04-arm
-          - platform: ANDROID_GLIBC
-            os: ubuntu-latest
-          - platform: WOW64
-            os: ubuntu-latest
          - platform: WOW64
            type: Trace
          - platform: WOW64
            type: StaticBuild
          - platform: WOW64
            type: Box32
+          - platform: ARM64-GCC-8
+            type: Trace
+          - platform: ARM64-GCC-8
+            type: StaticBuild
+          - platform: ANDROID
+            type: Trace
+          - platform: TERMUX
+            type: Trace
+          - platform: LARCH64
+            type: Trace
+          - platform: PPC64LE
+            type: Trace
+        include:
+          - platform: X64
+            os: ubuntu-latest
+
+          - platform: RISCV
+            os: ubuntu-latest
+
+          - platform: RK3588
+            os: ubuntu-22.04-arm
+
+          - platform: ARM64
+            os: ubuntu-22.04-arm
+
+          - platform: ANDROID
+            os: ubuntu-latest
+
+          - platform: TERMUX
+            os: ubuntu-latest
+
+          - platform: LARCH64
+            os: ubuntu-latest
+
+          - platform: PPC64LE
+            os: ubuntu-latest
+
+          - platform: ANDROID_GLIBC
+            os: ubuntu-22.04-arm
+
+          - platform: WOW64
+            os: ubuntu-22.04-arm
+
+          - platform: ARM64-GCC-8
+            os: ubuntu-22.04-arm
+            image: arm64v8/ubuntu:18.04

    runs-on: ${{ matrix.os }}
+    container:
+      image: ${{ matrix.image || '' }}
    steps:
      - name: "Checkout Box64 Repository"
+        if: ${{ matrix.platform != 'ARM64-GCC-8' }}
        uses: actions/checkout@v4

-      - name: "Environment preparation"
+      - name: "Checkout Box64 Repository (No NodeJS)"
+        if: ${{ matrix.platform == 'ARM64-GCC-8' }}
+        uses: taiki-e/checkout-action@v1
+
+      - name: Get all the unique changed directory names
+        id: changed-files-dir-names
+        if: ${{ matrix.platform != 'ARM64-GCC-8' }}
+        uses: tj-actions/changed-files@ed68ef82c095e0d48ec87eccea555d944a631a4c # v46
+        with:
+          dir_names: "true"
+
+      - name: Early exit other jobs if this is a ARM64-only change
+        id: early-exit-arm64
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        if: steps.changed-files-dir-names.outputs.all_modified_files == 'src/dynarec/arm64'
+        continue-on-error: true
        run: |
-          sudo apt-get update
-          if [[ ${{ matrix.platform }} != 'X64' && ${{ matrix.platform }} != 'RISCV' && ${{ matrix.platform }} != 'LARCH64' ]]; then
-            sudo apt-get -y install git cmake make python3 patchelf
+          if [[ ${{ matrix.platform }} != 'RISCV' && ${{ matrix.platform }} != 'LARCH64' && ${{ matrix.platform }} != 'PPC64LE' && ${{ matrix.platform }} != 'X64' ]]; then
+            exit 1
+          fi
+
+      - name: Early exit other jobs if this is a LA64-only change
+        id: early-exit-la64
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        if: steps.changed-files-dir-names.outputs.all_modified_files == 'src/dynarec/la64'
+        continue-on-error: true
+        run: |
+          if [[ ${{ matrix.platform }} == 'LARCH64' ]]; then
+            exit 1
+          fi
+
+      - name: Early exit other jobs if this is a RV64-only change
+        id: early-exit-rv64
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        if: steps.changed-files-dir-names.outputs.all_modified_files == 'src/dynarec/rv64'
+        continue-on-error: true
+        run: |
+          if [[ ${{ matrix.platform }} == 'RISCV' ]]; then
+            exit 1
+          fi
+
+      - name: Early exit other jobs if this is a PPC64LE-only change
+        id: early-exit-ppc64le
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        if: steps.changed-files-dir-names.outputs.all_modified_files == 'src/dynarec/ppc64le'
+        continue-on-error: true
+        run: |
+          if [[ ${{ matrix.platform }} == 'PPC64LE' ]]; then
+            exit 1
+          fi
+
+      - name: Do not early exit otherwise
+        id: early-exit-never
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          DIRS: ${{ steps.changed-files-dir-names.outputs.all_modified_files }}
+        if: env.DIRS != 'src/dynarec/arm64' && env.DIRS != 'src/dynarec/la64' && env.DIRS != 'src/dynarec/rv64' && env.DIRS != 'src/dynarec/ppc64le'
+        continue-on-error: true
+        run: exit 1
+
+      - name: Merge early exit conditions
+        id: early-exit
+        if: steps.early-exit-arm64.outcome == 'failure' || steps.early-exit-la64.outcome == 'failure' || steps.early-exit-rv64.outcome == 'failure' || steps.early-exit-ppc64le.outcome == 'failure' || steps.early-exit-never.outcome == 'failure'
+        continue-on-error: true
+        run: exit 1
+
+      # Use `steps.early-exit.outcome == 'failure'` to check whether the current step should be processed.
+
+      - name: "Environment preparation"
+        if: steps.early-exit.outcome == 'failure'
+        run: |
+          if [[ ${{ matrix.platform }} == 'ARM64-GCC-8' ]]; then
+            apt-get update
+            apt install -y sudo
+          else
+            sudo apt-get update
+            zydis_package=libzydis-dev
+          fi
+          if [[ ${{ matrix.platform }} != 'X64' && ${{ matrix.platform }} != 'RISCV' && ${{ matrix.platform }} != 'LARCH64' && ${{ matrix.platform }} != 'PPC64LE' ]]; then
+            sudo apt-get -y install git cmake make python3 patchelf $zydis_package
            if [[ ${{ matrix.platform }} == 'ANDROID' || ${{ matrix.platform }} == 'TERMUX' ]]; then
              sudo apt-get -y install p7zip
              wget -q https://dl.google.com/android/repository/android-ndk-r26b-linux.zip
@@ -98,14 +215,26 @@ jobs:
              wget -q https://github.com/mstorsjo/llvm-mingw/releases/download/${LLVM_MINGW_VERSION}/llvm-mingw-${LLVM_MINGW_VERSION}-ucrt-ubuntu-22.04-aarch64.tar.xz
              tar -xf llvm-mingw-${LLVM_MINGW_VERSION}-ucrt-ubuntu-22.04-aarch64.tar.xz
              echo "MINGW_COMPILER_PATH=$PWD/llvm-mingw-${LLVM_MINGW_VERSION}-ucrt-ubuntu-22.04-aarch64/bin" >> $GITHUB_ENV
+            elif [[ ${{ matrix.platform }} == 'ARM64-GCC-8' ]]; then
+              # some of these dependencies may be unnecessary to explicitly install depending on the image
+              sudo apt-get -y install git software-properties-common lsb-release \
+              wget curl build-essential autoconf automake \
+              pkg-config ca-certificates \
+              python3 make gettext zstd \
+              gcc-8 g++-8
+              sudo add-apt-repository ppa:theofficialgman/cmake-bionic -y
+              sudo apt update
+              sudo apt install -y cmake
+              echo "BOX64_PLATFORM_MARCRO=-DARM64=1" >> $GITHUB_ENV
+              echo "BOX64_COMPILER=aarch64-linux-gnu-gcc-8" >> $GITHUB_ENV
            else
              sudo apt-get -y install git gcc-aarch64-linux-gnu
              echo "BOX64_PLATFORM_MARCRO=-D${{ matrix.platform }}=1" >> $GITHUB_ENV
              echo "BOX64_COMPILER=aarch64-linux-gnu-gcc" >> $GITHUB_ENV
            fi
            if [[ ${{ matrix.platform }} == 'TERMUX' ]]; then
-              wget https://packages.termux.dev/apt/termux-main/pool/main/liba/libandroid-sysv-semaphore/libandroid-sysv-semaphore_0.1_aarch64.deb
-              7z x libandroid-sysv-semaphore_0.1_aarch64.deb
+              wget https://packages.termux.dev/apt/termux-main/pool/main/liba/libandroid-sysv-semaphore/libandroid-sysv-semaphore_0.1-1_aarch64.deb
+              7z x libandroid-sysv-semaphore_0.1-1_aarch64.deb
              tar -xf data.tar.xz
              sudo cp -rf data /
              sudo chmod 755 -R /data
@@ -120,14 +249,14 @@ jobs:
            if [[ ${{ matrix.platform }} == 'X64' ]]; then
              echo "BOX64_PLATFORM_MARCRO=-DLD80BITS=1 -DNOALIGN=1" >> $GITHUB_ENV
              echo "BOX64_COMPILER=gcc" >> $GITHUB_ENV
-              sudo apt-get -y install git cmake make python3
+              sudo apt-get -y install git cmake make python3 libzydis-dev
            elif [[ ${{ matrix.platform }} == 'RISCV' ]]; then
              echo BOX64_PLATFORM_MARCRO="-DRV64=ON" >> $GITHUB_ENV
-              echo "BOX64_COMPILER=riscv64-linux-gnu-gcc" >> $GITHUB_ENV
-              sudo apt-get -y install git gcc-riscv64-linux-gnu cmake make python3 ninja-build libglib2.0-dev
+              echo "BOX64_COMPILER=riscv64-linux-gnu-gcc-14" >> $GITHUB_ENV
+              sudo apt-get -y install git gcc-14-riscv64-linux-gnu cmake make python3 ninja-build libglib2.0-dev libzydis-dev
            elif [[ ${{ matrix.platform }} == 'LARCH64' ]]; then
              sudo mkdir /usr/local/larch
-              wget -O- -q https://github.com/loongson/build-tools/releases/download/2023.08.08/CLFS-loongarch64-8.1-x86_64-cross-tools-gcc-glibc.tar.xz | sudo tar -C /usr/local/larch --strip-components=1 --xz -xf -
+              wget -O- -q https://github.com/loongson/build-tools/releases/download/2025.02.21/x86_64-cross-tools-loongarch64-binutils_2.44-gcc_14.2.0-glibc_2.41.tar.xz | sudo tar -C /usr/local/larch --strip-components=1 --xz -xf -
              sudo ln -sf /usr/local/larch/target /usr/loongarch64-linux-gnu
              sudo cp -r /usr/local/larch/loongarch64-unknown-linux-gnu/lib/* /usr/loongarch64-linux-gnu/lib64/
              for i in objdump objcopy strip; do
@@ -139,11 +268,15 @@ jobs:

              echo BOX64_PLATFORM_MARCRO="-DLARCH64=ON" >> $GITHUB_ENV
              echo "BOX64_COMPILER=loongarch64-linux-gnu-gcc" >> $GITHUB_ENV
-              sudo apt-get -y install git cmake make python3
+              sudo apt-get -y install git cmake make python3 libzydis-dev
+            elif [[ ${{ matrix.platform }} == 'PPC64LE' ]]; then
+              echo BOX64_PLATFORM_MARCRO="-DPPC64LE=ON" >> $GITHUB_ENV
+              echo "BOX64_COMPILER=powerpc64le-linux-gnu-gcc" >> $GITHUB_ENV
+              sudo apt-get -y install git gcc-powerpc64le-linux-gnu cmake make python3 libzydis-dev
            else
              echo BOX64_PLATFORM_MARCRO="-DARM_DYNAREC=ON" >> $GITHUB_ENV
              echo "BOX64_COMPILER=aarch64-linux-gnu-gcc" >> $GITHUB_ENV
-              sudo apt-get -y install git gcc-aarch64-linux-gnu cmake make python3
+              sudo apt-get -y install git gcc-aarch64-linux-gnu cmake make python3 libzydis-dev
            fi
          fi
          if [[ ${{ matrix.type }} == 'Release' ]]; then
@@ -167,15 +300,20 @@ jobs:
            echo BOX64_STATICBUILD=0 >> $GITHUB_ENV
            echo BOX64_BOX32=0 >> $GITHUB_ENV
          fi
+          if [[ ${{ matrix.os }} == 'ubuntu-22.04-arm' ]]; then
+            echo "BOX64_ZYDIS3=1" >> $GITHUB_ENV
+          else
+            echo "BOX64_ZYDIS3=0" >> $GITHUB_ENV
+          fi

      - name: "Get XuanTie QEMU Cache Key"
-        if: matrix.platform == 'RISCV'
+        if: matrix.platform == 'RISCV' && steps.early-exit.outcome == 'failure'
        id: get-xuantie-qemu-cache-key
        run: |
          echo "key=f2dfdd13014d51f957c7172acc2e791cb42dc400" >> $GITHUB_OUTPUT

      - name: "Cache XuanTie QEMU"
-        if: matrix.platform == 'RISCV'
+        if: matrix.platform == 'RISCV' && steps.early-exit.outcome == 'failure'
        id: cache-xuantie-qemu
        uses: actions/cache@v3
        with:
@@ -183,7 +321,7 @@ jobs:
          key: ${{ runner.os }}-${{ steps.get-xuantie-qemu-cache-key.outputs.key }}-xuantie-qemu

      - name: "Checkout XuanTie QEMU"
-        if: matrix.platform == 'RISCV' && steps.cache-xuantie-qemu.outputs.cache-hit != 'true'
+        if: matrix.platform == 'RISCV' && steps.cache-xuantie-qemu.outputs.cache-hit != 'true' && steps.early-exit.outcome == 'failure'
        uses: actions/checkout@v3
        with:
          repository: revyos/qemu
@@ -191,7 +329,7 @@ jobs:
          ref: f2dfdd13014d51f957c7172acc2e791cb42dc400

      - name: "Build XuanTie QEMU for XTheadVector"
-        if: matrix.platform == 'RISCV' && steps.cache-xuantie-qemu.outputs.cache-hit != 'true'
+        if: matrix.platform == 'RISCV' && steps.cache-xuantie-qemu.outputs.cache-hit != 'true' && steps.early-exit.outcome == 'failure'
        run: |
          cd xuantie_qemu
          ./configure --prefix=$GITHUB_WORKSPACE/xuantie_qemu_install --target-list=riscv64-linux-user --disable-system
@@ -199,6 +337,7 @@ jobs:
          make install

      - name: "Display Build info"
+        if: steps.early-exit.outcome == 'failure'
        run: |
          echo "CMake Platform Macro: ${{ env.BOX64_PLATFORM_MARCRO }}"
          echo "CMake C Compiler: ${{ env.BOX64_COMPILER }}"
@@ -208,6 +347,7 @@ jobs:
          echo "Box32 Enabled: ${{ env.BOX64_BOX32 }}"

      - name: "Build Box64"
+        if: steps.early-exit.outcome == 'failure'
        run: |
          export PATH=$PATH:${{ env.MINGW_COMPILER_PATH }}
          mkdir build
@@ -217,6 +357,7 @@ jobs:
                   ${{ env.BOX64_PLATFORM_MARCRO }}\
                   -DCMAKE_BUILD_TYPE=${{ env.BOX64_BUILD_TYPE }}\
                   -DHAVE_TRACE=${{ env.BOX64_HAVE_TRACE }}\
+                   -DZYDIS3=${{ env.BOX64_ZYDIS3 }}\
                   -DSTATICBUILD=${{ env.BOX64_STATICBUILD }}\
                   -DBOX32=${{ env.BOX64_BOX32 }}\
                   -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON\
@@ -224,19 +365,23 @@ jobs:
          make -j$(nproc) VERBOSE=1

      - name: "Test Box64"
+        if: steps.early-exit.outcome == 'failure'
        run: |
          if [[ ${{ matrix.platform }} != 'X64' ]]; then
-            mkdir qemu9
-            wget -O- -q https://archive.archlinux.org/packages/q/qemu-user-static/qemu-user-static-9.0.2-1-x86_64.pkg.tar.zst | tar -I zstd -C qemu9 -xf -
-            sudo cp qemu9/usr/bin/* /usr/bin/
+            mkdir qemu10
+            wget -O- -q https://archive.archlinux.org/packages/q/qemu-user-static/qemu-user-static-10.0.0-7-x86_64.pkg.tar.zst  | tar -I zstd -C qemu10 -xf -
+            sudo cp qemu10/usr/bin/* /usr/bin/
          fi

          cd build
-          export CTEST_OPTIONS="-j$(nproc) --timeout 120 --output-on-failure --repeat until-pass:20"
+          export CTEST_OPTIONS="-j$(nproc) --timeout 300 --output-on-failure --repeat until-pass:20"
          if [[ ${{ matrix.platform }} == 'RISCV' ]]; then
            export INTERPRETER=qemu-riscv64-static
            export QEMU_LD_PREFIX=/usr/riscv64-linux-gnu/

+            if [[ ${{ env.BOX64_BOX32 }} == '1' ]]; then
+              export BOX32_PERSONA32BITS=1 # hack to disable personality setting
+            fi
            ctest $CTEST_OPTIONS
            QEMU_CPU=rv64,v=false BOX64_DYNAREC=0 ctest $CTEST_OPTIONS
            QEMU_CPU=rv64,v=false,zba=true,zbb=true,zbc=true,zbs=true ctest $CTEST_OPTIONS
@@ -257,10 +402,23 @@ jobs:
            export INTERPRETER=qemu-loongarch64-static
            export QEMU_LD_PREFIX=/usr/loongarch64-linux-gnu/

-            BOX64_DYNAREC_LA64NOEXT=1 ctest $CTEST_OPTIONS -E nocosim
+            if [[ ${{ env.BOX64_BOX32 }} == '1' ]]; then
+              export BOX32_PERSONA32BITS=1 # hack to disable personality setting
+            fi
+            ctest $CTEST_OPTIONS -E nocosim
+            BOX64_DYNAREC_NOHOSTEXT=1 ctest $CTEST_OPTIONS -E nocosim
            BOX64_DYNAREC=0 ctest $CTEST_OPTIONS -E nocosim

-            BOX64_DYNAREC_TEST=1 BOX64_DYNAREC_LA64NOEXT=1 ctest $CTEST_OPTIONS -E nocosim
+            BOX64_DYNAREC_TEST=1 ctest $CTEST_OPTIONS -E nocosim
+            BOX64_DYNAREC_TEST=1 BOX64_DYNAREC_NOHOSTEXT=1 ctest $CTEST_OPTIONS -E nocosim
+          elif [[ ${{ matrix.platform }} == 'PPC64LE' ]]; then
+            export INTERPRETER=qemu-ppc64le-static
+            export QEMU_LD_PREFIX=/usr/powerpc64le-linux-gnu/
+
+            ctest $CTEST_OPTIONS -E nocosim
+            BOX64_DYNAREC=0 ctest $CTEST_OPTIONS -E nocosim
+
+            BOX64_DYNAREC_TEST=1 ctest $CTEST_OPTIONS -E nocosim
          elif [[ ${{ matrix.platform }} == 'ANDROID' ]]; then
            export INTERPRETER=qemu-aarch64-static
            export QEMU_LD_PREFIX=/system/lib64
@@ -282,13 +440,15 @@ jobs:

      - name: "Get short Git commit"
        id: git-info
+        if: steps.early-exit.outcome == 'failure'
        run: echo "SHORT_COMMIT=$(git rev-parse --short HEAD)" >> $GITHUB_ENV

      - name: "Get Box64 Version"
+        if: steps.early-exit.outcome == 'failure'
        run: echo "BOX64_VERSION=$(cat src/box64version.h | grep BOX64_MAJOR | cut -d " " -f 3).$(cat src/box64version.h | grep BOX64_MINOR | cut -d " " -f 3).$(cat src/box64version.h | grep BOX64_REVISION | cut -d " " -f 3)" >> $GITHUB_ENV

      - name: "Packaging WCP file for Winlator"
-        if: matrix.platform == 'ANDROID_GLIBC' && matrix.type != 'StaticBuild'
+        if: matrix.platform == 'ANDROID_GLIBC' && matrix.type != 'StaticBuild' && steps.early-exit.outcome == 'failure'
        run: |
          cd build
          cat <<EOF > profile.json
@@ -309,7 +469,7 @@ jobs:
          tar --zstd -cf box64-latest.wcp box64 profile.json

      - name: "Package Rat File for MiceWine"
-        if: matrix.platform == 'ANDROID' && matrix.type != 'StaticBuild'
+        if: matrix.platform == 'ANDROID' && matrix.type != 'StaticBuild' && steps.early-exit.outcome == 'failure'
        run: |
          cd build

@@ -326,22 +486,22 @@ jobs:
          7z -tzip -mx=5 a box64-latest.rat files pkg-header

      - name: "Upload WCP file"
-        if: matrix.platform == 'ANDROID_GLIBC' && matrix.type != 'StaticBuild'
+        if: matrix.platform == 'ANDROID_GLIBC' && matrix.type == 'Release' && steps.early-exit.outcome == 'failure'
        uses: actions/upload-artifact@v4
        with:
          name: box64-latest-${{ matrix.type }}-wcp
          path: build/box64-latest.wcp

      - name: "Upload Rat File"
-        if: matrix.platform == 'ANDROID' && matrix.type != 'StaticBuild'
+        if: matrix.platform == 'ANDROID' && matrix.type == 'Release' && steps.early-exit.outcome == 'failure'
        uses: actions/upload-artifact@v4
        with:
          name: box64-MiceWine-${{ matrix.type }}
          path: build/box64-latest.rat

      - name: "Upload WowBox64 file"
-        if: matrix.platform == 'WOW64' && matrix.type == 'Release'
+        if: matrix.platform == 'WOW64' && matrix.type == 'Release' && steps.early-exit.outcome == 'failure'
        uses: actions/upload-artifact@v4
        with:
          name: wowbox64-${{ matrix.type }}
-          path: build/wowbox64-prefix/src/wowbox64-build/wowbox64.dll
+          path: build/wowbox64-prefix/src/wowbox64-build/wowbox64.dll
@@ -0,0 +1,15 @@
+name: 'Close stale issues and PRs'
+on:
+  schedule:
+    - cron: '* * * * 0'
+
+jobs:
+  stale:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/stale@v10
+        with:
+          stale-issue-message: 'This issue is stale because it has been open 1 year with no activity. Remove "Stale" label or comment or this will be closed in 30 days.'
+          days-before-stale: 365
+          days-before-close: 30
+          exempt-issue-labels: 'long-term'
@@ -68,6 +68,8 @@ backup/
 /wrapperhelper/sanleak
 /wrapperhelper/sanundefined
 /wrapperhelper/src/machine.gen
+/wrapperhelper/src/machine_x86.gen
+/wrapperhelper/src/machine_x86_64.gen
 /wrapperhelper/*.h
 !/wrapperhelper/example-libc.h

@@ -1,24 +0,0 @@
-language: c
-arch:
-  - amd64
-  - arm64-graviton2
-os: linux
-dist: focal
-
-compiler:
-    - gcc
-
-addons:
-  apt:
-    packages:
-      - python3
-
-      #Build steps
-before_script:
-    - mkdir build
-    - cd build
-    - if [ "$TRAVIS_CPU_ARCH" = "arm64" ]; then cmake .. -DRK3399=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo ; else cmake .. -DLD80BITS=1 -DNOALIGN=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo ; fi
-
-script:
-    - make
-    - ctest --output-on-failure
@@ -15,11 +15,13 @@ option(ODROIDN2 "Set to ON if targeting an Odroid-N2 device" ${ODROIDN2})
 option(TEGRAX1 "Set to ON if targeting an Tegra X1 based device" ${TEGRAX1})
 option(TEGRA_T194 "Set to ON if targeting an Tegra Xavier based device" ${TEGRA_T194})
 option(TEGRA_T234 "Set to ON if targeting an Tegra Orin based device" ${TEGRA_T234})
+option(NVIDIA_GB10 "Set to ON if targeting a DGX Spark(GB10) based device" ${NVIDIA_GB10})
 option(PHYTIUM "Set to ON if targeting an Phytium (D2000 or FT2000/4) based device" ${PHYTIUM})
-option(SD845 "Set to ON if targeting a Snapragon 845 based device" ${SD845})
-option(SD865 "Set to ON if targeting a Snapragon 865 based device" ${SD865})
-option(SD888 "Set to ON if targeting a Snapragon 888 based device" ${SD888})
-option(SD8G2 "Set to ON if targeting a Snapragon 8 Gen 2 based device" ${SD8G2})
+option(SD845 "Set to ON if targeting a Snapdragon 845 based device" ${SD845})
+option(SD865 "Set to ON if targeting a Snapdragon 865 based device" ${SD865})
+option(SD888 "Set to ON if targeting a Snapdragon 888 based device" ${SD888})
+option(SD8G2 "Set to ON if targeting a Snapdragon 8 Gen 2 based device" ${SD8G2})
+option(SD8EG5 "Set to ON if targeting a Snapdragon 8 Elite Gen 5 (SM8850) based device" ${SD8EG5})
 option(SDORYON1 "Set to ON if targeting a Snapdragon Oryon 1 (X1E80100/X1E78100) based device" ${SDORYON1})
 option(ADLINK "Set to ON if targeting an ADLink AmpereAltra based device" ${ADLINK})
 option(M1 "Set to ON if targeting a AppleM1 running on Asahi computer" ${M1})
@@ -32,11 +34,13 @@ option(ANDROID "Set to ON if targeting an Android device" ${ANDROID})
 option(TERMUX "Set to ON if targeting an Android device with Termux" ${TERMUX})
 option(WINLATOR_GLIBC "Set to ON if targeting an Android device with Winlator Glibc" ${WINLATOR_GLIBC})
 option(USE_CCACHE "Set to ON to use ccache if present in the system" ${USE_CCACHE})
-option(HAVE_TRACE "Set to ON to have Trace ability (needs ZydisInfo library)" ${HAVE_TRACE})
+option(HAVE_TRACE "Set to ON to have Trace ability (needs libzydis-dev)" ${HAVE_TRACE})
+option(ZYDIS3 "Set to ON if you have trace and an old libzydis v3.x" ${ZYDIS3})
 option(SAVE_MEM "Set to ON to build dynarec with some slower memory saving optimizations" ${SAVE_MEM})
 option(NOLOADADDR "Set to ON to avoid fixing the load address of Box64" OFF)
 option(NOGIT "Set to ON if not building from a git clone repo (like when building from a zip download from github)" ${NOGIT})
 option(BAD_SIGNAL "Set to ON to activate the workaround for incoherent si_info on SIGSEGV" ${BAD_SIGNAL})
+option(BAD_PKILL "Set to ON to activate the workaround for incoherent pthread_kill" ${BAD_PKILL})
 option(SW64 "Set ON if targeting an SW64 based device" ${SW64})
 option(CI "Set to ON if running in CI" ${CI})
 option(WITH_MOLD "Set to ON to use with mold" ${WITH_MOLD})
@@ -50,12 +54,23 @@ if(TERMUX)
    set(TERMUX_PATH "/data/data/com.termux/files")
    set(ANDROID ON CACHE BOOL "")
 endif()
+
+# Hack: limit the optimization level to -O1 for some platforms
+if(TERMUX OR ANDROID)
+    set(CMAKE_C_FLAGS_RELEASE "-O1")
+    set(CMAKE_CXX_FLAGS_RELEASE "-O1")
+    set(CMAKE_C_FLAGS_RELWITHDEBINFO "-O1 -g")
+    set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "-O1 -g")
+endif()
+
 if(LARCH64)
    set(LD80BITS OFF CACHE BOOL "")
    set(NOALIGN OFF CACHE BOOL "")
    set(ARM_DYNAREC OFF CACHE BOOL "")
    set(RV64_DYNAREC OFF CACHE BOOL "")
    set(LARCH64_DYNAREC ON CACHE BOOL "")
+    set(PPC64LE_DYNAREC OFF CACHE BOOL "")
+    set(BAD_PKILL ON CACHE BOOL "")
    if(LARCH64_ABI_1)
        message(STATUS "Build for Loongarch64 ABI 1.0 system")
        add_definitions(-DLA64_ABI_1)
@@ -67,6 +82,7 @@ if(RV64)
    set(ARM_DYNAREC OFF CACHE BOOL "")
    set(RV64_DYNAREC ON CACHE BOOL "")
    set(LARCH64_DYNAREC OFF CACHE BOOL "")
+    set(PPC64LE_DYNAREC OFF CACHE BOOL "")
 endif()
 if(PPC64LE)
    set(LD80BITS OFF CACHE BOOL "")
@@ -74,13 +90,15 @@ if(PPC64LE)
    set(ARM_DYNAREC OFF CACHE BOOL "")
    set(RV64_DYNAREC OFF CACHE BOOL "")
    set(LARCH64_DYNAREC OFF CACHE BOOL "")
+    set(PPC64LE_DYNAREC OFF CACHE BOOL "")
 endif()
-if(RK3399 OR RK3588 OR ODROIDN2 OR RPI3ARM64 OR RPI4ARM64 OR RPI5ARM64 OR RK3326 OR TEGRAX1 OR TEGRA_T194 OR TEGRA_T234 OR PHYTIUM OR SD845 OR SD865 OR SD888 OR SD8G2 OR SDORYON1 OR LX2160A OR M1 OR ARM64 OR ADLINK)
+if(RK3399 OR RK3588 OR ODROIDN2 OR RPI3ARM64 OR RPI4ARM64 OR RPI5ARM64 OR RK3326 OR TEGRAX1 OR TEGRA_T194 OR TEGRA_T234 OR NVIDIA_GB10 OR PHYTIUM OR SD845 OR SD865 OR SD888 OR SD8G2 OR SDORYON1 OR SD8EG5 OR LX2160A OR M1 OR ARM64 OR ADLINK)
    set(LD80BITS OFF CACHE BOOL "")
    set(NOALIGN OFF CACHE BOOL "")
    set(ARM_DYNAREC ON CACHE BOOL "")
    set(RV64_DYNAREC OFF CACHE BOOL "")
    set(LARCH64_DYNAREC OFF CACHE BOOL "")
+    set(PPC64LE_DYNAREC OFF CACHE BOOL "")
 endif()
 if(RK3399 OR ODROIDN2 OR RPI3ARM64 OR RPI4ARM64 OR RPI5ARM64)
    set(SAVE_MEM ON CACHE BOOL "")
@@ -108,6 +126,7 @@ option(NOALIGN "Set to ON if host device doesn't need re-align (i.e. i386)" ${NO
 option(ARM_DYNAREC "Set to ON to use ARM Dynamic Recompilation" ${ARM_DYNAREC})
 option(RV64_DYNAREC "Set to ON to use RISC-V Dynamic Recompilation" ${RV64_DYNAREC})
 option(LARCH64_DYNAREC "Set to ON to use LOONGARCH64 Dynamic Recompilation" ${LARCH64_DYNAREC})
+option(PPC64LE_DYNAREC "Set to ON to use PPC64LE Dynamic Recompilation" ${PPC64LE_DYNAREC})
 option(STATICBUILD "Set to ON to have a static build (Warning, not working)" ${STATICBUILD})
 option(NO_LIB_INSTALL "Set ON to not install a few x86_64 libs that are used by many program" ${NO_LIB_INSTALL})
 option(NO_CONF_INSTALL "Set ON to not install config files" ${NO_CONF_INSTALL})
@@ -173,6 +192,9 @@ elseif(TEGRA_T194)
 elseif(TEGRA_T234)
    add_definitions(-DTEGRA_T234)
    set(CFLAGS -pipe -march=armv8.2-a+fp16+simd+crypto+predres -mcpu=cortex-a78ae+crypto -mtune=cortex-a78ae)
+elseif(NVIDIA_GB10)
+    add_definitions(-DNVIDIA_GB10)
+    set(CFLAGS -pipe -mcpu=gb10 -mtune=gb10)
 elseif(PHYTIUM)
    add_definitions(-DPHYTIUM)
    set(CFLAGS -pipe -march=armv8.1-a+crc+simd+crypto)
@@ -189,6 +211,17 @@ elseif(SD888)
 elseif(SD8G2)
    add_definitions(-DSD8G2)
    set(CFLAGS -pipe -march=armv9-a+i8mm+sm4+sha3+rcpc+crypto+nosve+nosve2)
+elseif(SD8EG5)
+    add_definitions(-DSD8EG5)
+    # NOTE
+    # Even though Qualcomm Snapdragon 8 Elite Gen 5 relies on 3rd generation Oryon cores, it seems like the support for them 
+    # is not yet present in GCC and Clang, so for now we will use mostly the same optimization flags as SDORYON1, which houses
+    # 1st generation Oryon cores. This should still be fine since the main difference between the 
+    # two is the presence of SVE and SVE2, which are not yet used in Box64's ARM64 dynarec.
+    #
+    # As noted in the SDORYON1 section, GCC 14 and Clang 19 support oryon-1 as a -mcpu/-mtune argument, however it seems like 
+    # Fedora does not like them very much, so this option should be taken with a grain of salt. It worked in my test case just fine.
+    set(CFLAGS -pipe -march=armv8.7-a+crypto+sm4+sha3+fp16+sve+sve2 -mtune=oryon-1)
 elseif(SDORYON1)
    add_definitions(-DSDORYON1)
    # NOTE
@@ -214,6 +247,7 @@ elseif(M1)
    set(CFLAGS -pipe -march=armv8.5-a+simd+crypto)
 elseif(LARCH64)
    add_definitions(-DLA64)
+    add_definitions(-DBAD_PKILL)
    set(CFLAGS -pipe -march=loongarch64)
    set(ASMFLAGS  -pipe -march=loongarch64)
 elseif(RV64)
@@ -222,6 +256,8 @@ elseif(RV64)
    set(ASMFLAGS  -pipe -march=rv64gc)
 elseif(PPC64LE)
    add_definitions(-DPPC64LE)
+    set(CFLAGS -pipe -mcpu=power9)
+    set(ASMFLAGS  -pipe -mcpu=power9)
 elseif(LX2160A)
    add_definitions(-DLX2160A)
    set(CFLAGS -pipe -march=armv8-a+crypto+crc -mcpu=cortex-a72+crypto)
@@ -250,6 +286,9 @@ endif()
 if(BAD_SIGNAL)
    add_definitions(-DBAD_SIGNAL)
 endif()
+if(BAD_PKILL)
+    add_definitions(-DBAD_PKILL)
+endif()
 if(SAVE_MEM)
    add_definitions(-DSAVE_MEM)
 endif()
@@ -260,6 +299,9 @@ if(STATICBUILD)
    set(BUILD_SHARED_LIBS OFF)
    set(CMAKE_EXE_LINKER_FLAGS -static)
    set(HAVE_TRACE OFF)
+    if(PPC64LE)
+        add_definitions(-fno-stack-protector)
+    endif()
 endif()
 if(BOX32)
    add_definitions(-DBOX32)
@@ -294,6 +336,10 @@ if(HAVE_TRACE)
    add_definitions(-DHAVE_TRACE)
 endif()

+if(ZYDIS3)
+    add_definitions(-DZYDIS3)
+endif()
+
 if(ARM_DYNAREC)
    add_definitions(-DDYNAREC)
    add_definitions(-DARM64)
@@ -350,16 +396,20 @@ set_source_files_properties(
 set(OS_LINUX_SRC
    "${BOX64_ROOT}/src/os/backtrace.c"
    "${BOX64_ROOT}/src/os/emit_signals_linux.c"
+    "${BOX64_ROOT}/src/os/hostext_common.c"
+    "${BOX64_ROOT}/src/os/hostext_linux.c"
    "${BOX64_ROOT}/src/os/freq_linux.c"
    "${BOX64_ROOT}/src/os/os_linux.c"
    "${BOX64_ROOT}/src/os/perfmap.c"
    "${BOX64_ROOT}/src/os/symbolfuncs_linux.c"
-    "${BOX64_ROOT}/src/os/my_cpuid_linux.c"
-    "${BOX64_ROOT}/src/os/my_cpuid_common.c"
+    "${BOX64_ROOT}/src/os/random_linux.c"
+    "${BOX64_ROOT}/src/os/my_cpuid.c"
+    "${BOX64_ROOT}/src/os/sysinfo.c"
 )

 set(ELFLOADER_SRC
    "${BOX64_ROOT}/src/main.c"
+    "${BOX64_ROOT}/src/test.c"
    "${BOX64_ROOT}/src/core.c"
    "${BOX64_ROOT}/src/box64context.c"
    "${BOX64_ROOT}/src/build_info.c"
@@ -367,6 +417,7 @@ set(ELFLOADER_SRC
    "${BOX64_ROOT}/src/custommmap.c"
    "${BOX64_ROOT}/src/dynarec/dynarec.c"
    "${BOX64_ROOT}/src/elfs/elfloader.c"
+    "${BOX64_ROOT}/src/elfs/elfhacks.c"
    "${BOX64_ROOT}/src/elfs/elfparser.c"
    "${BOX64_ROOT}/src/elfs/elfhash.c"
    "${BOX64_ROOT}/src/elfs/elfload_dump.c"
@@ -390,6 +441,7 @@ set(ELFLOADER_SRC
    "${BOX64_ROOT}/src/libtools/auxval.c"
    "${BOX64_ROOT}/src/libtools/myalign.c"
    "${BOX64_ROOT}/src/libtools/signals.c"
+    "${BOX64_ROOT}/src/libtools/sigtools.c"
    "${BOX64_ROOT}/src/libtools/decopcode.c"
    "${BOX64_ROOT}/src/libtools/threads.c"
    "${BOX64_ROOT}/src/tools/bitutils.c"
@@ -427,6 +479,7 @@ if(BOX32)
        "${BOX64_ROOT}/src/tools/box32stack.c"
        "${BOX64_ROOT}/src/emu/x86int3.c"
        "${BOX64_ROOT}/src/librarian/globalsymbols32.c"
+        "${BOX64_ROOT}/src/libtools/box32_inputevent.c"
        "${BOX64_ROOT}/src/libtools/myalign32.c"
        "${BOX64_ROOT}/src/libtools/myalign64_32.c"
        "${BOX64_ROOT}/src/libtools/myalignxcb32.c"
@@ -452,26 +505,23 @@ if(NOT ANDROID)
    )
 endif()

+if(NOT STATICBUILD AND NOT ANDROID)
+    list(APPEND ELFLOADER_SRC
+        "${BOX64_ROOT}/src/libtools/libdl.c"
+        "${BOX64_ROOT}/src/libtools/vulkanoverlay.c"
+    )
+endif()
+
 set(INTERPRETER
    "${BOX64_ROOT}/src/emu/x64run.c"
    "${BOX64_ROOT}/src/emu/x64run0f.c"
-    "${BOX64_ROOT}/src/emu/x64run64.c"
    "${BOX64_ROOT}/src/emu/x64run66.c"
    "${BOX64_ROOT}/src/emu/x64run660f.c"
    "${BOX64_ROOT}/src/emu/x64run66f20f.c"
    "${BOX64_ROOT}/src/emu/x64run66f30f.c"
-    "${BOX64_ROOT}/src/emu/x64run6664.c"
    "${BOX64_ROOT}/src/emu/x64run66d9.c"
    "${BOX64_ROOT}/src/emu/x64run66dd.c"
    "${BOX64_ROOT}/src/emu/x64run66f0.c"
-    "${BOX64_ROOT}/src/emu/x64run67.c"
-    "${BOX64_ROOT}/src/emu/x64run6764.c"
-    "${BOX64_ROOT}/src/emu/x64run67avx.c"
-    "${BOX64_ROOT}/src/emu/x64run67_32.c"
-    "${BOX64_ROOT}/src/emu/x64run6764_32.c"
-    "${BOX64_ROOT}/src/emu/x64run670f.c"
-    "${BOX64_ROOT}/src/emu/x64run6766.c"
-    "${BOX64_ROOT}/src/emu/x64run67660f.c"
    "${BOX64_ROOT}/src/emu/x64rund8.c"
    "${BOX64_ROOT}/src/emu/x64rund9.c"
    "${BOX64_ROOT}/src/emu/x64runda.c"
@@ -519,8 +569,13 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedatomic.c"
    "${BOX64_ROOT}/src/wrapped/wrappedatspi.c"
    "${BOX64_ROOT}/src/wrapped/wrappedayatanaappindicator3.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedayatanaindicator3.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedayatanaido3.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedavahiclient.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedavahicommon.c"
    "${BOX64_ROOT}/src/wrapped/wrappedbz2.c"
    "${BOX64_ROOT}/src/wrapped/wrappedcap.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedcuda.c"
    "${BOX64_ROOT}/src/wrapped/wrappedcairo.c"
    "${BOX64_ROOT}/src/wrapped/wrappedcairogobject.c"
    "${BOX64_ROOT}/src/wrapped/wrappedcrashhandler.c"
@@ -574,6 +629,7 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedgthread2.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgtk3.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgtkx112.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedharfbuzzsubset.c"
    "${BOX64_ROOT}/src/wrapped/wrappedkrb5.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlcms2.c"
    "${BOX64_ROOT}/src/wrapped/wrappedldlinux.c"
@@ -581,11 +637,13 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedlibavcodec58.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibavformat58.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibavutil56.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibblas.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibbsd.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibc.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibcmusl.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibcrypt.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibcups.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibcupsimage.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibdl.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibdrm.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibdrmamdgpu.c"
@@ -597,9 +655,12 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedlibgl.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibglu.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibglx.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibglxnvidia.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibgpgme.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibharfbuzz.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibibus.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibice.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedliblapack.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibm.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibncurses.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibncurses6.c"
@@ -607,9 +668,11 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedlibncursesw6.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibnuma.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibogg.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibomp.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibpanel.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibpci.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibpcre.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibpcre216.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibpthread.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibresolv.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibrt.c"
@@ -626,6 +689,7 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedlibvawayland.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibvax11.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibvdpau.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibvpx.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibvorbis.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibx11.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibx11xcb.c"
@@ -678,6 +742,7 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappednspr4.c"
    "${BOX64_ROOT}/src/wrapped/wrappednss3.c"
    "${BOX64_ROOT}/src/wrapped/wrappednssutil3.c"
+    "${BOX64_ROOT}/src/wrapped/wrappednvml.c"
    "${BOX64_ROOT}/src/wrapped/wrappedopenal.c"
    "${BOX64_ROOT}/src/wrapped/wrappedopencl.c"
    "${BOX64_ROOT}/src/wrapped/wrappedpango.c"
@@ -705,6 +770,8 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedsdl2mixer.c"
    "${BOX64_ROOT}/src/wrapped/wrappedsdl2net.c"
    "${BOX64_ROOT}/src/wrapped/wrappedsdl2ttf.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedsdl3.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedsdl3ttf.c"
    "${BOX64_ROOT}/src/wrapped/wrappedselinux.c"
    "${BOX64_ROOT}/src/wrapped/wrappedsmime3.c"
    "${BOX64_ROOT}/src/wrapped/wrappedsmpeg2.c"
@@ -746,6 +813,8 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedicui18n74.c"
    "${BOX64_ROOT}/src/wrapped/wrappedicuuc75.c"
    "${BOX64_ROOT}/src/wrapped/wrappedicui18n75.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedicuuc76.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedicui18n76.c"
    "${BOX64_ROOT}/src/wrapped/wrappedidn2.c"
    "${BOX64_ROOT}/src/wrapped/wrappedanl.c"
    "${BOX64_ROOT}/src/wrapped/wrappedpsl5.c"
@@ -792,7 +861,7 @@ if(NOT CI)
        OUTPUT "${BOX64_ROOT}/src/wrapped/generated/functions_list.txt"
        COMMAND "${PYTHON_EXECUTABLE}" "${BOX64_ROOT}/rebuild_wrappers.py"
        "${BOX64_ROOT}"
-        "PANDORA" "HAVE_LD80BITS" "NOALIGN" "HAVE_TRACE" "ANDROID" "TERMUX" "STATICBUILD" "LA64" "--"
+        "PANDORA" "HAVE_LD80BITS" "NOALIGN" "HAVE_TRACE" "ANDROID" "TERMUX" "STATICBUILD" "LA64" "PPC64LE" "--"
        ${WRAPPEDS_HEAD}
        MAIN_DEPENDENCY "${BOX64_ROOT}/rebuild_wrappers.py"
        DEPENDS ${WRAPPEDS} ${WRAPPEDS_HEAD}
@@ -833,6 +902,7 @@ if(BOX32)
        "${BOX64_ROOT}/src/wrapped32/wrappedcrashhandler.c"
        "${BOX64_ROOT}/src/wrapped32/wrappedlibgl.c"
        "${BOX64_ROOT}/src/wrapped32/wrappedlibglu.c"
+        "${BOX64_ROOT}/src/wrapped32/wrappedlibglxnvidia.c"
        "${BOX64_ROOT}/src/wrapped32/wrappedsdl1.c"
        "${BOX64_ROOT}/src/wrapped32/wrappedsdl2.c"
        "${BOX64_ROOT}/src/wrapped32/wrappedsdl2image.c"
@@ -874,6 +944,9 @@ if(BOX32)
        "${BOX64_ROOT}/src/wrapped32/wrappedmount.c"
        "${BOX64_ROOT}/src/wrapped32/wrappeddbus.c"
        "${BOX64_ROOT}/src/wrapped32/wrappedcurl.c"
+        "${BOX64_ROOT}/src/wrapped32/wrappednvml.c"
+         "${BOX64_ROOT}/src/wrapped32/wrappedpulsesimple.c"
+         "${BOX64_ROOT}/src/wrapped32/wrappedlibogg.c"
    )
    endif()
    string(REPLACE ".c" "_private.h" MODROOT ${BOX64_ROOT})
@@ -911,6 +984,7 @@ if(DYNAREC)
        "${BOX64_ROOT}/src/dynarec/dynablock.c"
        "${BOX64_ROOT}/src/dynarec/dynarec_native.c"
        "${BOX64_ROOT}/src/dynarec/dynarec_native_functions.c"
+        "${BOX64_ROOT}/src/dynarec/dynacache_reloc.c"
        "${BOX64_ROOT}/src/emu/x64test.c"
    )
 endif()
@@ -924,6 +998,8 @@ if(ARM_DYNAREC)
        "${BOX64_ROOT}/src/dynarec/arm64/arm64_immenc.c"
        "${BOX64_ROOT}/src/dynarec/arm64/arm64_printer.c"
        "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_jmpnext.c"
+        "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_consts.c"
+        "${BOX64_ROOT}/src/dynarec/arm64/updateflags_arm64.c"
    )
    set(DYNAREC_ASM
        ${DYNAREC_ASM}
@@ -941,13 +1017,7 @@ if(ARM_DYNAREC)
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_emit_shift.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_00.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_0f.c"
-    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_64.c"
-    #"${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_65.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_66.c"
-    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_67.c"
-    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_67_32.c"
-    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_67_avx.c"
-    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_6764_32.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_d8.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_d9.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_da.c"
@@ -960,7 +1030,6 @@ if(ARM_DYNAREC)
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_660f.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_66f20f.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_66f30f.c"
-    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_6664.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_66f0.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_f20f.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_f30f.c"
@@ -975,6 +1044,8 @@ if(ARM_DYNAREC)
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_avx_f2_0f38.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_avx_f2_0f3a.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_avx_f3_0f38.c"
+
+    "${BOX64_ROOT}/src/dynarec/arm64/updateflags_arm64_pass.c"
    )
 endif()

@@ -986,7 +1057,7 @@ if(RV64_DYNAREC)
        "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_arch.c"
        "${BOX64_ROOT}/src/dynarec/rv64/rv64_printer.c"
        "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_jmpnext.c"
-        "${BOX64_ROOT}/src/rv64detect.c"
+        "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_consts.c"
        )
    set(DYNAREC_ASM
        ${DYNAREC_ASM}
@@ -1009,12 +1080,7 @@ if(RV64_DYNAREC)
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_00_3.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_0f.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_0f_vector.c"
-    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_64.c"
-    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_64_vector.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_66.c"
-    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_67.c"
-    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_67_vector.c"
-    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_67_32.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_d8.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_d9.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_da.c"
@@ -1029,13 +1095,17 @@ if(RV64_DYNAREC)
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_660f_vector.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_66f20f.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_66f30f.c"
-    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_6664.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_66f0.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_f20f.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_f20f_vector.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_f30f.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_f30f_vector.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_avx.c"
+    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_avx_0f.c"
+    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_avx_66_0f.c"
+    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_avx_66_0f38.c"
+    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_avx_66_0f3a.c"
+    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_avx_f2_0f.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_avx_f3_0f.c"
    )
 endif()
@@ -1047,6 +1117,7 @@ if(LARCH64_DYNAREC)
        "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_functions.c"
        "${BOX64_ROOT}/src/dynarec/la64/la64_printer.c"
        "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_jmpnext.c"
+        "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_consts.c"
    )
    set(DYNAREC_ASM
        ${DYNAREC_ASM}
@@ -1064,17 +1135,33 @@ if(LARCH64_DYNAREC)
    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_emit_logic.c"
    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_00.c"
    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_0f.c"
-    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_64.c"
    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_66.c"
-    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_6664.c"
    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_660f.c"
    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_66f0.c"
    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_66f20f.c"
    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_66f30f.c"
-    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_67.c"
    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_f0.c"
    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_f20f.c"
    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_f30f.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_avx.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_avx_0f.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_avx_0f38.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_avx_66_0f.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_avx_66_0f38.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_avx_66_0f3a.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_avx_f2_0f.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_avx_f2_0f38.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_avx_f2_0f3a.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_avx_f3_0f.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_avx_f3_0f38.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_d8.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_d9.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_da.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_db.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_dc.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_dd.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_de.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_df.c"
    )
 endif()

@@ -1143,13 +1230,13 @@ if(DYNAREC)
    add_custom_command(
        OUTPUT "${BOX64_ROOT}/src/git_head.h"
        COMMAND sh -c "echo \\\#define GITREV \\\"$(git rev-parse --short HEAD)\\\">\"${BOX64_ROOT}/src/git_head.h\""
-        DEPENDS ${DYNAREC_SRC} ${DYNAREC_ASM} ${DYNAREC_PASS} ${ELFLOADER_SRC} ${INTERPRETER} ${WRAPPEDS} ${WRAPPEDS32}
+        DEPENDS ${DYNAREC_SRC} ${DYNAREC_ASM} ${DYNAREC_PASS} ${ELFLOADER_SRC} ${INTERPRETER} ${WRAPPEDS} ${WRAPPEDS32} ${WRAPPEDS_HEAD} ${WRAPPEDS32_HEAD}
        VERBATIM)
 else()
    add_custom_command(
        OUTPUT "${BOX64_ROOT}/src/git_head.h"
        COMMAND sh -c "echo \\\#define GITREV \\\"$(git rev-parse --short HEAD)\\\">\"${BOX64_ROOT}/src/git_head.h\""
-        DEPENDS ${ELFLOADER_SRC} ${INTERPRETER} ${WRAPPEDS} ${WRAPPEDS32}
+        DEPENDS ${ELFLOADER_SRC} ${INTERPRETER} ${WRAPPEDS} ${WRAPPEDS32} ${WRAPPEDS_HEAD} ${WRAPPEDS32_HEAD}
        VERBATIM)
 endif()

@@ -1185,6 +1272,7 @@ else()
        endif()
    else()
 	    set_property(TARGET ${BOX64} APPEND_STRING PROPERTY LINK_FLAGS "-Wl,--no-as-needed -lc -lm -ldl -lrt -lpthread -lresolv -lutil -Wl,--as-needed -pthread ")
+		target_link_libraries(${BOX64} c m dl pthread resolv util c)
    endif()
 endif()
 if(DYNAREC)
@@ -1218,8 +1306,12 @@ string(COMPARE EQUAL "${CMAKE_SYSTEM_PROCESSOR}" "i686"  _x86)
 string(COMPARE EQUAL "${CMAKE_SYSTEM_PROCESSOR}" "x86_64"  _x86_64)
 string(COMPARE EQUAL "${CMAKE_SYSTEM_PROCESSOR}" "aarch64"  _aarch64)
 string(COMPARE EQUAL "${CMAKE_SYSTEM_PROCESSOR}" "riscv64"  _riscv64)
+string(COMPARE EQUAL "${CMAKE_SYSTEM_PROCESSOR}" "ppc64le"  _ppc64le)
+if(NOT _ppc64le)
+  string(COMPARE EQUAL "${CMAKE_SYSTEM_PROCESSOR}" "powerpc64le"  _ppc64le)
+endif()

-if(_x86_64 OR _aarch64)
+if(_x86_64 OR _aarch64 OR _ppc64le)
  add_definitions(-DCONFIG_64BIT)
 endif()

@@ -1341,12 +1433,13 @@ if(WOW64 AND ARM_DYNAREC)
    include(ExternalProject)
    ExternalProject_Add(
        wowbox64
-        SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/wow64"
+        SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/wine/wow64"
        CMAKE_ARGS 
-            -DCMAKE_TOOLCHAIN_FILE=${CMAKE_CURRENT_SOURCE_DIR}/wow64/toolchain_mingw.cmake
+            -DCMAKE_TOOLCHAIN_FILE=${CMAKE_CURRENT_SOURCE_DIR}/wine/toolchain_mingw.cmake
            -DDYNAREC_PASS_STR=${DYNAREC_PASS_STR}
            -DDYNAREC_ASM_STR=${DYNAREC_ASM_STR}
            -DINTERPRETER_STR=${INTERPRETER_STR}
+            -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}
        INSTALL_COMMAND ""
        TEST_COMMAND ""
        BUILD_ALWAYS ON
@@ -1385,8 +1478,10 @@ if(_aarch64)
 	else()
    		set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "arm64")
 	endif()
-elseif(__riscv64)
+elseif(_riscv64)
    set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "riscv64")
+elseif(_ppc64le)
+    set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "ppc64el")
 elseif(_x86_64)
    set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "x86_64")
 else()
@@ -1510,7 +1605,7 @@ add_test(irelative_reloc ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref20.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-add_test(longjumpInSignals ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+add_test(longjumpInSignals_nocosim ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test21 -D TEST_OUTPUT=tmpfile21.txt
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref21.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
@@ -1579,6 +1674,11 @@ add_test(x87pc ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}

 set_tests_properties(x87pc PROPERTIES ENVIRONMENT "BOX64_DYNAREC_X87DOUBLE=2")

+add_test(Boundary ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test33 -D TEST_OUTPUT=tmpfile33.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref33.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
 else()

 add_test(bootSyscall ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
@@ -1666,7 +1766,7 @@ add_test(irelative_reloc ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref20.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-# add_test(longjumpInSignals ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+# add_test(longjumpInSignals_nocosim ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
 #     -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test21_android -D TEST_OUTPUT=tmpfile21.txt
 #     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref21.txt
 #     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
@@ -1838,7 +1938,7 @@ if(BOX32)

    set_tests_properties(sse_32bits PROPERTIES ENVIRONMENT "BOX64_DYNAREC_FASTNAN=0;BOX64_DYNAREC_FASTROUND=0")

-    add_test(NAME longjumpInSignals_32bits COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    add_test(NAME longjumpInSignals_32bits_nocosim COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
        -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests32/test18 -D TEST_OUTPUT=tmpfile32_18.txt
        -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests32/ref18.txt
        -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
@@ -2,22 +2,21 @@

 # Box64: Linux Userspace x86-64 Emulator with a Twist

-[Changelog](https://github.com/ptitSeb/box64/blob/main/docs/CHANGELOG.md) | [中文](https://github.com/ptitSeb/box64/blob/main/README_CN.md) | [Українська](https://github.com/ptitSeb/box64/blob/main/README_UK.md) | [Report an Error](https://github.com/ptitSeb/box64/issues/new)
+[Website](https://box86.org/) | [Changelog](https://github.com/ptitSeb/box64/blob/main/docs/CHANGELOG.md) | [中文](https://github.com/ptitSeb/box64/blob/main/README_CN.md) | [Українська](https://github.com/ptitSeb/box64/blob/main/README_UK.md) | [Report an Error](https://github.com/ptitSeb/box64/issues/new)

-![Build Status](https://app.travis-ci.com/ptitSeb/box64.svg?branch=main) ![Stars](https://img.shields.io/github/stars/ptitSeb/box64) ![Forks](https://img.shields.io/github/forks/ptitSeb/box64) ![Contributors](https://img.shields.io/github/contributors/ptitSeb/box64) ![Pull Requests](https://img.shields.io/github/issues-pr/ptitSeb/box64) ![Issues](https://img.shields.io/github/issues/ptitSeb/box64)
+![Build Status](https://github.com/ptitSeb/box64/actions/workflows/release.yml/badge.svg) ![Stars](https://img.shields.io/github/stars/ptitSeb/box64) ![Forks](https://img.shields.io/github/forks/ptitSeb/box64) ![Contributors](https://img.shields.io/github/contributors/ptitSeb/box64) ![Pull Requests](https://img.shields.io/github/issues-pr/ptitSeb/box64) ![Issues](https://img.shields.io/github/issues/ptitSeb/box64)
+
+---
+
+**Check out our game compatibility list at [https://box86.org/app/](https://box86.org/app/), please help us [improve it](https://github.com/ptitSeb/box86-compatibility-list/issues)!**

 ---

 Box64 enables running x86_64 Linux programs, including games, on non-x86_64 Linux systems such as Arm (64-bit little-endian host system required).

-Check out Box64 in action on YouTube:
- [MicroLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ)
- [Pi Labs](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g)
- [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA)
-
 Box64 leverages native system libraries (libc, libm, SDL, OpenGL), offering ease of integration and surprising performance in many applications. For performance benchmarks, check [this analysis](https://box86.org/index.php/2021/06/game-performances/).

-With DynaRec for ARM64, RV64 and LA64 platforms, Box64 achieves a speed boost 5-10x faster than the interpreter alone. For a deeper look at DynaRec, see [Inner workings](https://box86.org/2021/07/inner-workings-a-high%e2%80%91level-view-of-box86-and-a-low%e2%80%91level-view-of-the-dynarec/).
+With DynaRec for Arm, RISC-V and LoongArch platforms, Box64 achieves a speed boost 5-10x faster than the interpreter alone. For a deeper look at DynaRec, see [Inner workings](https://box86.org/2021/07/inner-workings-a-high%e2%80%91level-view-of-box86-and-a-low%e2%80%91level-view-of-the-dynarec/).

 <img src="docs/img/Box64Icon.png" width="96" height="96" alt="Box64 Icon">

@@ -25,46 +24,53 @@ _Logo and icon by @grayduck - Thank you!_

 ---

-## 📖 Usage
+## Usage

-Box64 offers environment variables to control its behavior. For details, see [Usage Documentation](docs/USAGE.md).
+- `$ box64 ./program [args]` to run Linux x86_64 programs.
+- `$ box64 -k` to kill all the emulated processes.
+- `$ box64-bash` to have an x86_64 bash environment.
+- Want to use wine? see [Wine Documentation](docs/WINE.md).
+
+Box64 also offers environment variables and rcfile to control its behavior. For details, see [Usage Documentation](docs/USAGE.md).

 ---

-## 🚀 Compiling and Installation
+## Compiling and Installation

 - [Compilation Instructions](https://github.com/ptitSeb/box64/blob/main/docs/COMPILE.md)
 - [Bundle x86 Libraries](https://github.com/ptitSeb/box64/blob/main/docs/BUNDLE-X86-LIBS.md)
- [Wine Installation for Box64](https://github.com/ptitSeb/box64/blob/main/docs/X64WINE.md)
+- [Wine Installation for Box64](https://github.com/ptitSeb/box64/blob/main/docs/WINE.md)

 ---

-## 🔄 Version History
+## Version History

 See the [Changelog](docs/CHANGELOG.md) for version updates.

 ---

-## 🖥️ Platform Compatibility Notes
+## Platform Compatibility Notes

-Box64 requires 64-bit libraries on the host system, as it directly translates x86_64 function calls. For 32-bit binaries, use Box86.
+Box64 requires 64-bit libraries on the host system, as it directly translates x86_64 function calls. For 32-bit binaries, use Box86 or Box32.

 **Notes** 

-1. Box32 mode is in the making, which aims to support 32-bit binaries on Box64, you can enable it at build time, but do not expect everything to work. There is actually a few things that work, many that don't. It's a known issue, no need to create tickets about it for now.
-2. Some installers may default to x86 on 64-bit host OSes, causing compatibility issues. Workaround: Use a fake `uname` that returns "x86_64" for `-m`.
+1. Box32 is still experimental.
+2. Some shell scripts (such as the GOG game installer) may rely on `uname -m` to determine the current architecture. Please run them as `box64 script.sh` to allow Box64 to take over.

 ---

-## ⚙️ Configuration
+## Configuration

 Box64 has configuration files located in `/etc/box64.box64rc` and `~/.box64rc`, both formatted as `.ini` files. If you don’t want to use the default `/etc/box64.box64rc` file, copy it to `~/.box64rc` to ensure compatibility.

 Settings priority: `~/.box64rc` > `/etc/box64.box64rc` > Command line.

+For more information, check [Usage Documentation](docs/USAGE.md).
+
 ---

-## 📄 Additional Platform-Specific Notes
+## Additional Platform-Specific Notes

 ### Unity Game Emulation

@@ -72,13 +78,9 @@ Settings priority: `~/.box64rc` > `/etc/box64.box64rc` > Command line.
 - **Tip for Pi4 and Pi5 Users**: Set `MESA_GL_VERSION_OVERRIDE=3.2` with `BOX64_DYNAREC_STRONGMEM=1` to prevent freezes and enable strong memory mode.
 - **Using Panfrost**: For better compatibility on ARM, enable `PAN_MESA_DEBUG=gl3` to force higher OpenGL profiles. This can help if a game starts but quits unexpectedly before showing any content.

-### GTK Programs
-
-Box64 wraps GTK libraries, supporting both gtk2 and gtk3.
-
 ### Steam

-Steam requires Box86 due to its 32-bit client app, but uses 64-bit local server binaries. Systems with less than 6GB RAM may need a swap file for optimal performance.
+See [Steam Documentation](docs/STEAM.md).

 ### Wine

@@ -86,9 +88,7 @@ Box64 supports Wine64 and Proton. For 32-bit components, Box86 is required. Syst

 **Note**: You can use Wine WOW64 build to run x86 Windows programs in Box64-only environments, this is still experimental, but it works in most cases.

-### Vulkan
-
-Box64 wraps Vulkan libraries.
+For more information, check [Wine Documentation](docs/WINE.md).

 ----
 Final word
@@ -2,22 +2,21 @@

 # Box64：Linux 用户空间 x86-64 模拟器（支持原生库）

-[更新日志](https://github.com/ptitSeb/box64/blob/main/docs/CHANGELOG.md) | [English](https://github.com/ptitSeb/box64/blob/main/README.md) | [Українська](https://github.com/ptitSeb/box64/blob/main/README_UK.md) | [Report an Error](https://github.com/ptitSeb/box64/issues/new)
+[官网](https://box86.org/) | [更新日志](https://github.com/ptitSeb/box64/blob/main/docs/CHANGELOG.md) | [English](https://github.com/ptitSeb/box64/blob/main/README.md) | [Українська](https://github.com/ptitSeb/box64/blob/main/README_UK.md) | [Report an Error](https://github.com/ptitSeb/box64/issues/new)

 ![Build Status](https://app.travis-ci.com/ptitSeb/box64.svg?branch=main) ![Stars](https://img.shields.io/github/stars/ptitSeb/box64) ![Forks](https://img.shields.io/github/forks/ptitSeb/box64) ![Contributors](https://img.shields.io/github/contributors/ptitSeb/box64) ![Pull Requests](https://img.shields.io/github/issues-pr/ptitSeb/box64) ![Issues](https://img.shields.io/github/issues/ptitSeb/box64)

 ---

-Box64 可以在非 x86_64 Linux 主机系统（比如 Arm）上运行 x86_64 Linux 程序（比如游戏），目前仅支持 64 位小端主机系统。
+**请查看我们的游戏兼容性列表：[https://box86.org/app/](https://box86.org/app/)，并请帮助我们[改进它](https://github.com/ptitSeb/box86-compatibility-list/issues)！**

-在 YouTube 上查看 Box64 的运行效果:
- [MicroLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ)
- [Pi Labs](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g)
- [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA)
+---
+
+Box64 可以在非 x86_64 Linux 主机系统（比如 Arm）上运行 x86_64 Linux 程序（比如游戏），目前仅支持 64 位小端主机系统。

 Box64 可以使用本机原生的系统库（libc、libm、SDL、OpenGL），有着易于集成的特性和令人惊讶的性能。如需性能基准测试，请查看[此分析](https://box86.org/index.php/2021/06/game-performances/)。

-借助适用于 ARM64，RV64 以及 LA64 平台的 DynaRec，Box64 的速度比单独使用解释器快 5-10 倍。如需深入了解 DynaRec，请参考[此文章](https://box86.org/2021/07/inner-workings-a-high%e2%80%91level-view-of-box86-and-a-low%e2%80%91level-view-of-the-dynarec/)。
+借助适用于 Arm，RISC-V 以及龙架构平台的 DynaRec，Box64 的速度比单独使用解释器快 5-10 倍。如需深入了解 DynaRec，请参考[此文章](https://box86.org/2021/07/inner-workings-a-high%e2%80%91level-view-of-box86-and-a-low%e2%80%91level-view-of-the-dynarec/)。

 <img src="docs/img/Box64Icon.png" width="96" height="96" alt="Box64 Icon">

@@ -25,58 +24,63 @@ _Logo 由 @grayduck 制作，感谢！_

 ---

-## 📖 使用方法
+## 使用方法

-Box64 提供了环境变量来控制其行为。有关详细信息请参阅[使用文档](docs/USAGE.md)。
+- `$ box64 ./program [args]` 运行 Linux x86_64 程序。
+- `$ box64 -k` 杀死所有模拟的进程。
+- `$ box64-bash` 进入 x86_64 bash 环境。
+- 想要使用 wine？请参阅 [Wine 使用说明](docs/WINE.md)。
+
+Box64 还提供了环境变量和 Rc 文件来控制其行为。有关详细信息请参阅[使用文档](docs/USAGE.md)。

 ---

-## 🚀 编译和安装
+## 编译和安装

 - [编译说明](https://github.com/ptitSeb/box64/blob/main/docs/COMPILE.md)
- [安装 Wine](https://github.com/ptitSeb/box64/blob/main/docs/X64WINE.md)
+- [安装 Wine](https://github.com/ptitSeb/box64/blob/main/docs/WINE.md)

 ---

-## 🔄 版本历史
+## 版本历史

 请参阅[变更日志](docs/CHANGELOG.md)了解版本更新说明。

 ---

-## 🖥️ 平台兼容性说明
+## 平台兼容性说明

-Box64 直接转换 x86_64 函数调用到本机原生库，因此需要主机系统上的 64 位库。对于 32 位二进制文件，请使用 Box86。
+Box64 直接转换 x86_64 函数调用到本机原生库，因此需要主机系统上的 64 位库。对于 32 位二进制文件，请使用 Box86 或 Box32。

 **注意事项**

-1. Box32 模式正在制作中，旨在支持 Box64 上运行 32 位二进制文件，敬请期待！
-2. 某些安装程序可能会将 64 位的主机操作系统判定为 x86 导致兼容性问题。解决方法：使用假的 `uname` ，并使它在运行参数为 `-m` 时返回 `x86_64`。
+1. Box32 仍然是实验性质。
+2. 某些 Shell 脚本（比如 GOG 游戏安装脚本）可能会依赖 `uname -m` 判断当前架构，请使用 `box64 script.sh` 的方式运行以便让 Box64 接管。

 ---

-## ⚙️ 配置
+## 配置

 Box64 的配置文件位于 `/etc/box64.box64rc` 和 `~/.box64rc`，均为 `.ini` 格式的文件。如果您不想使用默认的 `/etc/box64.box64rc` 文件，请将其复制到 `~/.box64rc` 以确保兼容性。

 配置项的优先级：`~/.box64rc` > `/etc/box64.box64rc` > 命令行。

+更多信息，请参阅[使用文档](docs/USAGE.md)。
+
 ---

-## 📄 特定平台的其他说明
+## 特定平台的其他说明

 ### Unity 游戏
 - 许多 Unity 游戏需要 OpenGL 3+，这对 ARM/RISC-V SBC 来说可能具有挑战性。
 - **对 Pi4 和 Pi5 用户的提示**：设置 `MESA_GL_VERSION_OVERRIDE=3.2` 和 `BOX64_DYNAREC_STRONGMEM=1` 以防止卡死并启用强内存模式。
 - **Panfrost**：为了在 ARM 上实现更好的兼容性，启用 `PAN_MESA_DEBUG=gl3` 以强制更高的 OpenGL Profile。如果游戏可以启动但在显示任何内容之前意外退出，这个设置可能会有所帮助。

-### GTK 程序
-
-Box64 包装了 GTK 库，支持 gtk2 和 gtk3。
-
 ### Steam

-由于 Steam 客户端应用是 32 位的，需要使用 Box86，但其本地服务器是 64 位的二进制文件。对于内存少于 6GB 的系统，可能需要使用交换文件以获得最佳性能。
+Box64 支持 Linux Steam 和 Windows Steam，更多细节请参考[Steam 文档](docs/STEAM.md)。
+
+> 由于 Linux Steam 客户端应用是 32 位的，需要使用 Box86 或 Box32，但其本地服务器是 64 位的二进制文件。对于内存少于 6GB 的系统，可能需要使用交换文件以获得最佳性能。

 ### Wine

@@ -84,9 +88,7 @@ Box64 支持 Wine64 和 Proton。对于 32 位组件，需要 Box86。配备 Box

 **提示**：您可以使用 Wine WOW64 版本在仅限 Box64 的环境中运行 x86 Windows 程序，这项支持仍处于实验阶段，但在大多数情况下都可以工作。

-### Vulkan
-
-Box64 包装了 Vulkan 库。
+有关更多信息，请参阅 [Wine 使用说明](docs/WINE.md)。

 ----

@@ -1,3 +1,5 @@
+> Внимание: Этот документ устарел. Пожалуйста, помогите нам его перевести.
+
 ![Official logo](docs/img/Box64Logo.png "Official Logo")

 Емулятор простіра користувача x86_64 із родзинкою
@@ -10,8 +12,6 @@

 Box64 дозволяє запускати x86_64 Linux програми (наприклад, ігри) на системах Linux відмінних від x86_64, наприклад ARM (хост-система має мати 64-розрядний порядок редагування).

-Ви можете знайти багато відео про Box64 в YouTube каналах [MicroLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ), [Pi Labs](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g) або [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA)
-
 Оскільки Box64 використовує рідні версії деяких «системних» бібліотек, таких як libc, libm, SDL і OpenGL, його легко інтегрувати та використовувати з більшістю програм, а продуктивність у багатьох випадках може бути напрочуд високою. Подивіться на цей стендовий аналіз для прикладу [тут](https://box86.org/index.php/2021/06/game-performances/).

 Box64 інтегрується з DynaRec (динамічний рекомпілятор) для платформ ARM64 і RV64 забезпечуючи підвищення швидкості в 5-10 разів, ніж використання лише інтерпретатора. Деякі високорівневі відомості про те, як працює DynaRec, можна знайти [тут](https://box86.org/2021/07/inner-workings-a-high%e2%80%91level-view-of-box86-and-a-low%e2%80%91level-view-of-the-dynarec/).
@@ -37,7 +37,7 @@ Box64 інтегрується з DynaRec (динамічний рекомпіл
 Компіляція/Встановлення
 ----
 > Інструкцію зі складання Box64 можна знайти [тут](docs/COMPILE.md).\
-> Інструкцію з встановлення Wine для Box64 можна знайти [тут](docs/X64WINE.md).
+> Інструкцію з встановлення Wine для Box64 можна знайти [тут](docs/WINE.md).

 ----
 Історія версій/Журнал змін
@@ -1,66 +1,89 @@
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/bluez-libs-5.72-2.el9.i686.rpm,af1aba8e6fa5ccf33032d58a3acc59368c07d532293396db205fb9a4bc1241ae
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/bluez-libs-5.72-2.el9.x86_64.rpm,49c273813b4616240acdae8dae893192658c7541482f9bbb2c46dfe19b61877d
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/fribidi-1.0.10-6.el9.2.i686.rpm,75f808f0f5f93aceb4e477aa5f302edbed07a605a0644f0ea929d4ea609e36a0
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/fribidi-1.0.10-6.el9.2.x86_64.rpm,b4763ca410f8479b2775efb55ba09e6722948af7526180a56309e3e090c27010
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/gmp-6.2.0-13.el9.i686.rpm,fe352f7d197ebdf2c92bbd6e548f9f6b7ac9e69e7a9d6c9861a6274482e79d73
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/gmp-6.2.0-13.el9.x86_64.rpm,babe0b1385e723476a216d5f4b72453e031c3d683b19344786c2bddf7203cb04
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/gnutls-3.8.3-4.el9_4.i686.rpm,64cf0b0136d3f9b80b07ce2d4b8ce6a4ab86617e8ef9baf560bdfa5edef825ea
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/gnutls-3.8.3-4.el9_4.x86_64.rpm,16c48c504fe62f8e19cac7c6442d3e08bba487f610c2f533116dac32357cda86
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/harfbuzz-2.7.4-10.el9.i686.rpm,5157daf2ab0a03b403d37c60b150b8f0f866754714f81e3d96368e06a16339ce
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/harfbuzz-2.7.4-10.el9.x86_64.rpm,1f81073019abe4176d4496723a89b55a349c31f507e96397a0b3efa7cea0ff61
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libblkid-2.37.4-20.el9.i686.rpm,f9e62d2768c31948b53268b7de51bd1c7342cdf39a328a0d89d933b4aa0c5dd1
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libblkid-2.37.4-20.el9.x86_64.rpm,ed02d978a36488b38bad62af21d4045367c1f0b00d00c90ded01e2fc8e190536
 https://archive.debian.org/debian/pool/main/c/curl/libcurl3-gnutls_7.74.0-1.3+deb11u7~bpo11+1_amd64.deb,d5ae55e3a9838d12358530cd03218cc0d18127a84e9cfc73bdf79dacd40b25e4
 https://archive.debian.org/debian/pool/main/c/curl/libcurl3-gnutls_7.74.0-1.3+deb11u7~bpo11+1_i386.deb,625e5189304d4d3bfb27cb821e257816c5ce761b1696b5da5ca4fa1035c48463
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/libepoxy-1.5.5-4.el9.i686.rpm,fe35e8de9b68c8d7e7b6b525580fc36cf9c86412cdb2b022dc01f95801771981
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/libepoxy-1.5.5-4.el9.x86_64.rpm,b6d1e7f1c832882ceb05187cbc2d8fe1f459bcd88c7160487ee6f734e1a5d017
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/libffi-devel-3.4.2-8.el9.i686.rpm,0c6f2ecc324680c6424bee4b1dbbdad2863f0f7f3130b196fc6e2b2823f07bc4
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/libffi-devel-3.4.2-8.el9.x86_64.rpm,483e4c58cb8af94143a0ff4b3e49b62cb7d6896766c8c0d9fe3659e68945b50e
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libgomp-11.5.0-2.el9.alma.1.i686.rpm,686ac26afe2073a3607f26fd1592a8ccd9b78f6eec6e872528634b43c4e04dd7
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libgomp-11.5.0-2.el9.alma.1.x86_64.rpm,9a2bdc6807b22ffb6294b30dcc13e20198abe5facaeefcf1bfacec3e1013e3c2
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libgpg-error-1.42-5.el9.i686.rpm,23f828674bc30752687452dacf4a25ca8c25ef0d25e3e6953d0714b5b33cef08
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libgpg-error-1.42-5.el9.x86_64.rpm,f28fabeb02add41b8156c394427da9dbf99882d8c42055367f872130ee424c97
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/libICE-1.0.10-8.el9.i686.rpm,d9c44bb34c990fdcc6987feefacfc9e1e7997ca25f1e0a59c9adc9c16f2266d1
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/libICE-1.0.10-8.el9.x86_64.rpm,d0b867203384942985c3c4ad6fc9ec0000fcc9965408451a6cdef8df06a3833c
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libidn2-2.3.0-7.el9.i686.rpm,4458e2d416b5e6714ec664c4f88791eee7d1e707026a5854d948a6159911465b
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libidn2-2.3.0-7.el9.x86_64.rpm,c3ff43237c05258ee0dd77c28540fc8d0d824f1567b115ddfd555b78d2ee0e7a
-https://cdn.getsol.us/repo/shannon/libj/libjpeg-turbo/libjpeg-turbo-32bit-3.0.3-21-1-x86_64.eopkg,5624990281f6520026aa2252c711f02d4916099aeac63accc83bd2135b9d3220
-https://cdn.getsol.us/repo/shannon/libj/libjpeg-turbo/libjpeg-turbo-3.0.3-21-1-x86_64.eopkg,c92f19d446b4c2450306f13b1a908f2552400da09301996fe349fee243d964d2
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/libogg-1.3.4-6.el9.i686.rpm,9e4ff6fa0251c5357ae8f648b3e396d0043d5b3f2fb28d23325f5bd592c67aee
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/libogg-1.3.4-6.el9.x86_64.rpm,83b4b333e335654674032d1cda4d7deb328b5a409f95e3952c3dc482b684fd8f
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libpng-1.6.37-12.el9.i686.rpm,66bf967e84f0d8e7a0059f2b7dda33482ff5e4fe088d0e3bc86dc54e6c7c97be
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libpng-1.6.37-12.el9.x86_64.rpm,ebb3e927dfafa0b199e00e2bef4c10c85fc82a70c359faf9aeb599eda70bfa5e
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libbrotli-1.0.9-7.el9_5.i686.rpm,d18c20fccbb9497222059d475161a55c0c6279344f214fe551b8b0f87d3c054a
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libbrotli-1.0.9-7.el9_5.x86_64.rpm,77906ae44bcd42a9f6646d814874768e8befaabf413133d44f71a95ac6c6d68c
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/libsigc%2B%2B20-2.10.7-2.el9.i686.rpm,7c33f4c18cbd51dc0a7c9ae0190b1a27d0540ce2fd92ced4776dde66162d1a88
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/libsigc%2B%2B20-2.10.7-2.el9.x86_64.rpm,8600132b9e4efadfe23d3125e90287f5a7114165fd9ecb064d9930fcca446ef2
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/libSM-1.2.3-10.el9.i686.rpm,ec68b3def828ba442248f3cff495f42c4af3dd58f38c8daad07dfddc109d999a
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/libSM-1.2.3-10.el9.x86_64.rpm,d00318843a50138829b470a97b8db1a00723b10bca23c16a8386d9ab670e4c64
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libstdc%2B%2B-11.5.0-5.el9_5.alma.1.i686.rpm,05fd29569c312a661374aa69e28ed5cad20812822070a3138265c92a446e2c80
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libstdc%2B%2B-11.5.0-5.el9_5.alma.1.x86_64.rpm,7efb1ded0ec2763589dd3bc6a9d52fa9f898cc1f3db6e8761e1d223ce4506424
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libtasn1-4.16.0-8.el9_1.i686.rpm,9cbd11277e89bd0d4a80900bee01ca71b0c582c9cb5744f5035135b2ed907c85
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libtasn1-4.16.0-8.el9_1.x86_64.rpm,2dd3a1ad2785df642cb03dfed6dfbeec929ff8b14e52e61272ca7e89f86ba3cd
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/libtheora-1.1.1-31.el9.i686.rpm,2155c688b516ab359d0c47be6d73e51d3c65f199ed5591f7bb7ce6d5d9799977
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/libtheora-1.1.1-31.el9.x86_64.rpm,25ab4a8ad997b450b4a34a777f5a0678eea72ca1665944a7a461ac63de4b25bf
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libunistring-0.9.10-15.el9.i686.rpm,792600a879fbd94ce4bf249713b0cb3a166c147cfad397e8a15e6b4f01c75997
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libunistring-0.9.10-15.el9.x86_64.rpm,d2f86fbee729a4900a433a6c52a3cb95d9b52b1d79c9e719ee3ef509813eb6d9
+https://archive.debian.org/debian/pool/main/o/openssl/libssl1.0.0_1.0.2l-1~bpo8+1_amd64.deb,6e85968afe1a6643f4e82e1cc168f848c5967a131bb1c7392f11974953db0e67
+https://archive.debian.org/debian/pool/main/o/openssl/libssl1.0.0_1.0.2l-1~bpo8+1_i386.deb,4e8b924286a064c257ca50b8ecd44fe71226443daef62b6836010590e2157329
+https://archives.fedoraproject.org/pub/archive/epel/9.5/Everything/x86_64/Packages/m/mbedtls-2.28.8-1.el9.x86_64.rpm,54cb7e43e902a7378e912ecfec470ee30fe69af323d3812cbc39ddebd64790ff
+https://archives.fedoraproject.org/pub/archive/fedora/linux/releases/34/Everything/x86_64/os/Packages/l/libpng12-1.2.57-13.fc34.i686.rpm,5e4f525f44d538e3467dd13c66010d54f3d12ec4a319829e15a23ff2d0195b5b
+https://archives.fedoraproject.org/pub/archive/fedora/linux/releases/34/Everything/x86_64/os/Packages/l/libpng12-1.2.57-13.fc34.x86_64.rpm,49f956ad143399c29f5ddf6da6b6fc108ce130d1ed598b4ec5a68b1ff26091de
 https://archives.fedoraproject.org/pub/archive/fedora/linux/releases/34/Everything/x86_64/os/Packages/l/libunwind-1.4.0-5.fc34.i686.rpm,eb518e9d073c36fd29862c7afd00a8a7cb186074a0e162983f18fb6c9dc71d2c
 https://archives.fedoraproject.org/pub/archive/fedora/linux/releases/34/Everything/x86_64/os/Packages/l/libunwind-1.4.0-5.fc34.x86_64.rpm,0f44e37e4eb472508e0dba8505ebbb04c2c1390c6a43970a6af80aad886037e3
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/libvorbis-1.3.7-5.el9.i686.rpm,e38c5185fea611330ea253db020315dd564507bf95c48a81dbe7db186ce4214b
-https://repo.almalinux.org/almalinux/9.5/AppStream/x86_64/os/Packages/libvorbis-1.3.7-5.el9.x86_64.rpm,3c340337a073975636d198a282503f32a54bf85fb5345522a2c22e530b8d6c9e
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libzstd-1.5.1-2.el9.i686.rpm,e98f253d6f5f9acf2ded628cd98f50a54a031e354bf7bb2efc3fe39f9899fe49
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/libzstd-1.5.1-2.el9.x86_64.rpm,2e5215e225cec2db43d92c3b3380812a6f8684e881ed415670299e1e0fe7d1ce
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/lz4-libs-1.9.3-5.el9.i686.rpm,c308ba7380a3b5c5337c919150f0234342b64c1ea696d6207c2b790b82ba4669
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/lz4-libs-1.9.3-5.el9.x86_64.rpm,439e701491af6a957ac4d7dc69cccee1a945333f19d8bfcbd896b5119af01aab
+https://archives.fedoraproject.org/pub/archive/fedora/linux/updates/34/Everything/x86_64/Packages/h/hidapi-0.11.2-2.fc34.i686.rpm,386040fc502e2f02b4b072e0cd90c034dc0a30831fbd2ddb116d17c080b6f9a4
+https://archives.fedoraproject.org/pub/archive/fedora/linux/updates/34/Everything/x86_64/Packages/h/hidapi-0.11.2-2.fc34.x86_64.rpm,2a13461c7d5fcf72ffd682b3eb04833aea8306d61c3b8bdeb42a5081d96d5478
+https://archives.fedoraproject.org/pub/archive/fedora/linux/updates/40/Everything/x86_64/Packages/s/SDL3-3.2.12-1.fc40.i686.rpm,4a3bce6e6358ddd05ed2941e7b8fcc4c8cc02df20741d3f0f9295facbc687769
+https://archives.fedoraproject.org/pub/archive/fedora/linux/updates/40/Everything/x86_64/Packages/s/SDL3-3.2.12-1.fc40.x86_64.rpm,9524803ef42d80dc1b15190d5bb5925200995d5a6152eca04471383ff891af17
+https://cdn.getsol.us/repo/shannon/libj/libjpeg-turbo/libjpeg-turbo-3.0.3-21-1-x86_64.eopkg,c92f19d446b4c2450306f13b1a908f2552400da09301996fe349fee243d964d2
+https://cdn.getsol.us/repo/shannon/libj/libjpeg-turbo/libjpeg-turbo-32bit-3.0.3-21-1-x86_64.eopkg,5624990281f6520026aa2252c711f02d4916099aeac63accc83bd2135b9d3220
 https://repo.almalinux.org/almalinux/8.10/BaseOS/x86_64/os/Packages/nettle-3.4.1-7.el8.i686.rpm,13bc8758355c3ec885f414e0e86efad29a7fe82c0999cdb3754d666949c06cc5
 https://repo.almalinux.org/almalinux/8.10/BaseOS/x86_64/os/Packages/nettle-3.4.1-7.el8.x86_64.rpm,fb4aaa22b2e2f2fc0c451b87ac464e8746d240397d64e79f6bfd38f4e9128a5a
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/nettle-3.9.1-1.el9.i686.rpm,e5610fe3b25cfca1616440e1a8a7489fd37b0852f182fdda87ab8e196d981970
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/nettle-3.9.1-1.el9.x86_64.rpm,9786b6610d17aa495fb640ba6d046bf73704915b82d12e6d8339090ff4891de5
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/p11-kit-0.25.3-3.el9_5.i686.rpm,abe4ab924b405c236eaf3fa067fed50ac66d15d0ef42c782a745788c1a9c5b91
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/p11-kit-0.25.3-3.el9_5.x86_64.rpm,0e1bd97e697a9fa3b4e508b945cf352b73b06e4e8218c14ac5fb912b0022c646
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/pam-1.5.1-22.el9_5.i686.rpm,f2ae4441fc667cdcb930e7a2cfe90f7342e7d1c8848d6d18cb350ba68e025862
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/pam-1.5.1-22.el9_5.x86_64.rpm,15bfe177d3c10982c75e81d7b274883a9fb53322011c8d829667c983ee419b18
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/pciutils-libs-3.7.0-5.el9.i686.rpm,2cf468da5ecf307cc325c9a5da0a288aafd25dec0951af2b4d0954a77bdf900d
-https://repo.almalinux.org/almalinux/9.5/BaseOS/x86_64/os/Packages/pciutils-libs-3.7.0-5.el9.x86_64.rpm,308bd897d8f9a36020d70edc79089a3595b22b149200d67c37e2d4e6147a9f75
-https://dl.fedoraproject.org/pub/fedora/linux/updates/40/Everything/x86_64/Packages/s/SDL3-3.2.4-1.fc40.i686.rpm,f4776f6f56a8d3e62e95c06fb885832e22116323c0c2e0f57f918cd4b92e16a2
-https://dl.fedoraproject.org/pub/fedora/linux/updates/40/Everything/x86_64/Packages/s/SDL3-3.2.4-1.fc40.x86_64.rpm,96f52b7cf293ce335e2bf726c8b6aefd82ed76fbe9d9f0db1fe603780477c2d0
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/compat-openssl11-1.1.1k-4.el9_0.i686.rpm,6b9d8e1b34dcc5a5cb1196d746022b04677b7f9a105508392446508c6ae9b4cc
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/compat-openssl11-1.1.1k-4.el9_0.x86_64.rpm,102eb4355d2dfc8093ed59dd2ebc6bf8616965524faf7e886212fd542f4c1b4b
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/fribidi-1.0.10-6.el9.2.i686.rpm,75f808f0f5f93aceb4e477aa5f302edbed07a605a0644f0ea929d4ea609e36a0
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/fribidi-1.0.10-6.el9.2.x86_64.rpm,b4763ca410f8479b2775efb55ba09e6722948af7526180a56309e3e090c27010
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/libepoxy-1.5.5-4.el9.i686.rpm,fe35e8de9b68c8d7e7b6b525580fc36cf9c86412cdb2b022dc01f95801771981
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/libepoxy-1.5.5-4.el9.x86_64.rpm,b6d1e7f1c832882ceb05187cbc2d8fe1f459bcd88c7160487ee6f734e1a5d017
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/libICE-1.0.10-8.el9.i686.rpm,d9c44bb34c990fdcc6987feefacfc9e1e7997ca25f1e0a59c9adc9c16f2266d1
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/libICE-1.0.10-8.el9.x86_64.rpm,d0b867203384942985c3c4ad6fc9ec0000fcc9965408451a6cdef8df06a3833c
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/libogg-1.3.4-6.el9.i686.rpm,9e4ff6fa0251c5357ae8f648b3e396d0043d5b3f2fb28d23325f5bd592c67aee
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/libogg-1.3.4-6.el9.x86_64.rpm,83b4b333e335654674032d1cda4d7deb328b5a409f95e3952c3dc482b684fd8f
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/libsigc%2B%2B20-2.10.7-2.el9.i686.rpm,7c33f4c18cbd51dc0a7c9ae0190b1a27d0540ce2fd92ced4776dde66162d1a88
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/libsigc%2B%2B20-2.10.7-2.el9.x86_64.rpm,8600132b9e4efadfe23d3125e90287f5a7114165fd9ecb064d9930fcca446ef2
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/libSM-1.2.3-10.el9.i686.rpm,ec68b3def828ba442248f3cff495f42c4af3dd58f38c8daad07dfddc109d999a
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/libSM-1.2.3-10.el9.x86_64.rpm,d00318843a50138829b470a97b8db1a00723b10bca23c16a8386d9ab670e4c64
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/libtheora-1.1.1-31.el9.i686.rpm,2155c688b516ab359d0c47be6d73e51d3c65f199ed5591f7bb7ce6d5d9799977
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/libtheora-1.1.1-31.el9.x86_64.rpm,25ab4a8ad997b450b4a34a777f5a0678eea72ca1665944a7a461ac63de4b25bf
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/libvorbis-1.3.7-5.el9.i686.rpm,e38c5185fea611330ea253db020315dd564507bf95c48a81dbe7db186ce4214b
+https://vault.almalinux.org/9.5/AppStream/x86_64/os/Packages/libvorbis-1.3.7-5.el9.x86_64.rpm,3c340337a073975636d198a282503f32a54bf85fb5345522a2c22e530b8d6c9e
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/bluez-libs-5.72-2.el9.i686.rpm,af1aba8e6fa5ccf33032d58a3acc59368c07d532293396db205fb9a4bc1241ae
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/bluez-libs-5.72-2.el9.x86_64.rpm,49c273813b4616240acdae8dae893192658c7541482f9bbb2c46dfe19b61877d
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/glibc-2.34-125.el9_5.8.alma.1.i686.rpm,28bc1088e39483d145849e8804c5a0f2524faeb4f76631ae46c41100f40e08b9
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/glibc-2.34-125.el9_5.8.alma.1.x86_64.rpm,1848b9ced2dfec6c206791b23a5a8a6356939c5ed0492fc91c86bf1d3a255c1d
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/gmp-6.2.0-13.el9.i686.rpm,fe352f7d197ebdf2c92bbd6e548f9f6b7ac9e69e7a9d6c9861a6274482e79d73
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/gmp-6.2.0-13.el9.x86_64.rpm,babe0b1385e723476a216d5f4b72453e031c3d683b19344786c2bddf7203cb04
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/gnutls-3.8.3-4.el9_4.i686.rpm,64cf0b0136d3f9b80b07ce2d4b8ce6a4ab86617e8ef9baf560bdfa5edef825ea
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/gnutls-3.8.3-4.el9_4.x86_64.rpm,16c48c504fe62f8e19cac7c6442d3e08bba487f610c2f533116dac32357cda86
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/harfbuzz-2.7.4-10.el9.i686.rpm,5157daf2ab0a03b403d37c60b150b8f0f866754714f81e3d96368e06a16339ce
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/harfbuzz-2.7.4-10.el9.x86_64.rpm,1f81073019abe4176d4496723a89b55a349c31f507e96397a0b3efa7cea0ff61
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libblkid-2.37.4-20.el9.i686.rpm,f9e62d2768c31948b53268b7de51bd1c7342cdf39a328a0d89d933b4aa0c5dd1
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libblkid-2.37.4-20.el9.x86_64.rpm,ed02d978a36488b38bad62af21d4045367c1f0b00d00c90ded01e2fc8e190536
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libbrotli-1.0.9-7.el9_5.i686.rpm,d18c20fccbb9497222059d475161a55c0c6279344f214fe551b8b0f87d3c054a
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libbrotli-1.0.9-7.el9_5.x86_64.rpm,77906ae44bcd42a9f6646d814874768e8befaabf413133d44f71a95ac6c6d68c
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libffi-3.4.2-8.el9.i686.rpm,70daa216ed2a6540e41c455efc95d2b5671db164137f854bceae07e348f58645
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libffi-3.4.2-8.el9.x86_64.rpm,1cf2d122e5c423681394c1bc599ee2281d2eafff5a3749842bc93c37db4bed5e
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libgcc-11.5.0-2.el9.alma.1.i686.rpm,fed64dfadf74bd132ac2675f625ec601ca9d7ae9d05b8d7ff492a204260d0174
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libgcc-11.5.0-5.el9_5.alma.1.x86_64.rpm,7bd3ec130e1712442121b61b27164386006b9bfa6f0fa188316ecb35575a768f
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libgomp-11.5.0-2.el9.alma.1.i686.rpm,686ac26afe2073a3607f26fd1592a8ccd9b78f6eec6e872528634b43c4e04dd7
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libgomp-11.5.0-2.el9.alma.1.x86_64.rpm,9a2bdc6807b22ffb6294b30dcc13e20198abe5facaeefcf1bfacec3e1013e3c2
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libgpg-error-1.42-5.el9.i686.rpm,23f828674bc30752687452dacf4a25ca8c25ef0d25e3e6953d0714b5b33cef08
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libgpg-error-1.42-5.el9.x86_64.rpm,f28fabeb02add41b8156c394427da9dbf99882d8c42055367f872130ee424c97
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libidn2-2.3.0-7.el9.i686.rpm,4458e2d416b5e6714ec664c4f88791eee7d1e707026a5854d948a6159911465b
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libidn2-2.3.0-7.el9.x86_64.rpm,c3ff43237c05258ee0dd77c28540fc8d0d824f1567b115ddfd555b78d2ee0e7a
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libnsl-2.34-125.el9_5.8.alma.1.i686.rpm,7f4d1552b6036c49028db959f82fc0bc30773cd7ed0a62c3906da0e2388f4dc5
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libnsl-2.34-125.el9_5.8.alma.1.x86_64.rpm,7d0e521ab087807ee71d79da66a2829b827bf5080bb5d080dfab108e31cf218b
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libpng-1.6.37-12.el9.i686.rpm,66bf967e84f0d8e7a0059f2b7dda33482ff5e4fe088d0e3bc86dc54e6c7c97be
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libpng-1.6.37-12.el9.x86_64.rpm,ebb3e927dfafa0b199e00e2bef4c10c85fc82a70c359faf9aeb599eda70bfa5e
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libstdc++-11.5.0-5.el9_5.alma.1.i686.rpm,05fd29569c312a661374aa69e28ed5cad20812822070a3138265c92a446e2c80
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libstdc++-11.5.0-5.el9_5.alma.1.x86_64.rpm,7efb1ded0ec2763589dd3bc6a9d52fa9f898cc1f3db6e8761e1d223ce4506424
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libtasn1-4.16.0-8.el9_1.i686.rpm,9cbd11277e89bd0d4a80900bee01ca71b0c582c9cb5744f5035135b2ed907c85
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libtasn1-4.16.0-8.el9_1.x86_64.rpm,2dd3a1ad2785df642cb03dfed6dfbeec929ff8b14e52e61272ca7e89f86ba3cd
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libunistring-0.9.10-15.el9.i686.rpm,792600a879fbd94ce4bf249713b0cb3a166c147cfad397e8a15e6b4f01c75997
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libunistring-0.9.10-15.el9.x86_64.rpm,d2f86fbee729a4900a433a6c52a3cb95d9b52b1d79c9e719ee3ef509813eb6d9
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libzstd-1.5.1-2.el9.i686.rpm,e98f253d6f5f9acf2ded628cd98f50a54a031e354bf7bb2efc3fe39f9899fe49
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/libzstd-1.5.1-2.el9.x86_64.rpm,2e5215e225cec2db43d92c3b3380812a6f8684e881ed415670299e1e0fe7d1ce
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/lz4-libs-1.9.3-5.el9.i686.rpm,c308ba7380a3b5c5337c919150f0234342b64c1ea696d6207c2b790b82ba4669
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/lz4-libs-1.9.3-5.el9.x86_64.rpm,439e701491af6a957ac4d7dc69cccee1a945333f19d8bfcbd896b5119af01aab
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/ncurses-libs-6.2-10.20210508.el9.i686.rpm,285db6dc6f610f75ef6863dcb0a383eae394c2b33bff283189cb49ac7fae4c58
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/ncurses-libs-6.2-10.20210508.el9.x86_64.rpm,d06b0e491481a345faf21a69666cbdc302d19b54fb29ae1aa78fa6b205f9e604
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/nettle-3.9.1-1.el9.i686.rpm,e5610fe3b25cfca1616440e1a8a7489fd37b0852f182fdda87ab8e196d981970
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/nettle-3.9.1-1.el9.x86_64.rpm,9786b6610d17aa495fb640ba6d046bf73704915b82d12e6d8339090ff4891de5
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/p11-kit-0.25.3-3.el9_5.i686.rpm,abe4ab924b405c236eaf3fa067fed50ac66d15d0ef42c782a745788c1a9c5b91
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/p11-kit-0.25.3-3.el9_5.x86_64.rpm,0e1bd97e697a9fa3b4e508b945cf352b73b06e4e8218c14ac5fb912b0022c646
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/pam-1.5.1-22.el9_5.i686.rpm,f2ae4441fc667cdcb930e7a2cfe90f7342e7d1c8848d6d18cb350ba68e025862
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/pam-1.5.1-22.el9_5.x86_64.rpm,15bfe177d3c10982c75e81d7b274883a9fb53322011c8d829667c983ee419b18
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/pciutils-libs-3.7.0-5.el9.i686.rpm,2cf468da5ecf307cc325c9a5da0a288aafd25dec0951af2b4d0954a77bdf900d
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/pciutils-libs-3.7.0-5.el9.x86_64.rpm,308bd897d8f9a36020d70edc79089a3595b22b149200d67c37e2d4e6147a9f75
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/sqlite-libs-3.34.1-7.el9_3.i686.rpm,dc6946a8344b8f4f476fef9479d1786c86fdcd045e988ab5d85fef7780e1fae4
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/sqlite-libs-3.34.1-7.el9_3.x86_64.rpm,e7f87444f24a03b37caa02942082b6d4283067ecbbd3f7798f882d22b55713cf
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/zlib-1.2.11-40.el9.i686.rpm,659ee5222028744806be3f822369a726c721aa346eed2882dcf019d83c111e67
+https://vault.almalinux.org/9.5/BaseOS/x86_64/os/Packages/zlib-1.2.11-40.el9.x86_64.rpm,90d7d5d145c6d397645ffefdff40e3d2783e3bf3bfa7b6c1f089c2cad81a3191
+https://vault.centos.org/6.10/os/x86_64/Packages/compat-libstdc++-33-3.2.3-69.el6.i686.rpm,894abcc5f0b28e39c71d4b0ca27d19540e7eb9d3c32664d5621324c9efbfe11e
+https://vault.centos.org/6.10/os/x86_64/Packages/compat-libstdc++-33-3.2.3-69.el6.x86_64.rpm,41ea0fec16c0ca62722a32b7ffd70bb716beec8b6dd32b57f076be02a5aee00c
@@ -1,96 +1,134 @@
-#!/bin/bash
+#!/bin/sh

