Bumped version to v0.2.8

[RV64] Made Zba and Xthead* detection mutually exclusive (#1514 )
[ARM64_DYNAREC] Some small fixes to D9 E5 opcode
2026-11-08 17:39:19 -10:00 · 2024-05-21 17:33:16 +02:00 · 2024-05-21 15:35:48 +02:00 · 2024-05-21 14:35:45 +02:00 · 2024-05-21 13:34:31 +02:00 · 2024-05-21 11:54:08 +02:00
1472 changed files with 108926 additions and 19962 deletions
@@ -0,0 +1,17 @@
+---
+BasedOnStyle: WebKit
+AlignTrailingComments: true
+AllowShortBlocksOnASingleLine: true
+AllowShortCaseLabelsOnASingleLine: true
+AllowShortFunctionsOnASingleLine: true
+AllowShortIfStatementsOnASingleLine: true
+ColumnLimit: 0
+IndentWidth: 4
+SortIncludes: false
+MaxEmptyLinesToKeep: 2
+IndentCaseLabels: true
+AlignConsecutiveMacros: true
+WhitespaceSensitiveMacros: ['QUOTE']
+IfMacros: ['IFX', 'IFX2', 'IFXA', 'IFX_PENDOR0', 'IFXX', 'IFX2X', 'IFXN', 'UFLAG_IF', 'PASS2IF']
+UseTab: Never
+---
@@ -17,8 +17,12 @@ on:
        - RK3399
        - RK3588
        - RPI4ARM64
+        - RPI5ARM64
        - SD845
+        - SD888
+        - ADLINK
        - TEGRAX1
+        - ARM64
      build_type:
        description: 'Build type'
        required: true
@@ -45,7 +49,7 @@ jobs:
    runs-on: ubuntu-latest
    steps:
      - name: "Checkout Box64 Repository"
-        uses: actions/checkout@v2
+        uses: actions/checkout@v4

      - name: "Environment preparation"
        run: |
@@ -86,7 +90,7 @@ jobs:
          make -j$(nproc) VERBOSE=1

      - name: "Upload Artifact"
-        uses: actions/upload-artifact@v2
+        uses: actions/upload-artifact@v4
        with:
          name: box64-${{ github.event.inputs.platform }}-${{ github.event.inputs.build_type }}
          path: build/box64
@@ -27,7 +27,7 @@ jobs:
    strategy:
      fail-fast: false
      matrix:
-        platform: [X64, OTHER_ARM, RISCV, RPI4ARM64, RK3326, RK3399, RK3588, PHYTIUM, SD845]
+        platform: [X64, OTHER_ARM, RISCV, RPI4ARM64, RPI5ARM64, RK3326, RK3399, RK3588, PHYTIUM, SD845, SD888, ADLINK, ARM64, ANDROID, TERMUX, LARCH64]
        type: [Release, Trace]
        os: [ubuntu-latest]
        include:
@@ -41,15 +41,41 @@ jobs:
    runs-on: ${{ matrix.os }}
    steps:
      - name: "Checkout Box64 Repository"
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4

      - name: "Environment preparation"
        run: |
          sudo apt-get update
-          if [[ ${{ matrix.platform }} != 'X64' && ${{ matrix.platform }} != 'OTHER_ARM' && ${{ matrix.platform }} != 'RISCV' ]]; then
-            echo "BOX64_PLATFORM_MARCRO=-D${{ matrix.platform }}=1" >> $GITHUB_ENV
-            echo "BOX64_COMPILER=aarch64-linux-gnu-gcc" >> $GITHUB_ENV
-            sudo apt-get -y install git gcc-aarch64-linux-gnu cmake make python3
+          if [[ ${{ matrix.platform }} != 'X64' && ${{ matrix.platform }} != 'OTHER_ARM' && ${{ matrix.platform }} != 'RISCV' && ${{ matrix.platform }} != 'LARCH64' ]]; then
+            sudo apt-get -y install git cmake make python3
+            if [[ ${{ matrix.platform }} == 'ANDROID' || ${{ matrix.platform }} == 'TERMUX' ]]; then
+              wget https://dl.google.com/android/repository/android-ndk-r26b-linux.zip
+              unzip android-ndk-r26b-linux.zip
+              echo "BOX64_COMPILER=$PWD/android-ndk-r26b/toolchains/llvm/prebuilt/linux-x86_64/bin/aarch64-linux-android31-clang" >> $GITHUB_ENV
+              echo "BOX64_PLATFORM_MARCRO=-DANDROID=1 -DARM_DYNAREC=1 -DBAD_SIGNAL=1" >> $GITHUB_ENV
+              git clone https://github.com/termux/termux-docker.git
+              sudo cp -rf termux-docker/system/arm /system
+              sudo chown -R $(whoami):$(whoami) /system
+              sudo chmod 755 -R /system
+            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
+              sudo apt-get -y install p7zip
+              wget https://mirrors.utermux.dev/termux/termux-main/pool/main/liba/libandroid-sysv-semaphore/libandroid-sysv-semaphore_0.1_aarch64.deb
+              7z x libandroid-sysv-semaphore_0.1_aarch64.deb
+              tar -xf data.tar.xz
+              sudo cp -rf data /
+              sudo chmod 755 -R /data
+              sudo chown -R $(whoami):$(whoami) /data
+              echo "CFLAGS=-L/data/data/com.termux/files/usr/lib" >> $GITHUB_ENV
+              echo "IS_TERMUX=1" >> $GITHUB_ENV
+              echo "QEMU_SET_ENV=LD_LIBRARY_PATH=/data/data/com.termux/files/usr/lib" >> $GITHUB_ENV
+            else
+              echo "IS_TERMUX=0" >> $GITHUB_ENV
+            fi
          else
            if [[ ${{ matrix.platform }} == 'X64' ]]; then
              echo "BOX64_PLATFORM_MARCRO=-DLD80BITS=1 -DNOALIGN=1" >> $GITHUB_ENV
@@ -59,6 +85,21 @@ jobs:
              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
+            elif [[ ${{ matrix.platform }} == 'LARCH64' ]]; then
+              sudo mkdir /usr/local/larch
+              wget -O- -q https://github.com/loongson/build-tools/releases/download/2023.08.08/CLFS-loongarch64-8.1-x86_64-cross-tools-gcc-glibc.tar.xz | sudo tar -C /usr/local/larch --strip-components=1 --xz -xf -
+              sudo ln -sf /usr/local/larch/target /usr/loongarch64-linux-gnu
+              sudo cp -r /usr/local/larch/loongarch64-unknown-linux-gnu/lib/* /usr/loongarch64-linux-gnu/lib64/
+              for i in objdump objcopy strip; do
+                sudo ln -sf /usr/local/larch/bin/loongarch64-unknown-linux-gnu-$i /usr/bin/loongarch64-linux-gnu-$i
+              done
+              echo '/usr/local/larch/bin/loongarch64-unknown-linux-gnu-gcc -L/usr/local/larch/loongarch64-unknown-linux-gnu "$@"' | sudo tee /usr/bin/loongarch64-linux-gnu-gcc
+              echo '/usr/local/larch/bin/loongarch64-unknown-linux-gnu-g++ -L/usr/local/larch/loongarch64-unknown-linux-gnu "$@"' | sudo tee /usr/bin/loongarch64-linux-gnu-g++
+              sudo chmod 755 /usr/bin/loongarch64-linux-gnu-{gcc,g++}
+
+              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
            else
              echo BOX64_PLATFORM_MARCRO="-DARM_DYNAREC=ON" >> $GITHUB_ENV
              echo "BOX64_COMPILER=aarch64-linux-gnu-gcc" >> $GITHUB_ENV
@@ -85,30 +126,49 @@ jobs:
          mkdir build
          cd build
          cmake .. -DCMAKE_C_COMPILER=${{ env.BOX64_COMPILER }}\
+                   -DTERMUX=${{ env.IS_TERMUX }}\
                   ${{ env.BOX64_PLATFORM_MARCRO }}\
                   -DCMAKE_BUILD_TYPE=${{ env.BOX64_BUILD_TYPE }}\
                   -DHAVE_TRACE=${{ env.BOX64_HAVE_TRACE }}\
-                   -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON
+                   -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON\
+                   -DCI=${{ matrix.platform != 'ANDROID' }}
          make -j$(nproc) VERBOSE=1

      - name: "Test Box64"
        if: ${{ matrix.platform != 'TEGRAX1' }}
        run: |
          if [[ ${{ matrix.platform }} != 'X64' ]]; then
-            sudo apt-get -y install qemu-user
+            mkdir qemu9.0.0
+            wget -O- -q https://archive.archlinux.org/packages/q/qemu-user-static/qemu-user-static-9.0.0-1-x86_64.pkg.tar.zst | tar -I zstd -C qemu9.0.0 -xf -
+            sudo cp qemu9.0.0/usr/bin/* /usr/bin/
          fi

          cd build
          if [[ ${{ matrix.platform }} == 'RISCV' ]]; then
-            QEMU_LD_PREFIX=/usr/riscv64-linux-gnu/ ctest -j$(nproc) --rerun-failed --output-on-failure
-          elif [[ ${{ matrix.platform }} != 'X64' ]]; then
-            QEMU_LD_PREFIX=/usr/aarch64-linux-gnu/ ctest -j$(nproc) --rerun-failed --output-on-failure
+            INTERPRETER=qemu-riscv64-static QEMU_LD_PREFIX=/usr/riscv64-linux-gnu/ ctest -j$(nproc) --output-on-failure
+            INTERPRETER=qemu-riscv64-static QEMU_LD_PREFIX=/usr/riscv64-linux-gnu/ BOX64_DYNAREC=0 ctest -j$(nproc) --output-on-failure
+            INTERPRETER=qemu-riscv64-static QEMU_LD_PREFIX=/usr/riscv64-linux-gnu/ QEMU_CPU=rv64,zba=true,zbb=true,zbc=true,zbs=true ctest -j$(nproc) --output-on-failure
+            INTERPRETER=qemu-riscv64-static QEMU_LD_PREFIX=/usr/riscv64-linux-gnu/ QEMU_CPU=rv64,xtheadba=true,xtheadba=true,xtheadbb=true,xtheadbs=true,xtheadcondmov=true,xtheadmemidx=true,xtheadmempair=true,xtheadfmemidx=true,xtheadmac=true,xtheadfmv=true ctest -j$(nproc) --output-on-failure
+          elif [[ ${{ matrix.platform }} == 'LARCH64' ]]; then
+            INTERPRETER=qemu-loongarch64-static QEMU_LD_PREFIX=/usr/loongarch64-linux-gnu/ BOX64_DYNAREC_LA64NOEXT=1 ctest -j$(nproc) --repeat until-pass:20 --output-on-failure
+            INTERPRETER=qemu-loongarch64-static QEMU_LD_PREFIX=/usr/loongarch64-linux-gnu/ BOX64_DYNAREC_TEST=2 BOX64_DYNAREC_LA64NOEXT=1 ctest -j$(nproc) --repeat until-pass:20 --output-on-failure
+            INTERPRETER=qemu-loongarch64-static QEMU_LD_PREFIX=/usr/loongarch64-linux-gnu/ BOX64_DYNAREC=0 ctest -j$(nproc) --repeat until-pass:20 --output-on-failure
+          elif [[ ${{ matrix.platform }} == 'ANDROID' ]]; then
+            INTERPRETER=qemu-aarch64-static QEMU_LD_PREFIX=/system/lib64 BOX64_DYNAREC=0 ctest -j$(nproc) --repeat until-pass:20 --output-on-failure
+            INTERPRETER=qemu-aarch64-static QEMU_LD_PREFIX=/system/lib64 ctest -j$(nproc) --repeat until-pass:20 --output-on-failure
+          elif [[ ${{ matrix.platform }} == 'TERMUX' ]]; then
+            INTERPRETER=qemu-aarch64-static QEMU_SET_ENV=LD_LIBRARY_PATH=/data/data/com.termux/files/usr/lib QEMU_LD_PREFIX=/system/lib64:/data/data/com.termux/files/usr/lib BOX64_DYNAREC=0 ctest -j$(nproc) --repeat until-pass:20 --output-on-failure
+            INTERPRETER=qemu-aarch64-static QEMU_SET_ENV=LD_LIBRARY_PATH=/data/data/com.termux/files/usr/lib QEMU_LD_PREFIX=/system/lib64:/data/data/com.termux/files/usr/lib ctest -j$(nproc) --repeat until-pass:20 --output-on-failure
+          elif [[ ${{ matrix.platform }} != 'X64' ]]; then # AArch64
+            INTERPRETER=qemu-aarch64-static QEMU_LD_PREFIX=/usr/aarch64-linux-gnu/ BOX64_DYNAREC=0 ctest -j$(nproc) --output-on-failure
+            INTERPRETER=qemu-aarch64-static QEMU_LD_PREFIX=/usr/aarch64-linux-gnu/ ctest -j$(nproc) --output-on-failure
+            INTERPRETER=qemu-aarch64-static QEMU_LD_PREFIX=/usr/aarch64-linux-gnu/ BOX64_DYNAREC_TEST=2 ctest -j$(nproc) --output-on-failure
          else
-            ctest -j$(nproc) --rerun-failed --output-on-failure
+            ctest -j$(nproc) --output-on-failure
          fi

      - name: "Upload Artifact"
-        uses: actions/upload-artifact@v3
+        uses: actions/upload-artifact@v4
        with:
          name: box64-${{ matrix.platform }}-${{ matrix.type }}
          path: build/box64
@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 3.4)
+cmake_minimum_required(VERSION 3.13)

 cmake_policy(SET CMP0065 NEW)

@@ -7,73 +7,89 @@ SET(CMAKE_CXX_FLAGS_DEBUG  "-O0 -g")

 option(RPI3ARM64 "Set to ON if targeting an RaspberryPI3 device with multiarch arm64 and armhf" ${RPI3ARM64})
 option(RPI4ARM64 "Set to ON if targeting an RaspberryPI4 device with multiarch arm64 and armhf" ${RPI4ARM64})
+option(RPI5ARM64 "Set to ON if targeting an RaspberryPi5 device with multiarch arm64 and armhf" ${RPI5ARM64})
 option(RK3326 "Set to ON if targeting an Rockchip RK3326 based device" ${RK3326})
 option(RK3399 "Set to ON if targeting an Rockchip RK3399 based device" ${RK3399})
 option(RK3588 "Set to ON if targeting an Rockchip RK3588(S) based device" ${RK3588})
 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(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(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(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})
 option(LARCH64 "Set to ON if targeting an Loongarch64 based device" ${LARCH64})
 option(RV64 "Set to ON if targeting an RISC-V RV64GC based device" ${RV64})
 option(PPC64LE "Set to ON if targeting an PowerPC 64 LE based device" ${PPC64LE})
 option(LX2160A "Set to ON if targeting an LX2160A based device" ${LX2160A})
+option(ARM64 "Set to ON if targeting a generic ARM64 based device" ${ARM64})
+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(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(SAVE_MEM "Set to ON to build dynarec with some slower memory saving optimisations" ${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(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})

+if(TERMUX)
+    set(TERMUX_PATH "/data/data/com.termux/files")
+    set(ANDROID ON CACHE BOOL "")
+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 "")
 endif()
 if(RV64)
    set(LD80BITS OFF CACHE BOOL "")
    set(NOALIGN OFF CACHE BOOL "")
    set(ARM_DYNAREC OFF CACHE BOOL "")
    set(RV64_DYNAREC ON CACHE BOOL "")
-    set(PAGE16K OFF CACHE BOOL "")
+    set(LARCH64_DYNAREC OFF CACHE BOOL "")
 endif()
 if(PPC64LE)
    set(LD80BITS OFF CACHE BOOL "")
    set(NOALIGN OFF CACHE BOOL "")
    set(ARM_DYNAREC OFF CACHE BOOL "")
    set(RV64_DYNAREC OFF CACHE BOOL "")
-    set(PAGE16K OFF CACHE BOOL "")
+    set(LARCH64_DYNAREC OFF CACHE BOOL "")
 endif()
-if(RK3399 OR RK3588 OR ODROIDN2 OR RPI3ARM64 OR RPI4ARM64 OR RK3326 OR TEGRAX1 OR PHYTIUM OR SD845 OR LX2160A OR M1)
+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 SD888 OR SD8G2 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 "")
 endif()
-if(RK3399 OR RK3588 OR ODROIDN2 OR RPI3ARM64 OR RPI4ARM64 OR RK3326 OR TEGRAX1 OR PHYTIUM OR SD845 OR LX2160A)
-    set(PAGE16K OFF CACHE BOOL "")
+if(RK3399 OR ODROIDN2 OR RPI3ARM64 OR RPI4ARM64 OR RPI5ARM64)
+    set(SAVE_MEM ON CACHE BOOL "")
 endif()
 if(RK3588)
    set(BAD_SIGNAL ON CACHE BOOL "")
 endif()
-if(M1 OR LARCH64)
-    set(PAGE16K ON CACHE BOOL "")
-endif()
 if(SW64)
    set(LD80BITS OFF CACHE BOOL "")
    set(NOALIGN OFF CACHE BOOL "")
-    set(PAGE8K ON CACHE BOOL "")
+endif()
+if(ANDROID)
+    set(NOLOADADDR ON CACHE BOOL "")
+    set(BAD_SIGNAL ON CACHE BOOL "")
 endif()

 option(LD80BITS "Set to ON if host device have 80bits long double (i.e. i386)" ${LD80BITS})
 option(NOALIGN "Set to ON if host device doesn't need re-align (i.e. i386)" ${NOALIGN})
 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(PAGE8K "Set to ON if host device have PageSize of 8K (instead of 4K)" ${PAGE8K})
-option(PAGE16K "Set to ON if host device have PageSize of 16K (instead of 4K)" ${PAGE16K})
-option(PAGE64K "Set to ON if host device have PageSize of 64K (instead of 4K)" ${PAGE64K})
+option(LARCH64_DYNAREC "Set to ON to use LOONGARCH64 Dynamic Recompilation" ${LARCH64_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})
@@ -115,6 +131,11 @@ elseif(RPI4ARM64)
    add_definitions(-DRPI4ARM64)
    add_definitions(-pipe -march=armv8-a+crc -mtune=cortex-a72)
    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crc -mtune=cortex-a72")
+elseif(RPI5ARM64)
+    add_definitions(-DRPI)
+    add_definitions(-DRPI5ARM64)
+    add_definitions(-pipe -march=armv8.2-a+crc+crypto+fp16+rcpc+dotprod -mtune=cortex-a76)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8.2-a+crc+crypto+fp16+rcpc+dotprod -mtune=cortex-a76")
 elseif(RK3326)
    add_definitions(-DRK3326)
    add_definitions(-pipe -march=armv8-a+crc+simd+crypto -mcpu=cortex-a35+crypto)
@@ -136,20 +157,40 @@ elseif(TEGRAX1)
    add_definitions(-DTEGRAX1)
    add_definitions(-pipe -march=armv8-a+crc+simd+crypto -mcpu=cortex-a57+crypto)
    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crc+simd+crypto -mcpu=cortex-a57+crypto")
+elseif(TEGRA_T194)
+    add_definitions(-DTEGRA_T194)
+    add_definitions(-pipe -march=armv8.2-a+fp16+simd+crypto+predres -mcpu=cortex-a76+crypto -mtune=cortex-a76)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8.2-a+fp16+simd+crypto+predres -mcpu=cortex-a76+crypto -mtune=cortex-a76")
+elseif(TEGRA_T234)
+    add_definitions(-DTEGRA_T234)
+    add_definitions(-pipe -march=armv8.2-a+fp16+simd+crypto+predres -mcpu=cortex-a78ae+crypto -mtune=cortex-a78ae)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8.2-a+fp16+simd+crypto+predres -mcpu=cortex-a78ae+crypto -mtune=cortex-a78ae")
 elseif(PHYTIUM)
-    add_definitions(-DPHYSIUM)
+    add_definitions(-DPHYTIUM)
    add_definitions(-pipe -march=armv8.1-a+crc+simd+crypto)
    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8.1-a+crc+simd+crypto+aes")
 elseif(SD845)
    add_definitions(-DSD845)
    add_definitions(-pipe -march=armv8.2-a+simd+crypto -mtune=cortex-a75.cortex-a55)
    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8.2-a+simd+crypto -mtune=cortex-a75.cortex-a55")
+elseif(SD888)
+    add_definitions(-DSD888)
+    add_definitions(-pipe -march=armv8.4-a+simd+crypto)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8.4-a+simd+crypto")
+elseif(SD8G2)
+    add_definitions(-DSD8G2)
+    add_definitions(-pipe -march=armv9-a+i8mm+sm4+sha3+rcpc+crypto+nosve+nosve2)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv9-a+i8mm+sm4+sha3+rcpc+crypto+nosve+nosve2")
+elseif(ADLINK)
+    add_definitions(-DADLINK)
+    add_definitions(-pipe -mcpu=neoverse-n1 -fuse-ld=gold -fuse-linker-plugin)
+    set(CMAKE_ASM_FLAGS  "-pipe -mcpu=neoverse-n1")
 elseif(M1)
    add_definitions(-DM1)
    add_definitions(-pipe -march=armv8.5-a+simd+crypto)
    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8.5-a+simd+crypto")
 elseif(LARCH64)
-    add_definitions(-DLA464)
+    add_definitions(-DLA64)
    add_definitions(-pipe -march=loongarch64)
    set(CMAKE_ASM_FLAGS  "-pipe -march=loongarch64")
 elseif(RV64)
@@ -164,24 +205,33 @@ elseif(LX2160A)
    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crypto+crc -mcpu=cortex-a72+crypto")
 elseif(SW64)
    add_definitions(-DSW64)
+elseif(ARM64)
+    #add_definitions(-pipe -march=native)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+simd")
 elseif(ARM_DYNAREC)
-    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crc+simd+crypto")
+    if(NOT ANDROID)
+        set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crc+simd+crypto")
+    endif()
+endif()
+if(ANDROID)
+    add_definitions(-DANDROID)
+endif()
+if(TERMUX)
+    add_definitions(-DTERMUX)
 endif()
-
 if(BAD_SIGNAL)
    add_definitions(-DBAD_SIGNAL)
 endif()
-
-if(PAGE8K)
-    add_definitions(-DPAGE8K)
+if(SAVE_MEM)
+    add_definitions(-DSAVE_MEM)
 endif()
-
-if(PAGE16K)
-    add_definitions(-DPAGE16K)
-endif()
-
-if(PAGE64K)
-    add_definitions(-DPAGE64K)
+if(STATICBUILD)
+    #-fno-pic -mcmodel=large
+    add_definitions(-DSTATICBUILD)
+    add_definitions(-Wno-deprecated-declarations)   #by-design, deprecated function can be used
+    set(BUILD_SHARED_LIBS OFF)
+    set(CMAKE_EXE_LINKER_FLAGS "-static")
+    set(HAVE_TRACE OFF)
 endif()

 if(NOGIT)
@@ -221,6 +271,12 @@ elseif(RV64_DYNAREC)
    enable_language(ASM)
    include_directories("${BOX64_ROOT}/src/dynarec/rv64")
    set(DYNAREC ON)
+elseif(LARCH64_DYNAREC)
+    add_definitions(-DDYNAREC)
+    add_definitions(-DLA64)
+    enable_language(ASM)
+    include_directories("${BOX64_ROOT}/src/dynarec/la64")
+    set(DYNAREC ON)
 else()
    set(DYNAREC OFF)
 endif()
@@ -253,21 +309,23 @@ set_source_files_properties(

 set(ELFLOADER_SRC
    "${BOX64_ROOT}/src/main.c"
+    "${BOX64_ROOT}/src/core.c"
    "${BOX64_ROOT}/src/box64context.c"
    "${BOX64_ROOT}/src/build_info.c"
    "${BOX64_ROOT}/src/custommem.c"
-    "${BOX64_ROOT}/src/mallochook.c"
-    "${BOX64_ROOT}/src/steam.c"
    "${BOX64_ROOT}/src/dynarec/dynarec.c"
-    "${BOX64_ROOT}/src/elfs/elfdwarf_private.c"
    "${BOX64_ROOT}/src/elfs/elfloader.c"
    "${BOX64_ROOT}/src/elfs/elfparser.c"
+    "${BOX64_ROOT}/src/elfs/elfhash.c"
    "${BOX64_ROOT}/src/elfs/elfload_dump.c"
+    "${BOX64_ROOT}/src/elfs/elfdwarf_private.c"
+    "${BOX64_ROOT}/src/emu/x64compstrings.c"
    "${BOX64_ROOT}/src/emu/x64emu.c"
    "${BOX64_ROOT}/src/emu/x64int3.c"
    "${BOX64_ROOT}/src/emu/x87emu_private.c"
    "${BOX64_ROOT}/src/emu/x64primop.c"
    "${BOX64_ROOT}/src/emu/x64run_private.c"
+    "${BOX64_ROOT}/src/emu/x64shaext.c"
    "${BOX64_ROOT}/src/emu/x64syscall.c"
    "${BOX64_ROOT}/src/emu/x86syscall.c"
    "${BOX64_ROOT}/src/emu/x64tls.c"
@@ -275,13 +333,9 @@ set(ELFLOADER_SRC
    "${BOX64_ROOT}/src/librarian/librarian.c"
    "${BOX64_ROOT}/src/librarian/library.c"
    "${BOX64_ROOT}/src/librarian/dictionnary.c"
-    "${BOX64_ROOT}/src/librarian/globalsymbols.c"
    "${BOX64_ROOT}/src/librarian/symbols.c"
    "${BOX64_ROOT}/src/libtools/auxval.c"
-    "${BOX64_ROOT}/src/libtools/obstack.c"
    "${BOX64_ROOT}/src/libtools/myalign.c"
-    "${BOX64_ROOT}/src/libtools/sdl1rwops.c"
-    "${BOX64_ROOT}/src/libtools/sdl2rwops.c"
    "${BOX64_ROOT}/src/libtools/signals.c"
    "${BOX64_ROOT}/src/libtools/threads.c"
    "${BOX64_ROOT}/src/tools/bitutils.c"
@@ -289,13 +343,28 @@ set(ELFLOADER_SRC
    "${BOX64_ROOT}/src/tools/bridge.c"
    "${BOX64_ROOT}/src/tools/callback.c"
    "${BOX64_ROOT}/src/tools/my_cpuid.c"
-    "${BOX64_ROOT}/src/tools/gtkclass.c"
-    "${BOX64_ROOT}/src/tools/pathcoll.c"
    "${BOX64_ROOT}/src/tools/fileutils.c"
+    "${BOX64_ROOT}/src/tools/pathcoll.c"
+    "${BOX64_ROOT}/src/tools/rbtree.c"
    "${BOX64_ROOT}/src/tools/rcfile.c"
    "${BOX64_ROOT}/src/tools/wine_tools.c"
    "${BOX64_ROOT}/src/wrapped/generated/wrapper.c"
 )
+if(NOT STATICBUILD)
+    list(APPEND ELFLOADER_SRC
+        "${BOX64_ROOT}/src/mallochook.c"
+        "${BOX64_ROOT}/src/steam.c"
+        "${BOX64_ROOT}/src/libtools/sdl1rwops.c"
+        "${BOX64_ROOT}/src/libtools/sdl2rwops.c"
+        "${BOX64_ROOT}/src/tools/gtkclass.c"
+        "${BOX64_ROOT}/src/librarian/globalsymbols.c"
+        )
+endif()
+if(NOT ANDROID)
+    list(APPEND ELFLOADER_SRC
+        "${BOX64_ROOT}/src/libtools/obstack.c"
+    )
+endif()

 set(INTERPRETER
    "${BOX64_ROOT}/src/emu/x64run.c"
@@ -303,12 +372,15 @@ set(INTERPRETER
    "${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/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"
@@ -325,6 +397,20 @@ set(INTERPRETER
    "${BOX64_ROOT}/src/emu/x64runf30f.c"
 )

+if(STATICBUILD)
+set(WRAPPEDS
+    "${BOX64_ROOT}/src/wrapped/wrappedldlinux.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibbsd.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibc.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibcmusl.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibdl.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibm.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedutil.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibresolv.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibrt.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibpthread.c"
+)
+else()
 set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedalure.c"
    "${BOX64_ROOT}/src/wrapped/wrappedalut.c"
@@ -332,7 +418,9 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedatkbridge.c"
    "${BOX64_ROOT}/src/wrapped/wrappedatomic.c"
    "${BOX64_ROOT}/src/wrapped/wrappedatspi.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedayatanaappindicator3.c"
    "${BOX64_ROOT}/src/wrapped/wrappedbz2.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedcap.c"
    "${BOX64_ROOT}/src/wrapped/wrappedcairo.c"
    "${BOX64_ROOT}/src/wrapped/wrappedcairogobject.c"
    "${BOX64_ROOT}/src/wrapped/wrappedcrashhandler.c"
@@ -341,6 +429,9 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedcurl.c"
    "${BOX64_ROOT}/src/wrapped/wrappeddbus.c"
    "${BOX64_ROOT}/src/wrapped/wrappeddbusglib1.c"
+    "${BOX64_ROOT}/src/wrapped/wrappeddbusmenuglib.c"
+    "${BOX64_ROOT}/src/wrapped/wrappeddecor0.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedevent21.c"
    "${BOX64_ROOT}/src/wrapped/wrappedexpat.c"
    "${BOX64_ROOT}/src/wrapped/wrappedfaudio.c"
    "${BOX64_ROOT}/src/wrapped/wrappedflac.c"
@@ -354,26 +445,40 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedgdkpixbuf2.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgdkx112.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgio2.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedglesv2.c"
    "${BOX64_ROOT}/src/wrapped/wrappedglib2.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgmodule2.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgmp.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgnutls.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgobject2.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgomp.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgssapi.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgssapikrb5.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstallocators.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgstapp.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgstaudio.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgstbase.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstcheck.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstcontroller.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstfft.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgstgl.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstnet.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstpbutils.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgstreamer.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstriff.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstrtp.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstrtsp.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstsdp.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgsttag.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgstvideo.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgthread2.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgtk3.c"
    "${BOX64_ROOT}/src/wrapped/wrappedgtkx112.c"
    "${BOX64_ROOT}/src/wrapped/wrappedkrb5.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlcms2.c"
    "${BOX64_ROOT}/src/wrapped/wrappedldlinux.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibasound.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"
@@ -383,6 +488,7 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedlibegl.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibform.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibformw.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibformw6.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibfuse.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibgl.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibglu.c"
@@ -421,6 +527,7 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedlibx11xcb.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibxau.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibxcb.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibxcbcursor.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibxcbdri2.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibxcbdri3.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibxcbglx.c"
@@ -436,6 +543,7 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedlibxcbsync.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibxcbutil.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibxcbxinerama.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibxcbxinput.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibxcbxfixes.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibxcbxkb.c"
    "${BOX64_ROOT}/src/wrapped/wrappedlibxcbxtest.c"
@@ -465,13 +573,16 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappednss3.c"
    "${BOX64_ROOT}/src/wrapped/wrappednssutil3.c"
    "${BOX64_ROOT}/src/wrapped/wrappedopenal.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedopencl.c"
    "${BOX64_ROOT}/src/wrapped/wrappedpango.c"
    "${BOX64_ROOT}/src/wrapped/wrappedpangocairo.c"
    "${BOX64_ROOT}/src/wrapped/wrappedpangoft2.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedpam.c"
    "${BOX64_ROOT}/src/wrapped/wrappedplc4.c"
    "${BOX64_ROOT}/src/wrapped/wrappedplds4.c"
    "${BOX64_ROOT}/src/wrapped/wrappedpng16.c"
    "${BOX64_ROOT}/src/wrapped/wrappedpulse.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedpulsemainloopglib.c"
    "${BOX64_ROOT}/src/wrapped/wrappedpulsesimple.c"
    "${BOX64_ROOT}/src/wrapped/wrappedsdl1.c"
    "${BOX64_ROOT}/src/wrapped/wrappedsdl1image.c"
@@ -506,6 +617,7 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedxinerama.c"
    "${BOX64_ROOT}/src/wrapped/wrappedxkbcommon.c"
    "${BOX64_ROOT}/src/wrapped/wrappedxkbcommonx11.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedxkbregistry.c"
    "${BOX64_ROOT}/src/wrapped/wrappedxml2.c"
    "${BOX64_ROOT}/src/wrapped/wrappedxslt.c"
    "${BOX64_ROOT}/src/wrapped/wrappedldapr.c"
@@ -513,11 +625,34 @@ set(WRAPPEDS
    "${BOX64_ROOT}/src/wrapped/wrappedvulkan.c"
    "${BOX64_ROOT}/src/wrapped/wrappedxshmfence.c"
    "${BOX64_ROOT}/src/wrapped/wrappedd3dadapter9.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedicuuc64.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedicui18n64.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedicuuc66.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedicui18n66.c"
    "${BOX64_ROOT}/src/wrapped/wrappedicuuc67.c"
    "${BOX64_ROOT}/src/wrapped/wrappedicui18n67.c"
    "${BOX64_ROOT}/src/wrapped/wrappedicuuc72.c"
    "${BOX64_ROOT}/src/wrapped/wrappedicui18n72.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedicuuc73.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedicui18n73.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedicuuc74.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedicui18n74.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedidn2.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedanl.c"
 )
+endif()
+if(ANDROID)
+    list(APPEND WRAPPEDS
+        "${BOX64_ROOT}/src/wrapped/wrappedandroidshmem.c"
+    )
+endif()
+if(TERMUX)
+    list(APPEND WRAPPEDS
+        "${BOX64_ROOT}/src/wrapped/wrappediconv.c"
+        "${BOX64_ROOT}/src/wrapped/wrappedtermuxexec.c"
+        "${BOX64_ROOT}/src/wrapped/wrappedandroidsupport.c"
+    )
+endif()

 # If BOX64_ROOT contains a ".c", the build breaks...
 string(REPLACE ".c" "_private.h" MODROOT ${BOX64_ROOT})
@@ -532,16 +667,18 @@ endforeach()

 set(WRAPPER "${BOX64_ROOT}/src/wrapped/generated/wrapper.c" "${BOX64_ROOT}/src/wrapped/generated/wrapper.h")

-add_custom_command(
-    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" "--"
-    ${WRAPPEDS_HEAD}
-    MAIN_DEPENDENCY "${BOX64_ROOT}/rebuild_wrappers.py"
-    DEPENDS ${WRAPPEDS} ${WRAPPEDS_HEAD}
-    BYPRODUCTS ${WRAPPER}
-)
+if(NOT CI)
+    add_custom_command(
+        OUTPUT "${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" "--"
+        ${WRAPPEDS_HEAD}
+        MAIN_DEPENDENCY "${BOX64_ROOT}/rebuild_wrappers.py"
+        DEPENDS ${WRAPPEDS} ${WRAPPEDS_HEAD}
+        BYPRODUCTS ${WRAPPER}
+    )
+endif ()

 #add_custom_command(
 #    OUTPUT "${BOX64_ROOT}/src/dynarec/last_run.txt"
@@ -568,6 +705,7 @@ if(ARM_DYNAREC)
        ${DYNAREC_SRC}

        "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_functions.c"
+        "${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"

@@ -589,9 +727,11 @@ if(ARM_DYNAREC)
    #"${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_65.c"
    "${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"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_6764_32.c"
    "${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"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_da.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_db.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_dc.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_dd.c"
@@ -599,6 +739,8 @@ if(ARM_DYNAREC)
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_df.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_f0.c"
    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_660f.c"
+    "${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"
@@ -637,9 +779,10 @@ if(RV64_DYNAREC)
    #"${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_65.c"
    "${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_32.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_d8.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_d9.c"
-    #"${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_da.c"
+    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_da.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_db.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_dc.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_dd.c"
@@ -647,6 +790,8 @@ if(RV64_DYNAREC)
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_df.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_f0.c"
    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_660f.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"
@@ -654,6 +799,37 @@ if(RV64_DYNAREC)
    )
 endif()

+if(LARCH64_DYNAREC)
+    set(DYNAREC_SRC
+        ${DYNAREC_SRC}
+
+        "${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"
+
+        "${BOX64_ROOT}/src/dynarec/la64/la64_prolog.S"
+        "${BOX64_ROOT}/src/dynarec/la64/la64_epilog.S"
+        "${BOX64_ROOT}/src/dynarec/la64/la64_next.S"
+        "${BOX64_ROOT}/src/dynarec/la64/la64_lock.S"
+    )
+
+    set(DYNAREC_PASS
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_helper.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_emit_tests.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_emit_math.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_emit_shift.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_emit_logic.c"
+    "${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_f30f.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_660f.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_f0.c"
+    "${BOX64_ROOT}/src/dynarec/la64/dynarec_la64_f20f.c"
+    )
+endif()
+
 if(DYNAREC)
    set(DYNAREC_PASS
        "${BOX64_ROOT}/src/wrapped/generated/wrapper.h"
@@ -711,9 +887,18 @@ add_dependencies(${BOX64} WRAPPERS)
 #add_dependencies(${BOX64} PRINTER)
 #target_link_libraries(${BOX64} c m dl rt pthread resolv)
 if(STATICBUILD)
-    set_target_properties(${BOX64} PROPERTIES LINK_FLAGS "-static -Wl,--no-as-needed -Wl,--whole-archive -Wl,--allow-multiple-definition -lm -ldl -lrt -lpthread -lresolv -lc -Wl,-defsym,_DYNAMIC=0 -pthread")
+    #set_target_properties(${BOX64} PROPERTIES LINK_FLAGS "-fuse-ld=gold -static -Wl,--no-relax -Wl,--allow-multiple-definition -ldl -lrt -lpthread -lresolv -lc -lutil -lm -Wl,-defsym,_DYNAMIC=0 -pthread")
+    target_link_libraries(${BOX64} c m dl pthread resolv util bsd c)
 else()
-    set_target_properties(${BOX64} PROPERTIES LINK_FLAGS "-Wl,--no-as-needed -lc -lm -ldl -lrt -lpthread -lresolv -Wl,--as-needed -pthread")
+    if(ANDROID)
+        if(TERMUX)
+        target_link_libraries(${BOX64} c m dl android-sysv-semaphore)
+        else()
+        target_link_libraries(${BOX64} c m dl)
+        endif()
+    else()
+	    set_target_properties(${BOX64} PROPERTIES LINK_FLAGS "-Wl,--no-as-needed -lc -lm -ldl -lrt -lpthread -lresolv -lutil -Wl,--as-needed -pthread")
+    endif()
 endif()
 if(DYNAREC)
    target_link_libraries(${BOX64} dynarec)
@@ -722,7 +907,7 @@ target_link_libraries(${BOX64} interpreter)

 if(${CMAKE_VERSION} VERSION_LESS "3.13")
    if(NOT NOLOADADDR)
-        if(CMAKE_C_COMPILER_ID STREQUAL "Clang")
+        if(CMAKE_C_COMPILER_ID STREQUAL "Clang" OR WITH_MOLD)
            set_target_properties(${BOX64} PROPERTIES LINK_FLAGS "-Wl,--image-base=${BOX64_ELF_ADDRESS}")
        else()
            set_target_properties(${BOX64} PROPERTIES LINK_FLAGS "-Wl,-Ttext-segment,${BOX64_ELF_ADDRESS}")
@@ -731,7 +916,7 @@ if(${CMAKE_VERSION} VERSION_LESS "3.13")
 else()
    # If symbols are missing, try this: target_link_options(${BOX64} PUBLIC -rdynamic)
    if(NOT NOLOADADDR)
-        if(CMAKE_C_COMPILER_ID STREQUAL "Clang")
+        if(CMAKE_C_COMPILER_ID STREQUAL "Clang" OR WITH_MOLD)
            target_link_options(${BOX64} PUBLIC LINKER:--image-base=${BOX64_ELF_ADDRESS})
        else()
            target_link_options(${BOX64} PUBLIC LINKER:-Ttext-segment,${BOX64_ELF_ADDRESS})
@@ -739,28 +924,68 @@ else()
    endif()
 endif()

-string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "i686"  _x86)
-string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "x86_64"  _x86_64)
-string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "aarch64"  _aarch64)
-string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "riscv64"  _riscv64)
+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)

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

 if(NOT _x86 AND NOT _x86_64)
-  install(TARGETS ${BOX64}
+  if(NOT TERMUX)
+   install(TARGETS ${BOX64}
    RUNTIME DESTINATION bin)
+  else()
+   install(TARGETS ${BOX64}
+    RUNTIME DESTINATION ${TERMUX_PATH}/usr/bin)
+  endif()
  if(NOT NO_CONF_INSTALL)
    configure_file(system/box64.conf.cmake system/box64.conf)
-    install(FILES ${CMAKE_BINARY_DIR}/system/box64.conf DESTINATION /etc/binfmt.d/)
-    install(FILES ${CMAKE_SOURCE_DIR}/system/box64.box64rc DESTINATION /etc/)
+	  if(NOT TERMUX)
+    		install(FILES ${CMAKE_BINARY_DIR}/system/box64.conf DESTINATION /etc/binfmt.d/)
+    		install(FILES ${CMAKE_SOURCE_DIR}/system/box64.box64rc DESTINATION /etc/)
+	  else()
+		install(FILES ${CMAKE_BINARY_DIR}/system/box64.conf DESTINATION ${TERMUX_PATH}/usr/etc/binfmt.d/)
+		install(FILES ${CMAKE_SOURCE_DIR}/system/box64.box64rc DESTINATION ${TERMUX_PATH}/usr/etc/)
+	  endif()
+
  endif()
+  set(INSTALL_PATH "/usr/lib/x86_64-linux-gnu/")
  if(NOT NO_LIB_INSTALL)
-    install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libstdc++.so.5 DESTINATION /usr/lib/x86_64-linux-gnu/)
-    install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libstdc++.so.6 DESTINATION /usr/lib/x86_64-linux-gnu/)
-    install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libgcc_s.so.1 DESTINATION /usr/lib/x86_64-linux-gnu/)
-    install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libpng12.so.0 DESTINATION /usr/lib/x86_64-linux-gnu/)
+	if(NOT TERMUX)
+    		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libstdc++.so.5 DESTINATION ${INSTALL_PATH})
+    		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libstdc++.so.6 DESTINATION ${INSTALL_PATH})
+    		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libgcc_s.so.1 DESTINATION ${INSTALL_PATH})
+    		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libpng12.so.0 DESTINATION ${INSTALL_PATH})
+    		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libcrypto.so.1.1 DESTINATION ${INSTALL_PATH})
+    		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libssl.so.1.1 DESTINATION ${INSTALL_PATH})
+    		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libunwind.so.8 DESTINATION ${INSTALL_PATH})
+            install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libmbedx509.so.1 DESTINATION ${INSTALL_PATH})
+            install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libmbedx509.so.0 DESTINATION ${INSTALL_PATH})
+            install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libmbedtls.so.14 DESTINATION ${INSTALL_PATH})
+            install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libmbedtls.so.12 DESTINATION ${INSTALL_PATH})
+            install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libmbedcrypto.so.7 DESTINATION ${INSTALL_PATH})
+            install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libmbedcrypto.so.3 DESTINATION ${INSTALL_PATH})
+            # Create symlinks
+            install(CODE "execute_process(COMMAND ${CMAKE_COMMAND} -E create_symlink ${INSTALL_PATH}libssl.so.1.1 ${INSTALL_PATH}libssl.so.1.0.0)")
+            install(CODE "execute_process(COMMAND ${CMAKE_COMMAND} -E create_symlink ${INSTALL_PATH}libcrypto.so.1.1 ${INSTALL_PATH}libcrypto.so.1.0.0)")
+	else()
+		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libstdc++.so.5 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libstdc++.so.6 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libgcc_s.so.1 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libpng12.so.0 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libcrypto.so.1.1 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libssl.so.1.1 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libunwind.so.8 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libmbedx509.so.1 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+        install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libmbedx509.so.0 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+        install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libmbedtls.so.14 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+        install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libmbedtls.so.12 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+        install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libmbedcrypto.so.7 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+        install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libmbedcrypto.so.3 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+	endif()
   endif()
 endif()

@@ -778,7 +1003,11 @@ set(CPACK_GENERATOR "DEB")
 set(CPACK_DEBIAN_PACKAGE_MAINTAINER "ptitSeb")
 set(CPACK_PACKAGE_CONTACT "[email protected]")
 set(CPACK_PACKAGE_DESCRIPTION "Box64 - Linux Userspace x86_64 Emulator with a twist")
-set(CPACK_DEBIAN_PACKAGE_DEPENDS "libc6")
+if(NOT TERMUX)
+	set(CPACK_DEBIAN_PACKAGE_DEPENDS "libc6")
+else()
+	set(CPACK_DEBIAN_PACKAGE_DEPENDS "libandroid-sysv-semaphore")
+endif()
 set(CPACK_PACKAGE_HOMEPAGE_URL, "https://box86.org")
 file(STRINGS "${BOX64_ROOT}/src/box64version.h" TEMP_STRING REGEX "BOX64_MAJOR\\s*")
 string(REGEX REPLACE "BOX64_MAJOR" "" TEMP_STRING ${TEMP_STRING})
@@ -792,9 +1021,15 @@ string(REGEX MATCH "[0-9]" BOX64_REVISION ${TEMP_STRING})
 set(CPACK_PACKAGE_VERSION_MAJOR ${BOX64_MAJOR})
 set(CPACK_PACKAGE_VERSION_MINOR ${BOX64_MINOR})
 set(CPACK_PACKAGE_VERSION_PATCH ${BOX64_REVISION})
+if(NOT TERMUX)
 set(CPACK_DEBIAN_PACKAGE_CONTROL_EXTRA "${CMAKE_CURRENT_SOURCE_DIR}/postinst")
+endif()
 if(_aarch64)
-    set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "arm64")
+	if(TERMUX)
+		set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "aarch64")
+	else()
+    		set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "arm64")
+	endif()
 elseif(__riscv64)
    set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "riscv64")
 elseif(_x86_64)
@@ -803,9 +1038,17 @@ else()
    #probably wrong...
    execute_process(COMMAND "dpkg --print-architecture" OUTPUT_VARIABLE CPACK_DEBIAN_PACKAGE_ARCHITECTURE)
 endif()
-set(CPACK_DEBIAN_FILE_NAME "${BOX64}-${BOX64_MAJOR}.${BOX64_MINOR}.${BOX64_REVISION}_Linux-${CPACK_DEBIAN_PACKAGE_ARCHITECTURE}.deb")
+if(TERMUX)
+set(CPACK_PACKAGING_INSTALL_PREFIX "${TERMUX_PATH}/usr")
+endif()
+if(NOT TERMUX)
+set(CPACK_DEBIAN_FILE_NAME "${BOX64}-${BOX64_MAJOR}.${BOX64_MINOR}.${BOX64_REVISION}_${CMAKE_SYSTEM_NAME}-${CPACK_DEBIAN_PACKAGE_ARCHITECTURE}.deb")
+else()
+set(CPACK_DEBIAN_FILE_NAME "${BOX64}-${BOX64_MAJOR}.${BOX64_MINOR}.${BOX64_REVISION}_Termux-${CPACK_DEBIAN_PACKAGE_ARCHITECTURE}.deb")
+endif()
 INCLUDE(CPack)

+if(NOT ANDROID)
 add_test(bootSyscall ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test01 -D TEST_OUTPUT=tmpfile01.txt
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref01.txt
@@ -910,19 +1153,193 @@ 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 ${CMAKE_COMMAND} -D TEST_PROGRAM=${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 )

-    add_test(x87 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+add_test(x87 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test22 -D TEST_OUTPUT=tmpfile22.txt
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref22.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-    set_tests_properties(x87 PROPERTIES ENVIRONMENT "BOX64_DYNAREC_FASTROUND=0")
+set_tests_properties(x87 PROPERTIES ENVIRONMENT "BOX64_DYNAREC_FASTROUND=0")

-    file(GLOB extension_tests "${CMAKE_SOURCE_DIR}/tests/extensions/*.c")
+add_test(pshufb ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test23 -D TEST_OUTPUT=tmpfile23.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref23.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(bswap ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test24 -D TEST_OUTPUT=tmpfile24.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref24.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(x87cache ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test25 -D TEST_OUTPUT=tmpfile25.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref25.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(feround ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test26 -D TEST_OUTPUT=tmpfile26.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref26.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+set_tests_properties(feround PROPERTIES ENVIRONMENT "BOX64_SYNC_ROUNDING=1")
+
+add_test(sse4_2 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test27 -D TEST_OUTPUT=tmpfile27.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref27.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(shaext ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test28 -D TEST_OUTPUT=tmpfile28.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref28.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(lock ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test29 -D TEST_OUTPUT=tmpfile29.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref29.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+else()
+
+add_test(bootSyscall ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test01_android -D TEST_OUTPUT=tmpfile01.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref01.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(bootSyscallC ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test02_android -D TEST_OUTPUT=tmpfile02.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref02.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(printf ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test03_android -D TEST_OUTPUT=tmpfile03.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref03.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(args ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test04_android -D TEST_ARGS2=yeah -D TEST_OUTPUT=tmpfile04.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref04.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(maths1 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test05_android -D TEST_ARGS2=7 -D TEST_OUTPUT=tmpfile05.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref05.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(threadsStart ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test06_android -D TEST_OUTPUT=tmpfile06.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref06.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+#add_test(trig ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+#    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test07_android -D TEST_OUTPUT=tmpfile07.txt
+#    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref07.txt
+#    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(pi ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test08_android -D TEST_OUTPUT=tmpfile08.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref08.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(fork ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test09_android -D TEST_OUTPUT=tmpfile09.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref09.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(fpu ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test12_android -D TEST_OUTPUT=tmpfile12.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref12_android.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(linkingIndirectNoversion ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test15_android -D TEST_OUTPUT=tmpfile15.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref15.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+# Box64 for Android now have problems with test16, then disable it for now
+
+#add_test(sse_asm ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+#    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test16_android -D TEST_OUTPUT=tmpfile16.txt
+#    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref16.txt
+#    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+# Box64 for Android now have problems with test17, then disable it for now
+
+#set_tests_properties(sse_asm PROPERTIES ENVIRONMENT "BOX64_DYNAREC_FASTROUND=0")
+
+#add_test(sse_intrinsics ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+#    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test17_android -D TEST_OUTPUT=tmpfile17.txt
+#    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref17_android.txt
+#    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+#set_tests_properties(sse_intrinsics PROPERTIES ENVIRONMENT "BOX64_DYNAREC_FASTNAN=0;BOX64_DYNAREC_FASTROUND=0")
+
+add_test(aes ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test18_android -D TEST_OUTPUT=tmpfile18.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref18.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+# Android don't have backtrace symbols
+
+add_test(irelative_reloc ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test20_android -D TEST_OUTPUT=tmpfile20.txt
+    -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}/${BOX64}
+    -D TEST_ARGS=${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 )
+
+add_test(x87 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test22_android -D TEST_OUTPUT=tmpfile22.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref22_android.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+set_tests_properties(x87 PROPERTIES ENVIRONMENT "BOX64_DYNAREC_FASTROUND=0")
+
+add_test(pshufb ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test23_android -D TEST_OUTPUT=tmpfile23.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref23.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(bswap ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test24_android -D TEST_OUTPUT=tmpfile24.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref24.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(x87cache ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test25_android -D TEST_OUTPUT=tmpfile25.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref25.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(feround ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test26_android -D TEST_OUTPUT=tmpfile26.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref26.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+set_tests_properties(feround PROPERTIES ENVIRONMENT "BOX64_SYNC_ROUNDING=1")
+
+# TODO: need to regen test with latest test27.c sources
+
+#add_test(sse4_2 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+#    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test27_android -D TEST_OUTPUT=tmpfile27.txt
+#    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref27.txt
+#    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(shaext ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test28_android -D TEST_OUTPUT=tmpfile28.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref28.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(lock ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test29_android -D TEST_OUTPUT=tmpfile29.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref29.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+endif()
+
+if(NOT ANDROID)
+file(GLOB extension_tests "${CMAKE_SOURCE_DIR}/tests/extensions/*.c")
 foreach(file ${extension_tests})
    get_filename_component(testname "${file}" NAME_WE)
    add_test(NAME "${testname}" COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
@@ -930,3 +1347,14 @@ foreach(file ${extension_tests})
        -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/extensions/${testname}.txt
        -P ${CMAKE_SOURCE_DIR}/runTest.cmake)
 endforeach()
+else()
+
+file(GLOB extension_tests "${CMAKE_SOURCE_DIR}/tests/extensions/*.c")
+foreach(file ${extension_tests})
+    get_filename_component(testname "${file}" NAME_WE)
+    add_test(NAME "${testname}" COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+        -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/extensions/${testname}_android -D TEST_OUTPUT=tmpfile-${testname}.txt
+        -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/extensions/${testname}.txt
+        -P ${CMAKE_SOURCE_DIR}/runTest.cmake)
+endforeach()
+endif()
@@ -2,7 +2,7 @@

 Linux Userspace x86_64 Emulator with a twist

-[View changelog](https://github.com/ptitSeb/box64/blob/main/docs/CHANGELOG.md) | [中文](https://github.com/ptitSeb/box64/blob/main/README_CN.md) | [Report an error](https://github.com/ptitSeb/box64/issues/new)
+[View 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](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) ![prs](https://img.shields.io/github/issues-pr/ptitSeb/box64) ![issues](https://img.shields.io/github/issues/ptitSeb/box64)

@@ -37,8 +37,8 @@ Note: Box64's Dynarec uses a mechanism with Memory Protection and a SegFault sig

 Compiling/Installation
 ----
-> Compilation instructions can be found [here](docs/COMPILE.md)\
-> Instructions for installing Wine for Box64 can be found [here](docs/X64WINE.md)
+> Compilation instructions can be found [here](https://github.com/ptitSeb/box64/blob/main/docs/COMPILE.md) \
+> Instructions for installing Wine for Box64 can be found [here](https://github.com/ptitSeb/box64/blob/main/docs/X64WINE.md)

 ----

@@ -54,24 +54,26 @@ Notes about 32-bit platforms

 Because Box64 works by directly translating function calls from x86_64 to host system, the host system (the one Box64 is running on) needs to have 64-bit libraries. Box64 doesn't include any 64-bit <-> 32-bit translation. 

-So understand that box64 will only runs 64-bit linux binaries. For 32-bit binaries, you need box86 (with all the multiarch or proot trickery it imply on 64-bit OS).
+So understand that box64 will only run 64-bit linux binaries. For 32-bit binaries, you need box86 (with all the multiarch or proot trickery it implies on a 64-bit OS).
 Note that many installer (mojo setup based) will fall back to "x86" when detecting ARM64 OS, and so will try to use box86 for the setup, even if an x86_64 version exist. You can hack your way around with a fake "uname" that return "x86_64" when the argument is "-m"

+An alternative solution is to use Wine with new WoW64, but you will be limited to 32bits windows apps and games.
+
 ----

 Notes about Box64 configuration
 ----

-Box64 now have configurations files. There are 2 files loaded. `/etc/box4.box64rc` and `~/.box64rc`. Both files have the same syntax, and is basicaly an ini files. Section in square brakets define the process name, and the rest is the env. var. to set. Looke at [Usage](USAGE.md) for detail on what parameters can be put. Box64 comes with a default file that should be installed for better stability. The file in in `system/box64.box64rc` and should be installed to `/etc/box64.box64rc` If, for some reasons, you don't want to install that file here, at least copy it to `~/.box64rc` or some game may not function correctly.
-Note that the priority is: `~/.bashrc` > `/etc/box64.box64rc` > command line
-So, your settings in `~/.bashrc` may override the setting from your command line...
+Box64 now have configurations files. There are 2 files loaded. `/etc/box64.box64rc` and `~/.box64rc`. Both files have the same syntax, and is basicaly an ini files. Section in square brakets define the process name, and the rest is the env. var. to set. Looke at [Usage](USAGE.md) for detail on what parameters can be put. Box64 comes with a default file that should be installed for better stability. The file in in `system/box64.box64rc` and should be installed to `/etc/box64.box64rc` If, for some reason, you don't want to install that file here, at least copy it to `~/.box64rc` or some game may not function correctly.
+Note that the priority is: `~/.box64rc` > `/etc/box64.box64rc` > command line
+So, your settings in `~/.box64rc` may override the setting from your command line...

 ----

 Notes about Unity game emulation
 ----

-Running Unity games should just works, but you should also note that many Unity3D games require OpenGL 3+ which can be tricky to provide on ARM SBC (single-board computers). Also many newer Unity3D (like KSP) games use the BC7 compressed textures, wich is not supported on many ARM integrated GPU.
+Running Unity games should just work, but you should also note that many Unity3D games require OpenGL 3+ which can be tricky to provide on ARM SBC (single-board computers). Also many newer Unity3D (like KSP) games use the BC7 compressed textures, which is not supported on many ARM integrated GPU.
 Hint: on Pi4, use `MESA_GL_VERSION_OVERRIDE=3.2` and with Panfrost use `PAN_MESA_DEBUG=gl3` to use higher profile if the game starts then quits before showing anything.

 ----
@@ -86,23 +88,22 @@ GTK libraries are now wrapped on box64, both gtk2 and gtk3.
 Notes about Steam
 ----

-Note that Steam is a hybrid 32-bit / 64-bit. You NEED box86 to run Steam, as the client app is a 32-bit binary. It also uses a 64-bit local server binaries, but like most stuff using libcef/chromium, it's not working correctly on box64 for now.
-So, no Steam for now on box64.
+Note that Steam is a hybrid 32-bit / 64-bit. You NEED box86 to run Steam, as the client app is a 32-bit binary. It also uses a 64-bit local server binaries, and that steamwebhelper process is now mandatory, even on the "small mode". And that process will eat lots of memory. So machine with less the 6Gb of RAM will need a swapfile to use Steam.

 ----

 Notes about Wine
 ----

-Wine64 is supported on box64. Proton should also work. Be aware that 64-bit Wine also includes 32-bit components, to be able to run 32-bit Windows programs. The 32-bit apps will need box86 and will not run without it. On a system where both box64 and box86 are present and working, a wine 64-bit setup can run both 32-bit and 64-bit Windows programs (just use `wine` and `wine64` respectively).
-Note that the new 32bits PE in 64bits process that the Wine time in currently implementing in Wine 7.+ is not yet supported. I tested Wine 7.5 64bits and it worked, but more recent version will probably just not work yet.
+Wine64 is supported on box64. Proton too. Be aware that 64-bit Wine also includes 32-bit components, to be able to run 32-bit Windows programs. The 32-bit apps will need box86 and will not run without it. On a system where both box64 and box86 are present and working, a wine 64-bit setup can run both 32-bit and 64-bit Windows programs (just use `wine` and `wine64` respectively).
+Note that the new 32-bit PE in 64-bit processes that the Wine team is currently implementing in Wine 7.+ is now supported. It's called "new WoW64" and it's working fine on box64, allowing the use of 32bits windows program on 64bits only system.

 ----

 Notes about Vulkan
 ----

-Box64 wraps Vulkan libraries, but note that it as only been tested with a RX550 card, so some extensions may be missing depending on your graphics card.
+Box64 wraps Vulkan libraries, but note that it as mostly been tested with an AMD RX550 card and on the Freedreno driver, so some extensions may be missing depending on your graphics card.

 ----

@@ -112,12 +113,11 @@ Final word
 I want to thank everyone who has contributed to box64 development.
 There are many ways to contribute: code contribution, financial, hardware and advertisement!
 So, in no particular order, I want to thank:
- * For their major code contribution: rajdakin, mogery
- * For their major financial contribution: FlyingFathead, stormchaser3000, dennis1248, sll00, [libre-computer-project](https://libre.computer/)
+ * For their major code contribution: rajdakin, mogery, ksco, xctan
+ * For their major financial contribution: [Playtron](https://playtron.one), FlyingFathead, stormchaser3000, dennis1248, sll00, [libre-computer-project](https://libre.computer/), [CubeCoders Limited](http://cubecoders.com/)
 * For hardware contribution and LoongArch migration: [xiaoji](https://www.linuxgame.cn/), Deepin Beijing Develop Team
- * For their major code contribution on the RV64 Dynarec: ksco, xctan
- * For their hardware contribution: [Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/), [StarFive](https://rvspace.org/)
- * For their continous advertisement of box64 project: salva ([microLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ)), [PILab](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g)/[TwisterOS](https://twisteros.com/) team, [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA), [NicoD](https://www.youtube.com/channel/UCpv7NFr0-9AB5xoklh3Snhg), ekianjo ([Boilingsteam](https://boilingsteam.com/))
+ * For their hardware contribution: [ADLINK](https://www.adlinktech.com/Products/Computer_on_Modules/COM-HPC-Server-Carrier-and-Starter-Kit/Ampere_Altra_Developer_Platform?lang=en) with [Ampere](https://amperecomputing.com/home/edge), [SOPHGO](https://www.sophon.ai/), [Radxa](https://rockpi.org/), [StarFive](https://rvspace.org/), [Pine64](https://www.pine64.org/), [AYN](https://www.ayntec.com/), [AYANEO](https://ayaneo.com/), [jiangcuo](https://github.com/jiangcuo)
+ * For their continuous advertisements for the box64 project: salva ([microLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ)), [PILab](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g)/[TwisterOS](https://twisteros.com/) team, [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA), [NicoD](https://www.youtube.com/channel/UCpv7NFr0-9AB5xoklh3Snhg), ekianjo ([Boilingsteam](https://boilingsteam.com/))

 And I also thank the many other people who participated even once in this project.

@@ -6,7 +6,7 @@ Linux 上用户空间的 x86_64 应用模拟器（支持运行原生库）
 ----
 ![box64 build status](https://app.travis-ci.com/ptitSeb/box64.svg?branch=main)

-[English](docs/README.md)
+[English](docs/README.md) | [Українська](https://github.com/ptitSeb/box64/blob/main/README_UK.md) 

 Box64 可以在非 x86_64 Linux 系统（比如 ARM64）上运行 x86_64 Linux 程序（比如游戏），注意主机系统需要是 64 位小端。

@@ -14,7 +14,7 @@ Box64 可以在非 x86_64 Linux 系统（比如 ARM64）上运行 x86_64 Linux

 由于 Box64 使用一些“系统”库的原生版本，如 libc、libm、SDL 和 OpenGL 等，因此很容易与大多数应用程序集成和使用，并且在许多情况下性能会相当不错。可以在[这里](https://box86.org/index.php/2021/06/game-performances/)查看一些性能测试的样例。

-Box64 集成了适用于 ARM64 平台的 DynaRec（动态重编译器），速度可以比纯解释模式快 5 到 10 倍。可以在[这里](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/)找到有关 DynaRec 工作原理的一些信息。
+Box64 集成了适用于 ARM64 和 RV64 平台的 DynaRec（动态重编译器），速度可以比纯解释模式快 5 到 10 倍。可以在[这里](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/)找到有关 DynaRec 工作原理的一些信息。

 一些 x64 内部操作码使用 “Realmode X86 Emulator Library” 的部分内容，有关版权详细信息，请参见 [x64primop.c](../src/emu/x64primop.c)。

@@ -54,7 +54,7 @@ LOGO 由 @grayduck 制作，感谢！

 因为 Box64 的工作原理是直接将函数调用从 x86_64 转换为主机系统，所以主机系统（运行 Box64 的系统）需要有 64 位库。Box64 不包含任何 64 位 <-> 32 位的转换。

-所以 box64 只能运行 64 位的 Linux 二进制。对于 32 位二进制则需要使用 box86 来运行（它在 64 位操作系统上使用了 multiarch 和 proot 等技巧来实现运行）。请注意，许多（基于 mojo 的）安装程序在检测到 ARM64 操作系统时将回退到 “x86”，因此即使存在 x86_64 版本，也会尝试使用 box86。这时你可以使用一个假的 `uname`，当它的参数为 `-m` 时返回 `x86_64`。
+所以 box64 只能运行 64 位的 Linux 二进制。对于 32 位二进制则需要使用 box86 来运行（它在 64 位操作系统上使用了 multiarch 和 proot 等技巧来实现运行）。请注意，许多（基于 mojo 的）安装程序在检测到 ARM64 操作系统时将回退到 “x86”，因此即使存在 x86_64 版本，也会尝试使用 box86。这时你可以使用一个假的 `uname` ，并使它在运行参数为 `-m` 时返回 `x86_64`。

 ----

@@ -77,7 +77,7 @@ box64 封装了 GTK，包括 gtk2 和 gtk3。
 关于 Steam 的注意事项
 ----

-请注意，Steam 是 32/64 位的混合体，所以还需要 box86 才能运行，因为客户端应用程序是 32 位的。它还使用 64 位本地服务器，但与大多数使用 libcef/chromium 的东西一样，它目前无法在 box64 上正常工作。所以，现在 box64 上暂时不能运行 Steam。
+请注意，Steam 是 32/64 位混合的应用，所以你需要 box86 才能运行，因为客户端应用程序是 32 位的。它还使用 64 位本地服务器，它的 steamwebhelper 无法被关闭（即使是在最小模式）而且会吃掉大量的内存。对于内存小于 6 GB 的机型，你将会需要创建 swapfile 来运行 Steam。

 ----

@@ -102,10 +102,10 @@ Box64 封装了 Vulkan 库，但请注意，它仅在 RX550 显卡上进行过
 贡献方式有很多种：代码贡献、财务帮助、捐赠捐赠和广告合作！
 以下感谢不分先后： 

- * 代码贡献：rajdakin, mogery
- * 财务帮助：FlyingFathead, stormchaser3000, dennis1248, sll00, [libre-computer-project](https://libre.computer/)
+ * 代码贡献：rajdakin, mogery, ksco, xctan
+ * 财务帮助：[Playtron](https://playtron.one), FlyingFathead, stormchaser3000, dennis1248, sll00, [libre-computer-project](https://libre.computer/)
 * 硬件捐赠和 LoongArch 迁移：[xiaoji](https://www.linuxgame.cn/), Deepin Beijing Develop Team
- * 硬件捐赠：[Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/), [StarFive](https://rvspace.org/)
+ * 硬件捐赠：[ADLINK](https://www.adlinktech.com/Products/Computer_on_Modules/COM-HPC-Server-Carrier-and-Starter-Kit/Ampere_Altra_Developer_Platform?lang=en) with [Ampere](https://amperecomputing.com/home/edge), [SOPHGO](https://www.sophon.ai/), [Radxa](https://rockpi.org/), [StarFive](https://rvspace.org/), [Pine64](https://www.pine64.org/), [AYN](https://www.ayntec.com/), [AYANEO](https://ayaneo.com/), [jiangcuo](https://github.com/jiangcuo)
 * 为本项目持续的宣传：salva ([microLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ)), [PILab](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g)/[TwisterOS](https://twisteros.com/) team, [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA), [NicoD](https://www.youtube.com/channel/UCpv7NFr0-9AB5xoklh3Snhg), ekianjo ([Boilingsteam](https://boilingsteam.com/))