-set -x
+set -ex

-extract_pkg_auto() {
-    # Example x86_64 package name: libcurl3-gnutls_7.74.0-1.3+deb11u7~bpo11+1_amd64.deb
-    # Example x86 package name: libcurl3-gnutls_7.74.0-1.3+deb11u7~bpo11+1_i386.deb
-    if echo "${1}" | grep -q -P ".deb$"; then
-        echo "DEB (Debian) package detected."
-        ar x "$(basename "${1}")"
-        tar --extract --file data.tar*
-    # Example x86_64 package name: curl-gnutls-8.1.2-44-1-x86_64.eopkg
-    # Example x86 package name: curl-gnutls-32bit-8.1.2-44-1-x86_64.eopkg
-    elif echo "${1}" | grep -q -P ".eopkg$"; then
-        echo "EOPKG (Solus Linux) package detected."
-        unzip -o "$(basename "${1}")"
-        tar --extract --file install.tar.xz
-    # Example x86_64 package name: bluez-libs-5.63-1.el8.x86_64.rpm
-    # Example x86 package name: bluez-libs-5.63-1.el8.i686.rpm
-    elif echo "${1}" | grep -q -P ".rpm$"; then
-        echo "RPM (Fedora) detected."
-        rpm2cpio "$(basename "${1}")" | cpio -idmv
-    # Example x86_64 package name: libjpeg-turbo-3.0.1_1.x86_64.xbps
-    # Example x86 package name: libjpeg-turbo-32bit-3.0.1_1.x86_64.xbps
-    elif echo "${1}" | grep -q -P ".xbps$"; then
-        echo "XBPS (Void Linux) package detected."
-        tar --extract --file "$(basename "${1}")"
-    else
-        echo "Unsupported package type detected."
-        exit 1
-    fi
+trap cleanup_exit INT TERM EXIT
+
+cleanup_exit()
+{
+    echo "Running cleanup_exit..."
+    rm -rf /tmp/box64-bundle.*
 }

-box64_dir=$(pwd)
-dir_tmp_local="$(mktemp --directory)"
-cd "${dir_tmp_local}"
+# we must have few tools
+awk=$(which awk) || { echo "E: You must have awk" && exit 1; }
+cpio=$(which cpio) || { echo "E: You must have cpio" && exit 1; }
+curl=$(which curl) || { echo "E: You must have curl" && exit 1; }
+curl_cmd="${curl} --connect-timeout 5 --retry 5 --retry-delay 1 --create-dirs -fsSLo"
+rpm2cpio=$(which rpm2cpio) || { echo "E: You must have rpm2cpio" && exit 1; }
+sha256sum=$(which sha256sum) || { echo "E: You must have sha256sum" && exit 1; }
+unzip=$(which unzip) || { echo "E: You must have unzip" && exit 1; }

-for line in $(cat "${box64_dir}/box64-bundle-x86-libs.csv");
-    do pkg_url="$(echo "${line}" | cut -d, -f1)"
-    pkg_checksum_expected="$(echo "${line}" | cut -d, -f2)"
-    pkg_name="$(basename "${pkg_url}")"
-    curl --silent --location --remote-name "${pkg_url}"
-    pkg_checksum_actual="$(sha256sum ${pkg_name} | awk '{print $1}')"
-    if [[ "${pkg_checksum_expected}" -ne "${pkg_checksum_actual}" ]];
-        then echo "Invalid checksum for ${pkg_name}"
-        echo "Expected: ${pkg_checksum_expected}"
-        echo "Actual: ${pkg_checksum_actual}"
+current_dir=$(pwd)
+dir_tmp="$(mktemp -d /tmp/box64-bundle.XXXXXX)"
+
+# download the packages
+while IFS= read -r line; do
+    # shellcheck disable=SC2016
+    pkg_url=$(echo "${line}" | ${awk} -F',' '{print$(NF-1)}')
+    # shellcheck disable=SC2016
+    pkg_checksum=$(echo "${line}" | ${awk} -F',' '{print$(NF)}')
+    # shellcheck disable=SC2016
+    pkg_name=$(echo "${pkg_url}" | ${awk} -F'/' '{print$(NF)}')
+    ${curl_cmd} "${dir_tmp}/bundle-pkgs/${pkg_name}" "${pkg_url}"
+    checksum="$(${sha256sum} "${dir_tmp}/bundle-pkgs/${pkg_name}" | awk '{print $1}')"
+    if [ "${pkg_checksum}" != "${checksum}" ]; then
+        echo "E: Invalid checksum for ${pkg_name}"
+        echo "Expected: ${pkg_checksum}"
+        echo "Computed: ${checksum}"
        exit 1
    fi
-    extract_pkg_auto "${pkg_name}"
+done < box64-bundle-x86-libs.csv
+
+# generate the bundle packages archive
+tar -C "${dir_tmp}"/bundle-pkgs -czvf "${current_dir}"/box64-bundle-x86-pkgs.tar.gz .
+
+# extract the packages
+cd "${dir_tmp}"/bundle-pkgs
+for file in *.deb *.eopkg *.rpm *.xbps; do
+    # handle the case of no files, e.g. xbps
+    [ -e "${dir_tmp}"/bundle-pkgs/"${file}" ] || break
+    # shellcheck disable=SC2016
+    extension=$(echo "${file}" | ${awk} -F'.' '{print$(NF)}')
+    mkdir -p "${dir_tmp}"/bundle-libs
+    cd "${dir_tmp}"/bundle-libs
+    case "${extension}" in
+        deb)
+            echo "I: DEB (Debian) package detected (${file})"
+            ar x "${dir_tmp}"/bundle-pkgs/"${file}"
+            tar -xf data.tar*
+            ;;
+        eopkg)
+            echo "I: EOPKG (Solus Linux) package detected (${file})"
+            ${unzip} -o "${dir_tmp}"/bundle-pkgs/"${file}"
+            tar -xf install.tar.xz
+            ;;
+        rpm)
+            echo "I: RPM (Fedora) detected (${file})"
+            ${rpm2cpio} "${dir_tmp}"/bundle-pkgs/"${file}" | ${cpio} -idmv
+            ;;
+        xbps)
+            echo "I: XBPS (Void Linux) package detected (${file})"
+            tar -xf "${dir_tmp}"/bundle-pkgs/"${file}"
+            ;;
+    esac
 done

-# Start preperation of bundles.
-mkdir -p \
-  "${dir_tmp_local}/bundle-libs/usr/lib/box64-i386-linux-gnu" \
-  "${dir_tmp_local}/bundle-libs/usr/lib/box64-x86_64-linux-gnu" \
-  "${dir_tmp_local}/bundle-pkgs" \
+# reorganize the files
+cd "${dir_tmp}"/bundle-libs
+mkdir -p "${dir_tmp}"/bundle-libs/box64-i386-linux-gnu
+mv lib/*.so* usr/lib/*.so* usr/lib/i386-linux-gnu/*.so* usr/lib32/*.so* \
+    "${dir_tmp}"/bundle-libs/box64-i386-linux-gnu
+mkdir -p "${dir_tmp}"/bundle-libs/box64-x86_64-linux-gnu
+mv lib64/*.so* usr/lib/x86_64-linux-gnu/*.so* usr/lib64/*.so* \
+    "${dir_tmp}"/bundle-libs/box64-x86_64-linux-gnu
+rm -f ./*.tar.* debian-binary files.xml metadata.xml
+rm -rf ./lib* etc run sbin usr var
+mkdir -p usr/lib
+mv box64-*-linux-gnu usr/lib

-# Copy Debian family libraries.
-cp --archive "${dir_tmp_local}"/usr/lib/i386-linux-gnu/*.so* "${dir_tmp_local}/bundle-libs/usr/lib/box64-i386-linux-gnu"
-cp --archive "${dir_tmp_local}"/usr/lib/x86_64-linux-gnu/*.so* "${dir_tmp_local}/bundle-libs/usr/lib/box64-x86_64-linux-gnu"
-# Copy Fedora family libraries.
-cp --archive "${dir_tmp_local}"/usr/lib/*.so* "${dir_tmp_local}/bundle-libs/usr/lib/box64-i386-linux-gnu"
-cp --archive "${dir_tmp_local}"/usr/lib64/*.so* "${dir_tmp_local}/bundle-libs/usr/lib/box64-x86_64-linux-gnu"
-## This location is used by a few special packages such as "glibc".
-cp --archive "${dir_tmp_local}"/lib/*.so* "${dir_tmp_local}/bundle-libs/usr/lib/box64-i386-linux-gnu"
-cp --archive "${dir_tmp_local}"/lib64/*.so* "${dir_tmp_local}/bundle-libs/usr/lib/box64-x86_64-linux-gnu"
-# Copy Solus and Void Linux family libraries.
-cp --archive "${dir_tmp_local}"/usr/lib32/*.so* "${dir_tmp_local}/bundle-libs/usr/lib/box64-i386-linux-gnu"
+# remove libraries that cannot be emulated
+# https://github.com/ptitSeb/box64/blob/v0.3.8/src/librarian/library.c#L433
+set -- libc libpthread librt libGL libX11 libasound libdl libm libbsd libutil \
+  libresolv libXxf86vm libXinerama libXrandr libXext libXfixes libXcursor \
+  libXrender libXft libXi libXss libXpm libXau libXdmcp libX11-xcb libxcb \
+  libxcb-xfixes libxcb-shape libxcb-shm libxcb-randr libxcb-image \
+  libxcb-keysyms libxcb-xtest libxcb-glx libxcb-dri2 libxcb-dri3 libXtst libXt \
+  libXcomposite libXdamage libXmu libxkbcommon libxkbcommon-x11 \
+  libpulse-simple libpulse libvulkan ld-linux-x86-64 crashhandler \
+  libtcmalloc_minimal libanl ld-linux libthread_db
+for file in "$@"; do
+    rm -f "${dir_tmp}"/bundle-libs/usr/lib/box64-*-linux-gnu/"${file}".so*
+done

-# Create "<LIBRARY_FILE>.so" symlinks.
-for dir_lib in \
-  "${dir_tmp_local}/bundle-libs/usr/lib/box64-i386-linux-gnu" \
-  "${dir_tmp_local}/bundle-libs/usr/lib/box64-x86_64-linux-gnu"
-    # If two or more libraries of the same name exist, this will symlink the oldest version.
-    # Pipe this into a reverse sort `| sort -r` to symlink the newest version.
-    do for lib in $(ls -1 "${dir_lib}")
-        # Example output: libbluetooth.so
-        do lib_base="$(echo "${lib}" | cut -d. -f1,2)"
-        if ! ls "${dir_lib}/${lib_base}" 2> /dev/null
-            then ln -s "${lib}" "${dir_lib}/${lib_base}"
+# check broken symlinks
+if find "${dir_tmp}"/bundle-libs -type l ! -exec test -e {} \; -print | grep bundle-libs; then
+    echo "E: Broken symlinks found"
+    exit 1
+fi
+
+# Manually create additional symlinks that are useful to have.
+cd "${dir_tmp}"/bundle-libs/usr/lib/box64-x86_64-linux-gnu/
+ln -s libmbedcrypto.so.7 libmbedcrypto.so.3
+ln -s libmbedtls.so.14 libmbedtls.so.12
+ln -s libmbedx509.so.1 libmbedx509.so.0
+
+# create shared libraries (.so) symlinks
+# if multiple libraries of the same name exist, it will symlink the oldest version
+# e.g. libcurl-gnutls soname with versioned symbols (CURL_GNUTLS_3, CURL_GNUTLS_4)
+# libcurl-gnutls.so -> libcurl-gnutls.so.3
+# libcurl-gnutls.so.3 -> libcurl-gnutls.so.4
+# libcurl-gnutls.so.4 -> libcurl-gnutls.so.4.7.0
+# libcurl-gnutls.so.4.7.0
+for dir_lib in "${dir_tmp}"/bundle-libs/usr/lib/box64-*-linux-gnu; do
+    cd "${dir_lib}"
+    for lib in lib*.so*; do
+        lib_base="$(echo "${lib}" | awk -F'.' '{print $1"."$2}')"
+        if ! ls "${dir_lib}/${lib_base}" 2> /dev/null; then
+            ln -s "${lib}" "${lib_base}"
        fi
    done
 done

-mv "${dir_tmp_local}"/*.deb "${dir_tmp_local}/bundle-pkgs/"
-mv "${dir_tmp_local}"/*.eopkg "${dir_tmp_local}/bundle-pkgs/"
-mv "${dir_tmp_local}"/*.rpm "${dir_tmp_local}/bundle-pkgs/"
-mv "${dir_tmp_local}"/*.xbps "${dir_tmp_local}/bundle-pkgs/"
-
-tar --directory "${dir_tmp_local}/bundle-libs" --create --file "${box64_dir}/box64-bundle-x86-libs.tar.gz" .
-tar --directory "${dir_tmp_local}/bundle-pkgs" --create --file "${box64_dir}/box64-bundle-x86-pkgs.tar.gz" .
-
-# Clean up.
-cd -
-rm -r -f "${dir_tmp_local}"
+# generate the bundle libraries archive
+tar -C "${dir_tmp}"/bundle-libs -czvf "${current_dir}"/box64-bundle-x86-libs.tar.gz .
@@ -7,15 +7,15 @@ string(REGEX REPLACE "\n" ";" files "${files}")
 foreach(file ${files})
  message(STATUS "Uninstalling $ENV{DESTDIR}${file}")
  if(IS_SYMLINK "$ENV{DESTDIR}${file}" OR EXISTS "$ENV{DESTDIR}${file}")
-    exec_program(
-      "@CMAKE_COMMAND@" ARGS "-E remove \"$ENV{DESTDIR}${file}\""
+    execute_process(
+      COMMAND @CMAKE_COMMAND@ -E remove "$ENV{DESTDIR}${file}"
+      RESULT_VARIABLE rm_retval
      OUTPUT_VARIABLE rm_out
-      RETURN_VALUE rm_retval
+      ERROR_VARIABLE rm_out
+      OUTPUT_STRIP_TRAILING_WHITESPACE
      )
    if(NOT "${rm_retval}" STREQUAL 0)
      message(FATAL_ERROR "Problem when removing $ENV{DESTDIR}${file}")
-    endif(NOT "${rm_retval}" STREQUAL 0)
-  else(IS_SYMLINK "$ENV{DESTDIR}${file}" OR EXISTS "$ENV{DESTDIR}${file}")
-    message(STATUS "File $ENV{DESTDIR}${file} does not exist.")
-  endif(IS_SYMLINK "$ENV{DESTDIR}${file}" OR EXISTS "$ENV{DESTDIR}${file}")
+    endif()
+  endif()
 endforeach(file)
@@ -1,3 +1,24 @@
+box64 (0.4.0) unstable; urgency=low
+
+This version added many important refactors: the opcode prefix decoder for Dynarec, libGL and Vulkan, the FSGSBASE support, making segments handling easier and faster.
+    => Improved compatibility accross all architecture, allowing more DRM protected game to run
+    => the prefix refactor helped removed lots of redundant source file, simplifying maintenance
+    => BOX32 is getting more and more stable, even if still not 100% ok. But Steam is more stable than before, runs on all architecture. And some gameoverlay UI works, on OpenGL games. gameoverlay (and fossilize) support for Vulkan is planned for later.
+
+-- Sebastien Chevalier <[email protected]>  Mon, 3 Jan 2026 10:30:33 -0100
+
+box64 (0.3.8) unstable; urgency=low
+
+This version introduce DynaCache, a disk cache for the generated native code. It's not enabled by default for now (but it will use cache file if present by default). It can dramaticaly speedup the load of program (try with linux factorio for example). This version also improve in the volatile metadata handling. There is a massive compatiblity improvment for box32 and on all dynarec in general too. And few more function wrapping...
+
+    => DynaCache for faster loading of program, and avoid runtime hicups
+    => Batter box32 compatibility (much less "Out of Memory" error)
+    => Unity3D games on Wayland are working now
+    => Volatile Metadata fully supported and handled, making games like Starfield to just work.
+    => More work on AVX support for RV64 & LA64. LA64 almost there, while RV64 has Scalar only support (no hardware vector support yet for AVX)
+
+-- Sebastien Chevalier <[email protected]>  Mon, 10 Oct 2025 10:30:33 -0100
+
 box64 (0.3.6) unstable; urgency=low

 This version introduces wowbox64.dll for Hangover; the ability to use the Volatile Metadata of Windows executables for x64 wine; and some better x87/SSE/AVX handling
@@ -1,13 +1,33 @@
 #!/usr/bin/make -f

 %:
-	dh $@  
+	dh $@
+
+DEB_HOST_ARCH := $(shell dpkg-architecture -qDEB_HOST_ARCH 2>/dev/null || echo "unknown")

 override_dh_auto_configure:
+	@echo "Configuring for architecture: $(DEB_HOST_ARCH)"
+ifeq ($(DEB_HOST_ARCH),riscv64)
 	dh_auto_configure -- \
-	      -DCMAKE_BUILD_TYPE=RelWithDebInfo \
-	      -DARM_DYNAREC=1 \
-		  -DNOGIT=1
+		-DCMAKE_BUILD_TYPE=RelWithDebInfo \
+		-DRV64=1 \
+		-DRV64_DYNAREC=1 \
+		-DNOGIT=1
+else ifeq ($(DEB_HOST_ARCH),arm64)
+	dh_auto_configure -- \
+		-DCMAKE_BUILD_TYPE=RelWithDebInfo \
+		-DARM_DYNAREC=1 \
+		-DNOGIT=1
+else ifeq ($(DEB_HOST_ARCH),aarch64)
+	dh_auto_configure -- \
+		-DCMAKE_BUILD_TYPE=RelWithDebInfo \
+		-DARM_DYNAREC=1 \
+		-DNOGIT=1
+else
+	dh_auto_configure -- \
+		-DCMAKE_BUILD_TYPE=RelWithDebInfo \
+		-DNOGIT=1
+endif

 override_dh_shlibdeps:
 	dh_shlibdeps --exclude=libgcc_s.so.1 \
@@ -23,3 +43,7 @@ override_dh_shlibdeps:
 	--exclude=libcrypto.so.1.0.0 \
 	--exclude=libssl.so.1.0.0 \
 	--exclude=box64-bash
+
+override_dh_strip:
+	dh_strip --exclude=/usr/lib/box64-x86_64-linux-gnu/ \
+	--exclude=box64-bash
@@ -1,6 +1,6 @@
 # Bundling x86 Libraries

-The script `box64-bundle-x86-libs.sh` is provided to download, extract, and re-bundle x86_64 and x86 libraries for use with Box64/Box32 and Box86. These libraries improve Box emulation. This bundle provides a much smaller size compared to using a full root file system of an operating system.
+The script `box64-bundle-x86-libs.sh` is provided to download, extract, and re-bundle x86_64 and x86 libraries for use with Box64/Box32 and Box86. This bundle improves Box emulation by providing libraries that are not wrapped yet. In a perfect world, the application or game would already provide these libraries but that is not always the case. The bundle archive provides a much smaller size compared to using a full root file system of an operating system.

 ## Usage

@@ -8,14 +8,19 @@ The script `box64-bundle-x86-libs.sh` is provided to download, extract, and re-b

 Install required dependencies for the bundle script.

+- Arch Linux
+    ```
+    sudo pacman -S -y -u
+    sudo pacman -S binutils coreutils curl rpm-tools tar zstd
+    ```
 - Debian
    ```
    sudo apt-get update
-    sudo apt-get install coreutils curl binutils rpm2cpio tar zstd
+    sudo apt-get install binutils coreutils curl rpm2cpio tar zstd
    ```
 - Fedora
    ```
-    sudo dnf install coreutils curl binutils tar zstd
+    sudo dnf install binutils coreutils curl tar zstd
    ```

 Run `box64-bundle-x86-libs.sh`. This will create two archives:
@@ -25,7 +30,7 @@ Run `box64-bundle-x86-libs.sh`. This will create two archives:
 Extract the library archive. Box will automatically search these paths for library files to emulate.

 ```
-sudo tar --extract --file box64-bundle-x86-libs.tar.gz --directory /
+sudo tar --extract --no-same-owner --file box64-bundle-x86-libs.tar.gz --directory /
 ```

 ### Developer
@@ -45,9 +50,11 @@ Preference of operating system packages that provide a library:
 4. XBPS from Void Linux provide updated variants of obscure 32-bit libraries.
 5. Other (whereever a library is packaged)

-Some essential library files will never be emulated (only wrapped) and should not be bundled. This includes glibc, OpenGL, Vulkan, and X11/Xorg libraries. The full list can be found [here](https://github.com/ptitSeb/box64/blob/v0.3.2/src/librarian/library.c#L427). Other libraries with lots of dependencies, such as GTK, require all dependencies to be installed for emulation to work. This becomes dependency hell and should be avoided.
+Some essential libraries will always be emulated and should be bundled. The full list can be found in the [in the core source code](https://github.com/ptitSeb/box64/blob/v0.4.0/src/core.c#L413).

-For finding package names that contian a specific library file, use the [pkgs.org](https://pkgs.org/) website. Otherwise, use one of these package manager commands on a x86_64 Linux distribution:
+Some essential library files will never be emulated (only natively wrapped instead) and should not be bundled. This includes glibc, OpenGL, Vulkan, and X11/Xorg libraries. The full list can be found [in the library source code](https://github.com/ptitSeb/box64/blob/v0.4.0/src/librarian/library.c#L440). Other libraries with lots of dependencies, such as GTK, require all dependencies to be installed for emulation to work. This becomes dependency hell and should be avoided.
+
+For finding package names that contian a specific library file, use the [pkgs.org website](https://pkgs.org/). Otherwise, use one of these package manager commands on a x86_64 Linux distribution:
 - DEB
    - `apt-file <LIBRARY_FILE>`
 - EOPKG
@@ -70,4 +77,4 @@ Once the package name is found, find the exact URL for downloading it.

 Download the package and then find the checksum with the `sha256sum` command.

-Major Linux distributions such as AlmaLinux and Debian provide archives of packages. Rolling distributions such as Solus and Void Linux do not. In those cases, use the [Internet Archive](https://web.archive.org/) to archive any rolling release package used by Box by using the "Save Page Now" feature. If the error "Save Page Now browser crashed" appears, it can be safely ignored. The archive will still have been saved.
+Major Linux distributions such as AlmaLinux and Debian provide archives of packages. Most rolling distributions such as Solus and Void Linux do not (one exception is Arch Linux which has the [Arch Linux Archive (ALA)](https://wiki.archlinux.org/title/Arch_Linux_Archive)). In those cases, use the [Internet Archive](https://web.archive.org/) to archive any rolling release package used by Box by using the "Save Page Now" feature. If the error "Save Page Now browser crashed" appears, it can be safely ignored. The archive will still have been saved.
@@ -1,3 +1,104 @@
+v0.4.0
+======
+
+Highlights:
+* This version added many important refactors: the opcode prefix decoder for Dynarec, libGL and Vulkan, the FSGSBASE support, making segments handling easier and faster.
+    => Improved compatibility accross all architecture, allowing more DRM protected game to run
+    => the prefix refactor helped removed lots of redundant source file, simplifying maintenance
+    => BOX32 is getting more and more stable, even if still not 100% ok. But Steam is more stable than before, runs on all architecture. And some gameoverlay UI works, on OpenGL games. gameoverlay (and fossilize) support for Vulkan is planned for later.
+
+Version summary:
+
+* Added new BOX64_ARCH to have settings specific to one of the supported cpu architecture (arm64, rv64, la64)
+* Added BAD_PKILL build option, to workaround missing non-mandatory behaviour of pthread_kill (needed on LA64)
+* Improved the handling of invalid opcodes
+* Refactored handling of Selector, Segments and TLS data
+* Added FSGSBASE cpu extension
+* Added support for IFUNC relocation type to ElfLoader (64bits & 32bits)
+* Refactored some internal behaviour to get better Interpreter only build
+* Added BOX64_STEAM_VULKAN to have Vulkan renderer option injected when steamwebhelper is called
+* Introduce BOX64_PYTHON3 to force using an X64 version of python (and use that on arm64 steam)
+* Fixed signal handling when using Wine v10.19+
+* Added some cache to customMalloc / customFree internal functions
+* Some refactor on the cpuid helper function
+* Wrapping: most Wayland listener for wider compatibility
+* Wrapping: refactored libGL & libVulkan wrapping
+* Wrapping: more wrapped functions, reworked some existing one with wrapperhelper
+* Wrapping: Added a few more wrapped libs
+* Wrapping: Better handling of RTLD_NEXT in dlsym function call
+* Wrapping: Reworked wrapping of pthread_one (on Box32 too)
+* Wrapping: a few new syscall implemented
+* Wrapping: Make BOX64_EMULATED_LIBS to also affect gstreamer hacked loading of plugins
+* Wrapping: BOX32: more wrapped function and some fixes to existing ones
+* Wrapping: BOX32: a few new syscall implemented
+* Interpreter: Added LOCK prefix to the generic opcode prefix handling
+* Interpreter: A few fixes to existing opcodes
+* Interpreter: Added F3 0F 40..4F opcodes
+* Dynarec: Refactored HotPages Detection & Handling
+* Dynarec: Introduced DYNAREC_PURGE to allow purging of old (unused?) Dynarec block and save memory
+* Dynarec: Refactored prefix decoding, similar to the Interpreter
+* Dynarec: ARM64: Added detection of loops and preload of xmm/ymm regs before entering
+* Dynarec: ARM64: Refactored the handling of deferred flags
+* Dynarec: ARM64: A few optimisation in some existing opcodes
+* DYnarec: ARM64: Some fixes to existing opcodes
+* Dynarec: LA64: Added x87 opcodes
+* Dynarec: LA64: Added a few missing LOCK prefixed opcodes
+* Dynarec: LA64: Added some more shift/roll opcodes
+* Dynarec: LA64: Added support for native CRC opcodes
+* Dynarec: LA64: Many fixes to some AVX opcodes
+* Dynarec: LA64: Various fixes to some opcodes
+* Dynarec: LA64: Addes a few 66 prefixed opcodes
+* Dynarec: LA64: Added some "simple wrapper" support (allowing direct call of wrapped functions)
+* Dynarec: LA64: Added some more native flags optimisations
+* Dynarec: LA64: optimized REP MOVSB
+* Dynarec: RV64: Fixed/Fine tunning flags on existing opcodes
+* Dynarec: RV64: Various fixes to some opcodes
+
+v0.3.8
+======
+
+Highlights:
+
+* This version introduce DynaCache, a disk cache for the generated native code. It's not enabled by default for now (but it will use cache file if present by default). It can dramaticaly speedup the load of program (try with linux factorio for example). This version also improve in the volatile metadata handling. There is a massive compatiblity improvment for box32 and on all dynarec in general too. And few more function wrapping...
+
+    => DynaCache for faster loading of program, and avoid runtime hicups
+    => Batter box32 compatibility (much less "Out of Memory" error)
+    => Unity3D games on Wayland are working now
+    => Volatile Metadata fully supported and handled, making games like Starfield to just work.
+    => More work on AVX support for RV64 & LA64. LA64 almost there, while RV64 has Scalar only support (no hardware vector support yet for AVX)
+
+Version summary:
+
+* Improved and fixed memory tracking
+* Improved volatile metadata handling, and some light refactor on StrongMem emulation
+* Improved handling of TF flag
+* DynaCache: Introducing a way to cache on disk native generated code for faster startup time on subsequent release
+* DynaCache: Support for all 3 backends ARM64, RV64 & LA64
+* Interpreter: refactor opcodes decoding to have prefixes handled in a generic way
+* Interpreter: many small fixes
+* Dynarec: some optimisations in the jump table, to remove one memory fetch on each jump
+* Dynarec: create infrastructure to be able to change easily the reg mapping
+* Dynarec: optimised defered flag handling to limit cases where UpdateFlags is needed
+* Dynarec: Introduced BOX64_DYNAREC_NOHOTPAGE to disable detection of hotpages
+* Dynarec: ARM64: A few more opcode added
+* Dynarec: ARM64: Some small fixes and better handling of undefined behavior in opcodes, as well as undefined opcodes
+* Dynarec: ARM64: Implemented a Dynarec version of UpdateFlags, for improved performances
+* Dynarec; ARM64: Fixes/Improved SSE_FLUSHTO0 support
+* Dynarec: LA64: Added support for AVX (& friends) opcodes. Still need some debugging so AVX still disabled by default on LA64
+* Dynarec: LA64: Changed register mapping to be closer to the x86_64 ABI
+* Dynarec: LA64: RV64: Added limited support to SSE_FLUSHTO0
+* Dynarec: RV64: Many fixes and improvments for MMX opcodes
+* Dynarec: RV64: many Scalar version of AVX opcodes added
+* Dynarec: Many fixes to various OpCodes, on all 3 backends
+* WowBox64: Added minimal runtime, improved overall stability
+* BOX32: Wrapper: many new functions added, plus many fixes
+* BOX32: Improved memory hadnling, using 32bits personality and other tunning to get more reliable 32bits memory allocations
+* BOX64: Wrapper: some more wrapped functions added, including wayland (for Unity games on Wayland) and vulkan (more extensions)
+* BOX64: Wrapper: improved handling of some VarArg wrapped functions on non-ARM64 backend
+* RCFILE: Fixed MAXCPU not working properly with Wine programs
+* RCFILE: Generic Setup profile refined, and more games profile added
+* Trace: Added support for current libZydis
+    
 v0.3.6
 ======

@@ -1,11 +1,11 @@
 Compiling/Installing
 ----

-If you don't want to compile box64 yourself and prefer to use third-party pre-build version, go to the [end of the document](#pre-build-packages) for alternatives.
+If you don't want to compile box64 yourself and prefer to use third-party pre-build version, go to the [end of the document](#pre-built-packages) for alternatives.

 You can also generate your own package using the [instructions below](https://github.com/ptitSeb/box64/blob/main/docs/COMPILE.md#debian-packaging). 

-Additional installation steps may be necessary when copying only the box64 executable file without running make install in cross-build environments. See [Cross-compiling](https://github.com/ptitSeb/box64/blob/main/docs/COMPILE.md#Cross-compiling)
+Additional installation steps may be necessary when copying only the box64 executable file without running make install in cross-build environments. See [Cross-Compiling](https://github.com/ptitSeb/box64/blob/main/docs/COMPILE.md#Cross-Compiling)

 ## Per-platform compiling instructions

@@ -137,6 +137,22 @@ Note: use gcc-11 or higher, older gcc doesn't know cortex-a78ae
 -D TEGRA_T234=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
 ```

+#### for DGX Spark/GB10 based devices
+
+On a 64bit OS:
+
+Note: Requires GCC 15.2.0 or higher. Older GCC versions do not support gb10, cortex-a725, or cortex-x925.
+
+For GCC 14 or older, you can use the following march equivalent:
+```
+-march=armv9.2-a+crc+sve2-aes+sve2-bitperm+sve2-sha3+sve2-sm4+memtag+profile
+```
+
+With GCC 15.2.0 or higher:
+```
+-D NVIDIA_GB10=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
+```
+
 #### for ODROID N2/N2+

 On a 64bit OS:
@@ -149,16 +165,24 @@ On a 64bit OS:

 On a 64bit OS:

+Targeting Snapdragon 845:
+
 ```
 -D SD845=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
 ```

-or
+Targeting Snapdragon 888:

 ```
 -D SD888=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
 ```

+Targeting Snapdragon 8 Elite Gen 5:
+
+```
+-D SD8EG5=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
+```
+
 Depending on how recent your Snapdragon is

 #### for Phytium
@@ -260,7 +284,7 @@ Add `-DCMAKE_BUILD_TYPE=RelWithDebInfo` option for an optimized build with debug

 #### Build with trace

-To have a trace enabled build (***the interpreter will be slightly slower***), add `-DHAVE_TRACE=1`. You will need the [Zydis library](https://github.com/zyantific/zydis) in your `LD_LIBRARY_PATH` or in the system library folders at runtime to get x86 trace. Use version v3.2.1, as later versions have changed the API and no longer work with box64.
+To have a trace enabled build (***the interpreter will be slightly slower***), add `-DHAVE_TRACE=1`. You will need the [Zydis library](https://github.com/zyantific/zydis) devel package installed (`apt install libzydis-dev` on Debian for example).

 #### Build DynaRec

@@ -334,7 +358,7 @@ sudo apt update
 sudo apt install box64-generic-arm -y
 ```

-## Cross-compiling 
+## Cross-Compiling 
 ----
 ### Set Up the Cross-Compiler
 For example, to compile Box64 for RISC-V on an x86 machine, you can get prebuilt GNU toolchain from the [riscv-gnu-toolchain](https://github.com/riscv-collab/riscv-gnu-toolchain/releases) 
@@ -346,6 +370,9 @@ In particular, you must specify the cross-compiler. For example:
 -DCMAKE_C_COMPILER=riscv64-unknown-linux-gnu-gcc  # Or whichever cross-compiler you use
 ```

+Be aware of potential glibc capability errors when choosing a cross-compiler.
+To avoid this, run `ldd --version` on your target machine to check its glibc version.
+
 ### Running tests with QEMU (optional)
 To do a quick check, run:
 ```
@@ -361,6 +388,16 @@ ctest -j$(nproc)
 ### Installing on the Target Machine
 After successfully cross-compiling, copy the box64 executable to your RISC-V device. Note that simply copying the binary does not automatically install Box64’s shared libraries. Because `make install` does not run on the target during cross-compilation, libraries required for emulation may be missing.

-To fix this, copy the shared libraries folder from the Box64 repository (`x64lib` or `x86lib`) to your target. Place them into the proper library search paths, for example:
- `/usr/lib/box64-x86_64-linux-gnu/` (for x86_64) 
- `/usr/lib/box64-i386-linux-gnu/` (for i386) 
+To resolve this, copy the shared libraries folder from the Box64 repository (`x64lib` or `x86lib`) to your target device, then rename it and place it in the appropriate library search path (e.g., `/usr/lib`).
+
+For `x86_64`:
+```shell
+$ mv x64lib box64-x86_64-linux-gnu
+$ mv box64-x86_64-linux-gnu /usr/lib
+```
+For `i386`:
+```shell
+$ mv x86lib box64-i386-linux-gnu
+$ mv box64-i386-linux-gnu /usr/lib
+```
+
@@ -0,0 +1,19 @@
+# The guide on running Steam with Box64
+
+Steam works with Box64, both Linux version and Windows version. Systems with less than 6GB RAM may need a swap file for optimal performance.
+
+## Install and run Linux Steam
+
+Linux Steam requires Box86 or Box32 and it's binfmt enabled, check [Compilation Instructions](COMPILE.md) for details.
+
+```bash
+# This file can be found in the project root directory
+./install_steam.sh
+
+# And then, just run
+steam
+```
+
+## Install and run Windows Steam
+
+It's simple, just download it from Steam website and install it like other Windows applications.
@@ -0,0 +1,55 @@
+# The guide on using Wine with Box64
+
+<ins>W</ins>ine <ins>I</ins>s <ins>N</ins>ot an <ins>E</ins>mulator, but Box64 is. Box64 supports Wine in multiple ways. Before continuing reading, you should have a basic understanding of Wine.
+
+## Wine variants
+
+Wine ships in multiple forms:
+
+
+1. **x86 Wine**
+
+    The executable architecture of **x86 Wine** is x86, and it runs x32 Windows executables.
+
+    **x86 Wine** requires a 32-bit environment (read 32-bit Linux libraries), which is usually available on x86_64 machines because of the nature of the architecture.
+
+    However, when you try to run it via an emulator on RISC architectures, the situation is a bit different. For Arm64, there is armhf, which is a 32-bit environment, and there is also Box86, which is made for armhf, so you can use Box86 to run **x86 Wine**.
+
+    However, for RISC-V 64, LoongArch64, and many newer Arm64 cores, a 32-bit environment is not available. Box64 introduced a new mode called Box32, which can be enabled at compile time via a CMake option `BOX32`. It essentially emulates a 32-bit environment using the 64-bit environment, so it can be used to run **x86 Wine**. But be aware that Box32 is still highly experimental and unstable.
+
+2. **x86_64 Wine**
+
+    The executable architecture of **x86_64 Wine** is x86_64, and it runs x64 Windows executables.
+
+    It can be used with Box64. Note that **x86_64 Wine** usually ships with **x86 Wine** included. So you can use a combination of Box86 + Box64 or Box64 alone with Box32 enabled to run both x32 and x64 executables seamlessly.
+
+3. **x86_64 Wine WOW64**
+
+    The executable architecture of **x86_64 Wine WOW64** is x86_64, and it runs both x32 and x64 Windows executables.
+
+    What's great about **x86_64 Wine WOW64** is that it does not require a 32-bit environment for x32 Windows executable, as the 32-bit environment emulation is taken care of by Wine WOW64 itself.
+
+    That means you can just use Box64 (without Box32) to run x32 Windows executables. However, the downside is that the performance of WineD3D is pretty bad for 32-bit (use DXVK!).
+
+4. **Arm64 Wine WOW64**
+
+    The executable architecture of **Arm64 Wine WOW64** is Arm64, and it runs
+    x32, x64, Arm32, Arm64 Windows executables.
+
+    However, to run x32 and x64 ones, a cpu.dll is required, known as the 
+    WOW64 (for 32-bit) and Arm64EC (for 64-bit).
+
+    Box64 now supports 32-bit WOW64 DLL. You can have it compiled by enabling the `WOW64` CMake option. And we plan to support Arm64EC in the future.
+
+## Wine installation
+
+For x86 Wines, there are plenty of places to download, to name a few:
+
+- https://github.com/Kron4ek/Wine-Builds (Wine \[WOW64\])
+- https://github.com/GloriousEggroll/proton-ge-custom (Proton, known as the Valve-optimized Wine)
+
+After the download, just add the `bin` folder to your path in any way you like; Wine should work with Box64 just fine.
+
+To install components like DXVK, d3dcompiler_xx, or cjkfonts, it's recommended to use Winetricks: https://github.com/Winetricks/winetricks
+
+For Arm64 Wines, use of Hangover is recommended: https://github.com/AndreRH/hangover.
@@ -1,177 +0,0 @@
-# Installing Wine64
-Running Wine64/Wine with Box64/Box86 allows x64/x86 Windows programs to run on 64-bit ARM Linux devices.
- - Box64 needs `wine-amd64` to be installed **manually** on ARM64 (aarch64) devices.  This will run 64-bit Windows programs (aka x86_64).
- - Box86 needs `wine-i386` to be installed **manually** on ARM32 (armhf) devices (or on ARM64 with multiarch or chroot).  This will run 32-bit Windows programs (aka x86).
- - _Note that `wine:arm64` and `wine:armhf` will not work with Box64/Box86. Your device's package manager might try to install those if you ask it to install wine._
-
-See installation steps below (in the [Examples](#examples) section).
-
-## Overview
-The general procedure for installing Wine64 & Wine for Box64 & Box86 is to...
- - Download all the install files for the version of Wine you wish to install
- - Unzip or dpkg the install files into one folder
- - Move that folder to the directory that you wish Wine to run from (often `~/wine/` by default)
- - Go to `/usr/local/bin` and make symlinks or scripts that will point to your main wine binaries.
- - Boot wine to create a new wineprefix.
- - Download winetricks (which is just a complicated bash script), make it executable, then copy it to `/usr/local/bin`.
-
-## Examples
-### Installing Wine64 & Wine for Box64 & Box86 on Raspberry Pi OS from WineHQ .deb files
-_Links from the [WineHQ repo](https://dl.winehq.org/wine-builds/debian/dists/)_
-
-This install method allows you to install different versions of Wine64/Wine.  You may install any version/branch of Wine that you wish.
-```
-	# NOTE: Can only run on aarch64 (since box64 can only run on aarch64)
-	# box64 runs wine-amd64, box86 runs wine-i386.
-
-	### User-defined Wine version variables ################
-	# - Replace the variables below with your system's info.
-	# - Note that we need the amd64 version for Box64 even though we're installing it on our ARM processor.
-	# - Note that we need the i386 version for Box86 even though we're installing it on our ARM processor.
-	# - Wine download links from WineHQ: https://dl.winehq.org/wine-builds/
-  
-	local branch="devel" #example: devel, staging, or stable (wine-staging 4.5+ requires libfaudio0:i386)
-	local version="7.1" #example: "7.1"
-	local id="debian" #example: debian, ubuntu
-	local dist="bullseye" #example (for debian): bullseye, buster, jessie, wheezy, ${VERSION_CODENAME}, etc 
-	local tag="-1" #example: -1 (some wine .deb files have -1 tag on the end and some don't)
-
-  ########################################################
-
-	# Clean up any old wine instances
-	wineserver -k # stop any old wine installations from running
-	rm -rf ~/.cache/wine # remove any old wine-mono/wine-gecko install files
-	rm -rf ~/.local/share/applications/wine # remove any old program shortcuts
-
-	# Backup any old wine installs
-	rm -rf ~/wine-old 2>/dev/null; mv ~/wine ~/wine-old 2>/dev/null
-	rm -rf ~/.wine-old 2>/dev/null; mv ~/.wine ~/.wine-old 2>/dev/null
-	sudo mv /usr/local/bin/wine /usr/local/bin/wine-old 2>/dev/null
-	sudo mv /usr/local/bin/wine64 /usr/local/bin/wine-old 2>/dev/null
-	sudo mv /usr/local/bin/wineboot /usr/local/bin/wineboot-old 2>/dev/null
-	sudo mv /usr/local/bin/winecfg /usr/local/bin/winecfg-old 2>/dev/null
-	sudo mv /usr/local/bin/wineserver /usr/local/bin/wineserver-old 2>/dev/null
-
-	# Wine download links from WineHQ: https://dl.winehq.org/wine-builds/
-	LNKA="https://dl.winehq.org/wine-builds/${id}/dists/${dist}/main/binary-amd64/" #amd64-wine links
-	DEB_A1="wine-${branch}-amd64_${version}~${dist}${tag}_amd64.deb" #wine64 main bin
-	DEB_A2="wine-${branch}_${version}~${dist}${tag}_amd64.deb" #wine64 support files (required for wine64 / can work alongside wine_i386 main bin)
-		#DEB_A3="winehq-${branch}_${version}~${dist}${tag}_amd64.deb" #shortcuts & docs
-	LNKB="https://dl.winehq.org/wine-builds/${id}/dists/${dist}/main/binary-i386/" #i386-wine links
-	DEB_B1="wine-${branch}-i386_${version}~${dist}${tag}_i386.deb" #wine_i386 main bin
-	DEB_B2="wine-${branch}_${version}~${dist}${tag}_i386.deb" #wine_i386 support files (required for wine_i386 if no wine64 / CONFLICTS WITH wine64 support files)
-		#DEB_B3="winehq-${branch}_${version}~${dist}${tag}_i386.deb" #shortcuts & docs
-
-	# Install amd64-wine (64-bit) alongside i386-wine (32-bit)
-	echo -e "Downloading wine . . ."
-	wget -q ${LNKA}${DEB_A1} 
-	wget -q ${LNKA}${DEB_A2} 
-	wget -q ${LNKB}${DEB_B1} 
-	echo -e "Extracting wine . . ."
-	dpkg-deb -x ${DEB_A1} wine-installer
-	dpkg-deb -x ${DEB_A2} wine-installer
-	dpkg-deb -x ${DEB_B1} wine-installer
-	echo -e "Installing wine . . ."
-	mv wine-installer/opt/wine* ~/wine
-	
-	# Download wine dependencies
-	# - these packages are needed for running box86/wine-i386 on a 64-bit RPiOS via multiarch
-	sudo dpkg --add-architecture armhf && sudo apt-get update # enable multi-arch
-	sudo apt-get install -y libasound2:armhf libc6:armhf libglib2.0-0:armhf libgphoto2-6:armhf libgphoto2-port12:armhf \
-		libgstreamer-plugins-base1.0-0:armhf libgstreamer1.0-0:armhf libldap-2.4-2:armhf libopenal1:armhf libpcap0.8:armhf \
-		libpulse0:armhf libsane1:armhf libudev1:armhf libusb-1.0-0:armhf libvkd3d1:armhf libx11-6:armhf libxext6:armhf \
-		libasound2-plugins:armhf ocl-icd-libopencl1:armhf libncurses6:armhf libncurses5:armhf libcap2-bin:armhf libcups2:armhf \
-		libdbus-1-3:armhf libfontconfig1:armhf libfreetype6:armhf libglu1-mesa:armhf libglu1:armhf libgnutls30:armhf \
-		libgssapi-krb5-2:armhf libkrb5-3:armhf libodbc1:armhf libosmesa6:armhf libsdl2-2.0-0:armhf libv4l-0:armhf \
-		libxcomposite1:armhf libxcursor1:armhf libxfixes3:armhf libxi6:armhf libxinerama1:armhf libxrandr2:armhf \
-		libxrender1:armhf libxxf86vm1 libc6:armhf libcap2-bin:armhf # to run wine-i386 through box86:armhf on aarch64
-		# This list found by downloading...
-		#	wget https://dl.winehq.org/wine-builds/debian/dists/bullseye/main/binary-i386/wine-devel-i386_7.1~bullseye-1_i386.deb
-		#	wget https://dl.winehq.org/wine-builds/debian/dists/bullseye/main/binary-i386/winehq-devel_7.1~bullseye-1_i386.deb
-		#	wget https://dl.winehq.org/wine-builds/debian/dists/bullseye/main/binary-i386/wine-devel_7.1~bullseye-1_i386.deb
-		# then `dpkg-deb -I package.deb`. Read output, add `:armhf` to packages in dep list, then try installing them on Pi aarch64.
-		
-	# - these packages are needed for running box64/wine-amd64 on RPiOS (box64 only runs on 64-bit OS's)
-	sudo apt-get install -y libasound2:arm64 libc6:arm64 libglib2.0-0:arm64 libgphoto2-6:arm64 libgphoto2-port12:arm64 \
-		libgstreamer-plugins-base1.0-0:arm64 libgstreamer1.0-0:arm64 libldap-2.4-2:arm64 libopenal1:arm64 libpcap0.8:arm64 \
-		libpulse0:arm64 libsane1:arm64 libudev1:arm64 libunwind8:arm64 libusb-1.0-0:arm64 libvkd3d1:arm64 libx11-6:arm64 libxext6:arm64 \
-		ocl-icd-libopencl1:arm64 libasound2-plugins:arm64 libncurses6:arm64 libncurses5:arm64 libcups2:arm64 \
-		libdbus-1-3:arm64 libfontconfig1:arm64 libfreetype6:arm64 libglu1-mesa:arm64 libgnutls30:arm64 \
-		libgssapi-krb5-2:arm64 libjpeg62-turbo:arm64 libkrb5-3:arm64 libodbc1:arm64 libosmesa6:arm64 libsdl2-2.0-0:arm64 libv4l-0:arm64 \
-		libxcomposite1:arm64 libxcursor1:arm64 libxfixes3:arm64 libxi6:arm64 libxinerama1:arm64 libxrandr2:arm64 \
-		libxrender1:arm64 libxxf86vm1:arm64 libc6:arm64 libcap2-bin:arm64
-		# This list found by downloading...
-		#	wget https://dl.winehq.org/wine-builds/debian/dists/bullseye/main/binary-amd64/wine-devel_7.1~bullseye-1_amd64.deb
-		#	wget https://dl.winehq.org/wine-builds/debian/dists/bullseye/main/binary-amd64/wine-devel-amd64_7.1~bullseye-1_amd64.deb
-		# then `dpkg-deb -I package.deb`. Read output, add `:arm64` to packages in dep list, then try installing them on Pi aarch64.	
-
-	# These packages are needed for running wine-staging on RPiOS (Credits: chills340)
-	sudo apt install libstb0 -y
-	cd ~/Downloads
-	wget -r -l1 -np -nd -A "libfaudio0_*~bpo10+1_i386.deb" http://ftp.us.debian.org/debian/pool/main/f/faudio/ # Download libfaudio i386 no matter its version number
-	dpkg-deb -xv libfaudio0_*~bpo10+1_i386.deb libfaudio
-	sudo cp -TRv libfaudio/usr/ /usr/
-	rm libfaudio0_*~bpo10+1_i386.deb # clean up
-	rm -rf libfaudio # clean up
-
-	# Install symlinks
-	sudo ln -s ~/wine/bin/wine /usr/local/bin/wine
-	sudo ln -s ~/wine/bin/wine64 /usr/local/bin/wine64
-	sudo ln -s ~/wine/bin/wineboot /usr/local/bin/wineboot
-	sudo ln -s ~/wine/bin/winecfg /usr/local/bin/winecfg
-	sudo ln -s ~/wine/bin/wineserver /usr/local/bin/wineserver
-	sudo chmod +x /usr/local/bin/wine /usr/local/bin/wine64 /usr/local/bin/wineboot /usr/local/bin/winecfg /usr/local/bin/wineserver
-```
-
-## Installing winetricks
-Winetricks is a bash script which makes it easier to install & configure any desired Windows core system software packages which may be dependencies for certain Windows programs.  You can either install it with `apt`, or manually using the steps below.
-
-```
-sudo apt-get install cabextract -y                                                                   # winetricks needs this installed
-sudo mv /usr/local/bin/winetricks /usr/local/bin/winetricks-old                                      # Backup old winetricks
-cd ~/Downloads && wget https://raw.githubusercontent.com/Winetricks/winetricks/master/src/winetricks # Download
-sudo chmod +x winetricks && sudo mv winetricks /usr/local/bin/                                       # Install
-```
-
-Whenever we run winetricks, we must tell Box86 to supress its info banner, otherwise winetricks will crash.  Box86's info banner can be suppressed using the `BOX86_NOBANNER=1` environment variable before invoking winetricks (example: `BOX86_NOBANNER=1 winetricks`).
-
-If `apt` installed a desktop menu shortcut for winetricks (or if you make your own desktop shortcut for winetricks) then you may need to edit that shortcut to include Box86's BOX86_NOBANNER=1 env var.  With your favorite text editor, edit `/usr/share/applications/winetricks.desktop` and change `Exec=winetricks --gui` to `Exec=env BOX86_NOBANNER=1 winetricks --gui`.
-
-## Example commands
-Creating a 64-bit wineprefix:
-`wine64 wineboot` or `wine wineboot` (wineprefix is created in `~/.wine/`)
-`WINEPREFIX="$HOME/prefix64" wine wineboot` (wineprefix is created in `~/prefix64/`) _Note: You will need to invoke `WINEPREFIX="$HOME/prefix64"` before every command to use this wineprefix._
-
-Creating a 32-bit wineprefix:
-`WINEARCH=win32 wine wineboot` (wineprefix is created in `~/.wine/`)
-
-Fore-quitting wine:
-`wineserver -k`
-
-Running Wine Configuration:
-`winecfg`
-
-Using winetricks:
-`winetricks -q corefonts vcrun2010 dotnet20sp1`  
-_This command will silently install three packages in series: Windows core fonts, VC++ 2010 Runtimes, and .NET 2.0 SP1.  `-q` is the "install silent/quiet" command._
-
-Invoking Box86's logging features (with `BOX86_LOG=1` or similar) will cause winetricks to crash.
-
-For a list of all the different Windows packages & libraries that winetricks can help you install, run `winetricks list-all`  
-
-## Other notes
-
-### Wineprefixes (and Wine initialization)
-When you first run or boot Wine (`wine wineboot`), Wine will create a new user environment within which to install Windows software.  This user environment is called a "wineprefix" (or "wine bottle") and is located (by default) in `~/.wine` (note that Linux folders with a `.` in front of them are "hidden" folders).  Think of a wineprefix as Wine's virtual 'harddrive' where it installs software and saves settings.  Wineprefixes are portable and deletable.  For more Wine documentation, see [WineHQ](https://www.winehq.org/documentation).
-
-If you at any point corrupt something inside your default wineprefix, you can start "fresh" by deleting your `~/.wine` directory (with the `rm -rf ~/.wine` command) and boot wine again to create a new default wineprefix.
-
-### Transplanting wineprefixes
-If software isn't installing in Wine with Box86, but is installing for you in Wine on a regular x86 Linux computer, you can copy a wineprefix from your x86 Linux computer to the device you're running Box86 on. This is most easily done by tarring the `~/.wine` folder on your x86 Linux computer (`tar -cvf winebottle.tar ~/.wine`), transferring the tar file to your device, then un-tarring the tar file (`tar -xf winebottle.tar`) on your device running Box86 & Wine. Tarring the wineprefix preserves any symlinks that are inside it.
-
-### Swapping out different versions of Wine
-Some versions of Wine work better with certain software. It is best to install a version of Wine that is known to work with the software you would like to run. There are three main development branches of Wine you can pick from, referred to as wine-stable, wine-devel, and wine-staging. _Note that the wine-staging branch requires extra installation steps on Raspberry Pi._
-
-Your entire Wine installation can reside within a single folder on your Linux computer. TwisterOS assumes that your Wine installation is located inside the `~/wine/` directory. The actual directory where you put your `wine` folder doesn't matter as long as you have symlinks within the `/usr/local/bin/` directory which point to the `wine` folder so that Linux can find Wine when you type `wine` into the terminal).
-
-You can change which version of Wine you are running simply by renaming your old `wine` and `.wine` folders to something else, then putting a new `wine` folder (containing your new version of Wine) in its place. Running `wine wineboot` again will migrate a wineprefix from an older version of Wine to be usable by the newer version of Wine you just installed.  You can check which version of Wine you're running with the `wine --version` command.
@@ -57,6 +57,14 @@ Add extras to the needed libraries list, rarely needed.
 * XXXX:YYYY:ZZZZ : Add library XXXX, YYYY and ZZZZ to the needed libraries list. 


+=item B<BOX64_AES> =I<0|1>
+
+Expose AES capabilities. Available in WowBox64.
+
+ * 0 : Do not expose AES capabilities. 
+ * 1 : Expose AES capabilities. [Default]
+
+
 =item B<BOX64_ALLOWMISSINGLIBS> =I<0|1>

 Allow missing libraries to be ignored.
@@ -65,6 +73,16 @@ Allow missing libraries to be ignored.
 * 1 : Allow missing libraries. 


+=item B<BOX64_ARCH> =I<`arm64`|`la64`|`rv64`|`<empty>`>
+
+Specify the architecture in rcfiles to which the current entry applies. This option only takes effect when used inside rcfiles.
+
+ * `arm64` : Only apply current entry for AArch64. 
+ * `la64` : Only apply current entry for LoongArch64. 
+ * `rv64` : Only apply current entry for RISC-V. 
+ * `<empty>` : Apply current entry for all architectures, will be replaced by more specific entries. [Default]
+
+
 =item B<BOX64_ARGS> =I<XXXX|XXXX YYYY ZZZZ>

 Arguments to pass to the guest program, only valid if there is no existing arguments.
@@ -75,7 +93,7 @@ Arguments to pass to the guest program, only valid if there is no existing argum

 =item B<BOX64_AVX> =I<0|1|2>

-Expose AVX extension to CPUID and cpuinfo file. Default value is 2 on Arm64 because it's fully implemented in DynaRec, 0 otherwise. Availble in WowBox64.
+Expose AVX extension to CPUID and cpuinfo file. Default value is 2 on Arm64 because it's fully implemented in DynaRec, 0 otherwise. Available in WowBox64.

 * 0 : Do not expose AVX capabilities. 
 * 1 : Expose AVX, BMI1, F16C and VAES extensions to CPUID and cpuinfo file. 
@@ -89,25 +107,9 @@ Path to the bash executable.
 * XXXX : Use bash executable at path XXXX. 


-=item B<BOX64_CEFDISABLEGPU> =I<0|1>
-
-Add -cef-disable-gpu argument to the guest program.
-
- * 0 : Does nothing. [Default]
- * 1 : Add -cef-disable-gpu argument to the guest program. 
-
-
-=item B<BOX64_CEFDISABLEGPUCOMPOSITOR> =I<0|1>
-
-Add -cef-disable-gpu-compositor argument to the guest program.
-
- * 0 : Does nothing. [Default]
- * 1 : Add -cef-disable-gpu-compositor argument to the guest program. 
-
-
 =item B<BOX64_CPUTYPE> =I<0|1>

-Specify the CPU type to emulate. Availble in WowBox64.
+Specify the CPU type to emulate. Available in WowBox64.

 * 0 : Emulate a Intel CPU Model. [Default]
 * 1 : Emulate a AMD CPU Model. 
@@ -117,7 +119,7 @@ Specify the CPU type to emulate. Availble in WowBox64.

 Use a dummy crash handler or not.

- * 0 : Use emulated crashhandler.so library if nedded. 
+ * 0 : Use emulated crashhandler.so library if needed. 
 * 1 : Use an internal dummy (empty) crashhandler.so library. [Default]


@@ -131,7 +133,7 @@ Enable or disable the logging of dlsym errors.

 =item B<BOX64_DUMP> =I<0|1>

-Dump elfloader debug information. Availble in WowBox64.
+Dump elfloader debug information. Available in WowBox64.

 * 0 : Do not dump elfloader debug information. [Default]
 * 1 : Dump elfloader debug information. 
@@ -139,7 +141,7 @@ Dump elfloader debug information. Availble in WowBox64.

 =item B<BOX64_DYNAREC> =I<0|1>

-Enable/disable the Dynamic Recompiler (a.k.a DynaRec). This option defaults to 1 if it's enabled in the build options for a supported architecture. Availble in WowBox64.
+Enable/disable the Dynamic Recompiler (a.k.a DynaRec). This option defaults to 1 if it's enabled in the build options for a supported architecture. Available in WowBox64.

 * 0 : Disable DynaRec. 
 * 1 : Enable DynaRec. 
@@ -147,15 +149,15 @@ Enable/disable the Dynamic Recompiler (a.k.a DynaRec). This option defaults to 1

 =item B<BOX64_DYNAREC_ALIGNED_ATOMICS> =I<0|1>

-Generate aligned atomics only (only available on Arm64 for now). Availble in WowBox64.
+Generate aligned atomics only (only available on Arm64 for now). Available in WowBox64.

 * 0 : Generate unaligned atomics handling code. [Default]
- * 1 : Generate aligned atomics only, which is faster and smaller code size, but will cause SIGBUS for LOCK prefixed opcodes operating on aligned data addresses. 
+ * 1 : Generate aligned atomics only, which is faster and smaller code size, but will cause SIGBUS for LOCK prefixed opcodes operating on unaligned data addresses. 


 =item B<BOX64_DYNAREC_BIGBLOCK> =I<0|1|2|3>

-Enable building bigger DynaRec code blocks for better performance. Availble in WowBox64.
+Enable building bigger DynaRec code blocks for better performance. Available in WowBox64.

 * 0 : Do not try to build block as big as possible, suitable for programs using lots of threads and JIT, like Unity. 
 * 1 : Build Dynarec block as big as possible. 
@@ -165,7 +167,7 @@ Enable building bigger DynaRec code blocks for better performance. Availble in W

 =item B<BOX64_DYNAREC_BLEEDING_EDGE> =I<0|1>

-Detect MonoBleedingEdge and apply conservative settings.
+Detect MonoBleedingEdge and apply conservative settings (Linux only).

 * 0 : Do not detect MonoBleedingEdge. 
 * 1 : Detect MonoBleedingEdge and apply BOX64_DYNAREC_BIGBLOCK=0 BOX64_DYNAREC_STRONGMEM=1 when detected. [Default]
@@ -173,16 +175,25 @@ Detect MonoBleedingEdge and apply conservative settings.

 =item B<BOX64_DYNAREC_CALLRET> =I<0|1|2>

-Optimize CALL/RET opcodes. Availble in WowBox64.
+Optimize CALL/RET opcodes. Available in WowBox64.

 * 0 : Do not optimize CALL/RET, use jump table. [Default]
 * 1 : Try to optimize CALL/RET, skipping the jump table when possible. 
 * 2 : Try to optimize CALL/RET, skipping the jump table when possible, adding code to handle return to dirty/modified block. Does not work on WowBox64. 


+=item B<BOX64_DYNAREC_SEP> =I<0|1|2>
+
+Optimize CALL/RET further with Secondary Entry Points (SEP). Have no effect is CALLRET is not enabled Available in WowBox64.
+
+ * 0 : Do not add SEP at CALLRET return 
+ * 1 : Add SEP at CALLRET return, on meory that are from known binary files [Default]
+ * 2 : Add SEP at CALLRET return for all type of memory. 
+
+
 =item B<BOX64_DYNAREC_DF> =I<0|1>

-Enable or disable the use of deferred flags. Availble in WowBox64.
+Enable or disable the use of deferred flags. Available in WowBox64.

 * 0 : Disable the use of deferred flags. 
 * 1 : Enable the use of deferred flags. [Default]
@@ -197,17 +208,33 @@ Allow continue running a block that is unprotected and potentially dirty.
 * 2 : Will also, when it detect an HotPage, flag that page as NEVERCLEAN, and so it will not be write protected but Block build from that page will always be tested. It can be faster that way (but soem SMC case might not be trapped). 


+=item B<BOX64_DYNAREC_NOHOTPAGE> =I<0|1>
+
+Disable detection of hot page (where code is executed and data written at the same time).
+
+ * 0 : Detect hot page [Default]
+ * 1 : Do not detect hot page 
+
+
+=item B<BOX64_DYNAREC_HOTPAGE_ALT> =I<0|1>
+
+Use alternative way to detect HotPages (legacy way).
+
+ * 0 : Detect hot page, using new method from v0.3.9+ [Default]
+ * 1 : Use alternate method to detect HotPage (from v0.3.8 and earlier) 
+
+
 =item B<BOX64_DYNAREC_DIV0> =I<0|1>

-Enable or disable the generation of division-by-zero exception. Availble in WowBox64.
+Enable or disable the generation of division-by-zero exception. Available in WowBox64.

- * 0 : Do not generate thr division-by-zero exception. [Default]
+ * 0 : Do not generate the division-by-zero exception. [Default]
 * 1 : Generate the division-by-zero exception. 


 =item B<BOX64_DYNAREC_DUMP> =I<0|1|2>

-Enable DynaRec dump. Availble in WowBox64.
+Enable DynaRec dump. Available in WowBox64.

 * 0 : Do not dump DynaRec blocks. [Default]
 * 1 : Dump DynaRec blocks. 
@@ -216,14 +243,14 @@ Enable DynaRec dump. Availble in WowBox64.

 =item B<BOX64_DYNAREC_DUMP_RANGE> =I<0xXXXXXXXX-0xYYYYYYYY>

-Dump DynaRec blocks in the specified range. Availble in WowBox64.
+Dump DynaRec blocks in the specified range. Available in WowBox64.

 * 0xXXXXXXXX-0xYYYYYYYY : Define the range where dynablock gets dumped (inclusive-exclusive). 


 =item B<BOX64_DYNAREC_FASTNAN> =I<0|1>

-Enable or disable fast NaN handling. Availble in WowBox64.
+Enable or disable fast NaN handling. Available in WowBox64.

 * 0 : Precisely emulate the -NaN generation like on x86. 
 * 1 : Do not do anything special with -NAN generation, faster. [Default]
@@ -231,7 +258,7 @@ Enable or disable fast NaN handling. Availble in WowBox64.

 =item B<BOX64_DYNAREC_FASTROUND> =I<0|1|2>

-Enable or disable fast rounding. Availble in WowBox64.
+Enable or disable fast rounding. Available in WowBox64.

 * 0 : Generate float/double -> int rounding and use current rounding mode for float/double computation like on x86. 
 * 1 : Do not do anything special with edge case rounding, faster. [Default]
@@ -240,7 +267,7 @@ Enable or disable fast rounding. Availble in WowBox64.

 =item B<BOX64_DYNAREC_FORWARD> =I<0|128|XXXX>

-Define max allowed forward value when building block. Availble in WowBox64.
+Define max allowed forward value when building block. Available in WowBox64.

 * 0 : No forward value. When current block ends, do not try to go further even if there are previous forward jumps. 
 * 128 : Allow up to 128 bytes of gap between end of the block and the next forward jump. [Default]
@@ -260,7 +287,7 @@ The GDBJIT debugging support, only available on build with `-DGDBJIT=ON`, enable

 =item B<BOX64_DYNAREC_LOG> =I<0|1|2|3>

-Disable or enable DynaRec logs. Availble in WowBox64.
+Disable or enable DynaRec logs. Available in WowBox64.

 * 0 : Disable DynaRec logs. [Default]
 * 1 : Enable minimal DynaRec logs. 
@@ -270,7 +297,7 @@ Disable or enable DynaRec logs. Availble in WowBox64.

 =item B<BOX64_DYNAREC_MISSING> =I<0|1|2>

-Print missing opcodes. Availble in WowBox64.
+Print missing opcodes. Available in WowBox64.

 * 0 : Do not print the missing opcode. [Default]
 * 1 : Print missing opcodes. 
@@ -279,15 +306,24 @@ Print missing opcodes. Availble in WowBox64.

 =item B<BOX64_DYNAREC_NATIVEFLAGS> =I<0|1>

-Enable or disable the use of native flags. Availble in WowBox64.
+Enable or disable the use of native flags. Available in WowBox64.

 * 0 : Do not use native flags. 
 * 1 : Use native flags when possible. [Default]


+=item B<BOX64_DYNAREC_NOHOSTEXT> =I<0|1|xxx,yyy,zzz>
+
+Disable optional host extensions.
+
+ * 0 : Disable nothing. [Default]
+ * 1 : Disable all the optional extensions. 
+ * xxx,yyy,zzz : Disable selected optional extensions, choices are Arm: `crc32,pmull,aes,atomics,sha1,sha2,uscat,flagm,flagm2,frintts,afp,rndr`, LoongArch: `lasx,lbt,frecipe,lam_bh,lamcas,scq`, and RISC-V: `zba,zbb,zbc,zbs,vector,xtheadba,xtheadbb,xtheadbs,xtheadmemidx,xtheadmempair,xtheadcondmov`. 
+
+
 =item B<BOX64_DYNAREC_PAUSE> =I<0|1|2|3>

-Enable x86 PAUSE emulation, may help the performance of spinlocks. Availble in WowBox64.
+Enable x86 PAUSE emulation, may help the performance of spinlocks. Available in WowBox64.

 * 0 : Ignore x86 PAUSE instruction. [Default]
 * 1 : Use YIELD to emulate x86 PAUSE instruction. 
@@ -303,23 +339,40 @@ Generate map file for Linux perf tool.
 * 1 : Dynarec will generate perf map. 


+=item B<BOX64_DYNAREC_PURGE> =I<0|1>
+
+Purge dynablocks that haven't been executed recently, based on tick age threshold (BOX64_DYNAREC_PURGE_AGE).
+
+ * 0 : Dynarec will not purge old unused dynablocks. [Default]
+ * 1 : Dynarec will purge old unused dynablocks based on tick age. 
+
+
+=item B<BOX64_DYNAREC_PURGE_AGE> =I<4096|XXXX>
+
+Tick age threshold for dynablock purging (requires BOX64_DYNAREC_PURGE=1).
+
+ * 4096 : Default age threshold. [Default]
+ * XXXX : Custom age threshold (range: 10-65536). 
+
+
 =item B<BOX64_DYNAREC_SAFEFLAGS> =I<0|1|2>

-Behaviour of flags emulation on CALL/RET opcodes. Availble in WowBox64.
+Behaviour of flags emulation on CALL/RET opcodes and other edge cases. Available in WowBox64.

 * 0 : Treat CALL/RET as if it never needs any flags. 
 * 1 : Most of RETs will need flags, most of CALLs will not. [Default]
- * 2 : All CALL/RET will need flags. 
+ * 2 : All CALL/RET will need flags. All edge cases are also handled. 


-=item B<BOX64_DYNAREC_STRONGMEM> =I<0|1|2|3>
+=item B<BOX64_DYNAREC_STRONGMEM> =I<0|1|2|3|4>

-Enable the emulation of x86 strong memory model. Availble in WowBox64.
+Enable the emulation of x86 strong memory model. Available in WowBox64.

 * 0 : Do not try anything special. [Default]
 * 1 : Enable some memory barriers when writing to memory to emulate the x86 strong memory model in a limited way. 
 * 2 : All in 1, plus memory barriers on SIMD instructions. 
 * 3 : All in 2, plus more memory barriers on a regular basis. 
+ * 4 : Mimic x86 TSO similarly to QEMU's approach, for evaluation purposes. 


 =item B<BOX64_DYNAREC_TBB> =I<0|1>
@@ -332,13 +385,29 @@ Enable or disable libtbb detection.

 =item B<BOX64_DYNAREC_TEST> =I<0|1|0xXXXXXXXX-0xYYYYYYYY>

-Enable DynaRec execution comparison with the interpreter, very slow, only for testing. Availble in WowBox64.
+Enable DynaRec execution comparison with the interpreter, very slow, only for testing. Available in WowBox64.

 * 0 : No comparison. [Default]
 * 1 : Each opcode runs on interpreter and on Dynarec, regs and memory are compared and printed when they differ. 
 * 0xXXXXXXXX-0xYYYYYYYY : Define the range where dynarec is tested (inclusive-exclusive). 


+=item B<BOX64_DYNAREC_TEST_NODUP> =I<0|1>
+
+Enable deduplication of IP address where COSIM find a difference. Available in WowBox64.
+
+ * 0 : Show all addresses in error. [Default]
+ * 1 : Show address in error only 1 time, even if dump might differs. 
+
+
+=item B<BOX64_DYNAREC_TEST_NODUMP> =I<0|1>
+
+Do not dump test code when BOX64_DYNAREC_DUMP is enabled. Available in WowBox64.
+
+ * 0 : Dump test code. 
+ * 1 : Do not dump test code. [Default]
+
+
 =item B<BOX64_DYNAREC_TRACE> =I<0|1>

 Enable or disable DynaRec trace.
@@ -357,7 +426,7 @@ Use volatile metadata parsed from PE files, only valid for 64bit Windows games.

 =item B<BOX64_DYNAREC_WAIT> =I<0|1>

-Wait or not for the building of a DynaRec code block to be ready. Availble in WowBox64.
+Wait or not for the building of a DynaRec code block to be ready. Available in WowBox64.

 * 0 : Do not wait and use interpreter instead, might speedup a bit on massive multithread or JIT programs. 
 * 1 : Wait for a DynaRec code block to be ready. [Default]
@@ -365,7 +434,7 @@ Wait or not for the building of a DynaRec code block to be ready. Availble in Wo

 =item B<BOX64_DYNAREC_WEAKBARRIER> =I<0|1|2>

-Tweak the memory barriers to reduce the performance impact by strong memory emualtion. Availble in WowBox64.
+Tweak the memory barriers to reduce the performance impact by strong memory emualtion. Available in WowBox64.

 * 0 : Use regular safe barrier. 
 * 1 : Use weak barriers to have more performance boost. [Default]
@@ -374,13 +443,45 @@ Tweak the memory barriers to reduce the performance impact by strong memory emua

 =item B<BOX64_DYNAREC_X87DOUBLE> =I<0|1|2>

-Force the use of float/double for x87 emulation. Availble in WowBox64.
+Force the use of float/double for x87 emulation. Available in WowBox64.

 * 0 : Try to use float when possible for x87 emulation. [Default]
 * 1 : Only use Double for x87 emulation. 
 * 2 : Check Precision Control low precision on x87 emulation. 


+=item B<BOX64_DYNAREC_INTERP_SIGNAL> =I<0|1>
+
+Allow the use of Dynarec in Signal Handlers.
+
+ * 0 : Allow the use of Dynarec in signal handlers. [Default]
+ * 1 : Only use the Interpreter for signal handlers (as it was in v0.3.8 and earlier). 
+
+
+=item B<BOX64_DYNACACHE> =I<0|1|2>
+
+Enable/disable the Dynamic Recompiler Cache (a.k.a DynaCache). This option defaults to 2 (to read cache if present but not generate any). DynaCache write file to home folder by default, and can grow without limit.
+
+ * 0 : Disable DynaCache. 
+ * 1 : Enable DynaCache. 
+ * 2 : Use DynaCache files if present, but do not generate new one. [Default]
+
+
+=item B<BOX64_DYNACACHE_FOLDER> =I<XXXX>
+
+Set the folder for DynaCache files. Default is $XDG_CACHE_HOME/box64 or $HOME/.cache/box64 if $XDG_CACHE_HOME is not set.
+
+ * XXXX : Use folder XXXX for DynaCache files. 
+
+
+=item B<BOX64_DYNACACHE_MIN> =I<XXXX|350>
+
+Minimum size, in KB, for a DynaCache to be written to disk. Default size is 350KB
+
+ * XXXX : Set a minimum size of XXXX KB of Dynarec code to write the dynacache to disk. Will not be saved to disk else. 
+ * 350 : A size of 350 KB is the default value. [Default]
+
+
 =item B<BOX64_EMULATED_LIBS> =I<XXXX|XXXX:YYYY:ZZZZ>

 Force the use of emulated libraries.
@@ -419,6 +520,14 @@ Fix 64bit inodes.
 * 1 : Fix 64bit inodes. 


+=item B<BOX64_FORCE_LD_PRELOAD> =I<0|1>
+
+Force symbols from BOX64_LD_PRELOAD libraries to be resolved by RTLD_NEXT.
+
+ * 0 : Do nothing. [Default]
+ * 1 : Force symbols from BOX64_LD_PRELOAD libraries to be resolved by RTLD_NEXT. 
+
+
 =item B<BOX64_IGNOREINT3> =I<0|1>

 Ignore INT3 instructions.
@@ -480,8 +589,8 @@ Force loading libraries with the binary.

 Detect libcef and apply BOX64_MALLOC_HACK settings.

- * 0 : Does nothing. 
- * 1 : Detect libcef, and apply BOX64_MALLOC_HACK=2 if detected. [Default]
+ * 0 : Does nothing. [Default]
+ * 1 : Detect libcef, and apply BOX64_MALLOC_HACK=2 if detected. 


 =item B<BOX64_LIBGL> =I<libXXXX|/path/to/libXXXX>
@@ -501,7 +610,7 @@ Set the address where the program is loaded, only active for PIE guest programs.

 =item B<BOX64_LOG> =I<0|1|2|3>

-Enable or disable Box64 logs, default value is 0 if stdout is not terminal, 1 otherwise. Availble in WowBox64.
+Enable or disable Box64 logs, default value is 0 if stdout is not terminal, 1 otherwise. Available in WowBox64.

 * 0 : Disable Box64 logs. 
 * 1 : Enable minimal Box64 logs. 
@@ -520,7 +629,7 @@ Behaviour when hooking malloc operators.

 =item B<BOX64_MAXCPU> =I<0|XXXX>

-Maximum CPU cores exposed. Availble in WowBox64.
+Maximum CPU cores exposed. Available in WowBox64.

 * 0 : Use the actual number of CPU cores. [Default]
 * XXXX : Use XXXX CPU cores. 
@@ -536,7 +645,7 @@ Force 32-bit compatible memory mappings on 64-bit programs that run 32-bit code

 =item B<BOX64_NOBANNER> =I<0|1>

-Disable the Box64 banner. Availble in WowBox64.
+Disable the Box64 banner. Available in WowBox64.

 * 0 : Show the Box64 banner. 
 * 1 : Do not show the Box64 banner. 
@@ -544,11 +653,28 @@ Disable the Box64 banner. Availble in WowBox64.

 =item B<BOX64_NODYNAREC> =I<0xXXXXXXXX-0xYYYYYYYY>

-Forbid dynablock creation in the address range specified, helpful for debugging behaviour difference between Dynarec and Interpreter. Availble in WowBox64.
+Forbid dynablock creation in the address range specified, helpful for debugging behaviour difference between Dynarec and Interpreter. Available in WowBox64.

 * 0xXXXXXXXX-0xYYYYYYYY : Define the range where dynablock creation is forbidden (inclusive-exclusive). 


+=item B<BOX64_NODYNAREC_DELAY> =I<0|1>
+
+Remove NoDynarec range after a delay. Available in WowBox64.
+
+ * 0 : Do not remove NoDynarec range. [Default]
+ * 1 : Remove the NoDynarec range once 2 threads have been created. 
+
+
+=item B<BOX64_DYNAREC_NOARCH> =I<0|1|2>
+
+Remove Architecture metatdata from Dynarec blocks (used on signal). Available in WowBox64.
+
+ * 0 : Generate all architecture metadata for each blocks, including x87/mmx/SSE/AVX regs tracking. Wil use some more memory. [Default]
+ * 1 : Generate only essential metadata for each blocks (flags, alignments), saving some memory, but some encrypted software might fails. 
+ * 2 : Do not generate metadata for blocks. Saving a bit more memory but might fails on signal handling. 
+
+
 =item B<BOX64_NOGTK> =I<0|1>

 Do not load wrapped GTK libraries.
@@ -605,6 +731,14 @@ Do not load Vulkan libraries.
 * 1 : Do not load Vulkan libraries, both native and x86_64. 


+=item B<BOX64_NOVULKANOVERLAY> =I<0|1>
+
+Do not load x64 Vulkan overlay.
+
+ * 0 : Do try to Load x64 Vulkan overlay. [Default]
+ * 1 : Do not load x64 Vulkan overlay. 
+
+
 =item B<BOX64_PATH> =I<XXXX>

 Path to look for x86_64 binaries.
@@ -612,6 +746,14 @@ Path to look for x86_64 binaries.
 * XXXX : Add path XXXX to the binary path. 


+=item B<BOX64_PCLMULQDQ> =I<0|1>
+
+Expose PCLMULQDQ capabilities. Available in WowBox64.
+
+ * 0 : Do not expose PCLMULQDQ capabilities. 
+ * 1 : Expose PCLMULQDQ capabilities. [Default]
+
+
 =item B<BOX64_PREFER_EMULATED> =I<0|1>

 Prefer emulated libraries over native ones.
@@ -628,6 +770,24 @@ Prefer wrapped libs first even if the lib is specified with absolute path.
 * 1 : Prefer wrapped libs first even if the lib is specified with absolute path. 


+=item B<BOX64_PROFILE> =I<safest|safe|default|fast|fastest>
+
+Predefined sets of environment variables with compatibility or performance in mind. Available in WowBox64.
+
+ * safest : Profile with all the unsafe DynaRec optimizations disabled. 
+ * safe : Well, less safer than safest. 
+ * default : The default settings that most programs runs fine with a handy performance. [Default]
+ * fast : Enable many unsafe optimizations, but also enable strongmem emulation. 
+ * fastest : Enable many unsafe optimizations to have a better performance. 
+
+
+=item B<BOX64_PYTHON3> =I<XXXX>
+
+Path to a python3 executable.
+
+ * XXXX : Use python3 executable at path XXXX. 
+
+
 =item B<BOX64_RCFILE> =I<XXXX>

 Path to the rc file to load.
@@ -643,6 +803,14 @@ Use hardware counter for rdtsc if available.
 * 1 : Use hardware counter for rdtsc if and only if precision is at least 1GHz. 


+=item B<BOX64_RDTSC_INV> =I<0|1>
+
+Report Invariant TSC bit in CPUID for Intel CPU type.
+
+ * 0 : Do not report Invariant TSC for Intel CPU type (default). [Default]
+ * 1 : Report Invariant TSC bit in CPUID leaf 0x80000007 for Intel CPU type. 
+
+
 =item B<BOX64_RESERVE_HIGH> =I<0|1>

 Reserve high memory area for the program, always enabled on Box32.
@@ -670,7 +838,7 @@ Use a workaround for SDL_GetJoystickGUIDInfo function for wrapped SDL2.

 =item B<BOX64_SHAEXT> =I<0|1>

-Expose SHAEXT (a.k.a. SHA_NI) capabilities.
+Expose SHAEXT (a.k.a. SHA_NI) capabilities. Available in WowBox64.

 * 0 : Do not expose SHAEXT capabilities. 
 * 1 : Expose SHAEXT capabilities. [Default]
@@ -694,20 +862,28 @@ Always show SIGSEGV signal detailes.

 =item B<BOX64_SSE_FLUSHTO0> =I<0|1>

-Behaviour of SSE Flush to 0 flags. Availble in WowBox64.
+Behaviour of SSE Flush to 0 flags. Also tracking SSE exception flags. Available in WowBox64.

 * 0 : Just track the flag. [Default]
- * 1 : Apply SSE Flush to 0 flag directly. 
+ * 1 : Apply SSE Flush to 0 flag directly. Also reflect SSE exception flags in Dynarec (if available) 


 =item B<BOX64_SSE42> =I<0|1>

-Expose SSE4.2 capabilities. Availble in WowBox64.
+Expose SSE4.2 capabilities. Available in WowBox64.

 * 0 : Do not expose SSE4.2 capabilities. 
 * 1 : Expose SSE4.2 capabilities. [Default]


+=item B<BOX64_STEAM_VULKAN> =I<0|1>
+
+Force Vulkan for Linux Steam UI.
+
+ * 0 : Do nothing. [Default]
+ * 1 : Force Vulkan for Linux Steam UI. 
+
+
 =item B<BOX64_SYNC_ROUNDING> =I<0|1>

 Synchronize rounding mode between x86 and native.
@@ -732,21 +908,24 @@ Enable or disable EMM (i.e. MMX) trace output.
 * 1 : Enable EMM trace output. 


-=item B<BOX64_TRACE_FILE> =I<XXXX|XXXX%pid|stderr>
+=item B<BOX64_TRACE_FILE> =I<XXXX|XXXX%pid|stderr|stdout>

 Send all log and trace to a file instead of `stdout`.

 * XXXX : Send all log and trace to file XXXX. 
 * XXXX%pid : Send all log and trace to file XXXX with pid appended. 
- * stderr : Send all log and trace to `stderr`. 
+ * stderr : Send all log and trace to `stderr`. [Default]
+ * stdout : Send all log and trace to `stdout`. 


-=item B<BOX64_TRACE_INIT> =I<0|1>
+=item B<BOX64_TRACE_INIT> =I<0|1|symbolname|0xXXXXXXX-0xYYYYYYY>

 Same as BOX64_TRACE but starts the trace immediately.

 * 0 : Disable trace output. [Default]
 * 1 : Enable trace output. Trace starts before the initialization of dependencies. 
+ * symbolname : Enable trace output for `symbolname` only. Trace starts before the initialization of dependencies. 
+ * 0xXXXXXXX-0xYYYYYYY : Enable trace output for the range of address (inclusive-exclusive). Trace starts before the initialization of dependencies. 


 =item B<BOX64_TRACE_START> =I<0|1|XXXX>
@@ -781,7 +960,15 @@ Only available on box64 build with trace. Adds trace of all instructions execute
 Detect UnityPlayer and apply conservative settings.

 * 0 : Does nothing. 
- * 1 : Detect UnityPlayer, and apply BOX64_DYNAREC_STRONGMEM=1 when detected. [Default]
+ * 1 : Detect UnityPlayer (Windows or Linux), and apply BOX64_UNITY=1 when detected. [Default]
+
+
+=item B<BOX64_UNITY> =I<0|1>
+
+Games is a Unity one.
+
+ * 0 : Does nothing. 
+ * 1 : Unity Game, use special detecting code for Windows, apply BOX64_DYNAREC_STRONGMEM=1 for Linux. [Default]


 =item B<BOX64_WRAP_EGL> =I<0|1>
@@ -818,7 +1005,7 @@ Call XInitThreads when loading X11. This is mostly for old Loki games with the L

 =item B<BOX64_X87_NO80BITS> =I<0|1>

-Behavoiur of x87 80bits long double. Availble in WowBox64.
+Behavoiur of x87 80bits long double. Available in WowBox64.

 * 0 : Try to handle 80bits long double as precise as possible. [Default]
 * 1 : Use 64bits double for x87. 
@@ -63,10 +63,12 @@ This configuration will apply the specified settings application-wide to any exe
    for entry in data:
      categories[entry["category"]].append(entry)

-    for category, entries in categories.items():
+    # Put "Performance" at the top, then sort the rest alphabetically
+    sorted_categories = sorted(categories.items(), key=lambda x: (x[0] != "Performance", x[0]))
+    for category, entries in sorted_categories:
      md_file.write(f"## {category}\n\n")
      for entry in entries:
-        md_file.write(f"### {entry['name']}\n\n{entry['description']}{' Availble in WowBox64.' if entry['wine'] else ''}\n\n")
+        md_file.write(f"### {entry['name']}\n\n{entry['description']}{' Available in WowBox64.' if entry['wine'] else ''}\n\n")
        for option in entry['options']:
          md_file.write(f" * {option['key']}: {option['description']} {'[Default]' if option['default'] else ''}\n")
        md_file.write("\n")
@@ -126,7 +128,7 @@ Example:
 """)

    for entry in data:
-      pod_file.write(f"\n=item B<{entry['name']}> =I<{ '|'.join(option['key'] for option in entry['options']) }>\n\n{entry['description']}{' Availble in WowBox64.' if entry['wine'] else ''}\n\n")
+      pod_file.write(f"\n=item B<{entry['name']}> =I<{ '|'.join(option['key'] for option in entry['options']) }>\n\n{entry['description']}{' Available in WowBox64.' if entry['wine'] else ''}\n\n")
      for option in entry['options']:
        pod_file.write(f" * {option['key']} : {option['description']} {'[Default]' if option['default'] else ''}\n")
      pod_file.write("\n")
@@ -17,6 +17,24 @@
      }
    ]
  },
+  {
+    "name": "BOX64_AES",
+    "description": "Expose AES capabilities.",
+    "category": "Compatibility",
+    "wine": true,
+    "options": [
+      {
+        "key": "0",
+        "description": "Do not expose AES capabilities.",
+        "default": false
+      },
+      {
+        "key": "1",
+        "description": "Expose AES capabilities.",
+        "default": true
+      }
+    ]
+  },
  {
    "name": "BOX64_ALLOWMISSINGLIBS",
    "description": "Allow missing libraries to be ignored.",
@@ -35,6 +53,34 @@
      }
    ]
  },
+  {
+    "name": "BOX64_ARCH",
+    "description": "Specify the architecture in rcfiles to which the current entry applies. This option only takes effect when used inside rcfiles.",
+    "category": "Compatibility",
+    "wine": false,
+    "options": [
+      {
+        "key": "`arm64`",
+        "description": "Only apply current entry for AArch64.",
+        "default": false
+      },
+      {
+        "key": "`la64`",
+        "description": "Only apply current entry for LoongArch64.",
+        "default": false
+      },
+      {
+        "key": "`rv64`",
+        "description": "Only apply current entry for RISC-V.",
+        "default": false
+      },
+      {
+        "key": "`<empty>`",
+        "description": "Apply current entry for all architectures, will be replaced by more specific entries.",
+        "default": true
+      }
+    ]
+  },
  {
    "name": "BOX64_ARGS",
    "description": "Arguments to pass to the guest program, only valid if there is no existing arguments.",
@@ -89,42 +135,6 @@
      }
    ]
  },
-  {
-    "name": "BOX64_CEFDISABLEGPU",
-    "description": "Add -cef-disable-gpu argument to the guest program.",
-    "category": "Arguments",
-    "wine": false,
-    "options": [
-      {
-        "key": "0",
-        "description": "Does nothing.",
-        "default": true
-      },
-      {
-        "key": "1",
-        "description": "Add -cef-disable-gpu argument to the guest program.",
-        "default": false
-      }
-    ]
-  },
-  {
-    "name": "BOX64_CEFDISABLEGPUCOMPOSITOR",
-    "description": "Add -cef-disable-gpu-compositor argument to the guest program.",
-    "category": "Arguments",
-    "wine": false,
-    "options": [
-      {
-        "key": "0",
-        "description": "Does nothing.",
-        "default": true
-      },
-      {
-        "key": "1",
-        "description": "Add -cef-disable-gpu-compositor argument to the guest program.",
-        "default": false
-      }
-    ]
-  },
  {
    "name": "BOX64_CPUTYPE",
    "description": "Specify the CPU type to emulate.",
@@ -151,7 +161,7 @@
    "options": [
      {
        "key": "0",
-        "description": "Use emulated crashhandler.so library if nedded.",
+        "description": "Use emulated crashhandler.so library if needed.",
        "default": false
      },
      {
@@ -228,7 +238,7 @@
      },
      {
        "key": "1",
-        "description": "Generate aligned atomics only, which is faster and smaller code size, but will cause SIGBUS for LOCK prefixed opcodes operating on aligned data addresses.",
+        "description": "Generate aligned atomics only, which is faster and smaller code size, but will cause SIGBUS for LOCK prefixed opcodes operating on unaligned data addresses.",
        "default": false
      }
    ]
@@ -263,7 +273,7 @@
  },
  {
    "name": "BOX64_DYNAREC_BLEEDING_EDGE",
-    "description": "Detect MonoBleedingEdge and apply conservative settings.",
+    "description": "Detect MonoBleedingEdge and apply conservative settings (Linux only).",
    "category": "Compatibility",
    "wine": false,
    "options": [
@@ -302,6 +312,29 @@
      }
    ]
  },
+  {
+    "name": "BOX64_DYNAREC_SEP",
+    "description": "Optimize CALL/RET further with Secondary Entry Points (SEP). Have no effect is CALLRET is not enabled",
+    "category": "Performance",
+    "wine": true,
+    "options": [
+      {
+        "key": "0",
+        "description": "Do not add SEP at CALLRET return",
+        "default": false
+      },
+      {
+        "key": "1",
+        "description": "Add SEP at CALLRET return, on meory that are from known binary files",
+        "default": true
+      },
+      {
+        "key": "2",
+        "description": "Add SEP at CALLRET return for all type of memory.",
+        "default": false
+      }
+    ]
+  },
  {
    "name": "BOX64_DYNAREC_DF",
    "description": "Enable or disable the use of deferred flags.",
@@ -343,6 +376,42 @@
      }
    ]
  },
+  {
+    "name": "BOX64_DYNAREC_NOHOTPAGE",
+    "description": "Disable detection of hot page (where code is executed and data written at the same time).",
+    "category": "Performance",
+    "wine": false,
+    "options": [
+      {
+        "key": "0",
+        "description": "Detect hot page",
+        "default": true
+      },
+      {
+        "key": "1",
+        "description": "Do not detect hot page",
+        "default": false
+      }
+    ]
+  },
+  {
+    "name": "BOX64_DYNAREC_HOTPAGE_ALT",
+    "description": "Use alternative way to detect HotPages (legacy way).",
+    "category": "Performance",
+    "wine": false,
+    "options": [
+      {
+        "key": "0",
+        "description": "Detect hot page, using new method from v0.3.9+",
+        "default": true
+      },
+      {
+        "key": "1",
+        "description": "Use alternate method to detect HotPage (from v0.3.8 and earlier)",
+        "default": false
+      }
+    ]
+  },
  {
    "name": "BOX64_DYNAREC_DIV0",
    "description": "Enable or disable the generation of division-by-zero exception.",
@@ -351,7 +420,7 @@
    "options": [
      {
        "key": "0",
-        "description": "Do not generate thr division-by-zero exception.",
+        "description": "Do not generate the division-by-zero exception.",
        "default": true
      },
      {
@@ -563,6 +632,29 @@
      }
    ]
  },
+  {
+    "name": "BOX64_DYNAREC_NOHOSTEXT",
+    "description": "Disable optional host extensions.",
+    "category": "Debugging",
+    "wine": false,
+    "options": [
+      {
+        "key": "0",
+        "description": "Disable nothing.",
+        "default": true
+      },
+      {
+        "key": "1",
+        "description": "Disable all the optional extensions.",
+        "default": false
+      },
+      {
+        "key": "xxx,yyy,zzz",
+        "description": "Disable selected optional extensions, choices are Arm: `crc32,pmull,aes,atomics,sha1,sha2,uscat,flagm,flagm2,frintts,afp,rndr`, LoongArch: `lasx,lbt,frecipe,lam_bh,lamcas,scq`, and RISC-V: `zba,zbb,zbc,zbs,vector,xtheadba,xtheadbb,xtheadbs,xtheadmemidx,xtheadmempair,xtheadcondmov`.",
+        "default": false
+      }
+    ]
+  },
  {
    "name": "BOX64_DYNAREC_PAUSE",
    "description": "Enable x86 PAUSE emulation, may help the performance of spinlocks.",
@@ -609,9 +701,45 @@
      }
    ]
  },
+  {
+    "name": "BOX64_DYNAREC_PURGE",
+    "description": "Purge dynablocks that haven't been executed recently, based on tick age threshold (BOX64_DYNAREC_PURGE_AGE).",
+    "category": "Performance",
+    "wine": false,
+    "options": [
+      {
+        "key": "0",
+        "description": "Dynarec will not purge old unused dynablocks.",
+        "default": true
+      },
+      {
+        "key": "1",
+        "description": "Dynarec will purge old unused dynablocks based on tick age.",
+        "default": false
+      }
+    ]
+  },
+  {
+    "name": "BOX64_DYNAREC_PURGE_AGE",
+    "description": "Tick age threshold for dynablock purging (requires BOX64_DYNAREC_PURGE=1).",
+    "category": "Performance",
+    "wine": false,
+    "options": [
+      {
+        "key": "4096",
+        "description": "Default age threshold.",
+        "default": true
+      },
+      {
+        "key": "XXXX",
+        "description": "Custom age threshold (range: 10-65536).",
+        "default": false
+      }
+    ]
+  },
  {
    "name": "BOX64_DYNAREC_SAFEFLAGS",
-    "description": "Behaviour of flags emulation on CALL/RET opcodes.",
+    "description": "Behaviour of flags emulation on CALL/RET opcodes and other edge cases.",
    "category": "Performance",
    "wine": true,
    "options": [
@@ -627,7 +755,7 @@
      },
      {
        "key": "2",
-        "description": "All CALL/RET will need flags.",
+        "description": "All CALL/RET will need flags. All edge cases are also handled.",
        "default": false
      }
    ]
@@ -657,6 +785,11 @@
        "key": "3",
        "description": "All in 2, plus more memory barriers on a regular basis.",
        "default": false
+      },
+      {
+        "key": "4",
+        "description": "Mimic x86 TSO similarly to QEMU's approach, for evaluation purposes.",
+        "default": false
      }
    ]
  },
@@ -701,6 +834,42 @@
      }
    ]
  },
+  {
+    "name": "BOX64_DYNAREC_TEST_NODUP",
+    "description": "Enable deduplication of IP address where COSIM find a difference.",
+    "category": "Debugging",
+    "wine": true,
+    "options": [
+      {
+        "key": "0",
+        "description": "Show all addresses in error.",
+        "default": true
+      },
+      {
+        "key": "1",
+        "description": "Show address in error only 1 time, even if dump might differs.",
+        "default": false
+      }
+    ]
+  },
+  {
+    "name": "BOX64_DYNAREC_TEST_NODUMP",
+    "description": "Do not dump test code when BOX64_DYNAREC_DUMP is enabled.",
+    "category": "Debugging",
+    "wine": true,
+    "options": [
+      {
+        "key": "0",
+        "description": "Dump test code.",
+        "default": false
+      },
+      {
+        "key": "1",
+        "description": "Do not dump test code.",
+        "default": true
+      }
+    ]
+  },
  {
    "name": "BOX64_DYNAREC_TRACE",
    "description": "Enable or disable DynaRec trace.",
@@ -801,6 +970,78 @@
      }
    ]
  },
+  {
+    "name": "BOX64_DYNAREC_INTERP_SIGNAL",
+    "description": "Allow the use of Dynarec in Signal Handlers.",
+    "category": "Compatibility",
+    "wine": false,
+    "options": [
+      {
+        "key": "0",
+        "description": "Allow the use of Dynarec in signal handlers.",
+        "default": true
+      },
+      {
+        "key": "1",
+        "description": "Only use the Interpreter for signal handlers (as it was in v0.3.8 and earlier).",
+        "default": false
+      }
+    ]
+  },
+  {
+    "name": "BOX64_DYNACACHE",
+    "description": "Enable/disable the Dynamic Recompiler Cache (a.k.a DynaCache). This option defaults to 2 (to read cache if present but not generate any). DynaCache write file to home folder by default, and can grow without limit.",
+    "category": "Performance",
+    "wine": false,
+    "options": [
+      {
+        "key": "0",
+        "description": "Disable DynaCache.",
+        "default": false
+      },
+      {
+        "key": "1",
+        "description": "Enable DynaCache.",
+        "default": false
+      },
+      {
+        "key": "2",
+        "description": "Use DynaCache files if present, but do not generate new one.",
+        "default": true
+      }
+    ]
+  },
+  {
+    "name": "BOX64_DYNACACHE_FOLDER",
+    "description": "Set the folder for DynaCache files. Default is $XDG_CACHE_HOME/box64 or $HOME/.cache/box64 if $XDG_CACHE_HOME is not set.",
+    "category": "Performance",
+    "wine": false,
+    "options": [
+      {
+        "key": "XXXX",
+        "description": "Use folder XXXX for DynaCache files.",
+        "default": false
+      }
+    ]
+  },
+  {
+    "name": "BOX64_DYNACACHE_MIN",
+    "description": "Minimum size, in KB, for a DynaCache to be written to disk. Default size is 350KB",
+    "category": "Performance",
+    "wine": false,
+    "options": [
+      {
+        "key": "XXXX",
+        "description": "Set a minimum size of XXXX KB of Dynarec code to write the dynacache to disk. Will not be saved to disk else.",
+        "default": false
+      },
+      {
+        "key": "350",
+        "description": "A size of 350 KB is the default value.",
+        "default": true
+      }
+    ]
+  },
  {
    "name": "BOX64_EMULATED_LIBS",
    "description": "Force the use of emulated libraries.",
@@ -881,6 +1122,24 @@
      }
    ]
  },
+  {
+    "name": "BOX64_FORCE_LD_PRELOAD",
+    "description": "Force symbols from BOX64_LD_PRELOAD libraries to be resolved by RTLD_NEXT.",
+    "category": "Compatibility",
+    "wine": false,
+    "options": [
+      {
+        "key": "0",
+        "description": "Do nothing.",
+        "default": true
+      },
+      {
+        "key": "1",
+        "description": "Force symbols from BOX64_LD_PRELOAD libraries to be resolved by RTLD_NEXT.",
+        "default": false
+      }
+    ]
+  },
  {
    "name": "BOX64_IGNOREINT3",
    "description": "Ignore INT3 instructions.",
@@ -1021,12 +1280,12 @@
      {
        "key": "0",
        "description": "Does nothing.",
-        "default": false
+        "default": true
      },
      {
        "key": "1",
        "description": "Detect libcef, and apply BOX64_MALLOC_HACK=2 if detected.",
-        "default": true
+        "default": false
      }
    ]
  },
@@ -1179,6 +1438,47 @@
      }
    ]
  },
+  {
+    "name": "BOX64_NODYNAREC_DELAY",
+    "description": "Remove NoDynarec range after a delay.",
+    "category": "Debugging",
+    "wine": true,
+    "options": [
+      {
+        "key": "0",
+        "description": "Do not remove NoDynarec range.",
+        "default": true
+      },
+      {
+        "key": "1",
+        "description": "Remove the NoDynarec range once 2 threads have been created.",
+        "default": false
+      }
+    ]
+  },
+  {
+    "name": "BOX64_DYNAREC_NOARCH",
+    "description": "Remove Architecture metatdata from Dynarec blocks (used on signal).",
+    "category": "Performance",
+    "wine": true,
+    "options": [
+      {
+        "key": "0",
+        "description": "Generate all architecture metadata for each blocks, including x87/mmx/SSE/AVX regs tracking. Wil use some more memory.",
+        "default": true
+      },
+      {
+        "key": "1",
+        "description": "Generate only essential metadata for each blocks (flags, alignments), saving some memory, but some encrypted software might fails.",
+        "default": false
+      },
+      {
+        "key": "2",
+        "description": "Do not generate metadata for blocks. Saving a bit more memory but might fails on signal handling.",
+        "default": false
+      }
+    ]
+  },
  {
    "name": "BOX64_NOGTK",
    "description": "Do not load wrapped GTK libraries.",
@@ -1305,6 +1605,24 @@
      }
    ]
  },
+  {
+    "name": "BOX64_NOVULKANOVERLAY",
+    "description": "Do not load x64 Vulkan overlay.",
+    "category": "Libraries",
+    "wine": false,
+    "options": [
+      {
+        "key": "0",
+        "description": "Do try to Load x64 Vulkan overlay.",
+        "default": true
+      },
+      {
+        "key": "1",
+        "description": "Do not load x64 Vulkan overlay.",
+        "default": false
+      }
+    ]
+  },
  {
    "name": "BOX64_PATH",
    "description": "Path to look for x86_64 binaries.",
@@ -1318,6 +1636,24 @@
      }
    ]
  },
+  {
+    "name": "BOX64_PCLMULQDQ",
+    "description": "Expose PCLMULQDQ capabilities.",
+    "category": "Compatibility",
+    "wine": true,
+    "options": [
+      {
+        "key": "0",
+        "description": "Do not expose PCLMULQDQ capabilities.",
+        "default": false
+      },
+      {
+        "key": "1",
+        "description": "Expose PCLMULQDQ capabilities.",
+        "default": true
+      }
+    ]
+  },
  {
    "name": "BOX64_PREFER_EMULATED",
    "description": "Prefer emulated libraries over native ones.",
@@ -1354,6 +1690,52 @@
      }
    ]
  },
+  {
+    "name": "BOX64_PROFILE",
+    "description": "Predefined sets of environment variables with compatibility or performance in mind.",
+    "category": "Compatibility",
+    "wine": true,
+    "options": [
+      {
+        "key": "safest",
+        "description": "Profile with all the unsafe DynaRec optimizations disabled.",
+        "default": false
+      },
+      {
+        "key": "safe",
+        "description": "Well, less safer than safest.",
+        "default": false
+      },
+      {
+        "key": "default",
+        "description": "The default settings that most programs runs fine with a handy performance.",
+        "default": true
+      },
+      {
+        "key": "fast",
+        "description": "Enable many unsafe optimizations, but also enable strongmem emulation.",
+        "default": false
+      },
+      {
+        "key": "fastest",
+        "description": "Enable many unsafe optimizations to have a better performance.",
+        "default": false
+      }
+    ]
+  },
+  {
+    "name": "BOX64_PYTHON3",
+    "description": "Path to a python3 executable.",
+    "category": "Compatibility",
+    "wine": false,
+    "options": [
+      {
+        "key": "XXXX",
+        "description": "Use python3 executable at path XXXX.",
+        "default": false
+      }
+    ]
+  },
  {
    "name": "BOX64_RCFILE",
    "description": "Path to the rc file to load.",
@@ -1385,6 +1767,24 @@
      }
    ]
  },
+  {
+    "name": "BOX64_RDTSC_INV",
+    "description": "Report Invariant TSC bit in CPUID for Intel CPU type.",
+    "category": "Compatibility",
+    "wine": false,
+    "options": [
+      {
+        "key": "0",
+        "description": "Do not report Invariant TSC for Intel CPU type (default).",
+        "default": true
+      },
+      {
+        "key": "1",
+        "description": "Report Invariant TSC bit in CPUID leaf 0x80000007 for Intel CPU type.",
+        "default": false
+      }
+    ]
+  },
  {
    "name": "BOX64_RESERVE_HIGH",
    "description": "Reserve high memory area for the program, always enabled on Box32.",
@@ -1448,7 +1848,7 @@
    "name": "BOX64_SHAEXT",
    "description": "Expose SHAEXT (a.k.a. SHA_NI) capabilities.",
    "category": "Compatibility",
-    "wine": false,
+    "wine": true,
    "options": [
      {
        "key": "0",
@@ -1500,7 +1900,7 @@
  },
  {
    "name": "BOX64_SSE_FLUSHTO0",
-    "description": "Behaviour of SSE Flush to 0 flags.",
+    "description": "Behaviour of SSE Flush to 0 flags. Also tracking SSE exception flags.",
    "category": "Compatibility",
    "wine": true,
    "options": [
@@ -1511,7 +1911,7 @@
      },
      {
        "key": "1",
-        "description": "Apply SSE Flush to 0 flag directly.",
+        "description": "Apply SSE Flush to 0 flag directly. Also reflect SSE exception flags in Dynarec (if available)",
        "default": false
      }
    ]
@@ -1534,6 +1934,24 @@
      }
    ]
  },
+  {
+    "name": "BOX64_STEAM_VULKAN",
+    "description": "Force Vulkan for Linux Steam UI.",
+    "category": "Compatibility",
+    "wine": false,
+    "options": [
+      {
+        "key": "0",
+        "description": "Do nothing.",
+        "default": true
+      },
+      {
+        "key": "1",
+        "description": "Force Vulkan for Linux Steam UI.",
+        "default": false
+      }
+    ]
+  },
  {
    "name": "BOX64_SYNC_ROUNDING",
    "description": "Synchronize rounding mode between x86 and native.",
@@ -1607,6 +2025,11 @@
      {
        "key": "stderr",
        "description": "Send all log and trace to `stderr`.",
+        "default": true
+      },
+      {
+        "key": "stdout",
+        "description": "Send all log and trace to `stdout`.",
        "default": false
      }
    ]
@@ -1626,6 +2049,16 @@
        "key": "1",
        "description": "Enable trace output. Trace starts before the initialization of dependencies.",
        "default": false
+      },
+      {
+        "key": "symbolname",
+        "description": "Enable trace output for `symbolname` only. Trace starts before the initialization of dependencies.",
+        "default": false
+      },
+      {
+        "key": "0xXXXXXXX-0xYYYYYYY",
+        "description": "Enable trace output for the range of address (inclusive-exclusive). Trace starts before the initialization of dependencies.",
+        "default": false
      }
    ]
  },
@@ -1711,7 +2144,25 @@
      },
      {
        "key": "1",
-        "description": "Detect UnityPlayer, and apply BOX64_DYNAREC_STRONGMEM=1 when detected.",
+        "description": "Detect UnityPlayer (Windows or Linux), and apply BOX64_UNITY=1 when detected.",
+        "default": true
+      }
+    ]
+  },
+  {
+    "name": "BOX64_UNITY",
+    "description": "Games is a Unity one.",
+    "category": "Compatibility",
+    "wine": false,
+    "options": [
+      {
+        "key": "0",
+        "description": "Does nothing.",
+        "default": false
+      },
+      {
+        "key": "1",
+        "description": "Unity Game, use special detecting code for Windows, apply BOX64_DYNAREC_STRONGMEM=1 for Linux.",
        "default": true
      }
    ]
@@ -1,26 +1,23 @@
 #!/bin/bash