 我还要感谢很多人，即使他们只参与了一次本项目。
@@ -0,0 +1,118 @@
+![Official logo](docs/img/Box64Logo.png "Official Logo")
+
+Емулятор простіра користувача x86_64 із родзинкою
+
+[Дивитись журнал змін](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) | [Повідомити про помилку](https://github.com/ptitSeb/box64/issues/new)
+
+![build](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) ![prs](https://img.shields.io/github/issues-pr/ptitSeb/box64) ![issues](https://img.shields.io/github/issues/ptitSeb/box64)
+
+----
+
+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/).
+
+Деякі внутрішні коди операцій x64 використовують частини «Бібліотеки емулятора Realmode X86», дивіться [x64primop.c](src/emu/x64primop.c) для детальної інформації про авторські права.
+
+<img src="docs/img/Box64Icon.png" width="96" height="96">
+
+Логотип і піктограма зроблені @grayduck, дякую!
+
+----
+
+Використання
+----
+
+Є кілька змінних середовища для керування поведінкою Box64.
+
+Дивіться [тут](docs/USAGE.md) для перегляду всіх змінних середовища та того, що вони роблять.
+
+Примітка: Dynarec від Box64 використовує механізм із захистом пам’яті та обробником сигналу помилки сегментів для обробки JIT коду. Простіше кажучи, якщо ви хочете використовувати GDB для налагодження запущеної програми, яка використовує JIT-код (наприклад, mono/Unity3D), ви все одно матимете багато «звичайних» помилок сегментів. Пропонується використовувати щось на зразок `handle SIGSEGV nostop` в GDB, щоб не зупинятися на кожній помилці сегмента, і, можливо, поставити точку зупину всередині `my_memprotectionhandler` в `signals.c`, якщо ви хочете перехопити помилки сегментів.
+
+----
+Компіляція/Встановлення
+----
+> Інструкцію зі складання Box64 можна знайти [тут](docs/COMPILE.md).\
+> Інструкцію з встановлення Wine для Box64 можна знайти [тут](docs/X64WINE.md).
+
+----
+
+Історія версій/Журнал змін
+----
+
+Історія версій доступна [тут](docs/CHANGELOG.md).
+
+----
+
+Примітки щодо 32 бітних платформ
+----
+Оскільки Box64 працює шляхом прямого перекладу викликів функцій з x86_64 на хост-систему, хост-система (та, на якій працює Box64) повинна мати 64-розрядні бібліотеки. Box64 не містить 64-біт <-> 32-біт перекладу.
+
+Також розумійте що Box64 запускатиме лише 64-розрядні двійкові Linux файли. Для 32-розрядних двійкових файлів вам потрібен Box86 (з усіма хитрощами з мультиархітектурною системою або корінними правами). Зауважте, що багато інсталяторів (на основі mojo) повертаються до "x86" при виявленні ОС ARM64, і тому намагатимуться використовувати Box86 для налаштування, навіть якщо існує версія x86_64. Ви можете зламати свій шлях за допомогою фальшивого "uname", який повертає "x86_64", коли аргумент "-m"
+
+----
+
+Примітки щодо конфігурації Box64
+----
+
+Обидва файли мають однаковий синтаксис і в основному є файлами ini. Розділ у квадратних дужках визначає назву процесу, а решта — встановлють змінну середовища. Подивіться [використання](USAGE.md), щоб дізнатися, які параметри можна завдати. Box64 поставляється з файлом за замовчуванням, який слід встановити для кращої стабільності. Файл знаходиться в `system/box64.box64rc` і має бути встановлен в `/etc/box64.box64rc`. Якщо з певних причин ви не хочете встановлювати цей файл у `/etc`, принаймні скопіюйте його в `~/.box64rc` або гра може не працювати належним чином. Зауважте, що пріоритетом є: `~/.bashrc` > `/etc/box64.box64rc` > командний рядок
+Таким чином, ваші налаштування в `~/.bashrc` можуть замінити налаштування з вашого командного рядка...
+
+----
+
+Примітки щодо емуляції ігор Unity
+----
+
+Запуск ігор Unity має працювати, але ви також повинні зауважити, що для багатьох ігор Unity3D потрібен OpenGL 3+, який може бути складно забезпечити на ARM SBC (одноплатних комп’ютерах). Крім того, багато нових ігор Unity3D (наприклад, KSP) використовують стиснуті текстури BC7, які не підтримуються багатьма інтегрованими GPU ARM.
+Підказка: в Pi4 використовуйте `MESA_GL_VERSION_OVERRIDE=3.2`, а в Panfrost — `PAN_MESA_DEBUG=gl3`, щоб використовувати вищий профіль, якщо гра запускається, а потім завершується, перш ніж щось показує.
+
+----
+
+Примітки щодо програм GTK
+----
+
+Бібліотеки GTK2 і GTK3 тепер використовуються в Box64.
+
+----
+
+Примітки щодо Steam
+----
+
+Зауважте, що Steam є гібридом 32-біт / 64-біт. Вам ПОТРІБЕН box86, щоб запустити Steam, оскільки клієнта є 32-розрядною двійковою програмою. Він також використовує 64-розрядні двійкові файли локального сервера, і цей процес steamwebhelper тепер є обов’язковим навіть у «малому режимі». І цей процес з’їсть багато оперативної пам’яті, таким чином машині з 6 Гб оперативної пам’яті знадобиться файл підкачки для використання Steam.
+
+----
+
+Примітки щодо Wine
+----
+
+Wine64 підтримується Box64, Proton теж. Майте на увазі, що 64-розрядний Wine також містить 32-розрядні компоненти щоб мати можливість запускати 32-розрядні програми Windows. Для 32-розрядних програм потрібен Box86, і вони не працюватимуть без нього. У системі, де присутні і працюють Box64, і Box86, 64-розрядне встановлення Wine може запускати як 32-розрядні, так і 64-розрядні програми Windows (просто використовуйте `wine` і `wine64` відповідно).
+Зауважте, що новий 32-бітний PE в 64-бітному процесі, який Wine зараз реалізує у Wine 7+ тепер підтримується, але підтримка є досить молодою тому можуть виникнути деякі проблеми.
+
+----
+
+Примітки щодо Vulkan
+----
+
+Box64 використовує бібліотеки Vulkan, але зауважте що він тестувався з картою AMD Radeon RX550 і драйвером Freedreno, тому деякі розширення можуть бути відсутні залежно від вашої відеокарти.
+
+----
+
+Останнє слово
+----
+
+Я хочу подякувати всім, хто зробив внесок у розвиток box64.
+Є багато способів зробити внесок: внесок коду, фінансовий внесок, апаратне забезпечення та реклама!
+Отже, без особливого порядку, я хочу подякувати:
+* За основний внесок у код: rajdakin, mogery, ksco, xctan
+* За великий фінансовий внесок: [Playtron](https://playtron.one), FlyingFathead, stormchaser3000, dennis1248, sll00, [libre-computer-project](https://libre.computer/)
+* За апаратне забезпечення та міграцію LoongArch: [xiaoji](https://www.linuxgame.cn/), команда розробників Deepin Beijing
+* За внесок у апаратне забезпечення: [ADLINK](https://www.adlinktech.com/Products/Computer_on_Modules/COM-HPC-Server-Carrier-and-Starter-Kit/Ampere_Altra_Developer_Platform?lang=en) with [Ampere](https://amperecomputing.com/home/edge), [SOPHGO](https://www.sophon.ai/), [Radxa](https://rockpi.org/), [StarFive](https://rvspace.org/), [Pine64](https://www.pine64.org/), [AYN](https://www.ayntec.com/), [AYANEO](https://ayaneo.com/), [jiangcuo](https://github.com/jiangcuo)
+* За безперервну рекламу Box64: salva ([microLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ)), [PILab](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g)/Команда [TwisterOS](https://twisteros.com/), [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA), [NicoD](https://www.youtube.com/channel/UCpv7NFr0-9AB5xoklh3Snhg), ekianjo ([Boilingsteam](https://boilingsteam.com/))
+
+І я також дякую багатьом іншим людям, які хоч раз брали участь у цьому проекті.
+
+(Якщо ви використовуєте Box64 у своєму проекті, будь ласка, не забудьте згадати про це!)
@@ -1,3 +1,84 @@
+box64 (0.2.8) unstable; urgency=low
+
+* Wrapping: More libs and function wrapping (gtk3, vulkan...)
+* Reworked X11 Callback handling, for better stability
+* Reworked exit and the load/unload of libs and symbol resolution
+* Some more profiles in box64rc
+* Better way to rename executable (including wine ones) so driver like mesa can apply the correct profile
+* Reworked memory managment in the Dynarec, limiting amount of dynamic allocation
+* Improved memory tracking and managment. Also introduce BOX64_MMAP32 to force 32bits compatible mmap for external libs (like mesa)
+* => Allow Vulkan drvier on Wine/WOW64 games to allocate 32bits compatible memory, allowing dxvk on 32bits.
+* Reworked memory managment in the Dynarec, limiting amount of dynamic allocation
+* Lots of improvments in x87 handling (virtual stack, rounding, ffree, etc.)
+* => should improve games stability
+* Dynarec: ARM64 some opcodes optimisation, especialy shift and rotation ones
+* Dynarec: RV64 lots of opcoded added.
+* => Some games on linux now have a good speed. Game on Wine are still a bit slow, and dxvk still have issues
+* Dynarec: Added Loogarch backend! Note that Loongarch are 16K pagesize platform by default.
+* => Linux games are running, and with good speed, especialy on 3A6000 platforms. Games on Wine are not working yet due to 16k pagesize limitation
+* Dynarec: Better handling of large block in the dynarec
+* Dynarec: Lots of improvments and fixes to BOX86_DYNAREC_CALLRET optimisation.
+* => It can be used on more software now, and combine with BOX64_DYNAREC_BOGBLOCKS 2 or 3 can gives some good speed boosts.
+
+-- Sebastien Chevalier <[email protected]>  Thu, 31 May 2024 17:30:33 -0100
+
+box64 (0.2.6) unstable; urgency=low
+
+ * X64Libs: Updated libstdc++ & libgcc_s
+ * X64Libs: Use x86_64 vrsion of libunwind (it's needed by wine)
+ * Dynarec: More opcodes on ARM64 & RV64
+ * Dynarec: Fixed some flags propagation issues inside blocks
+ * Dynarec: Improved Strong Memory Model emulation, introducing a 3rd level
+ * Dynarec: ARM64: reworked shift and bit opcodes flags handling
+ * Dynarec: RV64: Many fixes, making Wow64 apps running
+ * Dynarec: ARM64: Added support for SSE4.2 (with CRC32 if supported by CPU)
+ * Dynarec: ARM64: Added support for SHA (with hardware support if CPU allows it)
+ * Dynarec: ARM64: The CALLRET optimisation has been improved and is now default
+ * CPU: Remoworked how INT, priviliged opcodes and teh Trace flags works
+ * CPU: Added full support for SSE4.2
+ * CPU: Added full support for SHA extension
+ * ElfLoader: rework how elf memory is handled
+ * Emulation: Added some more syscalls
+ * Emulation: Better handling of Signal (also handling SIGMAX correctly)
+ * Emulation: Better support for POKEUSER/PEEKUSER (helps Windows Unity games)
+ * Wrapping: More libs and some fixes in function wrapping (gtk3, vulkan...)
+ * Wrapping: Removed faked libunwind wrapping
+ * Hardware: Added more profile, with support ADLink Ampere Altra, Qualcomm 8Gen3, and Pi5 machines
+ * OS: Added profile to build for Android and Termux.
+ * OS: New option to use mold for a high speed link
+ * => More hadware support, more functions wrapped for an improve compatibity
+ * => More CPU extension, with ARM64 hardware support when possible, and improved CALLRET optimisation for an improved speed of emulation
+ * => Some work also to reduce the memory footprint
+
+-- Sebastien Chevalier <[email protected]>  Thu, 15 Dec 2023 17:30:33 -0100
+
+box64 (0.2.4) unstable; urgency=low
+
+  * Added Dynarec for RISC-V
+  * Added partial 32 bits support (WoW64 works)
+  * Changed cpuid to display box64
+  * Fixed steam, again
+  * Improved hacked version for overridden malloc/free functions
+  * More wrapped libs
+  * More wrapped functions and syscalls
+  * Reduced memory footprint
+  * Added support for the BIND_NOW flags in ELF files
+  * Added RTLD_NOLOAD support
+  * Fixed some typos and general cleanup, again
+  * Added a cosimulation mode (to compare the behaviour of the dynarec with the interpreter)
+  * Added FASTNAN/FASTROUND handling for more opcodes
+  * Better float/double/long double handling
+  * Better dlopen/dlclose handling
+  * More games in the rc file
+  * Added BOX64_DYNAREC_MISSING to show only missing opcode, independantly of the log level
+  * Added BOX64_RESERVE_HIGH to reserve addresses above 47bits
+  * Improved multitreading support
+  * Ukrainian translation of the README
+  * => Dynarec now support RISC-V, large speedup on RISC-V 64, were games are now playable (like Stardew Valley on a Vision Five 2)
+  * => Improved compatibility with more fixes in elf handling, more wrapped libs and functiond, more fixed opcodes and refined profile
+
+-- Sebastien Chevalier <[email protected]>  Thu, 10 Mar 2023 10:20:33 -0500
+
 box64 (0.2.2) unstable; urgency=low

 * Added a hacked version for overridden malloc/free functions, enabling libcef (and derivative) to work.
@@ -1,3 +1,78 @@
+v0.2.8
+======
+* Wrapping: More libs and function wrapping (gtk3, vulkan...)
+* Reworked X11 Callback handling, for better stability
+* Reworked exit and the load/unload of libs and symbol resolution
+* Some more profiles in box64rc
+* Better way to rename executable (including wine ones) so driver like mesa can apply the correct profile
+* Reworked memory managment in the Dynarec, limiting amount of dynamic allocation
+* Improved memory tracking and managment. Also introduce BOX64_MMAP32 to force 32bits compatible mmap for external libs (like mesa)
+=> Allow Vulkan drvier on Wine/WOW64 games to allocate 32bits compatible memory, allowing dxvk on 32bits.
+* Reworked memory managment in the Dynarec, limiting amount of dynamic allocation
+* Lots of improvments in x87 handling (virtual stack, rounding, ffree, etc.)
+=> should improve games stability
+* Dynarec: ARM64 some opcodes optimisation, especialy shift and rotation ones
+* Dynarec: RV64 lots of opcoded added.
+=> Some games on linux now have a good speed. Game on Wine are still a bit slow, and dxvk still have issues
+* Dynarec: Added Loogarch backend! Note that Loongarch are 16K pagesize platform by default.
+=> Linux games are running, and with good speed, especialy on 3A6000 platforms. Games on Wine are not working yet due to 16k pagesize limitation
+* Dynarec: Better handling of large block in the dynarec
+* Dynarec: Lots of improvments and fixes to BOX86_DYNAREC_CALLRET optimisation.
+=> It can be used on more software now, and combine with BOX64_DYNAREC_BOGBLOCKS 2 or 3 can gives some good speed boosts.
+
+v0.2.6
+======
+* X64Libs: Updated libstdc++ & libgcc_s
+* X64Libs: Use x86_64 vrsion of libunwind (it's needed by wine)
+* Dynarec: More opcodes on ARM64 & RV64
+* Dynarec: Fixed some flags propagation issues inside blocks
+* Dynarec: Improved Strong Memory Model emulation, introducing a 3rd level
+* Dynarec: ARM64: reworked shift and bit opcodes flags handling
+* Dynarec: RV64: Many fixes, making Wow64 apps running
+* Dynarec: ARM64: Added support for SSE4.2 (with CRC32 if supported by CPU)
+* Dynarec: ARM64: Added support for SHA (with hardware support if CPU allows it)
+* Dynarec: ARM64: The CALLRET optimisation has been improved and is now default
+* CPU: Remoworked how INT, priviliged opcodes and teh Trace flags works
+* CPU: Added full support for SSE4.2
+* CPU: Added full support for SHA extension
+* ElfLoader: rework how elf memory is handled
+* Emulation: Added some more syscalls
+* Emulation: Better handling of Signal (also handling SIGMAX correctly)
+* Emulation: Better support for POKEUSER/PEEKUSER (helps Windows Unity games)
+* Wrapping: More libs and some fixes in function wrapping (gtk3, vulkan...)
+* Wrapping: Removed faked libunwind wrapping
+* Hardware: Added more profile, with support ADLink Ampere Altra, Qualcomm 8Gen3, and Pi5 machines
+* OS: Added profile to build for Android and Termux.
+* OS: New option to use mold for a high speed link
+=> More hadware support, more functions wrapped for an improve compatibity
+=> More CPU extension, with ARM64 hardware support when possible, and improved CALLRET optimisation for an improved speed of emulation
+=> Some work also to reduce the memory footprint
+
+v0.2.4
+======
+* Added Dynarec for RISC-V
+* Added partial 32 bits support (WoW64 works)
+* Changed cpuid to display box64
+* Fixed steam, again
+* Improved hacked version for overridden malloc/free functions
+* More wrapped libs
+* More wrapped functions and syscalls
+* Reduced memory footprint
+* Added support for the BIND_NOW flags in ELF files
+* Added RTLD_NOLOAD support
+* Fixed some typos and general cleanup, again
+* Added a cosimulation mode (to compare the behaviour of the dynarec with the interpreter)
+* Added FASTNAN/FASTROUND handling for more opcodes
+* Better float/double/long double handling
+* Better dlopen/dlclose handling
+* More games in the rc file
+* Added BOX64_DYNAREC_MISSING to show only missing opcode, independantly of the log level
+* Added BOX64_RESERVE_HIGH to reserve addresses above 47bits
+* Improved multitreading support
+* Ukrainian translation of the README
+=> Dynarec now support RISC-V, large speedup on RISC-V 64, were games are now playable (like Stardew Valley on a Vision Five 2)
+=> Improved compatibility with more fixes in elf handling, more wrapped libs and functiond, more fixed opcodes and refined profile
+
 v0.2.2
 ======
 * Added a hacked version for overridden malloc/free functions, enabling libcef (and derivative) to work.
@@ -7,7 +7,7 @@ You can use [@ryanfortner](https://github.com/ryanfortner)'s apt repository to i
 ```
 sudo wget https://ryanfortner.github.io/box64-debs/box64.list -O /etc/apt/sources.list.d/box64.list
 wget -qO- https://ryanfortner.github.io/box64-debs/KEY.gpg | sudo gpg --dearmor -o /etc/apt/trusted.gpg.d/box64-debs-archive-keyring.gpg
-sudo apt update && sudo apt install box64-generic-arm -y
+sudo apt update && sudo apt install box64-arm64 -y
 ```

 Alternatively, you can generate your own package using the [instructions below](https://github.com/ptitSeb/box64/blob/main/docs/COMPILE.md#debian-packaging). 
@@ -27,9 +27,9 @@ If it's the first install, you also need:
 sudo systemctl restart systemd-binfmt
 ```
 - You can use `make -j1`, `make -j2` to prevent running out of memory
- You can also add `-DBAD_SIGNAL=ON` to the cmake command if you are on Linux Kernel mixed with Android, like on RK3588 or maybe Termux
+- You can also add `-DBAD_SIGNAL=ON` to the cmake command if you are on Linux Kernel mixed with Android, like on RK3588.

-#### For instance, if you want to build box64 for Generic ARM64 Linux platforms, it would look like this:
+#### For Instance, if you want to build box64 for Generic ARM64 Linux platforms, it would look like this:
 ```
 git clone https://github.com/ptitSeb/box64
 cd box64
@@ -68,7 +68,6 @@ Still, this can be faster if your build is attended.

 ```
 -D RPI3ARM64=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
-make 
 ```

 #### for Raspberry Pi 4
@@ -79,6 +78,12 @@ Warning, you need a 64bit OS:
 -D RPI4ARM64=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
 ```

+#### for Raspberry Pi 5
+
+```
+-D RPI5ARM64=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
+```
+
 #### for TEGRA X1

 Using a 64bit OS:
@@ -87,6 +92,22 @@ Using a 64bit OS:
 -D TEGRAX1=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
 ```

+#### for Jetson Xavier/T194
+
+Using a 64bit OS:
+
+```
+-D TEGRA_T194=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
+```
+
+#### for Jetson Orin/T234
+
+Using a 64bit OS:
+
+Caution: please use gcc-11 or higher, older gcc dosen't know cortex-a78ae
+```
+-D TEGRA_T234=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
+```

 #### for ODROID N2/N2+

@@ -96,7 +117,7 @@ Using a 64bit OS:
 -D ODROIDN2=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
 ```

-#### for Snapdragon 845
+#### for Snapdragon

 Using a 64bit OS:

@@ -104,6 +125,14 @@ Using a 64bit OS:
 -D SD845=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
 ```

+or
+
+```
+-D SD888=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
+```
+
+Depending how recent your Snapdragon is
+
 #### for Phytium

 Using a 64bit OS:
@@ -111,9 +140,16 @@ Using a 64bit OS:
 -D PHYTIUM=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
 ```

+#### for ADLink machines
+
+Using a 64bit OS:
+```
+-D ADLINK=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
+```
+
 #### for M1

-Only test on Asahi for now, using the default "16K page" kernel
+Only test on Asahi with Fedora, using the default "16K page" kernel

 ```
 -D M1=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
@@ -151,6 +187,26 @@ Using a 64bit OS:
 -D LX2160A=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
 ```

+#### for Termux
+
+You must have ARM64 Device for build box64.
+
+##### in CHRoot/PRoot
+
+```
+-D ARM64=1 -DCMAKE_C_COMPILER=gcc -DBAD_SIGNAL=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo
+```
+
+##### in Termux (Native)
+
+NOTE: BUILDING BOX64 IN TERMUX NATIVE IS EXPERIMENTAL AND DOESN'T GONNA RUN LINUX BINARIES IN NATIVE TERMUX BOX64!!!
+
+You also needed have `libandroid-sysv-semaphore` and `libandroid-spawn` libraries:
+
+```
+-D TERMUX=1 -DCMAKE_C_COMPILER=clang -D CMAKE_BUILD_TYPE=RelWithDebInfo
+```
+
 #### for x86_64 Linux
 ```
 -D LD80BITS=1 -D NOALIGN=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
@@ -159,7 +215,7 @@ If you encounter some linking errors, try using `NOLOADADDR=ON` (`cmake -D NOLOA

 ----

-### use ccmake
+### Use ccmake

 Alternatively, you can **use the curses-bases ccmake (or any other gui frontend for cmake)** to select wich platform to use interactively.

@@ -180,12 +236,24 @@ To have a trace enabled build (***the interpreter will be slightly slower***), a
 #### To have ARM Dynarec

 Dynarec is only available on ARM (for the meantime), Activate it by using `-DARM_DYNAREC=1`.
-###### *Note: VFPv3 and NEON are required for Dynarec.*
+
+#### To have a build using less memory
+
+You can use `-DSAVE_MEM` to have a build that will try to save some memory. It's, for now, only increasing the jumptable from 4 level to 5 levels. The added granularity avoid wasting space, but the 1 level more to the jumptable means there is on read from memory more when jumping between blocks.

 #### Not building from a git clone

 If you are not building from a git clone (for example, downloading a release source code zip from github), you need to use `-DNOGIT=1` from cmake to be able to build (box64 uses git SHA1 to show last commit in version number).

+#### Use faster linker
+
+You need to add `-DWITH_MOLD=1` if GNU ld is extremely slow. Then run `mold -run make -j4` to build (make sure [Mold](https://github.com/rui314/mold) is installed).
+
+#### Build a statically linked box64
+
+You can now build box64 staticaly linked, with `-DSTATICBUILD`. This is to use inside an x86_64 chroot. Note that this version of box64 will have just the minimum of wrapped libs. So only libc, libm and libpthread basically are wrapped. Other libs (like libGL or libvulkan, SDL2, etc...) will not be wrapped and x86_64 version will be used. It's designed to be used in docker image, or in headless server.
+Also, the Static Build is highly experimental, but feedback are always welcomed.
+
 ----

 Testing
@@ -3,7 +3,10 @@ Usage
 ----

 There are many environment variables to control Box64 behaviour. 
-Env. var with * can also be put inside box64rc files
+Env. var with * can also be put inside box64rc files. 
+Box64 look for 2 places for rcfile: `/etc/box64.box64rc` and `~/.box64rc`
+The second takes precedence to the first, on an APP level 
+(that means if an [MYAPP] my appears in both file, only the settings in `~/.box64rc` will be applied)

 #### BOX64_LOG *
 Controls the Verbosity level of the logs
@@ -49,7 +52,7 @@ Only on build with trace enabled. Trace allow the logging of all instruction exe
 * 0xXXXXXXX-0xYYYYYYY : Trace only between the 2 addresses.

 #### BOX64_TRACE_INIT *
-Use BOX64_TRACE_INIT instead of BOX_TRACE to start trace before the initialisation of libraries and the running program
+Use BOX64_TRACE_INIT instead of BOX64_TRACE to start trace before the initialisation of libraries and the running program
 * 0 : No trace. (Default.)
 * 1 : Trace enabled. The trace start with the initialisation of all depending libraries is done.

@@ -101,6 +104,16 @@ Call XInitThreads when loading X11. (This is mostly for old Loki games with the
 * 0 : Don't force call XInitThreads. (Default.)
 * 1 : Call XInitThreads as soon as libX11 is loaded.

+#### BOX64_MMAP32 *
+Will use 32bits address in priority for external MMAP (when 32bits process are detected)
+ * 0 : Use regular mmap (default, except for Snapdragron build)
+ * 1 : Use 32bits address space mmap in priority for external mmap as soon a 32bits process are detected (default for SnapDragon and TegraX1 build)
+
+#### BOX64_IGNOREINT3 *
+What to do when a CC INT3 opcode is encounter in the code being run
+ * 0 : Trigger a TRAP signal if a handler is present
+ * 1 : Just skip silently the opcode
+
 #### BOX64_X11GLX *
 Force libX11's GLX extension to be present.
 * 0 : Do not force libX11's GLX extension to be present. 
@@ -135,8 +148,9 @@ Forbid dynablock creation in the interval specified (helpfull for debugging beha

 #### BOX64_DYNAREC_TEST *
 Dynarec will compare it's execution with the interpreter (super slow, only for testing)
- * 0 : No comparison
- * 1 : Each opcode runs on interepter and on Dynarec, and regs and memory are compared and print if different
+ * 0 : No comparison. (Default.)
+ * 1 : Each opcode runs on interepter and on Dynarec, and regs and memory are compared and print if different.
+ * 2 : Thread-safe tests, extremely slow.

 #### BOX64_DYNAREC_BIGBLOCK *
 Enables/Disables Box64's Dynarec building BigBlock.
@@ -153,8 +167,9 @@ Define Box64's Dynarec max allowed forward value when building Block.
 #### BOX64_DYNAREC_STRONGMEM *
 Enable/Disable simulation of Strong Memory model
 * 0 : Don't try anything special (Default.)
-* 1 : Enable some Memory Barrier when reading from memory (on some MOV opcode) to simulate Strong Memory Model while trying to limit performance impact (Default when libmonobdwgc-2.0.so is loaded)
-* 2 : Enable some Memory Barrier when reading from memory (on some MOV opcode) to simulate Strong Memory Model
+* 1 : Enable some Memory Barrier when writting to memory (on some MOV opcode) to simulate Strong Memory Model while trying to limit performance impact (Default when libmonobdwgc-2.0.so is loaded)
+* 2 : All 1. plus a memory barrier on every write to memory using MOV
+* 3 : All 2. plus Memory Barrier when reading from memory and on some SSE/SSE2 opcodes too

 #### BOX64_DYNAREC_X87DOUBLE *
 Force the use of Double for x87 emulation
@@ -180,17 +195,12 @@ Handling of flags on CALL/RET opcodes
 #### BOX64_DYNAREC_CALLRET *
 Optimisation of CALL/RET opcodes (not compatible with jit/dynarec/smc)
 * 0 : Don't optimize CALL/RET, use Jump Table for boths (Default)
-* 1 : Try to optimized CALL/RET, skipping the use of the JumpTable when possible (will crash if blacks are invalidate, so probably incompatible with JIT/Dynarec)
+* 1 : Try to optimized CALL/RET, skipping the use of the JumpTable when possible

-#### BOX64_DYNAREC_HOTPAGE *
-Handling of HotPage (Page being both executed and written)
-* 0 : Don't track hotpage (Default)
-* 1-255 : Track HotPage, and disable execution of a page being written for N attempts
-
-#### BOX64_DYNAREC_FASTPAGE *
-Will use a faster handling of HotPage (Page being both executed and written)
-* 0 : use regular hotpage (Default)
-* 1 : Use faster hotpage, taking the risk of running obsolete JIT code (might be faster, but more prone to crash)
+#### BOX64_DYNAREC_ALIGNED_ATOMICS *
+Generated code for aligned atomics only
+* 0 : The code generated can handle unaligned atomics (Default)
+* 1 : Generated code only for aligned atomics (faster and less code generated, but will SEGBUS if LOCK prefix is unsed on unaligned data)

 #### BOX64_DYNAREC_BLEEDING_EDGE *
 Detect MonoBleedingEdge and apply conservative settings
@@ -198,9 +208,9 @@ Detect MonoBleedingEdge and apply conservative settings
 * 1 : Detect MonoBleedingEdge, and apply BIGBLOCK=0 STRONGMEM=1 if detected (Default)

 #### BOX64_DYNAREC_JVM *
-Detect libjvm and apply conservative settings
+Detect libjvm and apply conservative settings. Obsolete, use BOX64_JVM instead.
 * 0 : Don't detect libjvm
-* 1 : Detect libjvm, and apply BIGBLOCK=0 STRONGMEM=1 if detected (Default)
+* 1 : Detect libjvm, and apply BIGBLOCK=0 STRONGMEM=1 SSE42=0 if detected (Default)

 #### BOX64_DYNAREC_WAIT *
 Behavior with FillBlock is not availble (FillBlock build Dynarec blocks and is not multithreaded)
@@ -222,11 +232,31 @@ Handling of x87 80bits long double
 * 0 : Try to handle 80bits long double as precise as possible (Default)
 * 1 : Handle them as double

+#### BOX64_MAXCPU
+Maximum CPU Core exposed
+* 0 : Don't cap the number of cpu core exposed (Default)
+* XXX : Cap the maximum CPU Core exposed to XXX (usefull with wine64 or GridAutosport for example)
+
+#### BOX64_SYNC_ROUNDING *
+Box64 will sync rounding mode with fesetround/fegetround.
+* 0 : Disable rounding mode syncing. (Default.)
+* 1 : Enable rounding mode syncing.
+
 #### BOX64_LIBCEF *
 Detect libcef and apply malloc_hack settings
 * 0 : Don't detect libcef
 * 1 : Detect libcef, and apply MALLOC_HACK=2 if detected (Default)

+#### BOX64_JVM *
+Detect libjvm and apply conservative settings
+* 0 : Don't detect libjvm
+* 1 : Detect libjvm, and apply BIGBLOCK=0 STRONGMEM=1 SSE42=0 if detected (Default)
+
+#### BOX64_SDL2_JGUID *
+Need a workaround for SDL_GetJoystickGUIDInfo function for wrapped SDL2
+* 0 : Don't use any workaround
+* 1 : Use a workaround for program that use the private SDL_GetJoystickGUIDInfo function with 1 missing argument
+
 #### BOX64_LIBGL *
 * libXXXX set the name for libGL (defaults to libGL.so.1).
 * /PATH/TO/libGLXXX : Sets the name and path for libGL
@@ -282,19 +312,33 @@ Disables the load of vulkan libraries.
 * 0 : Load vulkan libraries if found.
 * 1 : Disables the load of vulkan libraries, both the native and the i386 version (can be useful on Pi4, where the vulkan driver is not quite there yet.)

+#### BOX64_SSE42 *
+Expose or not SSE 4.2 capabilites
+ * 0 : Do not expose SSE 4.2 capabilites (default when libjvm is detected)
+ * 1 : Expose SSE 4.2 capabilites (Default.)
+
+#### BOX64_FUTEX_WAITV *
+Use of the new fuext_waitc syscall
+ * 0 : Do not try to use it, return unsupported (Default for BAD_SIGNAL build)
+ * 1 : let program use the syscall if the host system support it (Default for other build)
+
 #### BOX64_BASH *
 Define x86_64 bash to launch script
 * yyyy
 Will use yyyy as x86_64 bash to launch script. yyyy needs to be a full path to a valid x86_64 version of bash

-#### BOX64_ENV
+#### BOX64_ENV *
 * XXX=yyyy
 will add XXX=yyyy env. var.

-#### BOX64_ENV1
+#### BOX64_ENV1 *
 * XXX=yyyy
 will add XXX=yyyy env. var. and continue with BOX86_ENV2 ... until var doesn't exist

+#### BOX64_RESERVE_HIGH
+* 0 : Don't try to pe-reserve high memory (beyond 47bits) (Default)
+* 1 : Try to reserve (without allocating it) memory beyond 47bits (seems unstable)
+
 #### BOX64_JITGDB *
 * 0 : Just print the Segfault message on segfault (default)
 * 1 : Launch `gdb` when a segfault, bus error or illegal instruction signal is trapped, attached to the offending process and go in an endless loop, waiting.
@@ -305,10 +349,14 @@ Define x86_64 bash to launch script
 * 2 : Launch `gdbserver` when a segfault, bus error or illegal instruction signal is trapped, attached to the offending process, and go in an endless loop, waiting.
 Use `gdb /PATH/TO/box64` and then `target remote 127.0.0.1:1234` to connect to the gdbserver (or use actual IP if not on the machine). After that, the procedure is the same as with ` BOX64_JITGDB=1`.
 This mode can be usefullwhen programs redirect all console output to a file (like Unity3D Games)
+ * 3 : Launch `lldb` when a segfault, bus error or illegal instruction signal is trapped, attached to the offending process and go in an endless loop, waiting.

 #### BOX64_NORCFILES
 If the env var exist, no rc files (like /etc/box64.box64rc and ~/.box64rc) will be loaded

+#### BOX64_RCFILE
+If the env var is set and file exists, this variable will be used as path to the box64rc file instead of default paths (BOX64_RCFILE is loaded first, default paths are not loaded)
+
 ----

 Those variables are only valid inside a rcfile:
@@ -318,10 +366,18 @@ Those variables are only valid inside a rcfile:
 * 0 : Nothing special
 * 1 : Added "--no-sandbox" to command line arguments (usefull for chrome based programs)

+#### BOX64_INPROCESSGPU
+ * 0 : Nothing special
+ * 1 : Added "--in-process-gpu" to command line arguments (usefull for chrome based programs)
+
+#### BOX64_CEFDISABLEGPU
+ * 0 : Nothing special
+ * 1 : Added "-cef-disable-gpu" to command line arguments (usefull for steamwebhelper/cef based programs)
+
+#### BOX64_CEFDISABLEGPUCOMPOSITOR
+ * 0 : Nothing special
+ * 1 : Added "-cef-disable-gpu-compositor" to command line arguments (usefull for steamwebhelper/cef based programs)
+
 #### BOX64_EXIT
 * 0 : Nothing special
 * 1 : Just exit, don't try to run the program
-
-#### BOX64_RESERVE_HIGH
-* 0 : Don't try to pe-reserve high memory (beyond 47bits) (Default)
-* 1 : Try to reserve (without allocating it) memory beyond 47bits (seems unstable)
@@ -55,6 +55,11 @@ F<~/.bashrc> may override the setting from your command line. Example:
 When set to 1, add C<--no-sandbox> to command line arguments (useful for chrome
 based programs).

+=item B<BOX64_INPROCESSGPU=0|1>
+
+When set to 1, add C<--in-process-gpu> to command line arguments (useful for chrome
+based programs).
+
 =item B<BOX64_EXIT>=I<0|1>

 When set to 1, just exit, don't try to run the program.
@@ -146,7 +151,7 @@ Only on build with trace enabled. Trace allow the logging of all instruction exe

 =item B<BOX64_TRACE_INIT>=I<0|1>

-Use B<BOX64_TRACE_INIT> instead of B<BOX_TRACE> to start trace before the
+Use B<BOX64_TRACE_INIT> instead of B<BOX64_TRACE> to start trace before the
 initialization of libraries and the running program

    * 0 : No trace. (Default.)
@@ -259,15 +264,15 @@ down the program a lot and generate huge logs.
 =item B<BOX64_NODYNAREC>=I<0xXXXXXXXX-0xYYYYYYYY>

 Forbid dynablock creation in the interval specified (helpfull for debugging
-behaviour difference between Dynarec and Interpretor)
+behaviour difference between Dynarec and Interpreter)

 =item B<BOX64_DYNAREC_TEST>=I<0|1>

-Enables/Disables test for generated code. This will run interpreter along the Dynacrec, so this will slow
-down the program a lot.
+Dynarec will compare it's execution with the interpreter (super slow, only for testing)

-    * 0 : No test for generated code. (Default.)
-    * 1 : Enable Tests for generated code
+    * 0 : No comparison. (Default.)
+    * 1 : Each opcode runs on interepter and on Dynarec, and regs and memory are compared and print if different.
+    * 2 : Thread-safe tests, extremely slow.

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

@@ -329,20 +334,6 @@ Optimisation of CALL/RET opcodes (not compatible with jit/dynarec/smc)
    * 0 : Don't optimize CALL/RET, use Jump Table for boths (Default)
    * 1 : Try to optimized CALL/RET, skipping the use of the JumpTable when possible (will crash if blacks are invalidate, so probably incompatible with JIT/Dynarec)

-=item B<BOX64_DYNAREC_HOTPAGE>=I<0|1-255>
-
-Handling of HotPage (Page beeing both executed and writen)
-
-    * 0 : Don't track hotpage
-    * 1-255 : Track HotPage, and disable execution of a page beeing writen for N attempts (default is 4)
-
-=item B<BOX64_DYNAREC_FASTPAGE>=I<0|1>
-
-Will use a faster handling of HotPage (Page beeing both executed and writen)
-
-    * 0 : use regular hotpage (Default)
-    * 1 : Use faster hotpage, taking the risk of running obsolete JIT code (might be faster, but more prone to crash)
-
 =item B<BOX64_DYNAREC_BLEEDING_EDGE>=I<0|1>

 Detect MonoBleedingEdge and apply conservative settings
@@ -371,6 +362,13 @@ Handling of x87 80bits long double
    * 0 : Try to handle 80bits long double as precise as possible (Default)
    * 1 : Handle them as double

+=item B<BOX64_SYNC_ROUNDING>=I<0|1>
+
+Box64 will sync rounding mode with fesetround/fegetround.
+
+    * 0 : Disable rounding mode syncing. (Default.)
+    * 1 : Enable rounding mode syncing.
+
 =item B<BOX64_LIBCEF>=I<0|1>

 Detect libcef and apply malloc_hack settings
@@ -451,6 +449,13 @@ Disables the load of vulkan libraries.
    * 0 : Load vulkan libraries if found.
    * 1 : Disables the load of vulkan libraries, both the native and the i386 version (can be useful on Pi4, where the vulkan driver is not quite there yet.)

+=item B<BOX64_FUTEX_WAITV>=I<0|1>
+
+Use of the new fuext_waitc syscall
+
+     * 0 : Do not try to use it, return unsupported (default for BAD_SIGNAL build)
+    * 1 : let program use the syscall if the host system support it (Default for other build)
+
 =item B<BOX64_BASH>=I<yyyy>

 Define x86_64 bash to launch script. Will use yyyy as x86_64 bash to launch
@@ -462,4 +467,17 @@ script. yyyy needs to be a full path to a valid x86_64 version of bash
    * 1 : Launch `gdb` when a segfault, bus error or illegal instruction signal is trapped, attached to the offending process and go in an endless loop, waiting. When in gdb, you need to find the correct thread yourself (the one with `my_box64signalhandler` in is stack) then probably need to `finish` 1 or 2 functions (inside `usleep(..)`) and then you'll be in `my_box64signalhandler`, just before the printf of the Segfault message. Then simply  `set waiting=0` to exit the infinite loop.
    * 2 : Launch `gdbserver` when a segfault, bus error or illegal instruction signal is trapped, attached to the offending process, and go in an endless loop, waiting. Use `gdb /PATH/TO/box64` and then `target remote 127.0.0.1:1234` to connect to the gdbserver (or use actual IP if not on the machine). After that, the procedure is the same as with ` BOX64_JITGDB=1`. This mode can be usefullwhen programs redirect all console output to a file (like Unity3D Games)

+=head1 VARIABLES FOR ENVIRONMENT ONLY
+
+=over 8
+
+=item B<BOX64_VERSION>
+
+When set, B<box64> will only print its version and then exit. This option is
+analogous to B<QEMU_VERSION> used by C<binfmt_misc> emulation of qemu-user. Its
+purpose is the same: provide a way to soft-disable C<binfmt_misc> emulation as
+well as for identifying B<box64> as the registered C<binfmt_misc> emulator.
+
+=back
+
 =cut
@@ -150,12 +150,12 @@ class FunctionConvention(object):
 		self.ident = ident
 		self.name = convname
 		self.values = valid_chars
-# Free letters:  B   FG  J      QR T   XYZab  e gh jk mno qrst   xyz
+# Free letters:  B   FG  J      QR T    YZa   e gh jk mno qrst    z
 conventions = {
-	'F': FunctionConvention('F', "System V", ['E', 'v', 'c', 'w', 'i', 'I', 'C', 'W', 'u', 'U', 'f', 'd', 'D', 'K', 'l', 'L', 'p', 'V', 'O', 'S', 'N', 'M', 'H', 'P', 'A', 'x', 'X']),
+	'F': FunctionConvention('F', "System V", ['E', 'v', 'c', 'w', 'i', 'I', 'C', 'W', 'u', 'U', 'f', 'd', 'D', 'K', 'l', 'L', 'p', 'V', 'O', 'S', 'N', 'M', 'H', 'P', 'A', 'x', 'X', 'Y', 'y', 'b']),
 	'W': FunctionConvention('W', "Windows",  ['E', 'v', 'c', 'w', 'i', 'I', 'C', 'W', 'u', 'U', 'f', 'd',      'K', 'l', 'L', 'p', 'V', 'O', 'S', 'N', 'M',      'P', 'A'])
 }
-sortedvalues = ['E', 'v', 'c', 'w', 'i', 'I', 'C', 'W', 'u', 'U', 'f', 'd', 'D', 'K', 'l', 'L', 'p', 'V', 'O', 'S', 'N', 'M', 'H', 'P', 'A', 'x', 'X', '0', '1']
+sortedvalues = ['E', 'v', 'c', 'w', 'i', 'I', 'C', 'W', 'u', 'U', 'f', 'd', 'D', 'K', 'l', 'L', 'p', 'V', 'O', 'S', 'N', 'M', 'H', 'P', 'A', 'x', 'X', 'Y', 'y', 'b', '0', '1']
 assert(all(all(c not in conv.values[:i] and c in sortedvalues for i, c in enumerate(conv.values)) for conv in conventions.values()))

 class FunctionType(str):
@@ -197,7 +197,7 @@ class FunctionType(str):
 		except ValueError as e:
 			raise ValueError("Value is " + self) from e
 	
-	def __getitem__(self, i: Union[int, slice]) -> 'FunctionType':
+	def __getitem__(self, i: Union[int, slice]) -> 'FunctionType': # type: ignore [override]
 		return FunctionType(super().__getitem__(i))

 RedirectType = NewType('RedirectType', FunctionType)
@@ -399,9 +399,12 @@ def readFiles(files: Iterable[Filename]) -> Tuple[JumbledGlobals, JumbledRedirec
 							 or match("sdl1mixer", "sdl2mixer") \
 							 or match("sdl1net", "sdl2net") \
 							 or match("sdl1ttf", "sdl2ttf") \
-							 or match("libGL", "libEGL") \
+							 or match("libgl", "libegl") \
+							 or match("libgl", "glesv2") \
+							 or match("libegl", "glesv2") \
 							 or match("libc", "tcmallocminimal") \
 							 or match("libc", "tbbmallocproxy") \
+							 or match("libc", "androidshmem") \
 							 or match("tcmallocminimal", "tbbmallocproxy"):
 								continue
 							
@@ -815,57 +818,84 @@ def main(root: str, files: Iterable[Filename], ver: str):
 		return 0
 	
 	# Detect simple wrappings
-	simple_wraps: Dict[ClausesStr, List[Tuple[FunctionType, int]]] = {}
 	allowed_conv_ident = "F"
 	allowed_conv = conventions[allowed_conv_ident]
 	
 	# H could be allowed maybe?
-	allowed_simply: str = "v"
-	allowed_regs  : str = "cCwWiuIUlLp"
-	allowed_fpr   : str = "fd"
+	allowed_simply: Dict[str, str] = {"ARM64": "v", "RV64": "v"}
+	allowed_regs  : Dict[str, str] = {"ARM64": "cCwWiuIUlLp", "RV64": "CWuIUlLp"}
+	allowed_fpr   : Dict[str, str] = {"ARM64": "fd", "RV64": "fd"}
+	
+	# Detect functions which return in an x87 register
+	retx87_wraps: Dict[ClausesStr, List[FunctionType]] = {}
+	return_x87: str = "DK"
 	
 	# Sanity checks
-	forbidden_simple: str = "EDKVOSNMHPAxX"
-	assert(len(allowed_simply) + len(allowed_regs) + len(allowed_fpr) + len(forbidden_simple) == len(allowed_conv.values))
-	assert(all(c not in allowed_regs for c in allowed_simply))
-	assert(all(c not in allowed_simply + allowed_regs for c in allowed_fpr))
-	assert(all(c not in allowed_simply + allowed_regs + allowed_fpr for c in forbidden_simple))
-	assert(all(c in allowed_simply + allowed_regs + allowed_fpr + forbidden_simple for c in allowed_conv.values))
+	forbidden_simple: Dict[str, str] = {"ARM64": "EDKVOSNMHPAxXYyb", "RV64": "EcwiDKVOSNMHPAxXYyb"}
+	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))
+	for k1 in forbidden_simple:
+		assert(len(allowed_simply[k1]) + len(allowed_regs[k1]) + len(allowed_fpr[k1]) + len(forbidden_simple[k1]) == len(allowed_conv.values))
+		assert(all(c not in allowed_regs[k1] for c in allowed_simply[k1]))
+		assert(all(c not in allowed_simply[k1] + allowed_regs[k1] for c in allowed_fpr[k1]))
+		assert(all(c not in allowed_simply[k1] + allowed_regs[k1] + allowed_fpr[k1] for c in forbidden_simple[k1]))
+		assert(all(c in allowed_simply[k1] + allowed_regs[k1] + allowed_fpr[k1] + forbidden_simple[k1] for c in allowed_conv.values))
+	assert(all(c in allowed_conv.values for c in return_x87))
 	
-	def check_simple(v: FunctionType) -> Optional[int]:
+	simple_wraps: Dict[str, Dict[ClausesStr, List[Tuple[FunctionType, int]]]] = {
+		k1: {} for k1 in forbidden_simple
+	}
+	
+	def check_simple(v: FunctionType) -> Dict[str, int]:
 		regs_count: int = 0
 		fpr_count : int = 0
 		
-		if v.get_convention() is not allowed_conv:
-			return None
-		if v[0] in forbidden_simple:
-			return None
-		for c in v[2:]:
-			if c in allowed_regs:
-				regs_count = regs_count + 1
-			elif c in allowed_fpr:
-				fpr_count = fpr_count + 1
-			elif c in allowed_simply:
+		ret = {}
+		for k in forbidden_simple:
+			if v.get_convention() is not allowed_conv:
 				continue
+			if v[0] in forbidden_simple[k]:
+				continue
+			for c in v[2:]:
+				if c in allowed_regs[k]:
+					regs_count = regs_count + 1
+				elif c in allowed_fpr[k]:
+					fpr_count = fpr_count + 1
+				elif c in allowed_simply[k]:
+					continue
+				else:
+					break
 			else:
-				return None
-		# No character in forbidden_simply
-		if (regs_count <= 6) and (fpr_count <= 8):
-			# All checks passed!
-			ret_val = 1 + fpr_count
-			if v[0] in allowed_fpr:
-				ret_val = -ret_val
-			return ret_val
-		else:
-			# Too many arguments...
-			return None
+				# No character in forbidden_simply
+				if (regs_count <= 6) and (fpr_count <= 8):
+					# All checks passed!
+					ret_val = 1 + fpr_count
+					if v[0] in allowed_fpr[k]:
+						ret_val = -ret_val
+					ret[k] = ret_val
+				# Else, too many arguments
+		return ret
 	
 	# Only search in real wrappers (mapped ones are nearly always not simple)
 	for k in gbls:
-		tmp = [ (v, ret_val) for v, ret_val in map(lambda v: (v, check_simple(v)), gbls[k]) if ret_val is not None ]
+		for v in gbls[k]:
+			simples = check_simple(v)
+			for k1, i in simples.items():
+				if k in simple_wraps[k1]:
+					simple_wraps[k1][k].append((v, i))
+				else:
+					simple_wraps[k1][k] = [(v, i)]
+	simple_idxs = { k1: sorted(v1.keys(), key=lambda x: Clauses(x).splitdef()) for k1, v1 in simple_wraps.items() }
+	
+	def check_return_x87(v: FunctionType) -> bool:
+		return v[0] in return_x87
+    
+	for k in gbls:
+		tmp = [v for v in gbls[k] if check_return_x87(v)]
 		if tmp:
-			simple_wraps[k] = tmp
-	simple_idxs = sorted(simple_wraps.keys(), key=lambda x: Clauses(x).splitdef())
+			retx87_wraps[k] = tmp
+	retx87_idxs = sorted(retx87_wraps.keys(), key=lambda x: Clauses(x).splitdef())
 	
 	# Now the files rebuilding part
 	# File headers and guards
@@ -903,6 +933,8 @@ def main(root: str, files: Iterable[Filename], ver: str):
 		#define ST0val ST0.d
 		
 		int of_convert(int);
+		void* align_xcb_connection(void* src);
+		void unalign_xcb_connection(void* src, void* dst);
 		
 		""",
 		"wrapper.h": """
@@ -939,6 +971,7 @@ def main(root: str, files: Iterable[Filename], ver: str):
 		// 0 = constant 0, 1 = constant 1
 		// x = float complex
 		// X = double complex
+		// b = xcb_connection_t*
 		
 		""",
 		"fntypes.h": """
@@ -1000,8 +1033,8 @@ def main(root: str, files: Iterable[Filename], ver: str):
 	# Rewrite the wrapper.c file:
 	# i and u should only be 32 bits
 	td_types = {
-		#      E            v       c         w          i          I          C          W           u           U           f        d         D              K         l           L            p        V        O          S        N      M      H                    P        A			x				X
-		'F': ["x64emu_t*", "void", "int8_t", "int16_t", "int32_t", "int64_t", "uint8_t", "uint16_t", "uint32_t", "uint64_t", "float", "double", "long double", "double", "intptr_t", "uintptr_t", "void*", "void*", "int32_t", "void*", "...", "...", "unsigned __int128", "void*", "void*", "complexf_t", "complex_t"],
+		#      E            v       c         w          i          I          C          W           u           U           f        d         D              K         l           L            p        V        O          S        N      M      H                    P        A			x				X           Y              y         b
+		'F': ["x64emu_t*", "void", "int8_t", "int16_t", "int32_t", "int64_t", "uint8_t", "uint16_t", "uint32_t", "uint64_t", "float", "double", "long double", "double", "intptr_t", "uintptr_t", "void*", "void*", "int32_t", "void*", "...", "...", "unsigned __int128", "void*", "void*", "complexf_t", "complex_t", "complexl_t", "complex_t", "void*"],
 		#      E            v       c         w          i          I          C          W           u           U           f        d         K         l           L            p        V        O          S        N      M      P        A
 		'W': ["x64emu_t*", "void", "int8_t", "int16_t", "int32_t", "int64_t", "uint8_t", "uint16_t", "uint32_t", "uint64_t", "float", "double", "double", "intptr_t", "uintptr_t", "void*", "void*", "int32_t", "void*", "...", "...", "void*", "void*"]
 	}
@@ -1059,8 +1092,11 @@ def main(root: str, files: Iterable[Filename], ver: str):
 				"unsigned __int128 u128 = fn({0}); R_RAX=(u128&0xFFFFFFFFFFFFFFFFL); R_RDX=(u128>>64)&0xFFFFFFFFFFFFFFFFL;", # H
 				"\n#error Invalid return type: pointer in the stack\n",    # P
 				"\n#error Invalid return type: va_list\n",                 # A
-				'from_complexf(emu, fn({0})));', 	                          # x
-				'from_complex(emu, fn({0})));'                             # X
+				'from_complexf(emu, fn({0}));', 	                       # x
+				'from_complex(emu, fn({0}));',                             # X
+				'from_complexl(emu, fn({0}));', 	                       # Y
+				'from_complexk(emu, fn({0}));',                            # y
+				"\n#error Invalid return type: xcb_connection_t*\n",       # b
 			],
 			conventions['W']: [
 				"\n#error Invalid return type: emulator\n",                # E
@@ -1090,17 +1126,17 @@ def main(root: str, files: Iterable[Filename], ver: str):
 		}
 		
 		# vreg: value is in a general register
-		#         E  v  c  w  i  I  C  W  u  U  f  d  D  K  l  L  p  V  O  S  N  M  H  P  A  x  X
-		vreg   = [0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 0, 1, 1, 1, 2, 2, 0, 1, 0, 0]
+		#         E  v  c  w  i  I  C  W  u  U  f  d  D  K  l  L  p  V  O  S  N  M  H  P  A  x  X  Y  y  b
+		vreg   = [0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 0, 1, 1, 1, 2, 2, 0, 1, 0, 0, 0, 0, 1]
 		# vxmm: value is in a XMM register
-		#         E  v  c  w  i  I  C  W  u  U  f  d  D  K  l  L  p  V  O  S  N  M  H  P  A  x  X
-		vxmm   = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 2]
+		#         E  v  c  w  i  I  C  W  u  U  f  d  D  K  l  L  p  V  O  S  N  M  H  P  A  x  X  Y  y  b
+		vxmm   = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 2, 0, 0, 0]
 		# vother: value is elsewere
-		#         E  v  c  w  i  I  C  W  u  U  f  d  D  K  l  L  p  V  O  S  N  M  H  P  A  x  X
-		vother = [1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0]
+		#         E  v  c  w  i  I  C  W  u  U  f  d  D  K  l  L  p  V  O  S  N  M  H  P  A  x  X  Y  y  b
+		vother = [1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
 		# vstack: value is on the stack (or out of register)
-		#         E  v  c  w  i  I  C  W  u  U  f  d  D  K  l  L  p  V  O  S  N  M  H  P  A  x  X
-		vstack = [0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 1, 1, 1, 0, 1, 1, 1, 2, 2, 1, 1, 1, 2]
+		#         E  v  c  w  i  I  C  W  u  U  f  d  D  K  l  L  p  V  O  S  N  M  H  P  A  x  X  Y  y  b
+		vstack = [0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 1, 1, 1, 0, 1, 1, 1, 2, 2, 1, 1, 1, 2, 4, 4, 1]
 		arg_r = [
 			"",                            # E
 			"",                            # v
@@ -1129,64 +1165,73 @@ def main(root: str, files: Iterable[Filename], ver: str):
 			"(void*){p}, ",                # A
 			"",	                           # x
 			"",                            # X
+			"",							   # Y
+			"",							   # y
+			"aligned_xcb, ",               # b
 		]
 		arg_x = [
-			"",                      # E
-			"",                      # v
-			"",                      # c
-			"",                      # w
-			"",                      # i
-			"",                      # I
-			"",                      # C
-			"",                      # W
-			"",                      # u
-			"",                      # U
-			"emu->xmm[{p}].f[0], ",  # f
-			"emu->xmm[{p}].d[0], ",  # d
-			"",                      # D
-			"",                      # K
-			"",                      # l
-			"",                      # L
-			"",                      # p
-			"",                      # V
-			"",                      # O
-			"",                      # S
-			"",                      # N
-			"",                      # M
-			"",                      # H
-			"",                      # P
-			"",                      # A
-			"to_complexf(emu, {p}),",# x
-			"to_complex(emu, {p}),"  # X
+			"",                       # E
+			"",                       # v
+			"",                       # c
+			"",                       # w
+			"",                       # i
+			"",                       # I
+			"",                       # C
+			"",                       # W
+			"",                       # u
+			"",                       # U
+			"emu->xmm[{p}].f[0], ",   # f
+			"emu->xmm[{p}].d[0], ",   # d
+			"",                       # D
+			"",                       # K
+			"",                       # l
+			"",                       # L
+			"",                       # p
+			"",                       # V
+			"",                       # O
+			"",                       # S
+			"",                       # N
+			"",                       # M
+			"",                       # H
+			"",                       # P
+			"",                       # A
+			"to_complexf(emu, {p}), ", # x
+			"to_complex(emu, {p}), ",  # X
+			"", 				      # Y
+			"", 				      # y
+			"",                       # b
 		]
 		arg_o = [
-			"emu, ",                   # E
-			"",                        # v
-			"",                        # c
-			"",                        # w
-			"",                        # i
-			"",                        # I
-			"",                        # C
-			"",                        # W
-			"",                        # u
-			"",                        # U
-			"",                        # f
-			"",                        # d
-			"",                        # D
-			"",                        # K
-			"",                        # l
-			"",                        # L
-			"",                        # p
-			"(void*)(R_RSP + {p}), ",  # V
-			"",                        # O
-			"",                        # S
-			"",                        # N
-			"",                        # M
-			"",                        # H
-			"",                        # P
-			"",                        # A
-			"",                        # x
-			"",                        # X
+			"emu, ",                  # E
+			"",                       # v
+			"",                       # c
+			"",                       # w
+			"",                       # i
+			"",                       # I
+			"",                       # C
+			"",                       # W
+			"",                       # u
+			"",                       # U
+			"",                       # f
+			"",                       # d
+			"",                       # D
+			"",                       # K
+			"",                       # l
+			"",                       # L
+			"",                       # p
+			"(void*)(R_RSP + {p}), ", # V
+			"",                       # O
+			"",                       # S
+			"",                       # N
+			"",                       # M
+			"",                       # H
+			"",                       # P
+			"",                       # A
+			"",                       # x
+			"",                       # X
+			"",						  # Y
+			"",						  # y
+			"",                       # b
 		]
 		arg_s = [
 			"",                                         # E
@@ -1211,11 +1256,14 @@ def main(root: str, files: Iterable[Filename], ver: str):
 			"io_convert(*(void**)(R_RSP + {p})), ",     # S
 			"*(void**)(R_RSP + {p}), ",                 # N
 			"*(void**)(R_RSP + {p}),*(void**)(R_RSP + {p} + 8), ", # M
-			"*(unsigned __int128)(R_RSP + {p}), ",      # H
+			"*(unsigned __int128*)(R_RSP + {p}), ",     # H
 			"*(void**)(R_RSP + {p}), ",                 # P
 			"*(void**)(R_RSP + {p}), ",                 # A
 			"*(complexf_t*)(R_RSP + {p}), ",            # x
 			"*(complex_t*)(R_RSP + {p}), ",             # X
+			"to_complexl(emu, R_RSP + {p}), ",		    # Y
+			"to_complexk(emu, R_RSP + {p}), ",		    # y
+			"aligned_xcb, ",                            # b
 		]
 		
 		# Asserts
@@ -1243,7 +1291,7 @@ def main(root: str, files: Iterable[Filename], ver: str):
 		for k in conventions:
 			c = conventions[k]
 			if c not in vals:
-				raise NotImplementedError("[{k}]values not in vals".format(k=k, lenval=len(c.values), lenvals=len(vals[c])))
+				raise NotImplementedError("convention {k} not in vals".format(k=k))
 			if len(c.values) != len(vals[c]):
 				raise NotImplementedError("len([{k}]values) = {lenval} != len(vals[...]) = {lenvals}".format(k=k, lenval=len(c.values), lenvals=len(vals[c])))
 		# When arg_* is not empty, v* should not be 0
@@ -1267,6 +1315,50 @@ def main(root: str, files: Iterable[Filename], ver: str):
 		
 		# Helper functions to write the function definitions
 		systemVconv = conventions['F']
+		def function_pre_systemV(args: FunctionType, d: int = 8, r: int = 0, x: int = 0) -> Tuple[Optional[str], str]:
+			# args: string of argument types
+			# d: delta (in the stack)
+			# r: general register no
+			# x: XMM register no
+			if len(args) == 0:
+				return None, ""
+			
+			# Redirections
+			if args[0] == "0":
+				return function_pre_systemV(args[1:], d, r, x)
+			elif args[0] == "1":
+				return function_pre_systemV(args[1:], d, r, x)
+			
+			idx = systemVconv.values.index(args[0])
+			# Name of the registers
+			reg_arg = ["R_RDI", "R_RSI", "R_RDX", "R_RCX", "R_R8", "R_R9"]
+			if args[0] == "b":
+				if 'b' in args[1:]:
+					raise NotImplementedError("Multiple XCB connections unsupported")
+				content = ""
+				if r < len(reg_arg):
+					content = reg_arg[r]
+				else:
+					content = "(R_RSP + " + str(d) + ")"
+				return "(void*)" + content, f"void *aligned_xcb = align_xcb_connection((void*){content}); "
+			elif (r < len(reg_arg)) and (vreg[idx] > 0):
+				for _ in range(vreg[idx]):
+					if r < len(reg_arg):
+						r = r + 1
+					else:
+						d = d + 8
+				return function_pre_systemV(args[1:], d, r, x)
+			elif (x < 8) and (vxmm[idx] > 0):
+				return function_pre_systemV(args[1:], d, r, x+1)
+			elif vstack[idx] > 0:
+				return function_pre_systemV(args[1:], d+8*vstack[idx], r, x)
+			else:
+				return function_pre_systemV(args[1:], d, r, x)
+		def function_post_systemV(content: Optional[str]) -> str:
+			if content is not None:
+				return f" unalign_xcb_connection(aligned_xcb, {content});"
+			else:
+				return ""
 		def function_args_systemV(args: FunctionType, d: int = 8, r: int = 0, x: int = 0) -> str:
 			# args: string of argument types
 			# d: delta (in the stack)
@@ -1299,7 +1391,7 @@ def main(root: str, files: Iterable[Filename], ver: str):
 				return ret + function_args_systemV(args[1:], d, r, x)
 			elif (x < 8) and (vxmm[idx] > 0):
 				# Value is in an XMM register
-				return arg_x[idx].format(p=x) + function_args_systemV(args[1:], d, r, x+1)
+				return arg_x[idx].format(p=x) + function_args_systemV(args[1:], d, r, x+vxmm[idx])
 			elif vstack[idx] > 0:
 				# Value is in the stack
 				return arg_s[idx].format(p=d) + function_args_systemV(args[1:], d+8*vstack[idx], r, x)
@@ -1346,9 +1438,11 @@ def main(root: str, files: Iterable[Filename], ver: str):
 			# Generic function
 			conv = N.get_convention()
 			if conv is systemVconv:
-				f.write(vals[conv][conv.values.index(N[0])].format(function_args_systemV(N[2:])[:-2]) + " }\n")
+				prepost, pre = function_pre_systemV(N[2:])
+				f.write(pre + vals[conv][conv.values.index(N[0])].format(function_args_systemV(N[2:])[:-2]) + function_post_systemV(prepost))
 			else:
-				f.write(vals[conv][conv.values.index(N[0])].format(function_args_windows(N[2:])[:-2]) + " }\n")
+				f.write(vals[conv][conv.values.index(N[0])].format(function_args_windows(N[2:])[:-2]))
+			f.write(" }\n")
 		
 		for k in gbls:
 			if k != str(Clauses()):
@@ -1371,12 +1465,28 @@ def main(root: str, files: Iterable[Filename], ver: str):
 				file.write("#endif\n")
 		
 		# Write the isSimpleWrapper function
-		file.write("\nint isSimpleWrapper(wrapper_t fun) {\n")
-		for k in simple_idxs:
+		inttext = ""
+		file.write("\n")
+		for k1 in simple_idxs:
+			file.write("#{inttext}if defined({k1})\nint isSimpleWrapper(wrapper_t fun) {{\n".format(inttext=inttext, k1=k1))
+			inttext = "el"
+			for k in simple_idxs[k1]:
+				if k != str(Clauses()):
+					file.write("#if " + k + "\n")
+				for vf, val in simple_wraps[k1][k]:
+					file.write("\tif (fun == &" + vf + ") return " + str(val) + ";\n")
+				if k != str(Clauses()):
+					file.write("#endif\n")
+			file.write("\treturn 0;\n}\n")
+		file.write("\n#else\nint isSimpleWrapper(wrapper_t fun) {\n\treturn 0;\n}\n#endif\n")
+		
+		# Write the isRetX87Wrapper function
+		file.write("\nint isRetX87Wrapper(wrapper_t fun) {\n")
+		for k in retx87_idxs:
 			if k != str(Clauses()):
 				file.write("#if " + k + "\n")
-			for vf, val in simple_wraps[k]:
-				file.write("\tif (fun == &" + vf + ") return " + str(val) + ";\n")
+			for vf in retx87_wraps[k]:
+				file.write("\tif (fun == &" + vf + ") return 1;\n")
 			if k != str(Clauses()):
 				file.write("#endif\n")
 		file.write("\treturn 0;\n}\n")
@@ -1454,6 +1564,6 @@ if __name__ == '__main__':
 		if v == "--":
 			limit.append(i)
 	Define.defines = list(map(DefineType, sys.argv[2:limit[0]]))
-	if main(sys.argv[1], sys.argv[limit[0]+1:], "2.2.0.17") != 0:
+	if main(sys.argv[1], sys.argv[limit[0]+1:], "2.4.0.21") != 0:
 		exit(2)
 	exit(0)
@@ -11,6 +11,9 @@ endif( NOT TEST_OUTPUT )
 if( NOT TEST_REFERENCE )
  message( FATAL_ERROR "Require TEST_REFERENCE to be defined" )
 endif( NOT TEST_REFERENCE )
+if(NOT DEFINED ENV{INTERPRETER})
+    set(ENV{INTERPRETER} "")
+endif()

 set(ENV{BOX64_LOG} 0)
 set(ENV{BOX64_NOBANNER} 1)
@@ -24,7 +27,7 @@ endif( EXISTS ${CMAKE_SOURCE_DIR}/x64lib )

 # run the test program, capture the stdout/stderr and the result var
 execute_process(
-  COMMAND ${TEST_PROGRAM} ${TEST_ARGS} ${TEST_ARGS2}
+  COMMAND $ENV{INTERPRETER} ${TEST_PROGRAM} ${TEST_ARGS} ${TEST_ARGS2}
  OUTPUT_FILE ${TEST_OUTPUT}
  ERROR_VARIABLE TEST_ERROR
  RESULT_VARIABLE TEST_RESULT
@@ -21,6 +21,8 @@
 #include "signals.h"
 #include "rcfile.h"
 #include "gltools.h"
+#include "rbtree.h"
+#include "dynarec.h"

 EXPORTDYN
 void initAllHelpers(box64context_t* context)
@@ -68,6 +70,8 @@ void free_tlsdatasize(void* p)
    tlsdatasize_t *data = (tlsdatasize_t*)p;
    box_free(data->ptr);
    box_free(p);
+    if(my_context)
+        pthread_setspecific(my_context->tlskey, NULL);
 }

 void x64Syscall(x64emu_t *emu);
@@ -209,8 +213,12 @@ box64context_t *NewBox64Context(int argc)

    init_custommem_helper(context);

-    context->maplib = NewLibrarian(context, 1);
-    context->local_maplib = NewLibrarian(context, 1);
+    #ifdef DYNAREC
+    x64test_init();
+    #endif
+
+    context->maplib = NewLibrarian(context);
+    context->local_maplib = NewLibrarian(context);
    context->versym = NewDictionnary();
    context->system = NewBridge();
    // Cannot use Bridge name as the map is not initialized yet
@@ -227,7 +235,9 @@ box64context_t *NewBox64Context(int argc)
    // create exit bridge
    context->exit_bridge = AddBridge(context->system, NULL, NULL, 0, NULL);
    // get handle to box64 itself
+    #ifndef STATICBUILD
    context->box64lib = dlopen(NULL, RTLD_NOW|RTLD_GLOBAL);
+    #endif
    context->dlprivate = NewDLPrivate();

    context->argc = argc;
@@ -259,7 +269,14 @@ box64context_t *NewBox64Context(int argc)
    context->segtls[4].present = 1;
    context->segtls[4].is32bits = 1;

+    context->globdata = NewMapSymbols();
+    context->uniques = NewMapSymbols();
+
    initAllHelpers(context);
+    
+    #ifdef DYNAREC
+    context->db_sizes = init_rbtree();
+    #endif

    return context;
 }
@@ -276,6 +293,7 @@ void FreeBox64Context(box64context_t** context)

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

+    //clean_current_emuthread();    // cleaning main thread seems a bad idea
    if(ctx->local_maplib)
        FreeLibrarian(&ctx->local_maplib, NULL);
    if(ctx->maplib)
@@ -330,8 +348,10 @@ void FreeBox64Context(box64context_t** context)

    FreeBridge(&ctx->system);

+    #ifndef STATICBUILD
    freeGLProcWrapper(ctx);
    freeALProcWrapper(ctx);
+    #endif

    if(ctx->stack_clone)
        box_free(ctx->stack_clone);
@@ -340,7 +360,6 @@ void FreeBox64Context(box64context_t** context)
    void* ptr;
    if ((ptr = pthread_getspecific(ctx->tlskey)) != NULL) {
        free_tlsdatasize(ptr);
-        pthread_setspecific(ctx->tlskey, NULL);
    }
    pthread_key_delete(ctx->tlskey);

@@ -353,6 +372,14 @@ void FreeBox64Context(box64context_t** context)
    if(ctx->emu_sig)
        FreeX64Emu(&ctx->emu_sig);

+    FreeMapSymbols(&ctx->globdata);
+    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));
+    delete_rbtree(ctx->db_sizes);
+#endif
+
    finiAllHelpers(ctx);

 #ifdef DYNAREC
@@ -3,6 +3,6 @@

 #define BOX64_MAJOR 0
 #define BOX64_MINOR 2
-#define BOX64_REVISION 3
+#define BOX64_REVISION 8

 #endif //__BOX64_VERSION_H_
@@ -9,7 +9,7 @@

 void PrintBox64Version()
 {
-    printf("Box64%s%s v%d.%d.%d %s built on %s %s\n", 
+    printf_log(LOG_NONE, "Box64%s%s v%d.%d.%d %s built on %s %s\n", 
    #ifdef HAVE_TRACE
        " with trace",
    #else
@@ -70,64 +70,71 @@
 #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

+int convert_bitmask(uint64_t bitmask);
+#define convert_bitmask_w(A)    convert_bitmask(((uint64_t)(A) << 32) + (uint32_t)(A))
+#define convert_bitmask_x(A)    convert_bitmask((uint64_t)A)
+#define convert_bitmask_xw(A)   convert_bitmask(rex.w?((uint64_t)(A)):(((uint64_t)(A) << 32) + (uint32_t)(A)))
+
 #define invCond(cond)   ((cond)^0b0001)

 // MOVZ
 #define MOVZ_gen(sf, hw, imm16, Rd)         ((sf)<<31 | 0b10<<29 | 0b100101<<23 | (hw)<<21 | (imm16)<<5 | (Rd))
-#define MOVZx(Rd, imm16)                    EMIT(MOVZ_gen(1, 0, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVZx_LSL(Rd, imm16, shift)         EMIT(MOVZ_gen(1, (shift)/16, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVZw(Rd, imm16)                    EMIT(MOVZ_gen(0, 0, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVZw_LSL(Rd, imm16, shift)         EMIT(MOVZ_gen(0, (shift)/16, ((uint16_t)imm16)&0xffff, Rd))
+#define MOVZx(Rd, imm16)                    EMIT(MOVZ_gen(1, 0, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVZx_LSL(Rd, imm16, shift)         EMIT(MOVZ_gen(1, (shift)/16, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVZw(Rd, imm16)                    EMIT(MOVZ_gen(0, 0, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVZw_LSL(Rd, imm16, shift)         EMIT(MOVZ_gen(0, (shift)/16, ((uint16_t)(imm16))&0xffff, Rd))

 // MOVN
 #define MOVN_gen(sf, hw, imm16, Rd)         ((sf)<<31 | 0b00<<29 | 0b100101<<23 | (hw)<<21 | (imm16)<<5 | (Rd))
-#define MOVNx(Rd, imm16)                    EMIT(MOVN_gen(1, 0, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVNx_LSL(Rd, imm16, shift)         EMIT(MOVN_gen(1, (shift)/16, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVNw(Rd, imm16)                    EMIT(MOVN_gen(0, 0, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVNw_LSL(Rd, imm16, shift)         EMIT(MOVN_gen(0, (shift)/16, ((uint16_t)imm16)&0xffff, Rd))
+#define MOVNx(Rd, imm16)                    EMIT(MOVN_gen(1, 0, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVNx_LSL(Rd, imm16, shift)         EMIT(MOVN_gen(1, (shift)/16, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVNw(Rd, imm16)                    EMIT(MOVN_gen(0, 0, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVNw_LSL(Rd, imm16, shift)         EMIT(MOVN_gen(0, (shift)/16, ((uint16_t)(imm16))&0xffff, Rd))

 // MOVK
 #define MOVK_gen(sf, hw, imm16, Rd)         ((sf)<<31 | 0b11<<29 | 0b100101<<23 | (hw)<<21 | (imm16)<<5 | (Rd))
-#define MOVKx(Rd, imm16)                    EMIT(MOVK_gen(1, 0, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVKx_LSL(Rd, imm16, shift)         EMIT(MOVK_gen(1, (shift)/16, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVKw(Rd, imm16)                    EMIT(MOVK_gen(0, 0, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVKw_LSL(Rd, imm16, shift)         EMIT(MOVK_gen(0, (shift)/16, ((uint16_t)imm16)&0xffff, Rd))
+#define MOVKx(Rd, imm16)                    EMIT(MOVK_gen(1, 0, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVKx_LSL(Rd, imm16, shift)         EMIT(MOVK_gen(1, (shift)/16, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVKw(Rd, imm16)                    EMIT(MOVK_gen(0, 0, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVKw_LSL(Rd, imm16, shift)         EMIT(MOVK_gen(0, (shift)/16, ((uint16_t)(imm16))&0xffff, Rd))

-// This macro will give a -Wsign-compare warning, probably bug #38341
-#define MOV32w(Rd, imm32) \
-    if(~((uint32_t)(imm32))<0xffffu) {                                      \
-        MOVNw(Rd, (~(uint32_t)(imm32))&0xffff);                             \
-    } else {                                                                \
-        MOVZw(Rd, (imm32)&0xffff);                                          \
-        if((imm32)&0xffff0000) {MOVKw_LSL(Rd, ((imm32)>>16)&0xffff, 16);}   \
-    }
-#define MOV64x(Rd, imm64) \
-    if(~((uint64_t)(imm64))<0xffff) {                                                                       \
-        MOVNx(Rd, (~(uint64_t)(imm64))&0xffff);                                                             \
-    } else {                                                                                                \
-        MOVZx(Rd, ((uint64_t)(imm64))&0xffff);                                                              \
-        if(((uint64_t)(imm64))&0xffff0000) {MOVKx_LSL(Rd, (((uint64_t)(imm64))>>16)&0xffff, 16);}           \
-        if(((uint64_t)(imm64))&0xffff00000000LL) {MOVKx_LSL(Rd, (((uint64_t)(imm64))>>32)&0xffff, 32);}      \
-        if(((uint64_t)(imm64))&0xffff000000000000LL) {MOVKx_LSL(Rd, (((uint64_t)(imm64))>>48)&0xffff, 48);}  \
-    }
+#define MOV32w(Rd, imm32) arm64_move32(dyn, ninst, Rd, imm32)
+#define MOV64x(Rd, imm64) arm64_move64(dyn, ninst, Rd, imm64)

 #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);}
@@ -200,6 +207,12 @@
 #define SBCSw_REG(Rd, Rn, Rm)       EMIT(ADDSUBC_gen(0, 1, 1, Rm, Rn, Rd))
 #define SBCSxw_REG(Rd, Rn, Rm)      EMIT(ADDSUBC_gen(rex.w, 1, 1, Rm, Rn, Rd))

+// 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))
+#define CCMPw(Wn, Wm, nzcv, cond)   EMIT(CCMP_reg(0, Wm, cond, Wn, nzcv))
+#define CCMPx(Xn, Xm, nzcv, cond)   EMIT(CCMP_reg(1, Xm, cond, Xn, nzcv))
+#define CCMPxw(Xn, Xm, nzcv, cond)  EMIT(CCMP_reg(rex.w, Xm, cond, Xn, nzcv))
+
 // ADR
 #define ADR_gen(immlo, immhi, Rd)   ((immlo)<<29 | 0b10000<<24 | (immhi)<<5 | (Rd))
 #define ADR_S20(Rd, imm)            EMIT(ADR_gen((imm)&3, ((imm)>>2)&0x7ffff, (Rd)))
@@ -240,6 +253,7 @@
 #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 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 LDRH_REG(Rt, Rn, Rm)            EMIT(LDR_REG_gen(0b01, Rm, 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))
@@ -363,6 +377,9 @@
 #define POP2_32(reg1, reg2)     LDPw_S7_postindex(reg1, reg2, xRSP, 8)
 #define PUSH2_32(reg1, reg2)    STPw_S7_preindex(reg2, reg1, xRSP, -8)

+#define POP1_16(reg)            LDRH_S9_postindex(reg, xRSP, 2)
+#define PUSH1_16(reg)           STRH_S9_preindex(reg, xRSP, -2)
+
 #define POP1z(reg)              if(rex.is32bits) {POP1_32(reg);} else {POP1(reg);}
 #define PUSH1z(reg)             if(rex.is32bits) {PUSH1_32(reg);} else {PUSH1(reg);}
 #define POP2z(reg1, reg2)       if(rex.is32bits) {POP2_32(reg1, reg2);} else {POP2(reg1, reg2);}
@@ -424,6 +441,13 @@
 // Data Memory Barrier
 #define DMB_gen(CRm)                    (0b1101010100<<22 | 0b011<<16 | 0b0011<<12 | (CRm)<<8 | 1<<7 | 0b01<<5 | 0b11111)
 #define DMB_ISH()                       EMIT(DMB_gen(0b1011))
+#define DMB_SY()                        EMIT(DMB_gen(0b1111))
+
+// Data Synchronization Barrier
+#define DSB_gen(CRm)                    (0b1101010100<<22 | 0b011<<16 | 0b0011<<12 | (CRm)<<8 | 1<<7 | 0b00<<5 | 0b11111)
+#define DSB_ISH()                       EMIT(DSB_gen(0b1011))
+#define DSB_ISHST()                     EMIT(DSB_gen(0b1010))
+#define DSB_SY()                        EMIT(DSB_gen(0b1111))