 # create necessary directories
-mkdir -p ~/steam
-mkdir -p ~/steam/tmp
+mkdir -p ~/steam/tmp # makes both ~/steam and ~/steam/tmp in one cmd
 cd ~/steam/tmp

 # download latest deb and unpack
-wget https://cdn.cloudflare.steamstatic.com/client/installer/steam.deb
+curl -LO https://cdn.cloudflare.steamstatic.com/client/installer/steam.deb
 ar x steam.deb
 tar xf data.tar.xz

-# remove deb archives, not needed anymore
-rm ./*.tar.xz ./steam.deb
-
 # move deb contents to steam folder
-mv ./usr/* ../
-cd ../ && rm -rf ./tmp/
+mv ./usr/* ..
+cd .. && rm -rf ./tmp

 # create run script
 echo '#!/bin/bash
 export STEAMOS=1
 export STEAM_RUNTIME=1
+export PROTON_USE_WOW64=1
 export DBUS_FATAL_WARNINGS=0
 ~/steam/bin/steam $@' > steam

@@ -403,15 +403,23 @@ def readFiles(files: Iterable[Filename]) -> Tuple[JumbledGlobals, JumbledRedirec
 							 or match("sdl1mixer", "sdl2mixer") \
 							 or match("sdl1net", "sdl2net") \
 							 or match("sdl1ttf", "sdl2ttf") \
+							 or match("sdl1ttf", "sdl3ttf") \
+							 or match("sdl2ttf", "sdl3ttf") \
 							 or match("libgl", "libegl") \
 							 or match("libgl", "glesv2") \
+							 or match("libgl", "libglxnvidia") \
+							 or match("libglx", "libglxnvidia") \
 							 or match("libegl", "glesv2") \
+							 or match("libegl", "libglxnvidia") \
+							 or match("glesv2", "libglxnvidia") \
 							 or match("softokn3", "p11kit") \
 							 or match("libc", "tcmallocminimal") \
 							 or match("libc", "tbbmallocproxy") \
 							 or match("libc", "androidshmem") \
 							 or match("crypto", "libssl3") \
-							 or match("tcmallocminimal", "tbbmallocproxy"):
+							 or match("tcmallocminimal", "tbbmallocproxy") \
+							 or match("iconv", "libunistring2") \
+							 or match("gomp", "libomp"):
 								continue
 							
 							# Note: this test is very (too) simple. If it ever raises, comment
@@ -841,17 +849,17 @@ def main(root: str, files: Iterable[Filename], ver: str):
 	allowed_conv = conventions[allowed_conv_ident]
 	
 	# H could be allowed maybe?
-	allowed_simply: Dict[str, str] = {"ARM64": "v", "RV64": "v"}
-	allowed_regs  : Dict[str, str] = {"ARM64": "cCwWiuIUlLp", "RV64": "CWIUlLp"}
-	allowed_fpr   : Dict[str, str] = {"ARM64": "fd", "RV64": "fd"}
-	allowed_sextw : Dict[str, str] = {"ARM64": "", "RV64": "cwiu"}
+	allowed_simply: Dict[str, str] = {"ARM64": "v", "RV64": "v", "LA64": "v"}
+	allowed_regs  : Dict[str, str] = {"ARM64": "cCwWiuIUlLp", "RV64": "CWIUlLp", "LA64": "CWIUlLp"}
+	allowed_fpr   : Dict[str, str] = {"ARM64": "fd", "RV64": "fd", "LA64": "fd"}
+	allowed_sextw : Dict[str, str] = {"ARM64": "", "RV64": "cwiu", "LA64": "cwiu"}
 	
 	# Detect functions which return in an x87 register
 	retx87_wraps: Dict[ClausesStr, List[FunctionType]] = {}
 	return_x87: str = "D"
 	
 	# Sanity checks
-	forbidden_simple: Dict[str, str] = {"ARM64": "EDVOSNHPAxXYb", "RV64": "EDVOSNHPAxXYb"}
+	forbidden_simple: Dict[str, str] = {"ARM64": "EDVOSNHPAxXYb", "RV64": "EDVOSNHPAxXYb", "LA64": "EDVOSNHPAxXYb"}
 	assert(all(k in allowed_simply for k in forbidden_simple))
 	assert(all(k in allowed_regs for k in forbidden_simple))
 	assert(all(k in allowed_fpr for k in forbidden_simple))
@@ -1176,7 +1184,7 @@ def main(root: str, files: Iterable[Filename], ver: str):
 				"fn({0});",                                                # v
 				"R_RAX=(uint8_t)fn({0});",                                 # c
 				"R_RAX=(uint16_t)fn({0});",                                # w
-				"R_RAX=(uint32_t)fn({0});",                                # i
+				"R_RAX=(int)fn({0});",                                	   # i
 				"S_RAX=(int64_t)fn({0});",                                 # I
 				"R_RAX=(unsigned char)fn({0});",                           # C
 				"R_RAX=(unsigned short)fn({0});",                          # W
@@ -1205,8 +1213,8 @@ def main(root: str, files: Iterable[Filename], ver: str):
 				"fn({0});",                                                # v
 				"R_RAX=fn({0});",                                          # c
 				"R_RAX=fn({0});",                                          # w
-				"R_RAX=(int32_t)fn({0});",                                 # i
-				"R_RAX=(int64_t)fn({0});",                                 # I
+				"R_RAX=(int)fn({0});",                                     # i
+				"S_RAX=(int64_t)fn({0});",                                 # I
 				"R_RAX=(unsigned char)fn({0});",                           # C
 				"R_RAX=(unsigned short)fn({0});",                          # W
 				"R_RAX=(uint32_t)fn({0});",                                # u
@@ -876,6 +876,7 @@ def readFiles(files: Iterable[str]) -> Tuple[JumbledFunctions, JumbledFunctions,
 							 or match("appindicator","appindicator3")\
 							 or match("libgl",        "libegl")     \
 							 or match("libgl",        "libglx")     \
+							 or match("libgl",        "libglxnvidia")\
 							 \
 							 or match("libc",        "tcmallocminimal") \
 							 or match("libc",        "ldlinux") 	\
@@ -18,7 +18,7 @@ KHASH_MAP_INIT_STR(strings, ptr_t);
 static kh_from_t*   hash_from = NULL;
 static kh_to_t*     hash_to = NULL;
 #define HASH_MSK    0xf000000f
-#define HASH_VAL    0x00ffffff
+#define HASH_VAL    0x007fffff
 #define HASH_SHIFT  4
 static uint32_t     hash_cnt = 1;
 static pthread_rwlock_t hash_lock = {0};
@@ -6,6 +6,7 @@
 #include <signal.h>
 #include <sys/mman.h>
 #include <pthread.h>
+#include <inttypes.h>

 #include "os.h"
 #include "box64context.h"
@@ -72,17 +73,6 @@ int getrand(int maxval)

 }

-void free_tlsdatasize(void* p)
-{
-    if(!p)
-        return;
-    tlsdatasize_t *data = (tlsdatasize_t*)p;
-    actual_free(data->ptr);
-    actual_free(p);
-    if(my_context)
-        pthread_setspecific(my_context->tlskey, NULL);
-}
-
 void x64Syscall(x64emu_t *emu);
 void x86Syscall(x64emu_t *emu);

@@ -94,14 +84,14 @@ void relockMutex(int locks)
            mutex_trylock(&A);          \

    GO(my_context->mutex_trace, 7)
-    #ifdef DYNAREC
    GO(my_context->mutex_dyndump, 8)
-    #else
-    GO(my_context->mutex_lock, 8)
-    #endif
    GO(my_context->mutex_tls, 9)
    GO(my_context->mutex_thread, 10)
    GO(my_context->mutex_bridge, 11)
+    #ifdef DYNAREC
+    #else
+    GO(my_context->mutex_lock, 12)
+    #endif
    #undef GO
 }

@@ -116,6 +106,7 @@ void init_mutexes(box64context_t* context)
    pthread_mutex_init(&context->mutex_tls, &attr);
    pthread_mutex_init(&context->mutex_thread, &attr);
    pthread_mutex_init(&context->mutex_bridge, &attr);
+    pthread_mutex_init(&context->mutex_dyndump, &attr);

    pthread_mutexattr_destroy(&attr);
 #else
@@ -146,28 +137,25 @@ static void atfork_child_box64context(void)
    init_mutexes(my_context);
 }

+int box64_cycle_log_initialized = 0;
+
 void freeCycleLog(box64context_t* ctx)
 {
-    if(BOX64ENV(rolling_log)) {
-        for(int i=0; i<BOX64ENV(rolling_log); ++i) {
-            box_free(ctx->log_call[i]);
-            box_free(ctx->log_ret[i]);
-        }
+    if (BOX64ENV(rolling_log) && box64_cycle_log_initialized) {
        box_free(ctx->log_call);
        box_free(ctx->log_ret);
        ctx->log_call = NULL;
        ctx->log_ret = NULL;
+        box64_cycle_log_initialized = 0;
    }
 }
+
 void initCycleLog(box64context_t* context)
 {
    if(context && BOX64ENV(rolling_log)) {
-        context->log_call = (char**)box_calloc(BOX64ENV(rolling_log), sizeof(char*));
-        context->log_ret = (char**)box_calloc(BOX64ENV(rolling_log), sizeof(char*));
-        for(int i=0; i<BOX64ENV(rolling_log); ++i) {
-            context->log_call[i] = (char*)box_calloc(256, 1);
-            context->log_ret[i] = (char*)box_calloc(128, 1);
-        }
+        context->log_call = (char*)box_calloc(BOX64ENV(rolling_log), 256*sizeof(char));
+        context->log_ret = (char*)box_calloc(BOX64ENV(rolling_log), 128*sizeof(char));
+        box64_cycle_log_initialized = 1;
    }
 }

@@ -186,7 +174,6 @@ box64context_t *NewBox64Context(int argc)
    initCycleLog(context);

    context->deferredInit = 1;
-    context->sel_serial = 1;

    init_custommem_helper(context);

@@ -225,28 +212,25 @@ box64context_t *NewBox64Context(int argc)
    init_mutexes(context);
    pthread_atfork(NULL, NULL, atfork_child_box64context);

-    pthread_key_create(&context->tlskey, free_tlsdatasize);
-
-
    for (int i=0; i<8; ++i) context->canary[i] = 1 +  getrand(255);
    context->canary[getrand(4)] = 0;
    printf_log(LOG_DEBUG, "Setting up canary (for Stack protector) at FS:0x28, value:%08X\n", *(uint32_t*)context->canary);

    // init segments
    for(int i=0; i<16; i++) {
-        context->segtls[i].limit = (uintptr_t)-1LL;
+        context->seggdt[i].limit = (uintptr_t)-1LL;
    }
-    context->segtls[10].key_init = 0;    // 0x53 selector
-    context->segtls[10].present = 1;
-    context->segtls[8].key_init = 0;    // 0x43 selector
-    context->segtls[8].present = 1;
-    context->segtls[6].key_init = 0;    // 0x33 selector
-    context->segtls[6].present = 1;
-    context->segtls[5].key_init = 0;    // 0x2b selector
-    context->segtls[5].present = 1;
-    context->segtls[4].key_init = 0;    // 0x23 selector
-    context->segtls[4].present = 1;
-    context->segtls[4].is32bits = 1;
+    // 0x53 selector
+    context->seggdt[10].present = 1;
+    // 0x43 selector
+    context->seggdt[8].present = 1;
+    // 0x33 selector
+    context->seggdt[6].present = 1;
+    // 0x2b selector
+    context->seggdt[5].present = 1;
+    // 0x23 selector
+    context->seggdt[4].present = 1;
+    context->seggdt[4].is32bits = 1;

    context->globdata = NewMapSymbols();
    context->uniques = NewMapSymbols();
@@ -261,6 +245,42 @@ box64context_t *NewBox64Context(int argc)
 }

 void freeALProcWrapper(box64context_t* context);
+void freeCUDAProcWrapper(box64context_t* context);
+
+#ifdef DYNAREC
+//#define DYNAREC_HIST    // uncomment to print dynablock size histogram at exit
+#ifdef DYNAREC_HIST
+#define HIST_BUCKETS 12
+static const uintptr_t bucket_limits[HIST_BUCKETS] = {
+    8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192, (uintptr_t)-1
+};
+static const char* bucket_labels[HIST_BUCKETS] = {
+    "1-8", "9-16", "17-32", "33-64", "65-128", "129-256",
+    "257-512", "513-1024", "1025-2048", "2049-4096", "4097-8192", "8193+"
+};
+typedef struct {
+    uint64_t count[HIST_BUCKETS];
+    uint64_t total_blocks;
+    uint64_t total_bytes;
+    uintptr_t max_size;
+} hist_data_t;
+static void db_size_walk_cb(uintptr_t start, uintptr_t end, uint64_t data, void* userdata) {
+    (void)end;
+    hist_data_t* h = (hist_data_t*)userdata;
+    uintptr_t size = start;  // key = x64_size
+    uint64_t n = data;       // value = count of blocks
+    for (int i = 0; i < HIST_BUCKETS; i++) {
+        if (size <= bucket_limits[i]) {
+            h->count[i] += n;
+            break;
+        }
+    }
+    h->total_blocks += n;
+    h->total_bytes += size * n;
+    if (size > h->max_size) h->max_size = size;
+}
+#endif
+#endif
 EXPORTDYN
 void FreeBox64Context(box64context_t** context)
 {
@@ -272,6 +292,11 @@ void FreeBox64Context(box64context_t** context)

    box64context_t* ctx = *context;   // local copy to do the cleaning

+    if(ctx->video_mem) {
+        munmap(ctx->video_mem, 0x10000);
+        ctx->video_mem = NULL;
+    }
+
    //clean_current_emuthread();    // cleaning main thread seems a bad idea
    if(ctx->local_maplib)
        FreeLibrarian(&ctx->local_maplib, NULL);
@@ -324,12 +349,14 @@ void FreeBox64Context(box64context_t** context)
    box_free(ctx->fullpath);
    box_free(ctx->box64path);
    box_free(ctx->bashpath);
+    box_free(ctx->pythonpath);

    FreeBridge(&ctx->system);

    #ifndef STATICBUILD
    freeGLProcWrapper(ctx);
    freeALProcWrapper(ctx);
+    freeCUDAProcWrapper(ctx);
    #ifdef BOX32
    #endif
    #endif
@@ -338,12 +365,6 @@ void FreeBox64Context(box64context_t** context)
        box_free(ctx->stack_clone);


-    void* ptr;
-    if ((ptr = pthread_getspecific(ctx->tlskey)) != NULL) {
-        free_tlsdatasize(ptr);
-    }
-    pthread_key_delete(ctx->tlskey);
-
    if(ctx->tlsdata)
        box_free(ctx->tlsdata);

@@ -357,7 +378,21 @@ void FreeBox64Context(box64context_t** context)
    FreeMapSymbols(&ctx->uniques);

 #ifdef DYNAREC
-    //dynarec_log(LOG_INFO, "BOX64 Dynarec at exit: Max DB=%d, righter=%d\n", ctx->max_db_size, rb_get_righter(ctx->db_sizes));
+#ifdef DYNAREC_HIST
+    {
+        hist_data_t hist = {0};
+        rbtree_walk(ctx->db_sizes, db_size_walk_cb, &hist);
+        if(hist.total_blocks) {
+            printf_log(LOG_INFO, "BOX64 Dynarec block size histogram: %" PRIu64 " blocks, %" PRIu64 " total x86 bytes, max=%" PRIuPTR "\n",
+                hist.total_blocks, hist.total_bytes, hist.max_size);
+            for(int i = 0; i < HIST_BUCKETS; i++) {
+                if(hist.count[i])
+                    printf_log(LOG_INFO, "  %10s: %" PRIu64 " blocks (%.1f%%)\n",
+                        bucket_labels[i], hist.count[i], hist.count[i] * 100.0 / hist.total_blocks);
+            }
+        }
+    }
+#endif
    rbtree_delete(ctx->db_sizes);
 #endif

@@ -371,6 +406,7 @@ void FreeBox64Context(box64context_t** context)
    pthread_mutex_destroy(&ctx->mutex_tls);
    pthread_mutex_destroy(&ctx->mutex_thread);
    pthread_mutex_destroy(&ctx->mutex_bridge);
+    pthread_mutex_destroy(&ctx->mutex_dyndump);
 #endif

    freeCycleLog(ctx);
@@ -403,8 +439,6 @@ void RemoveElfHeader(box64context_t* ctx, elfheader_t* head) {
        /*if(tlsbase == -ctx->tlssize) {
            // not really correct, but will do for now
            ctx->tlssize -= GetTLSSize(head);
-            if(!(++ctx->sel_serial))
-                ++ctx->sel_serial;
        }*/
    }
    for(int i=0; i<ctx->elfsize; ++i)
@@ -421,12 +455,7 @@ int AddTLSPartition(box64context_t* context, int tlssize) {
    context->tlsdata = box_realloc(context->tlsdata, context->tlssize);
    memmove(context->tlsdata+tlssize, context->tlsdata, oldsize);   // move to the top, using memmove as regions will probably overlap
    memset(context->tlsdata, 0, tlssize);           // fill new space with 0 (not mandatory)
-    // clean GS segment for current emu
-    if(my_context) {
-        //ResetSegmentsCache(thread_get_emu());
-        if(!(++context->sel_serial))
-            ++context->sel_serial;
-    }
+    // clean GS segment for current emu?

    return -context->tlssize;   // negative offset
 }
@@ -2,7 +2,7 @@
 #define __BOX64_VERSION_H_

 #define BOX64_MAJOR 0
-#define BOX64_MINOR 3
-#define BOX64_REVISION 6
+#define BOX64_MINOR 4
+#define BOX64_REVISION 1

 #endif //__BOX64_VERSION_H_
@@ -9,13 +9,15 @@
 #define __BUILD_INFO_H__


-#if defined(DYNAREC) && (defined(ARM64) || defined(RV64) || defined(LA64))
+#if defined(DYNAREC) && (defined(ARM64) || defined(RV64) || defined(LA64) || defined(PPC64LE))
 #if defined(ARM64)
 #define ARCH_STR " arm64"
 #elif defined(RV64)
 #define ARCH_STR " riscv64"
 #elif defined(LA64)
 #define ARCH_STR " loongarch64"
+#elif defined(PPC64LE)
+#define ARCH_STR " ppc64le"
 #endif
 #else
 #define ARCH_STR ""
@@ -36,4 +38,8 @@ void PrintBox64Version(int prefix);
    BOX64_STR ARCH_STR          \
        " v" BOX64_BUILD_INFO_STR(BOX64_MAJOR) "." BOX64_BUILD_INFO_STR(BOX64_MINOR) "." BOX64_BUILD_INFO_STR(BOX64_REVISION) " " GITREV

+#define BOX64_BUILD_INFO_STRING_SHORT \
+    BOX64_STR                         \
+    " v" BOX64_BUILD_INFO_STR(BOX64_MAJOR) "." BOX64_BUILD_INFO_STR(BOX64_MINOR) "." BOX64_BUILD_INFO_STR(BOX64_REVISION)
+
 #endif //__BUILD_INFO_H__
@@ -15,11 +15,10 @@
 #include <sys/stat.h>
 #include <stdarg.h>
 #include <ctype.h>
-#ifdef DYNAREC
-#ifdef ARM64
-#include <linux/auxvec.h>
-#include <asm/hwcap.h>
+#ifdef BOX32
+#include <sys/personality.h>
 #endif
+#ifdef DYNAREC
 #include <sys/types.h>
 #include <sys/stat.h>
 #include <fcntl.h>
@@ -49,6 +48,8 @@
 #include "env.h"
 #include "cleanup.h"
 #include "freq.h"
+#include "hostext.h"
+#include "sysinfo.h"

 box64context_t *my_context = NULL;
 extern box64env_t box64env;
@@ -59,6 +60,7 @@ int box64_stdout_no_w = 0;
 uintptr_t box64_pagesize;
 path_collection_t box64_addlibs = {0};
 int box64_is32bits = 0;
+int box64_isAddressSpace32 = 0;
 int box64_rdtsc = 0;
 uint8_t box64_rdtsc_shift = 0;
 int box64_mapclean = 0;
@@ -66,6 +68,7 @@ int box64_zoom = 0;
 int box64_steam = 0;
 int box64_steamcmd = 0;
 int box64_musl = 0;
+int box64_nolibs = 0;
 char* box64_custom_gstreamer = NULL;
 int box64_tcmalloc_minimal = 0;
 uintptr_t fmod_smc_start = 0;
@@ -73,45 +76,11 @@ uintptr_t fmod_smc_end = 0;
 uint32_t default_gs = 0x53;
 uint32_t default_fs = 0;
 int box64_isglibc234 = 0;
+int box64_unittest_mode = 0;
+sysinfo_t box64_sysinfo = { 0 };

 #ifdef DYNAREC
-#ifdef ARM64
-int arm64_asimd = 0;
-int arm64_aes = 0;
-int arm64_pmull = 0;
-int arm64_crc32 = 0;
-int arm64_atomics = 0;
-int arm64_sha1 = 0;
-int arm64_sha2 = 0;
-int arm64_uscat = 0;
-int arm64_flagm = 0;
-int arm64_flagm2 = 0;
-int arm64_frintts = 0;
-int arm64_afp = 0;
-int arm64_rndr = 0;
-#elif defined(RV64)
-int rv64_zba = 0;
-int rv64_zbb = 0;
-int rv64_zbc = 0;
-int rv64_zbs = 0;
-int rv64_vector = 0; // rvv 1.0 or xtheadvector
-int rv64_xtheadvector = 0;
-int rv64_vlen = 0;
-int rv64_xtheadba = 0;
-int rv64_xtheadbb = 0;
-int rv64_xtheadbs = 0;
-int rv64_xtheadcondmov = 0;
-int rv64_xtheadmemidx = 0;
-int rv64_xtheadmempair = 0;
-int rv64_xtheadfmemidx = 0;
-int rv64_xtheadmac = 0;
-int rv64_xtheadfmv = 0;
-#elif defined(LA64)
-int la64_lbt = 0;
-int la64_lam_bh = 0;
-int la64_lamcas = 0;
-int la64_scq = 0;
-#endif
+cpu_ext_t cpuext = {0};
 #endif

 int box64_wine = 0;
@@ -125,9 +94,10 @@ char* trace_func = NULL;

 FILE* ftrace = NULL;
 char* ftrace_name = NULL;
-int ftrace_has_pid = 0;
+int ftrace_opened = 0;

-void openFTrace(int reopen)
+
+static void openFTrace(void)
 {
    const char* p = BOX64ENV(trace_file);
    #ifndef MAX_PATH
@@ -135,251 +105,61 @@ void openFTrace(int reopen)
    #endif
    char tmp[MAX_PATH];
    char tmp2[MAX_PATH];
-    int append=0;
-    if(p && strlen(p) && p[strlen(p)-1]=='+') {
+    int append = 0;
+
+    if (ftrace_name) box_free(ftrace_name);
+    ftrace_name = NULL;
+
+    if (p && strlen(p) && p[strlen(p) - 1] == '+') {
        strncpy(tmp2, p, sizeof(tmp2));
        tmp2[strlen(p)-1]='\0';
        p = tmp2;
        append = 1;
    }
-    if (reopen) {
-        p = ftrace_name;
-        append = 1;
-    } else {
-        if (p && strstr(p, "\%pid")) {
-            int next = 0;
-            do {
-                strcpy(tmp, p);
-                char* c = strstr(tmp, "%pid");
-                *c = 0; // cut
-                char pid[16];
-                if (next)
-                    sprintf(pid, "%d-%d", GetTID(), next);
-                else
-                    sprintf(pid, "%d", GetTID());
-                strcat(tmp, pid);
-                c = strstr(p, "\%pid") + strlen("\%pid");
-                strcat(tmp, c);
-                ++next;
-            } while (FileExist(tmp, IS_FILE) && !append);
-            p = tmp;
-            ftrace_has_pid = 1;
-        }
-        if (ftrace_name)
-            box_free(ftrace_name);
-        ftrace_name = NULL;
-    }
-    if(p) {
-        if(!strcmp(p, "stderr"))
-            ftrace = stderr;
-        else {
-            if(append)
-                ftrace = fopen(p, "a");
+
+    if (!p || ftrace_opened) return;
+    ftrace_opened = 1;
+
+    if (strstr(p, "\%pid")) {
+        int next = 0;
+        do {
+            strcpy(tmp, p);
+            char* c = strstr(tmp, "%pid");
+            *c = 0; // cut
+            char pid[16];
+            if (next)
+                sprintf(pid, "%d-%d", GetTID(), next);
            else
-                ftrace = fopen(p, "w");
-            if(!ftrace) {
-                ftrace = stdout;
-                printf_log(LOG_INFO, "Cannot open trace file \"%s\" for writing (error=%s)\n", p, strerror(errno));
-            } else {
-                if (!reopen) ftrace_name = box_strdup(p);
-                /*fclose(ftrace);
-                ftrace = NULL;*/
-                if (!BOX64ENV(nobanner)) {
-                    printf("BOX64 Trace %s to \"%s\"\n", append?"appended":"redirected", p);
-                    box64_stdout_no_w = 1;
-                }
-                PrintBox64Version(0);
-            }
-        }
+                sprintf(pid, "%d", GetTID());
+            strcat(tmp, pid);
+            c = strstr(p, "\%pid") + strlen("\%pid");
+            strcat(tmp, c);
+            ++next;
+        } while (FileExist(tmp, IS_FILE) && !append);
+        p = tmp;
    }
-}

-void my_prepare_fork()
-{
-    if (ftrace_has_pid && ftrace && (ftrace != stdout) && (ftrace != stderr)) {
-        printf_log(LOG_INFO, "%04d|Closed trace file of %s at prepare\n", GetTID(), GetLastApplyEntryName());
-        fclose(ftrace);
-    }
-}
-
-void my_parent_fork()
-{
-    if (ftrace_has_pid) {
-        openFTrace(1);
-        printf_log(LOG_INFO, "%04d|Reopened trace file of %s at parent\n", GetTID(), GetLastApplyEntryName());
-    }
-}
-
-void my_child_fork()
-{
-    if (ftrace_has_pid) {
-        openFTrace(0);
-        printf_log(LOG_INFO, "%04d|Created trace file of %s at child\n", GetTID(), GetLastApplyEntryName());
-    }
-}
-
-const char* getCpuName();
-int getNCpu();
-
-#ifdef DYNAREC
-void GatherDynarecExtensions()
-{
-#ifdef ARM64
-    unsigned long hwcap = real_getauxval(AT_HWCAP);
-    if(!hwcap)
-        hwcap = HWCAP_ASIMD;
-    // first, check all needed extensions, lif half, edsp and fastmult
-    if((hwcap&HWCAP_ASIMD) == 0) {
-        printf_log(LOG_INFO, "Missing ASMID cpu support, disabling Dynarec\n");
-        SET_BOX64ENV(dynarec, 0);
-        return;
-    }
-    if(hwcap&HWCAP_CRC32)
-        arm64_crc32 = 1;
-    if(hwcap&HWCAP_PMULL)
-        arm64_pmull = 1;
-    if(hwcap&HWCAP_AES)
-        arm64_aes = 1;
-    if(hwcap&HWCAP_ATOMICS)
-        arm64_atomics = 1;
-    #ifdef HWCAP_SHA1
-    if(hwcap&HWCAP_SHA1)
-        arm64_sha1 = 1;
-    #endif
-    #ifdef HWCAP_SHA2
-    if(hwcap&HWCAP_SHA2)
-        arm64_sha2 = 1;
-    #endif
-    #ifdef HWCAP_USCAT
-    if(hwcap&HWCAP_USCAT)
-        arm64_uscat = 1;
-    #endif
-    #ifdef HWCAP_FLAGM
-    if(hwcap&HWCAP_FLAGM)
-        arm64_flagm = 1;
-    #endif
-    unsigned long hwcap2 = real_getauxval(AT_HWCAP2);
-    #ifdef HWCAP2_FLAGM2
-    if(hwcap2&HWCAP2_FLAGM2)
-        arm64_flagm2 = 1;
-    #endif
-    #ifdef HWCAP2_FRINT
-    if(hwcap2&HWCAP2_FRINT)
-        arm64_frintts = 1;
-    #endif
-    #ifdef HWCAP2_AFP
-    if(hwcap2&HWCAP2_AFP)
-        arm64_afp = 1;
-    #endif
-    #ifdef HWCAP2_RNG
-    if(hwcap2&HWCAP2_RNG)
-        arm64_rndr = 1;
-    #endif
-    printf_log(LOG_INFO, "Dynarec for ARM64, with extension: ASIMD");
-    if(arm64_aes)
-        printf_log_prefix(0, LOG_INFO, " AES");
-    if(arm64_crc32)
-        printf_log_prefix(0, LOG_INFO, " CRC32");
-    if(arm64_pmull)
-        printf_log_prefix(0, LOG_INFO, " PMULL");
-    if(arm64_atomics)
-        printf_log_prefix(0, LOG_INFO, " ATOMICS");
-    if(arm64_sha1)
-        printf_log_prefix(0, LOG_INFO, " SHA1");
-    if(arm64_sha2)
-        printf_log_prefix(0, LOG_INFO, " SHA2");
-    if(arm64_uscat)
-        printf_log_prefix(0, LOG_INFO, " USCAT");
-    if(arm64_flagm)
-        printf_log_prefix(0, LOG_INFO, " FLAGM");
-    if(arm64_flagm2)
-        printf_log_prefix(0, LOG_INFO, " FLAGM2");
-    if(arm64_frintts)
-        printf_log_prefix(0, LOG_INFO, " FRINT");
-    if(arm64_afp)
-        printf_log_prefix(0, LOG_INFO, " AFP");
-    if(arm64_rndr)
-        printf_log_prefix(0, LOG_INFO, " RNDR");
-    printf_log_prefix(0, LOG_INFO, "\n");
-#elif defined(LA64)
-    printf_log(LOG_INFO, "Dynarec for LoongArch ");
-    char* p = getenv("BOX64_DYNAREC_LA64NOEXT");
-    if(p == NULL || p[0] == '0') {
-        uint32_t cpucfg2 = 0, idx = 2;
-        asm volatile("cpucfg %0, %1" : "=r"(cpucfg2) : "r"(idx));
-        if (((cpucfg2 >> 6) & 0b11) == 3) {
-            printf_log_prefix(0, LOG_INFO, "with extension LSX LASX");
+    if (!strcmp(p, "stdout"))
+        ftrace = stdout;
+    else if (!strcmp(p, "stderr"))
+        ftrace = stderr;
+    else {
+        if (append)
+            ftrace = fopen(p, "a");
+        else
+            ftrace = fopen(p, "w");
+        if (!ftrace) {
+            ftrace = stderr;
+            printf_log(LOG_INFO, "Cannot open trace file \"%s\" for writing (error=%s), fallback to stderr\n", p, strerror(errno));
        } else {
-            printf_log(LOG_INFO, "\nMissing LSX and/or LASX extension support, disabling Dynarec\n");
-            SET_BOX64ENV(dynarec, 0);
-            return;
-        }
-
-        if (la64_lbt = ((cpucfg2 >> 18) & 0b1))
-            printf_log_prefix(0, LOG_INFO, " LBT_X86");
-        if ((la64_lam_bh = (cpucfg2 >> 27) & 0b1))
-            printf_log_prefix(0, LOG_INFO, " LAM_BH");
-        if ((la64_lamcas = (cpucfg2 >> 28) & 0b1))
-            printf_log_prefix(0, LOG_INFO, " LAMCAS");
-        if ((la64_scq = (cpucfg2 >> 30) & 0b1))
-            printf_log_prefix(0, LOG_INFO, " SCQ");
-    }
-    printf_log_prefix(0, LOG_INFO, "\n");
-#elif defined(RV64)
-    void RV64_Detect_Function();
-    // private env. variable for the developer ;)
-    char *p = getenv("BOX64_DYNAREC_RV64NOEXT");
-    if(p == NULL || strcasecmp(p, "1")) {
-        RV64_Detect_Function();
-        if (p) {
-            p = strtok(p, ",");
-            while (p) {
-                if (!strcasecmp(p, "zba")) rv64_zba = 0;
-                if (!strcasecmp(p, "zbb")) rv64_zbb = 0;
-                if (!strcasecmp(p, "zbc")) rv64_zbc = 0;
-                if (!strcasecmp(p, "zbs")) rv64_zbs = 0;
-                if (!strcasecmp(p, "vector")) {
-                    rv64_vector = 0;
-                    rv64_xtheadvector = 0;
-                }
-                if (!strcasecmp(p, "xtheadba")) rv64_xtheadba = 0;
-                if (!strcasecmp(p, "xtheadbb")) rv64_xtheadbb = 0;
-                if (!strcasecmp(p, "xtheadbs")) rv64_xtheadbs = 0;
-                if (!strcasecmp(p, "xtheadmemidx")) rv64_xtheadmemidx = 0;
-                // if (!strcasecmp(p, "xtheadfmemidx")) rv64_xtheadfmemidx = 0;
-                // if (!strcasecmp(p, "xtheadmac")) rv64_xtheadmac = 0;
-                // if (!strcasecmp(p, "xtheadfmv")) rv64_xtheadfmv = 0;
-                if (!strcasecmp(p, "xtheadmempair")) rv64_xtheadmempair = 0;
-                if (!strcasecmp(p, "xtheadcondmov")) rv64_xtheadcondmov = 0;
-                p = strtok(NULL, ",");
+            ftrace_name = box_strdup(p);
+            if (!BOX64ENV(nobanner)) {
+                fprintf(stderr, "[BOX64] Trace %s to \"%s\" (set BOX64_NOBANNER=1 to suppress this log)\n", append ? "appended" : "redirected", p);
            }
+            PrintBox64Version(0);
        }
    }
-
-    printf_log(LOG_INFO, "Dynarec for rv64g");
-    if (rv64_vector && !rv64_xtheadvector) printf_log_prefix(0, LOG_INFO, "v");
-    if (rv64_zba) printf_log_prefix(0, LOG_INFO, "_zba");
-    if (rv64_zbb) printf_log_prefix(0, LOG_INFO, "_zbb");
-    if (rv64_zbc) printf_log_prefix(0, LOG_INFO, "_zbc");
-    if (rv64_zbs) printf_log_prefix(0, LOG_INFO, "_zbs");
-    if (rv64_vector && !rv64_xtheadvector) printf_log_prefix(0, LOG_INFO, "_zvl%d", rv64_vlen);
-    if (rv64_xtheadba) printf_log_prefix(0, LOG_INFO, "_xtheadba");
-    if (rv64_xtheadbb) printf_log_prefix(0, LOG_INFO, "_xtheadbb");
-    if (rv64_xtheadbs) printf_log_prefix(0, LOG_INFO, "_xtheadbs");
-    if (rv64_xtheadmempair) printf_log_prefix(0, LOG_INFO, "_xtheadmempair");
-    if (rv64_xtheadcondmov) printf_log_prefix(0, LOG_INFO, "_xtheadcondmov");
-    if (rv64_xtheadmemidx) printf_log_prefix(0, LOG_INFO, "_xtheadmemidx");
-    // Disable the display since these are only detected but never used.
-    // if(rv64_xtheadfmemidx) printf_log_prefix(0, LOG_INFO, " xtheadfmemidx");
-    // if(rv64_xtheadmac) printf_log_prefix(0, LOG_INFO, " xtheadmac");
-    // if(rv64_xtheadfmv) printf_log_prefix(0, LOG_INFO, " xtheadfmv");
-    if (rv64_xtheadvector) printf_log_prefix(0, LOG_INFO, "_xthvector");
-    printf_log_prefix(0, LOG_INFO, "\n");
-#else
-#error Unsupported architecture
-#endif
 }
-#endif

 void computeRDTSC()
 {
@@ -421,11 +201,11 @@ void computeRDTSC()
    printf_log_prefix(0, LOG_INFO, "\n");
 }

-static void displayMiscInfo()
+static void displayMiscInfo(void)
 {
-    openFTrace(0);
+    openFTrace();

-    if ((BOX64ENV(nobanner) || BOX64ENV(log)) && ftrace==stdout)
+    if ((BOX64ENV(nobanner) || BOX64ENV(log)) && ftrace == stdout)
        box64_stdout_no_w = 1;

 #if !defined(DYNAREC) && (defined(ARM64) || defined(RV64) || defined(LA64))
@@ -434,23 +214,27 @@ static void displayMiscInfo()

    char* p;

-    // grab pagesize
-    box64_pagesize = sysconf(_SC_PAGESIZE);
-    if(!box64_pagesize)
-        box64_pagesize = 4096;
-
 #ifdef DYNAREC
-    // grab cpu extensions for dynarec usage
-    GatherDynarecExtensions();
+    if (DetectHostCpuFeatures())
+        PrintHostCpuFeatures();
+    else {
+        printf_log(LOG_INFO, "Minimum CPU requirements not met, disabling DynaRec\n");
+        SET_BOX64ENV(dynarec, 0);
+    }
+
+#if defined(LA64)
+    if (box64env.avx && !cpuext.lasx) {
+        box64env.avx = 0;
+        box64env.avx2 = 0;
+    }
+#endif
 #endif

-    // grab ncpu and cpu name
-    int ncpu = getNCpu();
-    const char* cpuname = getCpuName();
-
-    printf_log(LOG_INFO, "Running on %s with %d core%s, pagesize: %zd\n", cpuname, ncpu, ncpu > 1 ? "s" : "", box64_pagesize);
-
-    // grab and calibrate hardware counter
+    printf_log(LOG_INFO, "Running on %s with %d core%s, pagesize: %zd", box64_sysinfo.cpuname, box64_sysinfo.ncpu, box64_sysinfo.ncpu > 1 ? "s" : "", box64_pagesize);
+    if (BOX64ENV(maxcpu))
+        printf_log_prefix(0, LOG_INFO, ", emulating %d core%s\n", BOX64ENV(maxcpu), BOX64ENV(maxcpu) > 1 ? "s" : "");
+    else
+        printf_log_prefix(0, LOG_INFO, "\n");
    computeRDTSC();
 }

@@ -545,12 +329,19 @@ void AddNewLibs(const char* list)
 }

 void PrintHelp() {
-    PrintfFtrace(0, "This is Box64, the Linux x86_64 emulator with a twist.\n");
-    PrintfFtrace(0, "Usage is 'box64 [options] path/to/software [args]' to launch x86_64 software.\n");
-    PrintfFtrace(0, " options are:\n");
-    PrintfFtrace(0, "    '-v'|'--version' to print box64 version and quit\n");
-    PrintfFtrace(0, "    '-h'|'--help' to print this and quit\n");
-    PrintfFtrace(0, "    '-k'|'--kill-all' to kill all box64 instances\n");
+    PrintfFtrace(0, "%s\n", BOX64_BUILD_INFO_STRING);
+    PrintfFtrace(0, "Linux userspace x86-64 emulator with a twist.\n");
+    PrintfFtrace(0, "There are many environment variables to control Box64's behaviour, checkout the documentation here: https://github.com/ptitSeb/box64/blob/main/docs/USAGE.md\n\n");
+    PrintfFtrace(0, "USAGE:\n");
+    PrintfFtrace(0, "\tbox64 [options] path/to/x86_64/executable [args]\n");
+    PrintfFtrace(0, "\tbox64-bash\n");
+    PrintfFtrace(0, "OPTIONS:\n");
+    PrintfFtrace(0, "\t-v, --version          print box64 version and quit\n");
+    PrintfFtrace(0, "\t-h, --help             print this and quit\n");
+    PrintfFtrace(0, "\t-k, --kill-all         kill all box64 instances\n");
+    PrintfFtrace(0, "\t-t, --test             run a unit test\n");
+    PrintfFtrace(0, "\t--dynacache-list       list of DynaCache file and their validity\n");
+    PrintfFtrace(0, "\t--dynacache-clean      remove invalid DynaCache files\n");
 }

 void KillAllInstances()
@@ -586,7 +377,7 @@ void KillAllInstances()
    snprintf(self_name_with_deleted, sizeof(self_name_with_deleted), "%s (deleted)", self_name);

    const pid_t self = getpid();
-    while (entry = readdir(proc_dir)) {
+    while ((entry = readdir(proc_dir))) {
        const pid_t pid = atoi(entry->d_name);
        if (!pid || pid == self) continue;

@@ -624,16 +415,37 @@ static void addLibPaths(box64context_t* context)
    }

    // Add libssl and libcrypto (and a few others) to prefer the emulated version because multiple versions exist
-    AddPath("libssl.so.1", &context->box64_emulated_libs, 0);
-    AddPath("libssl.so.1.0.0", &context->box64_emulated_libs, 0);
-    AddPath("libcrypto.so.1", &context->box64_emulated_libs, 0);
-    AddPath("libcrypto.so.1.0.0", &context->box64_emulated_libs, 0);
-    AddPath("libunwind.so.8", &context->box64_emulated_libs, 0);
-    AddPath("libpng12.so.0", &context->box64_emulated_libs, 0);
-    AddPath("libcurl.so.4", &context->box64_emulated_libs, 0);
-    //AddPath("libgnutls.so.30", &context->box64_emulated_libs, 0);
-    AddPath("libtbbmalloc.so.2", &context->box64_emulated_libs, 0);
-    AddPath("libtbbmalloc_proxy.so.2", &context->box64_emulated_libs, 0);
+    #define GO(A)   AddPath(A, &context->box64_emulated_libs, 0);
+    GO("libssl.so.1");
+    GO("libssl.so.1.0.0");
+    GO("libcrypto.so.1");
+    GO("libcrypto.so.1.0.0");
+    GO("libunwind.so.8");
+    GO("libpng12.so.0");
+    GO("libpng16.so.16");
+    GO("libcurl.so.4");
+    if(getenv("BOX64_PRESSURE_VESSEL_FILES"))   // use emulated gnutls in this case, it's safer
+        GO("libgnutls.so.30");
+    GO("libtbbmalloc.so.2");
+    GO("libtbbmalloc_proxy.so.2");
+    GO("libicuuc.so.64");
+    GO("libicui18n.so.64");
+    GO("libicuuc.so.66");
+    GO("libicui18n.so.66");
+    GO("libicuuc.so.67");
+    GO("libicui18n.so.67");
+    GO("libicuuc.so.72");
+    GO("libicui18n.so.72");
+    GO("libicuuc.so.73");
+    GO("libicui18n.so.73");
+    GO("libicuuc.so.74");
+    GO("libicui18n.so.74");
+    GO("libicuuc.so.75");
+    GO("libicui18n.so.75");
+    GO("libicuuc.so.76");
+    GO("libicui18n.so.76");
+    GO("libSDL3.so.0");
+    #undef GO

    if(BOX64ENV(nosigsegv)) {
        context->no_sigsegv = 1;
@@ -649,7 +461,7 @@ static void addLibPaths(box64context_t* context)
        AppendList(&context->box64_path, getenv("PATH"), 1);   // in case some of the path are for x86 world
 }

-void setupZydis(box64context_t* context)
+static void setupZydis(box64context_t* context)
 {
 #ifdef HAVE_TRACE
    if ((BOX64ENV(trace_init) && strcmp(BOX64ENV(trace_init), "0")) || (BOX64ENV(trace) && strcmp(BOX64ENV(trace), "0"))) {
@@ -685,6 +497,8 @@ void LoadLDPath(box64context_t *context)
            AddPath("/usr/i386-linux-gnu/lib", &context->box64_ld_lib, 1);
        if(FileExist("/usr/lib/box64-i386-linux-gnu", 0))
            AddPath("/usr/lib/box64-i386-linux-gnu", &context->box64_ld_lib, 1);
+        if(FileExist("/opt/box64/lib32", 0))
+            AddPath("/opt/box64/lib32", &context->box64_ld_lib, 1);
        if(FileExist("/data/data/com.termux/files/usr/glibc/lib/i386-linux-gnu", 0))
            AddPath("/data/data/com.termux/files/usr/glibc/lib/i386-linux-gnu", &context->box64_ld_lib, 1);
        if(FileExist("/data/data/com.termux/files/usr/glibc/lib/box64-i386-linux-gnu", 0))
@@ -701,6 +515,8 @@ void LoadLDPath(box64context_t *context)
            AddPath("/usr/x86_64-linux-gnu/lib", &context->box64_ld_lib, 1);
        if(FileExist("/usr/lib/box64-x86_64-linux-gnu", 0))
            AddPath("/usr/lib/box64-x86_64-linux-gnu", &context->box64_ld_lib, 1);
+        if(FileExist("/opt/box64/lib64", 0))
+            AddPath("/opt/box64/lib64", &context->box64_ld_lib, 1);
        if(FileExist("/data/data/com.termux/files/usr/glibc/lib/x86_64-linux-gnu", 0))
            AddPath("/data/data/com.termux/files/usr/glibc/lib/x86_64-linux-gnu", &context->box64_ld_lib, 1);
        if(FileExist("/data/data/com.termux/files/usr/glibc/lib/box64-x86_64-linux-gnu", 0))
@@ -734,6 +550,8 @@ void setupTraceInit()
        uintptr_t s_trace_start=0, s_trace_end=0;
        if (strcmp(p, "1")==0)
            SetTraceEmu(0, 0);
+        else if (strcmp(p, "0")==0)
+            ;
        else if (strchr(p,'-')) {
            if(sscanf(p, "%ld-%ld", &s_trace_start, &s_trace_end)!=2) {
                if(sscanf(p, "0x%lX-0x%lX", &s_trace_start, &s_trace_end)!=2)
@@ -803,6 +621,8 @@ void setupTrace()
        uintptr_t s_trace_start=0, s_trace_end=0;
        if (strcmp(p, "1")==0)
            SetTraceEmu(0, 0);
+        else if (strcmp(p, "0")==0)
+            ;
        else if (strchr(p,'-')) {
            if(sscanf(p, "%ld-%ld", &s_trace_start, &s_trace_end)!=2) {
                if(sscanf(p, "0x%lX-0x%lX", &s_trace_start, &s_trace_end)!=2)
@@ -851,6 +671,7 @@ void endBox64()
    if(!my_context || box64_quit)
        return;

+    SerializeAllMapping();   // just to be safe
    // then call all the fini
    dynarec_log(LOG_DEBUG, "endBox64() called\n");
    box64_quit = 1;
@@ -874,6 +695,7 @@ void endBox64()
    #ifndef STATICBUILD
    endMallocHook();
    #endif
+    SerializeAllMapping();   // to be safe
    FreeBox64Context(&my_context);
    #ifdef DYNAREC
    // disable dynarec now
@@ -922,9 +744,9 @@ extern char** environ;

 int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elfheader_t** elfheader, int exec)
 {
-    #ifndef STATICBUILD
+#ifndef STATICBUILD
    init_malloc_hook();
-    #endif
+#endif
    init_auxval(argc, argv, environ?environ:env);
    // analogue to QEMU_VERSION in qemu-user-mode emulation
    if(getenv("BOX64_VERSION")) {
@@ -943,13 +765,22 @@ int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elf
    if(argc>1 && !strcmp(argv[1], "/usr/bin/gdb") && BOX64ENV(trace_file))
        exit(0);
    // uname -m is redirected to box64 -m
-    if(argc==2 && (!strcmp(argv[1], "-m") || !strcmp(argv[1], "-p") || !strcmp(argv[1], "-i")))
-    {
+    if (argc == 2 && (!strcmp(argv[1], "-m") || !strcmp(argv[1], "-p") || !strcmp(argv[1], "-i"))) {
        printf("x86_64\n");
        exit(0);
    }

-    ftrace = stdout;
+    if (argc >= 3 && (!strcmp(argv[1], "--test") || !strcmp(argv[1], "-t"))) {
+        box64_unittest_mode = 1;
+        exit(unittest(argc, argv));
+    }
+
+    ftrace = stderr;
+
+    // grab pagesize
+    box64_pagesize = sysconf(_SC_PAGESIZE);
+    if(!box64_pagesize)
+        box64_pagesize = 4096;

    LoadEnvVariables();
    InitializeEnvFiles();
@@ -970,6 +801,14 @@ int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elf
            KillAllInstances();
            exit(0);
        }
+        if(!strcmp(prog, "--dynacache-list")) {
+            DynaCacheList(argv[nextarg+1]);
+            exit(0);
+        }
+        if(!strcmp(prog, "--dynacache-clean")) {
+            DynaCacheClean();
+            exit(0);
+        }
        // other options?
        if(!strcmp(prog, "--")) {
            prog = argv[++nextarg];
@@ -985,6 +824,7 @@ int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elf

    if (!BOX64ENV(nobanner)) PrintBox64Version(1);

+    InitializeSystemInfo();
    displayMiscInfo();

    hookMangoHud();
@@ -1001,6 +841,18 @@ int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elf
            }
        }
    }
+    char* pythonpath = NULL;
+    {
+        char* p = BOX64ENV(python3);
+        if(p) {
+            if(FileIsX64ELF(p)) {
+                pythonpath = p;
+                printf_log(LOG_INFO, "Using python3 \"%s\"\n", pythonpath);
+            } else {
+                printf_log(LOG_INFO, "The x86_64 python3 \"%s\" is not an x86_64 binary.\n", p);
+            }
+        }
+    }

    // precheck, for win-preload
    const char* prog_ = strrchr(prog, '/');
@@ -1021,13 +873,6 @@ int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elf
            if(!prog_) prog_ = prog; else ++prog_;
        }
    }
-    #ifndef STATICBUILD
-    // pre-check for pressure-vessel-wrap
-    if(!strcmp(prog_, "pressure-vessel-wrap")) {
-        printf_log(LOG_INFO, "pressure-vessel-wrap detected\n");
-        pressure_vessel(argc, argv, nextarg+1, prog);
-    }
-    #endif
    int ld_libs_args = -1;
    int is_custom_gstreamer = 0;
    // check if this is wine
@@ -1125,6 +970,17 @@ int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elf
    // allocate extra space for new environment variables such as BOX64_PATH
    my_context->envv = (char**)box_calloc(my_context->envc+1, sizeof(char*));

+    #ifndef STATICBUILD
+    // pre-check for pressure-vessel-wrap
+    if(!strcmp(prog_, "pressure-vessel-wrap")) {
+        printf_log(LOG_INFO, "pressure-vessel-wrap detected, bashpath=%s\n", my_context->bashpath?my_context->bashpath:"(nil)");
+        unsetenv("BOX64_ARG0");
+        if(!my_context->bashpath)
+            my_context->bashpath = ResolveFile("box64-bash", &my_context->box64_path);
+        pressure_vessel(argc, argv, nextarg+1, prog);
+    }
+    #endif
+
    path_collection_t ld_preload = {0};
    if(getenv("BOX64_LD_PRELOAD")) {
        char* p = getenv("BOX64_LD_PRELOAD");
@@ -1204,6 +1060,7 @@ int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elf
        box64_zoom = 1;
    }
    // special case for bash
+    int setup_bash_rcfile = 0;
    if (!strcmp(box64_guest_name, "bash") || !strcmp(box64_guest_name, "box64-bash")) {
        printf_log(LOG_INFO, "Bash detected, disabling banner\n");
        if (!BOX64ENV(nobanner)) {
@@ -1212,27 +1069,41 @@ int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elf
        }
        if (!bashpath) {
            bashpath = (char*)prog;
+            SET_BOX64ENV(bash, (char*)prog);
            setenv("BOX64_BASH", prog, 1);
        }
+        setup_bash_rcfile = 1;
    }
    if(!bashpath)
        bashpath = ResolveFile("box64-bash", &my_context->box64_path);
    if(bashpath)
        my_context->bashpath = box_strdup(bashpath);
+    if(pythonpath)
+        my_context->pythonpath = box_strdup(pythonpath);

-    ApplyEnvFileEntry(box64_guest_name);
+    int applied = ApplyEnvFileEntry(box64_guest_name);
    if (box64_wine && box64_wine_guest_name) {
-        ApplyEnvFileEntry(box64_wine_guest_name);
+        applied |= ApplyEnvFileEntry(box64_wine_guest_name);
        box64_wine_guest_name = NULL;
    }
-    openFTrace(0);
+    if (applied) {
+        printf_log(LOG_INFO, "Applied settings from rcfile\n");
+        displayMiscInfo();
+        PrintEnvVariables(&box64env, LOG_INFO);
+    }
    setupZydis(my_context);
-    PrintEnvVariables(&box64env, LOG_INFO);

    for(int i=1; i<my_context->argc; ++i) {
        my_context->argv[i] = box_strdup(argv[i+nextarg]);
        printf_log(LOG_INFO, "argv[%i]=\"%s\"\n", i, my_context->argv[i]);
    }
+
+    // Setup custom bash rcfile iff no args are present
+    if (setup_bash_rcfile && my_context->argc == 1) {
+        add_argv("--rcfile");
+        add_argv("box64-custom-bashrc-file"); // handled by my_open
+    }
+
    if(BOX64ENV(nosandbox))
    {
        add_argv("--no-sandbox");
@@ -1241,14 +1112,6 @@ int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elf
    {
        add_argv("--in-process-gpu");
    }
-    if(BOX64ENV(cefdisablegpu))
-    {
-        add_argv("-cef-disable-gpu");
-    }
-    if(BOX64ENV(cefdisablegpucompositor))
-    {
-        add_argv("-cef-disable-gpu-compositor");
-    }
    // add new args only if there is no args already
    if(BOX64ENV(args)) {
        char tmp[256];
@@ -1316,8 +1179,10 @@ int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elf
    }
    if(!(my_context->fullpath = box_realpath(my_context->argv[0], NULL)))
        my_context->fullpath = box_strdup(my_context->argv[0]);
-    if(getenv("BOX64_ARG0"))
+    if (getenv("BOX64_ARG0")) {
        my_context->argv[0] = box_strdup(getenv("BOX64_ARG0"));
+        unsetenv("BOX64_ARG0");
+    }
    FILE *f = fopen(my_context->fullpath, "rb");
    if(!f) {
        printf_log(LOG_NONE, "Error: Cannot open %s\n", my_context->fullpath);
@@ -1328,6 +1193,32 @@ int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elf
    }
    #ifdef BOX32
    box64_is32bits = FileIsX86ELF(my_context->fullpath);
+    // try to switch personality, but only if not already tried
+    if(box64_is32bits && !box64env.nopersona32bits) {
+        int tried = getenv("BOX32_PERSONA32BITS")?1:0;
+        if(tried) {
+            unsetenv("BOX32_PERSONA32BITS");
+            int p = personality(0xffffffff);
+            if(p==ADDR_LIMIT_32BIT) {
+                box64_isAddressSpace32 = 1;
+                printf_log(LOG_INFO, "Personality set to 32bits\n");
+            }
+        } else {
+            if(personality(ADDR_LIMIT_32BIT)!=-1) {
+                int nenv = 0;
+                while(env[nenv]) nenv++;
+                // alloc + "LD_PRELOAD" if needd + last NULL ending
+                char** newenv = (char**)box_calloc(nenv+1+1, sizeof(char*));
+                // copy strings
+                for (int i=0; i<nenv; ++i)
+                    newenv[i] = box_strdup(env[i]);
+                newenv[nenv] = "BOX32_PERSONA32BITS=1";
+                // re-launch...
+                if(execve(my_context->box64path, (void*)argv, newenv)<0)
+                    printf_log(LOG_NONE, "Failed to relaunch. Error is %d/%s (argv[0]=\"%s\")\n", errno, strerror(errno), argv[0]);
+            }
+        }
+    }
    if(box64_is32bits) {
        printf_log(LOG_INFO, "Using Box32 to load 32bits elf\n");
        loadProtectionFromMap();
@@ -1336,10 +1227,16 @@ int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elf
    }
    #endif
    LoadLDPath(my_context);
+    my_context->video_mem = mmap((void*)0xb0000, 0x10000, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
+    if(my_context->video_mem != (void*)0xb0000) {
+        my_context->video_mem = NULL;
+        printf_log(LOG_INFO, "Warning, could not allocate text video memory");
+    }
    elfheader_t *elf_header = LoadAndCheckElfHeader(f, my_context->fullpath, 1);
    if(!elf_header) {
        int x86 = my_context->box86path?FileIsX86ELF(my_context->fullpath):0;
        int script = my_context->bashpath?FileIsShell(my_context->fullpath):0;
+        int python3 = my_context->pythonpath?FileIsPython(my_context->fullpath):0;
        printf_log(LOG_NONE, "Error: Reading elf header of %s, Try to launch %s instead\n", my_context->fullpath, x86?"using box86":(script?"using bash":"natively"));
        fclose(f);
        FreeCollection(&ld_preload);
@@ -1359,6 +1256,14 @@ int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elf
            for(int i=0; i<my_context->argc; ++i)
                newargv[i+2] = my_context->argv[i];
            ret = execvp(newargv[0], (char * const*)newargv);
+        } else if (python3) {
+            // duplicate the array and insert 1st arg as box64, 2nd is python3
+            const char** newargv = (const char**)box_calloc(my_context->argc+3, sizeof(char*));
+            newargv[0] = my_context->box64path;
+            newargv[1] = my_context->pythonpath;
+            for(int i=0; i<my_context->argc; ++i)
+                newargv[i+2] = my_context->argv[i];
+            ret = execvp(newargv[0], (char * const*)newargv);
        } else {
            const char** newargv = (const char**)box_calloc(my_context->argc+1, sizeof(char*));
            for(int i=0; i<my_context->argc; ++i)
@@ -1481,6 +1386,8 @@ int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elf
        my_context->orig_argc = argc;
        my_context->orig_argv = (char**)argv;
    }
+    box64_nolibs = (NeededLibs(elf_header)==0);
+    if(box64_nolibs) printf_log(LOG_INFO, "Warning, box64 is not really compatible with staticaly linked binaries. Expect crash!\n");
    box64_isglibc234 = GetNeededVersionForLib(elf_header, "libc.so.6", "GLIBC_2.34");
    if(box64_isglibc234)
        printf_log(LOG_DEBUG, "Program linked with GLIBC 2.34+\n");
@@ -1507,9 +1414,6 @@ int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elf
        SetRBP(emu, 0); // Frame pointer so to "No more frame pointer"
    }

-    // child fork to handle traces
-    //pthread_atfork(my_prepare_fork, my_parent_fork, my_child_fork);
-
    thread_set_emu(emu);

    // export symbols
@@ -1567,7 +1471,7 @@ int initialize(int argc, const char **argv, char** env, x64emu_t** emulator, elf
    setupTraceInit();
    RunDeferredElfInit(emu);
    // update TLS of main elf
-    RefreshElfTLS(elf_header);
+    RefreshElfTLS(elf_header, emu);
    // do some special case check, _IO_2_1_stderr_ and friends, that are setup by libc, but it's already done here, so need to do a copy
    ResetSpecialCaseMainElf(elf_header);
    // init...
@@ -5,148 +5,7 @@

 */

-/* 
-    ARM64 Linux Call Convention
-
-SP          The Stack Pointer.
-r30 LR      The Link Register.
-r29 FP      The Frame Pointer
-r19…r28     Callee-saved registers
-r18         The Platform Register, if needed; otherwise a temporary register. See notes.
-r17 IP1     The second intra-procedure-call temporary register (can be used by call veneers and PLT code); at other times may be used as a temporary register.
-r16 IP0     The first intra-procedure-call scratch register (can be used by call veneers and PLT code); at other times may be used as a temporary register.
-r9…r15      Temporary registers
-r8          Indirect result location register
-r0…r7       Parameter/result registers
-
-For SIMD:
-The first eight registers, v0-v7, are used to pass argument values into a subroutine and to return result values from a function. 
-    They may also be used to hold intermediate values within a routine (but, in general, only between subroutine calls).
-
-Registers v8-v15 must be preserved by a callee across subroutine calls; 
-    the remaining registers (v0-v7, v16-v31) do not need to be preserved (or should be preserved by the caller).
-    Additionally, only the bottom 64 bits of each value stored in v8-v15 need to be preserved [8];
-    it is the responsibility of the caller to preserve larger values.
-
-For SVE:
-z0-z7 are used to pass scalable vector arguments to a subroutine, and to return scalable vector results from a function.
-    If a subroutine takes at least one argument in scalable vector registers or scalable predicate registers,
-    or returns results in such regisers, the subroutine must ensure that the entire contents of z8-z23 are preserved across the call.
-    In other cases it need only preserve the low 64 bits of z8-z15, as described in SIMD and Floating-Point registers.
-p0-p3 are used to pass scalable predicate arguments to a subroutine and to return scalable predicate results from a function.
-    If a subroutine takes at least one argument in scalable vector registers or scalable predicate registers,
-    or returns results in such registers, the subroutine must ensure that p4-p15 are preserved across the call.
-    In other cases it need not preserve any scalable predicate register contents.
-
-*/
-
-// x86 Register mapping
-#define xRAX    10
-#define xRCX    11
-#define xRDX    12
-#define xRBX    13
-#define xRSP    14
-#define xRBP    15
-#define xRSI    16
-#define xRDI    17
-#define xR8     18
-#define xR9     19
-#define xR10    20
-#define xR11    21
-#define xR12    22
-#define xR13    23
-#define xR14    24
-#define xR15    25
-#define xFlags  26
-#define xRIP    27
-#define xSavedSP 28
-
-// convert a x86 register to native according to the register mapping
-#define TO_NAT(A) (xRAX + (A))
-
-// 32bits version
-#define wEAX    xRAX
-#define wECX    xRCX
-#define wEDX    xRDX
-#define wEBX    xRBX
-#define wESP    xRSP
-#define wEBP    xRBP
-#define wESI    xRSI
-#define wEDI    xRDI
-#define wR8     xR8
-#define wR9     xR9
-#define wR10    xR10
-#define wR11    xR11
-#define wR12    xR12
-#define wR13    xR13
-#define wR14    xR14
-#define wR15    xR15
-#define wFlags  xFlags
-// scratch registers
-#define x1      1
-#define x2      2
-#define x3      3
-#define x4      4
-#define x5      5
-#define x6      6
-#define x87pc   7
-// x87 can be a scratch, but check if it's used as x87 PC and restore if needed in that case
-// 32bits version of scratch
-#define w1      x1
-#define w2      x2
-#define w3      x3
-#define w4      x4
-#define w5      x5
-#define w6      x6
-#define w87pc   x87pc
-// emu is r0
-#define xEmu    0
-// ARM64 LR
-#define xLR     30
-// ARM64 SP is r31 but is a special register
-#define xSP     31      
-// xZR regs is 31
-#define xZR     31
-#define wZR     xZR
-
-// conditions
-// Z == 1
-#define cEQ 0b0000
-// Z != 1
-#define cNE 0b0001
-// C == 1
-#define cCS 0b0010
-// C == 1
-#define cHS cCS
-// C != 1
-#define cCC 0b0011
-// C != 1
-#define cLO cCC
-// N == 1
-#define cMI 0b0100
-// N != 1
-#define cPL 0b0101
-// V == 1
-#define cVS 0b0110
-// V != 1
-#define cVC 0b0111
-// C == 1 && Z == 0
-#define cHI 0b1000
-// C !=1 || Z == 1
-#define cLS 0b1001
-// N == V
-#define cGE 0b1010
-// N != V
-#define cLT 0b1011
-// N == V && Z == 0
-#define cGT 0b1100
-// N != V || Z == 1
-#define cLE 0b1101
-// always
-#define c__ 0b1110
-
-//FCMP type of opcode produce:
-// if any NAN: CV / v1 == v2: ZC / v1 < v2: N / v1 > v2: C
+#include "arm64_mapping.h"

 int convert_bitmask(uint64_t bitmask);
 #define convert_bitmask_w(A)    convert_bitmask(((uint64_t)(A) << 32) + (uint32_t)(A))
@@ -181,6 +40,7 @@ int convert_bitmask(uint64_t bitmask);

 #define MOV64xw(Rd, imm64)   if(rex.w) {MOV64x(Rd, imm64);} else {MOV32w(Rd, imm64);}
 #define MOV64z(Rd, imm64)    if(rex.is32bits) {MOV32w(Rd, imm64);} else {MOV64x(Rd, imm64);}
+#define MOV64y(Rd, imm64)    if(rex.is32bits || rex.is67) {MOV32w(Rd, imm64);} else {MOV64x(Rd, imm64);}


 // ADD / SUB
@@ -188,13 +48,17 @@ int convert_bitmask(uint64_t bitmask);
 #define ADDx_REG(Rd, Rn, Rm)                EMIT(ADDSUB_REG_gen(1, 0, 0, 0b00, Rm, 0, Rn, Rd))
 #define ADDSx_REG(Rd, Rn, Rm)              FEMIT(ADDSUB_REG_gen(1, 0, 1, 0b00, Rm, 0, Rn, Rd))
 #define ADDx_REG_LSL(Rd, Rn, Rm, lsl)       EMIT(ADDSUB_REG_gen(1, 0, 0, 0b00, Rm, lsl, Rn, Rd))
-#define ADDz_REG_LSL(Rd, Rn, Rm, lsl)       EMIT(ADDSUB_REG_gen(rex.is32bits?0:1, 0, 0, 0b00, Rm, lsl, Rn, Rd))
+#define ADDx_REG_LSR(Rd, Rn, Rm, lsr)       EMIT(ADDSUB_REG_gen(1, 0, 0, 0b01, Rm, lsr, Rn, Rd))
+#define ADDz_REG_LSL(Rd, Rn, Rm, lsl)       EMIT(ADDSUB_REG_gen((rex.is32bits)?0:1, 0, 0, 0b00, Rm, lsl, Rn, Rd))
+#define ADDy_REG_LSL(Rd, Rn, Rm, lsl)       EMIT(ADDSUB_REG_gen((rex.is32bits || rex.is67)?0:1, 0, 0, 0b00, Rm, lsl, Rn, Rd))
 #define ADDw_REG(Rd, Rn, Rm)                EMIT(ADDSUB_REG_gen(0, 0, 0, 0b00, Rm, 0, Rn, Rd))
 #define ADDSw_REG(Rd, Rn, Rm)              FEMIT(ADDSUB_REG_gen(0, 0, 1, 0b00, Rm, 0, Rn, Rd))
 #define ADDw_REG_LSL(Rd, Rn, Rm, lsl)       EMIT(ADDSUB_REG_gen(0, 0, 0, 0b00, Rm, lsl, Rn, Rd))
+#define ADDw_REG_LSR(Rd, Rn, Rm, lsr)       EMIT(ADDSUB_REG_gen(0, 0, 0, 0b01, Rm, lsr, Rn, Rd))
 #define ADDSw_REG_LSL(Rd, Rn, Rm, lsl)     FEMIT(ADDSUB_REG_gen(0, 0, 1, 0b00, Rm, lsl, Rn, Rd))
 #define ADDxw_REG(Rd, Rn, Rm)               EMIT(ADDSUB_REG_gen(rex.w, 0, 0, 0b00, Rm, 0, Rn, Rd))
-#define ADDz_REG(Rd, Rn, Rm)                EMIT(ADDSUB_REG_gen(rex.is32bits?0:1, 0, 0, 0b00, Rm, 0, Rn, Rd))
+#define ADDz_REG(Rd, Rn, Rm)                EMIT(ADDSUB_REG_gen((rex.is32bits)?0:1, 0, 0, 0b00, Rm, 0, Rn, Rd))
+#define ADDy_REG(Rd, Rn, Rm)                EMIT(ADDSUB_REG_gen((rex.is32bits || rex.is67)?0:1, 0, 0, 0b00, Rm, 0, Rn, Rd))
 #define ADDSxw_REG(Rd, Rn, Rm)             FEMIT(ADDSUB_REG_gen(rex.w, 0, 1, 0b00, Rm, 0, Rn, Rd))
 #define ADDxw_REG_LSR(Rd, Rn, Rm, lsr)      EMIT(ADDSUB_REG_gen(rex.w, 0, 0, 0b01, Rm, lsr, Rn, Rd))

@@ -205,7 +69,8 @@ int convert_bitmask(uint64_t bitmask);
 #define ADDSw_U12(Rd, Rn, imm12)   FEMIT(ADDSUB_IMM_gen(0, 0, 1, 0b00, (imm12)&0xfff, Rn, Rd))
 #define ADDxw_U12(Rd, Rn, imm12)    EMIT(ADDSUB_IMM_gen(rex.w, 0, 0, 0b00, (imm12)&0xfff, Rn, Rd))
 #define ADDSxw_U12(Rd, Rn, imm12)  FEMIT(ADDSUB_IMM_gen(rex.w, 0, 1, 0b00, (imm12)&0xfff, Rn, Rd))
-#define ADDz_U12(Rd, Rn, imm12)     EMIT(ADDSUB_IMM_gen(rex.is32bits?0:1, 0, 0, 0b00, (imm12)&0xfff, Rn, Rd))
+#define ADDz_U12(Rd, Rn, imm12)     EMIT(ADDSUB_IMM_gen((rex.is32bits)?0:1, 0, 0, 0b00, (imm12)&0xfff, Rn, Rd))
+#define ADDy_U12(Rd, Rn, imm12)     EMIT(ADDSUB_IMM_gen((rex.is32bits || rex.is67)?0:1, 0, 0, 0b00, (imm12)&0xfff, Rn, Rd))

 #define SUBx_REG(Rd, Rn, Rm)                EMIT(ADDSUB_REG_gen(1, 1, 0, 0b00, Rm, 0, Rn, Rd))
 #define SUBSx_REG(Rd, Rn, Rm)              FEMIT(ADDSUB_REG_gen(1, 1, 1, 0b00, Rm, 0, Rn, Rd))
@@ -214,12 +79,14 @@ int convert_bitmask(uint64_t bitmask);
 #define SUBw_REG(Rd, Rn, Rm)                EMIT(ADDSUB_REG_gen(0, 1, 0, 0b00, Rm, 0, Rn, Rd))
 #define SUBw_REG_LSL(Rd, Rn, Rm, lsl)       EMIT(ADDSUB_REG_gen(0, 1, 0, 0b00, Rm, lsl, Rn, Rd))
 #define SUBw_REG_ASR(Rd, Rn, Rm, asr)       EMIT(ADDSUB_REG_gen(0, 1, 0, 0b10, Rm, asr, Rn, Rd))
+#define SUBx_REG_ASR(Rd, Rn, Rm, asr)       EMIT(ADDSUB_REG_gen(1, 1, 0, 0b10, Rm, asr, Rn, Rd))
 #define SUBSw_REG(Rd, Rn, Rm)              FEMIT(ADDSUB_REG_gen(0, 1, 1, 0b00, Rm, 0, Rn, Rd))
 #define SUBSw_REG_LSL(Rd, Rn, Rm, lsl)     FEMIT(ADDSUB_REG_gen(0, 1, 1, 0b00, Rm, lsl, Rn, Rd))
 #define SUBSw_REG_LSR(Rd, Rn, Rm, lsr)     FEMIT(ADDSUB_REG_gen(0, 1, 1, 0b01, Rm, lsr, Rn, Rd))
 #define SUBSw_REG_ASR(Rd, Rn, Rm, asr)     FEMIT(ADDSUB_REG_gen(0, 1, 1, 0b10, Rm, asr, Rn, Rd))
 #define SUBxw_REG(Rd, Rn, Rm)               EMIT(ADDSUB_REG_gen(rex.w, 1, 0, 0b00, Rm, 0, Rn, Rd))
 #define SUBz_REG(Rd, Rn, Rm)                EMIT(ADDSUB_REG_gen(rex.is32bits?0:1, 1, 0, 0b00, Rm, 0, Rn, Rd))
+#define SUBy_REG(Rd, Rn, Rm)                EMIT(ADDSUB_REG_gen((rex.is32bits || rex.is67)?0:1, 1, 0, 0b00, Rm, 0, Rn, Rd))
 #define SUBSxw_REG(Rd, Rn, Rm)             FEMIT(ADDSUB_REG_gen(rex.w, 1, 1, 0b00, Rm, 0, Rn, Rd))
 #define SUBSxw_REG_ASR(Rd, Rn, Rm, asr)    FEMIT(ADDSUB_REG_gen(rex.w, 1, 1, 0b10, Rm, asr, Rn, Rd))
 #define CMPSx_REG(Rn, Rm)                   SUBSx_REG(xZR, Rn, Rm)
@@ -242,6 +109,7 @@ int convert_bitmask(uint64_t bitmask);
 #define SUBSw_U12(Rd, Rn, imm12)   FEMIT(ADDSUB_IMM_gen(0, 1, 1, 0b00, (imm12)&0xfff, Rn, Rd))
 #define SUBxw_U12(Rd, Rn, imm12)    EMIT(ADDSUB_IMM_gen(rex.w, 1, 0, 0b00, (imm12)&0xfff, Rn, Rd))
 #define SUBz_U12(Rd, Rn, imm12)     EMIT(ADDSUB_IMM_gen(rex.is32bits?0:1, 1, 0, 0b00, (imm12)&0xfff, Rn, Rd))
+#define SUBy_U12(Rd, Rn, imm12)     EMIT(ADDSUB_IMM_gen((rex.is32bits || rex.is67)?0:1, 1, 0, 0b00, (imm12)&0xfff, Rn, Rd))
 #define SUBSxw_U12(Rd, Rn, imm12)  FEMIT(ADDSUB_IMM_gen(rex.w, 1, 1, 0b00, (imm12)&0xfff, Rn, Rd))
 #define CMPSx_U12(Rn, imm12)        SUBSx_U12(xZR, Rn, imm12)
 #define CMPSw_U12(Rn, imm12)        SUBSw_U12(wZR, Rn, imm12)
@@ -261,9 +129,19 @@ int convert_bitmask(uint64_t bitmask);
 #define SBCSw_REG(Rd, Rn, Rm)      FEMIT(ADDSUBC_gen(0, 1, 1, Rm, Rn, Rd))
 #define SBCSxw_REG(Rd, Rn, Rm)     FEMIT(ADDSUBC_gen(rex.w, 1, 1, Rm, Rn, Rd))

-#define SUB_ext(sf, op, S, Rm, option, imm3, Rn, Rd)    ((sf)<<31 | (op)<<30 | (S)<<29 | 0b01011<<24 | 1<<21 | (Rm)<<16 | (option)<<13 | (imm3)<<10 | (Rn)<<5 | (Rd))
-#define SUBxw_UXTB(Rd, Rn, Rm)      EMIT(SUB_ext(rex.w, 1, 0, Rm, 0b000, 0, Rn, Rd))
-#define SUBw_UXTB(Rd, Rn, Rm)       EMIT(SUB_ext(0, 1, 0, Rm, 0b000, 0, Rn, Rd))
+#define ADDSUB_ext(sf, op, S, Rm, option, imm3, Rn, Rd)    ((sf)<<31 | (op)<<30 | (S)<<29 | 0b01011<<24 | 1<<21 | (Rm)<<16 | (option)<<13 | (imm3)<<10 | (Rn)<<5 | (Rd))
+#define SUBxw_UXTB(Rd, Rn, Rm)      EMIT(ADDSUB_ext(rex.w, 1, 0, Rm, 0b000, 0, Rn, Rd))
+#define SUBw_UXTB(Rd, Rn, Rm)       EMIT(ADDSUB_ext(0, 1, 0, Rm, 0b000, 0, Rn, Rd))
+#define SUBw_SXTB(Rd, Rn, Rm)       EMIT(ADDSUB_ext(0, 1, 0, Rm, 0b100, 0, Rn, Rd))
+#define SUBw_UXTH(Rd, Rn, Rm)       EMIT(ADDSUB_ext(0, 1, 0, Rm, 0b001, 0, Rn, Rd))
+#define SUBx_UXTW(Rd, Rn, Rm)       EMIT(ADDSUB_ext(1, 1, 0, Rm, 0b010, 0, Rn, Rd))
+#define SUBw_SXTH(Rd, Rn, Rm)       EMIT(ADDSUB_ext(0, 1, 0, Rm, 0b101, 0, Rn, Rd))
+#define SUBx_SXTW(Rd, Rn, Rm)       EMIT(ADDSUB_ext(1, 1, 0, Rm, 0b110, 0, Rn, Rd))
+#define ADDz_UXTB(Rd, Rn, Rm)       EMIT(ADDSUB_ext(rex.is32bits?0:1, 0, 0, Rm, 0b000, 0, Rn, Rd))
+#define ADDw_UXTH(Rd, Rn, Rm)       EMIT(ADDSUB_ext(0, 0, 0, Rm, 0b001, 0, Rn, Rd))
+#define ADDx_UXTW(Rd, Rn, Rm)       EMIT(ADDSUB_ext(1, 0, 0, Rm, 0b010, 0, Rn, Rd))
+#define ADDx_SXTW(Rd, Rn, Rm)       EMIT(ADDSUB_ext(1, 0, 0, Rm, 0b110, 0, Rn, Rd))
+#define ADDx_REGy(Rd, Rn, Rm)       do {if(rex.is32bits || rex.is67) ADDx_SXTW(Rd, Rn, Rm); else ADDx_REG(Rd, Rn, Rm); } while(0)

 // CCMP compare if cond is true, set nzcv if false
 #define CCMP_reg(sf, Rm, cond, Rn, nzcv)    ((sf)<<31 | 1<<30 | 1<<29 | 0b11010010<<21 | (Rm)<<16 | (cond)<<12 | (Rn)<<5 | (nzcv))
@@ -286,6 +164,7 @@ int convert_bitmask(uint64_t bitmask);
 #define LDRH_S9_postindex(Rt, Rn, imm9)   EMIT(LDR_gen(0b01, 0b00, (imm9)&0x1ff, 0b01, Rn, Rt))
 #define LDRH_S9_preindex(Rt, Rn, imm9)    EMIT(LDR_gen(0b01, 0b00, (imm9)&0x1ff, 0b11, Rn, Rt))
 #define LDRxw_S9_postindex(Rt, Rn, imm9)  EMIT(LDR_gen(rex.w?0b11:0b10, 0b00, (imm9)&0x1ff, 0b01, Rn, Rt))
+#define LDRz_S9_preindex(Rt, Rn, imm9)  EMIT(LDR_gen(rex.is32bits?0b10:0b11, 0b00, (imm9)&0x1ff, 0b11, Rn, Rt))

 #define LDRS_gen(size, op1, imm9, op2, Rn, Rt)   ((size)<<30 | 0b111<<27 | (op1)<<24 | 0b10<<22 | (imm9)<<12 | (op2)<<10 | (Rn)<<5 | (Rt))
 #define LDRSW_S9_postindex(Rt, Rn, imm9)  EMIT(LDRS_gen(0b10, 0b00, (imm9)&0x1ff, 0b01, Rn, Rt))
@@ -301,30 +180,24 @@ int convert_bitmask(uint64_t bitmask);

 #define LDS_gen(size, op1, imm12, Rn, Rt)       ((size)<<30 | 0b111<<27 | (op1)<<24 | 0b10<<22 | (imm12)<<10 | (Rn)<<5 | (Rt))
 #define LDRSW_U12(Rt, Rn, imm12)          EMIT(LDS_gen(0b10, 0b01, ((uint32_t)((imm12)>>2))&0xfff, Rn, Rt))
+#define LDRSH_U12(Rt, Rn, imm12)          EMIT(LDS_gen(0b01, 0b01, ((uint32_t)((imm12)>>1))&0xfff, Rn, Rt))
+#define LDRSB_U12(Rt, Rn, imm12)          EMIT(LDS_gen(0b00, 0b01, ((uint32_t)(imm12))&0xfff, Rn, Rt))

 #define LDR_REG_gen(size, Rm, option, S, Rn, Rt)    ((size)<<30 | 0b111<<27 | 0b01<<22 | 1<<21 | (Rm)<<16 | (option)<<13 | (S)<<12 | (0b10)<<10 | (Rn)<<5 | (Rt))
 #define LDRx_REG(Rt, Rn, Rm)            EMIT(LDR_REG_gen(0b11, Rm, 0b011, 0, Rn, Rt))
 #define LDRx_REG_LSL3(Rt, Rn, Rm)       EMIT(LDR_REG_gen(0b11, Rm, 0b011, 1, Rn, Rt))
 #define LDRx_REG_UXTW3(Rt, Rn, Rm)      EMIT(LDR_REG_gen(0b11, Rm, 0b010, 1, Rn, Rt))
 #define LDRx_REG_SXTW(Rt, Rn, Rm)       EMIT(LDR_REG_gen(0b11, Rm, 0b110, 0, Rn, Rt))
-#define LDRx_REGz(Rt, Rn, Rm)           EMIT(LDR_REG_gen(0b11, Rm, rex.is32bits?0b110:0b011, 0, Rn, Rt))
 #define LDRw_REG(Rt, Rn, Rm)            EMIT(LDR_REG_gen(0b10, Rm, 0b011, 0, Rn, Rt))
 #define LDRw_REG_LSL2(Rt, Rn, Rm)       EMIT(LDR_REG_gen(0b10, Rm, 0b011, 1, Rn, Rt))
 #define LDRw_REG_SXTW(Rt, Rn, Rm)       EMIT(LDR_REG_gen(0b10, Rm, 0b110, 0, Rn, Rt))
-#define LDRw_REGz(Rt, Rn, Rm)           EMIT(LDR_REG_gen(0b10, Rm, rex.is32bits?0b110:0b011, 0, Rn, Rt))
 #define LDRxw_REG(Rt, Rn, Rm)           EMIT(LDR_REG_gen(0b10+rex.w, Rm, 0b011, 0, Rn, Rt))
-#define LDRz_REG(Rt, Rn, Rm)            EMIT(LDR_REG_gen(rex.is32bits?0b10:0b11, Rm, 0b011, 0, Rn, Rt))
 #define LDRxw_REG_SXTW(Rt, Rn, Rm)      EMIT(LDR_REG_gen(0b10+rex.w, Rm, 0b110, 0, Rn, Rt))
-#define LDRxw_REGz(Rt, Rn, Rm)          EMIT(LDR_REG_gen(0b10+rex.w, Rm, rex.is32bits?0b110:0b011, 0, Rn, Rt))
-#define LDRz_REG_SXTW(Rt, Rn, Rm)       EMIT(LDR_REG_gen(rex.is32bits?0b10:0b11, Rm, 0b110, 0, Rn, Rt))
-#define LDRz_REGz(Rt, Rn, Rm)           EMIT(LDR_REG_gen(rex.is32bits?0b10:0b11, Rm, rex.is32bits?0b110:0b011, 0, Rn, Rt))
 #define LDRB_REG(Rt, Rn, Rm)            EMIT(LDR_REG_gen(0b00, Rm, 0b011, 0, Rn, Rt))
 #define LDRB_REG_UXTW(Rt, Rn, Rm)       EMIT(LDR_REG_gen(0b00, Rm, 0b010, 0, Rn, Rt))
 #define LDRB_REG_SXTW(Rt, Rn, Rm)       EMIT(LDR_REG_gen(0b00, Rm, 0b110, 0, Rn, Rt))
-#define LDRB_REGz(Rt, Rn, Rm)           EMIT(LDR_REG_gen(0b00, Rm, rex.is32bits?0b110:0b011, 0, Rn, Rt))
 #define LDRH_REG(Rt, Rn, Rm)            EMIT(LDR_REG_gen(0b01, Rm, 0b011, 0, Rn, Rt))
 #define LDRH_REG_SXTW(Rt, Rn, Rm)       EMIT(LDR_REG_gen(0b01, Rm, 0b110, 0, Rn, Rt))
-#define LDRH_REGz(Rt, Rn, Rm)           EMIT(LDR_REG_gen(0b01, Rm, rex.is32bits?0b110:0b011, 0, Rn, Rt))

 #define LDRS_U12_gen(size, op1, opc, imm12, Rn, Rt)    ((size)<<30 | 0b111<<27 | (op1)<<24 | (opc)<<22 | (imm12)<<10 | (Rn)<<5 | (Rt))
 #define LDRSHx_U12(Rt, Rn, imm12)           EMIT(LDRS_U12_gen(0b01, 0b01, 0b10, ((uint32_t)(imm12>>1))&0xfff, Rn, Rt))
@@ -337,7 +210,6 @@ int convert_bitmask(uint64_t bitmask);
 #define LDRS_REG_gen(size, Rm, option, S, Rn, Rt)    ((size)<<30 | 0b111<<27 | 0b10<<22 | 1<<21 | (Rm)<<16 | (option)<<13 | (S)<<12 | (0b10)<<10 | (Rn)<<5 | (Rt))
 #define LDRSW_REG(Rt, Rn, Rm)           EMIT(LDRS_REG_gen(0b10, Rm, 0b011, 0, Rn, Rt))
 #define LDRSW_REG_SXTW(Rt, Rn, Rm)      EMIT(LDRS_REG_gen(0b10, Rm, 0b110, 0, Rn, Rt))
-#define LDRSW_REGz(Rt, Rn, Rm)          EMIT(LDRS_REG_gen(0b10, Rm, rex.is32bits?0b110:0b011, 0, Rn, Rt))

 #define LDR_PC_gen(opc, imm19, Rt)      ((opc)<<30 | 0b011<<27 | (imm19)<<5 | (Rt))
 #define LDRx_literal(Rt, imm21)         EMIT(LDR_PC_gen(0b01, (((int64_t)(imm21))>>2)&0x7FFFF, Rt))
@@ -537,6 +409,9 @@ int convert_bitmask(uint64_t bitmask);
 #define DSB_ST()                        EMIT(DSB_gen(0b1110))
 #define DSB_SY()                        EMIT(DSB_gen(0b1111))

+#define DC_gen(op1, CRm, op2, Xt)       (0b1101010100001<<19 | (op1)<<16 | 0b0111<<12 | (CRm)<<8 | (op2)<<5 | Xt)
+#define DC_CIVAC(Xt)                    EMIT(DC_gen(0b011, 0b1110, 0b001, Xt))
+
 // Break
 #define BRK_gen(imm16)                  (0b11010100<<24 | 0b001<<21 | (((imm16)&0xffff)<<5))
 #define BRK(imm16)                      EMIT(BRK_gen(imm16))
@@ -551,11 +426,11 @@ int convert_bitmask(uint64_t bitmask);
 #define CBNZx(Rt, imm19)                EMIT(CB_gen(1, 1, ((imm19)>>2)&0x7FFFF, Rt))
 #define CBNZw(Rt, imm19)                EMIT(CB_gen(0, 1, ((imm19)>>2)&0x7FFFF, Rt))
 #define CBNZxw(Rt, imm19)               EMIT(CB_gen(rex.w, 1, ((imm19)>>2)&0x7FFFF, Rt))
-#define CBNZz(Rt, imm19)                EMIT(CB_gen(rex.is32bits?0:1, 1, ((imm19)>>2)&0x7FFFF, Rt))
+#define CBNZy(Rt, imm19)                EMIT(CB_gen((rex.is32bits || rex.is67)?0:1, 1, ((imm19)>>2)&0x7FFFF, Rt))
 #define CBZx(Rt, imm19)                 EMIT(CB_gen(1, 0, ((imm19)>>2)&0x7FFFF, Rt))
 #define CBZw(Rt, imm19)                 EMIT(CB_gen(0, 0, ((imm19)>>2)&0x7FFFF, Rt))
 #define CBZxw(Rt, imm19)                EMIT(CB_gen(rex.w, 0, ((imm19)>>2)&0x7FFFF, Rt))
-#define CBZz(Rt, imm19)                 EMIT(CB_gen(rex.is32bits?0:1, 0, ((imm19)>>2)&0x7FFFF, Rt))
+#define CBZy(Rt, imm19)                 EMIT(CB_gen((rex.is32bits || rex.is67)?0:1, 0, ((imm19)>>2)&0x7FFFF, Rt))

 #define TB_gen(b5, op, b40, imm14, Rt)  ((b5)<<31 | 0b011011<<25 | (op)<<24  | (b40)<<19 | (imm14)<<5 | (Rt))
 #define TBZ(Rt, bit, imm16)             EMIT(TB_gen(((bit)>>5)&1, 0, (bit)&0x1f, ((imm16)>>2)&0x3FFF, Rt))
@@ -688,7 +563,7 @@ int convert_bitmask(uint64_t bitmask);
 #define BICw_LSL(Rd, Rn, Rm, lsl)       EMIT(LOGIC_REG_gen(0, 0b00, 0b00, 1, Rm, lsl, Rn, Rd))
 #define BICSx(Rd, Rn, Rm)              FEMIT(LOGIC_REG_gen(1, 0b11, 0b00, 1, Rm, 0, Rn, Rd))
 #define BICSw(Rd, Rn, Rm)              FEMIT(LOGIC_REG_gen(0, 0b11, 0b00, 1, Rm, 0, Rn, Rd))
-#define BICxw(Rd, Rn, Rm)              FEMIT(LOGIC_REG_gen(rex.w, 0b00, 0b00, 1, Rm, 0, Rn, Rd))
+#define BICxw(Rd, Rn, Rm)               EMIT(LOGIC_REG_gen(rex.w, 0b00, 0b00, 1, Rm, 0, Rn, Rd))
 #define BICSxw(Rd, Rn, Rm)             FEMIT(LOGIC_REG_gen(rex.w, 0b11, 0b00, 1, Rm, 0, Rn, Rd))
 #define BICx_REG    BICx
 #define BICw_REG    BICw
@@ -748,6 +623,7 @@ int convert_bitmask(uint64_t bitmask);
 #define LSLx(Rd, Rn, lsl)               UBFMx(Rd, Rn, ((-(lsl))%64)&63, 63-(lsl))
 #define LSLw(Rd, Rn, lsl)               UBFMw(Rd, Rn, ((-(lsl))%32)&31, 31-(lsl))
 #define LSLxw(Rd, Rn, lsl)              UBFMxw(Rd, Rn, rex.w?(((-(lsl))%64)&63):(((-(lsl))%32)&31), (rex.w?63:31)-(lsl))
+#define LSLy(Rd, Rn, lsl)               do {if(rex.is32bits || rex.is67) LSLw(Rd, Rn, lsl); else LSLx(Rd, Rn, lsl);} while(0)
 // Take width first bits from Rn, LSL lsb and create Rd
 #define UBFIZx(Rd, Rn, lsb, width)      UBFMx(Rd, Rn, ((-(lsb))%64)&63, width-1)
 // Take width first bits from Rn, LSL lsb and create Rd
@@ -2208,6 +2084,7 @@ int convert_bitmask(uint64_t bitmask);
 #define LDADDAxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 1, 0, Rs, 0b000, Rn, Rt))
 #define LDADDALxw(Rs, Rt, Rn)           EMIT(ATOMIC_gen(0b10+rex.w, 1, 1, Rs, 0b000, Rn, Rt))
 #define LDADDLxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b000, Rn, Rt))
+#define STADDLw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10, 0, 1, Rs, 0b000, Rn, 0b11111))
 #define STADDxw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b000, Rn, 0b11111))
 #define STADDLxw(Rs, Rn)                EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b000, Rn, 0b11111))
 #define LDADDB(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b000, Rn, Rt))
@@ -2487,4 +2364,14 @@ int convert_bitmask(uint64_t bitmask);
 //SHA256 hash update (part 2)
 #define SHA256H2(Vd, Vn, Vm)        EMIT(SHA256H2_gen(Vm, Vn, Vd))

+// LRCPC extension
+
+#define LDAPR_gen(sz, Rn, Rt)   ((sz)<<30 | 0b111<<27 | 1<<23 | 1<<21 | 0b11111<<16 | 1<<15 | 0b100<<12 | (Rn)<<5 | (Rt))
+#define LDAPRx(Rt, Rn)      EMIT(LDAPR_gen(0b11, Rn, Rt))
+#define LDAPRw(Rt, Rn)      EMIT(LDAPR_gen(0b10, Rn, Rt))
+#define LDAPRxw(Rt, Rn)     EMIT(LDAPR_gen(0b10+rex.w, Rn, Rt))
+#define LDAPRz(Rt, Rn)      EMIT(LDAPR_gen(rex.is32bits?0b10:0b11, Rn, Rt))
+#define LDAPRH(Rt, Rn)      EMIT(LDAPR_gen(0b01, Rn, Rt))
+#define LDAPRB(Rt, Rn)      EMIT(LDAPR_gen(0b00, Rn, Rt))
+
 #endif  //__ARM64_EMITTER_H__
@@ -20,7 +20,7 @@ arm64_epilog:
    stp     x26, x27, [x0, (8 * 16)] // put back reg value in emu, including EIP (so x27 must be EIP now)
    //restore all used register
 #ifdef _WIN32
-    ldr     x18, [x0, 3120]
+    ldr     x18, [x0, 3104]
 #endif
    add     sp, x28, 0
    //vpop     {d8-d15}
@@ -5,7 +5,7 @@
 .text
 .align 4