 // Break
 #define BRK_gen(imm16)                  (0b11010100<<24 | 0b001<<21 | (((imm16)&0xffff)<<5))
@@ -487,27 +511,33 @@
 #define CSNEGx(Rd, Rn, Rm, cond)            EMIT(CSNEG_gen(1, Rm, cond, Rn, Rd))
 #define CSNEGw(Rd, Rn, Rm, cond)            EMIT(CSNEG_gen(0, Rm, cond, Rn, Rd))
 #define CSNEGxw(Rd, Rn, Rm, cond)           EMIT(CSNEG_gen(rex.w, Rm, cond, Rn, Rd))
-#define CNEGx(Rd, Rn, cond)                 CSNEGx(Rn, Rn, Rn, invCond(cond))
-#define CNEGw(Rd, Rn, cond)                 CSNEGw(Rn, Rn, Rn, invCond(cond))
-#define CNEGxw(Rd, Rn, cond)                CSNEGxw(Rn, Rn, Rn, invCond(cond))
+#define CNEGx(Rd, Rn, cond)                 CSNEGx(Rn, Rn, Rd, invCond(cond))
+#define CNEGw(Rd, Rn, cond)                 CSNEGw(Rn, Rn, Rd, invCond(cond))
+#define CNEGxw(Rd, Rn, cond)                CSNEGxw(Rn, Rn, Rd, invCond(cond))

 // AND / ORR
 #define LOGIC_gen(sf, opc, N, immr, imms, Rn, Rd)  ((sf)<<31 | (opc)<<29 | 0b100100<<23 | (N)<<22 | (immr)<<16 | (imms)<<10 | (Rn)<<5 | Rd)
 // logic to get the mask is ... convoluted... list of possible value there: https://gist.github.com/dinfuehr/51a01ac58c0b23e4de9aac313ed6a06a
 #define ANDx_mask(Rd, Rn, N, immr, imms)    EMIT(LOGIC_gen(1, 0b00, N, immr, imms, Rn, Rd))
 #define ANDw_mask(Rd, Rn, immr, imms)       EMIT(LOGIC_gen(0, 0b00, 0, immr, imms, Rn, Rd))
+#define ANDxw_mask(Rd, Rn, N, immr, imms)   EMIT(LOGIC_gen(rex.w, 0b00, rex.w?(N):0, immr, imms, Rn, Rd))
 #define ANDSx_mask(Rd, Rn, N, immr, imms)   EMIT(LOGIC_gen(1, 0b11, N, immr, imms, Rn, Rd))
 #define ANDSw_mask(Rd, Rn, immr, imms)      EMIT(LOGIC_gen(0, 0b11, 0, immr, imms, Rn, Rd))
+#define ANDSxw_mask(Rd, Rn, N, immr, imms)  EMIT(LOGIC_gen(rex.w, 0b11, rex.w?(N):0, immr, imms, Rn, Rd))
 #define ORRx_mask(Rd, Rn, N, immr, imms)    EMIT(LOGIC_gen(1, 0b01, N, immr, imms, Rn, Rd))
 #define ORRw_mask(Rd, Rn, immr, imms)       EMIT(LOGIC_gen(0, 0b01, 0, immr, imms, Rn, Rd))
+#define ORRxw_mask(Rd, Rn, N, immr, imms)   EMIT(LOGIC_gen(rex.w, 0b01, rex.w?(N):0, immr, imms, Rn, Rd))
 #define EORx_mask(Rd, Rn, N, immr, imms)    EMIT(LOGIC_gen(1, 0b10, N, immr, imms, Rn, Rd))
 #define EORw_mask(Rd, Rn, immr, imms)       EMIT(LOGIC_gen(0, 0b10, 0, immr, imms, Rn, Rd))
+#define EORxw_mask(Rd, Rn, N, immr, imms)   EMIT(LOGIC_gen(rex.w, 0b10, rex.w?(N):0, immr, imms, Rn, Rd))
 #define TSTx_mask(Rn, N, immr, imms)        ANDSx_mask(xZR, Rn, N, immr, imms)
 #define TSTw_mask(Rn, immr, imms)           ANDSw_mask(wZR, Rn, immr, imms)
+#define TSTxw_mask(Rn, N, immr, imms)       ANDSxw_mask(xZR, Rn, N, immr, imms)

 #define LOGIC_REG_gen(sf, opc, shift, N, Rm, imm6, Rn, Rd)    ((sf)<<31 | (opc)<<29 | 0b01010<<24 | (shift)<<22 | (N)<<21 | (Rm)<<16 | (imm6)<<10 | (Rn)<<5 | (Rd))
 #define ANDx_REG(Rd, Rn, Rm)            EMIT(LOGIC_REG_gen(1, 0b00, 0b00, 0, Rm, 0, Rn, Rd))
 #define ANDw_REG(Rd, Rn, Rm)            EMIT(LOGIC_REG_gen(0, 0b00, 0b00, 0, Rm, 0, Rn, Rd))
+#define ANDw_REG_LSR(Rd, Rn, Rm, lsr)   EMIT(LOGIC_REG_gen(0, 0b00, 0b01, 0, Rm, lsr, Rn, Rd))
 #define ANDxw_REG(Rd, Rn, Rm)           EMIT(LOGIC_REG_gen(rex.w, 0b00, 0b00, 0, Rm, 0, Rn, Rd))
 #define ANDSx_REG(Rd, Rn, Rm)           EMIT(LOGIC_REG_gen(1, 0b11, 0b00, 0, Rm, 0, Rn, Rd))
 #define ANDSw_REG(Rd, Rn, Rm)           EMIT(LOGIC_REG_gen(0, 0b11, 0b00, 0, Rm, 0, Rn, Rd))
@@ -534,6 +564,15 @@
 #define EORx_REG_LSR(Rd, Rn, Rm, lsr)   EMIT(LOGIC_REG_gen(1, 0b10, 0b01, 0, Rm, lsr, Rn, Rd))
 #define EORw_REG_LSR(Rd, Rn, Rm, lsr)   EMIT(LOGIC_REG_gen(0, 0b10, 0b01, 0, Rm, lsr, Rn, Rd))
 #define EORxw_REG_LSR(Rd, Rn, Rm, lsr)  EMIT(LOGIC_REG_gen(rex.w, 0b10, 0b01, 0, Rm, lsr, Rn, Rd))
+#define EONx_REG(Rd, Rn, Rm)            EMIT(LOGIC_REG_gen(1, 0b10, 0b00, 1, Rm, 0, Rn, Rd))
+#define EONw_REG(Rd, Rn, Rm)            EMIT(LOGIC_REG_gen(0, 0b10, 0b00, 1, Rm, 0, Rn, Rd))
+#define EONxw_REG(Rd, Rn, Rm)           EMIT(LOGIC_REG_gen(rex.w, 0b10, 0b00, 1, Rm, 0, Rn, Rd))
+#define EONx_REG_LSL(Rd, Rn, Rm, lsl)   EMIT(LOGIC_REG_gen(1, 0b10, 0b00, 1, Rm, lsl, Rn, Rd))
+#define EONw_REG_LSL(Rd, Rn, Rm, lsl)   EMIT(LOGIC_REG_gen(0, 0b10, 0b00, 1, Rm, lsl, Rn, Rd))
+#define EONxw_REG_LSL(Rd, Rn, Rm, lsl)  EMIT(LOGIC_REG_gen(rex.w, 0b10, 0b00, 1, Rm, lsl, Rn, Rd))
+#define EONx_REG_LSR(Rd, Rn, Rm, lsr)   EMIT(LOGIC_REG_gen(1, 0b10, 0b01, 1, Rm, lsr, Rn, Rd))
+#define EONw_REG_LSR(Rd, Rn, Rm, lsr)   EMIT(LOGIC_REG_gen(0, 0b10, 0b01, 1, Rm, lsr, Rn, Rd))
+#define EONxw_REG_LSR(Rd, Rn, Rm, lsr)  EMIT(LOGIC_REG_gen(rex.w, 0b10, 0b01, 1, Rm, lsr, Rn, Rd))
 #define MOVx_REG(Rd, Rm)                ORRx_REG(Rd, xZR, Rm)
 #define MOVw_REG(Rd, Rm)                ORRw_REG(Rd, xZR, Rm)
 #define MOVxw_REG(Rd, Rm)               ORRxw_REG(Rd, xZR, Rm)
@@ -560,6 +599,12 @@
 #define BICx_REG    BICx
 #define BICw_REG    BICw
 #define BICxw_REG   BICxw
+#define BICx_REG_LSL(Rd, Rn, Rm, lsl)   EMIT(LOGIC_REG_gen(1, 0b00, 0b00, 1, Rm, lsl, Rn, Rd))
+#define BICw_REG_LSL(Rd, Rn, Rm, lsl)   EMIT(LOGIC_REG_gen(0, 0b00, 0b00, 1, Rm, lsl, Rn, Rd))
+#define BICxw_REG_LSL(Rd, Rn, Rm, lsl)  EMIT(LOGIC_REG_gen(rex.w, 0b00, 0b00, 1, Rm, lsl, Rn, Rd))
+#define BICx_REG_LSR(Rd, Rn, Rm, lsl)   EMIT(LOGIC_REG_gen(1, 0b00, 0b01, 1, Rm, lsl, Rn, Rd))
+#define BICw_REG_LSR(Rd, Rn, Rm, lsl)   EMIT(LOGIC_REG_gen(0, 0b00, 0b01, 1, Rm, lsl, Rn, Rd))
+#define BICxw_REG_LSR(Rd, Rn, Rm, lsl)  EMIT(LOGIC_REG_gen(rex.w, 0b00, 0b01, 1, Rm, lsl, Rn, Rd))
 #define TSTx_REG(Rn, Rm)                ANDSx_REG(xZR, Rn, Rm)
 #define TSTw_REG(Rn, Rm)                ANDSw_REG(wZR, Rn, Rm)
 #define TSTxw_REG(Rn, Rm)               ANDSxw_REG(xZR, Rn, Rm)
@@ -575,12 +620,13 @@
 #define BFMx(Rd, Rn, immr, imms)        EMIT(BFM_gen(1, 0b01, 1, immr, imms, Rn, Rd))
 #define BFMw(Rd, Rn, immr, imms)        EMIT(BFM_gen(0, 0b01, 0, immr, imms, Rn, Rd))
 #define BFMxw(Rd, Rn, immr, imms)       EMIT(BFM_gen(rex.w, 0b01, rex.w, immr, imms, Rn, Rd))
-#define BFIx(Rd, Rn, lsb, width)        BFMx(Rd, Rn, ((-lsb)%64)&0x3f, (width)-1)
-#define BFIw(Rd, Rn, lsb, width)        BFMw(Rd, Rn, ((-lsb)%32)&0x1f, (width)-1)
+#define BFIx(Rd, Rn, lsb, width)        BFMx(Rd, Rn, ((-(lsb))%64)&0x3f, (width)-1)
+#define BFIw(Rd, Rn, lsb, width)        BFMw(Rd, Rn, ((-(lsb))%32)&0x1f, (width)-1)
 #define BFIxw(Rd, Rn, lsb, width)       if(rex.w) {BFIx(Rd, Rn, lsb, width);} else {BFIw(Rd, Rn, lsb, width);}
-#define BFCx(Rd, lsb, width)            BFMx(Rd, xZR, ((-lsb)%64)&0x3f, (width)-1)
-#define BFCw(Rd, lsb, width)            BFMw(Rd, xZR, ((-lsb)%32)&0x1f, (width)-1)
-#define BFCxw(Rd, lsb, width)           BFMxw(Rd, xZR, rex.w?(((-lsb)%64)&0x3f):(((-lsb)%32)&0x1f), (width)-1)
+#define BFIz(Rd, Rn, lsb, width)        if(rex.is32bits) {BFIw(Rd, Rn, lsb, width);} else {BFIx(Rd, Rn, lsb, width);}
+#define BFCx(Rd, lsb, width)            BFMx(Rd, xZR, ((-(lsb))%64)&0x3f, (width)-1)
+#define BFCw(Rd, lsb, width)            BFMw(Rd, xZR, ((-(lsb))%32)&0x1f, (width)-1)
+#define BFCxw(Rd, lsb, width)           BFMxw(Rd, xZR, rex.w?(((-(lsb))%64)&0x3f):(((-(lsb))%32)&0x1f), (width)-1)
 // Insert lsb:width part of Rn into low part of Rd (leaving rest of Rd untouched)
 #define BFXILx(Rd, Rn, lsb, width)      EMIT(BFM_gen(1, 0b01, 1, (lsb), (lsb)+(width)-1, Rn, Rd))
 // Insert lsb:width part of Rn into low part of Rd (leaving rest of Rd untouched)
@@ -619,9 +665,9 @@
 #define SBFMx(Rd, Rn, immr, imms)       EMIT(SBFM_gen(1, 1, immr, imms, Rn, Rd))
 #define SBFMw(Rd, Rn, immr, imms)       EMIT(SBFM_gen(0, 0, immr, imms, Rn, Rd))
 #define SBFMxw(Rd, Rn, immr, imms)      EMIT(SBFM_gen(rex.w, rex.w, immr, imms, Rn, Rd))
-#define SBFXx(Rd, Rn, lsb, width)       SBFMx(Rd, Rn, lsb, lsb+width-1)
-#define SBFXw(Rd, Rn, lsb, width)       SBFMw(Rd, Rn, lsb, lsb+width-1)
-#define SBFXxw(Rd, Rn, lsb, width)      SBFMxw(Rd, Rn, lsb, lsb+width-1)
+#define SBFXx(Rd, Rn, lsb, width)       SBFMx(Rd, Rn, lsb, (lsb)+(width)-1)
+#define SBFXw(Rd, Rn, lsb, width)       SBFMw(Rd, Rn, lsb, (lsb)+(width)-1)
+#define SBFXxw(Rd, Rn, lsb, width)      SBFMxw(Rd, Rn, lsb, (lsb)+(width)-1)
 #define SXTBx(Rd, Rn)                   SBFMx(Rd, Rn, 0, 7)
 #define SXTBw(Rd, Rn)                   SBFMw(Rd, Rn, 0, 7)
 #define SXTHx(Rd, Rn)                   SBFMx(Rd, Rn, 0, 15)
@@ -734,6 +780,10 @@
 // mrs    x0, fpsr : 1101010100 1 1 1 011 0100 0100 001 00000    o0=1(op0=3), op1=0b011(3) CRn=0b0100(4) CRm=0b0100(4) op2=1
 #define MRS_fpsr(Rt)                    EMIT(MRS_gen(1, 1, 3, 4, 4, 1, Rt))
 #define MSR_fpsr(Rt)                    EMIT(MRS_gen(0, 1, 3, 4, 4, 1, Rt))
+// mrs   x0, cntvct_el0     op0=0b11 op1=0b011 CRn=0b1110 CRm=0b0000 op2=0b010
+#define MRS_cntvct_el0(Rt)              EMIT(MRS_gen(1, 1, 0b011, 0b1110, 0b0000, 0b010, Rt))
+// mrs   x0, cntpctss_el0     op0=0b11 op1=0b011 CRn=0b1110 CRm=0b0000 op2=0b101
+#define MRS_cntpctss_el0(Rt)            EMIT(MRS_gen(1, 1, 0b011, 0b1110, 0b0000, 0b101, Rt))
 // NEON Saturation Bit
 #define FPSR_QC 27
 // NEON Input Denormal Cumulative
@@ -765,13 +815,13 @@
 // imm16 must be 4-aligned
 #define VLDR128_U12(Qt, Rn, imm16)          EMIT(VMEM_gen(0b00, 0b11, ((uint32_t)((imm16)>>4))&0xfff, Rn, Qt))
 // (imm14) must be 3-aligned
-#define VSTR32_U12(Dt, Rn, imm14)           EMIT(VMEM_gen(0b10, 0b00, ((uint32_t)(imm14>>2))&0xfff, Rn, Dt))
+#define VSTR32_U12(Dt, Rn, imm14)           EMIT(VMEM_gen(0b10, 0b00, ((uint32_t)((imm14)>>2))&0xfff, Rn, Dt))
 // (imm15) must be 3-aligned
-#define VSTR64_U12(Dt, Rn, imm15)           EMIT(VMEM_gen(0b11, 0b00, ((uint32_t)(imm15>>3))&0xfff, Rn, Dt))
+#define VSTR64_U12(Dt, Rn, imm15)           EMIT(VMEM_gen(0b11, 0b00, ((uint32_t)((imm15)>>3))&0xfff, Rn, Dt))
 // imm16 must be 4-aligned
 #define VSTR128_U12(Qt, Rn, imm16)          EMIT(VMEM_gen(0b00, 0b10, ((uint32_t)((imm16)>>4))&0xfff, Rn, Qt))
 // (imm13) must be 1-aligned
-#define VSTR16_U12(Ht, Rn, imm13)           EMIT(VMEM_gen(0b01, 0b00, ((uint32_t)(imm13>>1))&0xfff, Rn, Ht))
+#define VSTR16_U12(Ht, Rn, imm13)           EMIT(VMEM_gen(0b01, 0b00, ((uint32_t)((imm13)>>1))&0xfff, Rn, Ht))

 #define VMEMUR_vector(size, opc, imm9, Rn, Rt)  ((size)<<30 | 0b111<<27 | 1<<26 | (opc)<<22 | (imm9)<<12 | (Rn)<<5 | (Rt))
 // signed offset, no alignement!
@@ -933,6 +983,15 @@
 #define VSRI_16(Vd, Vn, shift)              EMIT(VSRI_vector(0, 0b0010 | ((shift)>>3)&1, (shift)&7, Vn, Vd))
 #define VSRI_32(Vd, Vn, shift)              EMIT(VSRI_vector(0, 0b0100 | (((shift)>>3)&3), (shift)&7, Vn, Vd))

+// Reverse elements in 64-bit doublewords (vector). This instruction reverses the order of 8-bit, 16-bit, or 32-bit elements in each doubleword
+#define VREVx_vector(Q, U, size, o0, Rn, Rd)    ((Q)<<30 | (U)<<29 | 0b01110<<24 | (size)<<22 | 0b10000<<17 | (o0)<<12 | 0b10<<10 | (Rn)<<5 | (Rd))
+#define VREV64_32(Vd, Vn)                   EMIT(VREVx_vector(0, 0, 0b10, 0, Vn, Vd))
+#define VREV64_16(Vd, Vn)                   EMIT(VREVx_vector(0, 0, 0b01, 0, Vn, Vd))
+#define VREV64_8(Vd, Vn)                    EMIT(VREVx_vector(0, 0, 0b00, 0, Vn, Vd))
+#define VREV64Q_32(Vd, Vn)                  EMIT(VREVx_vector(1, 0, 0b10, 0, Vn, Vd))
+#define VREV64Q_16(Vd, Vn)                  EMIT(VREVx_vector(1, 0, 0b01, 0, Vn, Vd))
+#define VREV64Q_8(Vd, Vn)                   EMIT(VREVx_vector(1, 0, 0b00, 0, Vn, Vd))
+
 // Integer MATH
 #define ADDSUB_vector(Q, U, size, Rm, Rn, Rd)   ((Q)<<30 | (U)<<29 | 0b01110<<24 | (size)<<22 | 1<<21 | (Rm)<<16 | 0b10000<<11 | 1<<10 | (Rn)<<5 | (Rd))
 #define VADDQ_8(Vd, Vn, Vm)                 EMIT(ADDSUB_vector(1, 0, 0b00, Vm, Vn, Vd))
@@ -950,6 +1009,10 @@
 #define VSUB_16(Vd, Vn, Vm)                 EMIT(ADDSUB_vector(0, 1, 0b01, Vm, Vn, Vd))
 #define VSUB_32(Vd, Vn, Vm)                 EMIT(ADDSUB_vector(0, 1, 0b10, Vm, Vn, Vd))

+#define ADDSUB_scalar(U, size, Rm, Rn, Rd)  (01<<30 | (U)<<29 | 0b11110<<24 | (size)<<22 | 1<<21 | (Rm)<<16 | 0b10000<<11 | 1<<10 | (Rn)<<5 | (Rd))
+#define ADD_64(Vd, Vn, Vm)                  EMIT(ADDSUB_scalar(0, 0b11, Vm, Vn, Vd))
+#define SUB_64(Vd, Vn, Vm)                  EMIT(ADDSUB_scalar(1, 0b11, Vm, Vn, Vd))
+
 #define NEGABS_vector(Q, U, size, Rn, Rd)   ((Q)<<30 | (U)<<29 | 0b01110<<24 | (size)<<22 | 0b10000<<17 | 0b01011<<12 | 0b10<<10 | (Rn)<<5 | (Rd))
 #define NEG_8(Vd, Vn)                       EMIT(NEGABS_vector(0, 1, 0b00, Vn, Vd))
 #define NEG_16(Vd, Vn)                      EMIT(NEGABS_vector(0, 1, 0b01, Vn, Vd))
@@ -1072,6 +1135,12 @@
 #define FADDP_vector(Q, sz, Rm, Rn, Rd) ((Q)<<30 | 1<<29 | 0b01110<<24 | (sz)<<22 | 1<<21 | (Rm)<<16 | 0b11010<<11 | 1<<10 | (Rn)<<5 | (Rd))
 #define VFADDPQS(Vd, Vn, Vm)        EMIT(FADDP_vector(1, 0, Vm, Vn, Vd))
 #define VFADDPQD(Vd, Vn, Vm)        EMIT(FADDP_vector(1, 1, Vm, Vn, Vd))
+#define VFADDPS(Vd, Vn, Vm)         EMIT(FADDP_vector(0, 0, Vm, Vn, Vd))
+#define VFADDPD(Vd, Vn, Vm)         EMIT(FADDP_vector(0, 1, Vm, Vn, Vd))
+
+#define FADDP_scalar(sz, Rn, Rd)        (0b01<<30 | 1<<29 | 0b11110<<24 | (sz)<<22 | 0b11000<<17 | 0b01101<<12 | 0b10<<10 | (Rn)<<5 | (Rd))
+#define FADDPS(Sd, Sn)              EMIT(FADDP_scalar(0, Sn, Sd))
+#define FADDPD(Dd, Dn)              EMIT(FADDP_scalar(1, Dn, Dd))

 // NEG / ABS
 #define FNEGABS_scalar(type, opc, Rn, Rd)  (0b11110<<24 | (type)<<22 | 1<<21 | (opc)<<15 | 0b10000<<10 | (Rn)<<5 | (Rd))
@@ -1141,8 +1210,8 @@
 #define FCMP_scalar(type, Rn, Rm, opc)  (0b11110<<24 | (type)<<22 | 1<<21 | (Rm)<<16 | 0b1000<<10 | (Rn)<<5 | (opc)<<3)
 #define FCMPS(Sn, Sm)               EMIT(FCMP_scalar(0b00, Sn, Sm, 0b00))
 #define FCMPD(Dn, Dm)               EMIT(FCMP_scalar(0b01, Dn, Dm, 0b00))
-#define FCMPS_0(Sn)                 EMIT(FCMP_scalar(0b00, 0, Sn, 0b01))
-#define FCMPD_0(Dn)                 EMIT(FCMP_scalar(0b01, 0, Dn, 0b01))
+#define FCMPS_0(Sn)                 EMIT(FCMP_scalar(0b00, Sn, 0, 0b01))
+#define FCMPD_0(Dn)                 EMIT(FCMP_scalar(0b01, Dn, 0, 0b01))

 // CVT
 #define FCVT_scalar(sf, type, rmode, opcode, Rn, Rd)    ((sf)<<31 | 0b11110<<24 | (type)<<22 | 1<<21 | (rmode)<<19 | (opcode)<<16 | (Rn)<<5 | (Rd))
@@ -1338,22 +1407,12 @@
 #define VFRINTRDQ(Vd,Vn, mode)      EMIT(FRINT_vector(1, 0, (mode)&1, 1, ((mode)>>1)&1, Vn, Vd))
 // round with mode, mode is 0 = TieEven, 1=+inf, 2=-inf, 3=zero
 #define VFRINTRSQ(Vd,Vn, mode)      EMIT(FRINT_vector(1, 0, (mode)&1, 0, ((mode)>>1)&1, Vn, Vd))
+#define VFRINTRS(Vd, Vn, mode)      EMIT(FRINT_vector(0, 0, (mode)&1, 0, ((mode)>>1)&1, Vn, Vd))

 #define FRINTI_scalar(type, Rn, Rd)  (0b11110<<24 | (type)<<22 | 1<<21 | 0b001<<18 | 0b111<<15 | 0b10000<<10 | (Rn)<<5 | (Rd))
 #define FRINTIS(Sd, Sn)             EMIT(FRINTI_scalar(0b00, Sn, Sd))
 #define FRINTID(Dd, Dn)             EMIT(FRINTI_scalar(0b01, Dn, Dd))

-
-#define FRINTxx_scalar(type, op, Rn, Rd)  (0b11110<<24 | (type)<<22 | 1<<21 | 0b0100<<17 | (op)<<15 | 0b10000<<10 | (Rn)<<5 | (Rd))
-#define FRINT32ZS(Sd, Sn)           EMIT(FRINTxx_scalar(0b00, 0b00, Sn, Sd))
-#define FRINT32ZD(Dd, Dn)           EMIT(FRINTxx_scalar(0b01, 0b00, Dn, Dd))
-#define FRINT32XS(Sd, Sn)           EMIT(FRINTxx_scalar(0b00, 0b01, Sn, Sd))
-#define FRINT32XD(Dd, Dn)           EMIT(FRINTxx_scalar(0b01, 0b01, Dn, Dd))
-#define FRINT64ZS(Sd, Sn)           EMIT(FRINTxx_scalar(0b00, 0b10, Sn, Sd))
-#define FRINT64ZD(Dd, Dn)           EMIT(FRINTxx_scalar(0b01, 0b10, Dn, Dd))
-#define FRINT64XS(Sd, Sn)           EMIT(FRINTxx_scalar(0b00, 0b11, Sn, Sd))
-#define FRINT64XD(Dd, Dn)           EMIT(FRINTxx_scalar(0b01, 0b11, Dn, Dd))
-
 #define FRINT_scalar(type, rmode, Rn, Rd)   (0b11110<<24 | (type)<<22 | 1<<21 | 0b001<<18 | (rmode)<<15 | 0b10000<<10 | (Rn)<<5 | (Rd))
 // round toward 0 (truncate)
 #define FRINTZS(Sd, Sn)             EMIT(FRINT_scalar(0b00, 0b011, Sn, Sd))
@@ -1424,14 +1483,23 @@
 #define VUZP1Q_64(Rt, Rn, Rm)       EMIT(UZP_gen(1, 0b11, Rm, 0, Rn, Rt))
 #define VUZP2Q_64(Rt, Rn, Rm)       EMIT(UZP_gen(1, 0b11, Rm, 1, Rn, Rt))

-#define DUP_gen(Q, imm5, Rn, Rd)    ((Q)<<30 | 0b01110000<<21 | (imm5)<<16 | 1<<10 | (Rn)<<5 | (Rd))
-#define VDUP_8(Vd, Vn, idx)         EMIT(DUP_gen(0, ((idx)<<1|1), Vn, Vd))
-#define VDUPQ_8(Vd, Vn, idx)        EMIT(DUP_gen(1, ((idx)<<1|1), Vn, Vd))
-#define VDUP_16(Vd, Vn, idx)        EMIT(DUP_gen(0, ((idx)<<2|0b10), Vn, Vd))
-#define VDUPQ_16(Vd, Vn, idx)       EMIT(DUP_gen(1, ((idx)<<2|0b10), Vn, Vd))
-#define VDUP_32(Vd, Vn, idx)        EMIT(DUP_gen(0, ((idx)<<3|0b100), Vn, Vd))
-#define VDUPQ_32(Vd, Vn, idx)       EMIT(DUP_gen(1, ((idx)<<3|0b100), Vn, Vd))
-#define VDUPQ_64(Vd, Vn, idx)       EMIT(DUP_gen(1, ((idx)<<4|0b1000), Vn, Vd))
+#define DUP_element(Q, imm5, Rn, Rd)    ((Q)<<30 | 0b01110000<<21 | (imm5)<<16 | 1<<10 | (Rn)<<5 | (Rd))
+#define VDUP_8(Vd, Vn, idx)         EMIT(DUP_element(0, ((idx)<<1|1), Vn, Vd))
+#define VDUPQ_8(Vd, Vn, idx)        EMIT(DUP_element(1, ((idx)<<1|1), Vn, Vd))
+#define VDUP_16(Vd, Vn, idx)        EMIT(DUP_element(0, ((idx)<<2|0b10), Vn, Vd))
+#define VDUPQ_16(Vd, Vn, idx)       EMIT(DUP_element(1, ((idx)<<2|0b10), Vn, Vd))
+#define VDUP_32(Vd, Vn, idx)        EMIT(DUP_element(0, ((idx)<<3|0b100), Vn, Vd))
+#define VDUPQ_32(Vd, Vn, idx)       EMIT(DUP_element(1, ((idx)<<3|0b100), Vn, Vd))
+#define VDUPQ_64(Vd, Vn, idx)       EMIT(DUP_element(1, ((idx)<<4|0b1000), Vn, Vd))
+
+#define DUP_general(Q, imm5, Rn, Rd)    ((Q)<<30 | 0b01110000<<21 | (imm5)<<16 | 0b11<<10 | (Rn)<<5 | (Rd))
+#define VDUPB(Vd, Wn)         EMIT(DUP_general(0, 0b1, Wn, Vd))
+#define VDUPQB(Vd, Wn)        EMIT(DUP_general(1, 0b1, Wn, Vd))
+#define VDUPH(Vd, Wn)        EMIT(DUP_general(0, 0b10, Wn, Vd))
+#define VDUPQH(Vd, Wn)       EMIT(DUP_general(1, 0b10, Wn, Vd))
+#define VDUPS(Vd, Wn)        EMIT(DUP_general(0, 0b100, Wn, Vd))
+#define VDUPQS(Vd, Wn)       EMIT(DUP_general(1, 0b100, Wn, Vd))
+#define VDUPQD(Vd, Xn)       EMIT(DUP_general(1, 0b1000, Xn, Vd))

 // TBL
 #define TBL_gen(Q, Rm, len, op, Rn, Rd) ((Q)<<30 | 0b001110<<24 | (Rm)<<16 | (len)<<13 | (op)<<12 | (Rn)<<5 | (Rd))
@@ -1459,7 +1527,6 @@
 #define VTRNQ1_32(Vd, Vn, Vm)       EMIT(TRN_gen(1, 0b10, Vm, 0, Vn, Vd))
 #define VTRNQ1_16(Vd, Vn, Vm)       EMIT(TRN_gen(1, 0b01, Vm, 0, Vn, Vd))
 #define VTRNQ1_8(Vd, Vn, Vm)        EMIT(TRN_gen(1, 0b00, Vm, 0, Vn, Vd))
-#define VSWP(Vd, Vn)                VTRNQ1_64(Vd, Vn, Vn)
 #define VTRNQ2_64(Vd, Vn, Vm)       EMIT(TRN_gen(1, 0b11, Vm, 1, Vn, Vd))
 #define VTRNQ2_32(Vd, Vn, Vm)       EMIT(TRN_gen(1, 0b10, Vm, 1, Vn, Vd))
 #define VTRNQ2_16(Vd, Vn, Vm)       EMIT(TRN_gen(1, 0b01, Vm, 1, Vn, Vd))
@@ -1514,6 +1581,15 @@
 // Signed saturating extract Unsigned Narrow, takes Rn element and reduce 16->8 with Unsigned saturation and fit higher part of Rd
 #define SQXTUN2_8(Rd, Rn)           EMIT(QXTUN_vector(1, 1, 0b00, Rn, Rd))

+#define XTN_vector(Q, size, Rn, Rd) ((Q)<<30 | 0b01110<<24 | (size)<<22 | 0b10000<<17 | 0b10010<<12 | 0b10<<10 | (Rn)<<5 | (Rd))
+// Xtract narrow to X bits
+#define XTN_8(Vd, Vn)               EMIT(XTN_vector(0, 0b00, Vn, Vd))
+#define XTN_16(Vd, Vn)              EMIT(XTN_vector(0, 0b01, Vn, Vd))
+#define XTN_32(Vd, Vn)              EMIT(XTN_vector(0, 0b10, Vn, Vd))
+#define XTN2_8(Vd, Vn)              EMIT(XTN_vector(1, 0b00, Vn, Vd))
+#define XTN2_16(Vd, Vn)             EMIT(XTN_vector(1, 0b01, Vn, Vd))
+#define XTN2_32(Vd, Vn)             EMIT(XTN_vector(1, 0b10, Vn, Vd))
+
 // Integer CMP
 // EQual
 #define CMEQ_vector(Q, U, size, Rm, Rn, Rd)     ((Q)<<30 | (U)<<29 | 0b01110<<24 | (size)<<22 | 1<<21 | (Rm)<<16 | 0b10001<<11 | 1<<10 | (Rn)<<5 | (Rd))
@@ -1568,6 +1644,15 @@
 #define CMEQQ_0_16(Rd, Rn)          EMIT(CMEQ_0_vector(1, 0b01, Rn, Rd))
 #define CMEQQ_0_32(Rd, Rn)          EMIT(CMEQ_0_vector(1, 0b10, Rn, Rd))
 #define CMEQQ_0_64(Rd, Rn)          EMIT(CMEQ_0_vector(1, 0b11, Rn, Rd))
+// Greater Than 0
+#define CMCond_0_vector(Q, U, size, op, Rn, Rd) ((Q)<<30 | (U)<<29 | 0b01110<<24 | (size)<<22 | 0b10000<<17 | 0b0100<<13 | (op)<<12 | 0b10<<10 | (Rn)<<5 | (Rd))
+#define CMGT_0_8(Rd, Rn)            EMIT(CMCond_0_vector(0, 0, 0b00, 0, Rn, Rd))
+#define CMGT_0_16(Rd, Rn)           EMIT(CMCond_0_vector(0, 0, 0b01, 0, Rn, Rd))
+#define CMGT_0_32(Rd, Rn)           EMIT(CMCond_0_vector(0, 0, 0b10, 0, Rn, Rd))
+#define CMGTQ_0_8(Rd, Rn)           EMIT(CMCond_0_vector(1, 0, 0b00, 0, Rn, Rd))
+#define CMGTQ_0_16(Rd, Rn)          EMIT(CMCond_0_vector(1, 0, 0b01, 0, Rn, Rd))
+#define CMGTQ_0_32(Rd, Rn)          EMIT(CMCond_0_vector(1, 0, 0b10, 0, Rn, Rd))
+#define CMGTQ_0_64(Rd, Rn)          EMIT(CMCond_0_vector(1, 0, 0b11, 0, Rn, Rd))

 // Vector Float CMP
 // EQual
@@ -1726,9 +1811,11 @@
 // MOV Immediate
 #define MOVI_vector(Q, op, abc, cmode, defgh, Rd)   ((Q)<<30 | (op)<<29 | 0b0111100000<<19 | (abc)<<16 | (cmode)<<12 | 1<<10 | (defgh)<<5 | (Rd))
 #define MOVIQ_8(Rd, imm8)           EMIT(MOVI_vector(1, 0, (((imm8)>>5)&0b111), 0b1110, ((imm8)&0b11111), Rd))
+#define MOVIQ_16(Rd, imm8, lsl8)    EMIT(MOVI_vector(1, 0, (((imm8)>>5)&0b111), 0b1000|((lsl8)?0b10:0), ((imm8)&0b11111), Rd))
 #define MOVI_8(Rd, imm8)            EMIT(MOVI_vector(0, 0, (((imm8)>>5)&0b111), 0b1110, ((imm8)&0b11111), Rd))
-#define MOVI_16(Rd, imm8)           EMIT(MOVI_vector(0, 0, (((imm8)>>5)&0b111), 0b1000, ((imm8)&0b11111), Rd))
+#define MOVI_16(Rd, imm8, lsl8)     EMIT(MOVI_vector(0, 0, (((imm8)>>5)&0b111), 0b1000|((lsl8)?0b10:0), ((imm8)&0b11111), Rd))
 #define MOVI_32(Rd, imm8)           EMIT(MOVI_vector(0, 0, (((imm8)>>5)&0b111), 0b0000, ((imm8)&0b11111), Rd))
+#define MOVI_64(Rd, imm8)           EMIT(MOVI_vector(0, 1, (((imm8)>>5)&0b111), 0b1110, ((imm8)&0b11111), Rd))

 // SHLL and eXtend Long
 #define SHLL_vector(Q, U, immh, immb, Rn, Rd)  ((Q)<<30 | (U)<<29 | 0b011110<<23 | (immh)<<19 | (immb)<<16 | 0b10100<<11 | 1<<10 | (Rn)<<5 | (Rd))
@@ -1888,4 +1975,290 @@
 #define PMULL_128(Rd, Rn, Rm)   EMIT(PMULL_gen(0, 0b11, Rm, Rn, Rd))
 #define PMULL2_128(Rd, Rn, Rm)  EMIT(PMULL_gen(1, 0b11, Rm, Rn, Rd))

+// ATOMIC extension
+#define ATOMIC_gen(size, A, R, Rs, opc, Rn, Rt) ((size)<<30 | 0b111<<27 | (A)<<23 | (R)<<22 | 1<<21 | (Rs)<<16 | (opc)<<12 | (Rn)<<5 | (Rt))
+// Atomic ADD
+#define LDADDxw(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b000, Rn, Rt))
+#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 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))
+#define LDADDAB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 1, 0, Rs, 0b000, Rn, Rt))
+#define LDADDALB(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b00, 1, 1, Rs, 0b000, Rn, Rt))
+#define LDADDLB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b000, Rn, Rt))
+#define STADDB(Rs, Rn)                  EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b000, Rn, 0b11111))
+#define STADDLB(Rs, Rn)                 EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b000, Rn, 0b11111))
+#define LDADDH(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b000, Rn, Rt))
+#define LDADDAH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 1, 0, Rs, 0b000, Rn, Rt))
+#define LDADDALH(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b01, 1, 1, Rs, 0b000, Rn, Rt))
+#define LDADDLH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b000, Rn, Rt))
+#define STADDH(Rs, Rn)                  EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b000, Rn, 0b11111))
+#define STADDLH(Rs, Rn)                 EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b000, Rn, 0b11111))
+// Atomic AND with complement (i.e. BIC)
+#define LDCLRxw(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b001, Rn, Rt))
+#define LDCLRAxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 1, 0, Rs, 0b001, Rn, Rt))
+#define LDCLRALxw(Rs, Rt, Rn)           EMIT(ATOMIC_gen(0b10+rex.w, 1, 1, Rs, 0b001, Rn, Rt))
+#define LDCLRLxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b001, Rn, Rt))
+#define STCLRxw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b001, Rn, 0b11111))
+#define STCLRLxw(Rs, Rn)                EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b001, Rn, 0b11111))
+#define LDCLRB(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b001, Rn, Rt))
+#define LDCLRAB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 1, 0, Rs, 0b001, Rn, Rt))
+#define LDCLRALB(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b00, 1, 1, Rs, 0b001, Rn, Rt))
+#define LDCLRLB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b001, Rn, Rt))
+#define STCLRB(Rs, Rn)                  EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b001, Rn, 0b11111))
+#define STCLRLB(Rs, Rn)                 EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b001, Rn, 0b11111))
+#define LDCLRH(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b001, Rn, Rt))
+#define LDCLRAH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 1, 0, Rs, 0b001, Rn, Rt))
+#define LDCLRALH(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b01, 1, 1, Rs, 0b001, Rn, Rt))
+#define LDCLRLH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b001, Rn, Rt))
+#define STCLRH(Rs, Rn)                  EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b001, Rn, 0b11111))
+#define STCLRLH(Rs, Rn)                 EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b001, Rn, 0b11111))
+// Atomic EOR
+#define LDEORxw(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b010, Rn, Rt))
+#define LDEORAxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 1, 0, Rs, 0b010, Rn, Rt))
+#define LDEORALxw(Rs, Rt, Rn)           EMIT(ATOMIC_gen(0b10+rex.w, 1, 1, Rs, 0b010, Rn, Rt))
+#define LDEORLxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b010, Rn, Rt))
+#define STEORxw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b010, Rn, 0b11111))
+#define STEORLxw(Rs, Rn)                EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b010, Rn, 0b11111))
+#define LDEORB(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b010, Rn, Rt))
+#define LDEORAB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 1, 0, Rs, 0b010, Rn, Rt))
+#define LDEORALB(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b00, 1, 1, Rs, 0b010, Rn, Rt))
+#define LDEORLB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b010, Rn, Rt))
+#define STEORB(Rs, Rn)                  EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b010, Rn, 0b11111))
+#define STEORLB(Rs, Rn)                 EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b010, Rn, 0b11111))
+#define LDEORH(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b010, Rn, Rt))
+#define LDEORAH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 1, 0, Rs, 0b010, Rn, Rt))
+#define LDEORALH(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b01, 1, 1, Rs, 0b010, Rn, Rt))
+#define LDEORLH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b010, Rn, Rt))
+#define STEORH(Rs, Rn)                  EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b010, Rn, 0b11111))
+#define STEORLH(Rs, Rn)                 EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b010, Rn, 0b11111))
+// Atomic OR
+#define LDSETxw(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b011, Rn, Rt))
+#define LDSETAxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 1, 0, Rs, 0b011, Rn, Rt))
+#define LDSETALxw(Rs, Rt, Rn)           EMIT(ATOMIC_gen(0b10+rex.w, 1, 1, Rs, 0b011, Rn, Rt))
+#define LDSETLxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b011, Rn, Rt))
+#define STSETxw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b011, Rn, 0b11111))
+#define STSETLxw(Rs, Rn)                EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b011, Rn, 0b11111))
+#define LDSETB(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b011, Rn, Rt))
+#define LDSETAB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 1, 0, Rs, 0b011, Rn, Rt))
+#define LDSETALB(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b00, 1, 1, Rs, 0b011, Rn, Rt))
+#define LDSETLB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b011, Rn, Rt))
+#define STSETB(Rs, Rn)                  EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b011, Rn, 0b11111))
+#define STSETLB(Rs, Rn)                 EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b011, Rn, 0b11111))
+#define LDSETH(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b011, Rn, Rt))
+#define LDSETAH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 1, 0, Rs, 0b011, Rn, Rt))
+#define LDSETALH(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b01, 1, 1, Rs, 0b011, Rn, Rt))
+#define LDSETLH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b011, Rn, Rt))
+#define STSETH(Rs, Rn)                  EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b011, Rn, 0b11111))
+#define STSETLH(Rs, Rn)                 EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b011, Rn, 0b11111))
+// Atomic Signel Max
+#define LDSMAXxw(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b100, Rn, Rt))
+#define LDSMAXAxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 1, 0, Rs, 0b100, Rn, Rt))
+#define LDSMAXALxw(Rs, Rt, Rn)           EMIT(ATOMIC_gen(0b10+rex.w, 1, 1, Rs, 0b100, Rn, Rt))
+#define LDSMAXLxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b100, Rn, Rt))
+#define STSMAXxw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b100, Rn, 0b11111))
+#define STSMAXLxw(Rs, Rn)                EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b100, Rn, 0b11111))
+#define LDSMAXB(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b100, Rn, Rt))
+#define LDSMAXAB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 1, 0, Rs, 0b100, Rn, Rt))
+#define LDSMAXALB(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b00, 1, 1, Rs, 0b100, Rn, Rt))
+#define LDSMAXLB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b100, Rn, Rt))
+#define STSMAXB(Rs, Rn)                  EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b100, Rn, 0b11111))
+#define STSMAXLB(Rs, Rn)                 EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b100, Rn, 0b11111))
+#define LDSMAXH(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b100, Rn, Rt))
+#define LDSMAXAH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 1, 0, Rs, 0b100, Rn, Rt))
+#define LDSMAXALH(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b01, 1, 1, Rs, 0b100, Rn, Rt))
+#define LDSMAXLH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b100, Rn, Rt))
+#define STSMAXH(Rs, Rn)                  EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b100, Rn, 0b11111))
+#define STSMAXLH(Rs, Rn)                 EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b100, Rn, 0b11111))
+// Atomic Signed Min
+#define LDSMINxw(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b101, Rn, Rt))
+#define LDSMINAxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 1, 0, Rs, 0b101, Rn, Rt))
+#define LDSMINALxw(Rs, Rt, Rn)           EMIT(ATOMIC_gen(0b10+rex.w, 1, 1, Rs, 0b101, Rn, Rt))
+#define LDSMINLxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b101, Rn, Rt))
+#define STSMINxw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b101, Rn, 0b11111))
+#define STSMINLxw(Rs, Rn)                EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b101, Rn, 0b11111))
+#define LDSMINB(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b101, Rn, Rt))
+#define LDSMINAB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 1, 0, Rs, 0b101, Rn, Rt))
+#define LDSMINALB(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b00, 1, 1, Rs, 0b101, Rn, Rt))
+#define LDSMINLB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b101, Rn, Rt))
+#define STSMINB(Rs, Rn)                  EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b101, Rn, 0b11111))
+#define STSMINLB(Rs, Rn)                 EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b101, Rn, 0b11111))
+#define LDSMINH(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b101, Rn, Rt))
+#define LDSMINAH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 1, 0, Rs, 0b101, Rn, Rt))
+#define LDSMINALH(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b01, 1, 1, Rs, 0b101, Rn, Rt))
+#define LDSMINLH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b101, Rn, Rt))
+#define STSMINH(Rs, Rn)                  EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b101, Rn, 0b11111))
+#define STSMINLH(Rs, Rn)                 EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b101, Rn, 0b11111))
+// Atomic Unsigned Max
+#define LDUMAXxw(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b110, Rn, Rt))
+#define LDUMAXAxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 1, 0, Rs, 0b110, Rn, Rt))
+#define LDUMAXALxw(Rs, Rt, Rn)           EMIT(ATOMIC_gen(0b10+rex.w, 1, 1, Rs, 0b110, Rn, Rt))
+#define LDUMAXLxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b110, Rn, Rt))
+#define STUMAXxw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b110, Rn, 0b11111))
+#define STUMAXLxw(Rs, Rn)                EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b110, Rn, 0b11111))
+#define LDUMAXB(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b110, Rn, Rt))
+#define LDUMAXAB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 1, 0, Rs, 0b110, Rn, Rt))
+#define LDUMAXALB(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b00, 1, 1, Rs, 0b110, Rn, Rt))
+#define LDUMAXLB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b110, Rn, Rt))
+#define STUMAXB(Rs, Rn)                  EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b110, Rn, 0b11111))
+#define STUMAXLB(Rs, Rn)                 EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b110, Rn, 0b11111))
+#define LDUMAXH(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b110, Rn, Rt))
+#define LDUMAXAH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 1, 0, Rs, 0b110, Rn, Rt))
+#define LDUMAXALH(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b01, 1, 1, Rs, 0b110, Rn, Rt))
+#define LDUMAXLH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b110, Rn, Rt))
+#define STUMAXH(Rs, Rn)                  EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b110, Rn, 0b11111))
+#define STUMAXLH(Rs, Rn)                 EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b110, Rn, 0b11111))
+// Atomic Unsigned Min
+#define LDUMINxw(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b111, Rn, Rt))
+#define LDUMINAxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 1, 0, Rs, 0b111, Rn, Rt))
+#define LDUMINALxw(Rs, Rt, Rn)           EMIT(ATOMIC_gen(0b10+rex.w, 1, 1, Rs, 0b111, Rn, Rt))
+#define LDUMINLxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b111, Rn, Rt))
+#define STUMINxw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b111, Rn, 0b11111))
+#define STUMINLxw(Rs, Rn)                EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b111, Rn, 0b11111))
+#define LDUMINB(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b111, Rn, Rt))
+#define LDUMINAB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 1, 0, Rs, 0b111, Rn, Rt))
+#define LDUMINALB(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b00, 1, 1, Rs, 0b111, Rn, Rt))
+#define LDUMINLB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b111, Rn, Rt))
+#define STUMINB(Rs, Rn)                  EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b111, Rn, 0b11111))
+#define STUMINLB(Rs, Rn)                 EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b111, Rn, 0b11111))
+#define LDUMINH(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b111, Rn, Rt))
+#define LDUMINAH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 1, 0, Rs, 0b111, Rn, Rt))
+#define LDUMINALH(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b01, 1, 1, Rs, 0b111, Rn, Rt))
+#define LDUMINLH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b111, Rn, Rt))
+#define STUMINH(Rs, Rn)                  EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b111, Rn, 0b11111))
+#define STUMINLH(Rs, Rn)                 EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b111, Rn, 0b11111))
+
+// SWAPxx(Xs, Xt, Xn) : [Xn] => Xt, Xs => [Xn]
+#define SWAP_gen(size, A, R, Rs, Rn, Rt)    ((size)<<30 | 0b111<<27 | (A)<<23 | (R)<<22 | 1<<21 | (Rs)<<16 | 1<<15 | (Rn)<<5 | (Rt))
+#define SWPxw(Rs, Rt, Rn)               EMIT(SWAP_gen(0b10+rex.w, 0, 0, Rs, Rn, Rt))
+#define SWPAxw(Rs, Rt, Rn)              EMIT(SWAP_gen(0b10+rex.w, 1, 0, Rs, Rn, Rt))
+#define SWPALxw(Rs, Rt, Rn)             EMIT(SWAP_gen(0b10+rex.w, 1, 1, Rs, Rn, Rt))
+#define SWPLxw(Rs, Rt, Rn)              EMIT(SWAP_gen(0b10+rex.w, 0, 1, Rs, Rn, Rt))
+#define SWPB(Rs, Rt, Rn)                EMIT(SWAP_gen(0b00, 0, 0, Rs, Rn, Rt))
+#define SWPAB(Rs, Rt, Rn)               EMIT(SWAP_gen(0b00, 1, 0, Rs, Rn, Rt))
+#define SWPALB(Rs, Rt, Rn)              EMIT(SWAP_gen(0b00, 1, 1, Rs, Rn, Rt))
+#define SWPLB(Rs, Rt, Rn)               EMIT(SWAP_gen(0b00, 0, 1, Rs, Rn, Rt))
+#define SWPH(Rs, Rt, Rn)                EMIT(SWAP_gen(0b01, 0, 0, Rs, Rn, Rt))
+#define SWPAH(Rs, Rt, Rn)               EMIT(SWAP_gen(0b01, 1, 0, Rs, Rn, Rt))
+#define SWPALH(Rs, Rt, Rn)              EMIT(SWAP_gen(0b01, 1, 1, Rs, Rn, Rt))
+#define SWPLH(Rs, Rt, Rn)               EMIT(SWAP_gen(0b01, 0, 1, Rs, Rn, Rt))
+
+#define CAS_gen(size, L, Rs, O0, Rn, Rt)    ((size)<<30 | 0b001000<<24 | 1<<23 | (L)<<22 | 1<<21 | (Rs)<<16 | (O0)<<15 | 0b11111<<10 | (Rn)<<5 | (Rt))
+// Compare and Swap compare Rs with [Rn], write Rt is same, return old [Rn] in Rs
+#define CASxw(Rs, Rt, Rn)               EMIT(CAS_gen(0b10+rex.w, 0, Rs, 0, Rn, Rt))
+#define CASAxw(Rs, Rt, Rn)              EMIT(CAS_gen(0b10+rex.w, 1, Rs, 0, Rn, Rt))
+#define CASALxw(Rs, Rt, Rn)             EMIT(CAS_gen(0b10+rex.w, 1, Rs, 1, Rn, Rt))
+#define CASLxw(Rs, Rt, Rn)              EMIT(CAS_gen(0b10+rex.w, 0, Rs, 1, Rn, Rt))
+#define CASB(Rs, Rt, Rn)                EMIT(CAS_gen(0b00, 0, Rs, 0, Rn, Rt))
+#define CASAB(Rs, Rt, Rn)               EMIT(CAS_gen(0b00, 1, Rs, 0, Rn, Rt))
+#define CASALB(Rs, Rt, Rn)              EMIT(CAS_gen(0b00, 1, Rs, 1, Rn, Rt))
+#define CASLB(Rs, Rt, Rn)               EMIT(CAS_gen(0b00, 0, Rs, 1, Rn, Rt))
+#define CASH(Rs, Rt, Rn)                EMIT(CAS_gen(0b01, 0, Rs, 0, Rn, Rt))
+#define CASAH(Rs, Rt, Rn)               EMIT(CAS_gen(0b01, 1, Rs, 0, Rn, Rt))
+#define CASALH(Rs, Rt, Rn)              EMIT(CAS_gen(0b01, 1, Rs, 1, Rn, Rt))
+#define CASLH(Rs, Rt, Rn)               EMIT(CAS_gen(0b01, 0, Rs, 1, Rn, Rt))
+
+#define CASP_gen(size, L, Rs, O0, Rn, Rt)   ((size)<<30 | 0b001000<<24 | 0<<23 | (L)<<22 | 1<<21 | (Rs)<<16 | (O0)<<15 | 0b11111<<10 | (Rn)<<5 | (Rt))
+// Compare and Swap with pair, Rs, Rs+1 and Rt,Rt+1 with [Rt]
+#define CASPxw(Rs, Rt, Rn)              EMIT(CASP_gen(0b00+rex.w, 0, Rs, 0, Rn, Rt))
+#define CASPAxw(Rs, Rt, Rn)             EMIT(CASP_gen(0b00+rex.w, 1, Rs, 0, Rn, Rt))
+#define CASPALxw(Rs, Rt, Rn)            EMIT(CASP_gen(0b00+rex.w, 1, Rs, 1, Rn, Rt))
+#define CASPLxw(Rs, Rt, Rn)             EMIT(CASP_gen(0b00+rex.w, 0, Rs, 1, Rn, Rt))
+
+// FLAGM extension
+// Invert Carry Flag
+#define CFINV()             EMIT(0b1101010100<<22 | 0b0100<<12 | 0b000<<5 | 0b11111)
+
+#define RMIF_gen(imm6, Rn, mask)        (0b10111010000<<21 | (imm6)<<15 | 0b00001<<10 | (Rn)<<5 | (mask))
+// Rotate right reg and use as NZCV
+#define RMIF(Xn, shift, mask)           EMIT(RMIF_gen(shift, Xn, mask))
+
+#define SETF_gen(sz, Rn)                (0b00111010000<<21 | (sz)<<14 | 0b0010<<10 | (Rn)<<5 | 0b1101)
+// Set NZVc with 8bit value of reg: N=bit7, Z=[0..7]==0, V=bit8 eor bit7, C unchanged
+#define SETF8(Wn)                       EMIT(SETF_gen(0, Wn))
+// Set NZVc with 16bit value of reg: N=bit15, Z=[0..15]==0, V=bit16 eor bit15, C unchanged
+#define SETF16(Wn)                      EMIT(SETF_gen(1, Wn))
+
+// FLAGM2 extension
+// NZCV -> N=0 Z=C|V C=C&!V V=0
+#define AXFLAG()            EMIT(0b1101010100<<22 | 0b0100<<12 | 0b010<<5 | 0b11111)
+// NZCV -> N=!C&!Z Z=Z&C C=C|Z V=!C&Z
+#define XAFLAG()            EMIT(0b1101010100<<22 | 0b0100<<12 | 0b001<<5 | 0b11111)
+
+// FRINTTS extension
+#define FRINTxx_scalar(type, op, Rn, Rd)  (0b11110<<24 | (type)<<22 | 1<<21 | 0b0100<<17 | (op)<<15 | 0b10000<<10 | (Rn)<<5 | (Rd))
+#define FRINT32ZS(Sd, Sn)           EMIT(FRINTxx_scalar(0b00, 0b00, Sn, Sd))
+#define FRINT32ZD(Dd, Dn)           EMIT(FRINTxx_scalar(0b01, 0b00, Dn, Dd))
+#define FRINT32XS(Sd, Sn)           EMIT(FRINTxx_scalar(0b00, 0b01, Sn, Sd))
+#define FRINT32XD(Dd, Dn)           EMIT(FRINTxx_scalar(0b01, 0b01, Dn, Dd))
+#define FRINT64ZS(Sd, Sn)           EMIT(FRINTxx_scalar(0b00, 0b10, Sn, Sd))
+#define FRINT64ZD(Dd, Dn)           EMIT(FRINTxx_scalar(0b01, 0b10, Dn, Dd))
+#define FRINT64XS(Sd, Sn)           EMIT(FRINTxx_scalar(0b00, 0b11, Sn, Sd))
+#define FRINT64XD(Dd, Dn)           EMIT(FRINTxx_scalar(0b01, 0b11, Dn, Dd))
+
+#define FRINTxx_vector(Q, U, sz, op, Rn, Rd)    ((Q)<<30 | (U)<<29 | 0b01110<<24 | (sz)<<22 | 0b10000<<17 | 0b1111<<13 | (op)<<12 | 0b10<<10 | (Rn)<<5 | (Rd))
+#define VFRINT32ZS(Vd, Vn)          EMIT(FRINTxx_vector(0, 0, 0, 0, Vn, Vd))
+#define VFRINT32ZSQ(Vd, Vn)         EMIT(FRINTxx_vector(1, 0, 0, 0, Vn, Vd))
+#define VFRINT32XS(Vd, Vn)          EMIT(FRINTxx_vector(0, 1, 0, 0, Vn, Vd))
+#define VFRINT32XSQ(Vd, Vn)         EMIT(FRINTxx_vector(1, 1, 0, 0, Vn, Vd))
+#define VFRINT32ZD(Vd, Vn)          EMIT(FRINTxx_vector(0, 0, 1, 0, Vn, Vd))
+#define VFRINT32ZDQ(Vd, Vn)         EMIT(FRINTxx_vector(1, 0, 1, 0, Vn, Vd))
+#define VFRINT32XD(Vd, Vn)          EMIT(FRINTxx_vector(0, 1, 1, 0, Vn, Vd))
+#define VFRINT32XDQ(Vd, Vn)         EMIT(FRINTxx_vector(1, 1, 1, 0, Vn, Vd))
+#define VFRINT64ZS(Vd, Vn)          EMIT(FRINTxx_vector(0, 0, 0, 1, Vn, Vd))
+#define VFRINT64ZSQ(Vd, Vn)         EMIT(FRINTxx_vector(1, 0, 0, 1, Vn, Vd))
+#define VFRINT64XS(Vd, Vn)          EMIT(FRINTxx_vector(0, 1, 0, 1, Vn, Vd))
+#define VFRINT64XSQ(Vd, Vn)         EMIT(FRINTxx_vector(1, 1, 0, 1, Vn, Vd))
+#define VFRINT64ZD(Vd, Vn)          EMIT(FRINTxx_vector(0, 0, 1, 1, Vn, Vd))
+#define VFRINT64ZDQ(Vd, Vn)         EMIT(FRINTxx_vector(1, 0, 1, 1, Vn, Vd))
+#define VFRINT64XD(Vd, Vn)          EMIT(FRINTxx_vector(0, 1, 1, 1, Vn, Vd))
+#define VFRINT64XDQ(Vd, Vn)         EMIT(FRINTxx_vector(1, 1, 1, 1, Vn, Vd))
+
+// CRC32 extension
+#define CRC32C_gen(sf, Rm, sz, Rn, Rd)  ((sf)<<31 | 0b11010110<<21 | (Rm)<<16 | 0b010<<13 | 1<<12 | (sz)<<10 | (Rn)<<5 | (Rd))
+#define CRC32CB(Wd, Wn, Wm)         EMIT(CRC32C_gen(0, Wm, 0b00, Wn, Wd))
+#define CRC32CH(Wd, Wn, Wm)         EMIT(CRC32C_gen(0, Wm, 0b01, Wn, Wd))
+#define CRC32CW(Wd, Wn, Wm)         EMIT(CRC32C_gen(0, Wm, 0b10, Wn, Wd))
+#define CRC32CX(Wd, Wn, Xm)         EMIT(CRC32C_gen(1, Xm, 0b11, Wn, Wd))
+#define CRC32Cxw(Wd, Wn, Rm)        EMIT(CRC32C_gen(rex.w, Rm, 0b10|rex.w, Wn, Wd))
+
+// SHA1 extension
+#define SHA1H_gen(Rn, Rd)       (0b01011110<<24 | 0b10100<<17 | 0b10<<10 | (Rn)<<5 | (Rd))
+// SHA1 fixed rotate (ROL 30 of 32bits value)
+#define SHA1H(Sd, Sn)           EMIT(SHA1H_gen(Sn, Sd))
+
+#define SHA1SU1_gen(Rn, Rd)     (0b01011110<<24 | 0b10100<<17 | 0b00001<<12 | 0b10<<10 | (Rn)<<5 | (Rd))
+// SHA1 schedule update 1
+#define SHA1SU1(Vd, Vn)         EMIT(SHA1SU1_gen(Vn, Vd))
+
+#define SHA1C_gen(Rm, Rn, Rd)   (0b01011110<<24 | (Rm)<<16 | (Rn)<<5 | (Rd))
+// SHA1 hash update (choose)
+#define SHA1C(Qd, Sn, Vm)       EMIT(SHA1C_gen(Vm, Sn, Qd))
+
+#define SHA1M_gen(Rm, Rn, Rd)   (0b01011110<<24 | (Rm)<<16 | 0b010<<12 | (Rn)<<5 | (Rd))
+// SHA1 hash update (majority)
+#define SHA1M(Qd, Sn, Vm)       EMIT(SHA1M_gen(Vm, Sn, Qd))
+
+#define SHA1P_gen(Rm, Rn, Rd)   (0b01011110<<24 | (Rm)<<16 | 0b001<<12 | (Rn)<<5 | (Rd))
+// SHA1 hash update (parity)
+#define SHA1P(Qd, Sn, Vm)       EMIT(SHA1P_gen(Vm, Sn, Qd))
+
+#define SHA256SU0_gen(Rn,Rd)    (0b01011110<<24 | 0b10100<<17 | 0b00010<<12 | 0b10<<10 | (Rn)<<5 | (Rd))
+//SHA256 schedule update 0
+#define SHA256SU0(Vd, Vn)       EMIT(SHA256SU0_gen(Vn, Vd))
+
+#define SHA256SU1_gen(Rm, Rn, Rd)   (0b01011110<<24 | (Rm)<<16 | 0b110<<12 | (Rn)<<5 | (Rd))
+//SHA256 schedule update 1
+#define SHA256SU1(Vd, Vn, Vm)       EMIT(SHA256SU1_gen(Vm, Vn, Vd))
+
+#define SHA256H_gen(Rm, Rn, Rd) (0b01011110<<24 | (Rm)<<16 | 0b100<<12 | (Rn)<<5 | (Rd))
+//SHA256 hash update (part 1)
+#define SHA256H(Vd, Vn, Vm)     EMIT(SHA256H_gen(Vm, Vn, Vd))
+
+#define SHA256H2_gen(Rm, Rn, Rd)    (0b01011110<<24 | (Rm)<<16 | 0b101<<12 | (Rn)<<5 | (Rd))
+//SHA256 hash update (part 2)
+#define SHA256H2(Vd, Vn, Vm)        EMIT(SHA256H2_gen(Vm, Vn, Vd))
+
 #endif  //__ARM64_EMITTER_H__
@@ -34,23 +34,3 @@ arm64_epilog:
    ldp     x30, x29, [sp], 16  // saved lr
    //end, return...
    ret
-
-
-.global arm64_epilog_fast
-arm64_epilog_fast:
-    //restore all used register
-    add     sp, x28, 0
-    //vpop     {d8-d15}
-    ldp     x19, x20, [sp, (8 * 0)]
-    ldp     x21, x22, [sp, (8 * 2)]
-    ldp     x23, x24, [sp, (8 * 4)]
-    ldp     x25, x26, [sp, (8 * 6)]
-    ldp     x27, x28, [sp, (8 * 8)]
-    ldp     d8, d9,   [sp, (8 *10)]
-    ldp     d10, d11, [sp, (8 *12)]
-    ldp     d12, d13, [sp, (8 *14)]
-    ldp     d14, d15, [sp, (8 *16)]
-    add     sp,  sp, (8 * 18)
-    ldp     x30, x29, [sp], 16  // saved lr
-    //end, return...
-    ret
@@ -0,0 +1,86 @@
+#include <stdint.h>
+#include "arm64_immenc.h"
+
+#define one (uint64_t)1
+
+// Returns the packed valid-N-imms-immr (1-1-6-6 bits); returns 0 if bitmask is not encodable
+int convert_bitmask(uint64_t bitmask) {
+	if (!bitmask || !~bitmask) return 0;
+	
+	uint64_t size, mask, pat;
+	
+	for (size = 6; size > 1; --size) {
+		mask = (one << (1 << (size - 1))) - 1;
+		pat = bitmask & mask;
+		// printf("%016lX/%lu: %016lX %016lX %2d\n", bitmask, size, mask, pat, (1 << (size - 1)));
+		if (pat != ((bitmask >> (1 << (size - 1))) & mask)) {
+			// printf("%016lX/%lu: %016lX %016lX xx\n", bitmask, size, pat, (bitmask >> (1 << (size - 1))) & mask);
+			break;
+		}
+	}
+	mask = (size >= 6) ? ((uint64_t)-1) : ((one << (1 << size)) - 1);
+	pat = bitmask & mask;
+	for (uint64_t i = 1; i < 7 - size; ++i) {
+		uint64_t boff = i * (1 << size);
+		if (((bitmask >> boff) & mask) != pat) {
+			// printf("%016lX/%lu: no %lu %lu %016lX %016lX %016lX\n", bitmask, size, i, size, boff, mask, pat);
+			return 0;
+		}
+	}
+	// Note that here, pat != 0 and ~pat & (1 << size) != 0 (otherwise size = 1 and bitmask = all 0 or all 1)
+	int immr = 0;
+	uint64_t last_bit = one << ((1 << size) - 1);
+	// printf("%016lX/%lu: %016lX %016lX %lu\n", bitmask, size, mask, pat, last_bit);
+	if (pat & 1) {
+		while (pat & last_bit) {
+			pat = ((pat - last_bit) << 1) + 1;
+			++immr;
+		}
+	} else {
+		immr = 1 << size;
+		while (!(pat & 1)) {
+			pat >>= 1;
+			--immr;
+		}
+	}
+	// printf("%016lX/%lu: %016lX %016lX %lu %d\n", bitmask, size, mask, pat, last_bit, immr);
+	if (pat & (pat + 1)) return 0; // Not 0...01...1
+	int to = 1;
+	while (pat & (one << to)) ++to;
+	
+	// printf("%016lX/%lu: returning %c%c%02lX%02lX\n", bitmask, size, '2' + (size == 6), (uint64_t)(((0x1E << size) & 0x3F) + (to - 1)), (uint64_t)immr);
+	return 0x2000 + ((size == 6) << 12) + ((((0x1E << size) & 0x3F) + (to - 1)) << 6) + immr;
+}
+
+#if 0
+#include <unordered_set>
+int main() {
+	std::unordered_set<uint64_t> okvals;
+	uint64_t val;
+	for (int n = 1; n < 7; ++n) {
+		int imms0 = ((n == 6) ? 0xC0 : 0x80) + ((0x1E << n) & 0x3F);
+		for (int nones = 0; nones < (1 << n) - 1; ++nones) {
+			val = (one << (nones + 1)) - 1;
+			for (int j = 0; j < 6 - n; ++j) {
+				val = val + (val << (1 << (n + j)));
+			}
+			for (int immr = 0; immr < (1 << n); ++immr) {
+				int exp = ((imms0 + nones) << 6) + immr;
+				int got = convert_bitmask(val);
+				if (exp != got) {
+					printf("0x%016lX: expected %04X, got %04X\n", val, exp, got);
+				}
+				okvals.emplace(val);
+				val = (val >> 1) + ((val & 1) << 63);
+			}
+		}
+	}
+#pragma omp parallel for
+	for (uint64_t i = 1; i < 0xFFFFFFFFFFFFFFFFu; ++i) {
+		int got = convert_bitmask(val);
+		if (!!got != (okvals.find(val) != okvals.cend())) {
+			printf("0x%016lX: expected %s, got %04X\n", val, (!!got ? "0000" : "non0"), got);
+		}
+	}
+}
+#endif
@@ -0,0 +1,8 @@
+#ifndef __ARM64_IMMENC_H__
+#define __ARM64_IMMENC_H__
+
+#include <stdint.h>
+
+int convert_bitmask(uint64_t bitmask);
+
+#endif // __ARM64_IMMENC_H__
@@ -29,6 +29,10 @@
 .global arm64_lock_incif0
 .global arm64_lock_decifnot0
 .global arm64_lock_store
+.global arm64_lock_store_dd
+.global arm64_lock_get_b
+.global arm64_lock_get_d
+.global arm64_lock_get_dd

 arm64_lock_read_b:
    dmb     ish
@@ -254,3 +258,20 @@ arm64_lock_store:
    str     w1, [x0]
    dmb     ish
    ret
+
+arm64_lock_store_dd:
+    str     x1, [x0]
+    dmb     ish
+    ret
+
+arm64_lock_get_b:
+    ldaxrb  w0, [x0]
+    ret
+
+arm64_lock_get_d:
+    ldaxr   w0, [x0]
+    ret
+
+arm64_lock_get_dd:
+    ldaxr   x0, [x0]
+    ret
@@ -27,46 +27,58 @@ extern void arm64_lock_read_dq(uint64_t * a, uint64_t* b, void* addr);
 // STLXRD of ADDR, return 0 if ok, 1 if not
 extern int arm64_lock_write_dq(uint64_t a, uint64_t b, void* addr);

-// Atomicaly exchange value at [p] with val, return old p
+// Atomically exchange value at [p] with val, return old p
 extern uintptr_t arm64_lock_xchg_dd(void* p, uintptr_t val);

-// Atomicaly exchange value at [p] with val, return old p
+// Atomically exchange value at [p] with val, return old p
 extern uint32_t arm64_lock_xchg_d(void* p, uint32_t val);

-// Atomicaly exchange value at [p] with val, return old p
+// Atomically exchange value at [p] with val, return old p
 extern uint32_t arm64_lock_xchg_h(void* p, uint32_t val);

-// Atomicaly exchange value at [p] with val, return old p
+// Atomically exchange value at [p] with val, return old p
 extern uint32_t arm64_lock_xchg_b(void* p, uint32_t val);

-// Atomicaly store value to [p] only if [p] is NULL. Return old [p] value
+// Atomically store value to [p] only if [p] is NULL. Return old [p] value
 extern uint32_t arm64_lock_storeifnull_d(void*p, uint32_t val);

-// Atomicaly store value to [p] only if [p] is NULL. Return old [p] value
+// Atomically store value to [p] only if [p] is NULL. Return old [p] value
 extern void* arm64_lock_storeifnull(void*p, void* val);

-// Atomicaly 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 new [p] value (so val or old)
 extern void* arm64_lock_storeifref(void*p, void* val, void* ref);

-// Atomicaly 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 new [p] value (so val or old)
 extern uint32_t arm64_lock_storeifref_d(void*p, uint32_t val, uint32_t ref);

-// Atomicaly 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 new [p] value (so val or old)
 extern uint32_t arm64_lock_storeifref2_d(void*p, uint32_t val, uint32_t ref);

-// decrement atomicaly the byte at [p] (but only if p not 0)
+// decrement atomically the byte at [p] (but only if p not 0)
 extern void arm64_lock_decifnot0b(void*p);

 // atomic store (with memory barrier)
 extern void arm64_lock_storeb(void*p, uint8_t b);

-// increment atomicaly the int at [p] only if it was 0. Return the old value of [p]
+// increment atomically the int at [p] only if it was 0. Return the old value of [p]
 extern int arm64_lock_incif0(void*p);

-// decrement atomicaly the int at [p] (but only if p not 0)
+// decrement atomically the int at [p] (but only if p not 0)
 extern int arm64_lock_decifnot0(void*p);

 // atomic store (with memory barrier)
 extern void arm64_lock_store(void*p, uint32_t v);