-.extern arm64_atomics
+.extern cpuext
 .global arm64_lock_read_b
 .global arm64_lock_write_b
 .global arm64_lock_read_h
@@ -23,6 +23,7 @@
 .global arm64_lock_storeifnull
 .global arm64_lock_storeifnull_d
 .global arm64_lock_storeifref
+.global arm64_lock_storeifref2
 .global arm64_lock_storeifref_d
 .global arm64_lock_storeifref2_d
 .global arm64_lock_decifnot0b
@@ -106,10 +107,10 @@ arm64_lock_write_dq:


 arm64_lock_xchg_dd:
-    adrp    x3, arm64_atomics
-    add     x3, x3, #:lo12:arm64_atomics
+    adrp    x3, cpuext
+    add     x3, x3, #:lo12:cpuext
    ldr     w3, [x3]
-    cbnz    w3, arm64_atomic_xchg_dd
+    tbnz    w3, #0, arm64_atomic_xchg_dd
    dmb     ish
 arm64_lock_xchg_dd_0:
    // address is x0, value is x1, return old value in x0
@@ -126,10 +127,10 @@ arm64_atomic_xchg_dd:
    ret

 arm64_lock_xchg_d:
-    adrp    x3, arm64_atomics
-    add     x3, x3, #:lo12:arm64_atomics
+    adrp    x3, cpuext
+    add     x3, x3, #:lo12:cpuext
    ldr     w3, [x3]
-    cbnz    w3, arm64_atomic_xchg_d
+    tbnz    w3, #0, arm64_atomic_xchg_d
    dmb     ish
 arm64_lock_xchg_d_0:
    // address is x0, value is x1, return old value in x0
@@ -146,10 +147,10 @@ arm64_atomic_xchg_d:
    ret

 arm64_lock_xchg_h:
-    adrp    x3, arm64_atomics
-    add     x3, x3, #:lo12:arm64_atomics
+    adrp    x3, cpuext
+    add     x3, x3, #:lo12:cpuext
    ldr     w3, [x3]
-    cbnz    w3, arm64_atomic_xchg_h
+    tbnz    w3, #0, arm64_atomic_xchg_h
    dmb     ish
 arm64_lock_xchg_h_0:
    // address is x0, value is x1, return old value in x0
@@ -166,10 +167,10 @@ arm64_atomic_xchg_h:
    ret

 arm64_lock_xchg_b:
-    adrp    x3, arm64_atomics
-    add     x3, x3, #:lo12:arm64_atomics
+    adrp    x3, cpuext
+    add     x3, x3, #:lo12:cpuext
    ldr     w3, [x3]
-    cbnz    w3, arm64_atomic_xchg_b
+    tbnz    w3, #0, arm64_atomic_xchg_b
    dmb     ish
 arm64_lock_xchg_b_0:
    // address is x0, value is x1, return old value in x0
@@ -186,10 +187,10 @@ arm64_atomic_xchg_b:
    ret

 arm64_lock_storeifnull:
-    adrp    x3, arm64_atomics
-    add     x3, x3, #:lo12:arm64_atomics
+    adrp    x3, cpuext
+    add     x3, x3, #:lo12:cpuext
    ldr     w3, [x3]
-    cbnz    w3, arm64_atomic_storeifnull
+    tbnz    w3, #0, arm64_atomic_storeifnull
    dmb     ish
 1:
    // address is x0, value is x1, x1 store to x0 only if [x0] is 0. return old [x0] value
@@ -211,10 +212,10 @@ arm64_atomic_storeifnull:


 arm64_lock_storeifnull_d:
-    adrp    x3, arm64_atomics
-    add     x3, x3, #:lo12:arm64_atomics
+    adrp    x3, cpuext
+    add     x3, x3, #:lo12:cpuext
    ldr     w3, [x3]
-    cbnz    w3, arm64_atomic_storeifnull_d
+    tbnz    w3, #0, arm64_atomic_storeifnull_d
    dmb     ish
 1:
    // address is x0, value is w1, w1 store to x0 only if [x0] is 0. return old [x0] value
@@ -236,10 +237,10 @@ arm64_atomic_storeifnull_d:
    ret

 arm64_lock_storeifref:
-    adrp    x3, arm64_atomics
-    add     x3, x3, #:lo12:arm64_atomics
+    adrp    x3, cpuext
+    add     x3, x3, #:lo12:cpuext
    ldr     w3, [x3]
-    cbnz    w3, arm64_atomic_storeifref
+    tbnz    w3, #0, arm64_atomic_storeifref
    dmb     ish
 1:
    // address is x0, value is x1, x1 store to x0 only if [x0] is x2. return new [x0] value (so x1 or old value)
@@ -260,6 +261,7 @@ arm64_atomic_storeifref:
    mov     x3, x2
    casal   x2, x1, [x0]
    cmp     x2, x3
+    bne     2f
    mov     x0, x1
    ret
 2:
@@ -267,10 +269,10 @@ arm64_atomic_storeifref:
    ret

 arm64_lock_storeifref_d:
-    adrp    x3, arm64_atomics
-    add     x3, x3, #:lo12:arm64_atomics
+    adrp    x3, cpuext
+    add     x3, x3, #:lo12:cpuext
    ldr     w3, [x3]
-    cbnz    w3, arm64_atomic_storeifref_d
+    tbnz    w3, #0, arm64_atomic_storeifref_d
    dmb     ish
 1:
    // address is x0, value is w1, w1 store to x0 only if [x0] is w2. return new [x0] value (so x1 or old value)
@@ -290,6 +292,7 @@ arm64_atomic_storeifref_d:
    // address is x0, value is w1, w1 store to x0 only if [x0] is w2. return new [x0] value (so w1 or old value)
    mov     w3, w2
    casal   w2, w1, [x0]
+    bne     2f
    cmp     w2, w3
    mov     w0, w1
    ret
@@ -298,10 +301,10 @@ arm64_atomic_storeifref_d:
    ret

 arm64_lock_storeifref2_d:
-    adrp    x3, arm64_atomics
-    add     x3, x3, #:lo12:arm64_atomics
+    adrp    x3, cpuext
+    add     x3, x3, #:lo12:cpuext
    ldr     w3, [x3]
-    cbnz    w3, arm64_atomic_storeifref2_d
+    tbnz    w3, #0, arm64_atomic_storeifref2_d
    dmb     ish
 1:
    // address is x0, value is w1, w1 store to x0 only if [x0] is w2. return old [x0] value
@@ -321,6 +324,30 @@ arm64_atomic_storeifref2_d:
    mov     w0, w2
    ret

+arm64_lock_storeifref2:
+    adrp    x3, cpuext
+    add     x3, x3, #:lo12:cpuext
+    ldr     w3, [x3]
+    tbnz    w3, #0, arm64_atomic_storeifref2
+    dmb     ish
+1:
+    // address is x0, value is x1, x1 store to x0 only if [x0] is x2. return old [x0] value
+    ldaxr   x3, [x0]
+    cmp     x2, x3
+    bne     2f
+    stlxr   w4, x1, [x0]
+    cbnz    w4, 1b
+2:
+    mov     x0, x3
+    ret
+
+arm64_atomic_storeifref2:
+    dmb     ish
+    // address is x0, value is x1, x1 store to x0 only if [x0] is x2. return old [x0] value
+    casal   x2, x1, [x0]
+    mov     x0, x2
+    ret
+
 arm64_lock_decifnot0b:
    dmb     ish
 1:
@@ -352,10 +379,10 @@ arm64_lock_decifnot0:
    ret

 arm64_lock_incif0:
-    adrp    x3, arm64_atomics
-    add     x3, x3, #:lo12:arm64_atomics
+    adrp    x3, cpuext
+    add     x3, x3, #:lo12:cpuext
    ldr     w3, [x3]
-    cbnz    w3, arm64_atomic_incif0
+    tbnz    w3, #0, arm64_atomic_incif0
    dmb     ish
 1:
    ldaxr   w1, [x0]
@@ -55,11 +55,15 @@ extern void* arm64_atomic_storeifnull(void*p, void* val);
 extern void* arm64_lock_storeifref(void*p, void* val, void* ref);
 extern void* arm64_atomic_storeifref(void*p, void* val, void* ref);

+// Atomically store value to [p] only if [p] is ref. Return old [p] value (so val or old)
+extern void* arm64_lock_storeifref2(void*p, void* val, void* ref);
+extern void* arm64_atomic_storeifref2(void*p, void* val, void* ref);
+
 // Atomically store value to [p] only if [p] is ref. Return new [p] value (so val or old)
 extern uint32_t arm64_lock_storeifref_d(void*p, uint32_t val, uint32_t ref);
 extern uint32_t arm64_atomic_storeifref_d(void*p, uint32_t val, uint32_t ref);

-// Atomically store value to [p] only if [p] is ref. Return new [p] value (so val or old)
+// Atomically store value to [p] only if [p] is ref. Return old [p] value (so ref or old)
 extern uint32_t arm64_lock_storeifref2_d(void*p, uint32_t val, uint32_t ref);
 extern uint32_t arm64_atomic_storeifref2_d(void*p, uint32_t val, uint32_t ref);

@@ -0,0 +1,149 @@
+#ifndef __ARM64_MAPPING_H__
+#define __ARM64_MAPPING_H__
+
+
+/* 
+    ARM64 Linux Call Convention
+
+SP          The Stack Pointer.
+r30 LR      The Link Register.
+r29 FP      The Frame Pointer
+r19…r28     Callee-saved registers
+r18         The Platform Register, if needed; otherwise a temporary register. See notes.
+r17 IP1     The second intra-procedure-call temporary register (can be used by call veneers and PLT code); at other times may be used as a temporary register.
+r16 IP0     The first intra-procedure-call scratch register (can be used by call veneers and PLT code); at other times may be used as a temporary register.
+r9…r15      Temporary registers
+r8          Indirect result location register
+r0…r7       Parameter/result registers
+
+For SIMD:
+The first eight registers, v0-v7, are used to pass argument values into a subroutine and to return result values from a function. 
+    They may also be used to hold intermediate values within a routine (but, in general, only between subroutine calls).
+
+Registers v8-v15 must be preserved by a callee across subroutine calls; 
+    the remaining registers (v0-v7, v16-v31) do not need to be preserved (or should be preserved by the caller).
+    Additionally, only the bottom 64 bits of each value stored in v8-v15 need to be preserved [8];
+    it is the responsibility of the caller to preserve larger values.
+
+For SVE:
+z0-z7 are used to pass scalable vector arguments to a subroutine, and to return scalable vector results from a function.
+    If a subroutine takes at least one argument in scalable vector registers or scalable predicate registers,
+    or returns results in such regisers, the subroutine must ensure that the entire contents of z8-z23 are preserved across the call.
+    In other cases it need only preserve the low 64 bits of z8-z15, as described in SIMD and Floating-Point registers.
+p0-p3 are used to pass scalable predicate arguments to a subroutine and to return scalable predicate results from a function.
+    If a subroutine takes at least one argument in scalable vector registers or scalable predicate registers,
+    or returns results in such registers, the subroutine must ensure that p4-p15 are preserved across the call.
+    In other cases it need not preserve any scalable predicate register contents.
+
+*/
+
+// x86 Register mapping
+#define xRAX    10
+#define xRCX    11
+#define xRDX    12
+#define xRBX    13
+#define xRSP    14
+#define xRBP    15
+#define xRSI    16
+#define xRDI    17
+#define xR8     18
+#define xR9     19
+#define xR10    20
+#define xR11    21
+#define xR12    22
+#define xR13    23
+#define xR14    24
+#define xR15    25
+#define xFlags  26
+#define xRIP    27
+#define xSavedSP 28
+
+// convert a x86 register to native according to the register mapping
+#define TO_NAT(A) (xRAX + (A))
+#define IS_GPR(A) ((A)>=xRAX && (A)<=xRIP)
+
+// 32bits version
+#define wEAX    xRAX
+#define wECX    xRCX
+#define wEDX    xRDX
+#define wEBX    xRBX
+#define wESP    xRSP
+#define wEBP    xRBP
+#define wESI    xRSI
+#define wEDI    xRDI
+#define wR8     xR8
+#define wR9     xR9
+#define wR10    xR10
+#define wR11    xR11
+#define wR12    xR12
+#define wR13    xR13
+#define wR14    xR14
+#define wR15    xR15
+#define wFlags  xFlags
+// scratch registers
+#define x1      1
+#define x2      2
+#define x3      3
+#define x4      4
+#define x5      5
+#define x6      6
+#define x87pc   7
+// x87 can be a scratch, but check if it's used as x87 PC and restore if needed in that case
+// 32bits version of scratch
+#define w1      x1
+#define w2      x2
+#define w3      x3
+#define w4      x4
+#define w5      x5
+#define w6      x6
+#define w87pc   x87pc
+// emu is r0
+#define xEmu    0
+// ARM64 LR
+#define xLR     30
+// ARM64 SP is r31 but is a special register
+#define xSP     31      
+// xZR regs is 31
+#define xZR     31
+#define wZR     xZR
+
+// conditions
+// Z == 1
+#define cEQ 0b0000
+// Z != 1
+#define cNE 0b0001
+// C == 1
+#define cCS 0b0010
+// C == 1
+#define cHS cCS
+// C != 1
+#define cCC 0b0011
+// C != 1
+#define cLO cCC
+// N == 1
+#define cMI 0b0100
+// N != 1
+#define cPL 0b0101
+// V == 1
+#define cVS 0b0110
+// V != 1
+#define cVC 0b0111
+// C == 1 && Z == 0
+#define cHI 0b1000
+// C !=1 || Z == 1
+#define cLS 0b1001
+// N == V
+#define cGE 0b1010
+// N != V
+#define cLT 0b1011
+// N == V && Z == 0
+#define cGT 0b1100
+// N != V || Z == 1
+#define cLE 0b1101
+// always
+#define c__ 0b1110
+
+//FCMP type of opcode produce:
+// if any NAN: CV / v1 == v2: ZC / v1 < v2: N / v1 > v2: C
+
+#endif //__ARM64_MAPPING_H__
@@ -1,7 +1,5 @@
 //arm update linker table for dynarec
 //called with pointer to emu as 1st parameter
-//and address of table to as 2nd parameter
-//ip is at r12

 .text
 .align 4
@@ -23,8 +21,9 @@ arm64_next:
    stp     x18, x27, [sp, (8 * 10)]    // also save x27(rip) to allow change in LinkNext

 #ifdef _WIN32
-    ldr     x18, [x0, 3120]
+    ldr     x18, [x0, 3104]
 #endif
+    mov     x1, x27     // xRIP, should not be needed most of the time
    mov     x2, x30      // "from" is in lr, so put in x2
    add     x3, sp, 8*11    // x3 is address to change rip
    // call the function
@@ -565,6 +565,16 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
            snprintf(buff, sizeof(buff), "SUB %s, %s, %s %s %d", sf?Xt[Rd]:Wt[Rd], sf?Xt[Rn]:Wt[Rn], sf?Xt[Rm]:Wt[Rm], shifts[sf][shift], imm);
        return buff;
    }
+    if(isMask(opcode, "f0001011001mmmmmoooiiinnnnnddddd", &a)) {
+        const char* shifts[2][8] = {{ "UXTB", "UXTH", "LSL", "UXTX", "SXTB", "SXTH", "SXTW", "SXTX"}, {"UXTB", "UXTH", "UXTW", "LSL", "SXTB", "SXTH", "SXTW", "SXTX" }};
+        if(((sf==0 && shift==2) || (sf==1 && shift==3)) && imm==0)
+            snprintf(buff, sizeof(buff), "ADD %s, %s, %s", sf?XtSp[Rd]:WtSp[Rd], sf?XtSp[Rn]:WtSp[Rn], sf?XtSp[Rm]:WtSp[Rm]);
+        else if(shift==0)
+            snprintf(buff, sizeof(buff), "ADD %s, %s, %s %s", sf?Xt[Rd]:Wt[Rd], sf?Xt[Rn]:Wt[Rn], sf?Xt[Rm]:Wt[Rm], shifts[sf][shift]);
+        else
+            snprintf(buff, sizeof(buff), "ADD %s, %s, %s %s %d", sf?Xt[Rd]:Wt[Rd], sf?Xt[Rn]:Wt[Rn], sf?Xt[Rm]:Wt[Rm], shifts[sf][shift], imm);
+        return buff;
+    }
    // ---- LOGIC
    if(isMask(opcode, "f11100100Nrrrrrrssssssnnnnnddddd", &a)) {
        uint64_t i = DecodeBitMasks(a.N, imms, immr);
@@ -1849,9 +1859,9 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
        snprintf(buff, sizeof(buff), "SWP%s%s %s, %s, [%s]", a.A?"A":"", a.R?"L":"", sf?Xt[Rs]:Wt[Rs], sf?Xt[Rt]:Wt[Rt], XtSp[Rn]);
        return buff;
    }
-    // SWPxw
+    // SWP B/H
    if(isMask(opcode, "0f111000AR1sssss100000nnnnnttttt", &a)) {
-        snprintf(buff, sizeof(buff), "CAS%s%s%s %s, %s, [%s]", a.A?"A":"", a.R?"L":"", sf?"H":"B", Xt[Rs], Xt[Rt], XtSp[Rn]);
+        snprintf(buff, sizeof(buff), "SWP%s%s%s %s, %s, [%s]", a.A?"A":"", a.R?"L":"", sf?"H":"B", Xt[Rs], Xt[Rt], XtSp[Rn]);
        return buff;
    }
    // LDXXXxw
@@ -16,15 +16,17 @@
 #include "my_cpuid.h"
 #include "freq.h"
 #include "emu/x87emu_private.h"
-#include "emu/x64shaext.h"
 #include "arm64_printer.h"
 #include "dynarec_arm64_private.h"
 #include "dynarec_arm64_functions.h"
 #include "../dynarec_helper.h"
+#ifndef __WIN32
+#include "elfloader.h"
+#endif

-uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
 {
-    (void)ip; (void)rep; (void)need_epilog;
+    (void)ip; (void)need_epilog;

    uint8_t opcode = F8;
    uint8_t nextop, u8;
@@ -55,12 +57,25 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
    MAYUSE(s0);
    MAYUSE(j64);
    MAYUSE(cacheupd);
-    #if STEP > 1
-    static const int8_t mask_shift8[] = { -7, -6, -5, -4, -3, -2, -1, 0 };
-    #endif

    switch(opcode) {
-
+        case 0x00:
+            nextop = F8;
+            switch((nextop>>3)&7) {
+                case 0:
+                    INST_NAME("SLDT EW");
+                    if(MODREG) {
+                        ed = TO_NAT((nextop & 7) + (rex.b << 3));
+                        MOV32w(ed, 0);
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
+                        STRH_U12(xZR, wback, 0);
+                    }
+                    break;
+                default:
+                    DEFAULT;
+            }
+            break;
        case 0x01:
            // TODO:, /0 is SGDT. While 0F 01 D0 is XGETBV, etc...
            nextop = F8;
@@ -90,7 +105,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    INST_NAME("RDTSCP");
                    NOTEST(x1);
                    if(box64_rdtsc) {
-                        CALL_(ReadTSC, x1, x3);
+                        CALL_(const_readtsc, x1, x3);
                    } else {
                        MRS_cntvct_el0(x1);
                    }
@@ -99,7 +114,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    }
                    LSRx(xRDX, x1, 32);
                    MOVw_REG(xRAX, x1);   // wipe upper part
-                    CALL_(helper_getcpu, x1, x3);
+                    CALL_(const_helper_getcpu, x1, x3);
                    MOVw_REG(xRCX, x1);    // IA32_TSC, 0 for now
                    break;
                default:
@@ -149,7 +164,14 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            SMEND();
            GETIP(addr);
            STORE_XEMU_CALL(xRIP);
-            CALL_S(EmuX64Syscall, -1);
+            #ifndef __WIN32
+            if(!box64_wine || FindElfAddress(my_context, ip)) {
+                CALL_S(const_x64syscall_linux, -1);
+            } else
+            #endif
+            {
+                CALL_S(const_x64syscall, -1);
+            }
            LOAD_XEMU_CALL(xRIP);
            TABLE64(x3, addr); // expected return address
            SUBx_REG(x3, x3, xRIP);
@@ -170,7 +192,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            }
            GETIP(ip);
            STORE_XEMU_CALL(xRIP);
-            CALL(native_ud, -1);
+            CALL_S(const_native_priv, -1);
            LOAD_XEMU_CALL(xRIP);
            jump_to_epilog(dyn, 0, xRIP, ninst);
            *need_epilog = 0;
@@ -184,10 +206,9 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            } else {
                SETFLAGS(X_ALL, SF_SET_NODF);    // Hack to set flags in "don't care" state
            }
+            BARRIER(BARRIER_FLOAT);
            GETIP(ip);
-            STORE_XEMU_CALL(xRIP);
-            CALL(native_ud, -1);
-            LOAD_XEMU_CALL(xRIP);
+            UDF(0);
            jump_to_epilog(dyn, 0, xRIP, ninst);
            *need_epilog = 0;
            *ok = 0;
@@ -225,7 +246,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            }
            GETIP(ip);
            STORE_XEMU_CALL(xRIP);
-            CALL(native_ud, -1);
+            CALL_S(const_native_ud, -1);
            LOAD_XEMU_CALL(xRIP);
            jump_to_epilog(dyn, 0, xRIP, ninst);
            *need_epilog = 0;
@@ -293,12 +314,12 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            break;
        case 0x13:
            nextop = F8;
-            INST_NAME("MOVLPS Ex, Gx");
-            GETGX(v0, 0);
            if(MODREG) {
-                DEFAULT;
-                return addr;
+                INST_NAME("Illegal 0F 13");
+                UDF(0);
            } else {
+                INST_NAME("MOVLPS Ex, Gx");
+                GETGX(v0, 0);
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                VST1_64(v0, 0, ed);  // better to use VST1 than VSTR_64, to avoid NEON->VFPU transfert I assume
                SMWRITE2();
@@ -353,22 +374,22 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            } else
            switch((nextop>>3)&7) {
                case 0:
-                    INST_NAME("PREFETCHh Ed");
+                    INST_NAME("PREFETCHNTA Ed");
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0xfff, 7, rex, NULL, 0, 0);
                    PLD_L1_STREAM_U12(ed, fixedaddress);
                    break;
                case 1:
-                    INST_NAME("PREFETCHh Ed");
+                    INST_NAME("PREFETCHT0 Ed");
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0xfff, 7, rex, NULL, 0, 0);
                    PLD_L1_KEEP_U12(ed, fixedaddress);
                    break;
                case 2:
-                    INST_NAME("PREFETCHh Ed");
+                    INST_NAME("PREFETCHT1 Ed");
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0xfff, 7, rex, NULL, 0, 0);
                    PLD_L2_KEEP_U12(ed, fixedaddress);
                    break;
                case 3:
-                    INST_NAME("PREFETCHh Ed");
+                    INST_NAME("PREFETCHT2 Ed");
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0xfff, 7, rex, NULL, 0, 0);
                    PLD_L3_KEEP_U12(ed, fixedaddress);
                    break;
@@ -379,11 +400,41 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            break;

        case 0x19:
+        case 0x1A:
+        case 0x1B:
+        case 0x1C:
+        case 0x1D:
+        case 0x1E:
        case 0x1F:
            INST_NAME("NOP (multibyte)");
            nextop = F8;
            FAKEED;
            break;
+        case 0x20:                      /* MOV REG, crX */
+        case 0x21:                      /* MOV REG, drX */
+        case 0x22:                      /* MOV cxR, REG */
+        case 0x23:                      /* MOV drX, REG */
+            if(BOX64DRENV(dynarec_safeflags)>1) {
+                READFLAGS(X_PEND);
+            } else {
+                SETFLAGS(X_ALL, SF_SET_NODF);    // Hack to set flags in "don't care" state
+            }
+            nextop = F8;
+            GETIP(ip);
+            u8 = (rex.r*8)+(nextop>>3&7);
+            if((((opcode==0x20) || (opcode==0x22)) && ((u8==1) || (u8==5) || (u8==6) || (u8==7) || (u8>8))) || (((opcode==0x21) || (opcode==0x23)) && rex.r)) {
+                INST_NAME("Illegal 0F 20..23");
+                UDF(0);
+            } else {
+                INST_NAME(opcode==0x20?"MOV REG, crX":(opcode==0x21?"MOV REG, drX":(opcode==0x22?"MOV cxR, REG":"MOV dxR, REG")));
+                STORE_XEMU_CALL(xRIP);
+                CALL_S(const_native_priv, -1);
+                LOAD_XEMU_CALL(xRIP);
+            }
+            jump_to_epilog(dyn, 0, xRIP, ninst);
+            *need_epilog = 0;
+            *ok = 0;
+            break;

        case 0x28:
            INST_NAME("MOVAPS Gx,Ex");
@@ -428,13 +479,14 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            VMOVeD(v0, 0, d0, 0);
            break;
        case 0x2B:
-            INST_NAME("MOVNTPS Ex,Gx");
            nextop = F8;
-            GETG;
-            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
            if(MODREG) {
-                DEFAULT;
+                INST_NAME("Illegal 0F 2B");
+                UDF(0);
            } else {
+                INST_NAME("MOVNTPS Ex,Gx");
+                GETG;
+                v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
                VST128(v0, ed, fixedaddress);
            }
@@ -443,11 +495,11 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            INST_NAME("CVTTPS2PI Gm,Ex");
            nextop = F8;
            GETGM(q0);
-            GETEX(v1, 0, 0);
+            GETEXSD(v1, 0, 0);
            if (BOX64ENV(dynarec_fastround)) {
                VFCVTZSS(q0, v1);
            } else {
-                if(arm64_frintts) {
+                if(cpuext.frintts) {
                    VFRINT32ZS(q0, v1);
                    VFCVTZSS(q0, q0);
                } else {
@@ -477,7 +529,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            INST_NAME("CVTPS2PI Gm, Ex");
            nextop = F8;
            GETGM(q0);
-            GETEX(v1, 0, 0);
+            GETEXSD(v1, 0, 0);
            if (BOX64ENV(dynarec_fastround)) {
                u8 = sse_setround(dyn, ninst, x1, x2, x3);
                VFRINTIS(q0, v1);
@@ -485,7 +537,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                VFCVTZSS(q0, q0);
            } else {
                u8 = sse_setround(dyn, ninst, x1, x2, x3);
-                if(arm64_frintts) {
+                if(cpuext.frintts) {
                    VFRINT32XS(q0, v1);
                    VFCVTZSS(q0, q0);
                } else {
@@ -516,11 +568,13 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            // no special check...
        case 0x2F:
            if(opcode==0x2F) {INST_NAME("COMISS Gx, Ex");} else {INST_NAME("UCOMISS Gx, Ex");}
-            SETFLAGS(X_ALL, SF_SET_NODF);
+            SETFLAGS(X_ALL, SF_SET_DF);
            nextop = F8;
            GETGX(v0, 0);
            GETEXSS(s0, 0, 0);
-            FCMPS(v0, s0);
+            IFX(X_CF|X_PF|X_ZF) {
+                FCMPS(v0, s0);
+            }
            FCOMI(x1, x2);
            break;
        case 0x30:
@@ -532,7 +586,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            }
            GETIP(ip);
            STORE_XEMU_CALL(xRIP);
-            CALL(native_priv, -1);
+            CALL_S(const_native_priv, -1);
            LOAD_XEMU_CALL(xRIP);
            jump_to_epilog(dyn, 0, xRIP, ninst);
            // continue the block
@@ -541,7 +595,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            INST_NAME("RDTSC");
            NOTEST(x1);
            if(box64_rdtsc) {
-                CALL_(ReadTSC, x1, x3);
+                CALL_(const_readtsc, x1, x3);
            } else {
                MRS_cntvct_el0(x1);
            }
@@ -711,7 +765,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    GETEX(q1, 0, 0);
                    v0 = fpu_get_scratch(dyn, ninst);
                    VEORQ(v0, v0, v0);
-                    if(arm64_sha1) {
+                    if(cpuext.sha1) {
                        v1 = fpu_get_scratch(dyn, ninst);
                        VMOVeS(v1, 0, q0, 3);
                        SHA1H(v1, v1);
@@ -735,7 +789,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                case 0xCA:
                    INST_NAME("SHA1MSG2 Gx, Ex");
                    nextop = F8;
-                    if(arm64_sha1) {
+                    if(cpuext.sha1) {
                        GETGX(q0, 1);
                        GETEX(q1, 0, 0);
                        VEXTQ_8(q0, q0, q0, 8);
@@ -771,13 +825,13 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                        GETG;
                        sse_forget_reg(dyn, ninst, gd);
                        ADDx_U12(x1, xEmu, offsetof(x64emu_t, xmm[gd]));
-                        CALL(sha1msg2, -1);
+                        CALL(const_sha1msg2, -1);
                    }
                    break;
                case 0xCB:
                    INST_NAME("SHA256RNDS2 Gx, Ex (, XMM0)");
                    nextop = F8;
-                    if(arm64_sha2) {
+                    if(cpuext.sha2) {
                        GETGX(q0, 1);
                        GETEX(q1, 0, 0);
                        d0 = sse_get_reg(dyn, ninst, x1, 0, 0);
@@ -817,13 +871,13 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                        sse_forget_reg(dyn, ninst, gd);
                        ADDx_U12(x1, xEmu, offsetof(x64emu_t, xmm[gd]));
                        sse_reflect_reg(dyn, ninst, 0);
-                        CALL(sha256rnds2, -1);
+                        CALL(const_sha256rnds2, -1);
                    }
                    break;
                case 0xCC:
                    INST_NAME("SHA256MSG1 Gx, Ex");
                    nextop = F8;
-                    if(arm64_sha2) {
+                    if(cpuext.sha2) {
                        GETGX(q0, 1);
                        GETEX(q1, 0, 0);
                        SHA256SU0(q0, q1);
@@ -842,13 +896,13 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                        GETG;
                        sse_forget_reg(dyn, ninst, gd);
                        ADDx_U12(x1, xEmu, offsetof(x64emu_t, xmm[gd]));
-                        CALL(sha256msg1, -1);
+                        CALL(const_sha256msg1, -1);
                    }
                    break;
                case 0xCD:
                    INST_NAME("SHA256MSG2 Gx, Ex");
                    nextop = F8;
-                    if(arm64_sha2) {
+                    if(cpuext.sha2) {
                        GETGX(q0, 1);
                        GETEX(q1, 0, 0);
                        v0 = fpu_get_scratch(dyn, ninst);
@@ -876,7 +930,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                        GETG;
                        sse_forget_reg(dyn, ninst, gd);
                        ADDx_U12(x1, xEmu, offsetof(x64emu_t, xmm[gd]));
-                        CALL(sha256msg2, -1);
+                        CALL(const_sha256msg2, -1);
                    }
                    break;

@@ -935,7 +989,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                case 0xCC:
                    INST_NAME("SHA1RNDS4 Gx, Ex, Ib");
                    nextop = F8;
-                    if(arm64_sha1) {
+                    if(cpuext.sha1) {
                        GETGX(q0, 1);
                        GETEX(q1, 0, 1);
                        u8 = F8&3;
@@ -999,7 +1053,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                        sse_forget_reg(dyn, ninst, gd);
                        ADDx_U12(x1, xEmu, offsetof(x64emu_t, xmm[gd]));
                        MOV32w(x3, u8);
-                        CALL(sha1rnds4, -1);
+                        CALL(const_sha1rnds4, -1);
                    }
                    break;

@@ -1199,7 +1253,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
        case 0x5A:
            INST_NAME("CVTPS2PD Gx, Ex");
            nextop = F8;
-            GETEX(q0, 0, 0);
+            GETEXSD(q0, 0, 0);
            GETGX_empty(q1);
            FCVTL(q1, q0);
            break;
@@ -1237,13 +1291,9 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            GETEX(v1, 0, 0);
            // FMIN/FMAX wll not copy the value if v0[x] is NaN
            // but x86 will copy if either v0[x] or v1[x] is NaN, so lets force a copy if source is NaN
-            if(BOX64ENV(dynarec_fastnan)) {
-                VFMINQS(v0, v0, v1);
-            } else {
-                q0 = fpu_get_scratch(dyn, ninst);
-                VFCMGTQS(q0, v1, v0);   // 0 is NaN or v1 GT v0, so invert mask for copy
-                VBIFQ(v0, v1, q0);
-            }
+            q0 = fpu_get_scratch(dyn, ninst);
+            VFCMGTQS(q0, v1, v0);   // 0 is NaN or v1 GT v0, so invert mask for copy
+            VBIFQ(v0, v1, q0);
            break;
        case 0x5E:
            INST_NAME("DIVPS Gx, Ex");
@@ -1272,13 +1322,9 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            GETEX(v1, 0, 0);
            // FMIN/FMAX wll not copy the value if v0[x] is NaN
            // but x86 will copy if either v0[x] or v1[x] is NaN, or if values are equals, so lets force a copy if source is NaN
-            if(BOX64ENV(dynarec_fastnan)) {
-                VFMAXQS(v0, v0, v1);
-            } else {
-                q0 = fpu_get_scratch(dyn, ninst);
-                VFCMGTQS(q0, v0, v1);   // 0 is NaN or v0 GT v1, so invert mask for copy
-                VBIFQ(v0, v1, q0);
-            }
+            q0 = fpu_get_scratch(dyn, ninst);
+            VFCMGTQS(q0, v0, v1);   // 0 is NaN or v0 GT v1, so invert mask for copy
+            VBIFQ(v0, v1, q0);
            break;
        case 0x60:
            INST_NAME("PUNPCKLBW Gm,Em");
@@ -1630,6 +1676,11 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            INST_NAME("EMMS");
            // empty MMX, FPU now usable
            mmx_purgecache(dyn, ninst, 0, x1);
+            x87_purgecache(dyn, ninst, 1, x1, x2, x3);  // also purge x87 and hard reset it
+            STRw_U12(xZR, xEmu, offsetof(x64emu_t, top));
+            STRw_U12(xZR, xEmu, offsetof(x64emu_t, fpu_stack));
+            MOV64x(x1, TAGS_EMPTY);
+            STRx_U12(x1, xEmu, offsetof(x64emu_t, fpu_tags));
            /*emu->top = 0;
            emu->fpu_stack = 0;*/ //TODO: Check if something is needed here?
            break;
@@ -1677,7 +1728,6 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                j64 = (uint32_t)(addr+i32_);                            \
            else                                                        \
                j64 = addr+i32_;                                        \
-            BARRIER(BARRIER_MAYBE);                                     \
            JUMP(j64, 1);                                               \
            GETFLAGS;                                                   \
            if(dyn->insts[ninst].x64.jmp_insts==-1 ||                   \
@@ -1687,10 +1737,10 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                Bcond(NO, i32);                                         \
                if(dyn->insts[ninst].x64.jmp_insts==-1) {               \
                    if(!(dyn->insts[ninst].x64.barrier&BARRIER_FLOAT))  \
-                        fpu_purgecache(dyn, ninst, 1, x1, x2, x3);      \
+                        fpu_purgecache(dyn, ninst, 1, x1, x2, x3, 0);   \
                    jump_to_next(dyn, j64, 0, ninst, rex.is32bits);     \
                } else {                                                \
-                    CacheTransform(dyn, ninst, cacheupd, x1, x2, x3);   \
+                    CacheTransform(dyn, ninst, cacheupd);               \
                    i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->native_size);    \
                    SKIP_SEVL(i32);                                     \
                    B(i32);                                             \
@@ -1737,17 +1787,20 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            INST_NAME("POP FS");
            POP1z(x2);
            STRH_U12(x2, xEmu, offsetof(x64emu_t, segs[_FS]));
-            STRw_U12(xZR, xEmu, offsetof(x64emu_t, segs_serial[_FS]));
+            CBZw_NEXT(x2);
+            MOV32w(x1, _FS);
+            CALL(const_getsegmentbase, -1);
            break;
        case 0xA2:
            INST_NAME("CPUID");
            NOTEST(x1);
-            MOVx_REG(x1, xRAX);
-            CALL_(my_cpuid, -1, 0);
+            GETIP(ip);  // sync RIP for easier debugging
+            STRx_U12(xRIP, xEmu, offsetof(x64emu_t, ip));
+            CALL_(const_cpuid, -1, 0);
            break;
        case 0xA3:
            INST_NAME("BT Ed, Gd");
-            if(!BOX64ENV(dynarec_safeflags)) {
+            if (!BOX64DRENV(dynarec_safeflags)) {
                SETFLAGS(X_ALL&~X_ZF, SF_SUBSET);
            } else {
                SETFLAGS(X_CF, SF_SUBSET);
@@ -1761,7 +1814,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<(2+rex.w), (1<<(2+rex.w))-1, rex, NULL, 0, 0);
                ASRxw(x1, gd, 5+rex.w); // r1 = (gd>>5)
                if(!rex.w && !rex.is32bits) {SXTWx(x1, x1);}
-                ADDz_REG_LSL(x3, wback, x1, 2+rex.w); //(&ed)+=r1*4;
+                ADDy_REG_LSL(x3, wback, x1, 2+rex.w); //(&ed)+=r1*4;
                LDxw(x1, x3, fixedaddress);
                ed = x1;
            }
@@ -1787,6 +1840,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                WBACK;
            } else {
                FAKEED;
+                if (!rex.w && !rex.is32bits && MODREG) { MOVw_REG(ed, ed); }
                F8;
            }
            break;
@@ -1820,12 +1874,14 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            INST_NAME("POP GS");
            POP1z(x2);
            STRH_U12(x2, xEmu, offsetof(x64emu_t, segs[_GS]));
-            STRw_U12(xZR, xEmu, offsetof(x64emu_t, segs_serial[_GS]));
+            CBZw_NEXT(x2);
+            MOV32w(x1, _GS);
+            CALL(const_getsegmentbase, -1);
            break;

        case 0xAB:
            INST_NAME("BTS Ed, Gd");
-            if(!BOX64ENV(dynarec_safeflags)) {
+            if (!BOX64DRENV(dynarec_safeflags)) {
                SETFLAGS(X_ALL&~X_ZF, SF_SUBSET);
            } else {
                SETFLAGS(X_CF, SF_SUBSET);
@@ -1840,7 +1896,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<(2+rex.w), (1<<(2+rex.w))-1, rex, NULL, 0, 0);
                ASRxw(x1, gd, 5+rex.w); // r1 = (gd>>5)
                if(!rex.w && !rex.is32bits) {SXTWx(x1, x1);}
-                ADDz_REG_LSL(x3, wback, x1, 2+rex.w); //(&ed)+=r1*4;
+                ADDy_REG_LSL(x3, wback, x1, 2+rex.w); //(&ed)+=r1*4;
                LDxw(x1, x3, fixedaddress);
                ed = x1;
                wback = x3;
@@ -1874,6 +1930,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                WBACK;
            } else {
                FAKEED;
+                if (!rex.w && !rex.is32bits && MODREG) { MOVw_REG(ed, ed); }
                F8;
            }
            break;
@@ -1904,15 +1961,15 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                switch (nextop) {
                    case 0xE8:
                        INST_NAME("LFENCE");
-                        SMDMB();
+                        DMB_LD();
                        break;
                    case 0xF0:
                        INST_NAME("MFENCE");
-                        SMDMB();
+                        DMB_SY();
                        break;
                    case 0xF8:
                        INST_NAME("SFENCE");
-                        SMDMB();
+                        DMB_ST();
                        break;
                    default:
                        DEFAULT;
@@ -1922,18 +1979,18 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    case 0:
                        INST_NAME("FXSAVE Ed");
                        MESSAGE(LOG_DUMP, "Need Optimization (FXSAVE)\n");
-                        fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+                        BARRIER(BARRIER_FLOAT);
                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                        if(ed!=x1) {MOVx_REG(x1, ed);}
-                        CALL(rex.is32bits?((void*)fpu_fxsave32):((void*)fpu_fxsave64), -1);
+                        CALL(rex.is32bits?const_fpu_fxsave32:const_fpu_fxsave64, -1);
                        break;
                    case 1:
                        INST_NAME("FXRSTOR Ed");
                        MESSAGE(LOG_DUMP, "Need Optimization (FXRSTOR)\n");
-                        fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+                        BARRIER(BARRIER_FLOAT);
                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                        if(ed!=x1) {MOVx_REG(x1, ed);}
-                        CALL(rex.is32bits?((void*)fpu_fxrstor32):((void*)fpu_fxrstor64), -1);
+                        CALL(rex.is32bits?const_fpu_fxrstor32:const_fpu_fxrstor64, -1);
                        break;
                    case 2:
                        INST_NAME("LDMXCSR Md");
@@ -1950,21 +2007,21 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                        if(BOX64ENV(sse_flushto0)) {
                            // try to sync mxcsr with fpsr on the flag side
                            /* mapping is
-                                ARM -> X86
-                                0 -> 0  // Invalid operation
-                                1 -> 2  // Divide by 0
-                                2 -> 3  // Overflow
-                                3 -> 4  // underflow
-                                4 -> 5  // Inexact
-                                5 -> 1  // denormal
+                                ARM <-> X86
+                                 0  <->  0  // Invalid operation
+                                 1  <->  2  // Divide by 0
+                                 2  <->  3  // Overflow
+                                 3  <->  4  // underflow
+                                 4  <->  5  // Inexact
+                                 5  <->  1  // denormal
                            */
-                            // doing X86 -> ARM here, 0 1 2 3 4 5 -> 0 2 3 4 5 1
+                            // doing X86 -> ARM here, 543210 => 432150
                            if(ed!=x1)
                                MOVw_REG(x1, ed);   // x1 = 543210
-                            RORw(x3, x1, 2);    // x3 = 10.....5432
-                            BFIw(x1, x3, 1, 4); // x1 = 54320
-                            RORw(x3, x3, 32-1); // x3 = 0.....54321
-                            BFIw(x1, x3, 5, 1); // x1 = 154320
+                            UBFXw(x2, x1, 1, 5);   // x2 = 54321
+                            BFIw(x1, x2, 2, 4); // x1 = 432110
+                            LSRw(x2, x2, 4);    // x2 = 5
+                            BFIw(x1, x2, 1, 1); // x1 = 432150
                            MRS_fpsr(x2);
                            BFIx(x2, x1, 0, 6);
                            MSR_fpsr(x2);
@@ -1974,45 +2031,42 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                        INST_NAME("STMXCSR Md");
                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
                        LDRw_U12(x4, xEmu, offsetof(x64emu_t, mxcsr));
-                        STW(x4, ed, fixedaddress);
                        if(BOX64ENV(sse_flushto0)) {
                            // sync with fpsr, with mask from mxcsr
                            MRS_fpsr(x1);
-                            // doing ARM -> X86 here,  543210 => 432150
-                            UBFXw(x2, x1, 1, 5);   // x2 = 54321
-                            BFIw(x1, x2, 2, 4); // x1 = 432110
-                            LSRw(x2, x2, 4);    // x2 = 5
-                            BFIw(x1, x2, 1, 1); // x1 = 432150
+                            // doing ARM -> X86 here,  0 1 2 3 4 5 -> 0 2 3 4 5 1
+                            RORw(x3, x1, 2);    // x3 = 10.....5432
+                            BFIw(x1, x3, 1, 4); // x1 = 54320
+                            RORw(x3, x3, 32-1); // x3 = 0.....54321
+                            BFIw(x1, x3, 5, 1); // x1 = 154320
                            //BFXILw(x3, x4, 7, 6); // this would the mask, but let's ignore that for now
                            BFIw(x4, x1, 0, 6); // inject back the flags
                        }
+                        STW(x4, ed, fixedaddress);
                        break;
                    case 4:
                        INST_NAME("XSAVE Ed");
                        MESSAGE(LOG_DUMP, "Need Optimization (XSAVE)\n");
-                        fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+                        BARRIER(BARRIER_FLOAT);
                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                        if(ed!=x1) {MOVx_REG(x1, ed);}
                        MOV32w(x2, rex.w?0:1);
-                        CALL((void*)fpu_xsave, -1);
+                        CALL(const_fpu_xsave, -1);
                        break;
                    case 5:
                        INST_NAME("XRSTOR Ed");
                        MESSAGE(LOG_DUMP, "Need Optimization (XRSTOR)\n");
-                        fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+                        BARRIER(BARRIER_FLOAT);
                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                        if(ed!=x1) {MOVx_REG(x1, ed);}
                        MOV32w(x2, rex.w?0:1);
-                        CALL((void*)fpu_xrstor, -1);
+                        CALL(const_fpu_xrstor, -1);
                        break;
                    case 7:
                        INST_NAME("CLFLUSH Ed");
-                        MESSAGE(LOG_DUMP, "Need Optimization (CLFLUSH)?\n");
                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
-                        if(ed!=x1) {
-                            MOVx_REG(x1, ed);
-                        }
-                        CALL_(native_clflush, -1, 0);
+                        DC_CIVAC(ed);
+                        SMDMB();
                        break;
                    default:
                        DEFAULT;
@@ -2020,7 +2074,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            break;
        case 0xAF:
            INST_NAME("IMUL Gd, Ed");
-            if(BOX64ENV(dynarec_safeflags) && BOX64ENV(cputype)) {
+            if (BOX64DRENV(dynarec_safeflags) && BOX64ENV(cputype)) {
                SETFLAGS(X_OF|X_CF, SF_SET);
            } else {
                SETFLAGS(X_ALL, SF_SET);
@@ -2092,23 +2146,28 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    CMPSxw_REG(xRAX, ed);
                }
                MOVxw_REG(x1, ed); // save value
-                CSELxw(ed, gd, ed, cEQ);
+                Bcond(cNE, 4 + (rex.w ? 4 : 8));
+                MOVxw_REG(ed, gd);
+                if (!rex.w) { B_NEXT_nocond; }
            } else {
                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<(2+rex.w), (1<<(2+rex.w))-1, rex, NULL, 0, 0);
                LDxw(x1, wback, fixedaddress);
-                UFLAG_IF {emit_cmp32(dyn, ninst, rex, xRAX, x1, x3, x4, x5);}
+                UFLAG_IF {
+                    emit_cmp32(dyn, ninst, rex, xRAX, x1, x3, x4, x5);
+                }
                SUBxw_REG(x4, xRAX, x1);
                CBNZxw_MARK(x4);
                // EAX == Ed
                STxw(gd, wback, fixedaddress);
+                if (!rex.w) { B_NEXT_nocond; }
                MARK;
            }
-            MOVxw_REG(xRAX, x1);    // upper part of RAX will be erase on 32bits, no mater what
+            MOVxw_REG(xRAX, x1);
            break;

        case 0xB3:
            INST_NAME("BTR Ed, Gd");
-            if(!BOX64ENV(dynarec_safeflags)) {
+            if (!BOX64DRENV(dynarec_safeflags)) {
                SETFLAGS(X_ALL&~X_ZF, SF_SUBSET);
            } else {
                SETFLAGS(X_CF, SF_SUBSET);
@@ -2124,7 +2183,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<(2+rex.w), (1<<(2+rex.w))-1, rex, NULL, 0, 0);
                ASRxw(x1, gd, 5+rex.w); // r1 = (gd>>5)
                if(!rex.w && !rex.is32bits) {SXTWx(x1, x1);}
-                ADDz_REG_LSL(x3, wback, x1, 2+rex.w); //(&ed)+=r1*4;
+                ADDy_REG_LSL(x3, wback, x1, 2+rex.w); //(&ed)+=r1*4;
                LDxw(x1, x3, fixedaddress);
                ed = x1;
                wback = x3;
@@ -2186,7 +2245,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            switch((nextop>>3)&7) {
                case 4:
                    INST_NAME("BT Ed, Ib");
-                    if(!BOX64ENV(dynarec_safeflags)) {
+                    if (!BOX64DRENV(dynarec_safeflags)) {
                        SETFLAGS(X_ALL&~X_ZF, SF_SUBSET);
                    } else {
                        SETFLAGS(X_CF, SF_SUBSET);
@@ -2208,7 +2267,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    break;
                case 5:
                    INST_NAME("BTS Ed, Ib");
-                    if(!BOX64ENV(dynarec_safeflags)) {
+                    if (!BOX64DRENV(dynarec_safeflags)) {
                        SETFLAGS(X_ALL&~X_ZF, SF_SUBSET);
                    } else {
                        SETFLAGS(X_CF, SF_SUBSET);
@@ -2236,7 +2295,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    break;
                case 6:
                    INST_NAME("BTR Ed, Ib");
-                    if(!BOX64ENV(dynarec_safeflags)) {
+                    if (!BOX64DRENV(dynarec_safeflags)) {
                        SETFLAGS(X_ALL&~X_ZF, SF_SUBSET);
                    } else {
                        SETFLAGS(X_CF, SF_SUBSET);
@@ -2263,7 +2322,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    break;
                case 7:
                    INST_NAME("BTC Ed, Ib");
-                    if(!BOX64ENV(dynarec_safeflags)) {
+                    if (!BOX64DRENV(dynarec_safeflags)) {
                        SETFLAGS(X_ALL&~X_ZF, SF_SUBSET);
                    } else {
                        SETFLAGS(X_CF, SF_SUBSET);
@@ -2295,7 +2354,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            break;
        case 0xBB:
            INST_NAME("BTC Ed, Gd");
-            if(!BOX64ENV(dynarec_safeflags)) {
+            if (!BOX64DRENV(dynarec_safeflags)) {
                SETFLAGS(X_ALL&~X_ZF, SF_SUBSET);
            } else {
                SETFLAGS(X_CF, SF_SUBSET);
@@ -2310,7 +2369,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<(2+rex.w), (1<<(2+rex.w))-1, rex, NULL, 0, 0);
                ASRxw(x1, gd, 5+rex.w); // r1 = (gd>>5)
                if(!rex.w && !rex.is32bits) {SXTWx(x1, x1);}
-                ADDz_REG_LSL(x3, wback, x1, 2+rex.w); //(&ed)+=r1*4;
+                ADDy_REG_LSL(x3, wback, x1, 2+rex.w); //(&ed)+=r1*4;
                LDxw(x1, x3, fixedaddress);
                ed = x1;
                wback = x3;
@@ -2334,7 +2393,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            break;
        case 0xBC:
            INST_NAME("BSF Gd, Ed");
-            if(!BOX64ENV(dynarec_safeflags) || BOX64ENV(cputype)) {
+            if (!BOX64DRENV(dynarec_safeflags) || BOX64ENV(cputype)) {
                SETFLAGS(X_ZF, SF_SUBSET);
            } else {
                SETFLAGS(X_ALL, SF_SET);
@@ -2343,14 +2402,8 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            nextop = F8;
            GETED(0);
            GETGD;
-            if(ed!=gd)
-                MOVxw_REG(gd, ed);  // to handle ed=0, setting UD gd to 0
-            IFX(X_ZF) {
-                TSTxw_REG(ed, ed);
-                B_MARK(cEQ);
-            } else {
-                CBZxw_MARK(ed);
-            }
+            TSTxw_REG(ed, ed);
+            B_MARK(cEQ);
            RBITxw(x1, ed);   // reverse
            CLZxw(gd, x1);    // x2 gets leading 0 == BSF
            MARK;
@@ -2360,7 +2413,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    BFIw(xFlags, x2, F_ZF, 1);
                }
            }
-            if(BOX64ENV(dynarec_safeflags) && !BOX64ENV(cputype)) {
+            if (BOX64DRENV(dynarec_safeflags) && !BOX64ENV(cputype)) {
                IFX(X_CF) BFCw(xFlags, F_CF, 1);
                IFX(X_AF) BFCw(xFlags, F_AF, 1);
                IFX(X_SF) BFCw(xFlags, F_SF, 1);
@@ -2370,7 +2423,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            break;
        case 0xBD:
            INST_NAME("BSR Gd, Ed");
-            if(!BOX64ENV(dynarec_safeflags) || BOX64ENV(cputype)) {
+            if (!BOX64DRENV(dynarec_safeflags) || BOX64ENV(cputype)) {
                SETFLAGS(X_ZF, SF_SUBSET);
            } else {
                SETFLAGS(X_ALL, SF_SET);
@@ -2379,14 +2432,8 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            nextop = F8;
            GETED(0);
            GETGD;
-            if(ed!=gd)
-                MOVxw_REG(gd, ed);  // to handle ed=0, setting UD gd to 0
-            IFX(X_ZF) {
-                TSTxw_REG(ed, ed);
-                B_MARK(cEQ);
-            } else {
-                CBZxw_MARK(ed);
-            }
+            TSTxw_REG(ed, ed);
+            B_MARK(cEQ);
            CLZxw(gd, ed);    // x2 gets leading 0
            SUBxw_U12(gd, gd, rex.w?63:31);
            NEGxw_REG(gd, gd);   // complement
@@ -2397,7 +2444,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    BFIw(xFlags, x2, F_ZF, 1);
                }
            }
-            if(BOX64ENV(dynarec_safeflags) && !BOX64ENV(cputype)) {
+            if (BOX64DRENV(dynarec_safeflags) && !BOX64ENV(cputype)) {
                IFX(X_CF) BFCw(xFlags, F_CF, 1);
                IFX(X_AF) BFCw(xFlags, F_AF, 1);
                IFX(X_SF) BFCw(xFlags, F_SF, 1);
@@ -2591,11 +2638,11 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    MOV32w(x1, (1<<F_OF)|(1<<F_SF)|(1<<F_ZF)|(1<<F_PF)|(1<<F_AF));
                    BICw(xFlags, xFlags, x1);
                }
-                if(arm64_rndr) {
+                if(cpuext.rndr) {
                    MRS_rndr(x1);
                    IFX(X_CF) { CSETw(x3, cNE); }
                } else {
-                    CALL(rex.w?((void*)get_random64):((void*)get_random32), x1);
+                    CALL(rex.w?const_random64:const_random32, x1);
                    IFX(X_CF) { MOV32w(x3, 1); }
                }
                IFX(X_CF) { BFIw(xFlags, x3, F_CF, 1); }
@@ -2604,42 +2651,64 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            case 7:
                INST_NAME("RDPID Ed");
                GETED(0);
-                CALL_(helper_getcpu, ed, x2);
+                CALL_(const_helper_getcpu, ed, x2);
                break;
            default:
                DEFAULT;
            } else switch((nextop>>3)&7) {
            case 1:
-                INST_NAME("CMPXCHG8B Gq, Eq");
-                SETFLAGS(X_ZF, SF_SUBSET);
-                SMREAD();
-                addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
-                LDPxw_S7_offset(x2, x3, wback, 0);
-                CMPSxw_REG(xRAX, x2);
-                CCMPxw(xRDX, x3, 0, cEQ);
-                B_MARK(cNE);    // EAX!=ED[0] || EDX!=Ed[1]
-                STPxw_S7_offset(xRBX, xRCX, wback, 0);
-                UFLAG_IF {
-                    IFNATIVE(NF_EQ) {} else {MOV32w(x1, 1);}
-                }
-                B_MARK3_nocond;
-                MARK;
-                MOVxw_REG(xRAX, x2);
-                MOVxw_REG(xRDX, x3);
-                UFLAG_IF {
-                    IFNATIVE(NF_EQ) {} else {MOV32w(x1, 0);}
-                }
-                MARK3;
-                UFLAG_IF {
-                    IFNATIVE(NF_EQ) {} else {
-                        BFIw(xFlags, x1, F_ZF, 1);
+                if(MODREG) {
+                    INST_NAME("Invalid LOCK");
+                    UDF(0);
+                    *need_epilog = 1;
+                    *ok = 0;
+                } else {
+                    if (rex.w) {
+                        INST_NAME("CMPXCHG16B Gq, Eq");
+                    } else {
+                        INST_NAME("CMPXCHG8B Gq, Eq");
                    }
+                    SETFLAGS(X_ZF, SF_SUBSET);
+                    SMREAD();
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
+                    if(rex.w && BOX64DRENV(dynarec_safeflags)>1) {
+                        // unaligned memory cause a GPF
+                        TSTx_mask(wback, 1, 0, 3);
+                        B_MARK2(cEQ);   // alligned, continue...
+                        STORE_XEMU_CALL(xRIP);
+                        CALL_S(const_native_gpf, -1);
+                        LOAD_XEMU_CALL(xRIP);
+                        MARK2;
+                    }
+                    LDPxw_S7_offset(x2, x3, wback, 0);
+                    CMPSxw_REG(xRAX, x2);
+                    CCMPxw(xRDX, x3, 0, cEQ);
+                    B_MARK(cNE);    // EAX!=ED[0] || EDX!=Ed[1]
+                    STPxw_S7_offset(xRBX, xRCX, wback, 0);
+                    UFLAG_IF {
+                        IFNATIVE(NF_EQ) {} else {MOV32w(x1, 1);}
+                    }
+                    B_MARK3_nocond;
+                    MARK;
+                    MOVxw_REG(xRAX, x2);
+                    MOVxw_REG(xRDX, x3);
+                    UFLAG_IF {
+                        IFNATIVE(NF_EQ) {} else {MOV32w(x1, 0);}
+                    }
+                    MARK3;
+                    UFLAG_IF {
+                        IFNATIVE(NF_EQ) {} else {
+                            BFIw(xFlags, x1, F_ZF, 1);
+                        }
+                    }
+                    SMWRITE();
                }
-                SMWRITE();
                break;
            case 4:
                INST_NAME("Unsupported XSAVEC Ed");
                FAKEED;
+                BARRIER(BARRIER_FLOAT);
+                GETIP(ip);
                UDF(0);
                break;
            case 6:
@@ -2650,11 +2719,11 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    MOV32w(x1, (1<<F_OF)|(1<<F_SF)|(1<<F_ZF)|(1<<F_PF)|(1<<F_AF));
                    BICw(xFlags, xFlags, x1);
                }
-                if(arm64_rndr) {
+                if(cpuext.rndr) {
                    MRS_rndr(x1);
                    IFX(X_CF) { CSETw(x3, cNE); }
                } else {
-                    CALL(rex.w?((void*)get_random64):((void*)get_random32), x1);
+                    CALL(rex.w?const_random64:const_random32, x1);
                    IFX(X_CF) { MOV32w(x3, 1); }
                }
                IFX(X_CF) { BFIw(xFlags, x3, F_CF, 1); }
@@ -2711,8 +2780,10 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            GETGM(d0);
            GETEM(d1, 0);
            v0 = fpu_get_scratch(dyn, ninst);
+            v1 = fpu_get_scratch(dyn, ninst);
+            UQXTN_32(v1, d1);
            MOVI_32(v0, 64);
-            UMIN_32(v0, v0, d1);    // limit to 0 .. +64 values (will force 32bits upper part to 0)
+            UMIN_32(v0, v0, v1);    // limit to 0 .. +64 values
            NEG_32(v0, v0);
            UXTL_32(v0, v0);
            USHL_R_64(d0, d0, v0);
@@ -2740,7 +2811,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            q1 = fpu_get_scratch(dyn, ninst);
            GETEM(q0, 0);
            GETGD;
-            TABLE64(x1, (uintptr_t)&mask_shift8);
+            TABLE64C(x1, const_8b_m7_m6_m5_m4_m3_m2_m1_0);
            VLDR64_U12(v0, x1, 0);     // load shift
            MOVI_8(v1, 0x80);   // load mask
            VAND(q1, v1, q0);
@@ -22,7 +22,7 @@
 #include "dynarec_arm64_functions.h"


-uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
 {
    uint8_t opcode = F8;
    uint8_t nextop, u8;
@@ -42,17 +42,8 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
    MAYUSE(j64);
    MAYUSE(lock);

-    while((opcode==0x2E) || (opcode==0x36) || (opcode==0x26) || (opcode==0x66))   // ignoring CS:, SS:, ES: or multiple 0x66
-        opcode = F8;
-
-    while((opcode==0xF2) || (opcode==0xF3)) {
-        rep = opcode-0xF1;
-        opcode = F8;
-    }
-    GETREX();
-
-    if(rex.w && !(opcode==0x0f || opcode==0xf0 || opcode==0x64 || opcode==0x65))   // rex.w cancels "66", but not for 66 0f type of prefix
-        return dynarec64_00(dyn, addr-1, ip, ninst, rex, rep, ok, need_epilog); // addr-1, to "put back" opcode
+    if(rex.w && !(opcode==0x0f))   // rex.w cancels "66", but not for 66 0f type of prefix
+        return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog); // addr-1, to "put back" opcode

    switch(opcode) {
        case 0x01:
@@ -91,7 +82,6 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            INST_NAME("POP ES");
            POP1_16(x1);
            STRH_U12(x1, xEmu, offsetof(x64emu_t, segs[_ES]));
-            STRw_U12(xZR, xEmu, offsetof(x64emu_t, segs_serial[_ES]));
            break;

        case 0x09:
@@ -122,12 +112,18 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            break;

        case 0x0F:
-            switch(rep) {
-                case 0: addr = dynarec64_660F(dyn, addr, ip, ninst, rex, ok, need_epilog); break;
-                case 1: addr = dynarec64_66F20F(dyn, addr, ip, ninst, rex, ok, need_epilog); break;
-                case 2: addr = dynarec64_66F30F(dyn, addr, ip, ninst, rex, ok, need_epilog); break;
+            switch(rex.rep) {
+            case 1:
+                addr = dynarec64_66F20F(dyn, addr, ip, ninst, rex, ok, need_epilog);
+                break;
+            case 2:
+                addr = dynarec64_66F30F(dyn, addr, ip, ninst, rex, ok, need_epilog);
+                break;
+            default:
+                addr = dynarec64_660F(dyn, addr, ip, ninst, rex, ok, need_epilog);
            }
            break;
+
        case 0x11:
            INST_NAME("ADC Ew, Gw");
            READFLAGS(X_CF);
@@ -148,6 +144,8 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            emit_adc16(dyn, ninst, x1, x2, x4, x3);
            GWBACK;
            break;
+        case 0x14:
+            return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
        case 0x15:
            INST_NAME("ADC AX, Iw");
            READFLAGS(X_CF);
@@ -198,7 +196,6 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            INST_NAME("POP DS");
            POP1_16(x1);
            STRH_U12(x1, xEmu, offsetof(x64emu_t, segs[_DS]));
-            STRw_U12(xZR, xEmu, offsetof(x64emu_t, segs_serial[_DS]));
            break;

        case 0x21:
@@ -384,9 +381,8 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                PUSH1_16(xRBP);
                PUSH1_16(xRSI);
                PUSH1_16(xRDI);
-            } else {
-                DEFAULT;
-            }
+            } else
+                return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
            break;
        case 0x61:
            if(rex.is32bits) {
@@ -406,26 +402,19 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                BFIz(xRCX, x1, 0, 16);
                POP1_16(x1);
                BFIz(xRAX, x1, 0, 16);
-            } else {
-                DEFAULT;
-            }
-            break;
-
-        case 0x64:
-            addr = dynarec64_6664(dyn, addr, ip, ninst, rex, _FS, ok, need_epilog);
-            break;
-        case 0x65:
-            addr = dynarec64_6664(dyn, addr, ip, ninst, rex, _GS, ok, need_epilog);
-            break;
-        case 0x66:
-            addr = dynarec64_66(dyn, addr, ip, ninst, rex, rep, ok, need_epilog);
+            } else 
+                return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
            break;

        case 0x68:
            INST_NAME("PUSH Iw");
            u16 = F16;
-            MOV32w(x2, u16);
-            PUSH1_16(x2);
+            if (!u16) {
+                PUSH1_16(xZR);
+            } else {
+                MOV32w(x2, u16);
+                PUSH1_16(x2);
+            }
            break;
        case 0x69:
        case 0x6B:
@@ -434,7 +423,7 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            } else {
                INST_NAME("IMUL Gw,Ew,Ib");
            }
-            if(BOX64ENV(dynarec_safeflags)>1 && BOX64ENV(cputype)) {
+            if (BOX64DRENV(dynarec_safeflags) > 1 && BOX64ENV(cputype)) {
                SETFLAGS(X_OF|X_CF, SF_SET);
            } else {
                SETFLAGS(X_ALL, SF_SET);
@@ -469,8 +458,45 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
        case 0x6A:
            INST_NAME("PUSH Ib");
            i16 = F8S;
-            MOV32w(x2, (uint16_t)i16);
-            PUSH1_16(x2);
+            if (!i16) {
+                PUSH1_16(xZR);
+            } else {
+                MOV32w(x2, (uint16_t)i16);
+                PUSH1_16(x2);
+            }
+            break;
+        case 0x6C:
+        case 0x6E:
+            return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
+        case 0x6D:
+            INST_NAME("INSW");
+            if(BOX64DRENV(dynarec_safeflags)>1) {
+                READFLAGS(X_PEND);
+            } else {
+                SETFLAGS(X_ALL, SF_SET_NODF);    // Hack to set flags in "don't care" state
+            }
+            GETIP(ip);
+            STORE_XEMU_CALL(xRIP);
+            CALL_S(const_native_priv, -1);
+            LOAD_XEMU_CALL(xRIP);
+            jump_to_epilog(dyn, 0, xRIP, ninst);
+            *need_epilog = 0;
+            *ok = 0;
+            break;
+        case 0x6F:
+            INST_NAME("OUTSW");
+            if(BOX64DRENV(dynarec_safeflags)>1) {
+                READFLAGS(X_PEND);
+            } else {
+                SETFLAGS(X_ALL, SF_SET_NODF);    // Hack to set flags in "don't care" state
+            }
+            GETIP(ip);
+            STORE_XEMU_CALL(xRIP);
+            CALL_S(const_native_priv, -1);
+            LOAD_XEMU_CALL(xRIP);
+            jump_to_epilog(dyn, 0, xRIP, ninst);
+            *need_epilog = 0;
+            *ok = 0;
            break;

        case 0x70:
@@ -490,7 +516,7 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
        case 0x7e:
        case 0x7f:
            // just use regular conditional jump
-            return dynarec64_00(dyn, addr-1, ip, ninst, rex, rep, ok, need_epilog);
+            return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);

        case 0x81:
        case 0x83:
@@ -596,9 +622,8 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                if(!ALIGNED_ATOMICH) {
                    TBNZ_MARK(ed, 0);
                }
-                if(arm64_atomics) {
+                if(cpuext.atomics) {
                    SWPALH(gd, x1, ed);
-                    SMDMB();
                    if(!ALIGNED_ATOMICH) {
                        B_MARK2_nocond;
                    }
@@ -607,7 +632,6 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    LDAXRH(x1, ed);
                    STLXRH(x3, gd, ed);
                    CBNZx_MARKLOCK(x3);
-                    SMDMB();
                    if(!ALIGNED_ATOMICH) {
                        B_MARK2_nocond;
                    }
@@ -619,7 +643,6 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    STLXRB(x3, gd, ed);
                    CBNZx_MARK(x3);
                    STRH_U12(gd, ed, 0);
-                    SMDMB();
                    MARK2;
                }
                BFIz(gd, x1, 0, 16);
@@ -670,33 +693,62 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                SMWRITE2();
            }
            break;
-
+        case 0x8D:
+            INST_NAME("LEA Gd, Ed");
+            nextop=F8;
+            GETGD;
+            if(MODREG) {   // reg <= reg? that's an invalid operation
+                DEFAULT;
+            } else {                    // mem <= reg
+                rex.seg = 0;
+                addr = geted(dyn, addr, ninst, nextop, &ed, x3, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
+                BFIx(gd, ed, 0, 16);
+            }
+            break;
        case 0x8E:
            INST_NAME("MOV Seg,Ew");
            nextop = F8;
            u8 = (nextop&0x38)>>3;
-            if (MODREG) {
-                ed = TO_NAT((nextop & 7) + (rex.b << 3));
+            if((u8>5) || (u8==1)) {
+                INST_NAME("Invalid MOV Seg,Ew");
+                UDF(0);
+                *need_epilog = 1;
+                *ok = 0;
            } else {
-                SMREAD();
-                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
-                LDH(x1, wback, fixedaddress);
-                ed = x1;
+                if (MODREG) {
+                    ed = TO_NAT((nextop & 7) + (rex.b << 3));
+                } else {
+                    SMREAD();
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
+                    LDH(x1, wback, fixedaddress);
+                    ed = x1;
+                }
+                STRH_U12(ed, xEmu, offsetof(x64emu_t, segs[u8]));
+                if((u8==_FS) || (u8==_GS)) {
+                    // refresh offset if needed
+                    CBZw_NEXT(ed);
+                    MOV32w(x1, u8);
+                    CALL(const_getsegmentbase, -1);
+                }
            }
-            STRH_U12(ed, xEmu, offsetof(x64emu_t, segs[u8]));
-            STRw_U12(wZR, xEmu, offsetof(x64emu_t, segs_serial[u8]));
            break;
        case 0x8F:
-            INST_NAME("POP Ew");
            nextop = F8;
-            POP1_16(x1);
-            if (MODREG) {
-                wback = TO_NAT((nextop & 7) + (rex.b << 3));
-                BFIz(wback, x1, 0, 16);
-            } else {
-                SMREAD();
-                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
-                STH(x1, wback, fixedaddress);
+            switch((nextop>>3)&7) {
+                case 0:
+                    INST_NAME("POP Ew");
+                    POP1_16(x1);
+                    if (MODREG) {
+                        wback = TO_NAT((nextop & 7) + (rex.b << 3));
+                        BFIz(wback, x1, 0, 16);
+                    } else {
+                        SMREAD();
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
+                        STH(x1, wback, fixedaddress);
+                    }
+                    break;
+                default:
+                    DEFAULT;
            }
            break;
        case 0x90:
@@ -746,102 +798,89 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            if(box64_wine) {    // should this be done all the time?
                TBZ_NEXT(xFlags, F_TF);
                // go to epilog, TF should trigger at end of next opcode, so using Interpreter only
+                LDRw_U12(x4, xEmu, offsetof(x64emu_t, flags));
+                ORRw_mask(x4, x4, 32-FLAGS_NO_TF, 0);   //mask=1<<FLAGS_NO_TF
+                STRw_U12(x4, xEmu, offsetof(x64emu_t, flags));
                jump_to_epilog(dyn, addr, 0, ninst);
            }
            break;

        case 0xA1:
            INST_NAME("MOV AX,Od");
-            if(rex.is32bits)
-                u64 = F32;
+            if(rex.is32bits && rex.is67)
+                u64 = F16S;
+            else if(rex.is32bits || rex.is67)
+                u64 = F32S;
            else
                u64 = F64;
-            MOV64z(x1, u64);
-            if(isLockAddress(u64)) lock=1; else lock = 0;
+            unscaled = 0; fixedaddress = 0;
+            if(rex.seg && (u64<0x1000 || (int64_t)u64>-0x1000 || (u64<(0x1000<<1) && !(u64&1)))) {
+                grab_segdata(dyn, addr, ninst, x1, rex.seg);
+                if(u64) {
+                    if(u64<0x100)
+                        {fixedaddress = u64; unscaled = 1;}
+                    else if(u64<(0x1000<<1) && !(u64&1))
+                        fixedaddress = u64;
+                    else if(u64<0x1000)
+                        ADDx_U12(x1, x1, u64);
+                    else if((int64_t)u64 > -0x100)
+                        {fixedaddress = (int64_t)u64; unscaled = 1;}
+                    else
+                        SUBx_U12(x1, x1, -(int64_t)u64);
+                }
+            } else {
+                MOV64y(x1, u64);
+                if(rex.seg) {
+                    grab_segdata(dyn, addr, ninst, x3, rex.seg);
+                    ADDx_REGy(x1, x3, x1);
+                }
+            }
+            lock=(rex.seg)?0:isLockAddress(u64);
            SMREADLOCK(lock);
-            LDRH_U12(x2, x1, 0);
+            LDH(x2, x1, fixedaddress);
            BFIz(xRAX, x2, 0, 16);
            break;

        case 0xA3:
            INST_NAME("MOV Od,AX");
-            if(rex.is32bits)
-                u64 = F32;
+            if(rex.is32bits && rex.is67)
+                u64 = F16S;
+            else if(rex.is32bits || rex.is67)
+                u64 = F32S;
            else
                u64 = F64;
-            MOV64z(x1, u64);
-            lock = isLockAddress(u64);
-            STRH_U12(xRAX, x1, 0);
+            unscaled = 0; fixedaddress = 0;
+            if(rex.seg && (u64<0x1000 || (int64_t)u64>-0x1000 || (u64<(0x1000<<1) && !(u64&1)))) {
+                grab_segdata(dyn, addr, ninst, x1, rex.seg);
+                if(u64) {
+                    if(u64<0x100)
+                        {fixedaddress = u64; unscaled = 1;}
+                    else if(u64<(0x1000<<1) && !(u64&1))
+                        fixedaddress = u64;
+                    else if(u64<0x1000)
+                        ADDx_U12(x1, x1, u64);
+                    else if((int64_t)u64 > -0x100)
+                        {fixedaddress = (int64_t)u64; unscaled = 1;}
+                    else
+                        SUBx_U12(x1, x1, -(int64_t)u64);
+                }
+            } else {
+                MOV64y(x1, u64);
+                if(rex.seg) {
+                    grab_segdata(dyn, addr, ninst, x3, rex.seg);
+                    ADDx_REGy(x1, x3, x1);
+                }
+            }
+            lock=(rex.seg)?0:isLockAddress(u64);
+            SMREADLOCK(lock);
+            STH(xRAX, x1, fixedaddress);
            SMWRITELOCK(lock);
            break;
        case 0xA4:
-            SMREAD();
-            if(rep) {
-                INST_NAME("REP MOVSB");
-                CBZx_NEXT(xRCX);
-                TBNZ_MARK2(xFlags, F_DF);
-                IF_UNALIGNED(ip) {
-                    MESSAGE(LOG_DEBUG, "\tUnaligned path");
-                    // special optim for large RCX value on forward case only
-                    // but because it's unaligned path, check if a byte per byt is needed, and do 4-bytes per 4-bytes only instead
-                    if(BOX64DRENV(dynarec_safeflags)) {
-                        SUBx_REG(x2, xRDI, xRSI);
-                        CMPSx_U12(x2, 4);
-                        B_MARK(cCC);
-                    }
-                    ORRw_REG(x1, xRSI, xRDI);
-                    ANDw_mask(x1, x1, 0, 1);    //mask = 3
-                    CBNZw_MARK(x1);
-                    MARK3;
-                    CMPSx_U12(xRCX, 4);
-                    B_MARK(cCC);
-                    LDRw_S9_postindex(x1, xRSI, 4);
-                    STRw_S9_postindex(x1, xRDI, 4);
-                    SUBx_U12(xRCX, xRCX, 4);
-                    CBNZx_MARK3(xRCX);
-                    CBZx_MARKLOCK(xRCX);
-                } else {
-                    if(BOX64DRENV(dynarec_safeflags)) {
-                        SUBx_REG(x2, xRDI, xRSI);
-                        CMPSx_U12(x2, 8);
-                        B_MARK(cCC);
-                    }
-                    // special optim for large RCX value on forward case only
-                    MARK3;
-                    CMPSx_U12(xRCX, 8);
-                    B_MARK(cCC);
-                    LDRx_S9_postindex(x1, xRSI, 8);
-                    STRx_S9_postindex(x1, xRDI, 8);
-                    SUBx_U12(xRCX, xRCX, 8);
-                    CBNZx_MARK3(xRCX);
-                    CBZx_MARKLOCK(xRCX);
-                }
-                MARK;   // Part with DF==0
-                LDRB_S9_postindex(x1, xRSI, 1);
-                STRB_S9_postindex(x1, xRDI, 1);
-                SUBx_U12(xRCX, xRCX, 1);
-                CBNZx_MARK(xRCX);
-                B_MARKLOCK_nocond;
-                MARK2;  // Part with DF==1
-                LDRB_S9_postindex(x1, xRSI, -1);
-                STRB_S9_postindex(x1, xRDI, -1);
-                SUBx_U12(xRCX, xRCX, 1);
-                CBNZx_MARK2(xRCX);
-                MARKLOCK;
-                // done
-            } else {
-                INST_NAME("MOVSB");
-                GETDIR(x3, 1);
-                LDRB_U12(x1, xRSI, 0);
-                STRB_U12(x1, xRDI, 0);
-                ADDx_REG(xRSI, xRSI, x3);
-                ADDx_REG(xRDI, xRDI, x3);
-            }
-            SMWRITE();
-            break;
+            return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
        case 0xA5:
            SMREAD();
-            if(rep) {
+            if(rex.rep) {
                INST_NAME("REP MOVSW");
                CBZx_NEXT(xRCX);
                TBNZ_MARK2(xFlags, F_DF);
@@ -870,10 +909,10 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin

        case 0xA7:
            SMREAD();
-            switch(rep) {
+            switch(rex.rep) {
            case 1:
            case 2:
-                if(rep==1) {INST_NAME("REPNZ CMPSW");} else {INST_NAME("REPZ CMPSW");}
+                if(rex.rep==1) {INST_NAME("REPNZ CMPSW");} else {INST_NAME("REPZ CMPSW");}
                if(BOX64DRENV(dynarec_safeflags)>1) {
                    READFLAGS(X_ALL);
                    SETFLAGS(X_ALL, SF_SET);
@@ -886,7 +925,7 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                LDRH_S9_postindex(x2, xRDI, 2);
                SUBx_U12(xRCX, xRCX, 1);
                CMPSw_REG(x1, x2);
-                B_MARK3((rep==1)?cEQ:cNE);
+                B_MARK3((rex.rep==1)?cEQ:cNE);
                CBNZx_MARK(xRCX);
                B_MARK3_nocond;
                MARK2;  // Part with DF==1
@@ -894,7 +933,7 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                LDRH_S9_postindex(x2, xRDI, -2);
                SUBx_U12(xRCX, xRCX, 1);
                CMPSw_REG(x1, x2);
-                B_MARK3((rep==1)?cEQ:cNE);
+                B_MARK3((rex.rep==1)?cEQ:cNE);
                CBNZx_MARK2(xRCX);
                MARK3;  // end
                emit_cmp16(dyn, ninst, x1, x2, x3, x4, x5);
@@ -921,7 +960,7 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            break;

        case 0xAB:
-            if(rep) {
+            if(rex.rep) {
                INST_NAME("REP STOSW");
                CBZx_NEXT(xRCX);
                TBNZ_MARK2(xFlags, F_DF);
@@ -946,7 +985,7 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            break;

        case 0xAD:
-            if(rep) {
+            if(rex.rep) {
                INST_NAME("REP LODSW");
                CBZx_NEXT(xRCX);
                TBNZ_MARK2(xFlags, F_DF);
@@ -972,10 +1011,10 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            break;

        case 0xAF:
-            switch(rep) {
+            switch(rex.rep) {
            case 1:
            case 2:
-                if(rep==1) {INST_NAME("REPNZ SCASW");} else {INST_NAME("REPZ SCASW");}
+                if(rex.rep==1) {INST_NAME("REPNZ SCASW");} else {INST_NAME("REPZ SCASW");}
                if(BOX64DRENV(dynarec_safeflags)>1) {
                    READFLAGS(X_ALL);
                    SETFLAGS(X_ALL, SF_SET);
@@ -988,14 +1027,14 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                LDRH_S9_postindex(x2, xRDI, 2);
                SUBx_U12(xRCX, xRCX, 1);
                CMPSw_REG(x1, x2);
-                B_MARK3((rep==1)?cEQ:cNE);
+                B_MARK3((rex.rep==1)?cEQ:cNE);
                CBNZx_MARK(xRCX);
                B_MARK3_nocond;
                MARK2;  // Part with DF==1
                LDRH_S9_postindex(x2, xRDI, -2);
                SUBx_U12(xRCX, xRCX, 1);
                CMPSw_REG(x1, x2);
-                B_MARK3((rep==1)?cEQ:cNE);
+                B_MARK3((rex.rep==1)?cEQ:cNE);
                CBNZx_MARK2(xRCX);
                MARK3;  // end
                emit_cmp16(dyn, ninst, x1, x2, x3, x4, x5);
@@ -1132,26 +1171,32 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            break;

        case 0xC7:
-            INST_NAME("MOV Ew, Iw");
            nextop = F8;
-            if(MODREG) {
-                ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                u16 = F16;
-                if(u16) {
-                    MOV32w(x1, u16);
-                    BFIz(ed, x1, 0, 16);
-                } else
-                    BFCz(ed, 0, 16);
-            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, &lock, 0, 2);
-                u16 = F16;
-                if(u16) {
-                    MOV32w(x1, u16);
-                    STH(x1, ed, fixedaddress);
-                } else {
-                    STH(xZR, ed, fixedaddress);
-                }
-                SMWRITELOCK(lock);
+            switch((nextop>>3)&7) {
+                case 0:
+                    INST_NAME("MOV Ew, Iw");
+                    if(MODREG) {
+                        ed = TO_NAT((nextop & 7) + (rex.b << 3));
+                        u16 = F16;
+                        if(u16) {
+                            MOV32w(x1, u16);
+                            BFIz(ed, x1, 0, 16);
+                        } else
+                            BFCz(ed, 0, 16);
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, &lock, 0, 2);
+                        u16 = F16;
+                        if(u16) {
+                            MOV32w(x1, u16);
+                            STH(x1, ed, fixedaddress);
+                        } else {
+                            STH(xZR, ed, fixedaddress);
+                        }
+                        SMWRITELOCK(lock);
+                    }
+                    break;
+                default:
+                    DEFAULT;
            }
            break;

@@ -1276,31 +1321,84 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    break;
                case 2:
                    INST_NAME("RCL Ew, CL");
-                    MESSAGE(LOG_DUMP, "Need Optimization (RCL Ex, CL)\n");
                    if(BOX64DRENV(dynarec_safeflags)>1) {
                        READFLAGS(X_OF|X_CF);
                    } else {
                        READFLAGS(X_CF);
                    }
-                    SETFLAGS(X_OF|X_CF, SF_SET_DF);
-                    ANDw_mask(x2, xRCX, 0, 0b00100);
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    ANDw_mask(x2, xRCX, 0, 0b00100);  //mask=0x00000001f
+                    CBZw_NEXT(x2);
+                    // get CL % 17
+                    MOV32w(x3, 0xf10); // 0x10000 / 17 + 1 (this is precise enough in the 0..31 range)
+                    MULw(x3, x3, x2);
+                    LSRw(x3, x3, 16);   // x3 = CL / 17
+                    MOV32w(x4, 17);
+                    MSUBw(x2, x3, x4, x2);  // CL mod 17
                    GETEW(x1, 0);
-                    CALL_(rcl16, x1, x3);
+                    CBZw_MARK(x2);
+                    IFX2(X_OF, && !BOX64ENV(cputype)) {
+                        LSRw(x5, ed, 14);
+                        EORw_REG_LSR(x5, x5, x5, 1);
+                        BFIw(xFlags, x5, F_OF, 1);
+                    }
+                    BFIw(ed, xFlags, 16, 1); // insert CF
+                    ORRx_REG_LSL(ed, ed, ed, 17);    // insert rest of ed
+                    SUBw_REG(x2, x4, x2);
+                    IFX(X_OF|X_CF) {
+                        SUBw_U12(x5, x2, 1);
+                        LSRx_REG(x5, ed, x5);   // keep the new CF in x5
+                    }
+                    LSRx_REG(ed, ed, x2);
                    EWBACK;
+                    u8 = X_CF;
+                    if(BOX64ENV(cputype)) u8 |= X_OF;
+                    IFX(u8) {
+                        BFXILw(xFlags, x5, 0, 1);
+                    }
+                    MARK;
+                    IFX2(X_OF, && BOX64ENV(cputype)) {
+                        EORw_REG_LSR(x2, xFlags, ed, 15);
+                        BFIw(xFlags, x2, F_OF, 1);
+                    }
                    break;
                case 3:
                    INST_NAME("RCR Ew, CL");
-                    MESSAGE(LOG_DUMP, "Need Optimization (RCR Ew, CL)\n");
                    if(BOX64DRENV(dynarec_safeflags)>1) {
                        READFLAGS(X_OF|X_CF);
                    } else {
                        READFLAGS(X_CF);
                    }
-                    SETFLAGS(X_OF|X_CF, SF_SET_DF);
-                    ANDw_mask(x2, xRCX, 0, 0b00100);
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    ANDw_mask(x2, xRCX, 0, 0b00100);  //mask=0x00000001f
+                    CBZw_NEXT(x2);
+                    // get CL % 17
+                    MOV32w(x3, 0xf10); // 0x10000 / 17 + 1
+                    MULw(x3, x3, x2);
+                    LSRw(x3, x3, 16);   // x3 = CL / 17
+                    MOV32w(x4, 17);
+                    MSUBw(x2, x3, x4, x2);  // CL mod 17
                    GETEW(x1, 0);
-                    CALL_(rcr16, x1, x3);
+                    CBZw_MARK(x2);
+                    BFIw(ed, xFlags, 16, 1); // insert CF
+                    ORRx_REG_LSL(ed, ed, ed, 17);    // insert rest of ed
+                    IFX2(X_OF, && !BOX64ENV(cputype)) {
+                        EORw_REG_LSR(x5, xFlags, ed, 15);
+                        BFIw(xFlags, x5, F_OF, 1);
+                    }
+                    IFX(X_CF) {
+                        SUBw_U12(x4, x2, 1);
+                        LSRw_REG(x5, ed, x4);
+                        BFIw(xFlags, x5, F_CF, 1);
+                    }
+                    LSRx_REG(ed, ed, x2);
                    EWBACK;
+                    MARK;
+                    IFX2(X_OF, && BOX64ENV(cputype)) {
+                        LSRw(x4, ed, 14);
+                        EORw_REG_LSR(x4, x4, x4, 1);
+                        BFIw(xFlags, x4, F_OF, 1);
+                    }
                    break;
                case 4:
                case 6:
@@ -1361,11 +1459,11 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    case 6:
                        INST_NAME("FNSTENV Ed");
                        MESSAGE(LOG_DUMP, "Need Optimization (FNSTENV16)\n");
-                        fpu_purgecache(dyn, ninst, 0, x1, x2, x3); // maybe only x87, not SSE?
+                        BARRIER(BARRIER_FLOAT); // maybe only x87, not SSE?
                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                        if(ed!=x1) {MOVx_REG(x1, ed);}
                        MOV32w(x2, 1);
-                        CALL(fpu_savenv, -1);
+                        CALL(const_fpu_savenv, -1);
                        break;
                    default:
                        DEFAULT;
@@ -1381,27 +1479,38 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    case 4:
                        INST_NAME("FRSTOR Ed");
                        MESSAGE(LOG_DUMP, "Need Optimization (FRSTOR16)\n");
-                        fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+                        BARRIER(BARRIER_FLOAT);
                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                        if(ed!=x1) {MOVx_REG(x1, ed);}
-                        CALL(native_frstor16, -1);
+                        CALL(const_native_frstor16, -1);
                        break;
                    case 6:
                        INST_NAME("FNSAVE Ed");
                        MESSAGE(LOG_DUMP, "Need Optimization (FNSAVE16)\n");
-                        fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+                        BARRIER(BARRIER_FLOAT);
                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                        if(ed!=x1) {MOVx_REG(x1, ed);}
-                        CALL(native_fsave16, -1);
-                        CALL(reset_fpu, -1);
+                        CALL(const_native_fsave16, -1);
                        break;
                    default:
                        DEFAULT;
            }
            break;

-        case 0xF0:
-            return dynarec64_66F0(dyn, addr, ip, ninst, rex, rep, ok, need_epilog);
+        case 0xE4:                      /* IN AL, Ib */
+        case 0xE5:                      /* IN AX, Ib */
+        case 0xE6:                      /* OUT Ib, AL */
+        case 0xE7:                      /* OUT Ib, AX */
+            return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
+        case 0xE8:                      /* CALL Imm */
+        case 0xE9:                      /* Jmp Imm */
+        case 0xEB:                      /* Jmp Imm8 */
+            return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
+        case 0xEC:                      /* IN AL, DX */
+        case 0xED:                      /* IN AX, DX */
+        case 0xEE:                      /* OUT DX, AL */
+        case 0xEF:                      /* OUT DX, AX */
+            return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);

        case 0xF7:
            nextop = F8;
@@ -1435,7 +1544,7 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    break;
                case 4:
                    INST_NAME("MUL AX, Ew");
-                    if(BOX64ENV(dynarec_safeflags) && BOX64ENV(cputype)) {
+                    if (BOX64DRENV(dynarec_safeflags) && BOX64ENV(cputype)) {
                        SETFLAGS(X_OF|X_CF, SF_SET);
                    } else {
                        SETFLAGS(X_ALL, SF_SET);
@@ -1466,7 +1575,7 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    break;
                case 5:
                    INST_NAME("IMUL AX, Ew");
-                    if(BOX64ENV(dynarec_safeflags) && BOX64ENV(cputype)) {
+                    if (BOX64DRENV(dynarec_safeflags) && BOX64ENV(cputype)) {
                        SETFLAGS(X_OF|X_CF, SF_SET);
                    } else {
                        SETFLAGS(X_ALL, SF_SET);
@@ -1504,15 +1613,20 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    BFIw(x2, xRDX, 16, 16);
                    if(BOX64ENV(dynarec_div0)) {
                        CBNZw_MARK3(ed);
+                        MARK2;
                        GETIP_(ip);
                        STORE_XEMU_CALL(xRIP);
-                        CALL(native_div0, -1);
+                        CALL_S(const_native_div0, -1);
                        CLEARIP();
                        LOAD_XEMU_CALL(xRIP);
                        jump_to_epilog(dyn, 0, xRIP, ninst);
                        MARK3;
                    }
                    UDIVw(x3, x2, ed);
+                    if(BOX64ENV(dynarec_div0)) {
+                        SUBw_UXTH(x5, x3, x3);
+                        CBNZw_MARK2(x5);
+                    }
                    MSUBw(x4, x3, ed, x2);  // x4 = x2 mod ed (i.e. x2 - x3*ed)
                    BFIz(xRAX, x3, 0, 16);
                    BFIz(xRDX, x4, 0, 16);
@@ -1530,17 +1644,18 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                case 7:
                    INST_NAME("IDIV Ew");
                    SKIPTEST(x1);
-                    if(!BOX64ENV(dynarec_safeflags)) {
+                    if (!BOX64DRENV(dynarec_safeflags)) {
                        SETFLAGS(X_ALL, SF_SET);
-                    } else if(BOX64ENV(cputype)) {
+                    } else if (BOX64ENV(cputype)) {
                        SETFLAGS(X_SF|X_PF|X_ZF|X_AF, SF_SET);
                    }
                    GETSEW(x1, 0);
                    if(BOX64ENV(dynarec_div0)) {
                        CBNZw_MARK3(ed);
+                        MARK2;
                        GETIP_(ip);
                        STORE_XEMU_CALL(xRIP);
-                        CALL(native_div0, -1);
+                        CALL_S(const_native_div0, -1);
                        CLEARIP();
                        LOAD_XEMU_CALL(xRIP);
                        jump_to_epilog(dyn, 0, xRIP, ninst);
@@ -1549,10 +1664,14 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    UXTHw(x2, xRAX);
                    BFIw(x2, xRDX, 16, 16);
                    SDIVw(x3, x2, ed);
+                    if(BOX64ENV(dynarec_div0)) {
+                        SUBw_SXTH(x5, x3, x3);
+                        CBNZw_MARK2(x5);
+                    }
                    MSUBw(x4, x3, ed, x2);  // x4 = x2 mod ed (i.e. x2 - x3*ed)
                    BFIz(xRAX, x3, 0, 16);
                    BFIz(xRDX, x4, 0, 16);
-                    if(!BOX64ENV(dynarec_safeflags)) {
+                    if (!BOX64DRENV(dynarec_safeflags)) {
                        SET_DFNONE();
                    }
                    IFX2(X_AF, && BOX64ENV(cputype))  {ORRw_mask(xFlags, xFlags, 28, 0);}   //mask=0x10
@@ -1563,17 +1682,8 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            }
            break;
        case 0xF8:
-            INST_NAME("CLC");
-            SETFLAGS(X_CF, SF_SUBSET);
-            SET_DFNONE();
-            BFCx(xFlags, F_CF, 1);
-            break;
        case 0xF9:
-            INST_NAME("STC");
-            SETFLAGS(X_CF, SF_SUBSET);
-            SET_DFNONE();
-            ORRx_mask(xFlags, xFlags, 1, 0, 0); // xFlags | 1
-            break;
+            return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);

        case 0xFF:
            nextop = F8;
@@ -18,7 +18,6 @@
 #include "dynarec_arm64_private.h"
 #include "dynarec_arm64_functions.h"
 #include "../dynarec_helper.h"
-#include "emu/x64compstrings.h"

 uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
 {
@@ -47,9 +46,6 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
    MAYUSE(eb2);
    MAYUSE(j64);
    #if STEP > 1
-    static const int8_t mask_shift8[] = { -7, -6, -5, -4, -3, -2, -1, 0 };
-    static const int8_t mask_string8[] = { 7, 6, 5, 4, 3, 2, 1, 0 };
-    static const int8_t mask_string16[] = { 15, 14, 13, 12, 11, 10, 9, 8 };
    static const int8_t round_round[] = { 0, 2, 1, 3};
    #endif

@@ -168,6 +164,11 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n

        case 0x18:
        case 0x19:
+        case 0x1A:
+        case 0x1B:
+        case 0x1C:
+        case 0x1D:
+        case 0x1E:
        case 0x1F:
            INST_NAME("NOP (multibyte)");
            nextop = F8;
@@ -233,7 +234,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                VFCVTZSQD(q0, v1);
                SQXTN_32(q0, q0);
            } else {
-                if(arm64_frintts) {
+                if(cpuext.frintts) {
                    VFRINT32ZDQ(q0, v1);
                    VFCVTZSQD(q0, q0);
                    SQXTN_32(q0, q0);
@@ -273,7 +274,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                VFCVTZSS(q0, q0);
            } else {
                u8 = sse_setround(dyn, ninst, x1, x2, x3);
-                if(arm64_frintts) {
+                if(cpuext.frintts) {
                    VFRINT32XDQ(q0, v1);
                    VFCVTZSQD(q0, q0);
                    SQXTN_32(q0, q0);
@@ -309,7 +310,9 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            nextop = F8;
            GETGX(v0, 0);
            GETEXSD(q0, 0, 0);
-            FCMPD(v0, q0);
+            IFX(X_CF|X_PF|X_ZF) {
+                FCMPD(v0, q0);
+            }
            FCOMI(x1, x2);
            break;

@@ -631,7 +634,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    } else {
                        GETGX_empty(v0);
                        SMREAD();
-                        addr = geted32(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff << 4, 15, rex, NULL, 0, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff << 4, 15, rex, NULL, 0, 0);
                        VLD128(v0, ed, fixedaddress);
                    }
                    break;
@@ -769,7 +772,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                case 0xDB:
                    INST_NAME("AESIMC Gx, Ex");  // AES-NI
                    nextop = F8;
-                    if(arm64_aes) {
+                    if(cpuext.aes) {
                        GETEX(q1, 0, 0);
                        GETGX_empty(q0);
                        AESIMC(q0, q1);
@@ -781,13 +784,13 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        }
                        sse_forget_reg(dyn, ninst, gd);
                        MOV32w(x1, gd);
-                        CALL(native_aesimc, -1);
+                        CALL(const_native_aesimc, -1);
                    }
                    break;
                case 0xDC:
                    INST_NAME("AESENC Gx, Ex");  // AES-NI
                    nextop = F8;
-                    if(arm64_aes) {
+                    if(cpuext.aes) {
                        GETEX(q1, 0, 0);
                        GETGX(q0, 1);
                        v0 = fpu_get_scratch(dyn, ninst);  // ARM64 internal operation differs a bit from x86_64
@@ -804,7 +807,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        } else d0 = -1;
                        sse_forget_reg(dyn, ninst, gd);
                        MOV32w(x1, gd);
-                        CALL(native_aese, -1);
+                        CALL(const_native_aese, -1);
                        GETGX(q0, 1);
                        VEORQ(q0, q0, (d0!=-1)?d0:q1);
                    }
@@ -812,7 +815,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                case 0xDD:
                    INST_NAME("AESENCLAST Gx, Ex");  // AES-NI
                    nextop = F8;
-                    if(arm64_aes) {
+                    if(cpuext.aes) {
                        GETEX(q1, 0, 0);
                        GETGX(q0, 1);
                        v0 = fpu_get_scratch(dyn, ninst);  // ARM64 internal operation differs a bit from x86_64
@@ -828,7 +831,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        } else d0 = -1;
                        sse_forget_reg(dyn, ninst, gd);
                        MOV32w(x1, gd);
-                        CALL(native_aeselast, -1);
+                        CALL(const_native_aeselast, -1);
                        GETGX(q0, 1);
                        VEORQ(q0, q0, (d0!=-1)?d0:q1);
                    }
@@ -836,7 +839,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                case 0xDE:
                    INST_NAME("AESDEC Gx, Ex");  // AES-NI
                    nextop = F8;
-                    if(arm64_aes) {
+                    if(cpuext.aes) {
                        GETEX(q1, 0, 0);
                        GETGX(q0, 1);
                        v0 = fpu_get_scratch(dyn, ninst);  // ARM64 internal operation differs a bit from x86_64
@@ -853,7 +856,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        } else d0 = -1;
                        sse_forget_reg(dyn, ninst, gd);
                        MOV32w(x1, gd);
-                        CALL(native_aesd, -1);
+                        CALL(const_native_aesd, -1);
                        GETGX(q0, 1);
                        VEORQ(q0, q0, (d0!=-1)?d0:q1);
                    }
@@ -861,7 +864,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                case 0xDF:
                    INST_NAME("AESDECLAST Gx, Ex");  // AES-NI
                    nextop = F8;
-                    if(arm64_aes) {
+                    if(cpuext.aes) {
                        GETEX(q1, 0, 0);
                        GETGX(q0, 1);
                        v0 = fpu_get_scratch(dyn, ninst);  // ARM64 internal operation differs a bit from x86_64
@@ -877,7 +880,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        } else d0 = -1;
                        sse_forget_reg(dyn, ninst, gd);
                        MOV32w(x1, gd);
-                        CALL(native_aesdlast, -1);
+                        CALL(const_native_aesdlast, -1);
                        GETGX(q0, 1);
                        VEORQ(q0, q0, (d0!=-1)?d0:q1);
                    }
@@ -907,10 +910,10 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        REV16x(x1, gd);
                        BFIx(ed, x1, 0, 16);
                    } else {
-                        SMREAD();
                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<1, (1<<1)-1, rex, NULL, 0, 0);
                        REV16x(x1, gd);
                        STH(x1, ed, fixedaddress);
+                        SMWRITE();
                    }
                    break;

@@ -923,7 +926,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    GETGD;
                    IFNATIVE_BEFORE(NF_CF) {
                        if(INVERTED_CARRY_BEFORE) {
-                            if(arm64_flagm)
+                            if(cpuext.flagm)
                                CFINV();
                            else {
                                MRS_nzcv(x3);
@@ -1151,7 +1154,15 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    INST_NAME("PINSRB Gx, ED, Ib");
                    nextop = F8;
                    GETGX(q0, 1);
-                    GETED(1);
+                    if (MODREG) {
+                        ed = TO_NAT((nextop & 7) + (rex.b << 3));
+                        wback = 0;
+                    } else {
+                        SMREAD();
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff, 0, rex, NULL, 0, 1);
+                        LDB(x1, wback, fixedaddress);
+                        ed = x1;
+                    }
                    u8 = F8;
                    VMOVQBfrom(q0, (u8&15), ed);
                    break;
@@ -1238,7 +1249,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                case 0x44:
                    INST_NAME("PCLMULQDQ Gx, Ex, Ib");
                    nextop = F8;
-                    if(arm64_pmull) {
+                    if(cpuext.pmull) {
                        GETGX(q0, 1);
                        GETEX(q1, 0, 1);
                        u8 = F8;
@@ -1276,7 +1287,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        }
                        u8 = F8;
                        MOV32w(x4, u8);
-                        CALL(native_pclmul, -1);
+                        CALL(const_native_pclmul, -1);
                    }
                    break;