-#endif  //__ARM64_LOCK__H__
+// atomic store (with memory barrier)
+extern void arm64_lock_store_dd(void*p, uint64_t v);
+
+// atomic get (with memory barrier)
+extern uint8_t arm64_lock_get_b(void*p);
+
+// atomic get (with memory barrier)
+extern uint32_t arm64_lock_get_d(void*p);
+
+// atomic get (with memory barrier)
+extern void* arm64_lock_get_dd(void*p);
+
+#endif  //__ARM64_LOCK__H__
@@ -16,7 +16,7 @@ static const char* conds[] = {"cEQ", "cNE", "cCS", "cCC", "cMI", "cPL", "cVS", "
 #define abs(A) (((A)<0)?(-(A)):(A))

 typedef struct arm64_print_s {
-    int N, S, U, L, Q;
+    int N, S, U, L, Q, A, R;
    int t, n, m, d, t2, a;
    int f, c, o, h, p;
    int i, r, s;
@@ -34,10 +34,12 @@ uint64_t DecodeBitMasks(int N, int imms, int immr)
    if(s==levels) return 0;
    uint64_t mask = (1LL<<(s+1))-1;
    if(r) { // rotate
-         mask=(mask>>r)|(mask<<(e-r));
-         mask&=((1LL<<e)-1);
+        mask=(mask>>r)|(mask<<(e-r));
+        if(e<64) {
+            mask&=((1LL<<e)-1);
+        }
    }
-    while (e<64) {  // replicate
+    while(e<64) {  // replicate
        mask|=(mask<<e);
        e<<=1;
    }
@@ -60,6 +62,8 @@ int isMask(uint32_t opcode, const char* mask, arm64_print_t *a)
            case 'N': a->N = (a->N<<1) | v; break;
            case 'S': a->S = (a->S<<1) | v; break;
            case 'U': a->U = (a->U<<1) | v; break;
+            case 'A': a->A = (a->A<<1) | v; break;
+            case 'R': a->R = (a->R<<1) | v; break;
            case 'L': a->L = (a->L<<1) | v; break;
            case 'Q': a->Q = (a->Q<<1) | v; break;
            case 't': a->t = (a->t<<1) | v; break;
@@ -182,22 +186,22 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
    }
    if(isMask(opcode, "01111000010iiiiiiiii00nnnnnttttt", &a)) {
        int offset = signExtend(imm, 9);
-        snprintf(buff, sizeof(buff), "LDRUH %s, [%s, %s0x%x]", Wt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
+        snprintf(buff, sizeof(buff), "LDURH %s, [%s, %s0x%x]", Wt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
        return buff;
    }
    if(isMask(opcode, "001110001x0iiiiiiiii00nnnnnttttt", &a)) {
        int offset = signExtend(imm, 9);
-        snprintf(buff, sizeof(buff), "LDRUSB %s, [%s, %s0x%x]", a.x?Xt[Rt]:Wt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
+        snprintf(buff, sizeof(buff), "LDURSB %s, [%s, %s0x%x]", a.x?Xt[Rt]:Wt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
        return buff;
    }
    if(isMask(opcode, "011110001x0iiiiiiiii00nnnnnttttt", &a)) {
        int offset = signExtend(imm, 9);
-        snprintf(buff, sizeof(buff), "LDRUSH %s, [%s, %s0x%x]", a.x?Xt[Rt]:Wt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
+        snprintf(buff, sizeof(buff), "LDURSH %s, [%s, %s0x%x]", a.x?Xt[Rt]:Wt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
        return buff;
    }
    if(isMask(opcode, "10111000100iiiiiiiii00nnnnnttttt", &a)) {
        int offset = signExtend(imm, 9);
-        snprintf(buff, sizeof(buff), "LDRUSW %s, [%s, %s0x%x]", Xt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
+        snprintf(buff, sizeof(buff), "LDURSW %s, [%s, %s0x%x]", Xt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
        return buff;
    }
    if(isMask(opcode, "1x111000000iiiiiiiii00nnnnnttttt", &a)) {
@@ -213,7 +217,7 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
    }
    if(isMask(opcode, "01111000000iiiiiiiii00nnnnnttttt", &a)) {
        int offset = signExtend(imm, 9);
-        snprintf(buff, sizeof(buff), "STRUH %s, [%s, %s0x%x]", Wt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
+        snprintf(buff, sizeof(buff), "STURH %s, [%s, %s0x%x]", Wt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
        return buff;
    }
    if(isMask(opcode, "1x11100101iiiiiiiiiiiinnnnnttttt", &a)) {
@@ -385,13 +389,13 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
    }
    if(isMask(opcode, "f00100101wwiiiiiiiiiiiiiiiiddddd", &a)) {
        if(sf) {
-            uint64_t noti=~(uint64_t)imm;
+            uint64_t noti=~(((uint64_t)imm)<<(hw*16));
            if(!hw)
                snprintf(buff, sizeof(buff), "MOVN %s, 0x%x\t; 0x%lx", Xt[Rd], imm, noti);
            else
                snprintf(buff, sizeof(buff), "MOVN %s, 0x%x LSL %d\t; 0x%lx", Xt[Rd], imm, 16*hw, noti);
        } else {
-            uint32_t noti=~(uint32_t)imm;
+            uint32_t noti=~(((uint32_t)imm)<<(hw*16));
            if(!hw)
                snprintf(buff, sizeof(buff), "MOVN %s, 0x%x\t; 0x%x", Wt[Rd], imm, noti);
            else
@@ -795,6 +799,11 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
        snprintf(buff, sizeof(buff), "CSEL %s, %s, %s, %s", sf?Xt[Rd]:Wt[Rd], sf?Xt[Rn]:Wt[Rn], sf?Xt[Rm]:Wt[Rm], conds[cond]);    
        return buff;
    }
+    if(isMask(opcode, "f1111010010mmmmmcccc00nnnnn0iiii", &a)) {
+        snprintf(buff, sizeof(buff), "CCMP %s, %s, %s 0x%x", sf?Xt[Rn]:Wt[Rn], sf?Xt[Rm]:Wt[Rm], conds[cond], imm);
+        return buff;
+    }
+
    // MISC Bits
    if(isMask(opcode, "f10110101100000000010onnnnnddddd", &a)) {
        snprintf(buff, sizeof(buff), "CL%c %s, %s", option?'S':'Z', sf?Xt[Rd]:Wt[Rd], sf?Xt[Rn]:Wt[Rn]);
@@ -950,6 +959,22 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
        snprintf(buff, sizeof(buff), "VCMEQ V%d.%s, V%d.%s, V%d.%s", Rd, Vd, Rn, Vd, Rm, Vd);
        return buff;
    }
+    // CMP zero
+    if(isMask(opcode, "0QU01110ff100000100o10nnnnnddddd", &a)) {
+        const char* Y[] = {"8B", "16B", "4H", "8H", "2S", "4S", "??", "2D"};
+        const char* Z[] = {"GT", "GE", "EQ", "LE"};
+        const char* Vd = Y[((sf)<<1) | a.Q];
+        const char* Cond = Z[(a.o << 1 | a.U)];
+        snprintf(buff, sizeof(buff), "VCM%s V%d.%s, V%d.%s, #0", Cond, Rd, Vd, Rn, Vd);
+        return buff;
+    }
+    // CMPLT zero
+    if(isMask(opcode, "0Q001110ff100000101010nnnnnddddd", &a)) {
+        const char* Y[] = {"8B", "16B", "4H", "8H", "2S", "4S", "??", "2D"};
+        const char* Vd = Y[((sf)<<1) | a.Q];
+        snprintf(buff, sizeof(buff), "VCMLT V%d.%s, V%d.%s, #0", Rd, Vd, Rn, Vd);
+        return buff;
+    }
    // MIN/MAX
    if(isMask(opcode, "0QU01110ff1mmmmm0110o1nnnnnddddd", &a)) {
        const char* Y[] = {"8B", "16B", "4H", "8H", "2S", "4S", "??", "2D"};
@@ -965,11 +990,21 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
        snprintf(buff, sizeof(buff), "MOVI V%d.%s, #0x%x", Rd, Vd, imm);
        return buff;
    }
-    // MOV immediate (not)shifted 16bits & 32bits
+    // MOV immediate notshifted 16bits & 32bits
    if(isMask(opcode, "0Q00111100000iiif00001iiiiiddddd", &a)) {
        const char* Y[] = {"2S", "4S", "4H", "8H"};
        const char* Vd = Y[(sf<<1)| a.Q];
-        snprintf(buff, sizeof(buff), "MOVI V%d.%s, #0x%x", Rd, Vd, imm);
+        int sh = 0;
+
+        snprintf(buff, sizeof(buff), "MOVI V%d.%s, #0x%x", Rd, Vd, imm<<sh);
+        return buff;
+    }
+    // MOV immediate shifted 16bits
+    if(isMask(opcode, "0Q00111100000iii101001iiiiiddddd", &a)) {
+        const char* Y[] = {"4H", "8H"};
+        const char* Vd = Y[a.Q];
+
+        snprintf(buff, sizeof(buff), "MOVI V%d.%s, #0x%x", Rd, Vd, imm<<8);
        return buff;
    }

@@ -1034,7 +1069,30 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
        if(!offset)
            snprintf(buff, sizeof(buff), "%s %c%d, [%s]", op?"STR":"LDR", s, Rt, XtSp[Rn]);
        else
-            snprintf(buff, sizeof(buff), "%s %c%d, [%s, %d]", op?"STR":"LDR", s, Rt, XtSp[Rn], offset);
+            snprintf(buff, sizeof(buff), "%s %c%d, [%s, 0x%x]", op?"STR":"LDR", s, Rt, XtSp[Rn], offset);
+        return buff;
+    }
+    if(isMask(opcode, "ss111100cc1mmmmmoooS10nnnnnttttt", &a)) {
+        char s = '?';
+        int size=imms;
+        int op=0;
+        if(size==0 && opc==1) {s='B';}
+        else if(size==1 && opc==1) {s='H';}
+        else if(size==2 && opc==1) {s='S';}
+        else if(size==3 && opc==1) {s='D';}
+        else if(size==0 && opc==3) {s='Q'; size = 4;}
+        else if(size==0 && opc==0) {s='B'; op=1;}
+        else if(size==1 && opc==0) {s='H'; op=1;}
+        else if(size==2 && opc==0) {s='S'; op=1;}
+        else if(size==3 && opc==0) {s='D'; op=1;}
+        else if(size==0 && opc==2) {s='Q'; op=1; size = 4;}
+
+        const char* extend[] = {"?0", "?1", "UXTW", "LSL", "?4", "?5", "SXTW", "SXTX"};
+        int amount = size*a.S;
+        if(option==3 && !amount)
+            snprintf(buff, sizeof(buff), "%s %c%d, [%s, %s]", op?"STR":"LDR", s, Rt, XtSp[Rn], ((option&1)==0)?Wt[Rm]:Xt[Rm]);
+        else
+            snprintf(buff, sizeof(buff), "%s %c%d, [%s, %s, %s %d]", op?"STR":"LDR", s, Rt, XtSp[Rn], ((option&1)==0)?Wt[Rm]:Xt[Rm], extend[option], amount);
        return buff;
    }

@@ -1094,6 +1152,20 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
        return buff;
    }

+    // FADDP
+    if(isMask(opcode, "0Q1011100f1mmmmm110101nnnnnddddd", &a)) {
+        char s = a.Q?'V':'D';
+        char d = sf?'D':'S';
+        int n = (a.Q && !sf)?4:2;
+        snprintf(buff, sizeof(buff), "VFADDP %c%d.%d%c, %c%d.%d%c, %c%d.%d%c", s, Rd, n, d, s, Rn, n, d, s, Rm, n, d);
+        return buff;
+    }
+    if(isMask(opcode, "011111100f110000110110nnnnnddddd", &a)) {
+        char s = (sf==0)?'S':((sf==1)?'D':'?');
+        snprintf(buff, sizeof(buff), "FADDP %c%d, %c%d, %c%d", s, Rd, s, Rn, s, Rm);
+        return buff;
+    }
+
    // SQRT
    if(isMask(opcode, "0Q1011101f100001111110nnnnnddddd", &a)) {
        char s = a.Q?'V':'D';
@@ -1107,7 +1179,19 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
        snprintf(buff, sizeof(buff), "FSQRT %c%d, %c%d", s, Rd, s, Rn);
        return buff;
    }
-
+    // FPRINTX
+    if(isMask(opcode, "0Q1011100f100001100110nnnnnddddd", &a)) {
+        char s = a.Q?'V':'D';
+        char d = sf?'D':'S';
+        int n = (a.Q && !sf)?4:2;
+        snprintf(buff, sizeof(buff), "VFRINTX %c%d.%d%c, %c%d.%d%c", s, Rd, n, d, s, Rn, n, d);
+        return buff;
+    }
+    if (isMask(opcode, "00011110ff100111010000nnnnnddddd", &a)) {
+        char s = (sf==0)?'S':((sf==1)?'D':'?');
+        snprintf(buff, sizeof(buff), "FRINTX %c%d, %c%d", s, Rd, s, Rn);
+        return buff;
+    }
    // FRINTI
    if(isMask(opcode, "0Q1011101f100001100110nnnnnddddd", &a)) {
        char s = a.Q?'V':'D';
@@ -1290,6 +1374,14 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
        snprintf(buff, sizeof(buff), "FCVT%sS %s, %c%d", roundings[a.c], sf?Xt[Rd]:Wt[Rd], s, Rn);
        return buff;
    }
+    if(isMask(opcode, "0QU01110of100001101o10nnnnnddddd", &a)) {
+        const char* Y[] = {"2S", "4S", "??", "2D"};
+        const char* Z[] = {"S", "S", "??", "D"};
+        const char* Vd = Y[(sf<<1) | a.Q];
+        const char* roundings[] = {"N", "M", "P", "Z"};
+        snprintf(buff, sizeof(buff), "VFCVT%s%s%s%s V%d.%s, V%d.%s", roundings[option], a.U?"U":"S", a.Q?"Q":"", Z[(sf<<1)|a.Q], Rd, Vd, Rn, Vd);
+        return buff;
+    }

    // FMOV
    if(isMask(opcode, "00011110pp100000010000nnnnnddddd", &a)) {
@@ -1484,10 +1576,21 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
            snprintf(buff, sizeof(buff), "%cSHLL%s V%d.%s, V%d.%s, #%d", a.U?'U':'S', a.Q?"2":"", Rd, Va, Rn, Vd, sh);
        return buff;
    }
+    if(isMask(opcode, "0Q0011110hhhhiii010101nnnnnddddd", &a) && (a.h != 0b1000)) {
+        const char* Y[] = {"8B", "16B", "4H", "8H", "2S", "4S", "??", "2D"};
+        int sz = 3;
+        if((a.h&0b1111)==0b0001) sz=0;
+        else if((a.h&0b1110)==0b0010) sz=1;
+        else if((a.h&0b1100)==0b0100) sz=2;
+        int sh=(((a.h)<<3)|(imm)) - (8<<sz);
+        const char* Vd = Y[(sz<<1)|a.Q];
+        snprintf(buff, sizeof(buff), "SHL%s V%d.%s, V%d.%s, #%d", a.Q?"Q":"", Rd, Vd, Rn, Vd, sh);
+        return buff;
+    }

    // DUP
    if(isMask(opcode, "0Q001110000iiiii000001nnnnnddddd", &a)) {
-        const char* Y[] = {"8B", "16B", "4H", "8H", "2S", "4S", "2D", "??"};
+        const char* Y[] = {"8B", "16B", "4H", "8H", "2S", "4S", "??", "2D"};
        const char* Z[] = {"B", "H", "S", "D"};
        int sz = 3;
        if((imm&0b0001)==0b0001) sz=0;
@@ -1499,7 +1602,18 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
        snprintf(buff, sizeof(buff), "DUP V%d.%s, V%d.%s[%d]", Rd, Vd, Rn, Vn, sh);
        return buff;
    }
-    
+
+    if(isMask(opcode, "0Q001110000iiiii000011nnnnnddddd", &a)) {
+        const char* Y[] = {"8B", "16B", "4H", "8H", "2S", "4S", "??", "2D"};
+        int sz = 3;
+        if((imm&0b0001)==0b0001) sz=0;
+        else if((imm&0b0011)==0b0010) sz=1;
+        else if((imm&0b0111)==0b0100) sz=2;
+        const char* Vd = Y[(sz<<1)|a.Q];
+        snprintf(buff, sizeof(buff), "DUP V%d.%s, X%d", Rd, Vd, Rn);
+        return buff;
+    }
+
    // AES
    if(isMask(opcode, "0100111000101000010f10nnnnnddddd", &a)) {
        snprintf(buff, sizeof(buff), "AES%c V%d.16B, V%d.16B", sf?'D':'E', Rd, Rn);
@@ -1525,7 +1639,144 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
        snprintf(buff, sizeof(buff), "DMB %s", (Rn==0b1011)?"ISH":"???");
        return buff;
    }
+    // DSB ISH/ISHST
+    if(isMask(opcode, "11010101000000110011nnnn10011111", &a)) {
+        snprintf(buff, sizeof(buff), "DSB %s", (Rn==0b1011)?"ISH":((Rn==0b1010)?"ISHST":"???"));
+        return buff;
+    }

+    // CASxw
+    if(isMask(opcode, "1f0010001L1ssssso11111nnnnnttttt", &a)) {
+        snprintf(buff, sizeof(buff), "CAS%s%s %s, %s, [%s]", a.o?"A":"", a.L?"L":"", sf?Xt[Rs]:Wt[Rs], sf?Xt[Rt]:Wt[Rt], XtSp[Rn]);
+        return buff;
+    }
+    // CAS B/H
+    if(isMask(opcode, "0f0010001L1ssssso11111nnnnnttttt", &a)) {
+        snprintf(buff, sizeof(buff), "CAS%s%s%s %s, %s, [%s]", a.o?"A":"", a.L?"L":"", sf?"H":"B", Xt[Rs], Xt[Rt], XtSp[Rn]);
+        return buff;
+    }
+    // CASPxw
+    if(isMask(opcode, "0f0010000L1ssssso11111nnnnnttttt", &a)) {
+        snprintf(buff, sizeof(buff), "CASP%s%s %s,%s, %s,%s, [%s]", a.o?"A":"", a.L?"L":"", sf?Xt[Rs]:Wt[Rs], sf?Xt[Rs+1]:Wt[Rs+1], sf?Xt[Rt]:Wt[Rt], sf?Xt[Rt+1]:Wt[Rt+1], XtSp[Rn]);
+        return buff;
+    }
+    // SWPxw
+    if(isMask(opcode, "1f111000AR1sssss100000nnnnnttttt", &a)) {
+        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
+    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]);
+        return buff;
+    }
+    // LDXXXxw
+    if(isMask(opcode, "1f111000AR1sssss0ooo00nnnnnttttt", &a)) {
+        const char* ops[] = { "ADD", "CLR", "EOR", "SET", "SMAX, SMIN", "UMAX", "UMIN" };
+        if((Rt == 0b11111) && !a.A) {
+            snprintf(buff, sizeof(buff), "ST%s%s %s, [%s]", ops[a.o], a.R?"L":"", sf?Xt[Rs]:Wt[Rs], XtSp[Rn]);
+        } else {
+            snprintf(buff, sizeof(buff), "LD%s%s%s %s, %s, [%s]", ops[a.o], a.A?"A":"", a.R?"L":"", sf?Xt[Rs]:Wt[Rs], sf?Xt[Rt]:Wt[Rt], XtSp[Rn]);
+        }
+        return buff;
+    }
+    // LDXXX B/H
+    if(isMask(opcode, "0f111000AR1sssss0ooo00nnnnnttttt", &a)) {
+        const char* ops[] = { "ADD", "CLR", "EOR", "SET", "SMAX, SMIN", "UMAX", "UMIN" };
+        if((Rt == 0b11111) && !a.A) {
+            snprintf(buff, sizeof(buff), "ST%s%s%s %s, [%s]", ops[a.o], a.R?"L":"", sf?"H":"B", Xt[Rs], XtSp[Rn]);
+        } else {
+            snprintf(buff, sizeof(buff), "LD%s%s%s%s %s, %s, [%s]", ops[a.o], a.A?"A":"", a.R?"L":"", sf?"H":"B", Xt[Rs], Xt[Rt], XtSp[Rn]);
+        }
+        return buff;
+    }
+    // AXFLAG
+    if(isMask(opcode, "11010101000000000100000001011111", &a)) {
+        snprintf(buff, sizeof(buff), "AXFLAG");
+        return buff;
+    }
+    // XAFLAG
+    if(isMask(opcode, "11010101000000000100000000111111", &a)) {
+        snprintf(buff, sizeof(buff), "XAFLAG");
+        return buff;
+    }
+    // CFINV
+    if(isMask(opcode, "11010101000000000100000000011111", &a)) {
+        snprintf(buff, sizeof(buff), "CFINV");
+        return buff;
+    }
+    // RMIF
+    if(isMask(opcode, "10111010000iiiiii00001nnnnn0oooo", &a)) {
+        snprintf(buff, sizeof(buff), "RMIF %s, #%d, #0x%x", Xt[Rn], imm, opc);
+        return buff;
+    }
+    // SETF
+    if(isMask(opcode, "00111010000000000f0010nnnnn01101", &a)) {
+        snprintf(buff, sizeof(buff), "SETF%d %s", 8<<sf, Xt[Rn]);
+        return buff;
+    }
+    // REV64
+    if(isMask(opcode, "0Q001110ff100000000010nnnnnddddd", &a)) {
+        const char* T[] = {"8B", "16B", "4H", "8H", "2S", "4S", "??", "??"};
+        int sz = sf;
+        const char* Vn = T[(sz<<1)|a.Q];
+        const char* Vd = T[(sz<<1)|a.Q];
+        snprintf(buff, sizeof(buff), "REV64 V%d.%s, V%d.%s",Rd, Vd, Rn, Vn);
+        return buff;
+    }
+    //TRNx
+    if(isMask(opcode, "0Q001110ff0mmmmm0o1010nnnnnddddd", &a)) {
+        const char* T[] = {"8B", "16B", "4H", "8H", "2S", "4S", "??", "2D"};
+        int sz = sf;
+        const char* Vn = T[(sz<<1)|a.Q];
+        const char* Vm = T[(sz<<1)|a.Q];
+        const char* Vd = T[(sz<<1)|a.Q];
+        snprintf(buff, sizeof(buff), "TRN%d V%d.%s, V%d.%s, V%d.%s", a.o+1, Rd, Vd, Rn, Vn, Rm, Vm);
+        return buff;
+    }
+    //SHA1 stuffs
+    if(isMask(opcode, "0101111000101000000010nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA1H S%d, S%d", Rd, Rn);
+        return buff;
+    }
+    if(isMask(opcode, "0101111000101000000110nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA1SU1 V%d.4S, V%d.4S", Rd, Rn);
+        return buff;
+    }
+    if(isMask(opcode, "01011110000mmmmm000000nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA1C Q%d, S%d, V%d.4S", Rd, Rn, Rm);
+        return buff;
+    }
+    if(isMask(opcode, "01011110000mmmmm001000nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA1M Q%d, S%d, V%d.4S", Rd, Rn, Rm);
+        return buff;
+    }
+    if(isMask(opcode, "01011110000mmmmm000100nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA1P Q%d, S%d, V%d.4S", Rd, Rn, Rm);
+        return buff;
+    }
+    //SHA256 stuffs
+    if(isMask(opcode, "0101111000101000001010nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA256SU0 V%d.4S, V%d.4S", Rd, Rn);
+        return buff;
+    }
+    if(isMask(opcode, "01011110000mmmmm011000nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA256SU1 V%d.4S, V%d.4S, V%d.4S", Rd, Rn, Rm);
+        return buff;
+    }
+    if(isMask(opcode, "01011110000mmmmm010000nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA256H Q%d, Q%d, V%d.4S", Rd, Rn, Rm);
+        return buff;
+    }
+    if(isMask(opcode, "01011110000mmmmm010100nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA256H2 Q%d, Q%d, V%d.4S", Rd, Rn, Rm);
+        return buff;
+    }
+    // UDF
+    if(isMask(opcode, "0000000000000000iiiiiiiiiiiiiiii", &a)) {
+        snprintf(buff, sizeof(buff), "UDF 0x%x", a.i);
+        return buff;
+    }

    snprintf(buff, sizeof(buff), "%08X ???", __builtin_bswap32(opcode));
    return buff;
@@ -6,8 +6,6 @@
 .text
 .align 4

-.extern arm64_next
-
 .global arm64_prolog
 arm64_prolog:
    //save all 18 used register
@@ -34,8 +32,7 @@ arm64_prolog:
    ldp     x24, x25, [x0, (8 * 14)]
    ldp     x26, x27, [x0, (8 * 16)]
    // Push a jump next on the stack
-    adr     x28, arm64_next
-    stp     x28, xzr, [sp, -16]!
+    stp     xzr, xzr, [sp, -16]!
    // Save old xSP in x28
    add     x28, sp, 16
    //jump to function
@@ -58,6 +58,16 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin

    switch(opcode) {

+        case 0x01:
+            INST_NAME("ADD Seg:Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            GETGD;
+            GETEDO(x4, 0);
+            emit_add32(dyn, ninst, rex, ed, gd, x3, x5);
+            WBACKO(x4);
+            break;
        case 0x03:
            INST_NAME("ADD Gd, Seg:Ed");
            SETFLAGS(X_ALL, SF_SET_PENDING);
@@ -68,6 +78,16 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            emit_add32(dyn, ninst, rex, gd, ed, x3, x4);
            break;

+        case 0x0B:
+            INST_NAME("OR Gd, Seg:Ed");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            GETGD;
+            GETEDO(x4, 0);
+            emit_or32(dyn, ninst, rex, gd, ed, x3, x4);
+            break;
+
        case 0x0F:
            opcode = F8;
            switch(opcode) {
@@ -223,8 +243,8 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                                    UFLAG_RES(gd);
                                    LSRx(x3, gd, 32);
                                    UFLAG_OP1(x3);
-                                    UFLAG_DF(x3, d_imul32);
                                    MOVw_REG(gd, gd);
+                                    UFLAG_DF(x3, d_imul32);
                                } else {
                                    MULxw(gd, gd, ed);
                                }
@@ -268,6 +288,49 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            }
            break;

+        case 0x13:
+            INST_NAME("ADC Gd, Seg:Ed");
+            READFLAGS(X_CF);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            GETGD;
+            GETEDO(x4, 0);
+            emit_adc32(dyn, ninst, rex, gd, ed, x3, x5);
+            break;
+
+        case 0x21:
+            INST_NAME("AND Seg:Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            GETGD;
+            GETEDO(x4, 0);
+            emit_and32(dyn, ninst, rex, ed, gd, x3, x5);
+            WBACKO(x4);
+            break;
+
+        case 0x23:
+            INST_NAME("AND Gd, Seg:Ed");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            GETGD;
+            GETEDO(x4, 0);
+            emit_and32(dyn, ninst, rex, gd, ed, x3, x5);
+            break;
+
+        case 0x29:
+            INST_NAME("SUB Seg:Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            GETGD;
+            GETEDO(x4, 0);
+            emit_sub32(dyn, ninst, rex, ed, gd, x3, x5);
+            WBACKO(x4);
+            break;
+
        case 0x2B:
            INST_NAME("SUB Gd, Seg:Ed");
            SETFLAGS(X_ALL, SF_SET_PENDING);
@@ -278,6 +341,17 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            emit_sub32(dyn, ninst, rex, gd, ed, x3, x4);
            break;

+        case 0x31:
+            INST_NAME("XOR Seg:Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            GETGD;
+            GETEDO(x4, 0);
+            emit_xor32(dyn, ninst, rex, ed, gd, x3, x5);
+            WBACKO(x4);
+            break;
+
        case 0x33:
            INST_NAME("XOR Gd, Seg:Ed");
            SETFLAGS(X_ALL, SF_SET_PENDING);
@@ -337,10 +411,22 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                }
            }
            break;
-
+        case 0x64:
+            addr = dynarec64_64(dyn, addr, ip, ninst, rex, rep, _FS, ok, need_epilog);
+            break;
+        case 0x65:
+            addr = dynarec64_64(dyn, addr, ip, ninst, rex, rep, _GS, ok, need_epilog);
+            break;
        case 0x66:
            addr = dynarec64_6664(dyn, addr, ip, ninst, rex, seg, ok, need_epilog);
            break;
+        case 0x67:
+            if(rex.is32bits)
+                addr = dynarec64_6764_32(dyn, addr, ip, ninst, rex, rep, seg, ok, need_epilog);
+            else {
+                DEFAULT;
+            }
+            break;

        case 0x80:
            nextop = F8;
@@ -514,7 +600,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            } else {
                gd = (nextop&0x38)>>3;
                gb1 = xRAX+(gd&3);
-                gb2 = ((gd&4)>>2);
+                gb2 = ((gd&4)<<1);
            }
            if(MODREG) {
                if(rex.rex) {
@@ -538,7 +624,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                LDRB_REG(x4, wback, x4);
                ed = x4;
            }
-            BFIx(gb1, ed, gb2*8, 8);
+            BFIx(gb1, ed, gb2, 8);
            break;
        case 0x89:
            INST_NAME("MOV Seg:Ed, Gd");
@@ -584,7 +670,22 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                }
            }
            break;
-
+        case 0x8E:
+            INST_NAME("MOV Seg,Ew");
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            u8 = (nextop&0x38)>>3;
+            if((nextop&0xC0)==0xC0) {
+                ed = xRAX+(nextop&7)+(rex.b<<3);
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
+                LDRH_REG(x1, wback, x4);
+                ed = x1;
+            }
+            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 FS:Ed");
            grab_segdata(dyn, addr, ninst, x4, seg);
@@ -604,6 +705,9 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                }
            }
            break;
+        case 0x90:
+            INST_NAME("NOP");
+            break;

        case 0xA1:
            INST_NAME("MOV EAX,FS:Od");
@@ -689,27 +793,19 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    break;
                case 2:
                    INST_NAME("RCL Ed, 1");
-                    MESSAGE(LOG_DUMP, "Need Optimization\n");
                    READFLAGS(X_CF);
-                    SETFLAGS(X_OF|X_CF, SF_SET);
-                    MOV32w(x2, 1);
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET); // removed PENDING on purpose
                    GETEDO(x6, 0);
-                    if(wback) {ADDx_REG(x6, x6, wback); wback=x6;}
-                    if(ed!=x1) {MOVxw_REG(x1, ed);}
-                    CALL_(rcl32, ed, x6);
-                    WBACK;
+                    emit_rcl32c(dyn, ninst, rex, ed, 1, x3, x4);
+                    WBACKO(x6);
                    break;
                case 3:
                    INST_NAME("RCR Ed, 1");
-                    MESSAGE(LOG_DUMP, "Need Optimization\n");
                    READFLAGS(X_CF);
-                    SETFLAGS(X_OF|X_CF, SF_SET);
-                    MOV32w(x2, 1);
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET); // removed PENDING on purpose
                    GETEDO(x6, 0);
-                    if(wback) {ADDx_REG(x6, x6, wback); wback=x6;}
-                    if(ed!=x1) {MOVxw_REG(x1, ed);}
-                    CALL_(rcr32, ed, x6);
-                    WBACK;
+                    emit_rcr32c(dyn, ninst, rex, ed, 1, x3, x4);
+                    WBACKO(x6);
                    break;
                case 4:
                case 6:
@@ -794,7 +890,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    INST_NAME("RCL Ed, CL");
                    MESSAGE(LOG_DUMP, "Need Optimization\n");
                    READFLAGS(X_CF);
-                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    SETFLAGS(X_OF|X_CF, SF_SET_DF);
                    if(rex.w) {
                        ANDSx_mask(x2, xRCX, 1, 0, 0b00101);  //mask=0x000000000000003f
                    } else {
@@ -811,7 +907,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    INST_NAME("RCR Ed, CL");
                    MESSAGE(LOG_DUMP, "Need Optimization\n");
                    READFLAGS(X_CF);
-                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    SETFLAGS(X_OF|X_CF, SF_SET_DF);
                    if(rex.w) {
                        ANDSx_mask(x2, xRCX, 1, 0, 0b00101);  //mask=0x000000000000003f
                    } else {
@@ -884,7 +980,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    GETEDO(x6, 4);
                    i64 = F32S;
                    MOV64xw(x2, i64);
-                    emit_test32(dyn, ninst, rex, ed, x2, x3, x4);
+                    emit_test32(dyn, ninst, rex, ed, x2, x3, x4, x5);
                    break;
                case 2:
                    INST_NAME("NOT Ed");
@@ -902,7 +998,6 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                case 4:
                    INST_NAME("MUL EAX, Ed");
                    SETFLAGS(X_ALL, SF_PENDING);
-                    UFLAG_DF(x2, rex.w?d_mul64:d_mul32);
                    GETEDO(x6, 0);
                    if(rex.w) {
                        if(ed==xRDX) gd=x3; else gd=xRDX;
@@ -916,11 +1011,11 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    }
                    UFLAG_RES(xRAX);
                    UFLAG_OP1(xRDX);
+                    UFLAG_DF(x2, rex.w?d_mul64:d_mul32);
                    break;
                case 5:
                    INST_NAME("IMUL EAX, Ed");
                    SETFLAGS(X_ALL, SF_PENDING);
-                    UFLAG_DF(x2, rex.w?d_imul64:d_imul32);
                    GETEDO(x6, 0);
                    if(rex.w) {
                        if(ed==xRDX) gd=x3; else gd=xRDX;
@@ -934,13 +1029,24 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    }
                    UFLAG_RES(xRAX);
                    UFLAG_OP1(xRDX);
+                    UFLAG_DF(x2, rex.w?d_imul64:d_imul32);
                    break;
                case 6:
                    INST_NAME("DIV Ed");
                    SETFLAGS(X_ALL, SF_SET);
+                    SET_DFNONE(x2);
                    if(!rex.w) {
-                        SET_DFNONE(x2);
                        GETEDO(x6, 0);
+                        if(box64_dynarec_div0) {
+                            CBNZx_MARK3(ed);
+                            GETIP_(ip);
+                            STORE_XEMU_CALL(xRIP);
+                            CALL(native_div0, -1);
+                            CLEARIP();
+                            LOAD_XEMU_CALL(xRIP);
+                            jump_to_epilog(dyn, 0, xRIP, ninst);
+                            MARK3;
+                        }
                        MOVw_REG(x3, xRAX);
                        ORRx_REG_LSL(x3, x3, xRDX, 32);
                        if(MODREG) {
@@ -956,8 +1062,17 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                           && dyn->insts[ninst-1].x64.addr
                           && *(uint8_t*)(dyn->insts[ninst-1].x64.addr)==0x31
                           && *(uint8_t*)(dyn->insts[ninst-1].x64.addr+1)==0xD2) {
-                            SET_DFNONE(x2);
                            GETEDO(x6, 0);
+                            if(box64_dynarec_div0) {
+                                CBNZx_MARK3(ed);
+                                GETIP_(ip);
+                                STORE_XEMU_CALL(xRIP);
+                                CALL(native_div0, -1);
+                                CLEARIP();
+                                LOAD_XEMU_CALL(xRIP);
+                                jump_to_epilog(dyn, 0, xRIP, ninst);
+                                MARK3;
+                            }
                            UDIVx(x2, xRAX, ed);
                            MSUBx(xRDX, x2, ed, xRAX);
                            MOVx_REG(xRAX, x2);
@@ -968,10 +1083,19 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                            CALL(div64, -1);
                            B_NEXT_nocond;
                            MARK;
+                            if(box64_dynarec_div0) {
+                                CBNZx_MARK3(ed);
+                                GETIP_(ip);
+                                STORE_XEMU_CALL(xRIP);
+                                CALL(native_div0, -1);
+                                CLEARIP();
+                                LOAD_XEMU_CALL(xRIP);
+                                jump_to_epilog(dyn, 0, xRIP, ninst);
+                                MARK3;
+                            }
                            UDIVx(x2, xRAX, ed);
                            MSUBx(xRDX, x2, ed, xRAX);
                            MOVx_REG(xRAX, x2);
-                            SET_DFNONE(x2);
                        }
                    }
                    break;
@@ -979,11 +1103,21 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    INST_NAME("IDIV Ed");
                    NOTEST(x1);
                    SETFLAGS(X_ALL, SF_SET);
+                    SET_DFNONE(x2)
                    if(!rex.w) {
-                        SET_DFNONE(x2)
                        GETSEDOw(x6, 0);
                        MOVw_REG(x3, xRAX);
                        ORRx_REG_LSL(x3, x3, xRDX, 32);
+                        if(box64_dynarec_div0) {
+                            CBNZx_MARK3(wb);
+                            GETIP_(ip);
+                            STORE_XEMU_CALL(xRIP);
+                            CALL(native_div0, -1);
+                            CLEARIP();
+                            LOAD_XEMU_CALL(xRIP);
+                            jump_to_epilog(dyn, 0, xRIP, ninst);
+                            MARK3;
+                        }
                        SDIVx(x2, x3, wb);
                        MSUBx(x4, x2, wb, x3);
                        MOVw_REG(xRAX, x2);
@@ -993,13 +1127,32 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                           &&  dyn->insts[ninst-1].x64.addr
                           && *(uint8_t*)(dyn->insts[ninst-1].x64.addr)==0x48
                           && *(uint8_t*)(dyn->insts[ninst-1].x64.addr+1)==0x99) {
-                            SET_DFNONE(x2)
                            GETEDO(x6, 0);
+                            if(box64_dynarec_div0) {
+                                CBNZx_MARK3(ed);
+                                GETIP_(ip);
+                                STORE_XEMU_CALL(xRIP);
+                                CALL(native_div0, -1);
+                                CLEARIP();
+                                LOAD_XEMU_CALL(xRIP);
+                                jump_to_epilog(dyn, 0, xRIP, ninst);
+                                MARK3;
+                            }
                            SDIVx(x2, xRAX, ed);
                            MSUBx(xRDX, x2, ed, xRAX);
                            MOVx_REG(xRAX, x2);
                        } else {
                            GETEDO(x6, 0);
+                            if(box64_dynarec_div0) {
+                                CBNZx_MARK3(ed);
+                                GETIP_(ip);
+                                STORE_XEMU_CALL(xRIP);
+                                CALL(native_div0, -1);
+                                CLEARIP();
+                                LOAD_XEMU_CALL(xRIP);
+                                jump_to_epilog(dyn, 0, xRIP, ninst);
+                                MARK3;
+                            }
                            CBZxw_MARK(xRDX);
                            MVNx_REG(x2, xRDX);
                            CBZxw_MARK(x2);
@@ -1010,7 +1163,6 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                            SDIVx(x2, xRAX, ed);
                            MSUBx(xRDX, x2, ed, xRAX);
                            MOVx_REG(xRAX, x2);
-                            SET_DFNONE(x2)
                        }
                    }
                    break;
@@ -1043,11 +1195,12 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    {
                        READFLAGS(X_PEND);          // that's suspicious
                    } else {
-                        SETFLAGS(X_ALL, SF_SET);    //Hack to put flag in "don't care" state
+                        SETFLAGS(X_ALL, SF_SET_NODF);    //Hack to put flag in "don't care" state
                    }
                    GETEDOz(x6, 0);
                    if(box64_dynarec_callret && box64_dynarec_bigblock>1) {
                        BARRIER(BARRIER_FULL);
+                        BARRIER_NEXT(BARRIER_FULL);
                    } else {
                        BARRIER(BARRIER_FLOAT);
                        *need_epilog = 0;
@@ -1055,12 +1208,15 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    }
                    GETIP_(addr);
                    if(box64_dynarec_callret) {
+                        SET_HASCALLRET();
                        // Push actual return address
                        if(addr < (dyn->start+dyn->isize)) {
                            // there is a next...
                            j64 = (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->native_size)):0;
                            ADR_S20(x4, j64);
+                            MESSAGE(LOG_NONE, "\tCALLRET set return to +%di\n", j64>>2);
                        } else {
+                            MESSAGE(LOG_NONE, "\tCALLRET set return to Jmptable(%p)\n", (void*)addr);
                            j64 = getJumpTableAddress64(addr);
                            TABLE64(x4, j64);
                            LDRx_U12(x4, x4, 0);
@@ -1068,14 +1224,14 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                        STPx_S7_preindex(x4, xRIP, xSP, -16);
                    }
                    PUSH1z(xRIP);
-                    jump_to_next(dyn, 0, ed, ninst);
+                    jump_to_next(dyn, 0, ed, ninst, rex.is32bits);
                    break;
                case 4: // JMP Ed
                    INST_NAME("JMP Ed");
                    READFLAGS(X_PEND);
                    BARRIER(BARRIER_FLOAT);
                    GETEDOz(x6, 0);
-                    jump_to_next(dyn, 0, ed, ninst);
+                    jump_to_next(dyn, 0, ed, ninst, rex.is32bits);
                    *need_epilog = 0;
                    *ok = 0;
                    break;
@@ -27,8 +27,9 @@ uintptr_t dynarec64_6664(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n