@@ -1302,7 +1313,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    MOVx_REG(x4, xRAX);
                    u8 = F8;
                    MOV32w(x5, u8);
-                    CALL(sse42_compare_string_explicit_len, x1);
+                    CALL(const_sse42_compare_string_explicit_len, x1);
                    q0 = sse_get_reg_empty(dyn, ninst, x2, 0);
                    if(u8&0b1000000) {
                        q1 = fpu_get_scratch(dyn, ninst);
@@ -1312,7 +1323,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                                LSRw_IMM(x1, x1, 8);
                                VDUPQB(q1, x1); // load the high 8bits of the mask
                                VEXTQ_8(q0, q0, q1, 8); // low and hig bits mask
-                                TABLE64(x2, (uintptr_t)&mask_string8);
+                                TABLE64C(x2, const_8b_7_6_5_4_3_2_1_0);
                                VLDR64_U12(q1, x2, 0);     // load shift
                                VDUPQ_64(q1, q1, 0);
                                USHLQ_8(q0, q0, q1); // extract 1 bit
@@ -1322,7 +1333,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                                break;
                            case 0b01:
                                VDUPQH(q0, x1); // load the 8bits of the mask
-                                TABLE64(x2, (uintptr_t)&mask_string16);
+                                TABLE64C(x2, const_8b_15_14_13_12_11_10_9_8);
                                VLDR64_U12(q1, x2, 0);     // load shift
                                UXTL_8(q1, q1);     // extend mask to 16bits
                                USHLQ_16(q0, q0, q1); // extract 1 bit
@@ -1354,7 +1365,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                            // transform that a mask in x1
                            q1 = fpu_get_scratch(dyn, ninst);
                            VSHL_8(q0, q0, 7);  // keep only bit 0x80
-                            TABLE64(x1, (uintptr_t)&mask_shift8);
+                            TABLE64C(x1, const_8b_m7_m6_m5_m4_m3_m2_m1_0);
                            VLDR64_U12(q1, x1, 0);     // load shift
                            USHL_8(q0, q0, q1); // shift
                            UADDLV_8(q0, q0);   // accumulate
@@ -1366,7 +1377,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                            q1 = fpu_get_scratch(dyn, ninst);
                            d0 = fpu_get_scratch(dyn, ninst);
                            VSHL_8(d0, q0, 7);  // keep only bit 0x80
-                            TABLE64(x1, (uintptr_t)&mask_shift8);
+                            TABLE64C(x1, const_8b_m7_m6_m5_m4_m3_m2_m1_0);
                            VLDR64_U12(q1, x1, 0);     // load shift
                            USHL_8(d0, d0, q1); // shift
                            UADDLV_8(d0, d0);   // accumulate
@@ -1390,10 +1401,10 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        CMPSw_REG(x3, x4);
                        CSELw(x3, x3, x4, cLT); // x3 is lmem
                        CMPSw_REG(x2, x4);
-                        CSELw(x2, x2, x4, cLT); // x2 is lreg
-                        CMPSw_REG(x2, x3);
-                        CSELw(x5, x3, x2, cLT); // x5 is max(lmem, lreg)
-                        CSELw(x2, x2, x3, cLT); // x2 is min(lmem, lreg)
+                        CSELw(x6, x2, x4, cLT); // x6 is lreg
+                        CMPSw_REG(x6, x3);
+                        CSELw(x5, x3, x6, cLT); // x5 is max(lmem, lreg)
+                        CSELw(x2, x6, x3, cLT); // x2 is min(lmem, lreg)
                        // x2 is min length 0-n_packed
                        MVNw_REG(x4, xZR);
                        LSLw_REG(x87pc, x4, x2);
@@ -1429,7 +1440,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                                    BFIw(xFlags, x5, F_ZF, 1);
                                }
                                IFX(X_SF) {
-                                    CMPSw_REG(x2, x4);
+                                    CMPSw_REG(x6, x4);
                                    CSETw(x5, cLT);
                                    BFIw(xFlags, x5, F_SF, 1);
                                }
@@ -1471,7 +1482,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        MOVw_REG(x4, xRAX);
                        u8 = F8;
                        MOV32w(x5, u8);
-                        CALL(sse42_compare_string_explicit_len, x1);
+                        CALL(const_sse42_compare_string_explicit_len, x1);
                    }
                    if(u8&0b1000000) {
                        CBNZw_MARK(x1);
@@ -1507,7 +1518,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    }
                    u8 = F8;
                    MOV32w(x3, u8);
-                    CALL(sse42_compare_string_implicit_len, x1);
+                    CALL(const_sse42_compare_string_implicit_len, x1);
                    q0 = sse_get_reg_empty(dyn, ninst, x2, 0);
                    if(u8&0b1000000) {
                        q1 = fpu_get_scratch(dyn, ninst);
@@ -1517,7 +1528,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                                LSRw_IMM(x1, x1, 8);
                                VDUPQB(q1, x1); // load the high 8bits of the mask
                                VEXTQ_8(q0, q0, q1, 8); // low and hig bits mask
-                                TABLE64(x2, (uintptr_t)&mask_string8);
+                                TABLE64C(x2, const_8b_7_6_5_4_3_2_1_0);
                                VLDR64_U12(q1, x2, 0);     // load shift
                                VDUPQ_64(q1, q1, 0);
                                USHLQ_8(q0, q0, q1); // extract 1 bit
@@ -1527,7 +1538,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                                break;
                            case 0b01:
                                VDUPQH(q0, x1); // load the 8bits of the mask
-                                TABLE64(x2, (uintptr_t)&mask_string16);
+                                TABLE64C(x2, const_8b_15_14_13_12_11_10_9_8);
                                VLDR64_U12(q1, x2, 0);     // load shift
                                UXTL_8(q1, q1);     // extend mask to 16bits
                                USHLQ_16(q0, q0, q1); // extract 1 bit
@@ -1561,7 +1572,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    }
                    u8 = F8;
                    MOV32w(x3, u8);
-                    CALL(sse42_compare_string_implicit_len, x1);
+                    CALL(const_sse42_compare_string_implicit_len, x1);
                    CBNZw_MARK(x1);
                    MOV32w(xRCX, (u8&1)?8:16);
                    B_NEXT_nocond;
@@ -1596,7 +1607,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    }
                    u8 = F8;
                    MOV32w(x4, u8);
-                    CALL(native_aeskeygenassist, -1);
+                    CALL(const_native_aeskeygenassist, -1);
                    break;

                default:
@@ -1605,6 +1616,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            break;

        #define GO(GETFLAGS, NO, YES, F)                                                                                   \
+            if(rex.w) return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);                                   \
            READFLAGS(F);                                                                                                  \
            GETFLAGS;                                                                                                      \
            nextop = F8;                                                                                                   \
@@ -1752,7 +1764,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                x87_restoreround(dyn, ninst, u8);
                VFCVTZSQS(v0, v0);
            } else {
-                if(arm64_frintts) {
+                if(cpuext.frintts) {
                    u8 = sse_setround(dyn, ninst, x1, x2, x3);
                    VFRINT32XSQ(v0, v1); // handle overflow
                    VFCVTZSQS(v0, v0);
@@ -1807,13 +1819,9 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            GETEX(v1, 0, 0);
            // FMIN/FMAX wll not copy the value if v0[x] is NaN
            // but x86 will copy if either v0[x] or v1[x] is NaN, so lets force a copy if source is NaN
-            if(BOX64ENV(dynarec_fastnan)) {
-                VFMINQD(v0, v0, v1);
-            } else {
-                q0 = fpu_get_scratch(dyn, ninst);
-                VFCMGTQD(q0, v1, v0);   // 0 is NaN or v1 GT v0, so invert mask for copy
-                VBIFQ(v0, v1, q0);
-            }
+            q0 = fpu_get_scratch(dyn, ninst);
+            VFCMGTQD(q0, v1, v0);   // 0 is NaN or v1 GT v0, so invert mask for copy
+            VBIFQ(v0, v1, q0);
            break;
        case 0x5E:
            INST_NAME("DIVPD Gx, Ex");
@@ -1842,13 +1850,9 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            GETEX(v1, 0, 0);
            // FMIN/FMAX wll not copy the value if v0[x] is NaN
            // but x86 will copy if either v0[x] or v1[x] is NaN, or if values are equals, so lets force a copy if source is NaN
-            if(BOX64ENV(dynarec_fastnan)) {
-                VFMAXQD(v0, v0, v1);
-            } else {
-                q0 = fpu_get_scratch(dyn, ninst);
-                VFCMGTQD(q0, v0, v1);   // 0 is NaN or v0 GT v1, so invert mask for copy
-                VBIFQ(v0, v1, q0);
-            }
+            q0 = fpu_get_scratch(dyn, ninst);
+            VFCMGTQD(q0, v0, v1);   // 0 is NaN or v0 GT v1, so invert mask for copy
+            VBIFQ(v0, v1, q0);
            break;
        case 0x60:
            INST_NAME("PUNPCKLBW Gx, Ex");
@@ -2374,7 +2378,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n

        case 0xA3:
            INST_NAME("BT Ew, Gw");
-            if(!BOX64ENV(dynarec_safeflags)) {
+            if (!BOX64DRENV(dynarec_safeflags)) {
                SETFLAGS(X_ALL&~X_ZF, SF_SUBSET);
            } else {
                SETFLAGS(X_CF, SF_SUBSET);
@@ -2398,6 +2402,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            }
            break;
        case 0xA4:
+            if(rex.w) return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
            INST_NAME("SHLD Ew, Gw, Ib");
            nextop = F8;
            u8 = geted_ib(dyn, addr, ninst, nextop)&0x1f;
@@ -2414,6 +2419,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            }
            break;
        case 0xA5:
+            if(rex.w) return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
            nextop = F8;
            INST_NAME("SHLD Ew, Gw, CL");
            if(BOX64DRENV(dynarec_safeflags)>1) {
@@ -2433,7 +2439,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n

        case 0xAB:
            INST_NAME("BTS Ew, Gw");
-            if(!BOX64ENV(dynarec_safeflags)) {
+            if (!BOX64DRENV(dynarec_safeflags)) {
                SETFLAGS(X_ALL&~X_ZF, SF_SUBSET);
            } else {
                SETFLAGS(X_CF, SF_SUBSET);
@@ -2465,6 +2471,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            }
            break;
        case 0xAC:
+            if(rex.w) return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
            nextop = F8;
            INST_NAME("SHRD Ew, Gw, Ib");
            u8 = geted_ib(dyn, addr, ninst, nextop)&0x1f;
@@ -2481,6 +2488,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            }
            break;
        case 0xAD:
+            if(rex.w) return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
            nextop = F8;
            INST_NAME("SHRD Ew, Gw, CL");
            if(BOX64DRENV(dynarec_safeflags)>1) {
@@ -2509,21 +2517,13 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                switch((nextop>>3)&7) {
                    case 6:
                        INST_NAME("CLWB Ed");
-                        MESSAGE(LOG_DUMP, "Need Optimization (CLWB)?\n");
                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
-                        if(ed!=x1) {
-                            MOVx_REG(x1, ed);
-                        }
-                        CALL_(native_clflush, -1, 0);
+                        DC_CIVAC(ed);
                        break;
                    case 7:
                        INST_NAME("CLFLUSHOPT Ed");
-                        MESSAGE(LOG_DUMP, "Need Optimization (CLFLUSHOPT)?\n");
                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
-                        if(ed!=x1) {
-                            MOVx_REG(x1, ed);
-                        }
-                        CALL_(native_clflush, -1, 0);
+                        DC_CIVAC(ed);
                        break;
                    default:
                        DEFAULT;
@@ -2531,7 +2531,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            break;
        case 0xAF:
            INST_NAME("IMUL Gw,Ew");
-            if(BOX64ENV(dynarec_safeflags) && BOX64ENV(cputype)) {
+            if (BOX64DRENV(dynarec_safeflags) && BOX64ENV(cputype)) {
                SETFLAGS(X_OF|X_CF, SF_SET);
            } else {
                SETFLAGS(X_ALL, SF_SET);
@@ -2564,7 +2564,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n

        case 0xB3:
            INST_NAME("BTR Ew, Gw");
-            if(!BOX64ENV(dynarec_safeflags)) {
+            if (!BOX64DRENV(dynarec_safeflags)) {
                SETFLAGS(X_ALL&~X_ZF, SF_SUBSET);
            } else {
                SETFLAGS(X_CF, SF_SUBSET);
@@ -2596,6 +2596,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            break;

        case 0xB6:
+            if(rex.w) return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
            INST_NAME("MOVZX Gw, Eb");
            nextop = F8;
            if(MODREG) {
@@ -2617,6 +2618,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            BFIx(gd, x1, 0, 16);        // insert in Gw
            break;
        case 0xB7:
+            if(rex.w) return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
            INST_NAME("MOVZX Gw, Ew");
            nextop = F8;
            if(MODREG) {
@@ -2633,11 +2635,12 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n


        case 0xBA:
+            if(rex.w) return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
            nextop = F8;
            switch((nextop>>3)&7) {
                case 4:
                    INST_NAME("BT Ew, Ib");
-                    if(!BOX64ENV(dynarec_safeflags)) {
+                    if (!BOX64DRENV(dynarec_safeflags)) {
                        SETFLAGS(X_ALL&~X_ZF, SF_SUBSET);
                    } else {
                        SETFLAGS(X_CF, SF_SUBSET);
@@ -2652,7 +2655,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    break;
                case 5:
                    INST_NAME("BTS Ew, Ib");
-                    if(!BOX64ENV(dynarec_safeflags)) {
+                    if (!BOX64DRENV(dynarec_safeflags)) {
                        SETFLAGS(X_ALL&~X_ZF, SF_SUBSET);
                    } else {
                        SETFLAGS(X_CF, SF_SUBSET);
@@ -2669,7 +2672,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    break;
                case 6:
                    INST_NAME("BTR Ew, Ib");
-                    if(!BOX64ENV(dynarec_safeflags)) {
+                    if (!BOX64DRENV(dynarec_safeflags)) {
                        SETFLAGS(X_ALL&~X_ZF, SF_SUBSET);
                    } else {
                        SETFLAGS(X_CF, SF_SUBSET);
@@ -2685,7 +2688,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    break;
                case 7:
                    INST_NAME("BTC Ew, Ib");
-                    if(!BOX64ENV(dynarec_safeflags)) {
+                    if (!BOX64DRENV(dynarec_safeflags)) {
                        SETFLAGS(X_ALL&~X_ZF, SF_SUBSET);
                    } else {
                        SETFLAGS(X_CF, SF_SUBSET);
@@ -2705,8 +2708,9 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            }
            break;
        case 0xBB:
+            if(rex.w) return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
            INST_NAME("BTC Ew, Gw");
-            if(!BOX64ENV(dynarec_safeflags)) {
+            if (!BOX64DRENV(dynarec_safeflags)) {
                SETFLAGS(X_ALL&~X_ZF, SF_SUBSET);
            } else {
                SETFLAGS(X_CF, SF_SUBSET);
@@ -2739,8 +2743,9 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            }
            break;
        case 0xBC:
+            if(rex.w) return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
            INST_NAME("BSF Gw,Ew");
-            if(!BOX64ENV(dynarec_safeflags) || BOX64ENV(cputype)) {
+            if (!BOX64DRENV(dynarec_safeflags) || BOX64ENV(cputype)) {
                SETFLAGS(X_ZF, SF_SUBSET);
            } else {
                SETFLAGS(X_ALL, SF_SET);
@@ -2756,16 +2761,16 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                CBZw_MARK(x1);
            }
            RBITw(x1, x1);   // reverse
-            CLZw(x1, x1);    // x2 gets leading 0 == BSF
-            MARK;
+            CLZw(x1, x1);    // x1 gets leading 0
            BFIx(gd, x1, 0, 16);
+            MARK;
            IFX(X_ZF) {
                IFNATIVE(NF_EQ) {} else {
                    CSETw(x2, cEQ);    //ZF not set
                    BFIw(xFlags, x2, F_ZF, 1);
                }
            }
-            if(BOX64ENV(dynarec_safeflags) && !BOX64ENV(cputype)) {
+            if (BOX64DRENV(dynarec_safeflags) && !BOX64ENV(cputype)) {
                IFX(X_CF) BFCw(xFlags, F_CF, 1);
                IFX(X_AF) BFCw(xFlags, F_AF, 1);
                IFX(X_SF) BFCw(xFlags, F_SF, 1);
@@ -2774,8 +2779,9 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            }
            break;
        case 0xBD:
+            if(rex.w) return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
            INST_NAME("BSR Gw,Ew");
-            if(!BOX64ENV(dynarec_safeflags) || BOX64ENV(cputype)) {
+            if (!BOX64DRENV(dynarec_safeflags) || BOX64ENV(cputype)) {
                SETFLAGS(X_ZF, SF_SUBSET);
            } else {
                SETFLAGS(X_ALL, SF_SET);
@@ -2793,16 +2799,16 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            LSLw(x1, x1, 16);   // put bits on top
            CLZw(x2, x1);       // x2 gets leading 0
            SUBw_U12(x2, x2, 15);
-            NEGw_REG(x1, x2);   // complement
-            MARK;
+            NEGw_REG(x1, x2); // complement
            BFIx(gd, x1, 0, 16);
+            MARK;
            IFX(X_ZF) {
                IFNATIVE(NF_EQ) {} else {
                    CSETw(x2, cEQ);    //ZF not set
                    BFIw(xFlags, x2, F_ZF, 1);
                }
            }
-            if(BOX64ENV(dynarec_safeflags) && !BOX64ENV(cputype)) {
+            if (BOX64DRENV(dynarec_safeflags) && !BOX64ENV(cputype)) {
                IFX(X_CF) BFCw(xFlags, F_CF, 1);
                IFX(X_AF) BFCw(xFlags, F_AF, 1);
                IFX(X_SF) BFCw(xFlags, F_SF, 1);
@@ -2811,6 +2817,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            }
            break;
        case 0xBE:
+            if(rex.w) return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
            INST_NAME("MOVSX Gw, Eb");
            nextop = F8;
            GETGD;
@@ -2833,6 +2840,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            BFIx(gd, x1, 0, 16);
            break;
        case 0xBF:
+            if(rex.w) return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
            INST_NAME("MOVSX Gw, Ew");
            nextop = F8;
            GETGD;
@@ -2848,6 +2856,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            break;

        case 0xC1:
+            if(rex.w) return dynarec64_00(dyn, addr-1, ip, ninst, rex, ok, need_epilog);
            INST_NAME("XADD Gw, Ew");
            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
@@ -2949,8 +2958,8 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            if(rex.w) {
                REV64x(gd, gd);
            } else {
-                REV16w(x1, gd);
-                BFIx(gd, x1, 0, 16);
+                // this is undefined behaviour
+                BFCx(gd, 0, 16);
            }
            break;

@@ -3042,7 +3051,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                q1 = fpu_get_scratch(dyn, ninst);
                GETEX(q0, 0, 0);
                GETGD;
-                TABLE64(x2, 0x0706050403020100LL);
+                TABLE64_(x2, 0x0706050403020100LL);
                VDUPQD(v0, x2);
                VSHRQ_8(q1, q0, 7);
                USHLQ_8(q1, q1, v0); // shift
@@ -3188,7 +3197,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                VFCVTZSQD(v0, v1);  // convert double -> int64
                SQXTN_32(v0, v0);   // convert int64 -> int32 with saturation in lower part, RaZ high part
            } else {
-                if(arm64_frintts) {
+                if(cpuext.frintts) {
                    VFRINT32ZDQ(v0, v1); // handle overflow
                    VFCVTZSQD(v0, v0);  // convert double -> int64
                    SQXTN_32(v0, v0);   // convert int64 -> int32 with saturation in lower part, RaZ high part
@@ -1,273 +0,0 @@
-#include <stdio.h>
-#include <stdlib.h>
-#include <stddef.h>
-#include <errno.h>
-
-#include "debug.h"
-#include "box64context.h"
-#include "box64cpu.h"
-#include "emu/x64emu_private.h"
-#include "x64emu.h"
-#include "box64stack.h"
-#include "callback.h"
-#include "emu/x64run_private.h"
-#include "x64trace.h"
-#include "dynarec_native.h"
-
-#include "arm64_printer.h"
-#include "dynarec_arm64_private.h"
-#include "../dynarec_helper.h"
-#include "dynarec_arm64_functions.h"
-
-uintptr_t dynarec64_6664(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int seg, int* ok, int* need_epilog)
-{
-    (void)ip; (void)need_epilog;
-
-    uint8_t opcode = F8;
-    uint8_t nextop;
-    uint8_t gd, ed, wb1, wback;
-    int64_t j64;
-    int16_t i16;
-    int v0, v1;
-    int64_t fixedaddress;
-    int unscaled;
-    MAYUSE(j64);
-
-    GETREX();
-
-    /*if(rex.w && opcode!=0x0f) {   // rex.w cancels "66", but not for 66 0f type of prefix
-        MESSAGE(LOG_DUMP, "Here!\n");
-        return dynarec64_64(dyn, addr-2, ip, ninst, rex, rep, ok, need_epilog);
-    }*/
-
-    switch(opcode) {
-
-        case 0x0F:
-            opcode = F8;
-            switch(opcode) {
-                case 0x2E:
-                    // no special check...
-                case 0x2F:
-                    if(opcode==0x2F) {INST_NAME("COMISD Gx, Ex");} else {INST_NAME("UCOMISD Gx, Ex");}
-                    SETFLAGS(X_ALL, SF_SET);
-                    nextop = F8;
-                    GETG;
-                    v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
-                    if(MODREG) {
-                        v1 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3), 0);
-                    } else {
-                        grab_segdata(dyn, addr, ninst, x4, seg, (MODREG));
-                        SMREAD();
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                        ADDz_REG(x4, x4, ed);
-                        ed = x4;
-                        v1 = fpu_get_scratch(dyn, ninst);
-                        VLD64(v1, ed, fixedaddress);
-                    }
-                    FCMPD(v0, v1);
-                    FCOMI(x1, x2);
-                    break;
-
-                case 0x6F:
-                    INST_NAME("MOVDQA Gx,Ex");
-                    nextop = F8;
-                    if(MODREG) {
-                        v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);
-                        GETGX_empty(v0);
-                        VMOVQ(v0, v1);
-                    } else {
-                        GETGX_empty(v0);
-                        grab_segdata(dyn, addr, ninst, x4, seg, (MODREG));
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
-                        SMREAD();
-                        ADDz_REG(x4, x4, ed);
-                        VLD128(v0, x4, fixedaddress);
-                    }
-                    break;
-
-                case 0x7F:
-                    INST_NAME("MOVDQA Ex,Gx");
-                    nextop = F8;
-                    GETGX(v0, 0);
-                    if(MODREG) {
-                        v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 1);
-                        VMOVQ(v1, v0);
-                    } else {
-                        grab_segdata(dyn, addr, ninst, x4, seg, (MODREG));
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
-                        ADDz_REG(x4, x4, ed);
-                        VST128(v0, x4, fixedaddress);
-                        SMWRITE2();
-                    }
-                    break;
-
-                case 0xD6:
-                    INST_NAME("MOVQ Ex, Gx");
-                    nextop = F8;
-                    GETG;
-                    v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
-                    if(MODREG) {
-                        v1 = sse_get_reg_empty(dyn, ninst, x1, (nextop&7) + (rex.b<<3));
-                        FMOVD(v1, v0);
-                    } else {
-                        grab_segdata(dyn, addr, ninst, x4, seg, (MODREG));
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                        ADDz_REG(x4, x4, ed);
-                        VST64(v0, x4, fixedaddress);
-                        SMWRITE();
-                    }
-                    break;
-
-                default:
-                    DEFAULT;
-            }
-            break;
-
-        case 0x83:
-            nextop = F8;
-            switch((nextop>>3)&7) {
-                case 0: //ADD
-                    if(opcode==0x81) {INST_NAME("ADD Ew, Iw");} else {INST_NAME("ADD Ew, Ib");}
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    grab_segdata(dyn, addr, ninst, x1, seg, (MODREG));
-                    GETEWO(x1, (opcode==0x81)?2:1);
-                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    MOVZw(x5, i16);
-                    emit_add16(dyn, ninst, ed, x5, x2, x4);
-                    EWBACK;
-                    break;
-                case 1: //OR
-                    if(opcode==0x81) {INST_NAME("OR Ew, Iw");} else {INST_NAME("OR Ew, Ib");}
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    grab_segdata(dyn, addr, ninst, x1, seg, (MODREG));
-                    GETEWO(x1, (opcode==0x81)?2:1);
-                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    emit_or16c(dyn, ninst, x1, i16, x2, x4);
-                    EWBACK;
-                    break;
-                case 2: //ADC
-                    if(opcode==0x81) {INST_NAME("ADC Ew, Iw");} else {INST_NAME("ADC Ew, Ib");}
-                    READFLAGS(X_CF);
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    grab_segdata(dyn, addr, ninst, x1, seg, (MODREG));
-                    GETEWO(x1, (opcode==0x81)?2:1);
-                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    MOVZw(x5, i16);
-                    emit_adc16(dyn, ninst, x1, x5, x2, x4);
-                    EWBACK;
-                    break;
-                case 3: //SBB
-                    if(opcode==0x81) {INST_NAME("SBB Ew, Iw");} else {INST_NAME("SBB Ew, Ib");}
-                    READFLAGS(X_CF);
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    grab_segdata(dyn, addr, ninst, x1, seg, (MODREG));
-                    GETEWO(x1, (opcode==0x81)?2:1);
-                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    MOVZw(x5, i16);
-                    emit_sbb16(dyn, ninst, x1, x5, x2, x4);
-                    EWBACK;
-                    break;
-                case 4: //AND
-                    if(opcode==0x81) {INST_NAME("AND Ew, Iw");} else {INST_NAME("AND Ew, Ib");}
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    grab_segdata(dyn, addr, ninst, x1, seg, (MODREG));
-                    GETEWO(x1, (opcode==0x81)?2:1);
-                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    emit_and16c(dyn, ninst, x1, i16, x2, x4);
-                    EWBACK;
-                    break;
-                case 5: //SUB
-                    if(opcode==0x81) {INST_NAME("SUB Ew, Iw");} else {INST_NAME("SUB Ew, Ib");}
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    grab_segdata(dyn, addr, ninst, x1, seg, (MODREG));
-                    GETEWO(x1, (opcode==0x81)?2:1);
-                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    MOVZw(x5, i16);
-                    emit_sub16(dyn, ninst, x1, x5, x2, x4);
-                    EWBACK;
-                    break;
-                case 6: //XOR
-                    if(opcode==0x81) {INST_NAME("XOR Ew, Iw");} else {INST_NAME("XOR Ew, Ib");}
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    grab_segdata(dyn, addr, ninst, x1, seg, (MODREG));
-                    GETEWO(x1, (opcode==0x81)?2:1);
-                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    emit_xor16c(dyn, ninst, x1, i16, x2, x4);
-                    EWBACK;
-                    break;
-                case 7: //CMP
-                    if(opcode==0x81) {INST_NAME("CMP Ew, Iw");} else {INST_NAME("CMP Ew, Ib");}
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    grab_segdata(dyn, addr, ninst, x1, seg, (MODREG));
-                    GETEWO(x1, (opcode==0x81)?2:1);
-                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    if(i16) {
-                        MOVZw(x2, i16);
-                        emit_cmp16(dyn, ninst, x1, x2, x3, x4, x5);
-                    } else
-                        emit_cmp16_0(dyn, ninst, x1, x3, x4);
-                    break;
-            }
-            break;
-            
-        case 0x89:
-            INST_NAME("MOV FS:Ew, Gw");
-            nextop = F8;
-            GETGD;  // don't need GETGW here
-            if(MODREG) {
-                ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                if(rex.w) {
-                    MOVx_REG(ed, gd);
-                } else {
-                    if(ed!=gd) {
-                        BFIx(ed, gd, 0, 16);
-                    }
-                }
-            } else {
-                grab_segdata(dyn, addr, ninst, x4, seg, (MODREG));
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<(rex.w?3:1), rex.w?7:1, rex, NULL, 0, 0);
-                ADDz_REG(x4, x4, ed);
-                ed = x4;
-                if(rex.w) {
-                    STx(gd, ed, fixedaddress);
-                } else {
-                    STH(gd, ed, fixedaddress);
-                }
-                SMWRITE();
-            }
-            break;
-
-        case 0x8B:
-            INST_NAME("MOV Gw, FS:Ew");
-            nextop=F8;
-            GETGD;
-            if(MODREG) {   // reg <= reg
-                ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                if(rex.w) {
-                    MOVx_REG(gd, ed);
-                } else {
-                    if(ed!=gd) {
-                        BFIx(gd, ed, 0, 16);
-                    }
-                }
-            } else {                    // mem <= reg
-                grab_segdata(dyn, addr, ninst, x4, seg, (MODREG));
-                SMREAD();
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<(rex.w?3:1), rex.w?7:1, rex, NULL, 0, 0);
-                ADDz_REG(x4, x4, ed);
-                ed = x4;
-                if(rex.w) {
-                    LDx(gd, ed, fixedaddress);
-                } else {
-                    LDH(x1, ed, fixedaddress);
-                    BFIx(gd, x1, 0, 16);
-                }
-            }
-            break;
-
-
-        default:
-            DEFAULT;
-    }
-    return addr;
-}
@@ -21,9 +21,9 @@
 #include "dynarec_arm64_functions.h"


-uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
 {
-    (void)ip; (void)rep; (void)need_epilog;
+    (void)ip; (void)need_epilog;

    uint8_t opcode = F8;
    uint8_t nextop;
@@ -39,28 +39,21 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
    MAYUSE(wb2);
    MAYUSE(j64);

-    while((opcode==0xF2) || (opcode==0xF3)) {
-        rep = opcode-0xF1;
-        opcode = F8;
-    }
-
-    GETREX();
-
    switch(opcode) {

        case 0x01:
-            INST_NAME("LOCK ADD Ew, Gw");
-            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
-            GETGW(x5);
            if(MODREG) {
-                ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                UXTHw(x6, ed);
-                emit_add16(dyn, ninst, x6, x5, x3, x4);
-                BFIx(ed, x6, 0, 16);
+                INST_NAME("Invalid LOCK");
+                UDF(0);
+                *need_epilog = 1;
+                *ok = 0;
            } else {
+                INST_NAME("LOCK ADD Ew, Gw");
+                SETFLAGS(X_ALL, SF_SET_PENDING);
+                GETGW(x5);
                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                if(arm64_atomics) {
+                if(cpuext.atomics) {
                    UFLAG_IF {
                        LDADDALH(x5, x1, wback);
                        emit_add16(dyn, ninst, x1, x5, x3, x4);
@@ -74,23 +67,22 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    STLXRH(x3, x1, wback);
                    CBNZx_MARKLOCK(x3);
                }
-                SMDMB();
            }
            break;

        case 0x09:
-            INST_NAME("LOCK OR Ew, Gw");
-            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
-            GETGW(x5);
            if(MODREG) {
-                ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                UXTHw(x6, ed);
-                emit_or16(dyn, ninst, x6, x5, x3, x4);
-                BFIx(ed, x6, 0, 16);
+                INST_NAME("Invalid LOCK");
+                UDF(0);
+                *need_epilog = 1;
+                *ok = 0;
            } else {
+                INST_NAME("LOCK OR Ew, Gw");
+                SETFLAGS(X_ALL, SF_SET_PENDING);
+                GETGW(x5);
                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                if(arm64_atomics) {
+                if(cpuext.atomics) {
                    UFLAG_IF {
                        LDSETALH(x5, x1, wback);
                        emit_or16(dyn, ninst, x1, x5, x3, x4);
@@ -104,7 +96,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    STLXRH(x3, x1, wback);
                    CBNZx_MARKLOCK(x3);
                }
-                SMDMB();
            }
            break;

@@ -113,22 +104,20 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            switch(opcode) {

                case 0xB1:
-                    INST_NAME("LOCK CMPXCHG Ew, Gw");
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    nextop = F8;
-                    GETGD;
-                    UXTHw(x6, xRAX);
                    if(MODREG) {
-                        ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                        wback = 0;
-                        UXTHw(x1, ed);
-                        CMPSxw_REG(x6, x1);
-                        B_MARK(cNE);
-                        BFIx(ed, gd, 0, 16);
+                        INST_NAME("Invalid LOCK");
+                        UDF(0);
+                        *need_epilog = 1;
+                        *ok = 0;
                    } else {
+                        INST_NAME("LOCK CMPXCHG Ew, Gw");
+                        SETFLAGS(X_ALL, SF_SET_PENDING);
+                        GETGD;
+                        UXTHw(x6, xRAX);
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
                        if(!ALIGNED_ATOMICH) {
-                            if(arm64_uscat) {
+                            if(cpuext.uscat) {
                                ANDx_mask(x1, wback, 1, 0, 3);  // mask = F
                                CMPSw_U12(x1, 0xF);
                                B_MARK3(cGE);
@@ -138,7 +127,7 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                            }
                        }
                        // Aligned version
-                        if(arm64_atomics) {
+                        if(cpuext.atomics) {
                            MOVw_REG(x1, x6);
                            CASALH(x1, gd, wback);
                        } else {
@@ -166,28 +155,27 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                            CBNZx_MARK3(x4);
                            STRH_U12(gd, wback, 0);
                        }
+                        MARK;
+                        // Common part (and fallback for EAX != Ed)
+                        UFLAG_IF {emit_cmp16(dyn, ninst, x6, x1, x3, x4, x5);}
+                        BFIx(xRAX, x1, 0, 16);
                    }
-                    MARK;
-                    SMDMB();
-                    // Common part (and fallback for EAX != Ed)
-                    UFLAG_IF {emit_cmp16(dyn, ninst, x6, x1, x3, x4, x5);}
-                    BFIx(xRAX, x1, 0, 16);
                    break;

                case 0xC1:
-                    INST_NAME("LOCK XADD Gw, Ew");
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    nextop = F8;
-                    gd = TO_NAT(((nextop & 0x38) >> 3) + (rex.r << 3));
-                    UXTHx(x5, gd);
                    if(MODREG) {
-                        ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                        BFIx(gd, ed, 0, 16);
-                        emit_add16(dyn, ninst, x5, gd, x3, x4);
-                        BFIx(ed, x5, 0, 16);
+                        INST_NAME("Invalid LOCK");
+                        UDF(0);
+                        *need_epilog = 1;
+                        *ok = 0;
                    } else {
+                        INST_NAME("LOCK XADD Gw, Ew");
+                        SETFLAGS(X_ALL, SF_SET_PENDING);
+                        gd = TO_NAT(((nextop & 0x38) >> 3) + (rex.r << 3));
+                        UXTHx(x5, gd);
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                        if(arm64_atomics) {
+                        if(cpuext.atomics) {
                            LDADDALH(x5, x1, wback);
                        } else {
                            MARKLOCK;
@@ -196,7 +184,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                            STLXRH(x3, x4, wback);
                            CBNZx_MARKLOCK(x3);
                        }
-                        SMDMB();
                        IFX(X_ALL|X_PEND) {
                            MOVxw_REG(x2, x1);
                            emit_add16(dyn, ninst, x2, x5, x3, x4);
@@ -211,40 +198,39 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            break;

        case 0x11:
-            INST_NAME("LOCK ADC Ew, Gw");
-            READFLAGS(X_CF);
-            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
-            GETGW(x5);
            if(MODREG) {
-                ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                UXTHw(x6, ed);
-                emit_adc16(dyn, ninst, x6, x5, x3, x4);
-                BFIx(ed, x6, 0, 16);
+                INST_NAME("Invalid LOCK");
+                UDF(0);
+                *need_epilog = 1;
+                *ok = 0;
            } else {
+                INST_NAME("LOCK ADC Ew, Gw");
+                READFLAGS(X_CF);
+                SETFLAGS(X_ALL, SF_SET_PENDING);
+                GETGW(x5);
                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
                MARKLOCK;
                LDAXRH(x1, wback);
                emit_adc16(dyn, ninst, x1, x5, x3, x4);
                STLXRH(x3, x1, wback);
                CBNZx_MARKLOCK(x3);
-                SMDMB();
            }
            break;

        case 0x21:
-            INST_NAME("LOCK AND Ew, Gw");
-            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
-            GETGW(x5);
            if(MODREG) {
-                ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                UXTHw(x6, ed);
-                emit_and16(dyn, ninst, x6, gd, x3, x4);
-                BFIx(ed, x6, 0, 16);
+                INST_NAME("Invalid LOCK");
+                UDF(0);
+                *need_epilog = 1;
+                *ok = 0;
            } else {
+                INST_NAME("LOCK AND Ew, Gw");
+                SETFLAGS(X_ALL, SF_SET_PENDING);
+                GETGW(x5);
                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                if(arm64_atomics) {
+                if(cpuext.atomics) {
                    MVNw_REG(x3, gd);
                    UFLAG_IF {
                        LDCLRALH(x3, x1, wback);
@@ -259,7 +245,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    STLXRH(x3, x1, wback);
                    CBNZx_MARKLOCK(x3);
                }
-                SMDMB();
            }
            break;

@@ -268,25 +253,23 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            nextop = F8;
            switch((nextop>>3)&7) {
                case 0: //ADD
-                    if(opcode==0x81) {
-                        INST_NAME("LOCK ADD Ew, Iw");
-                    } else {
-                        INST_NAME("LOCK ADD Ew, Ib");
-                    }
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    if(MODREG) {
-                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                        ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                        MOV32w(x5, i16);
-                        UXTHw(x6, ed);
-                        emit_add16(dyn, ninst, x6, x5, x3, x4);
-                        BFIx(ed, x6, 0, 16);
+                        INST_NAME("Invalid LOCK");
+                        UDF(0);
+                        *need_epilog = 1;
+                        *ok = 0;
                    } else {
+                        if(opcode==0x81) {
+                            INST_NAME("LOCK ADD Ew, Iw");
+                        } else {
+                            INST_NAME("LOCK ADD Ew, Ib");
+                        }
+                        SETFLAGS(X_ALL, SF_SET_PENDING);
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
                        MOV32w(x5, i16);
                        if(!ALIGNED_ATOMICH) {
-                            if(arm64_uscat) {
+                            if(cpuext.uscat) {
                                ANDx_mask(x1, wback, 1, 0, 3);  // mask = F
                                CMPSw_U12(x1, 0xF);
                                B_MARK(cGE);
@@ -295,7 +278,7 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                                B_MARK(cNE);
                            }
                        }
-                        if(arm64_atomics) {
+                        if(cpuext.atomics) {
                            UFLAG_IF {
                                LDADDALH(x5, x1, wback);
                            } else {
@@ -308,7 +291,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                            STLXRH(x3, x4, wback);
                            CBNZx_MARKLOCK(x3);
                        }
-                        SMDMB();
                        if(!ALIGNED_ATOMICH) {
                            B_MARK2_nocond;
                            MARK;   // unaligned! also, not enough
@@ -320,7 +302,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                            STLXRB(x3, x4, wback);
                            CBNZx_MARK(x3);
                            STRH_U12(x4, wback, 0);    // put the whole value
-                            SMDMB();
                        }
                        MARK2;
                        UFLAG_IF {
@@ -329,19 +310,18 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    }
                    break;
                case 1: //OR
-                    if(opcode==0x81) {INST_NAME("LOCK OR Ew, Iw");} else {INST_NAME("LOCK OR Ew, Ib");}
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    if(MODREG) {
-                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                        ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                        UXTHw(x6, ed);
-                        emit_or16c(dyn, ninst, x6, i16, x3, x4);
-                        BFIx(ed, x6, 0, 16);
+                        INST_NAME("Invalid LOCK");
+                        UDF(0);
+                        *need_epilog = 1;
+                        *ok = 0;
                    } else {
+                        if(opcode==0x81) {INST_NAME("LOCK OR Ew, Iw");} else {INST_NAME("LOCK OR Ew, Ib");}
+                        SETFLAGS(X_ALL, SF_SET_PENDING);
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
                        if(!i64) {MOV32w(x5, i16);}
-                        if(arm64_atomics) {
+                        if(cpuext.atomics) {
                            UFLAG_IF {
                                LDSETALH(x5, x1, wback);
                                emit_or16c(dyn, ninst, x1, i16, x3, x4);
@@ -361,21 +341,18 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                            STLXRH(x3, x1, wback);
                            CBNZx_MARKLOCK(x3);
                        }
-                        SMDMB();
                    }
                    break;
                case 2: //ADC
-                    if(opcode==0x81) {INST_NAME("LOCK ADC Ew, Iw");} else {INST_NAME("LOCK ADC Ew, Ib");}
-                    READFLAGS(X_CF);
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    if(MODREG) {
-                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                        ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                        MOV32w(x5, i16);
-                        UXTHw(x6, ed);
-                        emit_adc16(dyn, ninst, x6, x5, x3, x4);
-                        BFIx(ed, x6, 0, 16);
+                        INST_NAME("Invalid LOCK");
+                        UDF(0);
+                        *need_epilog = 1;
+                        *ok = 0;
                    } else {
+                        if(opcode==0x81) {INST_NAME("LOCK ADC Ew, Iw");} else {INST_NAME("LOCK ADC Ew, Ib");}
+                        READFLAGS(X_CF);
+                        SETFLAGS(X_ALL, SF_SET_PENDING);
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
                        MOV32w(x5, i16);
@@ -384,21 +361,18 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        emit_adc16(dyn, ninst, x1, x5, x3, x4);
                        STLXRH(x3, x1, wback);
                        CBNZx_MARKLOCK(x3);
-                        SMDMB();
                    }
                    break;
                case 3: //SBB
-                    if(opcode==0x81) {INST_NAME("LOCK SBB Ew, Iw");} else {INST_NAME("LOCK SBB Ew, Ib");}
-                    READFLAGS(X_CF);
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    if(MODREG) {
-                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                        ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                        MOV32w(x5, i16);
-                        UXTHw(x6, ed);
-                        emit_sbb16(dyn, ninst, x6, x5, x3, x4);
-                        BFIx(ed, x6, 0, 16);
+                        INST_NAME("Invalid LOCK");
+                        UDF(0);
+                        *need_epilog = 1;
+                        *ok = 0;
                    } else {
+                        if(opcode==0x81) {INST_NAME("LOCK SBB Ew, Iw");} else {INST_NAME("LOCK SBB Ew, Ib");}
+                        READFLAGS(X_CF);
+                        SETFLAGS(X_ALL, SF_SET_PENDING);
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
                        MOV32w(x5, i16);
@@ -407,23 +381,21 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        emit_sbb16(dyn, ninst, x1, x5, x3, x4);
                        STLXRH(x3, x1, wback);
                        CBNZx_MARKLOCK(x3);
-                        SMDMB();
                    }
                    break;
                case 4: //AND
-                    if(opcode==0x81) {INST_NAME("LOCK AND Ew, Iw");} else {INST_NAME("LOCK AND Ew, Ib");}
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    if(MODREG) {
-                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                        ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                        UXTHw(x6, ed);
-                        emit_and16c(dyn, ninst, x6, i16, x3, x4);
-                        BFIx(ed, x6, 0, 16);
+                        INST_NAME("Invalid LOCK");
+                        UDF(0);
+                        *need_epilog = 1;
+                        *ok = 0;
                    } else {
+                        if(opcode==0x81) {INST_NAME("LOCK AND Ew, Iw");} else {INST_NAME("LOCK AND Ew, Ib");}
+                        SETFLAGS(X_ALL, SF_SET_PENDING);
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
                        i64 = convert_bitmask_w(i16);
-                        if(arm64_atomics) {
+                        if(cpuext.atomics) {
                            MOV32w(x5, ~i16);
                            UFLAG_IF {
                                LDCLRALH(x5, x1, wback);
@@ -447,26 +419,23 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                            }
                            STLXRH(x3, x1, wback);
                            CBNZx_MARKLOCK(x3);
-                            SMDMB();
                        }
                    }
                    break;
                case 5: //SUB
-                    if(opcode==0x81) {INST_NAME("LOCK SUB Ew, Iw");} else {INST_NAME("LOCK SUB Ew, Ib");}
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    if(MODREG) {
-                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                        ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                        MOV32w(x5, i16);
-                        UXTHw(x6, ed);
-                        emit_sub16(dyn, ninst, x6, x5, x3, x4);
-                        BFIx(ed, x6, 0, 16);
+                        INST_NAME("Invalid LOCK");
+                        UDF(0);
+                        *need_epilog = 1;
+                        *ok = 0;
                    } else {
+                        if(opcode==0x81) {INST_NAME("LOCK SUB Ew, Iw");} else {INST_NAME("LOCK SUB Ew, Ib");}
+                        SETFLAGS(X_ALL, SF_SET_PENDING);
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
                        MOV32w(x5, i16);
                        if(!ALIGNED_ATOMICH) {
-                            if(arm64_uscat) {
+                            if(cpuext.uscat) {
                                ANDx_mask(x1, wback, 1, 0, 3);  // mask = F
                                CMPSw_U12(x1, 0xF);
                                B_MARK(cGE);
@@ -475,7 +444,7 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                                B_MARK(cNE);
                            }
                        }
-                        if(arm64_atomics) {
+                        if(cpuext.atomics) {
                            NEGw_REG(x4, x5);
                            UFLAG_IF {
                                LDADDALH(x4, x1, wback);
@@ -489,7 +458,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                            STLXRH(x3, x1, wback);
                            CBNZx_MARKLOCK(x3);
                        }
-                        SMDMB();
                        if(!ALIGNED_ATOMICH) {
                            B_MARK2_nocond;
                            MARK;   // unaligned! also, not enough
@@ -501,7 +469,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                            STLXRB(x3, x4, wback);
                            CBNZx_MARK(x3);
                            STRH_U12(x4, wback, 0);    // put the whole value
-                            SMDMB();
                        }
                        MARK2;
                        UFLAG_IF {
@@ -510,19 +477,18 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    }
                    break;
                case 6: //XOR
-                    if(opcode==0x81) {INST_NAME("LOCK XOR Ew, Iw");} else {INST_NAME("LOCK XOR Ew, Ib");}
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    if(MODREG) {
-                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                        ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                        UXTHw(x6, ed);
-                        emit_xor16c(dyn, ninst, x6, i16, x3, x4);
-                        BFIx(ed, x6, 0, 16);
+                        INST_NAME("Invalid LOCK");
+                        UDF(0);
+                        *need_epilog = 1;
+                        *ok = 0;
                    } else {
+                        if(opcode==0x81) {INST_NAME("LOCK XOR Ew, Iw");} else {INST_NAME("LOCK XOR Ew, Ib");}
+                        SETFLAGS(X_ALL, SF_SET_PENDING);
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
                        i64 = convert_bitmask_w(i16);
-                        if(arm64_atomics) {
+                        if(cpuext.atomics) {
                            UFLAG_IF {
                                LDEORALH(x5, x1, wback);
                                emit_xor16c(dyn, ninst, x1, i16, x3, x4);
@@ -539,23 +505,14 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                                emit_xor16(dyn, ninst, x1, x5, x3, x4);
                            STLXRH(x3, x1, wback);
                            CBNZx_MARKLOCK(x3);
-                            SMDMB();
                        }
                    }
                    break;
                case 7: //CMP
-                    if(opcode==0x81) {INST_NAME("(LOCK) CMP Ew, Iw");} else {INST_NAME("(LOCK) CMP Ew, Ib");}
-                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    GETEW(x6, (opcode==0x81)?2:1);
-                    (void)wb1;
-                    // No need to LOCK, this is readonly
-                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    if(i16) {
-                        MOV32w(x5, i16);
-                        emit_cmp16(dyn, ninst, x6, x5, x3, x4, x6);
-                    } else {
-                        emit_cmp16_0(dyn, ninst, ed, x3, x4);
-                    }
+                    INST_NAME("Invalid LOCK");
+                    UDF(0);
+                    *need_epilog = 1;
+                    *ok = 0;
                    break;
            }
            break;
@@ -565,16 +522,16 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                switch((nextop>>3)&7)
                {
                    case 0: // INC Ew
-                        INST_NAME("LOCK INC Ew");
-                        SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
                        if(MODREG) {
-                            ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                            UXTHw(x6, ed);
-                            emit_inc16(dyn, ninst, x6, x5, x3);
-                            BFIx(ed, x6, 0, 16);
+                            INST_NAME("Invalid LOCK");
+                            UDF(0);
+                            *need_epilog = 1;
+                            *ok = 0;
                        } else {
+                            INST_NAME("LOCK INC Ew");
+                            SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
                            addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                            if(arm64_atomics) {
+                            if(cpuext.atomics) {
                                MOV32w(x3, 1);
                                UFLAG_IF {
                                    LDADDALH(x3, x1, wback);
@@ -589,20 +546,19 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                                STLXRH(x3, x1, wback);
                                CBNZx_MARKLOCK(x3);
                            }
-                            SMDMB();
                        }
                        break;
                    case 1: //DEC Ew
-                        INST_NAME("LOCK DEC Ew");
-                        SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
                        if(MODREG) {
-                            ed = TO_NAT((nextop & 7) + (rex.b << 3));
-                            UXTHw(x6, ed);
-                            emit_dec16(dyn, ninst, x6, x5, x3);
-                            BFIx(ed, x6, 0, 16);
+                            INST_NAME("Invalid LOCK");
+                            UDF(0);
+                            *need_epilog = 1;
+                            *ok = 0;
                        } else {
+                            INST_NAME("LOCK DEC Ew");
+                            SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
                            addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                            if(arm64_atomics) {
+                            if(cpuext.atomics) {
                                MOV32w(x3, -1);
                                UFLAG_IF {
                                    LDADDALH(x3, x1, wback);
@@ -617,7 +573,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                                STLXRH(x3, x1, wback);
                                CBNZx_MARKLOCK(x3);
                            }
-                            SMDMB();
                        }
                        break;
                    default:
@@ -45,12 +45,6 @@ uintptr_t dynarec64_66F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
    MAYUSE(eb1);
    MAYUSE(eb2);
    MAYUSE(j64);
-    #if STEP > 1
-    static const int8_t mask_shift8[] = { -7, -6, -5, -4, -3, -2, -1, 0 };
-    static const int8_t mask_string8[] = { 7, 6, 5, 4, 3, 2, 1, 0 };
-    static const int8_t mask_string16[] = { 15, 14, 13, 12, 11, 10, 9, 8 };
-    static const int8_t round_round[] = { 0, 2, 1, 3};
-    #endif

    switch(opcode) {

@@ -59,11 +53,12 @@ uintptr_t dynarec64_66F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
            switch(opcode) {

                case 0xF1:
+                    if (rex.w) return dynarec64_F20F(dyn, addr - 2, ip, ninst, rex, ok, need_epilog);
                    INST_NAME("CRC32 Gd, Ew");
                    nextop = F8;
                    GETEW(x1, 0);
                    GETGD;
-                    if(arm64_crc32) {
+                    if(cpuext.crc32) {
                        CRC32CH(gd, gd, ed);
                    } else {
                        MOV32w(x2, 0x82f63b78);
@@ -45,12 +45,6 @@ uintptr_t dynarec64_66F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
    MAYUSE(eb1);
    MAYUSE(eb2);
    MAYUSE(j64);
-    #if STEP > 1
-    static const int8_t mask_shift8[] = { -7, -6, -5, -4, -3, -2, -1, 0 };
-    static const int8_t mask_string8[] = { 7, 6, 5, 4, 3, 2, 1, 0 };
-    static const int8_t mask_string16[] = { 15, 14, 13, 12, 11, 10, 9, 8 };
-    static const int8_t round_round[] = { 0, 2, 1, 3};
-    #endif

    switch(opcode) {
        case 0xB8:
@@ -85,7 +79,7 @@ uintptr_t dynarec64_66F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int

        case 0xBC:
            INST_NAME("TZCNT Gw, Ew");
-            if(!BOX64ENV(dynarec_safeflags)) {
+            if (!BOX64DRENV(dynarec_safeflags)) {
                SETFLAGS(X_CF|X_ZF, SF_SUBSET);
            } else {
                if(BOX64ENV(cputype)) {
@@ -112,7 +106,7 @@ uintptr_t dynarec64_66F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
                CSETw(x3, cEQ);
                BFIw(xFlags, x3, F_ZF, 1);  // ZF = is dest 0?
            }
-            if(BOX64ENV(dynarec_safeflags)) {
+            if (BOX64DRENV(dynarec_safeflags)) {
                IFX(X_AF) BFCw(xFlags, F_AF, 1);
                IFX(X_PF) BFCw(xFlags, F_PF, 1);
                IFX(X_SF) BFCw(xFlags, F_SF, 1);
@@ -123,7 +117,7 @@ uintptr_t dynarec64_66F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int

        case 0xBD:
            INST_NAME("LZCNT Gw, Ew");
-            if(!BOX64ENV(dynarec_safeflags)) {
+            if (!BOX64DRENV(dynarec_safeflags)) {
                SETFLAGS(X_CF|X_ZF, SF_SUBSET);
            } else {
                if(BOX64ENV(cputype)) {
@@ -148,7 +142,7 @@ uintptr_t dynarec64_66F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
                CSETw(x3, cEQ);
                BFIw(xFlags, x3, F_ZF, 1);  // ZF = is dest 0?
            }
-            if(BOX64ENV(dynarec_safeflags)) {
+            if (BOX64DRENV(dynarec_safeflags)) {
                IFX(X_AF) BFCw(xFlags, F_AF, 1);
                IFX(X_PF) BFCw(xFlags, F_PF, 1);
                IFX(X_SF) BFCw(xFlags, F_SF, 1);
@@ -1,159 +0,0 @@
-#include <stdio.h>
-#include <stdlib.h>
-#include <stddef.h>
-#include <errno.h>
-
-#include "debug.h"
-#include "box64context.h"
-#include "box64cpu.h"
-#include "emu/x64emu_private.h"
-#include "x64emu.h"
-#include "box64stack.h"
-#include "callback.h"
-#include "emu/x64run_private.h"
-#include "x64trace.h"
-#include "dynarec_native.h"
-
-#include "arm64_printer.h"
-#include "dynarec_arm64_private.h"
-#include "../dynarec_helper.h"
-#include "dynarec_arm64_functions.h"
-
-#define GETGm   gd = ((nextop&0x38)>>3)
-
-uintptr_t dynarec64_6764_32(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int seg, int* ok, int* need_epilog)
-{
-    (void)ip; (void)need_epilog;
-
-    uint8_t opcode = F8;
-    uint8_t nextop;
-    uint8_t gd, ed, wback, wb, wb1, wb2, gb1, gb2, eb1, eb2;
-    int64_t fixedaddress;
-    int unscaled;
-    int8_t  i8;
-    uint8_t u8;
-    int32_t i32;
-    int64_t j64, i64;
-    int cacheupd = 0;
-    int lock;
-    int v0, v1, s0;
-    MAYUSE(i32);
-    MAYUSE(j64);
-    MAYUSE(v0);
-    MAYUSE(v1);
-    MAYUSE(s0);
-    MAYUSE(lock);
-    MAYUSE(cacheupd);
-
-    if(!rex.is32bits) {
-        // should do a different file
-        DEFAULT;
-        return addr;
-    }
-
-    rep = 0;
-    while((opcode==0xF2) || (opcode==0xF3)) {
-        rep = opcode-0xF1;
-        opcode = F8;
-    }
-
-    switch(opcode) {
-
-        case 0x89:
-            INST_NAME("MOV Seg:Ed, Gd");
-            nextop=F8;
-            GETGD;
-            if(MODREG) {   // reg <= reg
-                MOVxw_REG(TO_NAT((nextop & 7) + (rex.b << 3)), gd);
-            } else {                    // mem <= reg
-                grab_segdata(dyn, addr, ninst, x4, seg, (MODREG));
-                addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, 0);
-                ADDz_REG(x4, x4, ed);
-                STz(gd, x4, fixedaddress);
-            }
-            break;
-
-        case 0x8B:
-            INST_NAME("MOV Gd, Seg:Ed");
-            nextop=F8;
-            GETGD;
-            if(MODREG) {   // reg => reg
-                MOVxw_REG(gd, TO_NAT((nextop & 7) + (rex.b << 3)));
-            } else {                    // mem => reg
-                grab_segdata(dyn, addr, ninst, x4, seg, (MODREG));
-                addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, 0);
-                ADDz_REG(x4, x4, ed);
-                LDz(gd, x4, fixedaddress);
-            }
-            break;
-
-        case 0x8F:
-            INST_NAME("POP Seg:Ed");
-            nextop=F8;
-            if(MODREG) {   // reg <= reg
-                POP1_32(x1);
-                MOVxw_REG(TO_NAT((nextop & 7) + (rex.b << 3)), x1);
-            } else {                    // mem <= reg
-                grab_segdata(dyn, addr, ninst, x4, seg, (MODREG));
-                POP1_32(x1);
-                addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, 0);
-                ADDz_REG(x4, x4, ed);
-                STz(x1, x4, fixedaddress);
-            }
-            break;
-
-        case 0xA1:
-            INST_NAME("MOV EAX, Seg:[Od]");
-            i32 = F16;
-            grab_segdata(dyn, addr, ninst, x4, seg, 0);
-            if(i32<4096 && !(i32&3)) {
-                LDRw_U12(xRAX, x4, i32);
-            } else if(i32<256) {
-                LDURw_I9(xRAX, x4, i32);
-            } else {
-                MOV32w(x1, i32);
-                LDRw_REG(xRAX, x4, x1);
-            }
-            break;
-
-        case 0xA3:
-            INST_NAME("MOV Seg:[Od], EAX");
-            i32 = F16;
-            grab_segdata(dyn, addr, ninst, x4, seg, 0);
-            if(i32<4096 && !(i32&3)) {
-                STRw_U12(xRAX, x4, i32);
-            } else if(i32<256) {
-                STURw_I9(xRAX, x4, i32);
-            } else {
-                MOV32w(x1, i32);
-                STRw_REG(xRAX, x4, x1);
-            }
-            break;
-
-        case 0xFF:
-            nextop = F8;
-            grab_segdata(dyn, addr, ninst, x4, seg, (MODREG));
-            switch((nextop>>3)&7) {
-                case 6: // Push Ed
-                    INST_NAME("PUSH FS:Ew");
-                    if (MODREG) {
-                        DEFAULT;
-                    } else {
-                        SMREAD();
-                        addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, 0);
-                        ADDz_REG(x4, x4, ed);
-                        LDW(x3, x4, fixedaddress);
-                        PUSH1_32(x3);
-                    }
-                    break;
-                default:
-                    DEFAULT;
-            }
-            break;
-
-
-        default:
-            DEFAULT;
-    }
-    return addr;
-}
@@ -1,157 +0,0 @@
-#include <stdio.h>
-#include <stdlib.h>
-#include <stddef.h>
-#include <errno.h>
-
-#include "debug.h"
-#include "box64context.h"
-#include "box64cpu.h"
-#include "emu/x64emu_private.h"
-#include "x64emu.h"
-#include "box64stack.h"
-#include "callback.h"
-#include "emu/x64run_private.h"
-#include "x64trace.h"
-#include "dynarec_native.h"
-
-#include "arm64_printer.h"
-#include "dynarec_arm64_private.h"
-#include "../dynarec_helper.h"
-#include "dynarec_arm64_functions.h"
-
-#define GETGm   gd = ((nextop&0x38)>>3)
-
-uintptr_t dynarec64_67_32(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
-{
-    (void)ip; (void)need_epilog;
-
-    uint8_t opcode = F8;
-    uint8_t nextop;
-    uint8_t gd, ed, wback, wb, wb1, wb2, gb1, gb2, eb1, eb2;
-    int64_t fixedaddress;
-    int unscaled;
-    int8_t  i8;
-    uint8_t u8;
-    int32_t i32;
-    int64_t j64, i64;
-    int cacheupd = 0;
-    int lock;
-    int v0, v1, s0;
-    MAYUSE(i32);
-    MAYUSE(j64);
-    MAYUSE(v0);
-    MAYUSE(v1);
-    MAYUSE(s0);
-    MAYUSE(lock);
-    MAYUSE(cacheupd);
-
-    if(!rex.is32bits) {
-        // should do a different file
-        DEFAULT;
-        return addr;
-    }
-
-    rep = 0;
-    while((opcode==0xF2) || (opcode==0xF3)) {
-        rep = opcode-0xF1;
-        opcode = F8;
-    }
-
-    switch(opcode) {
-        
-        case 0x0F:
-            opcode=F8;
-            switch(opcode) {
-
-                case 0x29:
-                    INST_NAME("MOVAPS Ex,Gx");
-                    nextop = F8;
-                    GETG;
-                    v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
-                    if(MODREG) {
-                        ed = (nextop&7)+(rex.b<<3);
-                        v1 = sse_get_reg_empty(dyn, ninst, x1, ed);
-                        VMOVQ(v1, v0);
-                    } else {
-                        addr = geted16(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, 0);
-                        VST128(v0, ed, fixedaddress);
-                        SMWRITE2();
-                    }
-                    break;
-
-                default:
-                    DEFAULT;
-            }
-            break;
-
-        case 64:
-            addr = dynarec64_6764_32(dyn, addr, ip, ninst, rex, rep, _FS, ok, need_epilog);
-            break;
-
-        #define GO(NO, YES)                                             \
-            BARRIER(BARRIER_MAYBE);                                     \
-            JUMP(addr+i8, 1);                                           \
-            if(dyn->insts[ninst].x64.jmp_insts==-1 ||                   \
-                CHECK_CACHE()) {                                        \
-                /* out of the block */                                  \
-                i32 = dyn->insts[ninst].epilog-(dyn->native_size);      \
-                Bcond(NO, i32);                                         \
-                if(dyn->insts[ninst].x64.jmp_insts==-1) {               \
-                    if(!(dyn->insts[ninst].x64.barrier&BARRIER_FLOAT))  \
-                        fpu_purgecache(dyn, ninst, 1, x1, x2, x3);      \
-                    jump_to_next(dyn, addr+i8, 0, ninst, rex.is32bits); \
-                } else {                                                \
-                    CacheTransform(dyn, ninst, cacheupd, x1, x2, x3);   \
-                    i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->native_size);\
-                    SKIP_SEVL(i32);                                     \
-                    B(i32);                                             \
-                }                                                       \
-            } else {                                                    \
-                /* inside the block */                                  \
-                i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->native_size);    \
-                SKIP_SEVL(i32);                                         \
-                Bcond(YES, i32);                                        \
-            }
-        case 0xE0:
-            INST_NAME("LOOPNZ (16bits)");
-            READFLAGS(X_ZF);
-            i8 = F8S;
-            UXTHw(x1, xRCX);
-            SUBw_U12(x1, x1, 1);
-            BFIx(xRCX, x1, 0, 16);
-            CBZw_NEXT(x1);    // ECX is 0, no LOOP
-            TSTw_mask(xFlags, 0b011010, 0); //mask=0x40
-            GO(cNE, cEQ);
-            break;
-        case 0xE1:
-            INST_NAME("LOOPZ (16bits)");
-            READFLAGS(X_ZF);
-            i8 = F8S;
-            UXTHw(x1, xRCX);
-            SUBw_U12(x1, x1, 1);
-            BFIx(xRCX, x1, 0, 16);
-            CBZw_NEXT(x1);    // ECX is 0, no LOOP
-            TSTw_mask(xFlags, 0b011010, 0); //mask=0x40
-            GO(cEQ, cNE);
-            break;
-        case 0xE2:
-            INST_NAME("LOOP (16bits)");
-            i8 = F8S;
-            UXTHw(x1, xRCX);
-            SUBSw_U12(x1, x1, 1);
-            BFIx(xRCX, x1, 0, 16);
-            GO(cEQ, cNE);
-            break;
-        case 0xE3:
-            INST_NAME("JCXZ");
-            i8 = F8S;
-            TSTw_mask(xRCX, 0, 15); //mask=0xffff
-            GO(cNE, cEQ);
-            break;
-        #undef GO
-        
-        default:
-            DEFAULT;
-    }
-    return addr;
-}
@@ -1,154 +0,0 @@
-#include <stdio.h>
-#include <stdlib.h>
-#include <stddef.h>
-#include <errno.h>
-
-#include "debug.h"
-#include "box64context.h"
-#include "box64cpu.h"
-#include "emu/x64emu_private.h"
-#include "emu/x64run_private.h"
-#include "x64emu.h"
-#include "box64stack.h"
-#include "callback.h"
-#include "emu/x64run_private.h"
-#include "x64trace.h"
-#include "dynarec_native.h"
-
-#include "arm64_printer.h"
-#include "dynarec_arm64_private.h"
-#include "dynarec_arm64_functions.h"
-#include "../dynarec_helper.h"
-
-static const char* avx_prefix_string(uint16_t p)
-{
-    switch(p) {
-        case VEX_P_NONE: return "0";
-        case VEX_P_66: return "66";
-        case VEX_P_F2: return "F2";
-        case VEX_P_F3: return "F3";
-        default: return "??";
-    }
-}
-static const char* avx_map_string(uint16_t m)
-{
-    switch(m) {
-        case VEX_M_NONE: return "0";
-        case VEX_M_0F: return "0F";
-        case VEX_M_0F38: return "0F38";
-        case VEX_M_0F3A: return "0F3A";
-        default: return "??";
-    }
-}
-
-uintptr_t dynarec64_67_AVX(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, vex_t vex, int* ok, int* need_epilog)
-{
-    (void)ip; (void)need_epilog;
-    uint8_t opcode = F8;
-    uint8_t nextop, u8;
-    uint8_t gd, ed, vd;
-    uint8_t wback, wb1, wb2;
-    uint8_t eb1, eb2, gb1, gb2;
-    int32_t i32, i32_;
-    int cacheupd = 0;
-    int v0, v1, v2;
-    int q0, q1, q2;
-    int d0, d1, d2;
-    int s0;
-    uint64_t tmp64u;
-    int64_t j64;
-    int64_t fixedaddress;
-    int unscaled;
-    MAYUSE(wb1);
-    MAYUSE(wb2);
-    MAYUSE(eb1);
-    MAYUSE(eb2);
-    MAYUSE(gb1);
-    MAYUSE(gb2);
-    MAYUSE(q0);
-    MAYUSE(q1);
-    MAYUSE(d0);
-    MAYUSE(d1);
-    MAYUSE(s0);
-    MAYUSE(j64);
-    MAYUSE(cacheupd);
-
-    rex_t rex = vex.rex;
-
-    if( (vex.m==VEX_M_0F38) && (vex.p==VEX_P_F2)) {
-        switch(opcode) {
-
-            case 0xF6:
-                INST_NAME("MULX Gd, Vd, Ed (,RDX)");
-                nextop = F8;
-                GETGD;
-                GETED32(0);
-                GETVD;
-                    if(rex.w) {
-                        // 64bits mul
-                    if((gd==xRDX) || (gd==ed) || (gd==vd))
-                        gb1 = x3;
-                    else
-                        gb1 = gd;
-                    UMULH(gb1, xRDX, ed);
-                    if(gd!=vd) {MULx(vd, xRDX, ed);}
-                    if(gb1==x3) {
-                        MOVx_REG(gd, gb1);
-                    }
-                } else {
-                    // 32bits mul
-                    UMULL(x3, xRDX, ed);
-                    if(gd!=vd) {MOVw_REG(vd, x3);}
-                    LSRx(gd, x3, 32);
-                }
-                break;
-
-            default:
-                DEFAULT;
-        }
-        
-    }
-    else if((vex.m==VEX_M_0F) && (vex.p==VEX_P_66)) {
-        switch(opcode) {
-            case 0xD6:
-                INST_NAME("VMOVQ Ex, Gx");
-                nextop = F8;
-                GETG;
-                v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
-                if(MODREG) {
-                    ed = (nextop&7)+(rex.b<<3);
-                    v1 = sse_get_reg_empty(dyn, ninst, x1, ed);
-                    VMOV(v1, v0);
-                    YMM0(ed);
-                } else {
-                    addr = geted32(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                    VST64(v0, ed, fixedaddress);
-                }
-                break;
-
-            case 0x6E:
-                INST_NAME("VMOVD Gx, Ed");
-                nextop = F8;
-                GETGX_empty(v0);
-                GETED(0);
-                VEORQ(v0, v0, v0); // RAZ vector
-                if(rex.w) {
-                    FMOVDx(v0, ed);
-                } else {
-                    FMOVSw(v0, ed);
-                }
-                YMM0(gd);
-                break;
-
-            default:
-                DEFAULT;
-        }
-    }
-    else {DEFAULT;}
-
-    if((*ok==-1) && (BOX64ENV(dynarec_log)>=LOG_INFO || dyn->need_dump || BOX64ENV(dynarec_missing)==1))
-        if(!dyn->size || BOX64ENV(dynarec_log)>LOG_INFO || dyn->need_dump) {
-            dynarec_log(LOG_NONE, "  Dynarec unimplemented VEX opcode size %d prefix %s map %s opcode %02X\n", 128 << vex.l, avx_prefix_string(vex.p), avx_map_string(vex.m), opcode);
-    }
-    return addr;
-}
@@ -82,13 +82,13 @@ typedef struct arch_build_s

 static arch_build_t static_build[MAX_INSTS+2] = {0};

-static int arch_build(dynarec_arm_t* dyn, int ninst, arch_build_t* arch)
+static int arch_build(dynarec_arm_t* dyn, int ninst, arch_build_t* arch, int noarch)
 {
    memset(arch, 0, sizeof(arch_build_t));
    // flags
-    if((dyn->insts[ninst].f_entry.dfnone==0) || dyn->insts[ninst].need_nat_flags || dyn->insts[ninst].before_nat_flags) {
+    if(((dyn->insts[ninst].f_entry!=status_none && dyn->insts[ninst].f_entry!=status_none_pending)) || dyn->insts[ninst].need_nat_flags || dyn->insts[ninst].before_nat_flags) {
        arch->flags = 1;
-        arch->flags_.defered = dyn->insts[ninst].f_entry.dfnone==0;
+        arch->flags_.defered = (dyn->insts[ninst].f_entry!=status_none && dyn->insts[ninst].f_entry!=status_none_pending);
        uint8_t flags = dyn->insts[ninst].need_nat_flags | dyn->insts[ninst].before_nat_flags;
        arch->flags_.vf = flags&NF_VF;
        arch->flags_.nf = flags&NF_SF;
@@ -97,65 +97,67 @@ static int arch_build(dynarec_arm_t* dyn, int ninst, arch_build_t* arch)
        if(arch->flags_.cf)
            arch->flags_.inv_cf = !dyn->insts[ninst].normal_carry;
    }
-    // got through all naoncache to gather regs assignments
-    int idx;
-    for(int i=0; i<32; ++i)
-        if(dyn->insts[ninst].n.neoncache[i].v!=-1)
-            switch(dyn->insts[ninst].n.neoncache[i].t) {
-                case NEON_CACHE_XMMW:
-                    arch->sse = 1;
-                    arch->sse_.sse |= 1<<dyn->insts[ninst].n.neoncache[i].n;
-                    break;
-                case NEON_CACHE_MM:
-                    arch->mmx = 1;
-                    arch->mmx_.mmx |= 1<<dyn->insts[ninst].n.neoncache[i].n;
-                    break;
-                case NEON_CACHE_YMMW:
-                    arch->ymm = 1;
-                    arch->ymm_.ymm |= 1<<dyn->insts[ninst].n.neoncache[i].n;
-                    idx = i;
-                    if(idx>=EMM0 && idx<=EMM0+8)
-                        idx-=EMM0;
-                    else
-                        idx-=SCRATCH0-8;
-                    arch->ymm_.ymm_pos |= idx<<(dyn->insts[ninst].n.neoncache[i].n*4);
-                    break;
-                case NEON_CACHE_ST_D:
-                    arch->x87 = 1;
-                    arch->x87_.x87 |= 1<<dyn->insts[ninst].n.neoncache[i].n;
-                    arch->x87_.x87_pos = (i-EMM0)<<(dyn->insts[ninst].n.neoncache[i].n*4);
-                    arch->x87_.x87_type = (X87_ST_D)<<(dyn->insts[ninst].n.neoncache[i].n*2);
-                    break;
-                case NEON_CACHE_ST_F:
-                    arch->x87 = 1;
-                    arch->x87_.x87 |= 1<<dyn->insts[ninst].n.neoncache[i].n;
-                    arch->x87_.x87_pos = (i-EMM0)<<(dyn->insts[ninst].n.neoncache[i].n*4);
-                    arch->x87_.x87_type = (X87_ST_F)<<(dyn->insts[ninst].n.neoncache[i].n*2);
-                    break;
-                case NEON_CACHE_ST_I64:
-                    arch->x87 = 1;
-                    arch->x87_.x87 |= 1<<dyn->insts[ninst].n.neoncache[i].n;
-                    arch->x87_.x87_pos = (i-EMM0)<<(dyn->insts[ninst].n.neoncache[i].n*4);
-                    arch->x87_.x87_type = (X87_ST_I64)<<(dyn->insts[ninst].n.neoncache[i].n*2);
-                    break;
-                case NEON_CACHE_XMMR:
-                case NEON_CACHE_YMMR:
-                default:
-                    // doing nothing, it's just a value read in memory
-                    break;
-            }
-    // ymm0
-    if(dyn->insts[ninst].ymm0_out) {
-        arch->ymm = 1;
-        arch->ymm_.ymm0 = dyn->insts[ninst].ymm0_out;
-    }
-    // x87 top
-    if(dyn->insts[ninst].n.x87stack) {
-        arch->x87 = 1;
-        arch->x87_.delta = dyn->insts[ninst].n.x87stack;
-    }
    // opcode can handle unaligned
    arch->unaligned = dyn->insts[ninst].unaligned;
+    if(!noarch) {
+        // got through all naoncache to gather regs assignments
+        int idx;
+        for(int i=0; i<32; ++i)
+            if(dyn->insts[ninst].n.neoncache[i].v)
+                switch(dyn->insts[ninst].n.neoncache[i].t) {
+                    case NEON_CACHE_XMMW:
+                        arch->sse = 1;
+                        arch->sse_.sse |= 1<<dyn->insts[ninst].n.neoncache[i].n;
+                        break;
+                    case NEON_CACHE_MM:
+                        arch->mmx = 1;
+                        arch->mmx_.mmx |= 1<<dyn->insts[ninst].n.neoncache[i].n;
+                        break;
+                    case NEON_CACHE_YMMW:
+                        arch->ymm = 1;
+                        arch->ymm_.ymm |= 1<<dyn->insts[ninst].n.neoncache[i].n;
+                        idx = i;
+                        if(idx>=EMM0 && idx<=EMM0+8)
+                            idx-=EMM0;
+                        else
+                            idx-=SCRATCH0-8;
+                        arch->ymm_.ymm_pos |= idx<<(dyn->insts[ninst].n.neoncache[i].n*4);
+                        break;
+                    case NEON_CACHE_ST_D:
+                        arch->x87 = 1;
+                        arch->x87_.x87 |= 1<<dyn->insts[ninst].n.neoncache[i].n;
+                        arch->x87_.x87_pos = (i-EMM0)<<(dyn->insts[ninst].n.neoncache[i].n*4);
+                        arch->x87_.x87_type = (X87_ST_D)<<(dyn->insts[ninst].n.neoncache[i].n*2);
+                        break;
+                    case NEON_CACHE_ST_F:
+                        arch->x87 = 1;
+                        arch->x87_.x87 |= 1<<dyn->insts[ninst].n.neoncache[i].n;
+                        arch->x87_.x87_pos = (i-EMM0)<<(dyn->insts[ninst].n.neoncache[i].n*4);
+                        arch->x87_.x87_type = (X87_ST_F)<<(dyn->insts[ninst].n.neoncache[i].n*2);
+                        break;
+                    case NEON_CACHE_ST_I64:
+                        arch->x87 = 1;
+                        arch->x87_.x87 |= 1<<dyn->insts[ninst].n.neoncache[i].n;
+                        arch->x87_.x87_pos = (i-EMM0)<<(dyn->insts[ninst].n.neoncache[i].n*4);
+                        arch->x87_.x87_type = (X87_ST_I64)<<(dyn->insts[ninst].n.neoncache[i].n*2);
+                        break;
+                    case NEON_CACHE_XMMR:
+                    case NEON_CACHE_YMMR:
+                    default:
+                        // doing nothing, it's just a value read in memory
+                        break;
+                }
+        // ymm0
+        if(dyn->insts[ninst].ymm0_out) {
+            arch->ymm = 1;
+            arch->ymm_.ymm0 = dyn->insts[ninst].ymm0_out;
+        }
+        // x87 top
+        if(dyn->insts[ninst].n.x87stack) {
+            arch->x87 = 1;
+            arch->x87_.delta = dyn->insts[ninst].n.x87stack;
+        }
+    }
    return arch->flags + arch->x87 + arch->mmx + arch->sse + arch->ymm + arch->unaligned;
 }

@@ -179,6 +181,8 @@ static int sizeof_arch_build(arch_build_t* build)

 size_t get_size_arch(dynarec_arm_t* dyn)
 {
+    int noarch = BOX64ENV(dynarec_noarch);
+    if(noarch>1) return 0;
    arch_build_t* previous = NULL;
    size_t sz = 0;
    int seq = 0;
@@ -187,7 +191,7 @@ size_t get_size_arch(dynarec_arm_t* dyn)
    if(!dyn->size) return 0;
    for(int i=0; i<dyn->size; ++i) {
        arch_build_t* build = static_build+i;
-        last = arch_build(dyn, i, build);
+        last = arch_build(dyn, i, build, noarch);
        if(i && (!memcmp(build, previous, sizeof(arch_build_t))) && (seq<((1<<10)-1))) {
            // same sequence, increment
            ++seq;
@@ -282,10 +286,10 @@ void adjust_arch(dynablock_t* db, x64emu_t* emu, ucontext_t* p, uintptr_t x64pc)
    SUPER()
    #undef GO
    int i = 0;
-    while(i<ninst-1) {
+    while(i<(ninst)) {  // this will look for ninst-1 state
        arch = next;
+        dynarec_log_prefix(0, (i<(ninst-1))?LOG_DEBUG:LOG_INFO, "[ seq=%d%s%s%s%s%s ] ", arch->seq, arch->flags?" Flags":"", arch->x87?" x87":"", arch->mmx?" MMX":"", arch->sse?" SSE":"", arch->ymm?" YMM":"");
        i += 1+arch->seq;
-        dynarec_log_prefix(0, (i<ninst-1)?LOG_DEBUG:LOG_INFO, "[ seq=%d%s%s%s%s%s ] ", arch->seq, arch->flags?" Flags":"", arch->x87?" x87":"", arch->mmx?" MMX":"", arch->sse?" SSE":"", arch->ymm?" YMM":"");
        next = (arch_arch_t*)((uintptr_t)next + sizeof_arch(arch));
    }
    int sz = sizeof(arch_arch_t);
@@ -54,9 +54,6 @@ uintptr_t dynarec64_AVX_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
    MAYUSE(s0);
    MAYUSE(j64);
    MAYUSE(cacheupd);
-    #if STEP > 1
-    static const int8_t mask_shift8[] = { -7, -6, -5, -4, -3, -2, -1, 0 };
-    #endif

    rex_t rex = vex.rex;

@@ -141,12 +138,12 @@ uintptr_t dynarec64_AVX_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
            break;
        case 0x13:
            nextop = F8;
-            INST_NAME("VMOVLPS Ex, Gx");
-            GETGX(v0, 0);
            if(MODREG) {
-                DEFAULT;
-                return addr;
+                INST_NAME("Illegal AVX.0F 13");
+                UDF(0);
            } else {
+                INST_NAME("VMOVLPS Ex, Gx");
+                GETGX(v0, 0);
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                VST1_64(v0, 0, ed);
                SMWRITE2();
@@ -265,19 +262,14 @@ uintptr_t dynarec64_AVX_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
            break;

        case 0x2B:
-            INST_NAME("VMOVNTPS Ex, Gx");
            nextop = F8;
-            GETG;
-            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
            if(MODREG) {
-                ed = (nextop&7)+(rex.b<<3);
-                v1 = sse_get_reg_empty(dyn, ninst, x1, ed);
-                VMOVQ(v1, v0);
-                if(vex.l) {
-                    GETGYEY_empty(v0, v1);
-                    VMOVQ(v1, v0);
-                } else YMM0(ed);
+                INST_NAME("Illegal AVX.0F 2B");
+                UDF(0);
            } else {
+                INST_NAME("VMOVNTPS Ex, Gx");
+                GETG;
+                v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
                if(vex.l) {
                    GETGY(v1, 0, -1, -1, -1);
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
@@ -293,11 +285,13 @@ uintptr_t dynarec64_AVX_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
            // no special check...
        case 0x2F:
            if(opcode==0x2F) {INST_NAME("VCOMISS Gx, Ex");} else {INST_NAME("VUCOMISS Gx, Ex");}
-            SETFLAGS(X_ALL, SF_SET_NODF);
+            SETFLAGS(X_ALL, SF_SET);
            nextop = F8;
            GETGX(v0, 0);
            GETEXSS(s0, 0, 0);
-            FCMPS(v0, s0);
+            IFX(X_CF|X_PF|X_ZF) {
+                FCMPS(v0, s0);
+            }
            FCOMI(x1, x2);
            break;

@@ -509,7 +503,12 @@ uintptr_t dynarec64_AVX_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
        case 0x5A:
            INST_NAME("VCVTPS2PD Gx, Ex");
            nextop = F8;
-            GETGX_empty_EX(v0, v1, 0);
+            if (vex.l) {
+                GETGX_empty_EX(v0, v1, 0);
+            } else {
+                GETEXSD(v1, 0, 0);
+                GETGX_empty(v0);
+            }
            if(vex.l) {
                GETGY_empty(q0, -1, -1, -1);
                FCVTL2(q0, v1);
@@ -553,20 +552,14 @@ uintptr_t dynarec64_AVX_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
        case 0x5D:
            INST_NAME("VMINPS Gx, Vx, Ex");
            nextop = F8;
-            if(!BOX64ENV(dynarec_fastnan)) {
-                q0 = fpu_get_scratch(dyn, ninst);
-            }
+            q0 = fpu_get_scratch(dyn, ninst);
            for(int l=0; l<1+vex.l; ++l) {
                if(!l) { GETGX_empty_VXEX(v0, v2, v1, 0); } else { GETGY_empty_VYEY(v0, v2, v1); }
                // FMIN/FMAX wll not copy a NaN if either is NaN
                // but x86 will copy src2 if either value is NaN, so lets force a copy of Src2 (Ex) if result is NaN
-                if(BOX64ENV(dynarec_fastnan)) {
-                    VFMINQS(v0, v2, v1);
-                } else {
-                    VFCMGTQS(q0, v1, v2);   // 0 if NaN or v1 GT v2, so invert mask for copy
-                    if(v0!=v1) VBIFQ(v0, v1, q0);
-                    if(v0!=v2) VBITQ(v0, v2, q0);
-                }
+                VFCMGTQS(q0, v1, v2);   // 0 if NaN or v1 GT v2, so invert mask for copy
+                if(v0!=v1) VBIFQ(v0, v1, q0);
+                if(v0!=v2) VBITQ(v0, v2, q0);
            }
            if(!vex.l) YMM0(gd);
            break;
@@ -597,20 +590,14 @@ uintptr_t dynarec64_AVX_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
        case 0x5F:
            INST_NAME("VMAXPS Gx, Vx, Ex");
            nextop = F8;
-            if(!BOX64ENV(dynarec_fastnan)) {
-                q0 = fpu_get_scratch(dyn, ninst);
-            }
+            q0 = fpu_get_scratch(dyn, ninst);
            for(int l=0; l<1+vex.l; ++l) {
                if(!l) { GETGX_empty_VXEX(v0, v2, v1, 0); } else { GETGY_empty_VYEY(v0, v2, v1); }
                // FMIN/FMAX wll not copy a NaN if either is NaN
                // but x86 will copy src2 if either value is NaN, so lets force a copy of Src2 (Ex) if result is NaN
-                if(BOX64ENV(dynarec_fastnan)) {
-                    VFMAXQS(v0, v2, v1);
-                } else {
-                    VFCMGTQS(q0, v2, v1);   // 0 if NaN or v2 GT v1, so invert mask for copy
-                    if(v0!=v1) VBIFQ(v0, v1, q0);
-                    if(v0!=v2) VBITQ(v0, v2, q0);
-                }
+                VFCMGTQS(q0, v2, v1);   // 0 if NaN or v2 GT v1, so invert mask for copy
+                if(v0!=v1) VBIFQ(v0, v1, q0);
+                if(v0!=v2) VBITQ(v0, v2, q0);
            }
            if(!vex.l) YMM0(gd);
            break;
@@ -660,21 +647,21 @@ uintptr_t dynarec64_AVX_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
                        if(BOX64ENV(sse_flushto0)) {
                            // try to sync mxcsr with fpsr on the flag side
                            /* mapping is 
-                                ARM -> X86
-                                0 -> 0  // Invalid operation
-                                1 -> 2  // Divide by 0
-                                2 -> 3  // Overflow
-                                3 -> 4  // underflow
-                                4 -> 5  // Inexact
-                                5 -> 1  // denormal
+                                ARM <-> X86
+                                 0  <->  0  // Invalid operation
+                                 1  <->  2  // Divide by 0
+                                 2  <->  3  // Overflow
+                                 3  <->  4  // underflow
+                                 4  <->  5  // Inexact
+                                 5  <->  1  // denormal
                            */
-                            // doing X86 -> ARM here, 0 1 2 3 4 5 -> 0 2 3 4 5 1
+                            // doing X86 -> ARM here, 543210 => 432150
                            if(ed!=x1)
                                MOVw_REG(x1, ed);   // x1 = 543210
-                            RORw(x3, x1, 2);    // x3 = 10.....5432
-                            BFIw(x1, x3, 1, 4); // x1 = 54320
-                            RORw(x3, x3, 32-1); // x3 = 0.....54321
-                            BFIw(x1, x3, 5, 1); // x1 = 154320
+                            UBFXw(x2, x1, 1, 5);   // x2 = 54321
+                            BFIw(x1, x2, 2, 4); // x1 = 432110
+                            LSRw(x2, x2, 4);    // x2 = 5
+                            BFIw(x1, x2, 1, 1); // x1 = 432150
                            MRS_fpsr(x2);
                            BFIx(x2, x1, 0, 6);
                            MSR_fpsr(x2);
@@ -684,18 +671,18 @@ uintptr_t dynarec64_AVX_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int
                        INST_NAME("VSTMXCSR Md");
                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
                        LDRw_U12(x4, xEmu, offsetof(x64emu_t, mxcsr));
-                        STW(x4, ed, fixedaddress);
                        if(BOX64ENV(sse_flushto0)) {
                            // sync with fpsr, with mask from mxcsr
                            MRS_fpsr(x1);
-                            // doing ARM -> X86 here,  543210 => 432150
-                            UBFXw(x2, x1, 1, 5);   // x2 = 54321
-                            BFIw(x1, x2, 2, 4); // x1 = 432110
-                            LSRw(x2, x2, 4);    // x2 = 5
-                            BFIw(x1, x2, 1, 1); // x1 = 432150
+                            // doing ARM -> X86 here,  0 1 2 3 4 5 -> 0 2 3 4 5 1
+                            RORw(x3, x1, 2);    // x3 = 10.....5432
+                            BFIw(x1, x3, 1, 4); // x1 = 54320
+                            RORw(x3, x3, 32-1); // x3 = 0.....54321
+                            BFIw(x1, x3, 5, 1); // x1 = 154320
                            //BFXILw(x3, x4, 7, 6); // this would the mask, but let's ignore that for now
                            BFIw(x4, x1, 0, 6); // inject back the flags
                        }
+                        STW(x4, ed, fixedaddress);
                        break;
                    default:
                        DEFAULT;
@@ -132,9 +132,14 @@ uintptr_t dynarec64_AVX_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, i
                    IFX(X_OF) {
                        IFNATIVE(NF_VF) {} else { BFCw(xFlags, F_OF, 1); }
                    }
-                    if (BOX64ENV(dynarec_test)) {
+                    if (BOX64DRENV(dynarec_safeflags)) {
+                        // those are UD flags
                        IFX(X_AF) BFCw(xFlags, F_AF, 1);
-                        IFX(X_PF) BFCw(xFlags, F_PF, 1);
+                        if(BOX64ENV(cputype)) {
+                            IFX(X_PF) BFCw(xFlags, F_PF, 1);
+                        } else {
+                            IFX(X_PF) emit_pf(dyn, ninst, vd, x3);
+                        }
                    }
                    break;

@@ -171,9 +176,14 @@ uintptr_t dynarec64_AVX_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, i
                    IFX(X_OF) {
                        IFNATIVE(NF_VF) {} else { BFCw(xFlags, F_OF, 1); }
                    }
-                    if (BOX64ENV(dynarec_test)) {
+                    if (BOX64DRENV(dynarec_safeflags)) {
+                        // those are UD flags
                        IFX(X_AF) BFCw(xFlags, F_AF, 1);
-                        IFX(X_PF) BFCw(xFlags, F_PF, 1);
+                        if(BOX64ENV(cputype)) {
+                            IFX(X_PF) BFCw(xFlags, F_PF, 1);
+                        } else {
+                            IFX(X_PF) emit_pf(dyn, ninst, vd, x3);
+                        }
                    }
                    break;

@@ -184,7 +194,7 @@ uintptr_t dynarec64_AVX_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, i
                    GETVD;
                    IFX(X_CF) {
                        TSTxw_REG(ed, ed);
-                        CSETMw(x3, cEQ);
+                        CSETMw(x3, cNE);
                        BFIw(xFlags, x3, F_CF, 1);
                    }
                    NEGxw_REG(x3, ed);
@@ -211,9 +221,14 @@ uintptr_t dynarec64_AVX_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, i
                    IFX(X_OF) {
                        IFNATIVE(NF_VF) {} else { BFCw(xFlags, F_OF, 1); }
                    }
-                    if (BOX64ENV(dynarec_test)) {
+                    if (BOX64DRENV(dynarec_safeflags)) {
+                        // those are UD flags
                        IFX(X_AF) BFCw(xFlags, F_AF, 1);
-                        IFX(X_PF) BFCw(xFlags, F_PF, 1);
+                        if(BOX64ENV(cputype)) {
+                            IFX(X_PF) BFCw(xFlags, F_PF, 1);
+                        } else {
+                            IFX(X_PF) emit_pf(dyn, ninst, vd, x3);
+                        }
                    }
                    break;

@@ -259,11 +274,18 @@ uintptr_t dynarec64_AVX_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, i
                    BFIw(xFlags, x3, F_SF, 1);
                }
            }
-            IFX(X_AF) BFCw(xFlags, F_AF, 1);
-            IFX(X_PF) BFCw(xFlags, F_PF, 1);
            IFX(X_OF) {
                IFNATIVE(NF_VF) {} else { BFCw(xFlags, F_OF, 1); }
            }
+            if (BOX64DRENV(dynarec_safeflags)) {
+                // those are UD flags
+                IFX(X_AF) BFCw(xFlags, F_AF, 1);
+                if(BOX64ENV(cputype)) {
+                    IFX(X_PF) BFCw(xFlags, F_PF, 1);
+                } else {
+                    IFX(X_PF) emit_pf(dyn, ninst, vd, x3);
+                }
+            }
            break;

        case 0xF7:
@@ -273,22 +295,24 @@ uintptr_t dynarec64_AVX_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, i
            GETGD;
            GETED(0);
            GETVD;
-            MOV64xw(x1, 0);
-            UXTBw(x2, vd);          // start
-            BFXILw(x3, vd, 8, 8);   // length
+            MOV64xw(x5, 0);
+            UXTBw(x4, vd);          // start
+            UBFXw(x3, vd, 8, 8);   // length
            TSTw_REG(x3, x3);
            B_MARK(cEQ);
-            LSRxw_REG(x1, ed, x2);
-            CMPSw_U12(x3, rex.w?64:32);
+            CMPSw_U12(x4, rex.w?64:32);
            B_MARK(cGE);
-            MOV32w(x2, rex.w?64:32);
-            SUBw_REG(x2, x2, x3);
-            LSLxw_REG(x1, x1, x2);
-            LSRxw_REG(x1, x1, x2);
-            TSTxw_REG(x1, x1);
+            LSRxw_REG(x5, ed, x4);
+            CMPSw_U12(x3, rex.w?64:32);
+            B_MARK(cGT);
+            MOV32w(x4, rex.w?64:32);
+            SUBw_REG(x4, x4, x3);
+            LSLxw_REG(x5, x5, x4);
+            LSRxw_REG(x5, x5, x4);
            MARK;
-            MOVxw_REG(gd, x1);
+            MOVxw_REG(gd, x5);
            IFX(X_ZF) {
+                TSTxw_REG(x5, x5);
                IFNATIVE(NF_EQ) {} else {
                    CSETw(x3, cEQ);
                    BFIw(xFlags, x3, F_ZF, 1);
@@ -268,7 +268,9 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
            nextop = F8;
            GETGX(v0, 0);
            GETEXSD(q0, 0, 0);
-            FCMPD(v0, q0);
+            IFX(X_CF|X_PF|X_ZF) {
+                FCMPD(v0, q0);
+            }
            FCOMI(x1, x2);
            break;

@@ -285,8 +287,8 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
                GETEY(q0);
                VMOVQDto(x1, q0, 0);
                VMOVQDto(x2, q0, 1);
-                BFIx(x1, x2, 63, 1);
-                RORx(x2, x1, 63);
+                LSRx(x2, x2, 62);
+                BFXILx(x2, x1, 63, 1);
                BFIx(gd, x2, 2, 2);
            }
            break;
@@ -429,7 +431,7 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
            INST_NAME("VCVTPS2DQ Gx, Ex");
            nextop = F8;
            u8 = sse_setround(dyn, ninst, x1, x2, x6);
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                d0 = fpu_get_scratch(dyn, ninst);
                d1 = fpu_get_scratch(dyn, ninst);
                MOVI_32_lsl(d1, 0x80, 3);
@@ -440,7 +442,7 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
                    VFRINTISQ(v0, v1);
                    VFCVTZSQS(v0, v0);
                } else {
-                    if(arm64_frintts) {
+                    if(cpuext.frintts) {
                         VFRINT32XSQ(v0, v1);
                         VFCVTZSQS(v0, v0);
                    } else {
@@ -493,24 +495,19 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
        case 0x5D:
            INST_NAME("VMINPD Gx, Vx, Ex");
            nextop = F8;
-            if(!BOX64ENV(dynarec_fastnan))
-                q0 = fpu_get_scratch(dyn, ninst);
+            q0 = fpu_get_scratch(dyn, ninst);
            for(int l=0; l<1+vex.l; ++l) {
                if(!l) { GETGX_empty_VXEX(v0, v2, v1, 0); } else { GETGY_empty_VYEY(v0, v2, v1); }
                // FMIN/FMAX wll not copy a NaN if either is NaN
                // but x86 will copy src2 if either value is NaN, so lets force a copy of Src2 (Ex) if result is NaN
-                if(BOX64ENV(dynarec_fastnan)) {
-                    VFMINQD(v0, v2, v1);
-                } else {
-                    VFCMGTQD(q0, v1, v2);   // 0 if NaN or v1 GT v2, so invert mask for copy
-                    if(v0!=v1) VBIFQ(v0, v1, q0);
-                    if(v0!=v2) VBITQ(v0, v2, q0);
-                }
+                VFCMGTQD(q0, v1, v2);   // 0 if NaN or v1 GT v2, so invert mask for copy
+                if(v0!=v1) VBIFQ(v0, v1, q0);
+                if(v0!=v2) VBITQ(v0, v2, q0);
            }
            if(!vex.l) YMM0(gd);
            break;
        case 0x5E:
-            INST_NAME("DIVPD Gx, Ex");
+            INST_NAME("VDIVPD Gx, Ex");
            nextop = F8;
            if(!BOX64ENV(dynarec_fastnan)) {
                q0 = fpu_get_scratch(dyn, ninst);
@@ -536,19 +533,14 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
        case 0x5F:
            INST_NAME("VMAXPD Gx, Vx, Ex");
            nextop = F8;
-            if(!BOX64ENV(dynarec_fastnan))
-                q0 = fpu_get_scratch(dyn, ninst);
+            q0 = fpu_get_scratch(dyn, ninst);
            for(int l=0; l<1+vex.l; ++l) {
                if(!l) { GETGX_empty_VXEX(v0, v2, v1, 0); } else { GETGY_empty_VYEY(v0, v2, v1); }
                // FMIN/FMAX wll not copy a NaN if either is NaN
                // but x86 will copy src2 if either value is NaN, so lets force a copy of Src2 (Ex) if result is NaN
-                if(BOX64ENV(dynarec_fastnan)) {
-                    VFMAXQD(v0, v2, v1);
-                } else {
-                    VFCMGTQD(q0, v2, v1);   // 0 if NaN or v2 GT v1, so invert mask for copy
-                    if(v0!=v1) VBIFQ(v0, v1, q0);
-                    if(v0!=v2) VBITQ(v0, v2, q0);
-                }
+                VFCMGTQD(q0, v2, v1);   // 0 if NaN or v2 GT v1, so invert mask for copy
+                if(v0!=v1) VBIFQ(v0, v1, q0);
+                if(v0!=v2) VBITQ(v0, v2, q0);
            }
            if(!vex.l) YMM0(gd);
            break;
@@ -1459,7 +1451,7 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
                q1 = fpu_get_scratch(dyn, ninst);
                GETEX_Y(q0, 0, 0);
                GETGD;
-                TABLE64(x2, 0x0706050403020100LL);
+                TABLE64_(x2, 0x0706050403020100LL);
                VDUPQD(v0, x2);
                VSHRQ_8(q1, q0, 7);
                USHLQ_8(q1, q1, v0); // shift
@@ -1694,7 +1686,7 @@ uintptr_t dynarec64_AVX_66_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
                    else
                        SQXTN2_32(v0, d0);   // convert int64 -> int32 with saturation in higher part
                } else {
-                    if(arm64_frintts) {
+                    if(cpuext.frintts) {
                        VFRINT32ZDQ(l?d0:v0, v1); // handle overflow
                        VFCVTZSQD(l?d0:v0, l?d0:v0);  // convert double -> int64
                        if(!l)
@@ -22,11 +22,6 @@
 #include "dynarec_arm64_functions.h"
 #include "../dynarec_helper.h"

-static const float addsubps[4] = {-1.f, 1.f, -1.f, 1.f};
-static const double addsubpd[2] = {-1., 1.};
-static const float subaddps[4] = {1.f, -1.f, 1.f, -1.f};
-static const double subaddpd[2] = {1., -1.};
-
 uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, vex_t vex, int* ok, int* need_epilog)
 {
    (void)ip; (void)need_epilog;
@@ -110,7 +105,7 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
            if(!vex.l) YMM0(gd);
            break;
        case 0x04:
-            INST_NAME("PMADDUBSW Gx, Vx, Ex");
+            INST_NAME("VPMADDUBSW Gx, Vx, Ex");
            nextop = F8;
            q0 = fpu_get_scratch(dyn, ninst);
            q1 = fpu_get_scratch(dyn, ninst);
@@ -779,7 +774,7 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
            } else YMM0(gd);
            break;
        case 0x2E:
-            INST_NAME("VMASKMOVPS Ex, Gx, Vx");
+            INST_NAME("VMASKMOVPS Ex, Vx, Gx");
            nextop = F8;
            q0 = fpu_get_scratch(dyn, ninst);
            GETVX(v2, 0);
@@ -835,7 +830,7 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
            SMWRITE2();
            break;
        case 0x2F:
-            INST_NAME("VMASKMOVPD Ex, Gx, Vx");
+            INST_NAME("VMASKMOVPD Ex, Vx, Gx");
            nextop = F8;
            q0 = fpu_get_scratch(dyn, ninst);
            GETVX(v2, 0);
@@ -1390,14 +1385,14 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
                    VMOVQDto(x4, v2, i);
                    CBZw(x4, 4+4*4);
                    SMOVQSto(x4, v1, i+l*2);
-                    ADDx_REG_LSL(x4, ed, x4, wb1);
+                    ADDz_REG_LSL(x4, ed, x4, wb1);
                    VLD1_64(v0, i, x4);
                    VMOVQDfrom(v2, i, xZR);
                } else for(int i=0; i<4; ++i) {
                    VMOVSto(x4, v2, i);
                    CBZw(x4, 4+4*4);
                    SMOVQSto(x4, v1, i);
-                    ADDx_REG_LSL(x4, ed, x4, wb1);
+                    ADDz_REG_LSL(x4, ed, x4, wb1);
                    VLD1_32(v0, i, x4);
                    VMOVQSfrom(v2, i, xZR);
                }
@@ -1445,7 +1440,7 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
                    VMOVSto(x4, v2, i);
                    CBZw(x4, 4+4*4);
                    VMOVQDto(x4, (i&2)?q1:v1, i&1);
-                    ADDx_REG_LSL(x4, ed, x4, wb1);
+                    ADDz_REG_LSL(x4, ed, x4, wb1);
                    VLD1_32(v0, i, x4);
                    VMOVQSfrom(v2, i, xZR);
                }
@@ -1470,7 +1465,7 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
                        VMOVQDto(x4, v2, i);
                        CBZw(x4, 4+4*4);
                        VMOVQDto(x4, v1, i);
-                        ADDx_REG_LSL(x4, ed, x4, wb1);
+                        ADDz_REG_LSL(x4, ed, x4, wb1);
                        VLD1_64(v0, i, x4);
                        VMOVQDfrom(v2, i, xZR);
                    }
@@ -1587,7 +1582,7 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
            INST_NAME("VFMADDSUB213PS/D Gx, Vx, Ex");
            nextop = F8;
            q0 = fpu_get_scratch(dyn, ninst);
-            TABLE64(x2, (rex.w)?((uintptr_t)&addsubpd):((uintptr_t)&addsubps));
+            TABLE64C(x2, rex.w?const_2d_m1_1:const_4f_m1_1_m1_1);
            VLDR128_U12(q0, x2, 0);
            q1 = fpu_get_scratch(dyn, ninst);
            for(int l=0; l<1+vex.l; ++l) {
@@ -1738,7 +1733,7 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
            INST_NAME("VFMADDSUB231PS/D Gx, Vx, Ex");
            nextop = F8;
            q0 = fpu_get_scratch(dyn, ninst);
-            TABLE64(x2, (rex.w)?((uintptr_t)&addsubpd):((uintptr_t)&addsubps));
+            TABLE64C(x2, rex.w?const_2d_m1_1:const_4f_m1_1_m1_1);
            VLDR128_U12(q0, x2, 0);
            for(int l=0; l<1+vex.l; ++l) {
                if(!l) { GETGX_VXEX(v0, v2, v1, 0); if(v0==v2 || v0==v1) q1 = fpu_get_scratch(dyn, ninst); } else { GETGY_VYEY(v0, v2, v1); }
@@ -1761,7 +1756,7 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
            INST_NAME("VFMSUBADD231PS/D Gx, Vx, Ex");
            nextop = F8;
            q0 = fpu_get_scratch(dyn, ninst);
-            TABLE64(x2, (rex.w)?((uintptr_t)&subaddpd):((uintptr_t)&subaddps));
+            TABLE64C(x2, rex.w?const_2d_1_m1:const_4f_1_m1_1_m1);
            VLDR128_U12(q0, x2, 0);
            for(int l=0; l<1+vex.l; ++l) {
                if(!l) { GETGX_VXEX(v0, v2, v1, 0); if(v0==v2 || v0==v1) q1 = fpu_get_scratch(dyn, ninst); } else { GETGY_VYEY(v0, v2, v1); }
@@ -1879,7 +1874,7 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
            INST_NAME("VAESIMC Gx, Ex");
            nextop = F8;
            GETGX_empty_EX(v0, v1, 0);
-            if(arm64_aes) {
+            if(cpuext.aes) {
                AESIMC(v0, v1);
            } else {
                if(v0!=v1) {
@@ -1887,7 +1882,7 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
                }
                sse_forget_reg(dyn, ninst, gd);
                MOV32w(x1, gd);
-                CALL(native_aesimc, -1);
+                CALL(const_native_aesimc, -1);
            }
            YMM0(gd);
            break;
@@ -1895,7 +1890,7 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
        case 0xDC:
            INST_NAME("VAESENC Gx, Vx, Ex");  // AES-NI
            nextop = F8;
-            if(arm64_aes) {
+            if(cpuext.aes) {
                d0 = fpu_get_scratch(dyn, ninst);  // ARM64 internal operation differs a bit from x86_64
                for(int l=0; l<1+vex.l; ++l) {
                    if(!l) {GETGX_empty_VXEX(v0, v2, v1, 0);} else {GETGY_empty_VYEY(v0, v2, v1);}
@@ -1922,8 +1917,8 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
                }
                sse_forget_reg(dyn, ninst, gd);
                MOV32w(x1, gd);
-                CALL(native_aese, -1);
-                if(vex.l) {MOV32w(x1, gd); CALL(native_aese_y, -1);}
+                CALL(const_native_aese, -1);
+                if(vex.l) {MOV32w(x1, gd); CALL(const_native_aese_y, -1);}
                GETGX(q0, 1);
                VEORQ(q0, q0, (d0==-1)?q1:d0);
                if(vex.l) {
@@ -1936,7 +1931,7 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
        case 0xDD:
            INST_NAME("VAESENCLAST Gx, Vx, Ex");  // AES-NI
            nextop = F8;
-            if(arm64_aes) {
+            if(cpuext.aes) {
                d0 = fpu_get_scratch(dyn, ninst);  // ARM64 internal operation differs a bit from x86_64
                for(int l=0; l<1+vex.l; ++l) {
                    if(!l) { GETGX_empty_VXEX(v0, v2, v1, 0); } else { GETGY_empty_VYEY(v0, v2, v1); }
@@ -1962,8 +1957,8 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
                }
                sse_forget_reg(dyn, ninst, gd);
                MOV32w(x1, gd);
-                CALL(native_aeselast, -1);
-                if(vex.l) {MOV32w(x1, gd); CALL(native_aeselast_y, -1);}
+                CALL(const_native_aeselast, -1);
+                if(vex.l) {MOV32w(x1, gd); CALL(const_native_aeselast_y, -1);}
                GETGX(q0, 1);
                VEORQ(q0, q0, (d0==-1)?q1:d0);
                if(vex.l) {
@@ -1976,7 +1971,7 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
        case 0xDE:
            INST_NAME("VAESDEC Gx, Vx, Ex");  // AES-NI
            nextop = F8;
-            if(arm64_aes) {
+            if(cpuext.aes) {
                d0 = fpu_get_scratch(dyn, ninst);  // ARM64 internal operation differs a bit from x86_64
                for(int l=0; l<1+vex.l; ++l) {
                    if(!l) {GETGX_empty_VXEX(v0, v2, v1, 0);} else {GETGY_empty_VYEY(v0, v2, v1);}
@@ -2003,8 +1998,8 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
                }
                sse_forget_reg(dyn, ninst, gd);
                MOV32w(x1, gd);
-                CALL(native_aesd, -1);
-                if(vex.l) {MOV32w(x1, gd); CALL(native_aesd_y, -1);}
+                CALL(const_native_aesd, -1);
+                if(vex.l) {MOV32w(x1, gd); CALL(const_native_aesd_y, -1);}
                GETGX(q0, 1);
                VEORQ(q0, q0, (d0==-1)?q1:d0);
                if(vex.l) {
@@ -2017,7 +2012,7 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
        case 0xDF:
            INST_NAME("VAESDECLAST Gx, Vx, Ex");  // AES-NI
            nextop = F8;
-            if(arm64_aes) {
+            if(cpuext.aes) {
                d0 = fpu_get_scratch(dyn, ninst);  // ARM64 internal operation differs a bit from x86_64
                for(int l=0; l<1+vex.l; ++l) {
                    if(!l) {GETGX_empty_VXEX(v0, v2, v1, 0);} else {GETGY_empty_VYEY(v0, v2, v1);}
@@ -2043,8 +2038,8 @@ uintptr_t dynarec64_AVX_66_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
                }
                sse_forget_reg(dyn, ninst, gd);
                MOV32w(x1, gd);
-                CALL(native_aesdlast, -1);
-                if(vex.l) {MOV32w(x1, gd); CALL(native_aesdlast_y, -1);}
+                CALL(const_native_aesdlast, -1);
+                if(vex.l) {MOV32w(x1, gd); CALL(const_native_aesdlast_y, -1);}
                GETGX(q0, 1);
                VEORQ(q0, q0, (d0==-1)?q1:d0);
                if(vex.l) {
@@ -167,12 +167,12 @@ uintptr_t dynarec64_AVX_66_0F3A(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
            u8 = F8;
            GETVX(v2, 0);
            GETGX_empty(v0);
-            if((v0==v2) && ((u8&0xf0)==0)) {
+            if((v0==v2) && ((u8&0b10110000)==0)) {
                q0 = fpu_get_scratch(dyn, ninst);
                VMOVQ(q0, v2);
                v2 = q0;
            }
-            if((v0==v1) && ((u8&0xf0)==0x20)) {
+            if((v0==v1) && ((u8&0b10110000)==0x20)) {
                q1 = fpu_get_scratch(dyn, ninst);
                VMOVQ(q1, v1);
                v1 = q1;
@@ -184,8 +184,8 @@ uintptr_t dynarec64_AVX_66_0F3A(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
                case 3: if(MODREG) { d1 = ymm_get_reg(dyn, ninst, x1, s0, 0, gd, vex.v, -1); VMOVQ(v0, d1); } else { VLDR128_U12(v0, ed, fixedaddress+16); } break;
                default: VEORQ(v0, v0, v0);
            }
-            if((u8&0xf0)==0x10) { if((u8&0x0f)!=1) d2 = ymm_get_reg(dyn, ninst, x1, vex.v, 0, gd, s0, -1); }
-            if(MODREG && ((u8&0xf0)==0x30)) { if((u8&0x0f)!=3) d1 = ymm_get_reg(dyn, ninst, x1, s0, 0, gd, vex.v, -1); }
+            if((u8&0b10110000)==0x10) { if((u8&0b1011)!=1) d2 = ymm_get_reg(dyn, ninst, x1, vex.v, 0, gd, s0, -1); }
+            if(MODREG && ((u8&0b10110000)==0x30)) { if((u8&0b1011)!=3) d1 = ymm_get_reg(dyn, ninst, x1, s0, 0, gd, vex.v, -1); }
            v0 = ymm_get_reg_empty(dyn, ninst, x1, gd, vex.v, s0, -1);
            switch((u8>>4)&0b1011) {
                case 0: VMOVQ(v0, v2); break;
@@ -378,7 +378,6 @@ uintptr_t dynarec64_AVX_66_0F3A(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
            if(MODREG) {
                ed = TO_NAT((nextop & 7) + (rex.b << 3));
            } else {
-                SMREAD();
                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff, 0, rex, NULL, 0, 1);
                ed = x1;
            }
@@ -396,7 +395,6 @@ uintptr_t dynarec64_AVX_66_0F3A(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
            if(MODREG) {
                ed = TO_NAT((nextop & 7) + (rex.b << 3));
            } else {
-                SMREAD();
                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 1);
                ed = x1;
            }
@@ -414,7 +412,6 @@ uintptr_t dynarec64_AVX_66_0F3A(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
            if(MODREG) {
                ed = TO_NAT((nextop & 7) + (rex.b << 3));
            } else {
-                SMREAD();
                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<(2+rex.w), (1<<(2+rex.w))-1, rex, NULL, 0, 1);
                ed = x1;
            }
@@ -436,7 +433,6 @@ uintptr_t dynarec64_AVX_66_0F3A(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
            if(MODREG) {
                ed = TO_NAT((nextop & 7) + (rex.b << 3));
            } else {
-                SMREAD();
                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 1);
                ed = x1;
            }
@@ -497,7 +493,7 @@ uintptr_t dynarec64_AVX_66_0F3A(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
            break;

        case 0x1D:
-            INST_NAME("VCVTPS2PH Ex, Gx");
+            INST_NAME("VCVTPS2PH Ex, Gx, Ib");
            nextop = F8;
            GETGX(v0, 0);
            if(MODREG) {
@@ -544,13 +540,20 @@ uintptr_t dynarec64_AVX_66_0F3A(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
            break;

        case 0x20:
-            INST_NAME("VINSERTD Gx, Vx, Ex, Ib");
+            INST_NAME("VPINSRB Gx, Vx, ED, Ib");
            nextop = F8;
            GETGX_empty_VX(v0, v2);
-            GETED(1);
-            u8 = F8;
            if(v0!=v2) VMOVQ(v0, v2);
-            VMOVQBfrom(v0, u8&0xf, ed);
+            if(MODREG) {
+                u8 = (F8)&15;
+                ed = TO_NAT((nextop & 7) + (rex.b << 3));
+                VMOVQBfrom(v0, u8, ed);
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 1);
+                u8 = (F8)&15;
+                VLD1_8(v0, u8, wback);
+            }
            YMM0(gd);
            break;
        case 0x21:
@@ -586,7 +589,11 @@ uintptr_t dynarec64_AVX_66_0F3A(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
            YMM0(gd);
            break;
        case 0x22:
-            INST_NAME("VINSERTD Gx, Vx, Ex, Ib");
+            if (rex.w) {
+                INST_NAME("VPINSRQ Gx, Vx, ED, Ib");
+            } else {
+                INST_NAME("VPINSRD Gx, Vx, ED, Ib");
+            }
            nextop = F8;
            GETGX_empty_VX(v0, v2);
            GETED(1);
@@ -662,7 +669,7 @@ uintptr_t dynarec64_AVX_66_0F3A(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
        case 0x44:
            INST_NAME("VPCLMULQDQ Gx, Vx, Ex, Ib");
            nextop = F8;
-            if(arm64_pmull) {
+            if(cpuext.pmull) {
                d0 = fpu_get_scratch(dyn, ninst);
                for(int l=0; l<1+vex.l; ++l) {
                    if(!l) {
@@ -706,7 +713,7 @@ uintptr_t dynarec64_AVX_66_0F3A(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
                }
                u8 = F8;
                MOV32w(x4, u8);
-                CALL_(vex.l?native_pclmul_y:native_pclmul_x, -1, x3);
+                CALL_(vex.l?const_native_pclmul_y:const_native_pclmul_x, -1, x3);
            }
            if(!vex.l) YMM0(gd);
            break;
@@ -810,7 +817,7 @@ uintptr_t dynarec64_AVX_66_0F3A(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
            }
            u8 = F8;
            MOV32w(x4, u8);
-            CALL(native_aeskeygenassist, -1);
+            CALL(const_native_aeskeygenassist, -1);
            YMM0(gd);
            break;

@@ -144,12 +144,12 @@ uintptr_t dynarec64_AVX_F2_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
            nextop = F8;
            GETGD;
            GETEXSD(q0, 0, 0);
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                MRS_fpsr(x5);
                BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
                MSR_fpsr(x5);
            }
-            if(!BOX64ENV(dynarec_fastround) && arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && cpuext.frintts) {
                v0 = fpu_get_scratch(dyn, ninst);
                if(rex.w) {
                    FRINT64ZD(v0, q0);
@@ -160,7 +160,7 @@ uintptr_t dynarec64_AVX_F2_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
            } else {
                FCVTZSxwD(gd, q0);
            }
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                MRS_fpsr(x5);   // get back FPSR to check the IOC bit
                TBZ_NEXT(x5, FPSR_IOC);
                if(rex.w) {
@@ -175,14 +175,14 @@ uintptr_t dynarec64_AVX_F2_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
            nextop = F8;
            GETGD;
            GETEXSD(q0, 0, 0);
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                MRS_fpsr(x5);
                BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
                MSR_fpsr(x5);
            }
            u8 = sse_setround(dyn, ninst, x1, x2, x3);
            d1 = fpu_get_scratch(dyn, ninst);
-            if(!BOX64ENV(dynarec_fastround) && arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && cpuext.frintts) {
                if(rex.w) {
                    FRINT64XD(d1, q0);
                } else {
@@ -193,7 +193,7 @@ uintptr_t dynarec64_AVX_F2_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
            }
            x87_restoreround(dyn, ninst, u8);
            FCVTZSxwD(gd, d1);
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                MRS_fpsr(x5);   // get back FPSR to check the IOC bit
                TBZ_NEXT(x5, FPSR_IOC);
                if(rex.w) {
@@ -496,9 +496,9 @@ uintptr_t dynarec64_AVX_F2_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
                case 0x09: CSETMx(x2, cHI); break;  // Less than or unordered
                case 0x0a: CSETMx(x2, cCS); break;  // Less or equal or unordered
                case 0x0b: MOV32w(x2, 0); break;    // false
-                case 0x0c: CSETMw(x2, cNE); CSETMx(x3, cVS); BICx(x2, x2, x3); break;  // Not Equal not unordered
-                case 0x0d: CSETMw(x2, cLS); break;  // Greater or equal not unordered
-                case 0x0e: CSETMw(x2, cCC); break;  // Greater not unordered
+                case 0x0c: CSETMx(x2, cNE); CSETMx(x3, cVS); BICx(x2, x2, x3); break;  // Not Equal not unordered
+                case 0x0d: CSETMx(x2, cLS); break;  // Greater or equal not unordered
+                case 0x0e: CSETMx(x2, cCC); break;  // Greater not unordered
                case 0x0f: MOV64x(x2, 0xffffffffffffffffLL); break; // true
            }
            VMOVQDfrom(v0, 0, x2);
@@ -509,9 +509,7 @@ uintptr_t dynarec64_AVX_F2_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
            INST_NAME("VADDSUBPS Gx, Vx, Ex");
            nextop = F8;
            q0 = fpu_get_scratch(dyn, ninst);
-            static float addsubps[4] = {-1.f, 1.f, -1.f, 1.f};
-            MAYUSE(addsubps);
-            MOV64x(x2, (uintptr_t)&addsubps);
+            TABLE64C(x2, const_4f_m1_1_m1_1);
            VLDR128_U12(q0, x2, 0);
            for(int l=0; l<1+vex.l; ++l) {
                if(!l) { GETGX_empty_VXEX(v0, v2, v1, 0); } else { GETGY_empty_VYEY(v0, v2, v1); }
@@ -550,7 +548,7 @@ uintptr_t dynarec64_AVX_F2_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
                    else
                        SQXTN2_32(v0, d0);   // convert int64 -> int32 with saturation in higher part
                } else {
-                    if(arm64_frintts) {
+                    if(cpuext.frintts) {
                        VFRINT32XDQ(l?d0:v0, v1);    // round, handling of overflow and Nan to 0x80000000
                        VFCVTNSQD(l?d0:v0, l?d0:v0);  // convert double -> int64
                        if(!l)
@@ -598,11 +596,16 @@ uintptr_t dynarec64_AVX_F2_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
                    VMOVQ(v0, v1);
                }
            } else {
-                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
-                v1 = ymm_get_reg_empty(dyn, ninst, x1, gd, -1, -1, -1);
                SMREAD();
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, NULL, 0x3f<<4, 15, rex, NULL, 1, 0);
-                VLDP128_I7(v0, v1, ed, fixedaddress);
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                if (vex.l) {
+                    v1 = ymm_get_reg_empty(dyn, ninst, x1, gd, -1, -1, -1);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, NULL, 0x3f << 4, 15, rex, NULL, 1, 0);
+                    VLDP128_I7(v0, v1, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff << 4, 15, rex, NULL, 0, 0);
+                    VLD128(v0, ed, fixedaddress);
+                }
            }
            if(!vex.l) YMM0(gd);
            break;
@@ -81,7 +81,7 @@ uintptr_t dynarec64_AVX_F3_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
            nextop = F8;
            GETG;
            if(MODREG) {
-                GETGXVXEX_empty(v0, v2, v1, 0);
+                GETGXVXEX_empty(v0, v1, v2, 0);
                if((v2!=v1) && (v2!=v0)) VMOVQ(v2, v1);
                if(v2!=v0) VMOVeS(v2, 0, v0, 0);
                if((v2!=v1) && (v2==v0)) { VMOVeS(v2, 1, v0, 1); VMOVeD(v2, 1, v0, 1);}
@@ -144,12 +144,12 @@ uintptr_t dynarec64_AVX_F3_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
            nextop = F8;
            GETGD;
            GETEXSS(d0, 0, 0);
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                MRS_fpsr(x5);
                BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
                MSR_fpsr(x5);
            }
-            if(!BOX64ENV(dynarec_fastround) && arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && cpuext.frintts) {
                v0 = fpu_get_scratch(dyn, ninst);
                if(rex.w) {
                    FRINT64ZS(v0, d0);
@@ -160,7 +160,7 @@ uintptr_t dynarec64_AVX_F3_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
            } else {
                FCVTZSxwS(gd, d0);
            }
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                MRS_fpsr(x5);   // get back FPSR to check the IOC bit
                TBZ_NEXT(x5, FPSR_IOC);
                if(rex.w) {
@@ -175,14 +175,14 @@ uintptr_t dynarec64_AVX_F3_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
            nextop = F8;
            GETGD;
            GETEXSS(q0, 0, 0);
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                MRS_fpsr(x5);
                BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
                MSR_fpsr(x5);
            }
            u8 = sse_setround(dyn, ninst, x1, x2, x3);
            d1 = fpu_get_scratch(dyn, ninst);
-            if(!BOX64ENV(dynarec_fastround) && arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && cpuext.frintts) {
                if(rex.w) {
                    FRINT64XS(d1, q0);
                } else {
@@ -193,7 +193,7 @@ uintptr_t dynarec64_AVX_F3_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
            }
            x87_restoreround(dyn, ninst, u8);
            FCVTZSxwS(gd, d1);
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                MRS_fpsr(x5);   // get back FPSR to check the IOC bit
                TBZ_NEXT(x5, FPSR_IOC);
                if(rex.w) {
@@ -342,7 +342,7 @@ uintptr_t dynarec64_AVX_F3_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip,
                if(BOX64ENV(dynarec_fastround)) {
                    VFCVTZSQS(v0, v1);
                } else {
-                    if(arm64_frintts) {
+                    if(cpuext.frintts) {
                        VFRINT32ZSQ(v0, v1);
                        VFCVTZSQS(v0, v0);
                    } else {
@@ -0,0 +1,144 @@
+#include <stdint.h>
+#include <string.h>
+
+#include "dynarec_arm64_consts.h"
+#include "debug.h"
+#include "box64context.h"
+#include "box64cpu.h"
+#include "emu/x64emu_private.h"
+#include "x64emu.h"
+#include "box64stack.h"
+#include "callback.h"
+#include "emu/x64run_private.h"
+#include "emu/x87emu_private.h"
+#include "emu/x64primop.h"
+#include "my_cpuid.h"
+#include "freq.h"
+#include "debug.h"
+#include "custommem.h"
+#include "dynarec_arm64_functions.h"
+#include "emu/x64shaext.h"
+#include "emu/x87emu_private.h"
+#include "emu/x64compstrings.h"
+#include "x64test.h"
+#include "dynarec/dynarec_next.h"
+#include "random.h"
+
+static const int8_t mask_shift8[] = { -7, -6, -5, -4, -3, -2, -1, 0 };
+static const int8_t mask_string8[] = { 7, 6, 5, 4, 3, 2, 1, 0 };
+static const int8_t mask_string16[] = { 15, 14, 13, 12, 11, 10, 9, 8 };
+static const float addsubps[4] = {-1.f, 1.f, -1.f, 1.f};
+static const double addsubpd[2] = {-1., 1.};
+static const float subaddps[4] = {1.f, -1.f, 1.f, -1.f};
+static const double subaddpd[2] = {1., -1.};
+
+#ifndef HAVE_TRACE
+void PrintTrace() {}
+#endif
+
+uintptr_t getConst(arm64_consts_t which)
+{
+    switch(which) {
+        case const_none: dynarec_log(LOG_NONE, "Warning, const none used\n");
+            return 0;
+        case const_daa8: return (uintptr_t)daa8;
+        case const_das8: return (uintptr_t)das8;
+        case const_aaa16: return (uintptr_t)aaa16;
+        case const_aas16: return (uintptr_t)aas16;
+        case const_aam16: return (uintptr_t)aam16;
+        case const_aad16: return (uintptr_t)aad16;
+        case const_native_br: return (uintptr_t)native_br;
+        case const_native_ud: return (uintptr_t)native_ud;
+        case const_native_priv: return (uintptr_t)native_priv;
+        case const_native_gpf: return (uintptr_t)native_gpf;
+        case const_native_int3: return (uintptr_t)native_int3;
+        case const_native_int: return (uintptr_t)native_int;
+        case const_native_div0: return (uintptr_t)native_div0;
+        case const_native_frstor16: return (uintptr_t)native_frstor16;
+        case const_native_fsave16: return (uintptr_t)native_fsave16;
+        case const_native_fsave: return (uintptr_t)native_fsave;
+        case const_native_aesimc: return (uintptr_t)native_aesimc;
+        case const_native_aesd: return (uintptr_t)native_aesd;
+        case const_native_aesd_y: return (uintptr_t)native_aesd_y;
+        case const_native_aesdlast: return (uintptr_t)native_aesdlast;
+        case const_native_aesdlast_y: return (uintptr_t)native_aesdlast_y;
+        case const_native_aese: return (uintptr_t)native_aese;
+        case const_native_aese_y: return (uintptr_t)native_aese_y;
+        case const_native_aeselast: return (uintptr_t)native_aeselast;
+        case const_native_aeselast_y: return (uintptr_t)native_aeselast_y;
+        case const_native_aeskeygenassist: return (uintptr_t)native_aeskeygenassist;
+        case const_native_pclmul: return (uintptr_t)native_pclmul;
+        case const_native_pclmul_x: return (uintptr_t)native_pclmul_x;
+        case const_native_pclmul_y: return (uintptr_t)native_pclmul_y;
+        case const_direct_f2xm1: return (uintptr_t)direct_f2xm1;
+        case const_direct_fyl2x: return (uintptr_t)direct_fyl2x;
+        case const_direct_fyl2xp1: return (uintptr_t)direct_fyl2xp1;
+        case const_native_fxtract: return (uintptr_t)native_fxtract;
+        case const_direct_ftan: return (uintptr_t)direct_ftan;
+        case const_direct_fpatan: return (uintptr_t)direct_fpatan;
+        case const_direct_fcos: return (uintptr_t)direct_fcos;
+        case const_direct_fsin: return (uintptr_t)direct_fsin;
+        case const_native_fsincos: return (uintptr_t)native_fsincos;
+        case const_direct_fscale: return (uintptr_t)direct_fscale;
+        case const_native_fld: return (uintptr_t)native_fld;
+        case const_native_fstp: return (uintptr_t)native_fstp;
+        case const_native_frstor: return (uintptr_t)native_frstor;
+        case const_native_next: return (uintptr_t)native_next;
+        case const_int3: return (uintptr_t)EmuInt3;
+        case const_x86syscall: return (uintptr_t)EmuX86Syscall;
+        case const_x64syscall: return (uintptr_t)EmuX64Syscall;
+        case const_x64syscall_linux: return (uintptr_t)EmuX64Syscall_linux;
+        case const_rcl16: return (uintptr_t)rcl16;
+        case const_rcl32: return (uintptr_t)rcl32;
+        case const_rcl64: return (uintptr_t)rcl64;
+        case const_rcr16: return (uintptr_t)rcr16;
+        case const_rcr32: return (uintptr_t)rcr32;
+        case const_rcr64: return (uintptr_t)rcr64;
+        case const_div64: return (uintptr_t)div64;
+        case const_idiv64: return (uintptr_t)idiv64;
+        case const_random32: return (uintptr_t)get_random32;
+        case const_random64: return (uintptr_t)get_random64;
+        case const_readtsc: return (uintptr_t)ReadTSC;
+        case const_helper_getcpu: return (uintptr_t)helper_getcpu;
+        case const_cpuid: return (uintptr_t)my_cpuid;
+        case const_getsegmentbase: return (uintptr_t)GetSegmentBaseEmu;
+        case const_updateflags_arm64: return (uintptr_t)create_updateflags();
+        case const_reset_fpu: return (uintptr_t)reset_fpu;
+        case const_sha1msg2: return (uintptr_t)sha1msg2;
+        case const_sha1rnds4: return (uintptr_t)sha1rnds4;
+        case const_sha256msg1: return (uintptr_t)sha256msg1;
+        case const_sha256msg2: return (uintptr_t)sha256msg2;
+        case const_sha256rnds2: return (uintptr_t)sha256rnds2;
+        case const_fpu_loadenv: return (uintptr_t)fpu_loadenv;
+        case const_fpu_savenv: return (uintptr_t)fpu_savenv;
+        case const_fpu_fxsave32: return (uintptr_t)fpu_fxsave32;
+        case const_fpu_fxsave64: return (uintptr_t)fpu_fxsave64;
+        case const_fpu_fxrstor32: return (uintptr_t)fpu_fxrstor32;
+        case const_fpu_fxrstor64: return (uintptr_t)fpu_fxrstor64;
+        case const_fpu_xsave: return (uintptr_t)fpu_xsave;
+        case const_fpu_xrstor: return (uintptr_t)fpu_xrstor;
+        case const_fpu_fbld: return (uintptr_t)fpu_fbld;
+        case const_fpu_fbst: return (uintptr_t)fpu_fbst;
+        case const_sse42_compare_string_explicit_len: return (uintptr_t)sse42_compare_string_explicit_len;
+        case const_sse42_compare_string_implicit_len: return (uintptr_t)sse42_compare_string_implicit_len;
+        case const_x64test_step: return (uintptr_t)x64test_step;
+        case const_printtrace: return (uintptr_t)PrintTrace;
+        case const_epilog: return (uintptr_t)native_epilog;
+        case const_jmptbl32: return getJumpTable32();
+        case const_jmptbl48: return getJumpTable48();
+        case const_jmptbl64: return getJumpTable64();
+        case const_context: return (uintptr_t)my_context;
+        case const_8b_m7_m6_m5_m4_m3_m2_m1_0: return (uintptr_t)&mask_shift8;
+        case const_8b_7_6_5_4_3_2_1_0: return (uintptr_t)&mask_string8;
+        case const_8b_15_14_13_12_11_10_9_8: return (uintptr_t)&mask_string16;
+        case const_4f_m1_1_m1_1: return (uintptr_t)&addsubps;
+        case const_4f_1_m1_1_m1: return (uintptr_t)&subaddps;
+        case const_2d_m1_1: return (uintptr_t)&addsubpd;
+        case const_2d_1_m1: return (uintptr_t)&subaddpd;
+
+        case const_last: dynarec_log(LOG_NONE, "Warning, const last used\n");
+            return 0;
+    }
+    dynarec_log(LOG_NONE, "Warning, Unknown const %d used\n", which);
+    return 0;
+}
@@ -0,0 +1,110 @@
+#ifndef __DYNAREC_ARM64_CONSTS__
+#define __DYNAREC_ARM64_CONSTS__
+#include <stdint.h>
+
+typedef enum arm64_consts_s {
+    const_none,
+    const_daa8,
+    const_das8,
+    const_aaa16,
+    const_aas16,
+    const_aam16,
+    const_aad16,
+    const_native_br,
+    const_native_ud,
+    const_native_priv,
+    const_native_gpf,
+    const_native_int3,
+    const_native_int,
+    const_native_div0,
+    const_native_frstor16,
+    const_native_fsave16,
+    const_native_fsave,
+    const_native_aesimc,
+    const_native_aesd,
+    const_native_aesd_y,
+    const_native_aesdlast,
+    const_native_aesdlast_y,
+    const_native_aese,
+    const_native_aese_y,
+    const_native_aeselast,
+    const_native_aeselast_y,
+    const_native_aeskeygenassist,
+    const_native_pclmul,
+    const_native_pclmul_x,
+    const_native_pclmul_y,
+    const_direct_f2xm1,
+    const_direct_fyl2x,
+    const_direct_fyl2xp1,
+    const_native_fxtract,
+    const_direct_ftan,
+    const_direct_fpatan,
+    const_direct_fcos,
+    const_direct_fsin,
+    const_native_fsincos,
+    const_direct_fscale,
+    const_native_fld,
+    const_native_fstp,
+    const_native_frstor,
+    const_native_next,
+    const_int3,
+    const_x86syscall,
+    const_x64syscall,
+    const_x64syscall_linux,
+    const_rcl16,
+    const_rcl32,
+    const_rcl64,
+    const_rcr16,
+    const_rcr32,
+    const_rcr64,
+    const_div64,
+    const_idiv64,
+    const_random32,
+    const_random64,
+    const_readtsc,
+    const_helper_getcpu,
+    const_cpuid,
+    const_getsegmentbase,
+    const_updateflags_arm64,
+    const_reset_fpu,
+    const_sha1msg2,
+    const_sha1rnds4,
+    const_sha256msg1,
+    const_sha256msg2,
+    const_sha256rnds2,
+    const_fpu_loadenv,
+    const_fpu_savenv,
+    const_fpu_fxsave32,
+    const_fpu_fxsave64,
+    const_fpu_fxrstor32,
+    const_fpu_fxrstor64,
+    const_fpu_xsave,
+    const_fpu_xrstor,
+    const_fpu_fbld,
+    const_fpu_fbst,
+    const_sse42_compare_string_explicit_len,
+    const_sse42_compare_string_implicit_len,
+    const_x64test_step,
+    const_printtrace,
+    const_epilog,
+    const_jmptbl32,
+    const_jmptbl48,
+    const_jmptbl64,
+    const_context,
+    const_8b_m7_m6_m5_m4_m3_m2_m1_0,
+    const_8b_7_6_5_4_3_2_1_0,
+    const_8b_15_14_13_12_11_10_9_8,
+    const_4f_m1_1_m1_1,
+    const_2d_m1_1,
+    const_4f_1_m1_1_m1,
+    const_2d_1_m1,
+
+    const_last
+} arm64_consts_t;
+
+uintptr_t getConst(arm64_consts_t which);
+
+// temporary define...
+#define const_PrintTrace const_printtrace
+
+#endif //__DYNAREC_ARM64_CONSTS__
@@ -21,9 +21,9 @@
 #include "dynarec_arm64_functions.h"


-uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
 {
-    (void)ip; (void)rep; (void)need_epilog;
+    (void)ip; (void)need_epilog;

    uint8_t nextop = F8;
    uint8_t ed;
@@ -21,9 +21,9 @@
 #include "dynarec_arm64_functions.h"


-uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
 {
-    (void)ip; (void)rep; (void)need_epilog;
+    (void)ip; (void)need_epilog;

    uint8_t nextop = F8;
    uint8_t ed;
@@ -269,57 +269,54 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
        case 0xF0:
            INST_NAME("F2XM1");
            MESSAGE(LOG_DUMP, "Need Optimization (F2XM1)\n");
-            i1 = x87_stackcount(dyn, ninst, x1);
-            x87_forget(dyn, ninst, x1, x2, 0);
-            CALL(native_f2xm1, -1);
-            x87_unstackcount(dyn, ninst, x1, i1);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            CALL_D(const_direct_f2xm1, v1, v1, -1, -1, -1);
            break;
        case 0xF1:
            INST_NAME("FYL2X");
            MESSAGE(LOG_DUMP, "Need Optimization (FYL2X)\n");
-            i1 = x87_stackcount(dyn, ninst, x1);
-            x87_forget(dyn, ninst, x1, x2, 0);
-            x87_forget(dyn, ninst, x1, x2, 1);
-            CALL(native_fyl2x, -1);
-            x87_unstackcount(dyn, ninst, x1, i1);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            v2 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
+            CALL_D(const_direct_fyl2x, v2, v1, v2, -1, -1);
            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xF2:
            INST_NAME("FPTAN");
            MESSAGE(LOG_DUMP, "Need Optimization (FPTAN)\n");
-            i1 = x87_stackcount(dyn, ninst, x1);
-            x87_forget(dyn, ninst, x1, x2, 0);
-            if(!BOX64ENV(dynarec_fastround))
-                u8 = x87_setround(dyn, ninst, x1, x2, x4);
-            CALL_(native_ftan, -1, BOX64ENV(dynarec_fastround) ? 0 : u8);
-            x87_unstackcount(dyn, ninst, x1, i1);
-           if(!BOX64ENV(dynarec_fastround))
-                x87_restoreround(dyn, ninst, u8);
            if(PK(0)==0xdd && PK(1)==0xd8) {
                MESSAGE(LOG_DUMP, "Optimized next DD D8 fstp st0, st0, not emitting 1\n");
                u8 = F8;
                u8 = F8;
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
            } else {
-                X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_F);
+                X87_PUSH_OR_FAIL(v2, dyn, ninst, x1, NEON_CACHE_ST_F);
                if(ST_IS_F(0)) {
-                    FMOVS_8(v1, 0b01110000);
+                    FMOVS_8(v2, 0b01110000);
                } else {
-                    FMOVD_8(v1, 0b01110000);
+                    FMOVD_8(v2, 0b01110000);
                }
+                v1 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
            }
+            if(!BOX64ENV(dynarec_fastround))
+                u8 = x87_setround(dyn, ninst, x1, x2, x4);
+            else
+                u8 = 0;
+            CALL_D(const_direct_ftan, v1, v1, -1, u8, -1);
+            if(!BOX64ENV(dynarec_fastround))
+                x87_restoreround(dyn, ninst, u8);
            break;
        case 0xF3:
            INST_NAME("FPATAN");
            MESSAGE(LOG_DUMP, "Need Optimization (FPATAN)\n");
-            i1 = x87_stackcount(dyn, ninst, x1);
-            x87_forget(dyn, ninst, x1, x2, 0);
-            x87_forget(dyn, ninst, x1, x2, 1);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            v2 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
            if(!BOX64ENV(dynarec_fastround))
                u8 = x87_setround(dyn, ninst, x1, x2, x4);
-            CALL_(native_fpatan, -1, BOX64ENV(dynarec_fastround) ? 0 : u8);
+            else
+                u8 = 0;
+            CALL_D(const_direct_fpatan, v2, v1, v2, u8, -1);
            if(!BOX64ENV(dynarec_fastround))
                x87_restoreround(dyn, ninst, u8);
-            x87_unstackcount(dyn, ninst, x1, i1);
            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xF4:
@@ -328,19 +325,11 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, 0);
            i1 = x87_stackcount(dyn, ninst, x1);
            x87_forget(dyn, ninst, x1, x2, 1);
-            CALL(native_fxtract, -1);
+            CALL(const_native_fxtract, -1);
            x87_unstackcount(dyn, ninst, x1, i1);
            break;
        case 0xF5:
            INST_NAME("FPREM1");
-            #if 0
-            MESSAGE(LOG_DUMP, "Need Optimization\n");
-            i1 = x87_stackcount(dyn, ninst, x1);
-            x87_forget(dyn, ninst, x1, x2, 0);
-            x87_forget(dyn, ninst, x1, x2, 1);
-            CALL(native_fprem1, -1);
-            x87_unstackcount(dyn, ninst, x1, i1);
-            #else
            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
            v2 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
            s0 = fpu_get_scratch(dyn, ninst);
@@ -361,11 +350,10 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            LSRx(x4, x4, 1);
            BFIw(x1, x4, 8, 1);
            STRw_U12(x1, xEmu, offsetof(x64emu_t, sw));
-            #endif
            break;
        case 0xF6:
            INST_NAME("FDECSTP");
-            fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+            BARRIER(BARRIER_FLOAT);
            LDRw_U12(x2, xEmu, offsetof(x64emu_t, top));
            SUBw_U12(x2, x2, 1);
            ANDw_mask(x2, x2, 0, 2);    //mask=7
@@ -373,7 +361,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            break;
        case 0xF7:
            INST_NAME("FINCSTP");
-            fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+            BARRIER(BARRIER_FLOAT);
            LDRw_U12(x2, xEmu, offsetof(x64emu_t, top));
            ADDw_U12(x2, x2, 1);
            ANDw_mask(x2, x2, 0, 2);    //mask=7
@@ -381,14 +369,6 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            break;
        case 0xF8:
            INST_NAME("FPREM");
-            #if 0
-            MESSAGE(LOG_DUMP, "Need Optimization\n");
-            i1 = x87_stackcount(dyn, ninst, x1);
-            x87_forget(dyn, ninst, x1, x2, 0);
-            x87_forget(dyn, ninst, x1, x2, 1);
-            CALL(native_fprem, -1);
-            x87_unstackcount(dyn, ninst, x1, i1);
-            #else
            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
            v2 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
            s0 = fpu_get_scratch(dyn, ninst);
@@ -409,16 +389,13 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            LSRx(x4, x4, 1);
            BFIw(x1, x4, 8, 1);
            STRw_U12(x1, xEmu, offsetof(x64emu_t, sw));
-            #endif
            break;
        case 0xF9:
            INST_NAME("FYL2XP1");
            MESSAGE(LOG_DUMP, "Need Optimization (FYL2XP1)\n");
-            i1 = x87_stackcount(dyn, ninst, x1);
-            x87_forget(dyn, ninst, x1, x2, 0);
-            x87_forget(dyn, ninst, x1, x2, 1);
-            CALL(native_fyl2xp1, -1);
-            x87_unstackcount(dyn, ninst, x1, i1);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            v2 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
+            CALL_D(const_direct_fyl2xp1, v2, v1, v2, -1, -1);
            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xFA:
@@ -443,19 +420,13 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            x87_forget(dyn, ninst, x1, x2, 1);
            if(!BOX64ENV(dynarec_fastround))
                u8 = x87_setround(dyn, ninst, x1, x2, x4);
-            CALL_(native_fsincos, -1, BOX64ENV(dynarec_fastround) ? 0 : u8);
+            CALL_(const_native_fsincos, -1, BOX64ENV(dynarec_fastround) ? 0 : u8);
            if(!BOX64ENV(dynarec_fastround))
                x87_restoreround(dyn, ninst, u8);
            x87_unstackcount(dyn, ninst, x1, i1);
            break;
        case 0xFC:
            INST_NAME("FRNDINT");
-            #if 0
-            MESSAGE(LOG_DUMP, "Need Optimization\n");
-            // use C helper for now, nothing staightforward is available
-            x87_forget(dyn, ninst, x1, x2, 0);
-            CALL(native_frndint, -1);
-            #else
            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
            u8 = x87_setround(dyn, ninst, x1, x2, x3);
            if(ST_IS_F(0)) {
@@ -464,44 +435,37 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                FRINTID(v1, v1);
            }
            x87_restoreround(dyn, ninst, u8);
-            #endif
            break;
        case 0xFD:
            INST_NAME("FSCALE");
            MESSAGE(LOG_DUMP, "Need Optimization (FSCALE)\n");
-            i1 = x87_stackcount(dyn, ninst, x1);
-            x87_forget(dyn, ninst, x1, x2, 0);
-            x87_forget(dyn, ninst, x1, x2, 1);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            v2 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
            if(!BOX64ENV(dynarec_fastround))
                u8 = x87_setround(dyn, ninst, x1, x2, x4);
-            CALL_(native_fscale, -1, BOX64ENV(dynarec_fastround) ? 0 : u8);
+            CALL_D(const_direct_fscale, v1, v1, v2, -1, -1);
            if(!BOX64ENV(dynarec_fastround))
                x87_restoreround(dyn, ninst, u8);
-            x87_unstackcount(dyn, ninst, x1, i1);
            break;
        case 0xFE:
            INST_NAME("FSIN");
-            MESSAGE(LOG_DUMP, "Need Optimization (FSIN)\n");
-            i1 = x87_stackcount(dyn, ninst, x1);
-            x87_forget(dyn, ninst, x1, x2, 0);
+            MESSAGE(LOG_DUMP, "Need Optimization (FCOS)\n");
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
            if(!BOX64ENV(dynarec_fastround))
                u8 = x87_setround(dyn, ninst, x1, x2, x4);
-            CALL_(native_fsin, -1, BOX64ENV(dynarec_fastround) ? 0 : u8);
+            CALL_D(const_direct_fsin, v1, v1, -1, -1, -1);
            if(!BOX64ENV(dynarec_fastround))
                x87_restoreround(dyn, ninst, u8);
-            x87_unstackcount(dyn, ninst, x1, i1);
            break;
        case 0xFF:
            INST_NAME("FCOS");
            MESSAGE(LOG_DUMP, "Need Optimization (FCOS)\n");
-            i1 = x87_stackcount(dyn, ninst, x1);
-            x87_forget(dyn, ninst, x1, x2, 0);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
            if(!BOX64ENV(dynarec_fastround))
                u8 = x87_setround(dyn, ninst, x1, x2, x4);
-            CALL_(native_fcos, -1, BOX64ENV(dynarec_fastround) ? 0 : u8);
+            CALL_D(const_direct_fcos, v1, v1, -1, -1, -1);
            if(!BOX64ENV(dynarec_fastround))
                x87_restoreround(dyn, ninst, u8);
-            x87_unstackcount(dyn, ninst, x1, i1);
            break;
        default:
            DEFAULT;
@@ -550,13 +514,13 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            case 4:
                INST_NAME("FLDENV Ed");
                MESSAGE(LOG_DUMP, "Need Optimization (FLDENV)\n");
-                fpu_purgecache(dyn, ninst, 0, x1, x2, x3); // maybe only x87, not SSE?
+                BARRIER(BARRIER_FLOAT); // maybe only x87, not SSE?
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                if(ed!=x1) {
                    MOVx_REG(x1, ed);
                }
                MOV32w(x2, 0);
-                CALL(fpu_loadenv, -1);
+                CALL(const_fpu_loadenv, -1);
                NATIVE_RESTORE_X87PC();
                break;
            case 5:
@@ -569,13 +533,13 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            case 6:
                INST_NAME("FNSTENV Ed");
                MESSAGE(LOG_DUMP, "Need Optimization (FNSTENV)\n");
-                fpu_purgecache(dyn, ninst, 0, x1, x2, x3); // maybe only x87, not SSE?
+                BARRIER(BARRIER_FLOAT); // maybe only x87, not SSE?
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                if(ed!=x1) {
                    MOVx_REG(x1, ed);
                }
                MOV32w(x2, 0);
-                CALL(fpu_savenv, -1);
+                CALL(const_fpu_savenv, -1);
                break;
            case 7:
                INST_NAME("FNSTCW Ew");
@@ -21,7 +21,7 @@
 #include "dynarec_arm64_functions.h"


-uintptr_t dynarec64_DA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+uintptr_t dynarec64_DA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
 {
    uint8_t nextop = F8;
    int64_t j64;
@@ -21,9 +21,9 @@
 #include "dynarec_arm64_functions.h"


-uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
 {
-    (void)ip; (void)rep; (void)need_epilog;
+    (void)ip; (void)need_epilog;

    uint8_t nextop = F8;
    uint8_t ed;
@@ -133,7 +133,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            INST_NAME("FNINIT");
            MESSAGE(LOG_DUMP, "Need Optimization (FNINIT)\n");
            x87_purgecache(dyn, ninst, 0, x1, x2, x3);
-            CALL(reset_fpu, -1);
+            CALL(const_reset_fpu, -1);
            NATIVE_RESTORE_X87PC();
            break;
        case 0xE8:
@@ -148,10 +148,12 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            SETFLAGS(X_ALL, SF_SET);
            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7, X87_COMBINE(0, nextop&7));
-            if(ST_IS_F(0)) {
-                FCMPS(v1, v2);
-            } else {
-                FCMPD(v1, v2);
+            IFX(X_CF|X_PF|X_ZF) {
+                if(ST_IS_F(0)) {
+                    FCMPS(v1, v2);
+                } else {
+                    FCMPD(v1, v2);
+                }
            }
            FCOMI(x1, x2);
            break;
@@ -167,10 +169,12 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            SETFLAGS(X_ALL, SF_SET);
            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7, X87_COMBINE(0, nextop&7));
-            if(ST_IS_F(0)) {
-                FCMPS(v1, v2);
-            } else {
-                FCMPD(v1, v2);
+            IFX(X_CF|X_PF|X_ZF) {
+                if(ST_IS_F(0)) {
+                    FCMPS(v1, v2);
+                } else {
+                    FCMPD(v1, v2);
+                }
            }
            FCOMI(x1, x2);
            break;
@@ -193,7 +197,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
                s0 = fpu_get_scratch(dyn, ninst);
-                if(arm64_frintts) {
+                if(cpuext.frintts) {
                    FRINT32ZD(s0, v1);
                    FCVTZSwD(x5, s0);
                    STW(x5, wback, fixedaddress);
@@ -223,7 +227,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                u8 = x87_setround(dyn, ninst, x1, x2, x4); // x1 have the modified RPSCR reg
                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
                s0 = fpu_get_scratch(dyn, ninst);
-                if(arm64_frintts) {
+                if(cpuext.frintts) {
                    FRINT32XD(s0, v1);
                    FCVTZSwD(x5, s0);
                    STW(x5, wback, fixedaddress);
@@ -253,7 +257,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                u8 = x87_setround(dyn, ninst, x1, x2, x4); // x1 have the modified RPSCR reg
                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
                s0 = fpu_get_scratch(dyn, ninst);
-                if(arm64_frintts) {
+                if(cpuext.frintts) {
                    FRINT32XD(s0, v1);
                    FCVTZSwD(x5, s0);
                    STW(x5, wback, fixedaddress);
@@ -309,7 +313,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                        X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, 0);
                        // sync top
                        x87_reflectcount(dyn, ninst, x3, x4);
-                        CALL(native_fld, -1);
+                        CALL(const_native_fld, -1);
                        // go back with the top & stack counter
                        x87_unreflectcount(dyn, ninst, x3, x4);
                    }
@@ -327,7 +331,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                        if(ed!=x1) {MOVx_REG(x1, ed);}
                        x87_reflectcount(dyn, ninst, x3, x4);
-                        CALL(native_fstp, -1);
+                        CALL(const_native_fstp, -1);
                        x87_unreflectcount(dyn, ninst, x3, x4);
                    } else {
                        // Painfully long, straight conversion from the C code, shoud be optimized
@@ -21,9 +21,9 @@
 #include "dynarec_arm64_functions.h"


-uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
 {
-    (void)ip; (void)rep; (void)need_epilog;
+    (void)ip; (void)need_epilog;

    uint8_t nextop = F8;
    uint8_t wback;
@@ -21,9 +21,9 @@
 #include "dynarec_arm64_functions.h"


-uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
 {
-    (void)ip; (void)rep; (void)need_epilog;
+    (void)ip; (void)need_epilog;

    uint8_t nextop = F8;
    uint8_t ed;
@@ -49,7 +49,6 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
        case 0xC6:
        case 0xC7:
            INST_NAME("FFREE STx");
-            #if 1
            if((nextop&7)==0 && PK(0)==0xD9 && PK(1)==0xF7) {
                MESSAGE(LOG_DUMP, "Hack for FFREE ST0 / FINCSTP\n");
                x87_do_pop(dyn, ninst, x1);
@@ -57,12 +56,6 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                SKIPTEST(x1);
            } else
                x87_free(dyn, ninst, x1, x2, x3, nextop&7);
-            #else
-            MESSAGE(LOG_DUMP, "Need Optimization\n");
-            x87_purgecache(dyn, ninst, 0, x1, x2, x3);
-            MOV32w(x1, nextop&7);
-            CALL(fpu_do_free, -1);
-            #endif
            break;
        case 0xD0:
        case 0xD1:
@@ -157,7 +150,7 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    VST64(v1, ed, fixedaddress);
                } else {
                    s0 = fpu_get_scratch(dyn, ninst);
-                    if(arm64_frintts) {
+                    if(cpuext.frintts) {
                        FRINT64ZD(s0, v1);
                        VFCVTZSd(s0, s0);
                        VST64(s0, ed, fixedaddress);
@@ -193,24 +186,22 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            case 4:
                INST_NAME("FRSTOR m108byte");
                MESSAGE(LOG_DUMP, "Need Optimization (FRSTOR)\n");
-                fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+                BARRIER(BARRIER_FLOAT);
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                if(ed!=x1) {MOVx_REG(x1, ed);}
-                CALL(native_frstor, -1);
+                CALL(const_native_frstor, -1);
                break;
            case 6:
                INST_NAME("FNSAVE m108byte");
                MESSAGE(LOG_DUMP, "Need Optimization (FNSAVE)\n");
-                fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+                BARRIER(BARRIER_FLOAT);
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                if(ed!=x1) {MOVx_REG(x1, ed);}
-                CALL(native_fsave, -1);
-                CALL(reset_fpu, -1);
+                CALL(const_native_fsave, -1);
                NATIVE_RESTORE_X87PC();
                break;
            case 7:
                INST_NAME("FNSTSW m2byte");
-                //fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
                addr = geted(dyn, addr, ninst, nextop, &ed, x4, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
                LDRw_U12(x2, xEmu, offsetof(x64emu_t, top));
                LDRH_U12(x3, xEmu, offsetof(x64emu_t, sw));
@@ -21,9 +21,9 @@
 #include "dynarec_arm64_functions.h"


-uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
 {
-    (void)ip; (void)rep; (void)need_epilog;
+    (void)ip; (void)need_epilog;

    uint8_t nextop = F8;
    uint8_t wback;
@@ -21,9 +21,9 @@
 #include "dynarec_arm64_functions.h"


-uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
 {
-    (void)ip; (void)rep; (void)need_epilog;
+    (void)ip; (void)need_epilog;

    uint8_t nextop = F8;
    uint8_t ed, wback, u8;
@@ -101,11 +101,13 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            SETFLAGS(X_ALL, SF_SET);
            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7, X87_COMBINE(0, nextop&7));
-            if(ST_IS_F(0))
-            {
-                FCMPS(v1, v2);
-            } else {
-                FCMPD(v1, v2);
+            IFX(X_CF|X_PF|X_ZF) {
+                if(ST_IS_F(0))
+                {
+                    FCMPS(v1, v2);
+                } else {
+                    FCMPD(v1, v2);
+                }
            }
            FCOMI(x1, x2);
            X87_POP_OR_FAIL(dyn, ninst, x3);
@@ -122,11 +124,13 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            SETFLAGS(X_ALL, SF_SET);
            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7, X87_COMBINE(0, nextop&7));
-            if(ST_IS_F(0))
-            {
-                FCMPS(v1, v2);
-            } else {
-                FCMPD(v1, v2);
+            IFX(X_CF|X_PF|X_ZF) {
+                if(ST_IS_F(0))
+                {
+                    FCMPS(v1, v2);
+                } else {
+                    FCMPD(v1, v2);
+                }
            }
            FCOMI(x1, x2);
            X87_POP_OR_FAIL(dyn, ninst, x3);
@@ -273,7 +277,7 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                if(ed!=x1) {MOVx_REG(x1, ed);}
                s0 = x87_stackcount(dyn, ninst, x3);
-                CALL(fpu_fbld, -1);
+                CALL(const_fpu_fbld, -1);
                x87_unstackcount(dyn, ninst, x3, s0);
                break;
            case 5:
@@ -310,7 +314,7 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                x87_forget(dyn, ninst, x1, x2, 0);
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                if(ed!=x1) {MOVx_REG(x1, ed);}
-                CALL(fpu_fbst, -1);
+                CALL(const_fpu_fbst, -1);
                x87_unstackcount(dyn, ninst, x1, i1);
                X87_POP_OR_FAIL(dyn, ninst, x3);
                break;
@@ -354,7 +358,7 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                        B_MARK3(c__);
                        MARK2;
                    }
-                    if(arm64_frintts) {
+                    if(cpuext.frintts) {
                        FRINT64XD(s0, v1);
                        VFCVTZSd(s0, s0);
                        VST64(s0, wback, fixedaddress);
@@ -451,7 +451,7 @@ void emit_and8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
        BICw_REG(xFlags, xFlags, s3);
    }
-    if(arm64_flagm) {
+    if(cpuext.flagm) {
        COMP_ZFSF(s1, 8)
    } else {
        IFX(X_ZF) {
@@ -496,7 +496,7 @@ void emit_and8c(dynarec_arm_t* dyn, int ninst, int s1, uint8_t c, int s3, int s4
        MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
        BICw(xFlags, xFlags, s3);
    }
-    if(arm64_flagm) {
+    if(cpuext.flagm) {
        COMP_ZFSF(s1, 8)
    } else {
        IFX(X_ZF) {
@@ -639,7 +639,7 @@ void emit_and16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
        BICw_REG(xFlags, xFlags, s3);
    }
-    if(arm64_flagm) {
+    if(cpuext.flagm) {
        COMP_ZFSF(s1, 16)
    } else {
        IFX(X_ZF) {
@@ -684,7 +684,7 @@ void emit_and16c(dynarec_arm_t* dyn, int ninst, int s1, int16_t c, int s3, int s
        MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
        BICw(xFlags, xFlags, s3);
    }
-    if(arm64_flagm) {
+    if(cpuext.flagm) {
        COMP_ZFSF(s1, 16)
    } else {
        IFX(X_ZF) {
@@ -991,7 +991,7 @@ void emit_adc32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3
    }
    IFNATIVE_BEFORE(NF_CF) {
        if(INVERTED_CARRY_BEFORE) {
-            if(arm64_flagm)
+            if(cpuext.flagm)
                CFINV();
            else {
                MRS_nzcv(s3);
@@ -1134,7 +1134,7 @@ void emit_adc8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
    }
    IFNATIVE_BEFORE(NF_CF) {
        if(INVERTED_CARRY_BEFORE) {
-            if(arm64_flagm)
+            if(cpuext.flagm)
                CFINV();
            else {
                MRS_nzcv(s3);
@@ -1198,7 +1198,7 @@ void emit_adc16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
    }
    IFNATIVE_BEFORE(NF_CF) {
        if(INVERTED_CARRY_BEFORE) {
-            if(arm64_flagm)
+            if(cpuext.flagm)
                CFINV();
            else {
                MRS_nzcv(s3);
@@ -1321,7 +1321,7 @@ void emit_sbb32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3
    }
    IFNATIVE_BEFORE(NF_CF) {
        if(!INVERTED_CARRY_BEFORE) {
-            if(arm64_flagm)
+            if(cpuext.flagm)
                CFINV();
            else {
                MRS_nzcv(s3);
@@ -1470,7 +1470,7 @@ void emit_sbb8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
    }
    IFNATIVE_BEFORE(NF_CF) {
        if(!INVERTED_CARRY_BEFORE) {
-            if(arm64_flagm)
+            if(cpuext.flagm)
                CFINV();
            else {
                MRS_nzcv(s3);
@@ -1535,7 +1535,7 @@ void emit_sbb16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
    }
    IFNATIVE_BEFORE(NF_CF) {
        if(!INVERTED_CARRY_BEFORE) {
-            if(arm64_flagm)
+            if(cpuext.flagm)
                CFINV();
            else {
                MRS_nzcv(s3);
@@ -1030,7 +1030,7 @@ void emit_rol16c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int
    if (!c) return;

    if(!BOX64ENV(cputype))
-        IFX(X_OF) {
+        IFX2(X_OF, && c == 1) {
            LSRw(s3, s1, 14);
            EORw_REG_LSR(s3, s3, s3, 1);
            BFIw(xFlags, s3, F_OF, 1);
@@ -1058,7 +1058,7 @@ void emit_ror16c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int
    if (!c) return;

    if(!BOX64ENV(cputype))
-        IFX(X_OF) {
+        IFX2(X_OF, && c == 1) {
            EORw_REG_LSR(s3, s1, s1, 15);
            BFIw(xFlags, s3, F_OF, 1);
        }
@@ -1086,20 +1086,11 @@ void emit_rcl8c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s
    if (!c && !BOX64ENV(cputype)) return;
    if(c) {
        BFIw(s1, xFlags, 8, 1); // insert cf
-        if(BOX64ENV(cputype)) {
-            IFX(X_OF|X_CF) {
-                BFXILw(xFlags, s1, 8-c, 1);
-            }
-        } else {
-            IFX(X_CF) {
-                BFXILw(xFlags, s1, 8-c, 1);
-            }
-        }
-    }
-    if(!BOX64ENV(cputype)) {
        IFX(X_OF|X_CF) {
            BFXILw(xFlags, s1, 8-c, 1);
        }
+    }
+    if(!BOX64ENV(cputype)) {
        IFX(X_OF) {
            LSRw(s3, s1, 6);
            EORw_REG_LSR(s3, s3, s3, 1);
@@ -1281,6 +1272,102 @@ void emit_rcr32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, i
    }
 }

+// emit RCL32/RCL64 instruction, from s1 , shift s2 (destroyed), store result in s1 using s3 and s4 as scratch
+void emit_rcl32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4, int s5)
+{
+    MAYUSE(s1); MAYUSE(s3); MAYUSE(s4);
+
+    if(BOX64ENV(cputype)) {
+        IFX(X_OF|X_CF) {
+            MOVZw(s4, rex.w?64:32);
+            SUBx_REG(s4, s4, s2);
+            LSRxw_REG(s3, s1, s4);
+        }
+    } else {
+        IFX(X_CF) {
+            MOVZw(s4, rex.w?64:32);
+            SUBx_REG(s4, s4, s2);
+            LSRxw_REG(s3, s1, s4);
+        }
+    }
+    if(!BOX64ENV(cputype))
+        IFX(X_OF) {
+            LSRxw(s4, s1, rex.w?62:30);
+            EORw_REG_LSR(s4, s4, s4, 1);
+            BFIw(xFlags, s4, F_OF, 1);
+        }
+    SUBw_U12(s4, s2, 1);
+    CBNZw(s4, 4+4*3);
+    /*if(c==1)*/
+    {
+        LSLxw(s1, s1, 1);
+        BFIxw(s1, xFlags, 0, 1);
+        B(4+4*9);
+    }
+    /*else*/
+    {
+        LSLxw_REG(s4, s1, s2);
+        UBFXw(s5, xFlags, 0, 1);
+        SUBw_U12(s2, s2, 1);    //s2 = c-1 now
+        LSLxw_REG(s5, s5, s2);
+        ORRxw_REG(s4, s4, s5);
+        MOVZw(s5, rex.w?64:32); // -1 to compensate from the c-1 is s2
+        SUBw_REG(s5, s5, s2);
+        LSRxw_REG(s1, s1, s5);
+        ORRxw_REG(s1, s1, s4);
+    }
+    IFX(X_CF) {
+        BFIw(xFlags, s3, F_CF, 1);
+    }
+    IFX2(X_OF, && BOX64ENV(cputype)) {
+        EORxw_REG_LSR(s3, s3, s1, rex.w?63:31);
+        BFIw(xFlags, s3, F_OF, 1);
+    }
+}
+// emit RCR32/RCR64 instruction, from s1 , shift s2, store result in s1 using s3 and s4 as scratch
+void emit_rcr32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4, int s5)
+{
+    MAYUSE(s1); MAYUSE(s3); MAYUSE(s4);
+
+    IFX2(X_OF, && !BOX64ENV(cputype)) {
+        EORxw_REG_LSR(s3, xFlags, s1, rex.w?63:31);
+        BFIw(xFlags, s3, F_OF, 1);
+    }
+    SUBw_U12(s4, s2, 1);
+    IFX(X_CF) {
+        LSRxw_REG(s3, s1, s4);
+        BFIw(xFlags, s3, F_res2, 1); // saving the value without reserving s3
+    }
+    CBNZw(s4, 4+4*3);
+    /*if(c==1)*/
+    {
+        LSRxw(s1, s1, 1);
+        BFIxw(s1, xFlags, rex.w?63:31, 1);
+        B(4+4*9);
+    } 
+    /*else*/
+    {
+        LSRxw_REG(s5, s1, s2);
+        UBFXw(s4, xFlags, 0, 1);
+        MOVZw(s3, (rex.w?64:32));
+        SUBx_REG(s3, s3, s2);
+        LSLxw_REG(s4, s4, s3);
+        ORRxw_REG(s5, s5, s4);
+        ADDx_U12(s3, s3, 1);
+        LSLxw_REG(s1, s1, s3);
+        ORRxw_REG(s1, s1, s5);
+    }
+    IFX(X_CF) {
+        BFXILw(xFlags, xFlags, F_res2, 1);
+        BFCw(xFlags, F_res2, 1);
+    }
+    IFX2(X_OF, && BOX64ENV(cputype)) {
+        LSRxw(s4, s1, rex.w?62:30);
+        EORw_REG_LSR(s4, s4, s4, 1);
+        BFIw(xFlags, s4, F_OF, 1);
+    }
+}
+
 // emit SHRD32 instruction, from s1, fill s2 , constant c, store result in s1 using s3 and s4 as scratch
 void emit_shrd32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, uint32_t c, int s3, int s4)
 {
@@ -118,12 +118,12 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            nextop = F8;
            GETGD;
            GETEXSD(q0, 0, 0);
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                MRS_fpsr(x5);
                BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
                MSR_fpsr(x5);
            }
-            if(!BOX64ENV(dynarec_fastround) && arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && cpuext.frintts) {
                v0 = fpu_get_scratch(dyn, ninst);
                if(rex.w) {
                    FRINT64ZD(v0, q0);
@@ -134,7 +134,7 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            } else {
                FCVTZSxwD(gd, q0);
            }
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                MRS_fpsr(x5);   // get back FPSR to check the IOC bit
                TBZ_NEXT(x5, FPSR_IOC);
                if(rex.w) {
@@ -149,14 +149,14 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            nextop = F8;
            GETGD;
            GETEXSD(q0, 0, 0);
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                MRS_fpsr(x5);
                BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
                MSR_fpsr(x5);
            }
            u8 = sse_setround(dyn, ninst, x1, x2, x3);
            d1 = fpu_get_scratch(dyn, ninst);
-            if(!BOX64ENV(dynarec_fastround) && arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && cpuext.frintts) {
                if(rex.w) {
                    FRINT64XD(d1, q0);
                } else {
@@ -167,7 +167,7 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            }
            x87_restoreround(dyn, ninst, u8);
            FCVTZSxwD(gd, d1);
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                MRS_fpsr(x5);   // get back FPSR to check the IOC bit
                TBZ_NEXT(x5, FPSR_IOC);
                if(rex.w) {
@@ -187,7 +187,7 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    nextop = F8;
                    GETEB(x1, 0);
                    GETGD;
-                    if(arm64_crc32) {
+                    if(cpuext.crc32) {
                        CRC32CB(gd, gd, ed);
                    } else {
                        EORw_REG(gd, gd, ed);
@@ -204,7 +204,7 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    nextop = F8;
                    GETED(0);
                    GETGD;
-                    if(arm64_crc32) {
+                    if(cpuext.crc32) {
                        CRC32Cxw(gd, gd, ed);
                    } else {
                        MOV32w(x2, 0x82f63b78);
@@ -430,7 +430,7 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            INST_NAME("HSUBPS Gx, Ex");
            nextop = F8;
            GETGX(v0, 1);
-            GETEX(v1, 0, 1);
+            GETEX(v1, 0, 0);
            d0 = fpu_get_scratch(dyn, ninst);
            VUZP1Q_32(d0, v0, v1);
            VUZP2Q_32(v0, v0, v1);
@@ -456,7 +456,6 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                j64 = (uint32_t)(addr+i32_);                            \
            else                                                        \
                j64 = addr+i32_;                                        \
-            BARRIER(BARRIER_MAYBE);                                     \
            JUMP(j64, 1);                                               \
            GETFLAGS;                                                   \
            if(dyn->insts[ninst].x64.jmp_insts==-1 ||                   \
@@ -466,10 +465,10 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                Bcond(NO, i32);                                         \
                if(dyn->insts[ninst].x64.jmp_insts==-1) {               \
                    if(!(dyn->insts[ninst].x64.barrier&BARRIER_FLOAT))  \
-                        fpu_purgecache(dyn, ninst, 1, x1, x2, x3);      \
+                        fpu_purgecache(dyn, ninst, 1, x1, x2, x3, 0);   \
                    jump_to_next(dyn, j64, 0, ninst, rex.is32bits);     \
                } else {                                                \
-                    CacheTransform(dyn, ninst, cacheupd, x1, x2, x3);   \
+                    CacheTransform(dyn, ninst, cacheupd);               \
                    i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->native_size);    \
                    SKIP_SEVL(i32);                                     \
                    B(i32);                                             \
@@ -523,9 +522,7 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            GETGX(v0, 1);
            GETEX(v1, 0, 0);
            q0 = fpu_get_scratch(dyn, ninst);
-            static float addsubps[4] = {-1.f, 1.f, -1.f, 1.f};
-            MAYUSE(addsubps);
-            MOV64x(x2, (uintptr_t)&addsubps);   // no need to use table64, as box64 is loaded in low memory
+            TABLE64C(x2, const_4f_m1_1_m1_1);
            VLDR128_U12(q0, x2, 0);
            VFMLAQS(v0, v1, q0);
            break;
@@ -551,7 +548,7 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                SQXTN_32(v0, v0);   // convert int64 -> int32 with saturation in lower part, RaZ high part
            } else {
                u8 = sse_setround(dyn, ninst, x1, x2, x3);
-                if(arm64_frintts) {
+                if(cpuext.frintts) {
                    VFRINT32XDQ(v0, v1);    // round, handling of overflow and Nan to 0x80000000
                    VFCVTNSQD(v0, v0);  // convert double -> int64
                    SQXTN_32(v0, v0);   // convert int64 -> int32 with saturation in lower part, RaZ high part
@@ -6,11 +6,11 @@
 #include "debug.h"
 #include "box64context.h"
 #include "box64cpu.h"
+#include "emu/x64run_private.h"
 #include "emu/x64emu_private.h"
 #include "x64emu.h"
 #include "box64stack.h"
 #include "callback.h"
-#include "emu/x64run_private.h"
 #include "x64trace.h"
 #include "dynarec_native.h"

@@ -119,12 +119,12 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            nextop = F8;
            GETGD;
            GETEXSS(d0, 0, 0);
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                MRS_fpsr(x5);
                BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
                MSR_fpsr(x5);
            }
-            if(!BOX64ENV(dynarec_fastround) && arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && cpuext.frintts) {
                v0 = fpu_get_scratch(dyn, ninst);
                if(rex.w) {
                    FRINT64ZS(v0, d0);
@@ -135,7 +135,7 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            } else {
                FCVTZSxwS(gd, d0);
            }
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                MRS_fpsr(x5);   // get back FPSR to check the IOC bit
                TBZ_NEXT(x5, FPSR_IOC);
                if(rex.w) {
@@ -150,14 +150,14 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            nextop = F8;
            GETGD;
            GETEXSS(q0, 0, 0);
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                MRS_fpsr(x5);
                BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
                MSR_fpsr(x5);
            }
            u8 = sse_setround(dyn, ninst, x1, x2, x3);
            d1 = fpu_get_scratch(dyn, ninst);
-            if(!BOX64ENV(dynarec_fastround) && arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && cpuext.frintts) {
                if(rex.w) {
                    FRINT64XS(d1, q0);
                } else {
@@ -168,7 +168,7 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            }
            x87_restoreround(dyn, ninst, u8);
            FCVTZSxwS(gd, d1);
-            if(!BOX64ENV(dynarec_fastround) && !arm64_frintts) {
+            if(!BOX64ENV(dynarec_fastround) && !cpuext.frintts) {
                MRS_fpsr(x5);   // get back FPSR to check the IOC bit
                TBZ_NEXT(x5, FPSR_IOC);
                if(rex.w) {
@@ -212,6 +212,24 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            }
            break;

+        #define GO(GETFLAGS, NO, YES, F)                                                                                                                  \
+            READFLAGS(F);                                                                                                                                 \
+            GETFLAGS;                                                                                                                                     \
+            nextop = F8;                                                                                                                                  \
+            GETGD;                                                                                                                                        \
+            if (MODREG) {                                                                                                                                 \
+                ed = TO_NAT((nextop & 7) + (rex.b << 3));                                                                                                 \
+                CSELxw(gd, ed, gd, YES);                                                                                                                  \
+            } else {                                                                                                                                      \
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff << (2 + rex.w), (1 << (2 + rex.w)) - 1, rex, NULL, 0, 0); \
+                Bcond(NO, +8);                                                                                                                            \
+                LDxw(gd, ed, fixedaddress);                                                                                                               \
+                if (!rex.w) { MOVw_REG(gd, gd); }                                                                                                         \
+            }
+
+        GOCOND(0x40, "CMOV", "Gd, Ed");
+        #undef GO
+
        case 0x51:
            INST_NAME("SQRTSS Gx, Ex");
            nextop = F8;
@@ -322,7 +340,7 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            if(BOX64ENV(dynarec_fastround)) {
                VFCVTZSQS(v0, v1);
            } else {
-                if(arm64_frintts) {
+                if(cpuext.frintts) {
                    VFRINT32ZSQ(v0, v1);
                    VFCVTZSQS(v0, v0);
                } else {
@@ -503,15 +521,37 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
        case 0xAE:
            nextop = F8;
            switch((nextop>>3)&7) {
+                case 0:
+                case 1:
+                    if(rex.is32bits || !MODREG) {
+                        INST_NAME("Illegal AE");
+                        FAKEED;
+                        UDF(0);
+                    } else {
+                        if(((nextop>>3)&7)==1) {INST_NAME("RDGSBASE");} else {INST_NAME("RDFSBASE");}
+                        ed = TO_NAT((nextop & 7) + (rex.b << 3));
+                        int seg = _FS + ((nextop>>3)&7);
+                        grab_segdata(dyn, addr, ninst, x4, seg);
+                        MOVxw_REG(ed, x4);
+                    }
+                     break;
                case 2:
-                    INST_NAME("(unsupported) WRFSBASE Ed");
-                    FAKEED;
-                    UDF(0);
-                    break;
                case 3:
-                    INST_NAME("(unsupported) WRGSBASE Ed");
-                    FAKEED;
-                    UDF(0);
+                    if(rex.is32bits || !MODREG) {
+                        INST_NAME("Illegal AE");
+                        FAKEED;
+                        UDF(0);
+                    } else {
+                        if(((nextop>>3)&7)==3) {INST_NAME("WRGSBASE");} else {INST_NAME("WRFSBASE");}
+                        ed = TO_NAT((nextop & 7) + (rex.b << 3));
+                        int seg = _FS + ((nextop>>3)&7)-2;
+                        if (!rex.w) {
+                            MOVw_REG(x4, ed);
+                            STRx_U12(x4, xEmu, offsetof(x64emu_t, segs_offs[seg]));
+                        } else {
+                            STRx_U12(ed, xEmu, offsetof(x64emu_t, segs_offs[seg]));
+                        }
+                    }
                    break;
                case 5:
                    INST_NAME("(unsupported) INCSSPD/INCSSPQ Ed");
@@ -574,7 +614,7 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n

        case 0xBC:
            INST_NAME("TZCNT Gd, Ed");
-            if(!BOX64ENV(dynarec_safeflags)) {
+            if (!BOX64DRENV(dynarec_safeflags)) {
                SETFLAGS(X_CF|X_ZF, SF_SUBSET);
            } else {
                if(BOX64ENV(cputype)) {
@@ -599,7 +639,7 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                CSETw(x3, cEQ);
                BFIw(xFlags, x3, F_ZF, 1);  // ZF = is dest 0?
            }
-            if(BOX64ENV(dynarec_safeflags)) {
+            if (BOX64DRENV(dynarec_safeflags)) {
                IFX(X_AF) BFCw(xFlags, F_AF, 1);
                IFX(X_PF) BFCw(xFlags, F_PF, 1);
                IFX(X_SF) BFCw(xFlags, F_SF, 1);
@@ -608,7 +648,7 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            break;
        case 0xBD:
            INST_NAME("LZCNT Gd, Ed");
-            if(!BOX64ENV(dynarec_safeflags)) {
+            if (!BOX64DRENV(dynarec_safeflags)) {
                SETFLAGS(X_CF|X_ZF, SF_SUBSET);
            } else {
                if(BOX64ENV(cputype)) {
@@ -632,7 +672,7 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                CSETw(x3, cEQ);
                BFIw(xFlags, x3, F_ZF, 1);  // ZF = is dest 0?
            }
-            if(BOX64ENV(dynarec_safeflags)) {
+            if (BOX64DRENV(dynarec_safeflags)) {
                IFX(X_AF) BFCw(xFlags, F_AF, 1);
                IFX(X_PF) BFCw(xFlags, F_PF, 1);
                IFX(X_SF) BFCw(xFlags, F_SF, 1);
@@ -4,7 +4,6 @@
 #include <errno.h>
 #include <string.h>
 #include <math.h>
-#include <signal.h>
 #include <sys/types.h>
 #include <unistd.h>

@@ -92,6 +91,7 @@ int fpu_get_reg_x87(dynarec_arm_t* dyn, int ninst, int t, int n)
    dyn->n.neoncache[i].n = n;
    dyn->n.neoncache[i].t = t;
    dyn->n.news |= (1<<i);
+    dyn->use_x87 = 1;
    return i; // return a Dx
 }
 // Free a FPU double reg
@@ -131,6 +131,7 @@ int fpu_get_reg_emm(dynarec_arm_t* dyn, int ninst, int emm)
    dyn->n.neoncache[ret].t = NEON_CACHE_MM;
    dyn->n.neoncache[ret].n = emm;
    dyn->n.news |= (1<<(ret));
+    dyn->use_mmx = 1;
    return ret;
 }
 // Get an XMM quad reg
@@ -146,10 +147,12 @@ int fpu_get_reg_xmm(dynarec_arm_t* dyn, int t, int xmm)
    dyn->n.neoncache[i].t = t;
    dyn->n.neoncache[i].n = xmm;
    dyn->n.news |= (1<<i);
+    dyn->use_xmm = 1;
    return i;
 }
 int internal_mark_ymm(dynarec_arm_t* dyn, int t, int ymm, int reg)
 {
+    dyn->use_ymm = 1;
    if((dyn->n.neoncache[reg].t==NEON_CACHE_YMMR) || (dyn->n.neoncache[reg].t==NEON_CACHE_YMMW)) {
        if(dyn->n.neoncache[reg].n == ymm) {
            // already there!
@@ -182,7 +185,6 @@ int is_ymm_to_keep(dynarec_arm_t* dyn, int reg, int k1, int k2, int k3)
 // Reset fpu regs counter
 static void fpu_reset_reg_neoncache(neoncache_t* n)
 {
-    n->fpu_reg = 0;
    for (int i=0; i<32; ++i) {
        n->fpuused[i]=0;
        n->neoncache[i].v = 0;
@@ -563,7 +565,6 @@ void neoncacheUnwind(neoncache_t* cache)
    // And now, rebuild the x87cache info with neoncache
    cache->mmxcount = 0;
    cache->fpu_scratch = 0;
-    cache->fpu_reg = 0;
    for(int i=0; i<8; ++i) {
        cache->x87cache[i] = -1;
        cache->mmxcache[i] = -1;
@@ -579,13 +580,11 @@ void neoncacheUnwind(neoncache_t* cache)
                case NEON_CACHE_MM:
                    cache->mmxcache[cache->neoncache[i].n] = i;
                    ++cache->mmxcount;
-                    ++cache->fpu_reg;
                    break;
                case NEON_CACHE_XMMR:
                case NEON_CACHE_XMMW:
                    cache->ssecache[cache->neoncache[i].n].reg = i;
                    cache->ssecache[cache->neoncache[i].n].write = (cache->neoncache[i].t==NEON_CACHE_XMMW)?1:0;
-                    ++cache->fpu_reg;
                    break;
                case NEON_CACHE_YMMR:
                case NEON_CACHE_YMMW:
@@ -597,7 +596,6 @@ void neoncacheUnwind(neoncache_t* cache)
                    cache->x87cache[x87reg] = cache->neoncache[i].n;
                    cache->x87reg[x87reg] = i;
                    ++x87reg;
-                    ++cache->fpu_reg;
                    break;
                case NEON_CACHE_SCR:
                    cache->fpuused[i] = 0;
@@ -617,7 +615,6 @@ void neoncacheUnwind(neoncache_t* cache)
                cache->neoncache[reg].n = i;
                cache->ssecache[i].reg = reg;
                cache->ssecache[i].write = (cache->xmm_write&(1<<i))?1:0;
-                ++cache->fpu_reg;
            }
        cache->xmm_write = cache->xmm_removed = 0;
    }
@@ -774,28 +771,34 @@ static register_mapping_t register_mappings[] = {
    { "rip", "x27" },
 };

+static const char* df_status[] = {"unknown", "set", "none_pending", "none"};
+
 void printf_x64_instruction(dynarec_native_t* dyn, zydis_dec_t* dec, instruction_x64_t* inst, const char* name);
 void inst_name_pass3(dynarec_native_t* dyn, int ninst, const char* name, rex_t rex)
 {
-    if (!dyn->need_dump && !BOX64ENV(dynarec_gdbjit) && !BOX64ENV(dynarec_perf_map)) return;
+    if (!dyn->need_dump && !BOX64ENV(dynarec_gdbjit) && !BOX64ENV(dynarec_perf_map)) {
+        /*zydis_dec_t* dec = rex.is32bits ? my_context->dec32 : my_context->dec;
+        if(dec && !OpcodeOK(dec, dyn->insts[ninst].x64.addr) && !strstr(name, "Illegal")) {
+            uint8_t* p = (uint8_t*)dyn->insts[ninst].x64.addr;
+            printf_log(LOG_INFO, "Warning: %p invalid opcode %02x %02x %02x %02x %02x %02x treated as valid\n", p, p[0], p[1], p[2], p[3], p[4], p[5]);
+        }*/
+        return;
+    }

    static char buf[4096];
-    int length = sprintf(buf, "barrier=%d state=%d/%d/%d(%d:%d->%d:%d), %s=%X/%X, use=%X, need=%X/%X, sm=%d(%d/%d)",
+    int length = sprintf(buf, "barrier=%d state=%d/%s(%s->%s)/%d, set=%X/%X, use=%X, need=%X/%X, sm=%d(%d/%d/%d)",
        dyn->insts[ninst].x64.barrier,
        dyn->insts[ninst].x64.state_flags,
-        dyn->f.pending,
-        dyn->f.dfnone,
-        dyn->insts[ninst].f_entry.pending,
-        dyn->insts[ninst].f_entry.dfnone,
-        dyn->insts[ninst].f_exit.pending,
-        dyn->insts[ninst].f_exit.dfnone,
-        dyn->insts[ninst].x64.may_set ? "may" : "set",
+        df_status[dyn->f],
+        df_status[dyn->insts[ninst].f_entry],
+        df_status[dyn->insts[ninst].f_exit],
+        dyn->insts[ninst].df_notneeded,
        dyn->insts[ninst].x64.set_flags,
        dyn->insts[ninst].x64.gen_flags,
        dyn->insts[ninst].x64.use_flags,
        dyn->insts[ninst].x64.need_before,
        dyn->insts[ninst].x64.need_after,
-        dyn->smwrite, dyn->insts[ninst].will_write, dyn->insts[ninst].last_write);
+        dyn->smwrite, dyn->insts[ninst].will_write, dyn->insts[ninst].last_write, dyn->insts[ninst].lock);
    if (dyn->insts[ninst].nat_flags_op) {
        if (dyn->insts[ninst].nat_flags_op == NAT_FLAG_OP_TOUCH && dyn->insts[ninst].before_nat_flags)
            length += sprintf(buf + length, " NF:%d/read:%x", dyn->insts[ninst].nat_flags_op, dyn->insts[ninst].before_nat_flags);
@@ -868,6 +871,9 @@ void inst_name_pass3(dynarec_native_t* dyn, int ninst, const char* name, rex_t r
    if (dyn->insts[ninst].purge_ymm) {
        length += sprintf(buf + length, " purgeYmm=%04x", dyn->insts[ninst].purge_ymm);
    }
+    if (dyn->insts[ninst].preload_xmmymm) {
+        length += sprintf(buf + length, " preload=%x/%d", dyn->insts[ninst].preload_xmmymm, dyn->insts[ninst].preload_from);
+    }
    if (dyn->n.stack || dyn->insts[ninst].n.stack_next || dyn->insts[ninst].n.x87stack) {
        length += sprintf(buf + length, " X87:%d/%d(+%d/-%d)%d", dyn->n.stack, dyn->insts[ninst].n.stack_next, dyn->insts[ninst].n.stack_push, dyn->insts[ninst].n.stack_pop, dyn->insts[ninst].n.x87stack);
    }
@@ -1155,6 +1161,7 @@ void updateNativeFlags(dynarec_native_t* dyn)
 void rasNativeState(dynarec_arm_t* dyn, int ninst)
 {
    dyn->insts[ninst].nat_flags_op = dyn->insts[ninst].set_nat_flags = dyn->insts[ninst].use_nat_flags = dyn->insts[ninst].need_nat_flags = 0;
+    dyn->insts[ninst].ymm0_in = dyn->insts[ninst].ymm0_sub = dyn->insts[ninst].ymm0_add = dyn->insts[ninst].ymm0_out = dyn->insts[ninst].purge_ymm = 0;
 }

 int nativeFlagsNeedsTransform(dynarec_arm_t* dyn, int ninst)
@@ -1196,44 +1203,333 @@ void fpu_unwind_restore(dynarec_arm_t* dyn, int ninst, neoncache_t* cache)
    memcpy(&dyn->insts[ninst].n, cache, sizeof(neoncache_t));
 }

-static void propagateXMMUneeded(dynarec_arm_t* dyn, int ninst, int a)
+static uint32_t getXYMMMask(dynarec_arm_t* dyn, int ninst)
 {
-    if(!ninst) return;
-    ninst = getNominalPred(dyn, ninst);
-    while(ninst>=0) {
-        if(dyn->insts[ninst].n.xmm_used&(1<<a)) return; // used, value is needed
-        if(dyn->insts[ninst].x64.barrier&BARRIER_FLOAT) return; // barrier, value is needed
-        if(dyn->insts[ninst].n.xmm_unneeded&(1<<a)) return; // already handled
-        if(dyn->insts[ninst].x64.jmp) return;   // stop when a jump is detected, that gets too complicated
-        dyn->insts[ninst].n.xmm_unneeded |= (1<<a); // flags
-        ninst = getNominalPred(dyn, ninst); // continue
+    if(ninst<0) return 0;
+    uint32_t ret = 0;
+    for (int i=0; i<32; ++i) {
+        switch(dyn->insts[ninst].n.neoncache[i].t) {
+            case NEON_CACHE_XMMR:
+            case NEON_CACHE_XMMW:
+                ret |= 1<<dyn->insts[ninst].n.neoncache[i].n;
+                break;
+            case NEON_CACHE_YMMR:
+            case NEON_CACHE_YMMW:
+                ret |= 1<<(16+dyn->insts[ninst].n.neoncache[i].n);
+                break;
+        }
    }
+    return ret;
 }

-static void propagateYMMUneeded(dynarec_arm_t* dyn, int ninst, int a)
+static void propagateXYMMUneeded(dynarec_arm_t* dyn, int ninst, uint16_t mask_x, uint16_t mask_y)
 {
    if(!ninst) return;
    ninst = getNominalPred(dyn, ninst);
    while(ninst>=0) {
-        if(dyn->insts[ninst].n.ymm_used&(1<<a)) return; // used, value is needed
+        mask_x &= ~dyn->insts[ninst].n.xmm_used;
+        mask_y &= ~dyn->insts[ninst].n.ymm_used;
+        if(!mask_x && !mask_y) return; // used, value is needed
        if(dyn->insts[ninst].x64.barrier&BARRIER_FLOAT) return; // barrier, value is needed
-        if(dyn->insts[ninst].n.ymm_unneeded&(1<<a)) return; // already handled
+        mask_x &= ~dyn->insts[ninst].n.xmm_unneeded;
+        mask_y &= ~dyn->insts[ninst].n.ymm_unneeded;
+        if(!mask_x && !mask_y) return; // already handled
        if(dyn->insts[ninst].x64.jmp) return;   // stop when a jump is detected, that gets too complicated
-        dyn->insts[ninst].n.ymm_unneeded |= (1<<a); // flags
+        dyn->insts[ninst].n.xmm_unneeded |= mask_x; // flags
+        dyn->insts[ninst].n.ymm_unneeded |= mask_y; // flags
        ninst = getNominalPred(dyn, ninst); // continue
    }
 }

 void updateUneeded(dynarec_arm_t* dyn)
 {
+    if(!dyn->use_xmm && !dyn->use_ymm)
+        return;
    for(int ninst=0; ninst<dyn->size; ++ninst) {
-        if(dyn->insts[ninst].n.xmm_unneeded)
-            for(int i=0; i<16; ++i)
-                if(dyn->insts[ninst].n.xmm_unneeded&(1<<i))
-                    propagateXMMUneeded(dyn, ninst, i);
-        if(dyn->insts[ninst].n.ymm_unneeded)
-            for(int i=0; i<16; ++i)
-                if(dyn->insts[ninst].n.ymm_unneeded&(1<<i))
-                    propagateYMMUneeded(dyn, ninst, i);
+        if(dyn->insts[ninst].n.xmm_unneeded || dyn->insts[ninst].n.ymm_unneeded)
+            propagateXYMMUneeded(dyn, ninst, dyn->insts[ninst].n.xmm_unneeded, dyn->insts[ninst].n.ymm_unneeded);
+        if(dyn->insts[ninst].df_notneeded)
+            propagate_nodf(dyn, ninst);
    }
+    // try to add some preload of XYMM on jump were it would make sense
+    for(int ninst=0; ninst<dyn->size; ++ninst)
+        if((ninst && dyn->insts[ninst].pred_sz>1) || (!ninst && dyn->insts[ninst].pred_sz)) {
+            int i2 = dyn->size+1;
+            for(int j=0; j<dyn->insts[ninst].pred_sz; ++j)
+                if(dyn->insts[ninst].pred[j]!=ninst-1 && dyn->insts[ninst].pred[j]<i2)
+                    i2 = dyn->insts[ninst].pred[j];
+            if(i2!=dyn->size+1 && i2>ninst) {
+                uint32_t used = getXYMMMask(dyn, i2);
+                if(used) {
+                    int prev = getNominalPred(dyn, ninst);
+                    if((prev==ninst-1) || !ninst) {
+                        uint32_t dest = getXYMMMask(dyn, prev);
+                        // removed unneeded
+                        uint32_t unnedded = dyn->insts[ninst].n.xmm_unneeded | (dyn->insts[ninst].n.ymm_unneeded<<16);
+                        // create the preload mask
+                        uint32_t preload = (used & ~dest) & ~unnedded;
+                        if(preload)
+                            addSSEPreload(dyn, i2, ninst, preload);
+                    }
+                }
+            }
+        }
+}
+
+void tryEarlyFpuBarrier(dynarec_arm_t* dyn, int last_fpu_used, int ninst)
+{
+    // there is a barrier at ninst
+    // check if, up to last fpu_used, if there is some suspicious jump that would prevent the barrier to be put earlier
+    int usefull = 0;
+    for(int i=ninst-1; i>last_fpu_used; --i)
+    {
+        if(!dyn->insts[i].x64.has_next)
+            return; // break of chain, don't try to be smart for now
+        if(dyn->insts[i].x64.barrier&BARRIER_FLOAT)
+            return; // already done?
+        if(dyn->insts[i].x64.jmp && dyn->insts[i].x64.jmp_insts==-1)
+            usefull = 1;
+        if(dyn->insts[i].x64.jmp && dyn->insts[i].x64.jmp_insts!=-1) {
+            int i2 = dyn->insts[i].x64.jmp_insts;
+            if(i2<last_fpu_used || i2>ninst) {
+                // check if some xmm/ymm/x87 stack are used in landing point
+                if(i2>ninst) {
+                    if(dyn->insts[i2].n.xmm_used || dyn->insts[i2].n.ymm_used || dyn->insts[i2].n.stack)
+                        return;
+                }
+                // we will stop there, not trying to guess too much thing
+                if((usefull && (i+1)!=ninst)) {
+                    if(BOX64ENV(dynarec_dump) || BOX64ENV(dynarec_log)>1) dynarec_log(LOG_NONE, "Putting early Float Barrier in %d for %d\n", i+1, ninst);
+                    dyn->insts[i+1].x64.barrier|=BARRIER_FLOAT;
+                }
+                return;
+            }
+            usefull = 1;
+        }
+        for(int pred=0; pred<dyn->insts[i].pred_sz; ++pred) {
+            if(dyn->insts[i].pred[pred]<=last_fpu_used) {
+                if(usefull && ((i+1)!=ninst)) {
+                    if(BOX64ENV(dynarec_dump) || BOX64ENV(dynarec_log)>1) dynarec_log(LOG_NONE, "Putting early Float Barrier in %d for %d\n", i+1, ninst);
+                    dyn->insts[i+1].x64.barrier|=BARRIER_FLOAT;
+                }
+                return;
+            }
+        }
+        if(dyn->insts[i].pred_sz>1)
+            usefull = 1;
+    }
+    if(usefull) {
+        if(BOX64ENV(dynarec_dump) || BOX64ENV(dynarec_log)>1) dynarec_log(LOG_NONE, "Putting early Float Barrier in %d for %d\n", last_fpu_used, ninst);
+        dyn->insts[last_fpu_used+1].x64.barrier|=BARRIER_FLOAT;
+    }
+}
+
+void propagateFpuBarrier(dynarec_arm_t* dyn)
+{
+    if(!dyn->use_x87)
+        return;
+    int last_fpu_used = -1;
+    for(int ninst=0; ninst<dyn->size; ++ninst) {
+        int fpu_used = dyn->insts[ninst].n.xmm_used || dyn->insts[ninst].n.ymm_used || dyn->insts[ninst].mmx_used || dyn->insts[ninst].x87_used;
+        if(fpu_used) last_fpu_used = ninst;
+        dyn->insts[ninst].fpu_used = fpu_used;
+        if(dyn->insts[ninst].fpupurge && (last_fpu_used!=-1) && (last_fpu_used!=(ninst-1))) {
+            tryEarlyFpuBarrier(dyn, last_fpu_used, ninst);
+            last_fpu_used = -1;  // reset the last_fpu_used...
+        }
+    }
+}
+
+
+void updateYmm0s(dynarec_arm_t* dyn, int ninst, int max_ninst_reached)
+{
+    if(!dyn->use_ymm)
+        return;
+    int can_incr = ninst == max_ninst_reached; // Are we the top-level call?
+    int ok = 1;
+    while ((can_incr || ok) && ninst < dyn->size) {
+        // if(dyn->need_dump) dynarec_log(LOG_NONE, "update ninst=%d (%d): can_incr=%d\n", ninst, max_ninst_reached, can_incr);
+        uint16_t new_purge_ymm, new_ymm0_in, new_ymm0_out;
+
+        if (dyn->insts[ninst].pred_sz && dyn->insts[ninst].x64.alive) {
+            // The union of the empty set is empty (0), the intersection is the universe (-1)
+            // The first instruction is the entry point, which has a virtual pred with ymm0_out = 0
+            // Similarly, float barriers reset ymm0s
+            uint16_t ymm0_union = 0;
+            uint16_t ymm0_inter = (ninst && !(dyn->insts[ninst].x64.barrier & BARRIER_FLOAT)) ? ((uint16_t)-1) : (uint16_t)0;
+            for (int i = 0; i < dyn->insts[ninst].pred_sz; ++i) {
+                int pred = dyn->insts[ninst].pred[i];
+                // if(dyn->need_dump) dynarec_log(LOG_NONE, "\twith pred[%d] = %d", i, pred);
+                if (pred >= max_ninst_reached) {
+                    // if(dyn->need_dump) dynarec_log(LOG_NONE, " (skipped)\n");
+                    continue;
+                }
+
+                int pred_out = dyn->insts[pred].x64.has_callret ? 0 : dyn->insts[pred].ymm0_out;
+                // if(dyn->need_dump) dynarec_log(LOG_NONE, " ~> %04X\n", pred_out);
+                ymm0_union |= pred_out;
+                ymm0_inter &= pred_out;
+            }
+            // if(dyn->need_dump) dynarec_log(LOG_NONE, "\t=> %04X,%04X\n", ymm0_union, ymm0_inter);
+            //  Notice the default values yield something coherent here (if all pred are after ninst)
+            new_purge_ymm = ymm0_union & ~ymm0_inter;
+            new_ymm0_in = ymm0_inter;
+            new_ymm0_out = (ymm0_inter | dyn->insts[ninst].ymm0_add) & ~dyn->insts[ninst].ymm0_sub;
+
+            if ((dyn->insts[ninst].purge_ymm != new_purge_ymm) || (dyn->insts[ninst].ymm0_in != new_ymm0_in) || (dyn->insts[ninst].ymm0_out != new_ymm0_out)) {
+                // Need to update self and next(s)
+                dyn->insts[ninst].purge_ymm = new_purge_ymm;
+                dyn->insts[ninst].ymm0_in = new_ymm0_in;
+                dyn->insts[ninst].ymm0_out = new_ymm0_out;
+
+                if (can_incr) {
+                    // We always have ninst == max_ninst_reached when can_incr == 1
+                    ++max_ninst_reached;
+                } else {
+                    // We need to stop here if the opcode has no "real" next or if we reached the ninst of the toplevel
+                    ok = (max_ninst_reached - 1 != ninst) && dyn->insts[ninst].x64.has_next && !dyn->insts[ninst].x64.has_callret;
+                }
+
+                int jmp = (dyn->insts[ninst].x64.jmp) ? dyn->insts[ninst].x64.jmp_insts : -1;
+                if ((jmp != -1) && (jmp < max_ninst_reached)) {
+                    // if(dyn->need_dump) dynarec_log(LOG_NONE, "\t! jump to %d\n", jmp);
+                    //  The jump goes before the last instruction reached, update the destination
+                    //  If this is the top level call, this means the jump goes backward (jmp != ninst)
+                    //  Otherwise, since we don't update all instructions, we may miss the update (don't use jmp < ninst)
+                    updateYmm0s(dyn, jmp, max_ninst_reached);
+                }
+            } else {
+                if (can_incr) {
+                    // We always have ninst == max_ninst_reached when can_incr == 1
+                    ++max_ninst_reached;
+
+                    // Also update jumps to before (they are skipped otherwise)
+                    int jmp = (dyn->insts[ninst].x64.jmp) ? dyn->insts[ninst].x64.jmp_insts : -1;
+                    if ((jmp != -1) && (jmp < max_ninst_reached)) {
+                        // if(dyn->need_dump) dynarec_log(LOG_NONE, "\t! jump to %d\n", jmp);
+                        updateYmm0s(dyn, jmp, max_ninst_reached);
+                    }
+                } else {
+                    // We didn't update anything, we can leave
+                    ok = 0;
+                }
+            }
+        } else if (can_incr) {
+            // We always have ninst == max_ninst_reached when can_incr == 1
+            ++max_ninst_reached;
+        } else {
+            // We didn't update anything, we can leave
+            ok = 0;
+        }
+        ++ninst;
+    }
+}
+
+
+// AVX helpers
+void avx_mark_zero(dynarec_arm_t* dyn, int ninst, int reg)
+{
+    dyn->ymm_zero |= (1<<reg);
+}
+
+int is_avx_zero(dynarec_arm_t* dyn, int ninst, int reg)
+{
+    return (dyn->ymm_zero>>reg)&1;
+}
+
+int is_avx_zero_unset(dynarec_arm_t* dyn, int ninst, int reg)
+{
+    if((dyn->ymm_zero>>reg)&1) {
+        dyn->ymm_zero &= ~(1<<reg);
+        return 1;
+    }
+    return 0;
+}
+
+void avx_mark_zero_reset(dynarec_arm_t* dyn, int ninst)
+{
+    dyn->ymm_zero = 0;
+}
+
+static int xmm_preload_reg(dynarec_arm_t* dyn, int ninst, int last, int xmm)
+{
+    return (xmm>7)?(XMM8 + xmm - 8):(XMM0 + xmm);
+}
+
+static int ymm_preload_reg(dynarec_arm_t* dyn, int ninst, int last, int ymm)
+{
+    int i = -1;
+    // search for when it will be loaded the first time
+    int start = ninst;
+    for(int ii=0; ii<32 && i==-1; ++ii)
+        if(dyn->insts[last].n.neoncache[ii].n==ymm && (dyn->insts[last].n.neoncache[ii].t==NEON_CACHE_YMMR || dyn->insts[last].n.neoncache[ii].t==NEON_CACHE_YMMW))
+            i = ii;
+    while((ninst<last) && (i!=-1)) {
+        // check if the reg is always free
+        if(!(!dyn->insts[ninst].n.neoncache[i].v ||
+            (dyn->insts[ninst].n.neoncache[i].n==ymm 
+                && (dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_YMMR || dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_YMMW))))
+            i = -1; // nope
+        else
+            ++ninst;
+    }
+    return i;
+}
+
+void addSSEPreload(dynarec_arm_t* dyn, int ninst, int i2, uint32_t preload)
+{
+    if((i2<0) || (i2>ninst) || !preload || dyn->insts[i2].preload_from)
+        return;
+    if(i2 && !dyn->insts[i2-1].x64.has_next)    // need a previous opcode to set the preload, unless it's the #0
+        return;
+    // check if there is jump out or outside this loop
+    for(int i=i2; i<ninst; ++i) {
+        if(!dyn->insts[i].x64.has_next || (dyn->insts[i].x64.barrier&BARRIER_FLOAT))
+            return;
+        // for now, unneeded stop at first jump, so this should also stop there
+        if(dyn->insts[i].x64.jmp/* && ((dyn->insts[i].x64.jmp_insts>ninst) || (dyn->insts[i].x64.jmp_insts<i2))*/)
+            return;
+    }
+    uint16_t ymm_mask = 0;
+    for(int i=0; i<16; ++i) {
+        if(preload&(1<<i)) {
+            int v = xmm_preload_reg(dyn, i2?(i2-1):0, ninst, i);
+            // propagate until not needed
+            int i3 = i2;
+            while(i3<ninst) {
+                if(dyn->insts[i3].n.xmm_used&(1<<i))
+                    i3 = ninst;
+                else {
+                    dyn->insts[i3].n.ssecache[i].write = 0;
+                    dyn->insts[i3].n.ssecache[i].reg = v;
+                    dyn->insts[i3].n.neoncache[v].t = NEON_CACHE_XMMR;
+                    dyn->insts[i3].n.neoncache[v].n = i;
+                    ++i3;
+                }
+            }
+        }
+        if(preload&(1<<(16+i))) {
+            int v = ymm_preload_reg(dyn, i2?(i2-1):0, ninst, i);
+            if(v!=-1) {
+                ymm_mask |= 1<<i;
+                // propagate until not needed
+                int i3 = i2;
+                while(i3<ninst) {
+                    if(dyn->insts[i3].n.ymm_used&(1<<i))
+                        i3 = ninst;
+                    else {
+                        dyn->insts[i3].n.neoncache[v].t = NEON_CACHE_YMMR;
+                        dyn->insts[i3].n.neoncache[v].n = i;
+                        ++i3;
+                    }
+                }
+            }
+        }
+    }
+    preload = (preload&0xffff) | (ymm_mask<<16);    // update ymm_mask
+    //TODO: should check native calls too?
+    dyn->insts[i2].preload_xmmymm = preload;
+    dyn->insts[i2].preload_from = ninst;
+    dyn->insts[i2].n.xmm_unneeded |= preload&0xffff;
+    dyn->insts[i2].n.ymm_unneeded |= ymm_mask;
 }
@@ -86,7 +86,20 @@ void fpu_reset_ninst(dynarec_native_t* dyn, int ninst);
 // is st freed
 int fpu_is_st_freed(dynarec_native_t* dyn, int ninst, int st);

+// propage FPU_BARRIER to trigger it as soon as possible (avoiding fetching an FPU reg if it's unused)
+void propagateFpuBarrier(dynarec_arm_t* dyn);
 // propage the uneeded flags on XMM/YMM regs (done between step 0 and step 1)
 void updateUneeded(dynarec_arm_t* dyn);

+void updateYmm0s(dynarec_arm_t* dyn, int ninst, int max_ninst_reached);
+
+// add a preload of SSE/AVX register preload before i2, for a jump in ninst. Might not do it if there is jump in between that goes outside this ninst-i2 interval
+void addSSEPreload(dynarec_arm_t* dyn, int ninst, int i2, uint32_t preload);
+
+// AVX helpers
+void avx_mark_zero(dynarec_arm_t* dyn, int ninst, int reg);
+int is_avx_zero(dynarec_arm_t* dyn, int ninst, int reg);
+int is_avx_zero_unset(dynarec_arm_t* dyn, int ninst, int reg);
+void avx_mark_zero_reset(dynarec_arm_t* dyn, int ninst);
+
 #endif //__DYNAREC_ARM_FUNCTIONS_H__
@@ -4,6 +4,7 @@
 // undef to get Close to SSE Float->int conversions
 //#define PRECISE_CVT

+#ifndef STEP_PASS
 #if STEP == 0
 #include "dynarec_arm64_pass0.h"
 #elif STEP == 1
@@ -13,10 +14,13 @@
 #elif STEP == 3
 #include "dynarec_arm64_pass3.h"
 #endif
+#define STEP_PASS
+#endif

 #include "debug.h"
 #include "arm64_emitter.h"
 #include "../emu/x64primop.h"
+#include "dynarec_arm64_consts.h"

 #define F8      *(uint8_t*)(addr++)
 #define F8S     *(int8_t*)(addr++)
@@ -35,6 +39,9 @@
 #ifndef FEMIT
 #define FEMIT(A)    EMIT(A)
 #endif
+#ifndef ENDPREFIX
+#define ENDPREFIX
+#endif

 //LOCK_* define
 #define LOCK_LOCK   (int*)1
@@ -70,7 +77,7 @@
        wback = 0;                                                                                                                                        \
    } else {                                                                                                                                              \
        SMREAD();                                                                                                                                         \
-        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff << (3 - rex.is32bits), rex.is32bits ? 3 : 7, rex, NULL, 0, D); \
+        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff << (3 - (rex.is32bits || rex.is67)), (rex.is32bits || rex.is67) ? 3 : 7, rex, NULL, 0, D); \
        LDz(x1, wback, fixedaddress);                                                                                                                     \
        ed = x1;                                                                                                                                          \
    }
@@ -96,28 +103,6 @@
        LDSW(x1, wback, fixedaddress);                                                                                \
        wb = ed = x1;                                                                                                 \
    }
-#define GETED32(D)                                                                                                                                     \
-    if (MODREG) {                                                                                                                                      \
-        ed = TO_NAT((nextop & 7) + (rex.b << 3));                                                                                                      \
-        wback = 0;                                                                                                                                     \
-    } else {                                                                                                                                           \
-        SMREAD();                                                                                                                                      \
-        addr = geted32(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff << (2 + rex.w), (1 << (2 + rex.w)) - 1, rex, NULL, 0, D); \
-        LDxw(x1, wback, fixedaddress);                                                                                                                 \
-        ed = x1;                                                                                                                                       \
-    }
-#define GETSED32w(D)                                                                                                    \
-    if (MODREG) {                                                                                                       \
-        ed = TO_NAT((nextop & 7) + (rex.b << 3));                                                                       \
-        SXTWx(x1, ed);                                                                                                  \
-        wb = x1;                                                                                                        \
-        wback = 0;                                                                                                      \
-    } else {                                                                                                            \
-        SMREAD();                                                                                                       \
-        addr = geted32(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff << 2, 3, rex, NULL, 0, D); \
-        LDSW(x1, wback, fixedaddress);                                                                                  \
-        wb = ed = x1;                                                                                                   \
-    }
 //GETEDH can use hint for ed, and r1 or r2 for wback (depending on hint). wback is 0 if ed is xEAX..xEDI
 #define GETEDH(hint, D)                                                                                                                                                  \
    if (MODREG) {                                                                                                                                                        \
@@ -129,16 +114,6 @@
        LDxw(hint, wback, fixedaddress);                                                                                                                                 \
        ed = hint;                                                                                                                                                       \
    }
-#define GETED32H(hint, D)                                                                                                                                                  \
-    if (MODREG) {                                                                                                                                                          \
-        ed = TO_NAT((nextop & 7) + (rex.b << 3));                                                                                                                          \
-        wback = 0;                                                                                                                                                         \
-    } else {                                                                                                                                                               \
-        SMREAD();                                                                                                                                                          \
-        addr = geted32(dyn, addr, ninst, nextop, &wback, (hint == x2) ? x1 : x2, &fixedaddress, &unscaled, 0xfff << (2 + rex.w), (1 << (2 + rex.w)) - 1, rex, NULL, 0, D); \
-        LDxw(hint, wback, fixedaddress);                                                                                                                                   \
-        ed = hint;                                                                                                                                                         \
-    }
 //GETEDW can use hint for wback and ret for ed. wback is 0 if ed is xEAX..xEDI
 #define GETEDW(hint, ret, D)                                                                                                                           \
    if (MODREG) {                                                                                                                                      \
@@ -151,73 +126,19 @@
        ed = ret;                                                                                                                                      \
        LDxw(ed, wback, fixedaddress);                                                                                                                 \
    }
-#define GETED32W(hint, ret, D)                                                                                                                           \
-    if (MODREG) {                                                                                                                                        \
-        ed = TO_NAT((nextop & 7) + (rex.b << 3));                                                                                                        \
-        MOVxw_REG(ret, ed);                                                                                                                              \
-        wback = 0;                                                                                                                                       \
-    } else {                                                                                                                                             \
-        SMREAD();                                                                                                                                        \
-        addr = geted32(dyn, addr, ninst, nextop, &wback, hint, &fixedaddress, &unscaled, 0xfff << (2 + rex.w), (1 << (2 + rex.w)) - 1, rex, NULL, 0, D); \
-        ed = ret;                                                                                                                                        \
-        LDxw(ed, wback, fixedaddress);                                                                                                                   \
-    }
 // Write back ed in wback (if wback not 0)
 #define WBACK       if(wback) {STxw(ed, wback, fixedaddress); SMWRITE();}
 // Write back ed in wback (if wback not 0)
 #define WBACKx      if(wback) {STx(ed, wback, fixedaddress); SMWRITE();}
 // Write back ed in wback (if wback not 0)
 #define WBACKw      if(wback) {STW(ed, wback, fixedaddress); SMWRITE();}
-//GETEDO can use r1 for ed, and r2 for wback. wback is 0 if ed is xEAX..xEDI
-#define GETEDO(O, D)                                                                                    \
-    if (MODREG) {                                                                                       \
-        ed = TO_NAT((nextop & 7) + (rex.b << 3));                                                       \
-        wback = 0;                                                                                      \
-    } else {                                                                                            \
-        SMREAD();                                                                                       \
-        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, NULL, 0, D); \
-        LDRxw_REGz(x1, O, wback);                                                                       \
-        ed = x1;                                                                                        \
-    }
-#define WBACKO(O)   if(wback) {STRxw_REGz(ed, O, wback); SMWRITE2();}
-//GETEDOx can use r1 for ed, and r2 for wback. wback is 0 if ed is xEAX..xEDI
-#define GETEDOx(O, D)                                                                                   \
-    if (MODREG) {                                                                                       \
-        ed = TO_NAT((nextop & 7) + (rex.b << 3));                                                       \
-        wback = 0;                                                                                      \
-    } else {                                                                                            \
-        SMREAD();                                                                                       \
-        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, NULL, 0, D); \
-        LDRx_REGz(x1, O, wback);                                                                        \
-        ed = x1;                                                                                        \
-    }
-//GETEDOz can use r1 for ed, and r2 for wback. wback is 0 if ed is xEAX..xEDI
-#define GETEDOz(O, D)                                                                                   \
-    if (MODREG) {                                                                                       \
-        ed = TO_NAT((nextop & 7) + (rex.b << 3));                                                       \
-        wback = 0;                                                                                      \
-    } else {                                                                                            \
-        SMREAD();                                                                                       \
-        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, NULL, 0, D); \
-        LDRz_REGz(x1, O, wback);                                                                        \
-        ed = x1;                                                                                        \
-    }
-#define GETSEDOw(O, D)                                                                                  \
-    if (MODREG) {                                                                                       \
-        ed = TO_NAT((nextop & 7) + (rex.b << 3));                                                       \
-        SXTWx(x1, ed);                                                                                  \
-        wb = x1;                                                                                        \
-        wback = 0;                                                                                      \
-    } else {                                                                                            \
-        SMREAD();                                                                                       \
-        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, NULL, 0, D); \
-        LDRSW_REGz(x1, O, wback);                                                                       \
-        wb = ed = x1;                                                                                   \
-    }
 //FAKEELike GETED, but doesn't get anything
-#define FAKEED                                   \
-    if (!MODREG) {                               \
-        addr = fakeed(dyn, addr, ninst, nextop); \
+#define FAKEED                                    \
+    if (MODREG) {                                 \
+        ed = TO_NAT((nextop & 7) + (rex.b << 3)); \
+        wback = 0;                                \
+    } else {                                      \
+        addr = fakeed(dyn, addr, ninst, nextop);  \
    }

 // GETGW extract x64 register in gd, that is i
@@ -277,22 +198,6 @@
        ed = i;                                                                                                                    \
        wb1 = 1;                                                                                                                   \
    }
-//GETEWO will use i for ed, i is also Offset, and can use r3 for wback.
-#define GETEWO(i, D)                                                                                                             \
-    if (MODREG) {                                                                                                                \
-        wback = TO_NAT((nextop & 7) + (rex.b << 3));                                                                             \
-        UXTHw(i, wback);                                                                                                         \
-        ed = i;                                                                                                                  \
-        wb1 = 0;                                                                                                                 \
-    } else {                                                                                                                     \
-        SMREAD();                                                                                                                \
-        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff << 1, (1 << 1) - 1, rex, NULL, 0, D); \
-        ADDz_REG(x3, wback, i);                                                                                                  \
-        if (wback != x3) wback = x3;                                                                                             \
-        LDH(i, wback, fixedaddress);                                                                                             \
-        wb1 = 1;                                                                                                                 \
-        ed = i;                                                                                                                  \
-    }
 //GETSEW will use i for ed, and can use r3 for wback. This is the Signed version
 #define GETSEW(i, D)                                                                                                             \
    if (MODREG) {                                                                                                                \
@@ -344,29 +249,6 @@
        wb1 = 1;                                                                                                 \
        ed = i;                                                                                                  \
    }
-//GETEBO will use i for ed, i is also Offset, and can use r3 for wback.
-#define GETEBO(i, D)                                                                                             \
-    if (MODREG) {                                                                                                \
-        if (rex.rex) {                                                                                           \
-            wback = TO_NAT((nextop & 7) + (rex.b << 3));                                                         \
-            wb2 = 0;                                                                                             \
-        } else {                                                                                                 \
-            wback = (nextop & 7);                                                                                \
-            wb2 = (wback >> 2) * 8;                                                                              \
-            wback = TO_NAT(wback & 3);                                                                           \
-        }                                                                                                        \
-        UBFXx(i, wback, wb2, 8);                                                                                 \
-        wb1 = 0;                                                                                                 \
-        ed = i;                                                                                                  \
-    } else {                                                                                                     \
-        SMREAD();                                                                                                \
-        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff, 0, rex, NULL, 0, D); \
-        ADDz_REG(x3, wback, i);                                                                                  \
-        if (wback != x3) wback = x3;                                                                             \
-        LDB(i, wback, fixedaddress);                                                                             \
-        wb1 = 1;                                                                                                 \
-        ed = i;                                                                                                  \
-    }
 //GETSEB sign extend EB, will use i for ed, and can use r3 for wback.
 #define GETSEB(i, D)                                                                                             \
    if (MODREG) {                                                                                                \
@@ -728,9 +610,9 @@
 // CALL_S will use x7 for the call address. Return value can be put in ret (unless ret is -1)
 // R0 will not be pushed/popd if ret is -2. Flags are not save/restored
 #define CALL_S(F, ret) call_c(dyn, ninst, F, x87pc, ret, 0, 0)
-// CALL_ will use x7 for the call address.
-// All regs are saved, including scratch. This is use to call internal function that should not change state
-#define CALL_I(F) call_i(dyn, ninst, F)
+// CALL_D will use x7 for the call address. Use for function having 1 or 2 double args and returning 1 double
+// All regs are saved, including xmm0-7, will move arg1/arg2 to D0, D1, return D0 to ret (unless any is -1)
+#define CALL_D(F, ret, arg1, arg2, sav1, sav2) call_d(dyn, ninst, F, ret, arg1, arg2, sav1, sav2)

 #define MARK        dyn->insts[ninst].mark = dyn->native_size
 #define GETMARK     dyn->insts[ninst].mark
@@ -942,7 +824,7 @@
 #define GEN_INVERTED_CARRY()
 #endif
 #ifndef INVERT_CARRY
-#define INVERT_CARRY(A)     if(dyn->insts[ninst].normal_carry) {if(arm64_flagm) CFINV(); else {MRS_nzcv(A); EORx_mask(A, A, 1, 35, 0); MSR_nzcv(A);}}
+#define INVERT_CARRY(A)     if(dyn->insts[ninst].normal_carry) {if(cpuext.flagm) CFINV(); else {MRS_nzcv(A); EORx_mask(A, A, 1, 35, 0); MSR_nzcv(A);}}
 #endif

 // Generate FCOM with s1 and s2 scratch regs (the VCMP is already done)
@@ -961,19 +843,19 @@

 // Generate FCOMI with s1 and s2 scratch regs (the VCMP is already done)
 #define FCOMI(s1, s2)                                                       \
-    IFX(X_OF|X_AF|X_SF|X_PEND) {                                            \
+    IFX(X_OF|X_AF|X_SF) {                                                   \
        MOV32w(s2, 0b100011010101);                                         \
        BICw_REG(xFlags, xFlags, s2);                                       \
-        IFX(X_CF|X_PF|X_ZF|X_PEND) {                                        \
+        IFX(X_CF|X_PF|X_ZF) {                                               \
            MOV32w(s2, 0b01000101);                                         \
        }                                                                   \
    } else {                                                                \
-        IFX(X_CF|X_PF|X_ZF|X_PEND) {                                        \
+        IFX(X_CF|X_PF|X_ZF) {                                               \
            MOV32w(s2, 0b01000101);                                         \
            BICw_REG(xFlags, xFlags, s2);                                   \
        }                                                                   \
    }                                                                       \
-    IFX(X_CF|X_PF|X_ZF|X_PEND) {                                            \
+    IFX(X_CF|X_PF|X_ZF) {                                                   \
        CSETw(s1, cMI); /* 1 if less than, 0 else */                        \
        /*s2 already set */     /* unordered */                             \
        CSELw(s1, s2, s1, cVS);                                             \
@@ -993,7 +875,7 @@
 #endif

 #ifndef CALLRET_RET
-#define CALLRET_RET()   NOP
+#define CALLRET_RET(A)   do {if(BOX64DRENV(dynarec_callret)>1) {NOP;}} while(0)
 #endif
 #ifndef CALLRET_GETRET
 #define CALLRET_GETRET()    (dyn->callrets?(dyn->callrets[dyn->callret_size].offs-dyn->native_size):0)
@@ -1118,68 +1000,57 @@
    x87_do_pop(dyn, ninst, scratch)
 #endif

-#define SET_DFNONE()                                        \
-    do {                                                    \
-        if (!dyn->f.dfnone) {                               \
-            STRw_U12(wZR, xEmu, offsetof(x64emu_t, df));    \
-        }                                                   \
-        if(!dyn->insts[ninst].x64.may_set) {                \
-            dyn->f.dfnone_here = 1;                         \
-            dyn->f.dfnone = 1;                              \
-        }                                                   \
-    } while(0)
+#define FORCE_DFNONE()  STRw_U12(wZR, xEmu, offsetof(x64emu_t, df))
+#define CHECK_DFNONE(N)  do {if(dyn->f==status_none_pending) {FORCE_DFNONE(); if(N) dyn->f = status_none;}} while(0)
+
+#define SET_DFNONE()                      \
+    do {                                  \
+        if (dyn->f != status_none) {      \
+            dyn->f = status_none_pending; \
+        }                                 \
+    } while (0)

 #define SET_DF(S, N)                                                                                                            \
    if ((N) != d_none) {                                                                                                        \
        MOVZw(S, (N));                                                                                                          \
        STRw_U12(S, xEmu, offsetof(x64emu_t, df));                                                                              \
-        if (dyn->f.pending == SF_PENDING && dyn->insts[ninst].x64.need_after && !(dyn->insts[ninst].x64.need_after & X_PEND)) { \
-            CALL_I(UpdateFlags);                                                                                                \
-            dyn->f.pending = SF_SET;                                                                                            \
-            SET_NODF();                                                                                                         \
-        }                                                                                                                       \
-        dyn->f.dfnone = 0;                                                                                                      \
+        dyn->f = status_set;                                                                                                    \
    } else                                                                                                                      \
        SET_DFNONE()

-#ifndef SET_NODF
-#define SET_NODF()          dyn->f.dfnone = 0
-#endif
-#define SET_DFOK()          dyn->f.dfnone = 1; dyn->f.dfnone_here=1
-
 #ifndef READFLAGS
 #define READFLAGS(A) \
-    if(((A)!=X_PEND && dyn->f.pending!=SF_SET)          \
-    && (dyn->f.pending!=SF_SET_PENDING)) {              \
-        if(dyn->f.pending!=SF_PENDING) {                \
-            LDRw_U12(x3, xEmu, offsetof(x64emu_t, df)); \
-            j64 = (GETMARKF)-(dyn->native_size);        \
-            CBZw(x3, j64);                              \
-        }                                               \
-        CALL_I(UpdateFlags);                            \
-        MARKF;                                          \
-        dyn->f.pending = SF_SET;                        \
-        SET_DFOK();                                     \
+    if((A)!=X_PEND                                          \
+    && (dyn->f==status_unk)) {                              \
+        TABLE64C(x6, const_updateflags_arm64);              \
+        BLR(x6);                                            \
+        dyn->f = status_none;                               \
    }
 #endif

+#define GRABFLAGS(A) \
+    if((A)!=X_PEND                                          \
+    && ((dyn->f==status_unk) || (dyn->f==status_set))) {    \
+        TABLE64C(x6, const_updateflags_arm64);              \
+        BLR(x6);                                            \
+        dyn->f = status_none;                               \
+    }
+
 #ifndef SETFLAGS
-#define SETFLAGS(A, B)                                                                          \
-    if(dyn->f.pending!=SF_SET                                                                   \
-    && ((B)&SF_SUB)                                                                             \
+#define SETFLAGS(A, B) do {                                                                     \
+    if (((B)&SF_SUB)                                                                            \
    && (dyn->insts[ninst].x64.gen_flags&(~(A))))                                                \
-        READFLAGS(((dyn->insts[ninst].x64.gen_flags&X_PEND)?X_ALL:dyn->insts[ninst].x64.gen_flags)&(~(A)));\
+        { GRABFLAGS(((dyn->insts[ninst].x64.gen_flags&X_PEND)?X_ALL:dyn->insts[ninst].x64.gen_flags)&(~(A))); }\
    if(dyn->insts[ninst].x64.gen_flags) switch(B) {                                             \
-        case SF_SUBSET: SET_DFNONE(); dyn->f.pending = SF_SET; break;                           \
-        case SF_SET: dyn->f.pending = SF_SET; break;                                            \
-        case SF_SET_DF: dyn->f.pending = SF_SET; dyn->f.dfnone = 1; break;                      \
-        case SF_SET_NODF: dyn->f.pending = SF_SET; dyn->f.dfnone = 0; break;                    \
-        case SF_PENDING: dyn->f.pending = SF_PENDING; break;                                    \
+        case SF_SET_DF: dyn->f = status_set; break;                                             \
+        case SF_SET_NODF: break;                                                                \
+        case SF_SUBSET:                                                                         \
        case SF_SUBSET_PENDING:                                                                 \
+        case SF_SET:                                                                            \
+        case SF_PENDING:                                                                        \
        case SF_SET_PENDING:                                                                    \
-            dyn->f.pending = (dyn->insts[ninst].x64.gen_flags&X_PEND)?SF_SET_PENDING:SF_SET;    \
-            break;                                                                              \
-    } else dyn->f.pending = SF_SET
+            SET_DFNONE(); break;                                                                \
+    } else if((B)!=SF_SET_NODF) SET_DFNONE();} while(0)
 #endif
 #ifndef JUMP
 #define JUMP(A, C) SMEND()
@@ -1195,9 +1066,12 @@
 #define UFLAG_OP12(A1, A2) if(dyn->insts[ninst].x64.gen_flags) {STRxw_U12(A1, xEmu, offsetof(x64emu_t, op1));STRxw_U12(A2, xEmu, offsetof(x64emu_t, op2));}
 #define UFLAG_RES(A) if(dyn->insts[ninst].x64.gen_flags) {STRxw_U12(A, xEmu, offsetof(x64emu_t, res));}
 #define UFLAG_DF(r, A) if(dyn->insts[ninst].x64.gen_flags) {SET_DF(r, A);}
+#ifndef UFLAG_IF
 #define UFLAG_IF if(dyn->insts[ninst].x64.gen_flags)
-#define UFLAG_IF_DF if(dyn->insts[ninst].x64.gen_flags || (dyn->insts[ninst].f_entry.dfnone && dyn->insts[ninst].f_entry.dfnone_here))
+#endif
+#ifndef UFLAG_IF2
 #define UFLAG_IF2(A) if(dyn->insts[ninst].x64.gen_flags A)
+#endif
 #ifndef DEFAULT
 #define DEFAULT      *ok = -1; BARRIER(2)
 #endif
@@ -1207,9 +1081,15 @@
 #ifndef TABLE64
 #define TABLE64(A, V)
 #endif
+#ifndef TABLE64_
+#define TABLE64_(A, V)
+#endif
 #ifndef FTABLE64
 #define FTABLE64(A, V)
 #endif
+#ifndef TABLE64C
+#define TABLE64C(A, V)
+#endif

 #define ARCH_INIT()      \
    SMSTART();           \
@@ -1222,23 +1102,40 @@
 #define GETIP_(A) MOV64x(xRIP, A)
 #else
 // put value in the Table64 even if not using it for now to avoid difference between Step2 and Step3. Needs to be optimized later...
-#define GETIP(A)                                        \
-    if(dyn->last_ip && ((A)-dyn->last_ip)<0x1000) {     \
-        uint64_t _delta_ip = (A)-dyn->last_ip;          \
-        dyn->last_ip += _delta_ip;                      \
-        if(_delta_ip) {                                 \
-            ADDx_U12(xRIP, xRIP, _delta_ip);            \
-        }                                               \
-    } else {                                            \
-        dyn->last_ip = (A);                             \
-        MOV64x(xRIP, dyn->last_ip);                     \
+#define GETIP(A)                                            \
+    if(dyn->last_ip && ((A)-dyn->last_ip)<0x1000) {         \
+        uint64_t _delta_ip = (A)-dyn->last_ip;              \
+        dyn->last_ip += _delta_ip;                          \
+        if(_delta_ip) {                                     \
+            ADDx_U12(xRIP, xRIP, _delta_ip);                \
+        }                                                   \
+    } else if(dyn->last_ip && (dyn->last_ip-(A))<0x1000) {  \
+        uint64_t _delta_ip = dyn->last_ip-(A);              \
+        dyn->last_ip -= _delta_ip;                          \
+        if(_delta_ip) {                                     \
+            SUBx_U12(xRIP, xRIP, _delta_ip);                \
+        }                                                   \
+    } else {                                                \
+        dyn->last_ip = (A);                                 \
+        if(dyn->need_reloc) {                               \
+            TABLE64(xRIP, dyn->last_ip);                    \
+        } else {                                            \
+            MOV64x(xRIP, dyn->last_ip);                     \
+        }                                                   \
    }
-#define GETIP_(A)                                       \
-    if(dyn->last_ip && ((A)-dyn->last_ip)<0x1000) {     \
-        uint64_t _delta_ip = (A)-dyn->last_ip;          \
-        if(_delta_ip) {ADDx_U12(xRIP, xRIP, _delta_ip);}\
-    } else {                                            \
-        MOV64x(xRIP, (A));                              \
+#define GETIP_(A)                                           \
+    if(dyn->last_ip && ((A)-dyn->last_ip)<0x1000) {         \
+        uint64_t _delta_ip = (A)-dyn->last_ip;              \
+        if(_delta_ip) {ADDx_U12(xRIP, xRIP, _delta_ip);}    \
+    } else if(dyn->last_ip && (dyn->last_ip-(A))<0x1000) {  \
+        uint64_t _delta_ip = dyn->last_ip-(A);              \
+        if(_delta_ip) {SUBx_U12(xRIP, xRIP, _delta_ip);}    \
+    } else {                                                \
+        if(dyn->need_reloc) {                               \
+            TABLE64(xRIP, (A));                             \
+        } else {                                            \
+            MOV64x(xRIP, (A));                              \
+        }                                                   \
    }
 #endif
 #define CLEARIP()   dyn->last_ip=0
@@ -1255,9 +1152,6 @@

 #define MODREG  ((nextop&0xC0)==0xC0)

-void arm64_epilog(void);
-void* arm64_next(x64emu_t* emu, uintptr_t addr);
-
 #ifndef STEPNAME
 #define STEPNAME3(N,M) N##M
 #define STEPNAME2(N,M) STEPNAME3(N,M)
@@ -1266,15 +1160,11 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);

 #define native_pass        STEPNAME(native_pass)

+#define updateflags_pass   STEPNAME(updateflags_pass)
+
 #define dynarec64_00       STEPNAME(dynarec64_00)
 #define dynarec64_0F       STEPNAME(dynarec64_0F)
-#define dynarec64_64       STEPNAME(dynarec64_64)
-#define dynarec64_65       STEPNAME(dynarec64_65)
 #define dynarec64_66       STEPNAME(dynarec64_66)
-#define dynarec64_67       STEPNAME(dynarec64_67)
-#define dynarec64_67_32    STEPNAME(dynarec64_67_32)
-#define dynarec64_67_AVX   STEPNAME(dynarec64_67_AVX)
-#define dynarec64_6764_32  STEPNAME(dynarec64_6764_32)
 #define dynarec64_D8       STEPNAME(dynarec64_D8)
 #define dynarec64_D9       STEPNAME(dynarec64_D9)
 #define dynarec64_DA       STEPNAME(dynarec64_DA)
@@ -1287,7 +1177,6 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 #define dynarec64_660F     STEPNAME(dynarec64_660F)
 #define dynarec64_66F20F   STEPNAME(dynarec64_66F20F)
 #define dynarec64_66F30F   STEPNAME(dynarec64_66F30F)
-#define dynarec64_6664     STEPNAME(dynarec64_6664)
 #define dynarec64_66F0     STEPNAME(dynarec64_66F0)
 #define dynarec64_F20F     STEPNAME(dynarec64_F20F)
 #define dynarec64_F30F     STEPNAME(dynarec64_F30F)
@@ -1304,15 +1193,13 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 #define dynarec64_AVX_F3_0F38   STEPNAME(dynarec64_AVX_F3_0F38)

 #define geted           STEPNAME(geted)
-#define geted32         STEPNAME(geted32)
 #define geted16         STEPNAME(geted16)
 #define jump_to_epilog  STEPNAME(jump_to_epilog)
 #define jump_to_next    STEPNAME(jump_to_next)
-#define ret_to_epilog   STEPNAME(ret_to_epilog)
-#define retn_to_epilog  STEPNAME(retn_to_epilog)
-#define iret_to_epilog  STEPNAME(iret_to_epilog)
+#define ret_to_next     STEPNAME(ret_to_next)
+#define iret_to_next    STEPNAME(iret_to_next)
 #define call_c          STEPNAME(call_c)
-#define call_i          STEPNAME(call_i)
+#define call_d          STEPNAME(call_d)
 #define call_n          STEPNAME(call_n)
 #define grab_segdata    STEPNAME(grab_segdata)
 #define emit_cmp8       STEPNAME(emit_cmp8)
@@ -1408,6 +1295,8 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 #define emit_rcr16c     STEPNAME(emit_rcr16c)
 #define emit_rcl32c     STEPNAME(emit_rcl32c)
 #define emit_rcr32c     STEPNAME(emit_rcr32c)
+#define emit_rcl32      STEPNAME(emit_rcl32)
+#define emit_rcr32      STEPNAME(emit_rcr32)
 #define emit_shrd32c    STEPNAME(emit_shrd32c)
 #define emit_shrd32     STEPNAME(emit_shrd32)
 #define emit_shld32c    STEPNAME(emit_shld32c)
@@ -1447,6 +1336,9 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 #define ymm_get_reg_empty STEPNAME(ymm_get_reg_empty)
 #define ymm_mark_zero     STEPNAME(ymm_mark_zero)
 #define fpu_get_reg_ymm   STEPNAME(fpu_get_reg_ymm)
+#define doPreload         STEPNAME(doPreload)
+#define doEnterBlock      STEPNAME(doEnterBlock)
+#define doLeaveBlock      STEPNAME(doLeaveBlock)

 #define fpu_pushcache   STEPNAME(fpu_pushcache)
 #define fpu_popcache    STEPNAME(fpu_popcache)
@@ -1470,22 +1362,18 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 uintptr_t geted(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop, uint8_t* ed, uint8_t hint, int64_t* fixaddress, int* unscaled, int absmax, uint32_t mask, rex_t rex, int* l, int s, int delta);

 /* setup r2 to address pointed by */
-uintptr_t geted32(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop, uint8_t* ed, uint8_t hint, int64_t* fixaddress, int* unscaled, int absmax, uint32_t mask, rex_t rex, int* l, int s, int delta);
-
-/* setup r2 to address pointed by */
-uintptr_t geted16(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop, uint8_t* ed, uint8_t hint, int64_t* fixaddress, int* unscaled, int absmax, uint32_t mask, int s);
+uintptr_t geted16(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop, uint8_t* ed, uint8_t hint, int64_t* fixaddress, int* unscaled, int absmax, uint32_t mask, rex_t rex, int s);


 // generic x64 helper
 void jump_to_epilog(dynarec_arm_t* dyn, uintptr_t ip, int reg, int ninst);
 void jump_to_next(dynarec_arm_t* dyn, uintptr_t ip, int reg, int ninst, int is32bits);
-void ret_to_epilog(dynarec_arm_t* dyn, uintptr_t ip, int ninst, rex_t rex);
-void retn_to_epilog(dynarec_arm_t* dyn, uintptr_t ip, int ninst, rex_t rex, int n);
-void iret_to_epilog(dynarec_arm_t* dyn, uintptr_t ip, int ninst, int is32bits, int is64bits);
-void call_c(dynarec_arm_t* dyn, int ninst, void* fnc, int reg, int ret, int saveflags, int save_reg);
-void call_i(dynarec_arm_t* dyn, int ninst, void* fnc);
+void ret_to_next(dynarec_arm_t* dyn, uintptr_t ip, int ninst, rex_t rex);
+void iret_to_next(dynarec_arm_t* dyn, uintptr_t ip, int ninst, int is32bits, int is64bits);
+void call_c(dynarec_arm_t* dyn, int ninst, arm64_consts_t fnc, int reg, int ret, int saveflags, int save_reg);
+void call_d(dynarec_arm_t* dyn, int ninst, arm64_consts_t fnc, int ret, int arg1, int arg2, int sav1, int sav2);
 void call_n(dynarec_arm_t* dyn, int ninst, void* fnc, int w);
-void grab_segdata(dynarec_arm_t* dyn, uintptr_t addr, int ninst, int reg, int segment, int modreg);
+void grab_segdata(dynarec_arm_t* dyn, uintptr_t addr, int ninst, int reg, int segment);
 void emit_cmp8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, int s5);
 void emit_cmp16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, int s5);
 void emit_cmp32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4, int s5);
@@ -1579,6 +1467,8 @@ void emit_rcl16c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int
 void emit_rcr16c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
 void emit_rcl32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, int s3, int s4);
 void emit_rcr32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, int s3, int s4);
+void emit_rcl32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4, int s5);
+void emit_rcr32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4, int s5);
 void emit_shrd32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, uint32_t c, int s3, int s4);
 void emit_shld32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, uint32_t c, int s3, int s4);
 void emit_shrd32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s5, int s3, int s4);
@@ -1628,15 +1518,17 @@ int sse_setround(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
 // purge ymm_zero mask according to purge_ymm
 void avx_purge_ymm(dynarec_arm_t* dyn, int ninst, uint16_t mask, int s1);

-void CacheTransform(dynarec_arm_t* dyn, int ninst, int cacheupd, int s1, int s2, int s3);
+void CacheTransform(dynarec_arm_t* dyn, int ninst, int cacheupd);

 void arm64_move32(dynarec_arm_t* dyn, int ninst, int reg, uint32_t val);
 void arm64_move64(dynarec_arm_t* dyn, int ninst, int reg, uint64_t val);

 #if STEP < 2
 #define CHECK_CACHE()   0
+#elif STEP == 2
+#define CHECK_CACHE()   (cacheupd = dyn->insts[ninst].cacheupd = CacheNeedsTransform(dyn, ninst))
 #else
-#define CHECK_CACHE()   (cacheupd = CacheNeedsTransform(dyn, ninst))
+#define CHECK_CACHE()   (cacheupd = dyn->insts[ninst].cacheupd)
 #endif

 #define neoncache_st_coherency STEPNAME(neoncache_st_coherency)
@@ -1690,7 +1582,7 @@ void fpu_reset_cache(dynarec_arm_t* dyn, int ninst, int reset_n);
 // propagate stack state
 void fpu_propagate_stack(dynarec_arm_t* dyn, int ninst);
 // purge the FPU cache (needs 3 scratch registers)
-void fpu_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1, int s2, int s3);
+void fpu_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1, int s2, int s3, uint32_t unneeded);
 // purge MMX cache
 void mmx_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1);
 // purge x87 cache
@@ -1713,30 +1605,29 @@ int ymm_get_reg_empty(dynarec_arm_t* dyn, int ninst, int s1, int a, int k1, int
 void ymm_mark_zero(dynarec_arm_t* dyn, int ninst, int a);
 // Get an YMM upper quad reg, while keeping up to 3 other YMM reg (-1 to no keep)
 int fpu_get_reg_ymm(dynarec_arm_t* dyn, int ninst, int t, int ymm, int k1, int k2, int k3);
+// Preload XMM/YMM regs
+void doPreload(dynarec_arm_t* dyn, int ninst);
+// Enter a block (atomic inrement of hot and in_used)
+void doEnterBlock(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
+// Leave a block (atomic decrement of in_used)
+void doLeaveBlock(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);

-uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
-uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
-uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int seg, int* ok, int* need_epilog);
-//uintptr_t dynarec64_65(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep,int* ok, int* need_epilog);
-uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
-uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
-uintptr_t dynarec64_67_32(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
-uintptr_t dynarec64_67_AVX(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, vex_t vex, int* ok, int* need_epilog);
-uintptr_t dynarec64_6764_32(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int seg, int* ok, int* need_epilog);
-uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
-uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
-uintptr_t dynarec64_DA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
-uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
-uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
-uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
-uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
-uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
-uintptr_t dynarec64_F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
+uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
+uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
+uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
+uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
+uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
+uintptr_t dynarec64_DA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
+uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
+uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
+uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
+uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
+uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
+uintptr_t dynarec64_F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
 uintptr_t dynarec64_66F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
 uintptr_t dynarec64_66F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
 uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
-uintptr_t dynarec64_6664(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int seg, int* ok, int* need_epilog);
-uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
+uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
 uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
 uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
 uintptr_t dynarec64_AVX(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, vex_t vex, int* ok, int* need_epilog);
@@ -1950,7 +1841,7 @@ uintptr_t dynarec64_AVX_F3_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip

 #define COMP_ZFSF(s1, A)                        \
    IFX(X_ZF|X_SF) {                            \
-        if(arm64_flagm) {                       \
+        if(cpuext.flagm) {                      \
            SETF##A(s1);                        \
            IFX(X_ZF) {                         \
                IFNATIVE(NF_EQ) {} else {       \
@@ -1979,6 +1870,10 @@ uintptr_t dynarec64_AVX_F3_0F38(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip
        }                                       \
    }

-#define PURGE_YMM()    avx_purge_ymm(dyn, ninst, dyn->insts[ninst+1].purge_ymm, x1)
+#define PURGE_YMM()                                                         \
+    do {                                                                    \
+        if ((ok > 0) && reset_n == -1 && dyn->insts[ninst + 1].purge_ymm)   \
+            avx_purge_ymm(dyn, ninst, dyn->insts[ninst + 1].purge_ymm, x1); \
+    } while (0)

 #endif //__DYNAREC_ARM64_HELPER_H__
@@ -7,18 +7,17 @@

 #define MESSAGE(A, ...) do {} while (0)
 #define READFLAGS(A)    \
-        dyn->insts[ninst].x64.use_flags = A; dyn->f.dfnone = 1;\
+        dyn->insts[ninst].x64.use_flags = A; if(dyn->f!=status_none_pending) dyn->f = status_none;\
        if(!BOX64ENV(dynarec_df) && (A)&X_PEND) dyn->insts[ninst].x64.use_flags = X_ALL; \
-        dyn->f.pending=SF_SET
+        dyn->f = status_none
 #define SETFLAGS(A,B)   \
        dyn->insts[ninst].x64.set_flags = A;    \
        dyn->insts[ninst].x64.state_flags = (B)&~SF_DF;  \
-        dyn->f.pending=(B)&SF_SET_PENDING;      \
-        dyn->f.dfnone=((B)&SF_SET)?(((B)==SF_SET_NODF)?0:1):0;  \
-        if(!BOX64ENV(dynarec_df)) {dyn->f.dfnone=1; dyn->f.pending=0; if((A)==SF_PENDING){printf_log(LOG_INFO, "Warning, some opcode use SF_PENDING, forcing deferedflags ON\n"); SET_BOX64ENV(dynarec_df, 1); }}
+        dyn->f=((B)&SF_SET)?(((B)==SF_SET_NODF)?dyn->f:status_none_pending):(((B)&SF_SET_PENDING)?status_set:status_none_pending);  \
+        if(!BOX64ENV(dynarec_df)) {dyn->f=status_none; if((A)==SF_PENDING){printf_log(LOG_INFO, "Warning, some opcode use SF_PENDING, forcing deferedflags ON\n"); SET_BOX64ENV(dynarec_df, 1); }}
 #define EMIT(A)         dyn->native_size+=4
 #define JUMP(A, C)         add_jump(dyn, ninst); add_next(dyn, (uintptr_t)A); SMEND(); dyn->insts[ninst].x64.jmp = A; dyn->insts[ninst].x64.jmp_cond = C; dyn->insts[ninst].x64.jmp_insts = 0
-#define BARRIER(A)      if(A!=BARRIER_MAYBE) {fpu_purgecache(dyn, ninst, 0, x1, x2, x3); dyn->insts[ninst].x64.barrier = A;} else dyn->insts[ninst].barrier_maybe = 1
+#define BARRIER(A)      if(A!=BARRIER_MAYBE) {fpu_purgecache(dyn, ninst, 0, x1, x2, x3, 0); dyn->insts[ninst].x64.barrier = A;} else dyn->insts[ninst].barrier_maybe = 1
 #define SET_HASCALLRET()    dyn->insts[ninst].x64.has_callret = 1
 #define NEW_INST \
        ++dyn->size;                            \
@@ -26,7 +25,8 @@
        dyn->n.combined1 = dyn->n.combined2 = 0;\
        dyn->n.swapped = 0; dyn->n.barrier = 0; \
        dyn->insts[ninst].f_entry = dyn->f;     \
-        if(ninst) {dyn->insts[ninst-1].x64.size = dyn->insts[ninst].x64.addr - dyn->insts[ninst-1].x64.addr;}
+        if(ninst) {dyn->insts[ninst-1].x64.size = dyn->insts[ninst].x64.addr - dyn->insts[ninst-1].x64.addr;}   \
+        AREFLAGSNEEDED()

 #define INST_EPILOG                             \
        dyn->insts[ninst].f_exit = dyn->f;      \
@@ -39,25 +39,17 @@
    --dyn->size;                                                                                                              \
    *ok = -1;                                                                                                                 \
    if (ninst) { dyn->insts[ninst - 1].x64.size = ip - dyn->insts[ninst - 1].x64.addr; }                                      \
-    if (BOX64ENV(dynarec_log) >= LOG_INFO || dyn->need_dump || BOX64ENV(dynarec_missing) == 1)                        \
-        if (!dyn->size || BOX64ENV(dynarec_log) > LOG_INFO || dyn->need_dump) {                                       \
-            dynarec_log(LOG_NONE, "%p: Dynarec stopped because of %s Opcode ", (void*)ip, rex.is32bits ? "x86" : "x64");      \
-            zydis_dec_t* dec = rex.is32bits ? my_context->dec32 : my_context->dec;                                            \
-            if (dec) {                                                                                                        \
-                dynarec_log_prefix(0, LOG_NONE, "%s", DecodeX64Trace(dec, dyn->insts[ninst].x64.addr, 1));                    \
-            } else {                                                                                                          \
-                dynarec_log_prefix(0, LOG_NONE, "%02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X", \
-                    PKip(0), PKip(1), PKip(2), PKip(3), PKip(4), PKip(5), PKip(6), PKip(7), PKip(8), PKip(9),                 \
-                    PKip(10), PKip(11), PKip(12), PKip(13), PKip(14));                                                        \
-            }                                                                                                                 \
-            PrintFunctionAddr(ip, " => ");                                                                                    \
-            dynarec_log_prefix(0, LOG_NONE, "\n");                                                                            \
+    if (BOX64ENV(dynarec_log) >= LOG_INFO || dyn->need_dump || BOX64ENV(dynarec_missing) == 1)                                \
+        if (!dyn->size || BOX64ENV(dynarec_log) > LOG_INFO || dyn->need_dump) {                                               \
+            dynarec_stopped(dyn->insts[ninst].x64.addr, rex.is32bits);                                                        \
        }


 #define FEMIT(A)        dyn->insts[ninst].nat_flags_op = dyn->insts[ninst].x64.set_flags?NAT_FLAG_OP_TOUCH:NAT_FLAG_OP_UNUSABLE
 #define IFNATIVE(A)     if(mark_natflag(dyn, ninst, A, 0))
 #define IFNATIVEN(A)    if(mark_natflag(dyn, ninst, A, 0))
+#define UFLAG_IF if(dyn->insts[ninst].x64.set_flags)
+#define UFLAG_IF2(A) if(dyn->insts[ninst].x64.set_flags A)
 #define IFX(A)  if((dyn->insts[ninst].x64.set_flags&(A)))
 #define IFX2(A, B)  if((dyn->insts[ninst].x64.set_flags&(A)) B)
 #define IFX_PENDOR0  if((dyn->insts[ninst].x64.set_flags&(X_PEND) || !dyn->insts[ninst].x64.set_flags))
@@ -2,20 +2,17 @@
 #define FINI
 #define MESSAGE(A, ...) do {} while (0)
 #define EMIT(A) do {} while (0)
-#define SET_NODF()      \
-        if(!dyn->insts[ninst].x64.gen_flags && !dyn->insts[ninst].x64.use_flags)   \
-                propagate_nodf(dyn, ninst);     \
-        dyn->f.dfnone = 0
+
 #define SETFLAGS(A,B)                                                   \
        dyn->insts[ninst].x64.set_flags = A;                            \
        dyn->insts[ninst].x64.state_flags = (B)&~SF_DF;                 \
-        dyn->f.pending=(B)&SF_SET_PENDING;                              \
-        dyn->f.dfnone=((B)&SF_SET)?(((B)==SF_SET_NODF)?0:1):0;          \
-        if(!BOX64ENV(dynarec_df)) {dyn->f.dfnone=1; dyn->f.pending=0; }
+        dyn->f=((B)&SF_SET)?(((B)==SF_SET_NODF)?dyn->f:status_none_pending):(((B)&SF_SET_PENDING)?status_set:status_none_pending);  \
+        if(!BOX64ENV(dynarec_df)) {dyn->f = status_none; }
 #define NEW_INST                                \
        dyn->insts[ninst].f_entry = dyn->f;     \
        dyn->n.combined1 = dyn->n.combined2 = 0;\
-        dyn->n.swapped = 0; dyn->n.barrier = 0
+        dyn->n.swapped = 0; dyn->n.barrier = 0; \
+        AREFLAGSNEEDED()

 #define INST_EPILOG                             \
        dyn->insts[ninst].n = dyn->n;           \
@@ -7,15 +7,32 @@

 #define MESSAGE(A, ...) do {} while (0)
 #define EMIT(A)     do{dyn->insts[ninst].size+=4; dyn->native_size+=4;}while(0)
+#define ENDPREFIX   dyn->prefixsize = dyn->native_size; dyn->insts[ninst].address = dyn->native_size; dyn->insts[ninst].size = 0
 #define NEW_INST                                                                                        \
        if(ninst) {                                                                                     \
                dyn->insts[ninst].address = (dyn->insts[ninst-1].address+dyn->insts[ninst-1].size);     \
                dyn->insts_size += 1+((dyn->insts[ninst-1].x64.size>(dyn->insts[ninst-1].size/4))?dyn->insts[ninst-1].x64.size:(dyn->insts[ninst-1].size/4))/15; \
-                dyn->insts[ninst].ymm0_pass2 = dyn->ymm_zero;   \
-        }
+                dyn->insts[ninst].ymm0_pass2 = dyn->ymm_zero;                                           \
+        }                                                                                               \
+        AREFLAGSNEEDED()
 #define INST_EPILOG dyn->insts[ninst].epilog = dyn->native_size; 
 #define INST_NAME(name) 
-#define TABLE64(A, V)   {Table64(dyn, (V), 2); EMIT(0);}
-#define FTABLE64(A, V)  {mmx87_regs_t v = {.d = V}; Table64(dyn, v.q, 2); EMIT(0);}
-#define CALLRET_RET()   do {dyn->callrets[dyn->callret_size].type = 0; dyn->callrets[dyn->callret_size++].offs = dyn->native_size; EMIT(ARCH_NOP); } while(0)
-#define CALLRET_LOOP()   do {dyn->callrets[dyn->callret_size].type = 1; dyn->callrets[dyn->callret_size++].offs = dyn->native_size; EMIT(ARCH_NOP); } while(0)
+#define TABLE64(A, V)   do {                                                    \
+                if(dyn->need_reloc)                                             \
+                        AddRelocTable64Addr(dyn, ninst, (V), 2);                \
+                Table64(dyn, (V), 2); EMIT(0);                                  \
+        } while(0)
+#define TABLE64_(A, V)  do {                                                    \
+                Table64(dyn, (V), 2); EMIT(0);                                  \
+        } while(0)
+#define TABLE64C(A, V)  do {                                                    \
+                if(dyn->need_reloc)                                             \
+                        AddRelocTable64Const(dyn, ninst, (V), 2);               \
+                Table64(dyn, getConst(V), 2); EMIT(0);                          \
+        } while(0)
+#define FTABLE64(A, V)  do { mmx87_regs_t v = {.d = V}; Table64(dyn, v.q, 2); EMIT(0); } while(0)
+#define CALLRET_RET(A)   do {                                                                                                                                                   \
+                if((A) && ISSEP() && BOX64DRENV(dynarec_callret)) { dyn->insts[ninst].size+=sizeof(void*); dyn->native_size+=sizeof(void*); ++dyn->sep_size; }                  \
+                if(BOX64DRENV(dynarec_callret)>1) {dyn->callrets[dyn->callret_size].type = 0; dyn->callrets[dyn->callret_size++].offs = dyn->native_size; EMIT(ARCH_NOP);}      \
+        } while(0)
+#define CALLRET_LOOP()   do { dyn->callrets[dyn->callret_size].type = 1; dyn->callrets[dyn->callret_size++].offs = dyn->native_size; EMIT(ARCH_NOP); } while(0)
@@ -16,15 +16,48 @@
    do {                                                                  \
        if (dyn->need_dump) dynarec_log(LOG_NONE, __VA_ARGS__); \
    } while (0)
+#define ENDPREFIX   dyn->insts[ninst].size2 = 0
 #define NEW_INST        \
-    if(ninst) {                                                  \
+    if(ninst) {                                                                                             \
        if(dyn->insts[ninst].address!=(uintptr_t)dyn->block-(uintptr_t)dyn->native_start) dyn->abort = 1;   \
        addInst(dyn->instsize, &dyn->insts_size, dyn->insts[ninst-1].x64.size, dyn->insts[ninst-1].size/4); \
-        dyn->insts[ninst].ymm0_pass3 = dyn->ymm_zero;   \
-    }
+        dyn->insts[ninst].ymm0_pass3 = dyn->ymm_zero;                                                       \
+    }                                                                                                       \
+    AREFLAGSNEEDED()
 #define INST_EPILOG
 #define INST_NAME(name) inst_name_pass3(dyn, ninst, name, rex)
-#define TABLE64(A, V)   {int val64offset = Table64(dyn, (V), 3); MESSAGE(LOG_DUMP, "  Table64: 0x%lx\n", (V)); LDRx_literal(A, val64offset);}
-#define FTABLE64(A, V)  {mmx87_regs_t v = {.d = V}; int val64offset = Table64(dyn, v.q, 3); MESSAGE(LOG_DUMP, "  FTable64: %g\n", v.d); VLDR64_literal(A, val64offset);}
-#define CALLRET_RET()   do {dyn->callrets[dyn->callret_size].type = 0; dyn->callrets[dyn->callret_size++].offs = dyn->native_size; EMIT(ARCH_NOP); } while(0)
+#define TABLE64(A, V)  do {                                                                 \
+                if(dyn->need_reloc)                                                         \
+                        AddRelocTable64Addr(dyn, ninst, (V), 3);                            \
+                int val64offset = Table64(dyn, (V), 3);                                     \
+                MESSAGE(LOG_DUMP, "  Table64: 0x%lx\n", (V)); LDRx_literal(A, val64offset); \
+            } while(0)
+#define TABLE64_(A, V)   {                                                                  \
+                int val64offset = Table64(dyn, (V), 3);                                     \
+                MESSAGE(LOG_DUMP, "  Table64: 0x%lx\n", (V)); LDRx_literal(A, val64offset); \
+            }
+#define TABLE64C(A, V) do {                                                                 \
+                if(dyn->need_reloc)                                                         \
+                        AddRelocTable64Const(dyn, ninst, (V), 3);                           \
+                int val64offset = Table64(dyn, getConst(V), 3);                             \
+                MESSAGE(LOG_DUMP, "  Table64C: 0x%lx\n", (V)); LDRx_literal(A, val64offset);\
+            } while(0)
+#define FTABLE64(A, V)  do {mmx87_regs_t v = {.d = V}; int val64offset = Table64(dyn, v.q, 3); MESSAGE(LOG_DUMP, "  FTable64: %g\n", v.d); VLDR64_literal(A, val64offset);} while(0)
+#define CALLRET_RET(A)                                                          \
+    do {                                                                        \
+        if((A) && ISSEP() && BOX64DRENV(dynarec_callret)) {                     \
+            MESSAGE(LOG_DUMP, "   Dynablock*\n");                               \
+            dyn->block += sizeof(void*);                                        \
+            dyn->native_size+=sizeof(void*);                                    \
+            dyn->insts[ninst].size2 += sizeof(void*);                           \
+            dyn->sep[dyn->sep_size].x64_offs = addr - dyn->start;               \
+            dyn->sep[dyn->sep_size].nat_offs =  dyn->native_size;               \
+            ++dyn->sep_size;                                                    \
+        }                                                                       \
+        if(BOX64DRENV(dynarec_callret)>1) {                                     \
+            dyn->callrets[dyn->callret_size].type = 0;                          \
+            dyn->callrets[dyn->callret_size++].offs = dyn->native_size;         \
+            EMIT(ARCH_NOP);                                                     \
+        }                                                                       \
+    } while(0)
 #define CALLRET_LOOP()   do {dyn->callrets[dyn->callret_size].type = 1; dyn->callrets[dyn->callret_size++].offs = dyn->native_size; EMIT(ARCH_NOP); } while(0)
@@ -6,6 +6,7 @@
 typedef struct x64emu_s x64emu_t;
 typedef struct dynablock_s dynablock_t;
 typedef struct instsize_s instsize_t;
+typedef struct box64env_s box64env_t;

 #define BARRIER_MAYBE   8

@@ -48,6 +49,8 @@ typedef union sse_cache_s {
    };
 } sse_cache_t;
 typedef struct callret_s callret_t;
+typedef struct sep_s sep_t;
+
 typedef struct neoncache_s {
    // Neon cache
    neon_cache_t        neoncache[32];
@@ -72,7 +75,6 @@ typedef struct neoncache_s {
    int8_t              x87stack;       // cache stack counter
    int8_t              mmxcount;       // number of mmx register used (not both mmx and x87 at the same time)
    int8_t              fpu_scratch;    // scratch counter
-    int8_t              fpu_reg;        // x87/sse/mmx reg counter
    uint16_t            xmm_write;      // 1bit of xmmXX removed write
    uint16_t            xmm_removed;    // 1bit if xmmXX was removed
    uint16_t            xmm_used;       // mask of the xmm regs used in this opcode
@@ -84,11 +86,16 @@ typedef struct neoncache_s {
    uint64_t            ymm_regs;       // 4bits (0-15) position of 16 ymmXX regs removed
 } neoncache_t;

-typedef struct flagcache_s {
+typedef enum flagcache_s {
+    status_unk = 0,         // unknown deferred flags status
+    status_set,             // deferred flags set to something (not 0)
+    status_none_pending,    // deferred flags set to 0, but still pending the write to x64emu_t
+    status_none,            // deferred flags set to 0, written to x64emu_t
+} flagcache_t;
+/*typedef struct flagcache_s {
    int                 pending;    // is there a pending flags here, or to check?
    uint8_t             dfnone;     // if deferred flags is already set to df_none
-    uint8_t             dfnone_here;// defered flags is cleared in this opcode
-} flagcache_t;
+} flagcache_t;*/

 typedef struct instruction_arm64_s {
    instruction_x64_t   x64;
@@ -116,7 +123,6 @@ typedef struct instruction_arm64_s {
    uint8_t             will_read:1;     // [strongmem] will read from memory
    uint8_t             last_write:1;    // [strongmem] the last write in a SEQ
    uint8_t             lock:1;          // [strongmem] lock semantic
-    uint8_t             lock_prefixed:1; // [strongmem] the opcode is lock prefixed
    uint8_t             wfe:1;        // opcode uses sevl + wfe
    uint8_t             set_nat_flags;  // 0 or combinaison of native flags define
    uint8_t             use_nat_flags;  // 0 or combinaison of native flags define
@@ -132,9 +138,16 @@ typedef struct instruction_arm64_s {
    unsigned            df_notneeded:1;
    unsigned            unaligned:1;    // this opcode can be re-generated for unaligned special case
    unsigned            x87precision:1; // this opcode can handle x87pc
+    unsigned            mmx_used:1; // no fine tracking, just a global "any reg used"
+    unsigned            x87_used:1; // no fine tracking, just a global "any reg used"
+    unsigned            fpu_used:1; // any xmm/ymm/x87/mmx reg used
+    unsigned            fpupurge:1;   // this opcode will purge all fpu regs
    flagcache_t         f_exit;     // flags status at end of instruction
    neoncache_t         n;          // neoncache at end of instruction (but before poping)
    flagcache_t         f_entry;    // flags status before the instruction begin
+    int                 cacheupd;
+    uint32_t            preload_xmmymm;
+    int                 preload_from;
 } instruction_arm64_t;

 typedef struct dynarec_arm_s {
@@ -142,7 +155,9 @@ typedef struct dynarec_arm_s {
    int32_t             size;
    int32_t             cap;
    uintptr_t           start;      // start of the block
+    uintptr_t           end;        // maximum end of the block (only used in pass0)
    uint32_t            isize;      // size in bytes of x64 instructions included
+    uint32_t            prefixsize; // size in byte of the prefix of the block
    void*               block;      // memory pointer where next instruction is emitted
    uintptr_t           native_start;  // start of the arm code
    size_t              native_size;   // size of emitted arm code
@@ -165,7 +180,9 @@ typedef struct dynarec_arm_s {
    instsize_t*         instsize;
    size_t              insts_size; // size of the instruction size array (calculated)
    int                 callret_size;   // size of the array
-    callret_t*          callrets;   // arrey of callret return, with NOP / UDF depending if the block is clean or dirty
+    int                 sep_size;   // size of the array
+    callret_t*          callrets;   // array of callret return, with NOP / UDF depending if the block is clean or dirty
+    sep_t*              sep;        // array of secondary entry point
    uintptr_t           forward;    // address of the last end of code while testing forward
    uintptr_t           forward_to; // address of the next jump to (to check if everything is ok)
    int32_t             forward_size;   // size at the forward point
@@ -176,9 +193,19 @@ typedef struct dynarec_arm_s {
    uint8_t             doublepop;
    uint8_t             always_test;
    uint8_t             abort;      // abort the creation of the block
+    uint8_t             use_x87:1;  // set if x87 regs are used
+    uint8_t             use_mmx:1;
+    uint8_t             use_xmm:1;
+    uint8_t             use_ymm:1;
+    uint8_t             have_purge:1;   // set to 1 if block can be purged
+    uint8_t             is_file_mapped:1;   // if the memory is a mapped file (probably binary, not a memory)
    void*               gdbjit_block;
    uint32_t            need_x87check;  // needs x87 precision control check if non-null, or 0 if not
    uint32_t            need_dump;     // need to dump the block
+    int                 need_reloc; // does the dynablock need relocations
+    int                 reloc_size;
+    uint32_t*           relocs;
+    box64env_t*         env;
 } dynarec_arm_t;

 void add_next(dynarec_arm_t *dyn, uintptr_t addr);
@@ -187,20 +214,20 @@ void add_jump(dynarec_arm_t *dyn, int ninst);
 int get_first_jump(dynarec_arm_t *dyn, int next);
 int get_first_jump_addr(dynarec_arm_t *dyn, uintptr_t next);
 int is_nops(dynarec_arm_t *dyn, uintptr_t addr, int n);
-int is_instructions(dynarec_arm_t *dyn, uintptr_t addr, int n);

+int isTable64(dynarec_arm_t *dyn, uint64_t val); // return 1 if val already in Table64
 int Table64(dynarec_arm_t *dyn, uint64_t val, int pass);  // add a value to table64 (if needed) and gives back the imm19 to use in LDR_literal

 void CreateJmpNext(void* addr, void* next);

-#define GO_TRACE(A, B, s0)  \
-    GETIP(addr);            \
-    MOVx_REG(x1, xRIP);     \
-    MRS_nzcv(s0);           \
-    STORE_XEMU_CALL(xRIP);  \
-    MOV32w(x2, B);          \
-    CALL_(A, -1, s0);       \
-    MSR_nzcv(s0);           \
+#define GO_TRACE(A, B, s0)      \
+    GETIP(addr);                \
+    MOVx_REG(x1, xRIP);         \
+    MRS_nzcv(s0);               \
+    STORE_XEMU_CALL(xRIP);      \
+    MOV32w(x2, B);              \
+    CALL_(const_##A, -1, s0);   \
+    MSR_nzcv(s0);               \
    LOAD_XEMU_CALL(xRIP)

 #endif //__DYNAREC_ARM_PRIVATE_H_
@@ -0,0 +1,134 @@
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <errno.h>
+#include <string.h>
+
+#include "os.h"
+#include "debug.h"
+#include "box64context.h"
+#include "custommem.h"
+#include "box64cpu.h"
+#include "emu/x64emu_private.h"
+#include "x64emu.h"
+#include "box64stack.h"
+#include "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynablock.h"
+#include "../dynablock_private.h"
+
+#include "dynarec_native.h"
+#include "../dynarec_arch.h"
+
+void updateflags_pass0(dynarec_arm_t* dyn, uint64_t jmp_df[]);
+void updateflags_pass1(dynarec_arm_t* dyn, uint64_t jmp_df[]);
+void updateflags_pass2(dynarec_arm_t* dyn, uint64_t jmp_df[]);
+void updateflags_pass3(dynarec_arm_t* dyn, uint64_t jmp_df[]);
+
+static dynablock_t* updaflags_arm64 = NULL;
+
+static uint8_t dummy_code[] = {0x90, 0xc3}; // some dummy code so update_flags dynablock point to something
+
+void* create_updateflags()
+{
+    if(updaflags_arm64)
+        return updaflags_arm64->block;
+    uint64_t jmp_df[d_unknown+1] = {0};
+    dynarec_arm_t helper = {0};
+    instruction_arm64_t insts[1] = {0};
+    helper.insts = insts;
+    helper.need_dump = BOX64ENV(dynarec_dump);
+    helper.cap = 1;
+    helper.f = status_none;
+    helper.insts[0].x64.gen_flags = X_ALL;
+    // pass 0
+    updateflags_pass0(&helper, jmp_df);
+    // check if all flags are handled
+    helper.have_purge = 0;  // force no purge
+    int ok = 1;
+    for(int i=d_none; i<d_unknown; ++i)
+        if(!jmp_df[i]) {
+            printf_log(LOG_NONE, "Error, UpdateFlags case %d is not handled, will crash later\n", i);
+            ok = 0;
+        }
+    // pass 1
+    updateflags_pass1(&helper, jmp_df);
+    // pass 2
+    helper.native_size = 0;
+    updateflags_pass2(&helper, jmp_df);
+    // alloc memory for pass3
+    size_t native_size = (helper.native_size+7)&~7;   // round the size...
+    size_t sz = sizeof(void*) + native_size + helper.table64size*sizeof(uint64_t) + 4*sizeof(void*) +  0  +  0  +  0  + sizeof(dynablock_t) + 0;
+    //           dynablock_t*     block (arm insts)            table64               jmpnext code instsize arch callrets     dynablock      relocs
+    void* actual_p = (void*)AllocDynarecMap((uintptr_t)&dummy_code, sz, 1);   // arbitrary address
+    void* p = (void*)(((uintptr_t)actual_p) + sizeof(void*));
+    void* tablestart = p + native_size;
+    void* next = tablestart + helper.table64size*sizeof(uint64_t);
+    void* instsize = next + 4*sizeof(void*);
+    void* arch = instsize + 0;
+    void* callrets = arch + 0;
+    if(actual_p==NULL) {
+        dynarec_log(LOG_INFO, "AllocDynarecMap(%zu) failed, canceling UpdateBlock\n", sz);
+        return NULL;
+    }
+    helper.block = p;
+    dynablock_t* block = (dynablock_t*)(callrets+0);
+    memset(block, 0, sizeof(dynablock_t));
+    void* relocs = helper.need_reloc?(block+1):NULL;
+    // fill the block
+    block->x64_addr = &dummy_code;
+    block->isize = 0;
+    block->actual_block = actual_p;
+    helper.relocs = relocs;
+    block->relocs = relocs;
+    block->table64size = helper.table64size;
+    helper.native_start = (uintptr_t)p;
+    helper.tablestart = (uintptr_t)tablestart;
+    helper.jmp_next = (uintptr_t)next+sizeof(void*);
+    helper.instsize = (instsize_t*)instsize;
+    *(dynablock_t**)actual_p = block;
+    helper.table64cap = helper.table64size;
+    helper.table64 = (uint64_t*)helper.tablestart;
+    helper.callrets = (callret_t*)callrets;
+    block->table64 = helper.table64;
+    helper.callret_size = 0;
+    // pass 3, emit (log emit native opcode)
+    if(helper.need_dump) {
+        dynarec_log(LOG_NONE, "%s%04d|Emitting %zu bytes for UpdateFlags", (helper.need_dump>1)?"\e[01;36m":"", GetTID(), helper.native_size);
+        PrintFunctionAddr(helper.start, " => ");
+        dynarec_log_prefix(0, LOG_NONE, "%s\n", (helper.need_dump>1)?"\e[m":"");
+    }
+    helper.native_size = 0;
+    updateflags_pass3(&helper, jmp_df);
+    helper.jmp_sz = helper.jmp_cap = 0;
+    helper.jmps = NULL;
+    // keep size of instructions for signal handling
+    block->instsize = instsize;
+    helper.table64 = NULL;
+    helper.instsize = NULL;
+    helper.predecessor = NULL;
+    block->size = sz;
+    block->isize = helper.size;
+    block->block = p;
+    block->jmpnext = next+sizeof(void*);
+    block->always_test = helper.always_test;
+    block->dirty = block->always_test;
+    block->is32bits = 0;
+    block->relocsize = helper.reloc_size*sizeof(uint32_t);
+    block->arch = NULL;
+    block->arch_size = 0;
+    block->callret_size = helper.callret_size;
+    block->callrets = helper.callrets;
+    block->native_size = native_size;
+    *(dynablock_t**)next = block;
+    *(void**)(next+3*sizeof(void*)) = NULL;
+    CreateJmpNext(block->jmpnext, next+3*sizeof(void*));
+    ClearCache(block->jmpnext, 4*sizeof(void*));
+    block->x64_size = 0;
+    // all done...
+    ClearCache(actual_p+sizeof(void*), native_size);   // need to clear the cache before execution...
+
+    updaflags_arm64 = block;
+    return block->block;
+}
@@ -1,6 +1,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <errno.h>
+#include <signal.h>

 #include "os.h"
 #include "debug.h"
@@ -45,6 +46,9 @@ dynablock_t* InvalidDynablock(dynablock_t* db, int need_lock)
        dynarec_log(LOG_DEBUG, "InvalidDynablock(%p), db->block=%p x64=%p:%p already gone=%d\n", db, db->block, db->x64_addr, db->x64_addr+db->x64_size-1, db->gone);
        // remove jumptable without waiting
        setJumpTableDefault64(db->x64_addr);
+        for(int i=0; i<db->sep_size; ++i)
+            if(db->sep[i].active)
+                setJumpTableDefault64(db->x64_addr+db->sep[i].x64_offs);
        if(need_lock)
            mutex_lock(&my_context->mutex_dyndump);
        db->done = 0;
@@ -54,14 +58,14 @@ dynablock_t* InvalidDynablock(dynablock_t* db, int need_lock)
        if(db->callret_size) {
            // mark all callrets to UDF
            for(int i=0; i<db->callret_size; ++i)
-                *(uint32_t*)(db->block+db->callrets[i].offs) = ARCH_NOP;
+                *(uint32_t*)(db->block+db->callrets[i].offs) = ARCH_UDF;
            ClearCache(db->block, db->size);
        }
        #endif
-        if(db_size && my_context) {
+        if(db_size && my_context && !BOX64ENV(dynarec_dirty)) {
            uint32_t n = rb_dec(my_context->db_sizes, db_size, db_size+1);
            if(!n && (db_size >= my_context->max_db_size)) {
-                my_context->max_db_size = rb_get_righter(my_context->db_sizes);
+                my_context->max_db_size = rb_get_rightmost(my_context->db_sizes);
                dynarec_log(LOG_INFO, "BOX64 Dynarec: lower max_db=%d\n", my_context->max_db_size);
            }
        }
@@ -71,29 +75,71 @@ dynablock_t* InvalidDynablock(dynablock_t* db, int need_lock)
    return db;
 }

+dynablock_t* SwitchDynablock(dynablock_t* db, int need_lock)
+{
+    if(db) {
+        if(!db->done || !db->previous || !db->previous->gone)
+            return NULL; // not a correct block!
+        dynarec_log(LOG_DEBUG, "SwitchDynablock(%p/%p), db->block=%p->%p x64=%p:%p->%p hash->%x->%x\n", db, db->previous, db->block, db->previous->block, db->x64_addr, db->x64_addr+db->x64_size-1, db->previous->x64_addr+db->previous->x64_size-1, db->hash, db->previous->hash);
+        // remove jumptable without waiting
+        dynablock_t* db_new = db->previous;
+        db->previous = NULL;
+        if(need_lock)
+            mutex_lock(&my_context->mutex_dyndump);
+        InvalidDynablock(db, 0);
+        db_new->done = 1;
+        db_new->gone = 0;
+        db_new->previous = db;
+        #ifdef ARCH_NOP
+        if(db_new->callret_size) {
+            // mark all callrets to UDF
+            for(int i=0; i<db_new->callret_size; ++i)
+                *(uint32_t*)(db_new->block+db_new->callrets[i].offs) = ARCH_NOP;
+            ClearCache(db_new->block, db_new->size);
+        }
+        #endif
+        if(need_lock)
+            mutex_unlock(&my_context->mutex_dyndump);
+        return db_new;
+    }
+    return db;
+}
+
 void FreeInvalidDynablock(dynablock_t* db, int need_lock)
 {
    if(db) {
        if(!db->gone)
            return; // already in the process of deletion!
        dynarec_log(LOG_DEBUG, "FreeInvalidDynablock(%p), db->block=%p x64=%p:%p already gone=%d\n", db, db->block, db->x64_addr, db->x64_addr+db->x64_size-1, db->gone);
+        uintptr_t db_size = db->x64_size;
        if(need_lock)
            mutex_lock(&my_context->mutex_dyndump);
-        FreeDynarecMap((uintptr_t)db->actual_block);
-        customFree(db);
+        if(db_size && my_context && BOX64ENV(dynarec_dirty)) {
+            uint32_t n = rb_dec(my_context->db_sizes, db_size, db_size+1);
+            if(!n && (db_size >= my_context->max_db_size)) {
+                my_context->max_db_size = rb_get_rightmost(my_context->db_sizes);
+                dynarec_log(LOG_INFO, "BOX64 Dynarec: lower max_db=%d\n", my_context->max_db_size);
+            }
+        }
+        FreeDynarecMap((uintptr_t)db->actual_block);    // will also free db
        if(need_lock)
            mutex_unlock(&my_context->mutex_dyndump);
    }
 }

-void FreeDynablock(dynablock_t* db, int need_lock)
+void FreeDynablock(dynablock_t* db, int need_lock, int need_remove)
 {
    if(db) {
        if(db->gone)
            return; // already in the process of deletion!
        dynarec_log(LOG_DEBUG, "FreeDynablock(%p), db->block=%p x64=%p:%p already gone=%d\n", db, db->block, db->x64_addr, db->x64_addr+db->x64_size-1, db->gone);
        // remove jumptable without waiting
-        setJumpTableDefault64(db->x64_addr);
+        if(need_remove) {
+            setJumpTableDefault64(db->x64_addr);
+            for(int i=0; i<db->sep_size; ++i)
+                if(db->sep[i].active)
+                    setJumpTableDefault64(db->x64_addr+db->sep[i].x64_offs);
+        }
        if(need_lock)
            mutex_lock(&my_context->mutex_dyndump);
        dynarec_log(LOG_DEBUG, " -- FreeDyrecMap(%p, %d)\n", db->actual_block, db->size);
@@ -103,14 +149,13 @@ void FreeDynablock(dynablock_t* db, int need_lock)
        if(db_size && my_context) {
            uint32_t n = rb_dec(my_context->db_sizes, db_size, db_size+1);
            if(!n && (db_size >= my_context->max_db_size)) {
-                my_context->max_db_size = rb_get_righter(my_context->db_sizes);
+                my_context->max_db_size = rb_get_rightmost(my_context->db_sizes);
                dynarec_log(LOG_INFO, "BOX64 Dynarec: lower max_db=%d\n", my_context->max_db_size);
            }
        }
        if(db->previous)
            FreeInvalidDynablock(db->previous, 0);
-        FreeDynarecMap((uintptr_t)db->actual_block);
-        customFree(db);
+        FreeDynarecMap((uintptr_t)db->actual_block);    // will also free db
        if(need_lock)
            mutex_unlock(&my_context->mutex_dyndump);
    }
@@ -122,6 +167,9 @@ void MarkDynablock(dynablock_t* db)
 {
    if(db) {
        dynarec_log(LOG_DEBUG, "MarkDynablock %p %p-%p\n", db, db->x64_addr, db->x64_addr+db->x64_size-1);
+        for(int i=0; i<db->sep_size; ++i)
+            if(db->sep[i].active)
+                setJumpTableIfRef64(db->x64_addr+db->sep[i].x64_offs, db->jmpnext, db->block+db->sep[i].nat_offs);
        if(!setJumpTableIfRef64(db->x64_addr, db->jmpnext, db->block)) {
            dynablock_t* old = db;
            db = getDB((uintptr_t)old->x64_addr);
@@ -135,14 +183,16 @@ void MarkDynablock(dynablock_t* db)
                else
                    db->previous = old;
            }
-        } 
        #ifdef ARCH_NOP
-        else if(db->callret_size) {
-            // mark all callrets to UDF
-            for(int i=0; i<db->callret_size; ++i)
-                *(uint32_t*)(db->block+db->callrets[i].offs) = ARCH_UDF;
-        }
+        } else {
+            if(db->callret_size) {
+                // mark all callrets to UDF
+                for(int i=0; i<db->callret_size; ++i)
+                    *(uint32_t*)(db->block+db->callrets[i].offs) = ARCH_UDF;
+                ClearCache(db->block, db->size);
+            }
        #endif
+        }
    }
 }

@@ -170,37 +220,30 @@ int FreeRangeDynablock(dynablock_t* db, uintptr_t addr, uintptr_t size)

    int need_lock = my_context?1:0;
    if(IntervalIntersects((uintptr_t)db->x64_addr, (uintptr_t)db->x64_addr+db->x64_size-1, addr, addr+size+1)) {
-        FreeDynablock(db, need_lock);
+        FreeDynablock(db, need_lock, 1);
        return 0;
    }
    return 1;
 }

-dynablock_t *AddNewDynablock(uintptr_t addr)
+void dynablock_leave_runtime(dynablock_t* db)
 {
-    dynablock_t* block;
-    // create and add new block
-    dynarec_log(LOG_VERBOSE, "Ask for DynaRec Block creation @%p\n", (void*)addr);
-    block = (dynablock_t*)customCalloc(1, sizeof(dynablock_t));
-    return block;
-}
-
-NEW_JUMPBUFF(dynarec_jmpbuf);
-
-void cancelFillBlock()
-{
-    LongJmp(GET_JUMPBUFF(dynarec_jmpbuf), 1);
+    if(!db) return;
+    if(!db->tick) return;
+    __atomic_fetch_sub(&db->in_used, 1, __ATOMIC_ACQ_REL);
 }

 /* 
    return NULL if block is not found / cannot be created. 
    Don't create if create==0
 */
-static dynablock_t* internalDBGetBlock(x64emu_t* emu, uintptr_t addr, uintptr_t filladdr, int create, int need_lock, int is32bits)
+static dynablock_t* internalDBGetBlock(x64emu_t* emu, uintptr_t addr, uintptr_t filladdr, int create, int need_lock, int is32bits, int is_new)
 {
    if (hasAlternate((void*)filladdr))
        return NULL;
    const uint32_t req_prot = (box64_pagesize==4096)?(PROT_EXEC|PROT_READ):PROT_READ;
+    if(BOX64ENV(nodynarec_delay) && (addr>=BOX64ENV(nodynarec_start)) && (addr<BOX64ENV(nodynarec_end)))
+        return NULL;
    dynablock_t* block = getDB(addr);
    if(block || !create) {
        if(block && getNeedTest(addr) && (getProtection(addr)&req_prot)!=req_prot)
@@ -208,53 +251,63 @@ static dynablock_t* internalDBGetBlock(x64emu_t* emu, uintptr_t addr, uintptr_t
        return block;
    }

+    #ifndef WIN32
+    static int critical_filled = 0;
+    static sigset_t critical_prot = {0};
+    sigset_t old_sig = {0};
+    if(!critical_filled) {
+        critical_filled = 1;
+        sigfillset(&critical_prot);
+        sigdelset(&critical_prot, SIGSEGV);
+        sigdelset(&critical_prot, SIGILL);
+        sigdelset(&critical_prot, SIGBUS);
+        sigdelset(&critical_prot, SIGINT);
+        sigdelset(&critical_prot, SIGABRT);
+        sigdelset(&critical_prot, SIGFPE);
+    }
+    #endif
+
+    pthread_sigmask(SIG_BLOCK, &critical_prot, &old_sig);
    if(need_lock) {
        if(BOX64ENV(dynarec_wait)) {
            mutex_lock(&my_context->mutex_dyndump);
        } else {
-            if(mutex_trylock(&my_context->mutex_dyndump))   // FillBlock not available for now
+            if(mutex_trylock(&my_context->mutex_dyndump)) {   // FillBlock not available for now
+                pthread_sigmask(SIG_SETMASK, &old_sig, NULL);
                return NULL;
+            }
        }
        block = getDB(addr);    // just in case
        if(block) {
            if(block && getNeedTest(addr) && (getProtection_fast(addr)&req_prot)!=req_prot)
                block = NULL;
            mutex_unlock(&my_context->mutex_dyndump);
+            pthread_sigmask(SIG_SETMASK, &old_sig, NULL);
            return block;
        }
    }
 #ifndef _WIN32
-    if((getProtection_fast(addr)&req_prot)!=req_prot) {// cannot be run, get out of the Dynarec
+    if((getProtection_fast(filladdr)&req_prot)!=req_prot) {// cannot be run, get out of the Dynarec
        if(need_lock)
            mutex_unlock(&my_context->mutex_dyndump);
+        pthread_sigmask(SIG_SETMASK, &old_sig, NULL);
        return NULL;
    }
 #endif
-    block = AddNewDynablock(addr);
-
-    // fill the block
-    block->x64_addr = (void*)addr;
-    if (SigSetJmp(GET_JUMPBUFF(dynarec_jmpbuf), 1)) {
-        printf_log(LOG_INFO, "FillBlock at %p triggered a segfault, canceling\n", (void*)addr);
-        FreeDynablock(block, 0);
-        if(need_lock)
-            mutex_unlock(&my_context->mutex_dyndump);
-        return NULL;
-    }
-    void* ret = FillBlock64(block, filladdr, (addr==filladdr)?0:1, is32bits, MAX_INSTS);
-    if(!ret) {
+    block = FillBlock64(filladdr, (addr==filladdr)?0:1, is32bits, MAX_INSTS, is_new);
+    if(!block) {
        dynarec_log(LOG_DEBUG, "Fillblock of block %p for %p returned an error\n", block, (void*)addr);
-        customFree(block);
-        block = NULL;
    }
    // check size
    if(block) {
        // fill-in jumptable
-        if(!addJumpTableIfDefault64(block->x64_addr, block->dirty?block->jmpnext:block->block)) {
-            FreeDynablock(block, 0);
+        if(!addJumpTableIfDefault64(block->x64_addr, (block->dirty || block->always_test)?block->jmpnext:block->block)) {
+            FreeDynablock(block, 0, 0);
            block = getDB(addr);
            MarkDynablock(block);   // just in case...
        } else {
+            if(block->dirty)
+                block->dirty = 0;
            if(block->x64_size) {
                if(block->x64_size>my_context->max_db_size) {
                    my_context->max_db_size = block->x64_size;
@@ -263,10 +316,19 @@ static dynablock_t* internalDBGetBlock(x64emu_t* emu, uintptr_t addr, uintptr_t
                block->done = 1;    // don't validate the block if the size is null, but keep the block
                rb_inc(my_context->db_sizes, block->x64_size, block->x64_size+1);
            }
+            for(int i=0; i<block->sep_size; ++i) {
+                uintptr_t x64_offs = block->sep[i].x64_offs;
+                uintptr_t nat_offs = block->sep[i].nat_offs;
+                if(addJumpTableIfDefault64(block->x64_addr+x64_offs, (block->dirty || block->always_test)?block->jmpnext:(block->block+nat_offs)))
+                    block->sep[i].active = 1;
+                else
+                    block->sep[i].active = 0;
+            }
        }
    }
    if(need_lock)
        mutex_unlock(&my_context->mutex_dyndump);
+    pthread_sigmask(SIG_SETMASK, &old_sig, NULL);

    dynarec_log(LOG_DEBUG, "%04d| --- DynaRec Block %p created @%p:%p (%p, 0x%x bytes)\n", GetTID(), block, (void*)addr, (void*)(addr+((block)?block->x64_size:1)-1), (block)?block->block:0, (block)?block->size:0);

@@ -278,34 +340,63 @@ dynablock_t* DBGetBlock(x64emu_t* emu, uintptr_t addr, int create, int is32bits)
    int is_inhotpage = isInHotPage(addr);
    if(is_inhotpage && !BOX64ENV(dynarec_dirty))
        return NULL;
-    dynablock_t *db = internalDBGetBlock(emu, addr, addr, create, 1, is32bits);
+    dynablock_t *db = internalDBGetBlock(emu, addr, addr, create, 1, is32bits, 1);
    if(db && db->done && db->block && getNeedTest(addr)) {
-        if (db->always_test) SchedYield(); // just calm down...
+        //if (db->always_test) SchedYield(); // just calm down...
        uint32_t hash = X31_hash_code(db->x64_addr, db->x64_size);
-        if(is_inhotpage && hash!=db->hash)
-            return NULL;    // will be handle when hotpage is over
-        int need_lock = mutex_trylock(&my_context->mutex_dyndump);
+        mutex_lock(&my_context->mutex_dyndump)?1:0;
        if(hash!=db->hash) {
+            if(is_inhotpage && db->previous) {
+                // check alternate
+                if(db->previous && !db->dirty && X31_hash_code(db->previous->x64_addr, db->previous->x64_size)==db->previous->hash) {
+                    db = SwitchDynablock(db, 0);
+                    if(!addJumpTableIfDefault64(db->x64_addr, (db->always_test)?db->jmpnext:db->block)) {
+                        FreeDynablock(db, 0, 0);
+                        db = getDB(addr);
+                        MarkDynablock(db);   // just in case...
+                    } else {
+                        for(int i=0; i<db->sep_size; ++i) {
+                            uint32_t x64_offs = db->sep[i].x64_offs;
+                            uint32_t nat_offs = db->sep[i].nat_offs;
+                            if(addJumpTableIfDefault64(db->x64_addr+x64_offs, (db->always_test)?db->jmpnext:(db->block+nat_offs)))
+                                db->sep[i].active = 1;
+                            else
+                                db->sep[i].active = 0;
+                        }
+                    }
+                    mutex_unlock(&my_context->mutex_dyndump);
+                    return db;
+                }
+                mutex_unlock(&my_context->mutex_dyndump);
+                dynarec_log(LOG_DEBUG, "Cannot run block (or previous) %p from %p:%p (hash:%X/%X, always_test:%d, previous=%p/hash=%X) for %p\n", db, db->x64_addr, db->x64_addr+db->x64_size-1, hash, db->hash, db->always_test,db->previous, db->previous?db->previous->hash:0,(void*)addr);
+                return NULL;    // will be handle when hotpage is over
+            }
            db->done = 0;   // invalidating the block
-            dynarec_log(LOG_DEBUG, "Invalidating block %p from %p:%p (hash:%X/%X, always_test:%d) for %p\n", db, db->x64_addr, db->x64_addr+db->x64_size-1, hash, db->hash, db->always_test,(void*)addr);
+            dynarec_log(LOG_DEBUG, "Invalidating block %p from %p:%p (hash:%X/%X, always_test:%d, previous=%p/hash=%X) for %p\n", db, db->x64_addr, db->x64_addr+db->x64_size-1, hash, db->hash, db->always_test,db->previous, db->previous?db->previous->hash:0,(void*)addr);
            // Free db, it's now invalid!
-            dynablock_t* old = InvalidDynablock(db, need_lock);
+            dynablock_t* old = InvalidDynablock(db, 0);
            // start again... (will create a new block)
-            db = internalDBGetBlock(emu, addr, addr, create, need_lock, is32bits);
+            db = internalDBGetBlock(emu, addr, addr, create, 0, is32bits, 0);
            if(db) {
                if(db->previous)
-                    FreeInvalidDynablock(db->previous, need_lock);
+                    FreeInvalidDynablock(db->previous, 0);
                db->previous = old;
            } else
-                FreeInvalidDynablock(old, need_lock);
+                FreeInvalidDynablock(old, 0);
        } else {
            if(is_inhotpage) {
+                db->always_test = 2;
                // log?
            } else {
                dynarec_log(LOG_DEBUG, "Validating block %p from %p:%p (hash:%X, always_test:%d) for %p\n", db, db->x64_addr, db->x64_addr+db->x64_size-1, db->hash, db->always_test, (void*)addr);
-                if(db->always_test)
-                    protectDB((uintptr_t)db->x64_addr, db->x64_size);
-                else {
+                if(db->always_test) {
+                    if(db->always_test==2) {
+                        db->always_test = 0;
+                        protectDB((uintptr_t)db->x64_addr, db->x64_size);
+                    }
+                    // always_test==1 (NEVERCLEAN): skip protectDB, the page has mixed
+                    // code+data and mprotect would strip writability from data regions
+                } else {
                    #ifdef ARCH_NOP
                    if(db->callret_size) {
                        // mark all callrets to UDF
@@ -315,11 +406,19 @@ dynablock_t* DBGetBlock(x64emu_t* emu, uintptr_t addr, int create, int is32bits)
                    }
                    #endif
                    protectDBJumpTable((uintptr_t)db->x64_addr, db->x64_size, db->block, db->jmpnext);
+                    if(!db->always_test)
+                        for(int i=0; i<db->sep_size; ++i) {
+                            uint32_t x64_offs = db->sep[i].x64_offs;
+                            uint32_t nat_offs = db->sep[i].nat_offs;
+                            if(setJumpTableIfRef64(db->x64_addr+x64_offs, (db->always_test)?db->jmpnext:(db->block+nat_offs), db->jmpnext))
+                                db->sep[i].active = 1;
+                            else
+                                db->sep[i].active = 0;
+                        }
                }
            }
        }
-        if(!need_lock)
-            mutex_unlock(&my_context->mutex_dyndump);
+        mutex_unlock(&my_context->mutex_dyndump);
    } 
    if(!db || !db->block || !db->done)
        emu->test.test = 0;
@@ -330,24 +429,24 @@ dynablock_t* DBAlternateBlock(x64emu_t* emu, uintptr_t addr, uintptr_t filladdr,
 {
    dynarec_log(LOG_DEBUG, "Creating AlternateBlock at %p for %p%s\n", (void*)addr, (void*)filladdr, is32bits?" 32bits":"");
    int create = 1;
-    dynablock_t *db = internalDBGetBlock(emu, addr, filladdr, create, 1, is32bits);
-    if(db && db->done && db->block && getNeedTest(filladdr)) {
+    dynablock_t *db = internalDBGetBlock(emu, addr, filladdr, create, 1, is32bits, 1);
+    if(db && db->done && db->block && (db->dirty || getNeedTest(filladdr))) {
        if (db->always_test) SchedYield(); // just calm down...
-        int need_lock = mutex_trylock(&my_context->mutex_dyndump);
+        mutex_lock(&my_context->mutex_dyndump);
        uint32_t hash = X31_hash_code(db->x64_addr, db->x64_size);
        if(hash!=db->hash) {
            db->done = 0;   // invalidating the block
            dynarec_log(LOG_DEBUG, "Invalidating alt block %p from %p:%p (hash:%X/%X) for %p\n", db, db->x64_addr, db->x64_addr+db->x64_size, hash, db->hash, (void*)addr);
            // Free db, it's now invalid!
-            dynablock_t* old = InvalidDynablock(db, need_lock);
+            dynablock_t* old = InvalidDynablock(db, 0);
            // start again... (will create a new block)
-            db = internalDBGetBlock(emu, addr, filladdr, create, need_lock, is32bits);
+            db = internalDBGetBlock(emu, addr, filladdr, create, 0, is32bits, 0);
            if(db) {
                if(db->previous)
-                    FreeInvalidDynablock(db->previous, need_lock);
+                    FreeInvalidDynablock(db->previous, 0);
                db->previous = old;
            } else
-                FreeInvalidDynablock(old, need_lock);
+                FreeInvalidDynablock(old, 0);
        } else {
            if(db->always_test)
                protectDB((uintptr_t)db->x64_addr, db->x64_size);
@@ -361,10 +460,17 @@ dynablock_t* DBAlternateBlock(x64emu_t* emu, uintptr_t addr, uintptr_t filladdr,
                }
                #endif
                protectDBJumpTable((uintptr_t)db->x64_addr, db->x64_size, db->block, db->jmpnext);
+                for(int i=0; i<db->sep_size; ++i) {
+                    uint32_t x64_offs = db->sep[i].x64_offs;
+                    uint32_t nat_offs = db->sep[i].nat_offs;
+                    if(addJumpTableIfDefault64(db->x64_addr+x64_offs, (db->always_test)?db->jmpnext:(db->block+nat_offs)))
+                        db->sep[i].active = 1;
+                    else
+                        db->sep[i].active = 0;
+                }
            }
        }
-        if(!need_lock)
-            mutex_unlock(&my_context->mutex_dyndump);
+        mutex_unlock(&my_context->mutex_dyndump);
    } 
    if(!db || !db->block || !db->done)
        emu->test.test = 0;
@@ -373,10 +479,15 @@ dynablock_t* DBAlternateBlock(x64emu_t* emu, uintptr_t addr, uintptr_t filladdr,

 uintptr_t getX64Address(dynablock_t* db, uintptr_t native_addr)
 {
-    uintptr_t x64addr = (uintptr_t)db->x64_addr;
-    uintptr_t armaddr = (uintptr_t)db->block;
    if ((native_addr < (uintptr_t)db->block) || (native_addr > (uintptr_t)db->actual_block + db->size))
        return 0;
+    uintptr_t x64addr = (uintptr_t)db->x64_addr;
+    uintptr_t armaddr = (uintptr_t)db->block;
+    if (!db->isize) return x64addr;
+    if(native_addr<(uintptr_t)db->block+db->prefixsize)
+        // issue indide the prefix, return as the 1st opcode
+        return x64addr;
+    armaddr += db->prefixsize;
    int i = 0;
    do {
        int x64sz = 0;
@@ -402,6 +513,7 @@ int getX64AddressInst(dynablock_t* db, uintptr_t x64pc)
    int ret = 0;
    if (x64pc < (uintptr_t)db->x64_addr || x64pc > (uintptr_t)db->x64_addr + db->x64_size)
        return -1;
+    armaddr += db->prefixsize;
    int i = 0;
    do {
        int x64sz = 0;
@@ -420,3 +532,29 @@ int getX64AddressInst(dynablock_t* db, uintptr_t x64pc)
    } while (db->instsize[i].x64 || db->instsize[i].nat);
    return ret;
 }
+
+uintptr_t getX64InstAddress(dynablock_t* db, int inst)
+{
+    uintptr_t x64addr = (uintptr_t)db->x64_addr;
+    uintptr_t armaddr = (uintptr_t)db->block + db->prefixsize;
+    if (inst < 0 || inst > db->isize)
+        return (uintptr_t)-1LL;
+    int i = 0;
+    int ret = 0;
+    do {
+        if (inst == ret)
+            return x64addr;
+        int x64sz = 0;
+        int armsz = 0;
+        do {
+            x64sz += db->instsize[i].x64;
+            armsz += db->instsize[i].nat * 4;
+            ++i;
+        } while ((db->instsize[i - 1].x64 == 15) || (db->instsize[i - 1].nat == 15));
+        // if the opcode is a NOP on ARM side (so armsz==0), it cannot be an address to find
+        armaddr += armsz;
+        x64addr += x64sz;
+        ret++;
+    } while (db->instsize[i].x64 || db->instsize[i].nat);
+    return (uintptr_t)-1LL;
+}
@@ -11,26 +11,42 @@ typedef struct callret_s {
    uint32_t    type:1;
 } callret_t;

+typedef struct sep_s {
+    uint32_t    active:1;
+    uint32_t    nat_offs:31;
+    uint32_t    x64_offs;
+} sep_t;
+
 typedef struct dynablock_s {
    void*           block;  // block-sizeof(void*) == self
    void*           actual_block;   // the actual start of the block (so block-sizeof(void*))
    struct dynablock_s*    previous;   // a previous block that might need to be freed
+    uint32_t        in_used;// will be 0 if not in_used, >0 if used be some code
+    uint32_t        tick;    // last "tick" when dynablock was run
    void*           x64_addr;
    uintptr_t       x64_size;
+    size_t          native_size;
+    uint32_t        prefixsize; // size in byte of the prefix of the block
    int             size;
    uint32_t        hash;
    uint8_t         done;
    uint8_t         gone;
    uint8_t         dirty;      // if need to be tested as soon as it's created
-    uint8_t         always_test:1;
+    uint8_t         always_test:2;
    uint8_t         is32bits:1;
    int             callret_size;   // size of the array
    int             isize;
-    size_t          arch_size;  // size of of arch dependant infos
+    int             arch_size;  // size of of arch dependant infos
+    int             sep_size;   // SEP are the Secondary Entry Point
    instsize_t*     instsize;
    void*           arch;       // arch dependant per inst info (can be NULL)
    callret_t*      callrets;   // array of callret return, with NOP / UDF depending if the block is clean or dirty
+    sep_t*          sep;        // secondary entry point, offsets are relative to x64_addr & block
    void*           jmpnext;    // a branch jmpnext code when block is marked
+    size_t          table64size;// to check table64
+    void*           table64;    // to relocate the table64
+    size_t          relocsize;  // number of relocations (should be an int only)
+    void*           relocs;     // relocations, when block is loaded
    #ifdef GDBJIT
    void*           gdbjit_block;
    #endif
@@ -0,0 +1,243 @@
+#include <stdint.h>
+#include <string.h>
+
+#include "os.h"
+#include "dynarec_private.h"
+#include "emu/x64run_private.h"
+#include "dynablock_private.h"
+#include "dynarec_arch.h"
+#include "custommem.h"
+#include "debug.h"
+
+#define RELOC_TBL64C        1
+#define RELOC_TBL64ADDR     2
+#define RELOC_TBL64RETENDBL 3
+#define RELOC_CANCELBLOCK   4
+#define RELOC_TBL64TBLJMPH  5
+#define RELOC_TBL64TBLJMPL  6
+
+typedef union reloc_s {
+    uint8_t type;
+    struct {
+        uint8_t     type;
+        uint8_t     C;
+        uint16_t    idx;
+    } table64c;
+    struct {
+        uint8_t     type;
+        uint8_t     nothing;
+        uint16_t    idx;
+    } table64addr;
+    struct {
+        uint8_t     type;
+        uint8_t     delta;
+        uint16_t    idx;
+    } table64retendbl;
+    struct {
+        uint8_t     type;
+        uint8_t     deltah;
+        uint16_t    idx;
+    } table64jmptblh;
+    struct {
+        uint32_t    type:8;
+        uint32_t    deltal:24;
+    } table64jmptbll;
+    uint32_t x;
+} reloc_t;
+
+void AddRelocTable64Const(dynarec_native_t* dyn, int ninst, native_consts_t C, int pass)
+{
+    if(!dyn->need_reloc)
+        return;
+    if(isTable64(dyn, getConst(C)))
+        return;
+    if(pass<3)  {
+        dyn->reloc_size++;
+        return;
+    }
+    // generating the reloc
+    reloc_t reloc = {0};
+    reloc.table64c.type = RELOC_TBL64C;
+    reloc.table64c.C = C;
+    reloc.table64c.idx = dyn->table64size;
+    dyn->relocs[dyn->reloc_size++] = reloc.x;
+}
+
+void AddRelocTable64RetEndBlock(dynarec_native_t* dyn, int ninst, uintptr_t addr, int pass)
+{
+    if(!dyn->need_reloc)
+        return;
+    if(isTable64(dyn, getJumpTableAddress64(addr)))
+        return; // no need, already handled
+    // TODO: check iif addr is in current map
+    if(pass<3) {
+        dyn->reloc_size++;
+        return;
+    }
+    reloc_t reloc = {0};
+    intptr_t delta = addr-(dyn->start + dyn->isize);
+    if(delta>255 || delta<0) {
+        dynarec_log(LOG_INFO, "Warning Relocation Ret End Block not possible: delta=%lld, aborting block\n", delta);
+        dyn->abort = 1;
+    }
+    reloc.table64retendbl.type = RELOC_TBL64RETENDBL;
+    reloc.table64retendbl.delta = delta;
+    reloc.table64retendbl.idx = dyn->table64size;
+    dyn->relocs[dyn->reloc_size++] = reloc.x;
+}
+
+void AddRelocCancelBlock(dynarec_native_t* dyn, int ninst, int pass)
+{
+    if(!dyn->need_reloc)
+        return;
+    if(pass<3) {
+        dyn->reloc_size++;
+        return;
+    }
+    reloc_t reloc = {0};
+    reloc.type = RELOC_CANCELBLOCK;
+    dyn->relocs[dyn->reloc_size++] = reloc.x;
+}
+
+void AddRelocTable64Addr(dynarec_native_t* dyn, int ninst, uintptr_t addr, int pass)
+{
+    if(!dyn->need_reloc)
+        return;
+    if(isTable64(dyn, addr))
+        return; // no need, already handled
+    // check if addr is in the same mmap area
+    // else, use a cancel block instead
+    uintptr_t start;
+    int ok = 1;
+    if(!IsAddrFileMapped(dyn->start, NULL, &start))
+        ok = 0;
+    else {
+        uintptr_t end = start + SizeFileMapped(dyn->start);
+        if(addr<start || addr>=end)
+            ok = 0;
+        if(ok && ((addr-start)>=0x100000000LL))
+            ok = 0;
+    }
+    if(!ok) return AddRelocCancelBlock(dyn, ninst, pass);
+    if(pass<3) {
+        dyn->reloc_size++;
+        return;
+    }
+    reloc_t reloc = {0};
+    reloc.type = RELOC_TBL64ADDR;
+    reloc.table64addr.idx = dyn->table64size;
+    dyn->relocs[dyn->reloc_size++] = reloc.x;
+}
+
+
+void AddRelocTable64JmpTbl(dynarec_native_t* dyn, int ninst, uintptr_t addr, int pass)
+{
+    if(!dyn->need_reloc)
+        return;
+    if(isTable64(dyn, getJumpTableAddress64(addr)))
+        return; // no need, already handled
+    // check if addr is in the same mmap area
+    // else, use a cancel block instead
+    uintptr_t start;
+    int ok = 1;
+    if(!IsAddrFileMapped(dyn->start, NULL, &start))
+        ok = 0;
+    else {
+        uintptr_t end = start + SizeFileMapped(dyn->start);
+        if(addr<start || addr>=end)
+            ok = 0;
+        if(ok && ((addr-start)>=0x100000000LL))
+            ok = 0;
+    }
+    if(!ok) return AddRelocCancelBlock(dyn, ninst, pass);
+    if(pass<3) {
+        dyn->reloc_size+=2;
+        return;
+    }
+    uintptr_t delta = addr - start;
+    reloc_t reloc = {0};
+    reloc.type = RELOC_TBL64TBLJMPH;
+    reloc.table64jmptblh.idx = dyn->table64size;
+    reloc.table64jmptblh.deltah = delta>>24;
+    dyn->relocs[dyn->reloc_size++] = reloc.x;
+    reloc.type = RELOC_TBL64TBLJMPL;
+    reloc.table64jmptbll.deltal = delta&0xffffff;
+    dyn->relocs[dyn->reloc_size++] = reloc.x;
+}
+
+int ApplyRelocs(dynablock_t* block, intptr_t delta_block, intptr_t delta_map, uintptr_t mapping_start)
+{
+    if(!block || !block->relocs || !block->relocsize)
+        return 0;
+    size_t reloc_size = block->relocsize / sizeof(uint32_t);
+    reloc_t *relocs = block->relocs;
+    uint64_t *table64 = block->table64;
+    int idx;
+
+    int i = 0;
+    uintptr_t addr;
+    dynarec_log(LOG_DEBUG, "Will apply %zd reloc to dynablock starting at %p - %p\n", reloc_size, block->x64_addr, block->x64_addr + block->x64_size);
+    while(i<reloc_size) {
+        idx = -1;
+        switch(relocs[i].type) {
+            case RELOC_TBL64C:
+                idx = relocs[i].table64c.idx;
+                table64[idx] = getConst(relocs[i].table64c.C);
+                dynarec_log(LOG_DEBUG, "\tApply Relocs[%d]: TABLE64[%d]=Const:%d\n", i, idx, relocs[i].table64c.C);
+                break;
+            case RELOC_TBL64ADDR:
+                idx = relocs[i].table64addr.idx;
+                table64[idx] += delta_map;
+                dynarec_log(LOG_DEBUG, "\tApply Relocs[%d]: TABLE64[%d]=Addr in Map, delta=%zd\n", i, idx, delta_map);
+                break;
+            case RELOC_TBL64RETENDBL:
+                idx = relocs[i].table64retendbl.idx;
+                addr = (uintptr_t)block->x64_addr + block->x64_size + relocs[i].table64retendbl.delta;
+                table64[idx] = getJumpTableAddress64(addr);
+                dynarec_log(LOG_DEBUG, "\tApply Relocs[%d]: TABLE64[%d]=JmpTable64(%p)\n", i, idx, (void*)addr);
+                break;
+            case RELOC_CANCELBLOCK:
+                dynarec_log(LOG_DEBUG, "\tApply Relocs[%d]: Cancel Block\n", i);
+                block->dirty = 1;
+                block->hash = 0;
+                return 0;
+            case RELOC_TBL64TBLJMPH:
+                if(relocs[i+1].type!=RELOC_TBL64TBLJMPL)
+                    return -2;  // bad sequence
+                idx = relocs[i].table64jmptblh.idx;
+                addr = relocs[i].table64jmptblh.deltah;
+                addr = mapping_start + relocs[i+1].table64jmptbll.deltal + (addr<<24);
+                table64[idx] = getJumpTableAddress64(addr);
+                dynarec_log(LOG_DEBUG, "\tApply Relocs[%d,%d]: TABLE64[%d]=JmpTable64(%p)=%p\n", i, i+1, idx, (void*)addr, getJumpTableAddress64(addr));
+                break;
+            case RELOC_TBL64TBLJMPL:
+                break;
+            default: 
+                dynarec_log(LOG_DEBUG, "\tUnknown Relocs[%d]: %d\n", i, relocs[i].type);
+                return -1;
+        }
+        if(idx!=-1) {
+            if(idx>=block->table64size) {
+                dynarec_log(LOG_NONE, "Warning: Reloc Table64 idx out of range (%d vs %d)\n", idx, block->table64size);
+            }
+        }
+        ++i;
+    }
+    return 0;
+}
+
+int RelocsHaveCancel(dynablock_t* block)
+{
+    if(!block->relocs || !block->relocsize)
+        return 0;
+    size_t reloc_size = block->relocsize/sizeof(uint32_t);
+    for(size_t i=0; i<reloc_size; ++i)
+        if(((reloc_t*)block->relocs)[i].type == RELOC_CANCELBLOCK)
+            return 1;
+    return 0;
+}
+
+uintptr_t RelocGetNext()
+{
+    return getConst(const_native_next);
+}
@@ -0,0 +1,27 @@
+#ifndef __DYNACACHE_RELOC_H__
+#define __DYNACACHE_RELOC_H__
+
+#ifdef ARM64
+#include "dynarec/arm64/dynarec_arm64_consts.h"
+#define native_consts_t arm64_consts_t
+#elif defined(RV64)
+#include "dynarec/rv64/dynarec_rv64_consts.h"
+#define native_consts_t rv64_consts_t
+#elif defined(LA64)
+#include "dynarec/la64/dynarec_la64_consts.h"
+#define native_consts_t la64_consts_t
+#else
+#error Unsupported architecture
+#endif
+
+void AddRelocTable64Const(dynarec_native_t* dyn, int ninst, native_consts_t C, int pass);
+void AddRelocTable64Addr(dynarec_native_t* dyn, int ninst, uintptr_t addr, int pass);
+void AddRelocTable64RetEndBlock(dynarec_native_t* dyn, int ninst, uintptr_t addr, int pass);
+void AddRelocTable64JmpTbl(dynarec_native_t* dyn, int ninst, uintptr_t addr, int pass);
+void AddRelocCancelBlock(dynarec_native_t* dyn, int ninst, int pass);
+
+int ApplyRelocs(dynablock_t* block, intptr_t delta_block, intptr_t delta_map, uintptr_t mapping_start);
+int RelocsHaveCancel(dynablock_t* block);
+uintptr_t RelocGetNext();
+
+#endif
@@ -27,10 +27,12 @@
 #include "elfloader.h"
 #endif

+extern int running32bits;
 #ifdef DYNAREC
 void* LinkNext(x64emu_t* emu, uintptr_t addr, void* x2, uintptr_t* x3)
 {
    int is32bits = (R_CS == 0x23);
+    if(!running32bits && is32bits) running32bits=1;
    #ifdef HAVE_TRACE
    uintptr_t new_addr = (uintptr_t)getAlternate((void*)addr);
    if(!addr) {
@@ -43,17 +45,19 @@ void* LinkNext(x64emu_t* emu, uintptr_t addr, void* x2, uintptr_t* x3)
    } else if(emu->segs[_CS]==0x23 && addr>0x100000000LL) {
        dynablock_t* db = FindDynablockFromNativeAddress(x2-4);
        printf_log(LOG_INFO, "Warning, jumping to high address %p from %p (db=%p, x64addr=%p/%s)\n", (void*)addr, x2-4, db, db?(void*)getX64Address(db, (uintptr_t)x2-4):NULL, db?getAddrFunctionName(getX64Address(db, (uintptr_t)x2-4)):"(nil)");
+    #endif
    } else if(!memExist(new_addr)) {
        dynablock_t* db = FindDynablockFromNativeAddress(x2-4);
        printf_log(LOG_INFO, "Warning, jumping to an unmapped address %p->%p from %p (db=%p, x64addr=%p/%s)\n", (void*)new_addr, (void*)addr, x2-4, db, db?(void*)getX64Address(db, (uintptr_t)x2-4):NULL, db?getAddrFunctionName(getX64Address(db, (uintptr_t)x2-4)):"(nil)");
-    #endif
    }
+    #else   //HAVE_TRACE
+    uintptr_t new_addr = addr;
    #endif
    void * jblock;
    dynablock_t* block = NULL;
+    uintptr_t old_addr = addr;
    if(hasAlternate((void*)addr)) {
        printf_log(LOG_DEBUG, "Jmp address has alternate: %p\n", (void*)addr);
-        uintptr_t old_addr = addr;
        addr = (uintptr_t)getAlternate((void*)addr);    // set new address
        R_RIP = addr;   // but also new RIP!
        *x3 = addr; // and the RIP in x27 register
@@ -90,6 +94,17 @@ void* LinkNext(x64emu_t* emu, uintptr_t addr, void* x2, uintptr_t* x3)
        // null block, but done: go to epilog, no linker here
        return native_epilog;
    }
+    if(block->sep_size && (uintptr_t)block->x64_addr!=old_addr) {
+        jblock = NULL;
+        for(int i=0; i<block->sep_size && !jblock; ++i) {
+            if(old_addr==(uintptr_t)block->x64_addr + block->sep[i].x64_offs)
+                jblock = block->block + block->sep[i].nat_offs;
+        }
+        if(!jblock) {
+            printf_log(LOG_NONE, "Warning, cannot find Secondary Entry Point %p in dynablock %p\n", new_addr, block);
+            return native_epilog;
+        }
+    }
    //dynablock_t *father = block->father?block->father:block;
    return jblock;
 }
@@ -118,7 +133,9 @@ void DynaCall(x64emu_t* emu, uintptr_t addr)
        PushExit(emu);
    R_RIP = addr;
    emu->df = d_none;
-    DynaRun(emu);
+    if(emu->flags.quitonlongjmp)
+        emu->flags.need_jmpbuf = 1;
+    EmuRun(emu, 1);
    emu->quit = 0;  // reset Quit flags...
    emu->df = d_none;
    if(emu->flags.quitonlongjmp && emu->flags.longjmp) {
@@ -141,8 +158,7 @@ void DynaCall(x64emu_t* emu, uintptr_t addr)
    }
 }

-extern int running32bits;
-void DynaRun(x64emu_t* emu)
+void EmuRun(x64emu_t* emu, int use_dynarec)
 {
    // prepare setjump for signal handling
    JUMPBUFF jmpbuf[1] = {0};
@@ -151,7 +167,6 @@ void DynaRun(x64emu_t* emu)
    #ifdef RV64
    uintptr_t old_savesp = emu->xSPSave;
    #endif
-    emu->flags.jmpbuf_ready = 0;
    int is32bits = (emu->segs[_CS]==0x23);
    while(!(emu->quit)) {
        if(!emu->jmpbuf || (emu->flags.need_jmpbuf && emu->jmpbuf!=jmpbuf)) {
@@ -166,7 +181,7 @@ void DynaRun(x64emu_t* emu)
            if ((skip = SigSetJmp(emu->jmpbuf, 1)))
            #endif
            {
-                printf_log(LOG_DEBUG, "Setjmp DynaRun, fs=0x%x\n", emu->segs[_FS]);
+                dynarec_log(LOG_DEBUG, "Setjmp EmuRun, fs=0x%x will %sskip dynarec next\n", emu->segs[_FS], (skip==3)?"not ":"");
                #ifdef DYNAREC
                if(BOX64ENV(dynarec_test)) {
                    if(emu->test.clean)
@@ -182,7 +197,7 @@ void DynaRun(x64emu_t* emu)
            emu->flags.need_jmpbuf = 0;

 #ifdef DYNAREC
-        if(!BOX64ENV(dynarec))
+        if(!BOX64ENV(dynarec) || !use_dynarec)
 #endif
            Run(emu, 0);
 #ifdef DYNAREC
@@ -203,7 +218,7 @@ void DynaRun(x64emu_t* emu)
                    running32bits = 1;
                }
            }
-            dynablock_t* block = (skip)?NULL:DBGetBlock(emu, R_RIP, 1, is32bits);
+            dynablock_t* block = (skip || ACCESS_FLAG(F_TF))?NULL:DBGetBlock(emu, R_RIP, 1, is32bits);
            if(!block || !block->block || !block->done || ACCESS_FLAG(F_TF)) {
                skip = 0;
                // no block, of block doesn't have DynaRec content (yet, temp is not null)
@@ -223,7 +238,19 @@ void DynaRun(x64emu_t* emu)
                    CHECK_FLAGS(emu);
                }
                // block is here, let's run it!
-                native_prolog(emu, block->block);
+                void* jblock = block->block;
+                if(block->sep_size && R_RIP!=(uintptr_t)block->x64_addr) {
+                    jblock = NULL;
+                    for(int i=0; i<block->sep_size && !jblock; ++i) {
+                        if(R_RIP==(uintptr_t)block->x64_addr + block->sep[i].x64_offs)
+                            jblock = block->block + block->sep[i].nat_offs;
+                    }
+                }
+                if(!jblock) {
+                    printf_log(LOG_NONE, "Warning, cannot find Secondary Entry Point %p in dynablock %p\n", (void*)R_RIP, block);
+                    skip = 1;
+                } else
+                    native_prolog(emu, jblock);
            }
            if(emu->fork) {
                int forktype = emu->fork;
@@ -242,3 +269,8 @@ void DynaRun(x64emu_t* emu)
    emu->xSPSave = old_savesp;
    #endif
 }
+
+void DynaRun(x64emu_t *emu)
+{
+    EmuRun(emu, 1);
+}
@@ -24,7 +24,7 @@
 #define MAXBLOCK_SIZE ((1<<20)-200)

 #define RAZ_SPECIFIC(A, N)      rasNativeState(A, N)
-#define UPDATE_SPECIFICS(A)     updateNativeFlags(A)
+#define UPDATE_SPECIFICS(A)     updateYmm0s(dyn, 0, 0); updateNativeFlags(A); propagateFpuBarrier(A)
 #define PREUPDATE_SPECIFICS(A)
 #define POSTUPDATE_SPECIFICS(A) updateUneeded(A)
 #define ARCH_SIZE(A)    get_size_arch(A)
@@ -33,11 +33,14 @@
 #define STOP_NATIVE_FLAGS(A, B)   A->insts[B].nat_flags_op = NAT_FLAG_OP_UNUSABLE
 #define ARCH_UNALIGNED(A, B) arch_unaligned(A, B)
 extern uint32_t arm64_crc(void* p, uint32_t len);
-#define ARCH_CRC(A, B)  if(arm64_crc32) return arm64_crc(A, B)
+#define ARCH_CRC(A, B)  if(cpuext.crc32) return arm64_crc(A, B)
+extern void* create_updateflags();
+#define ARCH_UPDATEFLAGS()      create_updateflags()

-#define ARCH_NOP    0b11010101000000110010000000011111
-#define ARCH_UDF    0xcafe
-#elif defined(LA64)
+#define ARCH_NOP    0b11010101000000110010000000011111
+#define ARCH_UDF    0xcafe
+#define JMPNEXT_SIZE    (4*sizeof(void*))
+#elif defined(LA64)

 #define instruction_native_t        instruction_la64_t
 #define dynarec_native_t            dynarec_la64_t
@@ -56,7 +59,7 @@ extern uint32_t arm64_crc(void* p, uint32_t len);
 #define MAXBLOCK_SIZE ((1 << 27) - 200)

 #define RAZ_SPECIFIC(A, N)
-#define UPDATE_SPECIFICS(A)
+#define UPDATE_SPECIFICS(A)    propagateFpuBarrier(A)
 #define PREUPDATE_SPECIFICS(A) updateNativeFlags(A)
 #define POSTUPDATE_SPECIFICS(A)

@@ -65,10 +68,11 @@ extern uint32_t arm64_crc(void* p, uint32_t len);
 #define ARCH_ADJUST(A, B, C, D) {}
 #define STOP_NATIVE_FLAGS(A, B) {}
 #define ARCH_UNALIGNED(A, B) 0
-
-// NYI
-#define NATIVE_RESTORE_X87PC()
-#elif defined(RV64)
+extern uint32_t la64_crc(void* p, uint32_t len);
+#define ARCH_CRC(A, B)       return la64_crc(A, B)
+#define JMPNEXT_SIZE    (4*sizeof(void*))
+
+#elif defined(RV64)

 #define instruction_native_t        instruction_rv64_t
 #define dynarec_native_t            dynarec_rv64_t
@@ -98,9 +102,12 @@ extern uint32_t arm64_crc(void* p, uint32_t len);
 #define ARCH_FILL(A, B, C) populate_arch(A, B, C)
 #define ARCH_ADJUST(A, B, C, D) {}
 #define STOP_NATIVE_FLAGS(A, B) {}
-#define ARCH_UNALIGNED(A, B) arch_unaligned(A, B)
-#else
+#define ARCH_UNALIGNED(A, B) arch_unaligned(A, B)
+#define JMPNEXT_SIZE    (4*sizeof(void*))
+#else
 #error Unsupported platform
 #endif

+#include "dynacache_reloc.h"
+
 #endif //__DYNAREC_ARCH__H_
@@ -35,8 +35,9 @@
 #define STRONGMEM_SIMD_WRITE 2 // The level of SIMD memory writes will be tracked
 #define STRONGMEM_LAST_WRITE 1 // The level of a barrier before the last guest memory store will be put
 #define STRONGMEM_SEQ_WRITE  3 // The level of a barrier at every third memory store will be  put
+#define STRONGMEM_QEMU       4 // The level of a mimic to QEMU's strong memory model

-#define STRONGMEM_LEVEL() (box64_wine && VolatileRangesContains(ip) ? 0 : BOX64DRENV(dynarec_strongmem))
+#define STRONGMEM_LEVEL() ((box64_wine && VolatileRangesContains(ip)) ? 0 : BOX64DRENV(dynarec_strongmem))

 #if STEP == 1

@@ -49,10 +50,10 @@
        dyn->insts[ninst].will_write = 1;                        \
    } while (0)

-#define SMWRITELOCK(lock)              \
-    do {                               \
-        dyn->insts[ninst].lock = lock; \
-        SMWRITE();                     \
+#define SMWRITELOCK(lock)                   \
+    do {                                    \
+        if(lock) dyn->insts[ninst].lock = 1;\
+        SMWRITE();                          \
    } while (0)

 #define SMWRITE2()                                       \
@@ -96,23 +97,24 @@
 #else

 // An opcode writes guest memory, this need to be put after the STORE instruction manually. It will also end the SEQ automaticaly on last_write immediatly
-#define SMWRITE()                                                         \
-    do {                                                                  \
-        if (dyn->insts[ninst].last_write) {                               \
-            dyn->smwrite = 1;                                             \
-            SMEND();                                                      \
-        } else {                                                          \
-            /* Put a barrier at every third memory write. */              \
-            if (STRONGMEM_LEVEL() >= STRONGMEM_SEQ_WRITE) {               \
-                if (++dyn->smwrite >= 3 /* Every third memory write */) { \
-                    DMB_ISH();                                            \
-                    dyn->smwrite = 1;                                     \
-                }                                                         \
-            } else {                                                      \
-                /* Mark that current sequence writes to guest memory. */  \
-                dyn->smwrite = 1;                                         \
-            }                                                             \
-        }                                                                 \
+#define SMWRITE()                                                        \
+    do {                                                                 \
+        if (dyn->insts[ninst].last_write) {                              \
+            dyn->smwrite = 1;                                            \
+            SMEND();                                                     \
+        } else {                                                         \
+            /* Put a barrier at every third memory write. */             \
+            if (STRONGMEM_LEVEL() >= STRONGMEM_SEQ_WRITE) {              \
+                if (++dyn->smwrite >= 3 /* Every third memory write */   \
+                    || STRONGMEM_LEVEL() >= STRONGMEM_QEMU) {            \
+                    DMB_ISH();                                           \
+                    dyn->smwrite = 1;                                    \
+                }                                                        \
+            } else {                                                     \
+                /* Mark that current sequence writes to guest memory. */ \
+                dyn->smwrite = 1;                                        \
+            }                                                            \
+        }                                                                \
    } while (0)

 // Similar to SMWRITE, but checks lock.
@@ -133,7 +135,12 @@
    } while (0)

 // An opcode reads guest memory, this need to be put before the LOAD instruction manually.
-#define SMREAD()
+#define SMREAD()                                   \
+    do {                                           \
+        if (STRONGMEM_LEVEL() >= STRONGMEM_QEMU) { \
+            DMB_ISHLD();                           \
+        }                                          \
+    } while (0)

 // Similar to SMREAD, but checks lock.
 #define SMREADLOCK(lock) \
@@ -200,7 +207,17 @@
 #define SMDMB() DMB_ISH()
 #endif

+#define AREFLAGSNEEDED()    \
+    if((dyn->insts[ninst].x64.need_before&~X_PEND) && (!ninst || dyn->insts[ninst-1].x64.has_callret)) {    \
+        READFLAGS(dyn->insts[ninst].x64.need_before&~X_PEND);                                               \
+    }
+
+// Insert a Secondary Entry Point if it's asked, but only on block that are from file mapped memory (so some binary file)
+//  and that are not flagged as always_test (as this could slow down a bit, and the block is unsafe), unless SEP level 2 is asked
+#define ISSEP() (BOX64DRENV(dynarec_sep) && ((dyn->is_file_mapped && !dyn->always_test) || BOX64DRENV(dynarec_sep)>1))
+
 int is_addr_unaligned(uintptr_t addr);
+int is_addr_autosmc(uintptr_t addr);

 #ifdef ARM64
 #include "arm64/dynarec_arm64_helper.h"
@@ -7,6 +7,7 @@
 #include <sys/types.h>
 #include <unistd.h>

+#include "x64_signals.h"
 #include "os.h"
 #include "debug.h"
 #include "box64context.h"
@@ -37,7 +38,7 @@ void native_fstp(x64emu_t* emu, void* p)
 void native_print_armreg(x64emu_t* emu, uintptr_t reg, uintptr_t n)
 {
    (void)emu;
-    dynarec_log(LOG_DEBUG, "R%lu=0x%lx (%lu)\n", n, reg, reg);
+    dynarec_log(LOG_INFO, "Debug Register R%lu=0x%lx (%lu)\n", n, reg, reg);
 }

 void native_f2xm1(x64emu_t* emu)
@@ -79,24 +80,32 @@ void native_fxtract(x64emu_t* emu)
 }
 void native_fprem(x64emu_t* emu)
 {
-    int e0, e1;
-    int64_t ll;
-    frexp(ST0.d, &e0);
-    frexp(ST1.d, &e1);
-    int32_t tmp32s = e0 - e1;
-    if(tmp32s<64)
-    {
-        ll = (int64_t)floor(ST0.d/ST1.d);
-        ST0.d = ST0.d - (ST1.d*ll);
-        emu->sw.f.F87_C2 = 0;
-        emu->sw.f.F87_C1 = (ll&1)?1:0;
-        emu->sw.f.F87_C3 = (ll&2)?1:0;
-        emu->sw.f.F87_C0 = (ll&4)?1:0;
+    double x = ST0.d, y = ST1.d;
+    int64_t q = 0;
+    if (isnan(x) || isnan(y)) {
+        ST0.d = NAN;
+        q = 0;
+    } else if (isinf(x) || y == 0.0) {
+#if !defined(_WIN32) && !defined(__MINGW32__)
+        feraiseexcept(FE_INVALID);
+#endif
+        ST0.d = NAN;
+        q = 0;
    } else {
-        ll = (int64_t)(floor((ST0.d/ST1.d))/exp2(tmp32s - 32));
-        ST0.d = ST0.d - ST1.d*ll*exp2(tmp32s - 32);
-        emu->sw.f.F87_C2 = 1;
+#if defined(_WIN32) || defined(__MINGW32__)
+        q = (int64_t)trunc(x / y);
+        ST0.d = x - (y * q);
+#else
+        ST0.d = fmod(x, y);
+        q = (int64_t)trunc(x / y);
+#endif
    }
+    q &= 7;
+    emu->sw.f.F87_C2 = 0;
+    emu->sw.f.F87_C1 = q & 1;
+    emu->sw.f.F87_C3 = (q >> 1) & 1;
+    emu->sw.f.F87_C0 = (q >> 2) & 1;
+
 }
 void native_fyl2xp1(x64emu_t* emu)
 {
@@ -134,6 +143,50 @@ void native_fcos(x64emu_t* emu)
    emu->sw.f.F87_C2 = 0;
 }

+double direct_f2xm1(x64emu_t* emu, double a)
+{
+    return expm1(LN2 * a);
+}
+double direct_fyl2x(x64emu_t* emu, double a, double b)
+{
+    return log2(a)*b;
+}
+double direct_fyl2xp1(x64emu_t* emu, double a, double b)
+{
+    return log1p(a)*b/LN2;
+}
+double direct_fpatan(x64emu_t* emu, double a, double b)
+{
+#pragma STDC FENV_ACCESS ON
+    return atan2(b, a);
+}
+double direct_fsin(x64emu_t* emu, double a)
+{
+#pragma STDC FENV_ACCESS ON
+    // seems that sin of glib doesn't follow the rounding direction mode
+    emu->sw.f.F87_C2 = 0;
+    return sin(a);
+}
+double direct_fcos(x64emu_t* emu, double a)
+{
+#pragma STDC FENV_ACCESS ON
+    // seems that cos of glib doesn't follow the rounding direction mode
+    emu->sw.f.F87_C2 = 0;
+    return cos(a);
+}
+double direct_ftan(x64emu_t* emu, double a)
+{
+#pragma STDC FENV_ACCESS ON
+    // seems that tan of glib doesn't follow the rounding direction mode
+    emu->sw.f.F87_C2 = 0;
+    return tan(a);
+}
+double direct_fscale(x64emu_t* emu, double a, double b)
+{
+#pragma STDC FENV_ACCESS ON
+    return a?ldexp(a, trunc(b)):a;
+}
+
 void native_fbld(x64emu_t* emu, uint8_t* ed)
 {
    fpu_fbld(emu, ed);
@@ -186,20 +239,26 @@ void native_ud(x64emu_t* emu)
 {
    if(BOX64ENV(dynarec_test))
        emu->test.test = 0;
-    EmitSignal(emu, SIGILL, (void*)R_RIP, 0);
+    EmitSignal(emu, X64_SIGILL, (void*)R_RIP, 0);
 }

 void native_br(x64emu_t* emu)
 {
    if(BOX64ENV(dynarec_test))
        emu->test.test = 0;
-    EmitSignal(emu, SIGSEGV, (void*)R_RIP, 0xb09d);
+    EmitSignal(emu, X64_SIGSEGV, (void*)R_RIP, 0xb09d);
 }

 void native_priv(x64emu_t* emu)
 {
    emu->test.test = 0;
-    EmitSignal(emu, SIGSEGV, (void*)R_RIP, 0xbad0);
+    EmitSignal(emu, X64_SIGSEGV, (void*)R_RIP, 0xbad0);
+}
+
+void native_gpf(x64emu_t* emu)
+{
+    emu->test.test = 0;
+    EmitSignal(emu, X64_SIGSEGV, (void*)R_RIP, 0xbad0); // same effect has private opcode?
 }

 void native_int(x64emu_t* emu, int num)
@@ -216,7 +275,7 @@ void native_wineint(x64emu_t* emu, int num)
 #endif
 void native_int3(x64emu_t* emu)
 {
-    EmitSignal(emu, SIGTRAP, NULL, 3);
+    EmitSignal(emu, X64_SIGTRAP, NULL, 3);
 }

 void native_div0(x64emu_t* emu)
@@ -235,6 +294,7 @@ void native_fsave(x64emu_t* emu, uint8_t* ed)
        LD2D(p, &emu->x87[7-i].d);
        p+=10;
    }
+    reset_fpu(emu);
 }
 void native_fsave16(x64emu_t* emu, uint8_t* ed)
 {
@@ -246,6 +306,7 @@ void native_fsave16(x64emu_t* emu, uint8_t* ed)
        LD2D(p, &emu->x87[7-i].d);
        p+=10;
    }
+    reset_fpu(emu);
 }
 void native_frstor(x64emu_t* emu, uint8_t* ed)
 {
@@ -274,26 +335,171 @@ void native_frstor16(x64emu_t* emu, uint8_t* ed)

 void native_fprem1(x64emu_t* emu)
 {
-    int e0, e1;
-    int64_t ll;
-    frexp(ST0.d, &e0);
-    frexp(ST1.d, &e1);
-    int32_t tmp32s = e0 - e1;
-    if(tmp32s<64)
-    {
-        ll = (int64_t)round(ST0.d/ST1.d);
-        ST0.d = ST0.d - (ST1.d*ll);
-        emu->sw.f.F87_C2 = 0;
-        emu->sw.f.F87_C1 = (ll&1)?1:0;
-        emu->sw.f.F87_C3 = (ll&2)?1:0;
-        emu->sw.f.F87_C0 = (ll&4)?1:0;
+    double x = ST0.d, y = ST1.d;
+    int q = 0;
+    if (isnan(x) || isnan(y)) {
+        ST0.d = NAN;
+    } else if (isinf(x) || y == 0.0) {
+#if !defined(_WIN32) && !defined(__MINGW32__)
+        feraiseexcept(FE_INVALID);
+#endif
+        ST0.d = NAN;
    } else {
-        ll = (int64_t)(trunc((ST0.d/ST1.d))/exp2(tmp32s - 32));
-        ST0.d = ST0.d - ST1.d*ll*exp2(tmp32s - 32);
-        emu->sw.f.F87_C2 = 1;
+#if defined(_WIN32) || defined(__MINGW32__)
+        q = (int)round(x / y);
+        ST0.d = x - (y * q);
+#else
+        ST0.d = remquo(x, y, &q);
+#endif
    }
+    q &= 7;
+    emu->sw.f.F87_C2 = 0;
+    emu->sw.f.F87_C1 = q & 1;
+    emu->sw.f.F87_C3 = (q >> 1) & 1;
+    emu->sw.f.F87_C0 = (q >> 2) & 1;
 }

+const uint8_t ff_mult2[4][256] = {
+    // a = 0x0e
+    0x00, 0x0e, 0x1c, 0x12, 0x38, 0x36, 0x24, 0x2a, 0x70, 0x7e, 0x6c, 0x62, 0x48, 0x46, 0x54, 0x5a,
+    0xe0, 0xee, 0xfc, 0xf2, 0xd8, 0xd6, 0xc4, 0xca, 0x90, 0x9e, 0x8c, 0x82, 0xa8, 0xa6, 0xb4, 0xba,
+    0xdb, 0xd5, 0xc7, 0xc9, 0xe3, 0xed, 0xff, 0xf1, 0xab, 0xa5, 0xb7, 0xb9, 0x93, 0x9d, 0x8f, 0x81,
+    0x3b, 0x35, 0x27, 0x29, 0x03, 0x0d, 0x1f, 0x11, 0x4b, 0x45, 0x57, 0x59, 0x73, 0x7d, 0x6f, 0x61,
+    0xad, 0xa3, 0xb1, 0xbf, 0x95, 0x9b, 0x89, 0x87, 0xdd, 0xd3, 0xc1, 0xcf, 0xe5, 0xeb, 0xf9, 0xf7,
+    0x4d, 0x43, 0x51, 0x5f, 0x75, 0x7b, 0x69, 0x67, 0x3d, 0x33, 0x21, 0x2f, 0x05, 0x0b, 0x19, 0x17,
+    0x76, 0x78, 0x6a, 0x64, 0x4e, 0x40, 0x52, 0x5c, 0x06, 0x08, 0x1a, 0x14, 0x3e, 0x30, 0x22, 0x2c,
+    0x96, 0x98, 0x8a, 0x84, 0xae, 0xa0, 0xb2, 0xbc, 0xe6, 0xe8, 0xfa, 0xf4, 0xde, 0xd0, 0xc2, 0xcc,
+    0x41, 0x4f, 0x5d, 0x53, 0x79, 0x77, 0x65, 0x6b, 0x31, 0x3f, 0x2d, 0x23, 0x09, 0x07, 0x15, 0x1b,
+    0xa1, 0xaf, 0xbd, 0xb3, 0x99, 0x97, 0x85, 0x8b, 0xd1, 0xdf, 0xcd, 0xc3, 0xe9, 0xe7, 0xf5, 0xfb,
+    0x9a, 0x94, 0x86, 0x88, 0xa2, 0xac, 0xbe, 0xb0, 0xea, 0xe4, 0xf6, 0xf8, 0xd2, 0xdc, 0xce, 0xc0,
+    0x7a, 0x74, 0x66, 0x68, 0x42, 0x4c, 0x5e, 0x50, 0x0a, 0x04, 0x16, 0x18, 0x32, 0x3c, 0x2e, 0x20,
+    0xec, 0xe2, 0xf0, 0xfe, 0xd4, 0xda, 0xc8, 0xc6, 0x9c, 0x92, 0x80, 0x8e, 0xa4, 0xaa, 0xb8, 0xb6,
+    0x0c, 0x02, 0x10, 0x1e, 0x34, 0x3a, 0x28, 0x26, 0x7c, 0x72, 0x60, 0x6e, 0x44, 0x4a, 0x58, 0x56,
+    0x37, 0x39, 0x2b, 0x25, 0x0f, 0x01, 0x13, 0x1d, 0x47, 0x49, 0x5b, 0x55, 0x7f, 0x71, 0x63, 0x6d,
+    0xd7, 0xd9, 0xcb, 0xc5, 0xef, 0xe1, 0xf3, 0xfd, 0xa7, 0xa9, 0xbb, 0xb5, 0x9f, 0x91, 0x83, 0x8d,
+    // a = 0x09
+    0x00, 0x09, 0x12, 0x1b, 0x24, 0x2d, 0x36, 0x3f, 0x48, 0x41, 0x5a, 0x53, 0x6c, 0x65, 0x7e, 0x77,
+    0x90, 0x99, 0x82, 0x8b, 0xb4, 0xbd, 0xa6, 0xaf, 0xd8, 0xd1, 0xca, 0xc3, 0xfc, 0xf5, 0xee, 0xe7,
+    0x3b, 0x32, 0x29, 0x20, 0x1f, 0x16, 0x0d, 0x04, 0x73, 0x7a, 0x61, 0x68, 0x57, 0x5e, 0x45, 0x4c,
+    0xab, 0xa2, 0xb9, 0xb0, 0x8f, 0x86, 0x9d, 0x94, 0xe3, 0xea, 0xf1, 0xf8, 0xc7, 0xce, 0xd5, 0xdc,
+    0x76, 0x7f, 0x64, 0x6d, 0x52, 0x5b, 0x40, 0x49, 0x3e, 0x37, 0x2c, 0x25, 0x1a, 0x13, 0x08, 0x01,
+    0xe6, 0xef, 0xf4, 0xfd, 0xc2, 0xcb, 0xd0, 0xd9, 0xae, 0xa7, 0xbc, 0xb5, 0x8a, 0x83, 0x98, 0x91,
+    0x4d, 0x44, 0x5f, 0x56, 0x69, 0x60, 0x7b, 0x72, 0x05, 0x0c, 0x17, 0x1e, 0x21, 0x28, 0x33, 0x3a,
+    0xdd, 0xd4, 0xcf, 0xc6, 0xf9, 0xf0, 0xeb, 0xe2, 0x95, 0x9c, 0x87, 0x8e, 0xb1, 0xb8, 0xa3, 0xaa,
+    0xec, 0xe5, 0xfe, 0xf7, 0xc8, 0xc1, 0xda, 0xd3, 0xa4, 0xad, 0xb6, 0xbf, 0x80, 0x89, 0x92, 0x9b,
+    0x7c, 0x75, 0x6e, 0x67, 0x58, 0x51, 0x4a, 0x43, 0x34, 0x3d, 0x26, 0x2f, 0x10, 0x19, 0x02, 0x0b,
+    0xd7, 0xde, 0xc5, 0xcc, 0xf3, 0xfa, 0xe1, 0xe8, 0x9f, 0x96, 0x8d, 0x84, 0xbb, 0xb2, 0xa9, 0xa0,
+    0x47, 0x4e, 0x55, 0x5c, 0x63, 0x6a, 0x71, 0x78, 0x0f, 0x06, 0x1d, 0x14, 0x2b, 0x22, 0x39, 0x30,
+    0x9a, 0x93, 0x88, 0x81, 0xbe, 0xb7, 0xac, 0xa5, 0xd2, 0xdb, 0xc0, 0xc9, 0xf6, 0xff, 0xe4, 0xed,
+    0x0a, 0x03, 0x18, 0x11, 0x2e, 0x27, 0x3c, 0x35, 0x42, 0x4b, 0x50, 0x59, 0x66, 0x6f, 0x74, 0x7d,
+    0xa1, 0xa8, 0xb3, 0xba, 0x85, 0x8c, 0x97, 0x9e, 0xe9, 0xe0, 0xfb, 0xf2, 0xcd, 0xc4, 0xdf, 0xd6,
+    0x31, 0x38, 0x23, 0x2a, 0x15, 0x1c, 0x07, 0x0e, 0x79, 0x70, 0x6b, 0x62, 0x5d, 0x54, 0x4f, 0x46,
+    // a = 0x0d
+    0x00, 0x0d, 0x1a, 0x17, 0x34, 0x39, 0x2e, 0x23, 0x68, 0x65, 0x72, 0x7f, 0x5c, 0x51, 0x46, 0x4b,
+    0xd0, 0xdd, 0xca, 0xc7, 0xe4, 0xe9, 0xfe, 0xf3, 0xb8, 0xb5, 0xa2, 0xaf, 0x8c, 0x81, 0x96, 0x9b,
+    0xbb, 0xb6, 0xa1, 0xac, 0x8f, 0x82, 0x95, 0x98, 0xd3, 0xde, 0xc9, 0xc4, 0xe7, 0xea, 0xfd, 0xf0,
+    0x6b, 0x66, 0x71, 0x7c, 0x5f, 0x52, 0x45, 0x48, 0x03, 0x0e, 0x19, 0x14, 0x37, 0x3a, 0x2d, 0x20,
+    0x6d, 0x60, 0x77, 0x7a, 0x59, 0x54, 0x43, 0x4e, 0x05, 0x08, 0x1f, 0x12, 0x31, 0x3c, 0x2b, 0x26,
+    0xbd, 0xb0, 0xa7, 0xaa, 0x89, 0x84, 0x93, 0x9e, 0xd5, 0xd8, 0xcf, 0xc2, 0xe1, 0xec, 0xfb, 0xf6,
+    0xd6, 0xdb, 0xcc, 0xc1, 0xe2, 0xef, 0xf8, 0xf5, 0xbe, 0xb3, 0xa4, 0xa9, 0x8a, 0x87, 0x90, 0x9d,
+    0x06, 0x0b, 0x1c, 0x11, 0x32, 0x3f, 0x28, 0x25, 0x6e, 0x63, 0x74, 0x79, 0x5a, 0x57, 0x40, 0x4d,
+    0xda, 0xd7, 0xc0, 0xcd, 0xee, 0xe3, 0xf4, 0xf9, 0xb2, 0xbf, 0xa8, 0xa5, 0x86, 0x8b, 0x9c, 0x91,
+    0x0a, 0x07, 0x10, 0x1d, 0x3e, 0x33, 0x24, 0x29, 0x62, 0x6f, 0x78, 0x75, 0x56, 0x5b, 0x4c, 0x41,
+    0x61, 0x6c, 0x7b, 0x76, 0x55, 0x58, 0x4f, 0x42, 0x09, 0x04, 0x13, 0x1e, 0x3d, 0x30, 0x27, 0x2a,
+    0xb1, 0xbc, 0xab, 0xa6, 0x85, 0x88, 0x9f, 0x92, 0xd9, 0xd4, 0xc3, 0xce, 0xed, 0xe0, 0xf7, 0xfa,
+    0xb7, 0xba, 0xad, 0xa0, 0x83, 0x8e, 0x99, 0x94, 0xdf, 0xd2, 0xc5, 0xc8, 0xeb, 0xe6, 0xf1, 0xfc,
+    0x67, 0x6a, 0x7d, 0x70, 0x53, 0x5e, 0x49, 0x44, 0x0f, 0x02, 0x15, 0x18, 0x3b, 0x36, 0x21, 0x2c,
+    0x0c, 0x01, 0x16, 0x1b, 0x38, 0x35, 0x22, 0x2f, 0x64, 0x69, 0x7e, 0x73, 0x50, 0x5d, 0x4a, 0x47,
+    0xdc, 0xd1, 0xc6, 0xcb, 0xe8, 0xe5, 0xf2, 0xff, 0xb4, 0xb9, 0xae, 0xa3, 0x80, 0x8d, 0x9a, 0x97,
+    // a = 0x0b
+    0x00, 0x0b, 0x16, 0x1d, 0x2c, 0x27, 0x3a, 0x31, 0x58, 0x53, 0x4e, 0x45, 0x74, 0x7f, 0x62, 0x69,
+    0xb0, 0xbb, 0xa6, 0xad, 0x9c, 0x97, 0x8a, 0x81, 0xe8, 0xe3, 0xfe, 0xf5, 0xc4, 0xcf, 0xd2, 0xd9,
+    0x7b, 0x70, 0x6d, 0x66, 0x57, 0x5c, 0x41, 0x4a, 0x23, 0x28, 0x35, 0x3e, 0x0f, 0x04, 0x19, 0x12,
+    0xcb, 0xc0, 0xdd, 0xd6, 0xe7, 0xec, 0xf1, 0xfa, 0x93, 0x98, 0x85, 0x8e, 0xbf, 0xb4, 0xa9, 0xa2,
+    0xf6, 0xfd, 0xe0, 0xeb, 0xda, 0xd1, 0xcc, 0xc7, 0xae, 0xa5, 0xb8, 0xb3, 0x82, 0x89, 0x94, 0x9f,
+    0x46, 0x4d, 0x50, 0x5b, 0x6a, 0x61, 0x7c, 0x77, 0x1e, 0x15, 0x08, 0x03, 0x32, 0x39, 0x24, 0x2f,
+    0x8d, 0x86, 0x9b, 0x90, 0xa1, 0xaa, 0xb7, 0xbc, 0xd5, 0xde, 0xc3, 0xc8, 0xf9, 0xf2, 0xef, 0xe4,
+    0x3d, 0x36, 0x2b, 0x20, 0x11, 0x1a, 0x07, 0x0c, 0x65, 0x6e, 0x73, 0x78, 0x49, 0x42, 0x5f, 0x54,
+    0xf7, 0xfc, 0xe1, 0xea, 0xdb, 0xd0, 0xcd, 0xc6, 0xaf, 0xa4, 0xb9, 0xb2, 0x83, 0x88, 0x95, 0x9e,
+    0x47, 0x4c, 0x51, 0x5a, 0x6b, 0x60, 0x7d, 0x76, 0x1f, 0x14, 0x09, 0x02, 0x33, 0x38, 0x25, 0x2e,
+    0x8c, 0x87, 0x9a, 0x91, 0xa0, 0xab, 0xb6, 0xbd, 0xd4, 0xdf, 0xc2, 0xc9, 0xf8, 0xf3, 0xee, 0xe5,
+    0x3c, 0x37, 0x2a, 0x21, 0x10, 0x1b, 0x06, 0x0d, 0x64, 0x6f, 0x72, 0x79, 0x48, 0x43, 0x5e, 0x55,
+    0x01, 0x0a, 0x17, 0x1c, 0x2d, 0x26, 0x3b, 0x30, 0x59, 0x52, 0x4f, 0x44, 0x75, 0x7e, 0x63, 0x68,
+    0xb1, 0xba, 0xa7, 0xac, 0x9d, 0x96, 0x8b, 0x80, 0xe9, 0xe2, 0xff, 0xf4, 0xc5, 0xce, 0xd3, 0xd8,
+    0x7a, 0x71, 0x6c, 0x67, 0x56, 0x5d, 0x40, 0x4b, 0x22, 0x29, 0x34, 0x3f, 0x0e, 0x05, 0x18, 0x13,
+    0xca, 0xc1, 0xdc, 0xd7, 0xe6, 0xed, 0xf0, 0xfb, 0x92, 0x99, 0x84, 0x8f, 0xbe, 0xb5, 0xa8, 0xa3,
+};
+const uint8_t ff_mult3[4][256] = {
+    // a = 0x02
+    0x00, 0x02, 0x04, 0x06, 0x08, 0x0a, 0x0c, 0x0e, 0x10, 0x12, 0x14, 0x16, 0x18, 0x1a, 0x1c, 0x1e,
+    0x20, 0x22, 0x24, 0x26, 0x28, 0x2a, 0x2c, 0x2e, 0x30, 0x32, 0x34, 0x36, 0x38, 0x3a, 0x3c, 0x3e,
+    0x40, 0x42, 0x44, 0x46, 0x48, 0x4a, 0x4c, 0x4e, 0x50, 0x52, 0x54, 0x56, 0x58, 0x5a, 0x5c, 0x5e,
+    0x60, 0x62, 0x64, 0x66, 0x68, 0x6a, 0x6c, 0x6e, 0x70, 0x72, 0x74, 0x76, 0x78, 0x7a, 0x7c, 0x7e,
+    0x80, 0x82, 0x84, 0x86, 0x88, 0x8a, 0x8c, 0x8e, 0x90, 0x92, 0x94, 0x96, 0x98, 0x9a, 0x9c, 0x9e,
+    0xa0, 0xa2, 0xa4, 0xa6, 0xa8, 0xaa, 0xac, 0xae, 0xb0, 0xb2, 0xb4, 0xb6, 0xb8, 0xba, 0xbc, 0xbe,
+    0xc0, 0xc2, 0xc4, 0xc6, 0xc8, 0xca, 0xcc, 0xce, 0xd0, 0xd2, 0xd4, 0xd6, 0xd8, 0xda, 0xdc, 0xde,
+    0xe0, 0xe2, 0xe4, 0xe6, 0xe8, 0xea, 0xec, 0xee, 0xf0, 0xf2, 0xf4, 0xf6, 0xf8, 0xfa, 0xfc, 0xfe,
+    0x1b, 0x19, 0x1f, 0x1d, 0x13, 0x11, 0x17, 0x15, 0x0b, 0x09, 0x0f, 0x0d, 0x03, 0x01, 0x07, 0x05,
+    0x3b, 0x39, 0x3f, 0x3d, 0x33, 0x31, 0x37, 0x35, 0x2b, 0x29, 0x2f, 0x2d, 0x23, 0x21, 0x27, 0x25,
+    0x5b, 0x59, 0x5f, 0x5d, 0x53, 0x51, 0x57, 0x55, 0x4b, 0x49, 0x4f, 0x4d, 0x43, 0x41, 0x47, 0x45,
+    0x7b, 0x79, 0x7f, 0x7d, 0x73, 0x71, 0x77, 0x75, 0x6b, 0x69, 0x6f, 0x6d, 0x63, 0x61, 0x67, 0x65,
+    0x9b, 0x99, 0x9f, 0x9d, 0x93, 0x91, 0x97, 0x95, 0x8b, 0x89, 0x8f, 0x8d, 0x83, 0x81, 0x87, 0x85,
+    0xbb, 0xb9, 0xbf, 0xbd, 0xb3, 0xb1, 0xb7, 0xb5, 0xab, 0xa9, 0xaf, 0xad, 0xa3, 0xa1, 0xa7, 0xa5,
+    0xdb, 0xd9, 0xdf, 0xdd, 0xd3, 0xd1, 0xd7, 0xd5, 0xcb, 0xc9, 0xcf, 0xcd, 0xc3, 0xc1, 0xc7, 0xc5,
+    0xfb, 0xf9, 0xff, 0xfd, 0xf3, 0xf1, 0xf7, 0xf5, 0xeb, 0xe9, 0xef, 0xed, 0xe3, 0xe1, 0xe7, 0xe5,
+    // a = 0x01
+    0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f,
+    0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f,
+    0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28, 0x29, 0x2a, 0x2b, 0x2c, 0x2d, 0x2e, 0x2f,
+    0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 0x3a, 0x3b, 0x3c, 0x3d, 0x3e, 0x3f,
+    0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 0x49, 0x4a, 0x4b, 0x4c, 0x4d, 0x4e, 0x4f,
+    0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59, 0x5a, 0x5b, 0x5c, 0x5d, 0x5e, 0x5f,
+    0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69, 0x6a, 0x6b, 0x6c, 0x6d, 0x6e, 0x6f,
+    0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, 0x7a, 0x7b, 0x7c, 0x7d, 0x7e, 0x7f,
+    0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87, 0x88, 0x89, 0x8a, 0x8b, 0x8c, 0x8d, 0x8e, 0x8f,
+    0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, 0x98, 0x99, 0x9a, 0x9b, 0x9c, 0x9d, 0x9e, 0x9f,
+    0xa0, 0xa1, 0xa2, 0xa3, 0xa4, 0xa5, 0xa6, 0xa7, 0xa8, 0xa9, 0xaa, 0xab, 0xac, 0xad, 0xae, 0xaf,
+    0xb0, 0xb1, 0xb2, 0xb3, 0xb4, 0xb5, 0xb6, 0xb7, 0xb8, 0xb9, 0xba, 0xbb, 0xbc, 0xbd, 0xbe, 0xbf,
+    0xc0, 0xc1, 0xc2, 0xc3, 0xc4, 0xc5, 0xc6, 0xc7, 0xc8, 0xc9, 0xca, 0xcb, 0xcc, 0xcd, 0xce, 0xcf,
+    0xd0, 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, 0xdb, 0xdc, 0xdd, 0xde, 0xdf,
+    0xe0, 0xe1, 0xe2, 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, 0xea, 0xeb, 0xec, 0xed, 0xee, 0xef,
+    0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff,
+    // a = 0x01
+    0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f,
+    0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f,
+    0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28, 0x29, 0x2a, 0x2b, 0x2c, 0x2d, 0x2e, 0x2f,
+    0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 0x3a, 0x3b, 0x3c, 0x3d, 0x3e, 0x3f,
+    0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 0x49, 0x4a, 0x4b, 0x4c, 0x4d, 0x4e, 0x4f,
+    0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59, 0x5a, 0x5b, 0x5c, 0x5d, 0x5e, 0x5f,
+    0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69, 0x6a, 0x6b, 0x6c, 0x6d, 0x6e, 0x6f,
+    0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, 0x7a, 0x7b, 0x7c, 0x7d, 0x7e, 0x7f,
+    0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87, 0x88, 0x89, 0x8a, 0x8b, 0x8c, 0x8d, 0x8e, 0x8f,
+    0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, 0x98, 0x99, 0x9a, 0x9b, 0x9c, 0x9d, 0x9e, 0x9f,
+    0xa0, 0xa1, 0xa2, 0xa3, 0xa4, 0xa5, 0xa6, 0xa7, 0xa8, 0xa9, 0xaa, 0xab, 0xac, 0xad, 0xae, 0xaf,
+    0xb0, 0xb1, 0xb2, 0xb3, 0xb4, 0xb5, 0xb6, 0xb7, 0xb8, 0xb9, 0xba, 0xbb, 0xbc, 0xbd, 0xbe, 0xbf,
+    0xc0, 0xc1, 0xc2, 0xc3, 0xc4, 0xc5, 0xc6, 0xc7, 0xc8, 0xc9, 0xca, 0xcb, 0xcc, 0xcd, 0xce, 0xcf,
+    0xd0, 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, 0xdb, 0xdc, 0xdd, 0xde, 0xdf,
+    0xe0, 0xe1, 0xe2, 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, 0xea, 0xeb, 0xec, 0xed, 0xee, 0xef,
+    0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, 0xff,
+    // a = 0x03
+    0x00, 0x03, 0x06, 0x05, 0x0c, 0x0f, 0x0a, 0x09, 0x18, 0x1b, 0x1e, 0x1d, 0x14, 0x17, 0x12, 0x11,
+    0x30, 0x33, 0x36, 0x35, 0x3c, 0x3f, 0x3a, 0x39, 0x28, 0x2b, 0x2e, 0x2d, 0x24, 0x27, 0x22, 0x21,
+    0x60, 0x63, 0x66, 0x65, 0x6c, 0x6f, 0x6a, 0x69, 0x78, 0x7b, 0x7e, 0x7d, 0x74, 0x77, 0x72, 0x71,
+    0x50, 0x53, 0x56, 0x55, 0x5c, 0x5f, 0x5a, 0x59, 0x48, 0x4b, 0x4e, 0x4d, 0x44, 0x47, 0x42, 0x41,
+    0xc0, 0xc3, 0xc6, 0xc5, 0xcc, 0xcf, 0xca, 0xc9, 0xd8, 0xdb, 0xde, 0xdd, 0xd4, 0xd7, 0xd2, 0xd1,
+    0xf0, 0xf3, 0xf6, 0xf5, 0xfc, 0xff, 0xfa, 0xf9, 0xe8, 0xeb, 0xee, 0xed, 0xe4, 0xe7, 0xe2, 0xe1,
+    0xa0, 0xa3, 0xa6, 0xa5, 0xac, 0xaf, 0xaa, 0xa9, 0xb8, 0xbb, 0xbe, 0xbd, 0xb4, 0xb7, 0xb2, 0xb1,
+    0x90, 0x93, 0x96, 0x95, 0x9c, 0x9f, 0x9a, 0x99, 0x88, 0x8b, 0x8e, 0x8d, 0x84, 0x87, 0x82, 0x81,
+    0x9b, 0x98, 0x9d, 0x9e, 0x97, 0x94, 0x91, 0x92, 0x83, 0x80, 0x85, 0x86, 0x8f, 0x8c, 0x89, 0x8a,
+    0xab, 0xa8, 0xad, 0xae, 0xa7, 0xa4, 0xa1, 0xa2, 0xb3, 0xb0, 0xb5, 0xb6, 0xbf, 0xbc, 0xb9, 0xba,
+    0xfb, 0xf8, 0xfd, 0xfe, 0xf7, 0xf4, 0xf1, 0xf2, 0xe3, 0xe0, 0xe5, 0xe6, 0xef, 0xec, 0xe9, 0xea,
+    0xcb, 0xc8, 0xcd, 0xce, 0xc7, 0xc4, 0xc1, 0xc2, 0xd3, 0xd0, 0xd5, 0xd6, 0xdf, 0xdc, 0xd9, 0xda,
+    0x5b, 0x58, 0x5d, 0x5e, 0x57, 0x54, 0x51, 0x52, 0x43, 0x40, 0x45, 0x46, 0x4f, 0x4c, 0x49, 0x4a,
+    0x6b, 0x68, 0x6d, 0x6e, 0x67, 0x64, 0x61, 0x62, 0x73, 0x70, 0x75, 0x76, 0x7f, 0x7c, 0x79, 0x7a,
+    0x3b, 0x38, 0x3d, 0x3e, 0x37, 0x34, 0x31, 0x32, 0x23, 0x20, 0x25, 0x26, 0x2f, 0x2c, 0x29, 0x2a,
+    0x0b, 0x08, 0x0d, 0x0e, 0x07, 0x04, 0x01, 0x02, 0x13, 0x10, 0x15, 0x16, 0x1f, 0x1c, 0x19, 0x1a,
+};
+
 static uint8_t ff_mult(uint8_t a, uint8_t b)
 {
    int retval = 0;
@@ -320,10 +526,15 @@ void native_aesimc(x64emu_t* emu, int xmm)
    sse_regs_t eax1 = emu->xmm[xmm];

    for(int j=0; j<4; ++j) {
-        emu->xmm[xmm].ub[0+j*4] = ff_mult(0x0E, eax1.ub[0+j*4]) ^ ff_mult(0x0B, eax1.ub[1+j*4]) ^ ff_mult(0x0D, eax1.ub[2+j*4]) ^ ff_mult(0x09, eax1.ub[3+j*4]);
-        emu->xmm[xmm].ub[1+j*4] = ff_mult(0x09, eax1.ub[0+j*4]) ^ ff_mult(0x0E, eax1.ub[1+j*4]) ^ ff_mult(0x0B, eax1.ub[2+j*4]) ^ ff_mult(0x0D, eax1.ub[3+j*4]);
-        emu->xmm[xmm].ub[2+j*4] = ff_mult(0x0D, eax1.ub[0+j*4]) ^ ff_mult(0x09, eax1.ub[1+j*4]) ^ ff_mult(0x0E, eax1.ub[2+j*4]) ^ ff_mult(0x0B, eax1.ub[3+j*4]);
-        emu->xmm[xmm].ub[3+j*4] = ff_mult(0x0B, eax1.ub[0+j*4]) ^ ff_mult(0x0D, eax1.ub[1+j*4]) ^ ff_mult(0x09, eax1.ub[2+j*4]) ^ ff_mult(0x0E, eax1.ub[3+j*4]);
+        // 0x0E -> 0, 0x09 -> 1, 0x0D -> 2, 0x0B -> 3
+        // emu->xmm[xmm].ub[0+j*4] = ff_mult(0x0E, eax1.ub[0+j*4]) ^ ff_mult(0x0B, eax1.ub[1+j*4]) ^ ff_mult(0x0D, eax1.ub[2+j*4]) ^ ff_mult(0x09, eax1.ub[3+j*4]);
+        // emu->xmm[xmm].ub[1+j*4] = ff_mult(0x09, eax1.ub[0+j*4]) ^ ff_mult(0x0E, eax1.ub[1+j*4]) ^ ff_mult(0x0B, eax1.ub[2+j*4]) ^ ff_mult(0x0D, eax1.ub[3+j*4]);
+        // emu->xmm[xmm].ub[2+j*4] = ff_mult(0x0D, eax1.ub[0+j*4]) ^ ff_mult(0x09, eax1.ub[1+j*4]) ^ ff_mult(0x0E, eax1.ub[2+j*4]) ^ ff_mult(0x0B, eax1.ub[3+j*4]);
+        // emu->xmm[xmm].ub[3+j*4] = ff_mult(0x0B, eax1.ub[0+j*4]) ^ ff_mult(0x0D, eax1.ub[1+j*4]) ^ ff_mult(0x09, eax1.ub[2+j*4]) ^ ff_mult(0x0E, eax1.ub[3+j*4]);
+        emu->xmm[xmm].ub[0+j*4] = ff_mult2[0][eax1.ub[0+j*4]] ^ ff_mult2[3][eax1.ub[1+j*4]] ^ ff_mult2[2][eax1.ub[2+j*4]] ^ ff_mult2[1][eax1.ub[3+j*4]];
+        emu->xmm[xmm].ub[1+j*4] = ff_mult2[1][eax1.ub[0+j*4]] ^ ff_mult2[0][eax1.ub[1+j*4]] ^ ff_mult2[3][eax1.ub[2+j*4]] ^ ff_mult2[2][eax1.ub[3+j*4]];
+        emu->xmm[xmm].ub[2+j*4] = ff_mult2[2][eax1.ub[0+j*4]] ^ ff_mult2[1][eax1.ub[1+j*4]] ^ ff_mult2[0][eax1.ub[2+j*4]] ^ ff_mult2[3][eax1.ub[3+j*4]];
+        emu->xmm[xmm].ub[3+j*4] = ff_mult2[3][eax1.ub[0+j*4]] ^ ff_mult2[2][eax1.ub[1+j*4]] ^ ff_mult2[1][eax1.ub[2+j*4]] ^ ff_mult2[0][eax1.ub[3+j*4]];
    }
 }
 void native_aesimc_y(x64emu_t* emu, int ymm)
@@ -331,10 +542,14 @@ void native_aesimc_y(x64emu_t* emu, int ymm)
    sse_regs_t eay1 = emu->ymm[ymm];

    for(int j=0; j<4; ++j) {
-        emu->ymm[ymm].ub[0+j*4] = ff_mult(0x0E, eay1.ub[0+j*4]) ^ ff_mult(0x0B, eay1.ub[1+j*4]) ^ ff_mult(0x0D, eay1.ub[2+j*4]) ^ ff_mult(0x09, eay1.ub[3+j*4]);
-        emu->ymm[ymm].ub[1+j*4] = ff_mult(0x09, eay1.ub[0+j*4]) ^ ff_mult(0x0E, eay1.ub[1+j*4]) ^ ff_mult(0x0B, eay1.ub[2+j*4]) ^ ff_mult(0x0D, eay1.ub[3+j*4]);
-        emu->ymm[ymm].ub[2+j*4] = ff_mult(0x0D, eay1.ub[0+j*4]) ^ ff_mult(0x09, eay1.ub[1+j*4]) ^ ff_mult(0x0E, eay1.ub[2+j*4]) ^ ff_mult(0x0B, eay1.ub[3+j*4]);
-        emu->ymm[ymm].ub[3+j*4] = ff_mult(0x0B, eay1.ub[0+j*4]) ^ ff_mult(0x0D, eay1.ub[1+j*4]) ^ ff_mult(0x09, eay1.ub[2+j*4]) ^ ff_mult(0x0E, eay1.ub[3+j*4]);
+        // emu->ymm[ymm].ub[0+j*4] = ff_mult(0x0E, eay1.ub[0+j*4]) ^ ff_mult(0x0B, eay1.ub[1+j*4]) ^ ff_mult(0x0D, eay1.ub[2+j*4]) ^ ff_mult(0x09, eay1.ub[3+j*4]);
+        // emu->ymm[ymm].ub[1+j*4] = ff_mult(0x09, eay1.ub[0+j*4]) ^ ff_mult(0x0E, eay1.ub[1+j*4]) ^ ff_mult(0x0B, eay1.ub[2+j*4]) ^ ff_mult(0x0D, eay1.ub[3+j*4]);
+        // emu->ymm[ymm].ub[2+j*4] = ff_mult(0x0D, eay1.ub[0+j*4]) ^ ff_mult(0x09, eay1.ub[1+j*4]) ^ ff_mult(0x0E, eay1.ub[2+j*4]) ^ ff_mult(0x0B, eay1.ub[3+j*4]);
+        // emu->ymm[ymm].ub[3+j*4] = ff_mult(0x0B, eay1.ub[0+j*4]) ^ ff_mult(0x0D, eay1.ub[1+j*4]) ^ ff_mult(0x09, eay1.ub[2+j*4]) ^ ff_mult(0x0E, eay1.ub[3+j*4]);
+        emu->ymm[ymm].ub[0+j*4] = ff_mult2[0][eay1.ub[0+j*4]] ^ ff_mult2[3][eay1.ub[1+j*4]] ^ ff_mult2[2][eay1.ub[2+j*4]] ^ ff_mult2[1][eay1.ub[3+j*4]];
+        emu->ymm[ymm].ub[1+j*4] = ff_mult2[1][eay1.ub[0+j*4]] ^ ff_mult2[0][eay1.ub[1+j*4]] ^ ff_mult2[3][eay1.ub[2+j*4]] ^ ff_mult2[2][eay1.ub[3+j*4]];
+        emu->ymm[ymm].ub[2+j*4] = ff_mult2[2][eay1.ub[0+j*4]] ^ ff_mult2[1][eay1.ub[1+j*4]] ^ ff_mult2[0][eay1.ub[2+j*4]] ^ ff_mult2[3][eay1.ub[3+j*4]];
+        emu->ymm[ymm].ub[3+j*4] = ff_mult2[3][eay1.ub[0+j*4]] ^ ff_mult2[2][eay1.ub[1+j*4]] ^ ff_mult2[1][eay1.ub[2+j*4]] ^ ff_mult2[0][eay1.ub[3+j*4]];
    }
 }
 void native_aesmc(x64emu_t* emu, int xmm)
@@ -342,10 +557,15 @@ void native_aesmc(x64emu_t* emu, int xmm)
    sse_regs_t eax1 = emu->xmm[xmm];

    for(int j=0; j<4; ++j) {
-        emu->xmm[xmm].ub[0+j*4] = ff_mult(0x02, eax1.ub[0+j*4]) ^ ff_mult(0x03, eax1.ub[1+j*4]) ^               eax1.ub[2+j*4]  ^               eax1.ub[3+j*4] ;
-        emu->xmm[xmm].ub[1+j*4] =               eax1.ub[0+j*4]  ^ ff_mult(0x02, eax1.ub[1+j*4]) ^ ff_mult(0x03, eax1.ub[2+j*4]) ^               eax1.ub[3+j*4] ;
-        emu->xmm[xmm].ub[2+j*4] =               eax1.ub[0+j*4]  ^               eax1.ub[1+j*4]  ^ ff_mult(0x02, eax1.ub[2+j*4]) ^ ff_mult(0x03, eax1.ub[3+j*4]);
-        emu->xmm[xmm].ub[3+j*4] = ff_mult(0x03, eax1.ub[0+j*4]) ^               eax1.ub[1+j*4]  ^               eax1.ub[2+j*4]  ^ ff_mult(0x02, eax1.ub[3+j*4]);
+        // 0x02 -> 0, 0x01 -> 1, 0x01 -> 2, 0x03 -> 3
+        // emu->xmm[xmm].ub[0+j*4] = ff_mult(0x02, eax1.ub[0+j*4]) ^ ff_mult(0x03, eax1.ub[1+j*4]) ^               eax1.ub[2+j*4]  ^               eax1.ub[3+j*4] ;
+        // emu->xmm[xmm].ub[1+j*4] =               eax1.ub[0+j*4]  ^ ff_mult(0x02, eax1.ub[1+j*4]) ^ ff_mult(0x03, eax1.ub[2+j*4]) ^               eax1.ub[3+j*4] ;
+        // emu->xmm[xmm].ub[2+j*4] =               eax1.ub[0+j*4]  ^               eax1.ub[1+j*4]  ^ ff_mult(0x02, eax1.ub[2+j*4]) ^ ff_mult(0x03, eax1.ub[3+j*4]);
+        // emu->xmm[xmm].ub[3+j*4] = ff_mult(0x03, eax1.ub[0+j*4]) ^               eax1.ub[1+j*4]  ^               eax1.ub[2+j*4]  ^ ff_mult(0x02, eax1.ub[3+j*4]);
+        emu->xmm[xmm].ub[0+j*4] = ff_mult3[0][eax1.ub[0+j*4]] ^ ff_mult3[3][eax1.ub[1+j*4]] ^ ff_mult3[2][eax1.ub[2+j*4]] ^ ff_mult3[1][eax1.ub[3+j*4]];
+        emu->xmm[xmm].ub[1+j*4] = ff_mult3[1][eax1.ub[0+j*4]] ^ ff_mult3[0][eax1.ub[1+j*4]] ^ ff_mult3[3][eax1.ub[2+j*4]] ^ ff_mult3[2][eax1.ub[3+j*4]];
+        emu->xmm[xmm].ub[2+j*4] = ff_mult3[2][eax1.ub[0+j*4]] ^ ff_mult3[1][eax1.ub[1+j*4]] ^ ff_mult3[0][eax1.ub[2+j*4]] ^ ff_mult3[3][eax1.ub[3+j*4]];
+        emu->xmm[xmm].ub[3+j*4] = ff_mult3[3][eax1.ub[0+j*4]] ^ ff_mult3[2][eax1.ub[1+j*4]] ^ ff_mult3[1][eax1.ub[2+j*4]] ^ ff_mult3[0][eax1.ub[3+j*4]];
    }
 }
 void native_aesmc_y(x64emu_t* emu, int ymm)
@@ -353,16 +573,20 @@ void native_aesmc_y(x64emu_t* emu, int ymm)
    sse_regs_t eay1 = emu->ymm[ymm];

    for(int j=0; j<4; ++j) {
-        emu->ymm[ymm].ub[0+j*4] = ff_mult(0x02, eay1.ub[0+j*4]) ^ ff_mult(0x03, eay1.ub[1+j*4]) ^               eay1.ub[2+j*4]  ^               eay1.ub[3+j*4] ;
-        emu->ymm[ymm].ub[1+j*4] =               eay1.ub[0+j*4]  ^ ff_mult(0x02, eay1.ub[1+j*4]) ^ ff_mult(0x03, eay1.ub[2+j*4]) ^               eay1.ub[3+j*4] ;
-        emu->ymm[ymm].ub[2+j*4] =               eay1.ub[0+j*4]  ^               eay1.ub[1+j*4]  ^ ff_mult(0x02, eay1.ub[2+j*4]) ^ ff_mult(0x03, eay1.ub[3+j*4]);
-        emu->ymm[ymm].ub[3+j*4] = ff_mult(0x03, eay1.ub[0+j*4]) ^               eay1.ub[1+j*4]  ^               eay1.ub[2+j*4]  ^ ff_mult(0x02, eay1.ub[3+j*4]);
+        // emu->ymm[ymm].ub[0+j*4] = ff_mult(0x02, eay1.ub[0+j*4]) ^ ff_mult(0x03, eay1.ub[1+j*4]) ^               eay1.ub[2+j*4]  ^               eay1.ub[3+j*4] ;
+        // emu->ymm[ymm].ub[1+j*4] =               eay1.ub[0+j*4]  ^ ff_mult(0x02, eay1.ub[1+j*4]) ^ ff_mult(0x03, eay1.ub[2+j*4]) ^               eay1.ub[3+j*4] ;
+        // emu->ymm[ymm].ub[2+j*4] =               eay1.ub[0+j*4]  ^               eay1.ub[1+j*4]  ^ ff_mult(0x02, eay1.ub[2+j*4]) ^ ff_mult(0x03, eay1.ub[3+j*4]);
+        // emu->ymm[ymm].ub[3+j*4] = ff_mult(0x03, eay1.ub[0+j*4]) ^               eay1.ub[1+j*4]  ^               eay1.ub[2+j*4]  ^ ff_mult(0x02, eay1.ub[3+j*4]);
+        emu->ymm[ymm].ub[0+j*4] = ff_mult3[0][eay1.ub[0+j*4]] ^ ff_mult3[3][eay1.ub[1+j*4]] ^ ff_mult3[2][eay1.ub[2+j*4]] ^ ff_mult3[1][eay1.ub[3+j*4]];
+        emu->ymm[ymm].ub[1+j*4] = ff_mult3[1][eay1.ub[0+j*4]] ^ ff_mult3[0][eay1.ub[1+j*4]] ^ ff_mult3[3][eay1.ub[2+j*4]] ^ ff_mult3[2][eay1.ub[3+j*4]];
+        emu->ymm[ymm].ub[2+j*4] = ff_mult3[2][eay1.ub[0+j*4]] ^ ff_mult3[1][eay1.ub[1+j*4]] ^ ff_mult3[0][eay1.ub[2+j*4]] ^ ff_mult3[3][eay1.ub[3+j*4]];
+        emu->ymm[ymm].ub[3+j*4] = ff_mult3[3][eay1.ub[0+j*4]] ^ ff_mult3[2][eay1.ub[1+j*4]] ^ ff_mult3[1][eay1.ub[2+j*4]] ^ ff_mult3[0][eay1.ub[3+j*4]];
    }
 }
                                  //   A0 B1 C2 D3 E4 F5 G6 H7 I8 J9 Ka Lb Mc Nd Oe Pf
                                  //   A  N  K  H  E  B  O  L  I  F  C  P  M  J  G  D
-static const uint8_t invshiftrows[] = {0,13,10, 7, 4, 1,14,11, 8, 5, 2,15,12, 9, 6, 3};
-static const uint8_t invsubbytes[256] = {
+const uint8_t invshiftrows[] = {0,13,10, 7, 4, 1,14,11, 8, 5, 2,15,12, 9, 6, 3};
+const uint8_t invsubbytes[256] = {
    0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38, 0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb,
    0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87, 0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb,
    0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d, 0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e,
@@ -402,8 +626,8 @@ void native_aesdlast_y(x64emu_t* emu, int ymm)
        emu->ymm[ymm].ub[i] = invsubbytes[eay1.ub[i]];

 }
-static const uint8_t shiftrows[] = {0, 5,10,15, 4, 9,14, 3, 8,13, 2, 7,12, 1, 6,11};
-static const uint8_t subbytes[256] = {
+const uint8_t shiftrows[] = {0, 5,10,15, 4, 9,14, 3, 8,13, 2, 7,12, 1, 6,11};
+const uint8_t subbytes[256] = {
    0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5, 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76,
    0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0, 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0,
    0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc, 0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15,
@@ -527,15 +751,29 @@ void native_pclmul_y(x64emu_t* emu, int gy, int vy, void* p, uint32_t u8)
    GY->u128 = result;
 }

-void native_clflush(x64emu_t* emu, void* p)
-{
-    cleanDBFromAddressRange((uintptr_t)p, 8, 0);
-}
-
 static int flagsCacheNeedsTransform(dynarec_native_t* dyn, int ninst) {
    int jmp = dyn->insts[ninst].x64.jmp_insts;
    if(jmp<0)
        return 0;
+    #if defined(ARM64) || defined(LA64)
+    // df_none is now a defered information
+    if(dyn->insts[ninst].f_exit==dyn->insts[jmp].f_entry)
+        return 0;
+    if(dyn->insts[jmp].df_notneeded)
+        return 0;
+    if((dyn->insts[jmp].f_entry==status_none_pending) && (dyn->insts[ninst].f_exit!=status_none_pending))
+        return 1;
+    switch (dyn->insts[jmp].f_entry) {
+        case status_unk:
+            return (dyn->insts[ninst].f_exit==status_none_pending)?1:0;
+        case status_none:
+            return 1;
+        case status_set:
+            return (dyn->insts[ninst].f_exit==status_none)?0:1;
+        case status_none_pending:
+            return 1;
+    }
+#else
    if(dyn->insts[ninst].f_exit.dfnone)  // flags are fully known, nothing we can do more
        return 0;
    if(dyn->insts[jmp].f_entry.dfnone && !dyn->insts[ninst].f_exit.dfnone && !dyn->insts[jmp].df_notneeded)
@@ -556,6 +794,7 @@ static int flagsCacheNeedsTransform(dynarec_native_t* dyn, int ninst) {
                return 0;
            return (dyn->insts[jmp].f_entry.dfnone  == dyn->insts[ninst].f_exit.dfnone)?0:1;
    }
+    #endif
    return 0;
 }

@@ -617,34 +856,6 @@ uint8_t geted_ib(dynarec_native_t* dyn, uintptr_t addr, int ninst, uint8_t nexto
 }
 #undef F8

-// AVX
-void avx_mark_zero(dynarec_native_t* dyn, int ninst, int reg)
-{
-    dyn->ymm_zero |= (1<<reg);
-}
-
-int is_avx_zero(dynarec_native_t* dyn, int ninst, int reg)
-{
-    return (dyn->ymm_zero>>reg)&1;
-}
-int is_avx_zero_unset(dynarec_native_t* dyn, int ninst, int reg)
-{
-    if((dyn->ymm_zero>>reg)&1) {
-        dyn->ymm_zero &= ~(1<<reg);
-        return 1;
-    }
-    return 0;
-}
-void avx_mark_zero_reset(dynarec_native_t* dyn, int ninst)
-{
-    dyn->ymm_zero = 0;
-}
-
-void avx_unmark_zero(dynarec_native_t* dyn, int ninst, int reg)
-{
-    dyn->ymm_zero &= ~(1<<reg);
-}
-
 void propagate_nodf(dynarec_native_t* dyn, int ninst)
 {
    while(ninst>=0) {
@@ -653,7 +864,11 @@ void propagate_nodf(dynarec_native_t* dyn, int ninst)
        if(dyn->insts[ninst].x64.gen_flags || dyn->insts[ninst].x64.use_flags)
            return; // flags are use, so maybe it's needed
        dyn->insts[ninst].df_notneeded = 1;
-        --ninst;
+        if(!dyn->insts[ninst].pred_sz)
+            return;
+        for(int i=1; i<dyn->insts[ninst].pred_sz; ++i)
+            propagate_nodf(dyn, dyn->insts[ninst].pred[i]);
+        ninst = dyn->insts[ninst].pred[0];
    }
 }

@@ -685,3 +900,21 @@ void x64disas_add_register_mapping_annotations(char* buf, const char* disas, con
    if (tmp[0]) tmp[strlen(tmp) - 1] = '\0';
    sprintf(buf, "%-35s ;%s", disas, tmp);
 }
+
+void dynarec_stopped(uintptr_t ip, int is32bits)
+{
+    #define PKip(A) (((uint8_t*)ip)[A])
+    dynarec_log(LOG_NONE, "%p: Dynarec stopped because of %s Opcode ", (void*)ip, is32bits ? "x86" : "x64");
+    zydis_dec_t* dec = is32bits ? my_context->dec32 : my_context->dec;
+    if(getProtection(ip+14)&PROT_READ) {
+        if (dec) {
+            dynarec_log_prefix(0, LOG_NONE, "%s", DecodeX64Trace(dec, ip, 1));
+        } else {
+            dynarec_log_prefix(0, LOG_NONE, "%02hhX %02hhX %02hhX %02hhX %02hhX %02hhX %02hhX %02hhX %02hhX %02hhX %02hhX %02hhX %02hhX %02hhX %02hhX",
+                PKip(0), PKip(1), PKip(2), PKip(3), PKip(4), PKip(5), PKip(6), PKip(7), PKip(8), PKip(9),
+                PKip(10), PKip(11), PKip(12), PKip(13), PKip(14));
+        }
+    } else dynarec_log_prefix(0, LOG_NONE, "%02hhX", PKip(0));
+    PrintFunctionAddr(ip, " => ");
+    dynarec_log_prefix(0, LOG_NONE, "\n");
+}
@@ -35,6 +35,15 @@ void native_frstor(x64emu_t* emu, uint8_t* ed);
 void native_frstor16(x64emu_t* emu, uint8_t* ed);
 void native_fprem1(x64emu_t* emu);

+double direct_f2xm1(x64emu_t* emu, double a);
+double direct_fyl2x(x64emu_t* emu, double a, double b);
+double direct_fyl2xp1(x64emu_t* emu, double a, double b);
+double direct_fpatan(x64emu_t* emu, double a, double b);
+double direct_fsin(x64emu_t* emu, double a);
+double direct_fcos(x64emu_t* emu, double a);
+double direct_ftan(x64emu_t* emu, double a);
+double direct_fscale(x64emu_t* emu, double a, double b);
+
 void native_aesd(x64emu_t* emu, int xmm);
 void native_aesd_y(x64emu_t* emu, int ymm);
 void native_aese(x64emu_t* emu, int xmm);
@@ -50,11 +59,10 @@ void native_pclmul(x64emu_t* emu, int gx, int ex, void* p, uint32_t u8);
 void native_pclmul_x(x64emu_t* emu, int gx, int vx, void* p, uint32_t u8);
 void native_pclmul_y(x64emu_t* emu, int gy, int vy, void* p, uint32_t u8);

-void native_clflush(x64emu_t* emu, void* p);
-
 void native_ud(x64emu_t* emu);
 void native_br(x64emu_t* emu);
 void native_priv(x64emu_t* emu);
+void native_gpf(x64emu_t* emu);
 void native_int3(x64emu_t* emu);
 void native_int(x64emu_t* emu, int num);
 void native_wineint(x64emu_t* emu, int num);
@@ -74,19 +82,13 @@ uintptr_t fakeed(dynarec_native_t* dyn, uintptr_t addr, int ninst, uint8_t nexto
 // return Ib on a mod/rm opcode without emitting anything
 uint8_t geted_ib(dynarec_native_t* dyn, uintptr_t addr, int ninst, uint8_t nextop);

-// AVX utilities
-void avx_mark_zero(dynarec_native_t* dyn, int ninst, int reg);
-int is_avx_zero(dynarec_native_t* dyn, int ninst, int reg);
-int is_avx_zero_unset(dynarec_native_t* dyn, int ninst, int reg);
-void avx_mark_zero_reset(dynarec_native_t* dyn, int ninst);
-void avx_unmark_zero(dynarec_native_t* dyn, int ninst, int reg);
-
 typedef struct register_mapping_s {
    const char* name;
    const char* native;
 } register_mapping_t;

 void x64disas_add_register_mapping_annotations(char* buf, const char* disas, const register_mapping_t* mappings, size_t mappings_sz);
+void dynarec_stopped(uintptr_t x64addr, int is32bits);

 ADDITIONNAL_DEFINITION()

@@ -31,6 +31,27 @@
 #ifndef PROT_READ
 #define PROT_READ 0x1
 #endif
+
+// Fail-close guard for page-boundary decode hazards:
+// if we cannot read a full x86 max instruction window, stop block build early.
+static int dynarec_can_read_window(uintptr_t addr, uintptr_t size)
+{
+    if(!size)
+        return 1;
+    uintptr_t end = addr + size - 1;
+    if(end < addr)
+        return 0;
+
+    uintptr_t cur = addr;
+    while(1) {
+        if(!(getProtection(cur) & PROT_READ))
+            return 0;
+        uintptr_t page_end = (cur & ~(box64_pagesize - 1)) + box64_pagesize - 1;
+        if(end <= page_end)
+            return 1;
+        cur = page_end + 1;
+    }
+}
 #endif

 uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int is32bits, int inst_max)
@@ -44,21 +65,30 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
    int rep = 0;    // 0 none, 1=F2 prefix, 2=F3 prefix
    int need_epilog = 1;
    // Clean up (because there are multiple passes)
+    #if defined(ARM64) || defined(LA64)
+    dyn->f = status_unk;
+    #else
    dyn->f.pending = 0;
    dyn->f.dfnone = 0;
+    #endif
    dyn->forward = 0;
    dyn->forward_to = 0;
    dyn->forward_size = 0;
    dyn->forward_ninst = 0;
    dyn->ymm_zero = 0;
+    dyn->is_file_mapped = IsAddrElfOrFileMapped(addr);
+    int dynarec_dirty = BOX64ENV(dynarec_dirty);
    #if STEP == 0
    memset(&dyn->insts[ninst], 0, sizeof(instruction_native_t));
+    #ifdef ARM64
+    dyn->have_purge = BOX64ENV(dynarec_purge);
+    #endif
    #endif
    fpu_reset(dyn);
    ARCH_INIT();
    int reset_n = -1; // -1 no reset; -2 reset to 0; else reset to the state of reset_n
    dyn->last_ip = (alternate || (dyn->insts && dyn->insts[0].pred_sz))?0:ip;  // RIP is always set at start of block unless there is a predecessor!
-    int stopblock = 2 + !IsAddrElfOrFileMapped(addr);                          // if block is in elf memory or file mapped memory, it can be extended with BOX64DRENV(dynarec_bigblock)==2, else it needs 3
+    int stopblock = 2 + !dyn->is_file_mapped;                          // if block is in elf memory or file mapped memory, it can be extended with BOX64DRENV(dynarec_bigblock)==2, else it needs 3
    // ok, go now
    INIT;
    #if STEP == 0
@@ -66,16 +96,21 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
    #endif
    while(ok) {
        #if STEP == 0
+        int stop_for_guard = 0;
        if(cur_page != ((addr)&~(box64_pagesize-1))) {
            cur_page = (addr)&~(box64_pagesize-1);
            uint32_t prot = getProtection(addr);
-            if(!(prot&PROT_READ) || checkInHotPage(addr)) {
-                need_epilog = 1;
-                break;
+            if(!(prot&PROT_READ) || !(prot&PROT_EXEC) || checkInHotPage(addr) || (addr>dyn->end)) {
+                stop_for_guard = 1;
            }
            if(prot&PROT_NEVERCLEAN)
                dyn->always_test = 1;
        }
+        if(stop_for_guard) {
+            dynarec_log(LOG_INFO, "Stopping dynablock because of protection/hotpage/mmap/decode-window at %p -> %p inst=%d\n", (void*)dyn->start, (void*)addr, ninst);
+            need_epilog = 1;
+            break;
+        }
        // This test is here to prevent things like TABLE64 to be out of range
        // native_size is not exact at this point, but it should be larger, not smaller, and not by a huge margin anyway
        // so it's good enough to avoid overflow in relative to PC data fectching
@@ -87,14 +122,27 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
        if(!ninst && dyn->need_x87check) {
            NATIVE_RESTORE_X87PC();
        }
-        fpu_propagate_stack(dyn, ninst);
        ip = addr;
+        #ifdef ARM64
+        if(!ninst) {
+            if(dyn->have_purge)
+                doEnterBlock(dyn, 0, x1, x2, x3);
+            if(dyn->insts[0].preload_xmmymm)
+                doPreload(dyn, 0);
+            ENDPREFIX;
+        }
+        #endif
+        fpu_propagate_stack(dyn, ninst);
        if (reset_n!=-1) {
            dyn->last_ip = 0;
            if(reset_n==-2) {
                MESSAGE(LOG_DEBUG, "Reset Caches to zero\n");
+                #if defined(ARM64) || defined(LA64)
+                dyn->f = status_unk;
+                #else
                dyn->f.dfnone = 0;
                dyn->f.pending = 0;
+                #endif
                fpu_reset(dyn);
                ARCH_RESET();
            } else {
@@ -106,8 +154,12 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
                }
                if(dyn->insts[ninst].x64.barrier&BARRIER_FLAGS) {
                    MESSAGE(LOG_DEBUG, "Apply Barrier Flags\n");
+                    #if defined(ARM64) || defined(LA64)
+                    dyn->f = status_unk;
+                    #else
                    dyn->f.dfnone = 0;
                    dyn->f.pending = 0;
+                    #endif
                }
            }
            reset_n = -1;
@@ -116,7 +168,6 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
        else if(ninst && (dyn->insts[ninst].pred_sz>1 || (dyn->insts[ninst].pred_sz==1 && dyn->insts[ninst].pred[0]!=ninst-1)))
            dyn->last_ip = 0;   // reset IP if some jump are coming here
        #endif
-        dyn->f.dfnone_here = 0;
        NEW_INST;
        MESSAGE(LOG_DUMP, "New Instruction %s:%p, native:%p\n", is32bits?"x86":"x64",(void*)addr, (void*)dyn->block);
        #ifdef ARCH_NOP
@@ -134,14 +185,16 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
            WILLWRITE();
        }

-        int is_opcode_volatile = box64_wine && VolatileRangesContains(ip);
-        if (is_opcode_volatile && !dyn->insts[ninst].lock_prefixed && dyn->insts[ninst].will_write)
+        int is_opcode_volatile = /*box64_wine &&*/ VolatileRangesContains(ip) && VolatileOpcodesHas(ip);
+        if (is_opcode_volatile && !dyn->insts[ninst].lock)
            DMB_ISHST();
        #endif
-        if((dyn->insts[ninst].x64.need_before&~X_PEND) && !ninst) {
-            READFLAGS(dyn->insts[ninst].x64.need_before&~X_PEND);
+        if (BOX64DRENV(dynarec_dump) && (!BOX64ENV(dynarec_dump_range_end) || (ip >= BOX64ENV(dynarec_dump_range_start) && ip < BOX64ENV(dynarec_dump_range_end)))) {
+            dyn->need_dump = BOX64DRENV(dynarec_dump);
        }
        if(BOX64ENV(dynarec_test) && (!BOX64ENV(dynarec_test_end) || (ip>=BOX64ENV(dynarec_test_start) && ip<BOX64ENV(dynarec_test_end)))) {
+            int need_dump = dyn->need_dump;
+            if (BOX64ENV(dynarec_test_nodump)) dyn->need_dump = 0;
            MESSAGE(LOG_DUMP, "TEST STEP ----\n");
            extcache_native_t save;
            fpu_save_and_unwind(dyn, ninst, &save);
@@ -150,9 +203,7 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
            fpu_unreflectcache(dyn, ninst, x1, x2, x3);
            fpu_unwind_restore(dyn, ninst, &save);
            MESSAGE(LOG_DUMP, "----------\n");
-        }
-        if (BOX64DRENV(dynarec_dump) && (!BOX64ENV(dynarec_dump_range_end) || (ip >= BOX64ENV(dynarec_dump_range_start) && ip < BOX64ENV(dynarec_dump_range_end)))) {
-            dyn->need_dump = BOX64DRENV(dynarec_dump);
+            dyn->need_dump = need_dump;
        }
        #ifdef HAVE_TRACE
        else if(my_context->dec && BOX64ENV(dynarec_trace)) {
@@ -167,35 +218,59 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
        }
        #endif

-        rep = 0;
        uint8_t pk = PK(0);
-        while((pk==0xF2) || (pk==0xF3) || (pk==0x3E) || (pk==0x26)) {
-            switch(pk) {
-                case 0xF2: rep = 1; break;
-                case 0xF3: rep = 2; break;
-                case 0x3E:
-                case 0x26: /* ignored */ break;
+        
+        rex.rex = 0;
+        rex.seg = 0;
+        rex.offset = 0;
+        rex.is32bits = is32bits;
+        rex.is66 = 0;
+        rex.is67 = 0;
+        rex.isf0 = 0;
+        rex.rep = 0;
+        while((pk==0xF2) || (pk==0xF3) || (pk==0xf0)
+            || (pk==0x3E) || (pk==0x26) || (pk==0x2e) || (pk==0x36) 
+            || (pk==0x64) || (pk==0x65) || (pk==0x66) || (pk==0x67)
+            || (!is32bits && (pk>=0x40 && pk<=0x4f))) {
+            switch (pk) {
+                case 0xF0: rex.isf0 = 1; rex.rex = 0; break;
+                case 0xF2: rex.rep = 1; rex.rex = 0; break;
+                case 0xF3: rex.rep = 2; rex.rex = 0; break;
+                case 0x26: /* ES: */
+                case 0x2E: /* CS: */
+                case 0x36: /* SS; */
+                case 0x3E: /* DS; */ 
+                           rex.seg =   0; rex.rex = 0; break;
+                case 0x64: rex.seg = _FS; rex.rex = 0; break;
+                case 0x65: rex.seg = _GS; rex.rex = 0; break;
+                case 0x66: rex.is66 = 1; rex.rex = 0; break;
+                case 0x67: rex.is67 = 1; rex.rex = 0; break;
+                case 0x40 ... 0x4F: rex.rex = pk; break;
            }
            ++addr;
            pk = PK(0);
        }
-        rex.rex = 0;
-        rex.is32bits = is32bits;
-        if(!rex.is32bits)
-            while(pk>=0x40 && pk<=0x4f) {
-                rex.rex = pk;
-                ++addr;
-                pk = PK(0);
-            }
-
-        addr = dynarec64_00(dyn, addr, ip, ninst, rex, rep, &ok, &need_epilog);
+        if(rex.isf0) {
+            if(rex.is66 && !rex.w)
+                addr = dynarec64_66F0(dyn, addr, ip, ninst, rex, &ok, &need_epilog);
+            else
+                addr = dynarec64_F0(dyn, addr, ip, ninst, rex, &ok, &need_epilog);
+        } else if(rex.is66)
+            addr = dynarec64_66(dyn, addr, ip, ninst, rex, &ok, &need_epilog);
+        else
+            addr = dynarec64_00(dyn, addr, ip, ninst, rex, &ok, &need_epilog);
        if(dyn->abort)
            return ip;
        INST_EPILOG;

        #if STEP > 1
-        if (is_opcode_volatile && !dyn->insts[ninst].lock_prefixed && dyn->insts[ninst].will_read)
-            DMB_ISHLD();
+        if (is_opcode_volatile || dyn->insts[ninst].lock)
+            DMB_ISH();
+        #endif
+        #ifdef ARM64
+        if(dyn->insts[ninst].x64.has_next && dyn->insts[ninst+1].preload_xmmymm) {
+            doPreload(dyn, ninst+1);
+        }
        #endif

        fpu_reset_scratch(dyn);
@@ -209,12 +284,16 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
                else { tmp1=x1; tmp2=x2; tmp3=x3; }
                fpu_purgecache(dyn, ninst, 0, tmp1, tmp2, tmp3);
                #else
-                fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+                fpu_purgecache(dyn, ninst, 0, x1, x2, x3, 0);
                #endif
            }
            if(dyn->insts[next].x64.barrier&BARRIER_FLAGS) {
+                #if defined(ARM64) || defined(LA64)
+                dyn->f = status_unk;
+                #else
                dyn->f.pending = 0;
                dyn->f.dfnone = 0;
+                #endif
                dyn->last_ip = 0;
            }
        }
@@ -222,14 +301,25 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
        #ifndef PROT_READ
        #define PROT_READ 1
        #endif
+        #if STEP == 0
+        if(dynarec_dirty && ok && is_addr_autosmc(ip)) {
+            // this is the last opcode, because it will write in current block if not stopped
+            ok = 0;
+            need_epilog = 1;
+        }
+        #endif
        #if STEP != 0
        if(!ok && !need_epilog && (addr < (dyn->start+dyn->isize))) {
            ok = 1;
            // we use the 1st predecessor here
            if((ninst+1)<dyn->size && !dyn->insts[ninst+1].x64.alive) {
                // reset fpu value...
+                #if defined(ARM64) || defined(LA64)
+                dyn->f = status_unk;
+                #else
                dyn->f.dfnone = 0;
                dyn->f.pending = 0;
+                #endif
                fpu_reset(dyn);
                while((ninst+1)<dyn->size && !dyn->insts[ninst+1].x64.alive) {
                    // may need to skip opcodes to advance
@@ -255,7 +345,7 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
        }
        #else
        // check if block need to be stopped, because it's a 00 00 opcode (unreadeable is already checked earlier)
-        if((ok>0) && !dyn->forward && !(*(uint32_t*)addr)) {
+        if((ok>0) && !dyn->forward && (!(getProtection(addr+3)&PROT_READ) || !(*(uint32_t*)addr))) {
            if (dyn->need_dump) dynarec_log(LOG_NONE, "Stopping block at %p reason: %s\n", (void*)addr, "Next opcode is 00 00 00 00");
            ok = 0;
            need_epilog = 1;
@@ -290,8 +380,7 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
            if(*(uint32_t*)addr!=0) {   // check if need to continue (but is next 4 bytes are 0, stop)
                uintptr_t next = get_closest_next(dyn, addr);
                if(next && (
-                    (((next-addr)<15) && is_nops(dyn, addr, next-addr))
-                    /*||(((next-addr)<30) && is_instructions(dyn, addr, next-addr))*/ ))
+                    (((next-addr)<15) && is_nops(dyn, addr, next-addr))))
                {
                    ok = 1;
                    if(dyn->insts[ninst].x64.has_callret && !dyn->insts[ninst].x64.has_next) {
@@ -299,7 +388,7 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
                    } else {
                        // need to find back that instruction to copy the caches, as previous version cannot be used anymore
                        // and pred table is not ready yet
-                        reset_n = get_first_jump(dyn, next);
+                        reset_n = get_first_jump_addr(dyn, next);
                    }
                    if (dyn->need_dump) dynarec_log(LOG_NONE, "Extend block %p, %s%p -> %p (ninst=%d, jump from %d)\n", dyn, dyn->insts[ninst].x64.has_callret ? "(opt. call) " : "", (void*)addr, (void*)next, ninst + 1, dyn->insts[ninst].x64.has_callret ? ninst : reset_n);
                } else if (next && (int)(next - addr) < BOX64ENV(dynarec_forward) && (getProtection(next) & PROT_READ) /*BOX64DRENV(dynarec_bigblock)>=stopblock*/) {
@@ -346,8 +435,7 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
        }
        if((ok>0) && dyn->insts[ninst].x64.has_callret)
            reset_n = -2;
-        if((ok>0) && reset_n==-1 && dyn->insts[ninst+1].purge_ymm)
-            PURGE_YMM();
+        PURGE_YMM();
        ++ninst;
        #if STEP == 0
        memset(&dyn->insts[ninst], 0, sizeof(instruction_native_t));
@@ -380,14 +468,22 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
            #endif
            ++ninst;
            NOTEST(x3);
+            #if defined (RV64) || defined(LA64)
            fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+            #else
+            fpu_purgecache(dyn, ninst, 0, x1, x2, x3, 0);
+            #endif
            jump_to_next(dyn, addr, 0, ninst, rex.is32bits);
            ok=0; need_epilog=0;
        }
    }
    if(need_epilog) {
        NOTEST(x3);
+        #if defined (RV64) || defined(LA64)
        fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+        #else
+        fpu_purgecache(dyn, ninst, 0, x1, x2, x3, 0);
+        #endif
        jump_to_epilog(dyn, ip, 0, ninst);  // no linker here, it's an unknown instruction
    }
    FINI;
@@ -12,16 +12,20 @@ void arm64_epilog(void) EXPORTDYN;
 void la64_next(void) EXPORTDYN;
 void la64_prolog(x64emu_t* emu, void* addr) EXPORTDYN;
 void la64_epilog(void) EXPORTDYN;
+void la64_epilog_fast(void) EXPORTDYN;
 #define native_next         la64_next
 #define native_prolog       la64_prolog
 #define native_epilog       la64_epilog
+#define native_epilog_fast  la64_epilog_fast
 #elif defined(RV64)
 void rv64_next(void) EXPORTDYN;
 void rv64_prolog(x64emu_t* emu, void* addr) EXPORTDYN;
 void rv64_epilog(void) EXPORTDYN;
+void rv64_epilog_fast(void) EXPORTDYN;
 #define native_next         rv64_next
 #define native_prolog       rv64_prolog
 #define native_epilog       rv64_epilog
+#define native_epilog_fast  rv64_epilog_fast
 #else
 #error Unsupported architecture
 #endif