    uint8_t opcode = F8;
    uint8_t nextop;
-    uint8_t gd, ed;
+    uint8_t gd, ed, wb1, wback;
    int64_t j64;
+    int16_t i16;
    int v0, v1;
    int64_t fixedaddress;
    int unscaled;
@@ -66,27 +67,149 @@ uintptr_t dynarec64_6664(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    FCMPD(v0, v1);
                    FCOMI(x1, x2);
                    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);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
-                    VSTR64_REG(v0, ed, x4);
-                    SMWRITE();
-                }
-                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);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
+                        SMREAD();
+                        VLDR128_REG(v0, ed, x4);
+                    }
+                    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);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
+                        VSTR128_REG(v0, ed, x4);
+                        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);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
+                        VSTR64_REG(v0, ed, x4);
+                        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);
+                    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);
+                    GETEWO(x1, (opcode==0x81)?2:1);
+                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
+                    MOVZw(x5, i16);
+                    emit_or16(dyn, ninst, x1, x5, 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);
+                    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);
+                    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);
+                    GETEWO(x1, (opcode==0x81)?2:1);
+                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
+                    MOVZw(x5, i16);
+                    emit_and16(dyn, ninst, x1, x5, 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);
+                    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);
+                    GETEWO(x1, (opcode==0x81)?2:1);
+                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
+                    MOVZw(x5, i16);
+                    emit_xor16(dyn, ninst, x1, x5, 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);
+                    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;
@@ -48,12 +48,42 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
    GETREX();

    switch(opcode) {
+
+        case 0x01:
+            INST_NAME("LOCK ADD Ew, Gw");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETGW(x5);
+            if(MODREG) {
+                ed = xRAX+(nextop&7)+(rex.b<<3);
+                UXTHw(x6, ed);
+                emit_add16(dyn, ninst, x6, x5, x3, x4);
+                BFIx(ed, x6, 0, 16);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
+                if(arm64_atomics) {
+                    UFLAG_IF {
+                        LDADDALH(x5, x1, wback);
+                        emit_add16(dyn, ninst, x1, x5, x3, x4);
+                    } else {
+                        STADDLH(x5, wback);
+                    }
+                } else {
+                    MARKLOCK;
+                    LDAXRH(x1, wback);
+                    emit_add16(dyn, ninst, x1, x5, x3, x4);
+                    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);
-            SMDMB();
            if(MODREG) {
                ed = xRAX+(nextop&7)+(rex.b<<3);
                UXTHw(x6, ed);
@@ -61,13 +91,22 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                BFIx(ed, x6, 0, 16);
            } else {
                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                MARKLOCK;
-                LDAXRH(x1, wback);
-                emit_or16(dyn, ninst, x1, x5, x3, x4);
-                STLXRH(x3, x1, wback);
-                CBNZx_MARKLOCK(x3);
+                if(arm64_atomics) {
+                    UFLAG_IF {
+                        LDSETALH(x5, x1, wback);
+                        emit_or16(dyn, ninst, x1, x5, x3, x4);
+                    } else {
+                        STSETLH(x5, wback);
+                    }
+                } else {
+                    MARKLOCK;
+                    LDAXRH(x1, wback);
+                    emit_or16(dyn, ninst, x1, x5, x3, x4);
+                    STLXRH(x3, x1, wback);
+                    CBNZx_MARKLOCK(x3);
+                }
+                SMDMB();
            }
-            SMDMB();
            break;

        case 0x0F:
@@ -79,7 +118,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    nextop = F8;
                    GETGD;
-                    SMDMB();
                    UXTHw(x6, xRAX);
                    if(MODREG) {
                        ed = xRAX+(nextop&7)+(rex.b<<3);
@@ -90,34 +128,49 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        BFIx(ed, gd, 0, 16);
                    } else {
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                        TSTx_mask(wback, 1, 0, 0);    // mask=1
-                        B_MARK3(cNE);
+                        if(!ALIGNED_ATOMICH) {
+                            if(arm64_uscat) {
+                                ANDx_mask(x1, wback, 1, 0, 3);  // mask = F
+                                CMPSw_U12(x1, 0xF);
+                                B_MARK3(cGE);
+                            } else {
+                                TSTx_mask(wback, 1, 0, 0);    // mask=1
+                                B_MARK3(cNE);
+                            }
+                        }
                        // Aligned version
-                        MARKLOCK;
-                        LDAXRH(x1, wback);
-                        CMPSw_REG(x6, x1);
-                        B_MARK(cNE);
-                        // EAX == Ed
-                        STLXRH(x4, gd, wback);
-                        CBNZx_MARKLOCK(x4);
-                        // done
-                        B_MARK_nocond;
-                        // Unaligned version
-                        MARK3;
-                        LDRH_U12(x1, wback, 0);
-                        LDAXRB(x3, wback); // dummy read, to arm the write...
-                        CMPSw_REG(x6, x1);
-                        B_MARK(cNE);
-                        // EAX == Ed
-                        STLXRB(x4, gd, wback);
-                        CBNZx_MARK3(x4);
-                        STRH_U12(gd, wback, 0);
+                        if(arm64_atomics) {
+                            MOVw_REG(x1, x6);
+                            CASALH(x1, gd, wback);
+                        } else {
+                            MARKLOCK;
+                            LDAXRH(x1, wback);
+                            CMPSw_REG(x6, x1);
+                            B_MARK(cNE);
+                            // EAX == Ed
+                            STLXRH(x4, gd, wback);
+                            CBNZx_MARKLOCK(x4);
+                            // done
+                        }
+                        if(!ALIGNED_ATOMICH) {
+                            B_MARK_nocond;
+                            // Unaligned version
+                            MARK3;
+                            LDRH_U12(x1, wback, 0);
+                            LDAXRB(x3, wback); // dummy read, to arm the write...
+                            CMPSw_REG(x6, x1);
+                            B_MARK(cNE);
+                            // EAX == Ed
+                            STLXRB(x4, gd, wback);
+                            CBNZx_MARK3(x4);
+                            STRH_U12(gd, wback, 0);
+                        }
                    }
+                    SMDMB();
                    MARK;
                    // Common part (and fallback for EAX != Ed)
                    UFLAG_IF {emit_cmp16(dyn, ninst, x6, x1, x3, x4, x5);}
                    BFIx(xRAX, x1, 0, 16);
-                    SMDMB();
                    break;

                case 0xC1:
@@ -126,7 +179,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    nextop = F8;
                    gd = xRAX+((nextop&0x38)>>3)+(rex.r<<3);
                    UXTHx(x5, gd);
-                    SMDMB();
                    if(MODREG) {
                        ed = xRAX+(nextop&7)+(rex.b<<3);
                        BFIx(gd, ed, 0, 16);
@@ -134,18 +186,22 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        BFIx(ed, x5, 0, 16);
                    } else {
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                        MARKLOCK;
-                        LDAXRH(x1, wback);
-                        ADDxw_REG(x4, x1, x5);
-                        STLXRH(x3, x4, wback);
-                        CBNZx_MARKLOCK(x3);
+                        if(arm64_atomics) {
+                            LDADDALH(x5, x1, wback);
+                        } else {
+                            MARKLOCK;
+                            LDAXRH(x1, wback);
+                            ADDxw_REG(x4, x1, x5);
+                            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);
                        }
                        BFIx(gd, x1, 0, 16);
                    }
-                    SMDMB();
                    break;

                default:
@@ -153,10 +209,62 @@ 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 = xRAX+(nextop&7)+(rex.b<<3);
+                UXTHw(x6, ed);
+                emit_adc16(dyn, ninst, x6, x5, x3, x4);
+                BFIx(ed, x6, 0, 16);
+            } else {
+                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 = xRAX+(nextop&7)+(rex.b<<3);
+                UXTHw(x6, ed);
+                emit_and16(dyn, ninst, x6, gd, x3, x4);
+                BFIx(ed, x6, 0, 16);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
+                if(arm64_atomics) {
+                    MVNw_REG(x3, gd);
+                    UFLAG_IF {
+                        LDCLRALH(x3, x1, wback);
+                        emit_and16(dyn, ninst, x1, gd, x3, x4);
+                    } else {
+                        STCLRLH(x3, wback);
+                    }
+                } else {
+                    MARKLOCK;
+                    LDAXRH(x1, wback);
+                    emit_and16(dyn, ninst, x1, gd, x3, x4);
+                    STLXRH(x3, x1, wback);
+                    CBNZx_MARKLOCK(x3);
+                }
+                SMDMB();
+            }
+            break;
+
        case 0x81:
        case 0x83:
            nextop = F8;
-            SMDMB();
            switch((nextop>>3)&7) {
                case 0: //ADD
                    if(opcode==0x81) {
@@ -176,22 +284,41 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                        if(opcode==0x81) i32 = F16S; else i32 = F8S;
                        MOV32w(x5, i32);
-                        TSTx_mask(wback, 1, 0, 0);    // mask=1
-                        B_MARK(cNE);
-                        MARKLOCK;
-                        LDAXRH(x1, wback);
-                        emit_add16(dyn, ninst, x1, x5, x3, x4);
-                        STLXRH(x3, x1, wback);
-                        CBNZx_MARKLOCK(x3);
-                        B_NEXT_nocond;
-                        MARK;   // unaligned! also, not enough
-                        LDRH_U12(x1, wback, 0);
-                        LDAXRB(x4, wback);
-                        BFIw(x1, x4, 0, 8); // re-inject
-                        emit_add16(dyn, ninst, x1, x5, x3, x4);
-                        STLXRB(x3, x1, wback);
-                        CBNZx_MARK(x3);
-                        STRH_U12(x1, wback, 0);    // put the whole value
+                        if(!ALIGNED_ATOMICH) {
+                            if(arm64_uscat) {
+                                ANDx_mask(x1, wback, 1, 0, 3);  // mask = F
+                                CMPSw_U12(x1, 0xF);
+                                B_MARK(cGE);
+                            } else {
+                                TSTx_mask(wback, 1, 0, 0);    // mask=1
+                                B_MARK(cNE);
+                            }
+                        }
+                        if(arm64_atomics) {
+                            LDADDALH(x5, x1, wback);
+                            UFLAG_IF {
+                                emit_add16(dyn, ninst, x1, x5, x3, x4);    
+                            }
+                        } else {
+                            MARKLOCK;
+                            LDAXRH(x1, wback);
+                            emit_add16(dyn, ninst, x1, x5, x3, x4);
+                            STLXRH(x3, x1, wback);
+                            CBNZx_MARKLOCK(x3);
+                        }
+                        SMDMB();
+                        if(!ALIGNED_ATOMICH) {
+                            B_NEXT_nocond;
+                            MARK;   // unaligned! also, not enough
+                            LDRH_U12(x1, wback, 0);
+                            LDAXRB(x4, wback);
+                            BFIw(x1, x4, 0, 8); // re-inject
+                            emit_add16(dyn, ninst, x1, x5, x3, x4);
+                            STLXRB(x3, x1, wback);
+                            CBNZx_MARK(x3);
+                            STRH_U12(x1, wback, 0);    // put the whole value
+                            SMDMB();
+                        }
                    }
                    break;
                case 1: //OR
@@ -208,11 +335,21 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                        if(opcode==0x81) i32 = F16S; else i32 = F8S;
                        MOV32w(x5, i32);
-                        MARKLOCK;
-                        LDAXRH(x1, wback);
-                        emit_or16(dyn, ninst, x1, x5, x3, x4);
-                        STLXRH(x3, x1, wback);
-                        CBNZx_MARKLOCK(x3);
+                        if(arm64_atomics) {
+                            UFLAG_IF {
+                                LDSETALH(x5, x1, wback);
+                                emit_or16(dyn, ninst, x1, x5, x3, x4);
+                            } else {
+                                STSETLH(x5, wback);
+                            }
+                        } else {
+                            MARKLOCK;
+                            LDAXRH(x1, wback);
+                            emit_or16(dyn, ninst, x1, x5, x3, x4);
+                            STLXRH(x3, x1, wback);
+                            CBNZx_MARKLOCK(x3);
+                        }
+                        SMDMB();
                    }
                    break;
                case 2: //ADC
@@ -235,6 +372,7 @@ 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
@@ -257,6 +395,7 @@ 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
@@ -272,12 +411,24 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    } else {
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                        if(opcode==0x81) i32 = F16S; else i32 = F8S;
-                        MOV32w(x5, i32);
-                        MARKLOCK;
-                        LDAXRH(x1, wback);
-                        emit_and16(dyn, ninst, x1, x5, x3, x4);
-                        STLXRH(x3, x1, wback);
-                        CBNZx_MARKLOCK(x3);
+                        if(arm64_atomics) {
+                            MOV32w(x5, ~i32);
+                            UFLAG_IF {
+                                LDCLRALH(x5, x1, wback);
+                                MVNw_REG(x5, x5);
+                                emit_and16(dyn, ninst, x1, x5, x3, x4);
+                            } else {
+                                STCLRLH(x5, wback);
+                            }
+                        } else {
+                            MOV32w(x5, i32);
+                            MARKLOCK;
+                            LDAXRH(x1, wback);
+                            emit_and16(dyn, ninst, x1, x5, x3, x4);
+                            STLXRH(x3, x1, wback);
+                            CBNZx_MARKLOCK(x3);
+                            SMDMB();
+                        }
                    }
                    break;
                case 5: //SUB
@@ -294,22 +445,44 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                        if(opcode==0x81) i32 = F16S; else i32 = F8S;
                        MOV32w(x5, i32);
-                        TSTx_mask(wback, 1, 0, 0);    // mask=1
-                        B_MARK(cNE);
-                        MARKLOCK;
-                        LDAXRH(x1, wback);
-                        emit_sub16(dyn, ninst, x1, x5, x3, x4);
-                        STLXRH(x3, x1, wback);
-                        CBNZx_MARKLOCK(x3);
-                        B_NEXT_nocond;
-                        MARK;   // unaligned! also, not enough
-                        LDRH_U12(x1, wback, 0);
-                        LDAXRB(x4, wback);
-                        BFIw(x1, x4, 0, 8); // re-inject
-                        emit_sub16(dyn, ninst, x1, x5, x3, x4);
-                        STLXRB(x3, x1, wback);
-                        CBNZx_MARK(x3);
-                        STRH_U12(x1, wback, 0);    // put the whole value
+                        if(!ALIGNED_ATOMICH) {
+                            if(arm64_uscat) {
+                                ANDx_mask(x1, wback, 1, 0, 3);  // mask = F
+                                CMPSw_U12(x1, 0xF);
+                                B_MARK(cGE);
+                            } else {
+                                TSTx_mask(wback, 1, 0, 0);    // mask=1
+                                B_MARK(cNE);
+                            }
+                        }
+                        if(arm64_atomics) {
+                            NEGw_REG(x4, x5);
+                            UFLAG_IF {
+                                LDADDALH(x4, x1, wback);
+                                emit_sub16(dyn, ninst, x1, x5, x3, x4);
+                            } else {
+                                STADDLH(x4, wback);
+                            }
+                        } else {
+                            MARKLOCK;
+                            LDAXRH(x1, wback);
+                            emit_sub16(dyn, ninst, x1, x5, x3, x4);
+                            STLXRH(x3, x1, wback);
+                            CBNZx_MARKLOCK(x3);
+                        }
+                        SMDMB();
+                        if(!ALIGNED_ATOMICH) {
+                            B_NEXT_nocond;
+                            MARK;   // unaligned! also, not enough
+                            LDRH_U12(x1, wback, 0);
+                            LDAXRB(x4, wback);
+                            BFIw(x1, x4, 0, 8); // re-inject
+                            emit_sub16(dyn, ninst, x1, x5, x3, x4);
+                            STLXRB(x3, x1, wback);
+                            CBNZx_MARK(x3);
+                            STRH_U12(x1, wback, 0);    // put the whole value
+                            SMDMB();
+                        }
                    }
                    break;
                case 6: //XOR
@@ -326,11 +499,21 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                        if(opcode==0x81) i32 = F16S; else i32 = F8S;
                        MOV32w(x5, i32);
-                        MARKLOCK;
-                        LDAXRH(x1, wback);
-                        emit_xor16(dyn, ninst, x1, x5, x3, x4);
-                        STLXRH(x3, x1, wback);
-                        CBNZx_MARKLOCK(x3);
+                        if(arm64_atomics) {
+                            UFLAG_IF {
+                                LDEORALH(x5, x1, wback);
+                                emit_xor16(dyn, ninst, x1, x5, x3, x4);
+                            } else {
+                                STEORLH(x5, wback);
+                            }
+                        } else {
+                            MARKLOCK;
+                            LDAXRH(x1, wback);
+                            emit_xor16(dyn, ninst, x1, x5, x3, x4);
+                            STLXRH(x3, x1, wback);
+                            CBNZx_MARKLOCK(x3);
+                            SMDMB();
+                        }
                    }
                    break;
                case 7: //CMP
@@ -348,7 +531,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    }
                    break;
            }
-            SMDMB();
            break;

            case 0xFF:
@@ -358,7 +540,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                    case 0: // INC Ew
                        INST_NAME("LOCK INC Ew");
                        SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
-                        SMDMB();
                        if(MODREG) {
                            ed = xRAX+(nextop&7)+(rex.b<<3);
                            UXTHw(x6, ed);
@@ -366,18 +547,27 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                            BFIx(ed, x6, 0, 16);
                        } else {
                            addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                            MARKLOCK;
-                            LDAXRH(x1, wback);
-                            emit_inc16(dyn, ninst, x1, x3, x4);
-                            STLXRH(x3, x1, wback);
-                            CBNZx_MARKLOCK(x3);
+                            if(arm64_atomics) {
+                                MOV32w(x3, 1);
+                                UFLAG_IF {
+                                    LDADDALH(x3, x1, wback);
+                                    emit_inc16(dyn, ninst, x1, x3, x4);
+                                } else {
+                                    STADDLH(x3, wback);
+                                }
+                            } else {
+                                MARKLOCK;
+                                LDAXRH(x1, wback);
+                                emit_inc16(dyn, ninst, x1, x3, x4);
+                                STLXRH(x3, x1, wback);
+                                CBNZx_MARKLOCK(x3);
+                            }
+                            SMDMB();
                        }
-                        SMDMB();
                        break;
                    case 1: //DEC Ew
                        INST_NAME("LOCK DEC Ew");
                        SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
-                        SMDMB();
                        if(MODREG) {
                            ed = xRAX+(nextop&7)+(rex.b<<3);
                            UXTHw(x6, ed);
@@ -385,13 +575,23 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                            BFIx(ed, x6, 0, 16);
                        } else {
                            addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                            MARKLOCK;
-                            LDAXRH(x1, wback);
-                            emit_dec16(dyn, ninst, x1, x3, x4);
-                            STLXRH(x3, x1, wback);
-                            CBNZx_MARKLOCK(x3);
+                            if(arm64_atomics) {
+                                MOV32w(x3, -1);
+                                UFLAG_IF {
+                                    LDADDALH(x3, x1, wback);
+                                    emit_dec16(dyn, ninst, x1, x3, x4);
+                                } else {
+                                    STADDLH(x3, wback);
+                                }
+                            } else {
+                                MARKLOCK;
+                                LDAXRH(x1, wback);
+                                emit_dec16(dyn, ninst, x1, x3, x4);
+                                STLXRH(x3, x1, wback);
+                                CBNZx_MARKLOCK(x3);
+                            }
+                            SMDMB();
                        }
-                        SMDMB();
                        break;
                    default:
                        DEFAULT;
@@ -0,0 +1,93 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "emu/x64run_private.h"
+#include "x64run.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_arm64_helper.h"
+#include "emu/x64compstrings.h"
+
+uintptr_t dynarec64_66F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
+{
+    (void)ip; (void)need_epilog;
+
+    uint8_t opcode = F8;
+    uint8_t nextop, u8;
+    int32_t i32;
+    uint8_t gd, ed;
+    uint8_t wback, wb1, wb2;
+    uint8_t eb1, eb2;
+    int64_t j64;
+    uint64_t tmp64u, tmp64u2;
+    int v0, v1;
+    int q0, q1;
+    int d0, d1;
+    int64_t fixedaddress;
+    int unscaled;
+
+    MAYUSE(d0);
+    MAYUSE(d1);
+    MAYUSE(q0);
+    MAYUSE(q1);
+    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 0x38:  // SSE 4.x
+            opcode = F8;
+            switch(opcode) {
+
+                case 0xF1:
+                    INST_NAME("CRC32 Gd, Ew");
+                    nextop = F8;
+                    GETEW(x1, 0);
+                    GETGD;
+                    if(arm64_crc32) {
+                        CRC32CH(gd, gd, ed);
+                    } else {
+                        MOV32w(x2, 0x82f63b78);
+                        for(int j=0; j<2; ++j) {
+                            UBFXxw(x3, ed, 8*j, 8);
+                            EORw_REG(gd, gd, x3);
+                            for(int i=0; i<8; ++i) {
+                                LSRw_IMM((i&1)?gd:x4, (i&1)?x4:gd, 1);
+                                TBZ((i&1)?x4:gd, 0, 4+4);
+                                EORw_REG((i&1)?gd:x4, (i&1)?gd:x4, x2);
+                            }
+                        }
+                    }
+                    break;
+
+                default:
+                    DEFAULT;
+            }
+            break;
+
+        default:
+            DEFAULT;
+    }
+    return addr;
+}
@@ -0,0 +1,64 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "emu/x64run_private.h"
+#include "x64run.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_arm64_helper.h"
+#include "emu/x64compstrings.h"
+
+uintptr_t dynarec64_66F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
+{
+    (void)ip; (void)need_epilog;
+
+    uint8_t opcode = F8;
+    uint8_t nextop, u8;
+    int32_t i32;
+    uint8_t gd, ed;
+    uint8_t wback, wb1, wb2;
+    uint8_t eb1, eb2;
+    int64_t j64;
+    uint64_t tmp64u, tmp64u2;
+    int v0, v1;
+    int q0, q1;
+    int d0, d1;
+    int64_t fixedaddress;
+    int unscaled;
+
+    MAYUSE(d0);
+    MAYUSE(d1);
+    MAYUSE(q0);
+    MAYUSE(q1);
+    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) {
+
+
+        default:
+            DEFAULT;
+    }
+    return addr;
+}
@@ -15,6 +15,7 @@
 #include "emu/x64run_private.h"
 #include "x64trace.h"
 #include "dynarec_native.h"
+#include "custommem.h"

 #include "arm64_printer.h"
 #include "dynarec_arm64_private.h"
@@ -166,6 +167,28 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            opcode=F8;
            switch(opcode) {

+                case 0x11:
+                    switch(rep) {
+                        case 0:
+                            INST_NAME("MOVUPS 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 = geted32(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                                VST128(v0, ed, fixedaddress);
+                                SMWRITE2();
+                            }
+                            break;
+                        default:
+                            DEFAULT;
+                    }
+                    break;
+
                case 0x2E:
                    // no special check...
                case 0x2F:
@@ -208,6 +231,20 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                                VLD64(v0, ed, fixedaddress);
                            }
                            break;
+                        case 2:
+                            INST_NAME("MOVDQU 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);
+                                SMREAD();
+                                addr = geted32(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                                VLD128(v0, ed, fixedaddress);
+                            }
+                            break;
                        default:
                            DEFAULT;
                    }
@@ -240,7 +277,7 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    if(MODREG) {
                        if(rex.rex) {
                            eb1 = xRAX+(nextop&7)+(rex.b<<3);
-                            eb2 = 0;                \
+                            eb2 = 0;
                        } else {
                            ed = (nextop&7);
                            eb1 = xRAX+(ed&3);  // Ax, Cx, Dx or Bx
@@ -616,6 +653,7 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin

        case 0x66:
            opcode=F8;
+            GETREX();
            switch(opcode) {

                case 0x89:
@@ -629,16 +667,66 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                        }
                    } else {
                        addr = geted32(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, &lock, 0, 0);
+                        WILLWRITELOCK(lock);
                        STH(gd, ed, fixedaddress);
                        SMWRITELOCK(lock);
                    }
                    break;

+                case 0x8D:
+                    INST_NAME("LEA Gw, Ed");
+                    nextop=F8;
+                    //GETGW(x1);
+                    gd = x1;    // no need to read
+                    if(MODREG) {   // reg <= reg? that's an invalid operation
+                        DEFAULT;
+                    } else {                    // mem <= reg
+                        addr = geted32(dyn, addr, ninst, nextop, &ed, gd, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
+                        if(ed!=gd)
+                            gd = ed;
+                    }
+                    GWBACK;
+                    break;
+
                default:
                    DEFAULT;
            }
            break;

+        case 0x6B:
+            INST_NAME("IMUL Gd, Ed, Ib");
+            SETFLAGS(X_ALL, SF_PENDING);
+            nextop = F8;
+            GETGD;
+            GETED32(1);
+            i64 = F8S;
+            MOV64xw(x4, i64);
+            if(rex.w) {
+                // 64bits imul
+                UFLAG_IF {
+                    SMULH(x3, ed, x4);
+                    MULx(gd, ed, x4);
+                    UFLAG_OP1(x3);
+                    UFLAG_RES(gd);
+                    UFLAG_DF(x3, d_imul64);
+                } else {
+                    MULxw(gd, ed, x4);
+                }
+            } else {
+                // 32bits imul
+                UFLAG_IF {
+                    SMULL(gd, ed, x4);
+                    UFLAG_RES(gd);
+                    LSRx(x3, gd, 32);
+                    UFLAG_OP1(x3);
+                    UFLAG_DF(x3, d_imul32);
+                    MOVw_REG(gd, gd);
+                } else {
+                    MULxw(gd, ed, x4);
+                }
+            }
+            break;
+
        case 0x81:
        case 0x83:
            nextop = F8;
@@ -725,12 +813,12 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                gb2 = 0;
                gb1 = xRAX + gd;
            } else {
-                gb2 = ((gd&4)>>2);
+                gb2 = ((gd&4)<<1);
                gb1 = xRAX+(gd&3);
            }
            if(gb2) {
                gd = x4;
-                UBFXw(gd, gb1, gb2*8, 8);
+                UBFXw(gd, gb1, gb2, 8);
            } else {
                gd = gb1;   // no need to extract
            }
@@ -746,6 +834,7 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                BFIx(eb1, gd, eb2*8, 8);
            } else {
                addr = geted32(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff, 0, rex, &lock, 0, 0);
+                WILLWRITELOCK(lock);
                STB(gd, ed, fixedaddress);
                SMWRITELOCK(lock);
            }
@@ -758,11 +847,45 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                MOVxw_REG(xRAX+(nextop&7)+(rex.b<<3), gd);
            } else {                    // mem <= reg
                addr = geted32(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<(2+rex.w), (1<<(2+rex.w))-1, rex, &lock, 0, 0);
+                WILLWRITELOCK(lock);
                STxw(gd, ed, fixedaddress);
                SMWRITELOCK(lock);
            }
            break;
-
+        case 0x8A:
+            INST_NAME("MOV Gb, Eb");
+            nextop = F8;
+            if(rex.rex) {
+                gb1 = gd = xRAX+((nextop&0x38)>>3)+(rex.r<<3);
+                gb2=0;
+            } else {
+                gd = (nextop&0x38)>>3;
+                gb1 = xRAX+(gd&3);
+                gb2 = ((gd&4)<<1);
+            }
+            if(MODREG) {
+                if(rex.rex) {
+                    wback = xRAX+(nextop&7)+(rex.b<<3);
+                    wb2 = 0;
+                } else {
+                    wback = (nextop&7);
+                    wb2 = (wback>>2);
+                    wback = xRAX+(wback&3);
+                }
+                if(wb2) {
+                    UBFXw(x4, wback, wb2*8, 8);
+                    ed = x4;
+                } else {
+                    ed = wback;
+                }
+            } else {
+                addr = geted32(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff, 0, rex, &lock, 0, 0);
+                SMREADLOCK(lock);
+                LDB(x4, wback, fixedaddress);
+                ed = x4;
+            }
+            BFIx(gb1, ed, gb2, 8);
+            break;
        case 0x8B:
            INST_NAME("MOV Gd, Ed");
            nextop=F8;
@@ -811,25 +934,21 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    break;
                case 2:
                    INST_NAME("RCL Ed, Ib");
-                    MESSAGE(LOG_DUMP, "Need Optimization\n");
                    READFLAGS(X_CF);
-                    SETFLAGS(X_OF|X_CF, SF_SET);
-                    GETED32W(x4, x1, 1);
-                    u8 = F8;
-                    MOV32w(x2, u8);
-                    CALL_(rex.w?((void*)rcl64):((void*)rcl32), ed, x4);
-                    WBACK;
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET); // removed PENDING on purpose
+                    GETED32(1);
+                    u8 = (F8)&(rex.w?0x3f:0x1f);
+                    emit_rcl32c(dyn, ninst, rex, ed, u8, x3, x4);
+                    if(u8) { WBACK; }
                    break;
                case 3:
                    INST_NAME("RCR Ed, Ib");
-                    MESSAGE(LOG_DUMP, "Need Optimization\n");
                    READFLAGS(X_CF);
-                    SETFLAGS(X_OF|X_CF, SF_SET);
-                    GETED32W(x4, x1, 1);
-                    u8 = F8;
-                    MOV32w(x2, u8);
-                    CALL_(rex.w?((void*)rcr64):((void*)rcr32), ed, x4);
-                    WBACK;
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET); // removed PENDING on purpose
+                    GETED32(1);
+                    u8 = (F8)&(rex.w?0x3f:0x1f);
+                    emit_rcr32c(dyn, ninst, rex, ed, u8, x3, x4);
+                    if(u8) { WBACK; }
                    break;
                case 4:
                case 6:
@@ -863,6 +982,34 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            }
            break;

+        case 0xC6:
+            INST_NAME("MOV Eb, Ib");
+            nextop=F8;
+            if(MODREG) {   // reg <= u8
+                u8 = F8;
+                if(!rex.rex) {
+                    ed = (nextop&7);
+                    eb1 = xRAX+(ed&3);  // Ax, Cx, Dx or Bx
+                    eb2 = (ed&4)>>2;    // L or H
+                } else {
+                    eb1 = xRAX+(nextop&7)+(rex.b<<3);
+                    eb2 = 0;
+                }
+                MOV32w(x3, u8);
+                BFIx(eb1, x3, eb2*8, 8);
+            } else {                    // mem <= u8
+                addr = geted32(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, &unscaled, 0xfff, 0, rex, &lock, 0, 1);
+                u8 = F8;
+                if(u8) {
+                    MOV32w(x3, u8);
+                    ed = x3;
+                } else
+                    ed = xZR;
+                WILLWRITELOCK(lock);
+                STB(ed, wback, fixedaddress);
+                SMWRITELOCK(lock);
+            }
+            break;
        case 0xC7:
            INST_NAME("MOV Ed, Id");
            nextop=F8;
@@ -874,6 +1021,7 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                addr = geted32(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<(2+rex.w), (1<<(2+rex.w))-1, rex, &lock, 0, 4);
                i64 = F32S;
                MOV64xw(x3, i64);
+                WILLWRITELOCK(lock);
                STxw(x3, ed, fixedaddress);
                SMWRITELOCK(lock);
            }
@@ -891,7 +1039,7 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                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);               \
+                    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);\
@@ -954,7 +1102,7 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    GETED32H(x1, 4);
                    i64 = F32S;
                    MOV64xw(x2, i64);
-                    emit_test32(dyn, ninst, rex, ed, x2, x3, x4);
+                    emit_test32(dyn, ninst, rex, ed, x2, x3, x4, x5);
                    break;
                case 2:
                    INST_NAME("NOT Ed");
@@ -972,7 +1120,6 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                case 4:
                    INST_NAME("MUL EAX, Ed");
                    SETFLAGS(X_ALL, SF_PENDING);
-                    UFLAG_DF(x2, rex.w?d_mul64:d_mul32);
                    GETED32(0);
                    if(rex.w) {
                        if(ed==xRDX) gd=x3; else gd=xRDX;
@@ -986,11 +1133,11 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    }
                    UFLAG_RES(xRAX);
                    UFLAG_OP1(xRDX);
+                    UFLAG_DF(x2, rex.w?d_mul64:d_mul32);
                    break;
                case 5:
                    INST_NAME("IMUL EAX, Ed");
                    SETFLAGS(X_ALL, SF_PENDING);
-                    UFLAG_DF(x2, rex.w?d_imul64:d_imul32);
                    GETED32(0);
                    if(rex.w) {
                        if(ed==xRDX) gd=x3; else gd=xRDX;
@@ -1004,12 +1151,13 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    }
                    UFLAG_RES(xRAX);
                    UFLAG_OP1(xRDX);
+                    UFLAG_DF(x2, rex.w?d_imul64:d_imul32);
                    break;
                case 6:
                    INST_NAME("DIV Ed");
                    SETFLAGS(X_ALL, SF_SET);
+                    SET_DFNONE(x2);
                    if(!rex.w) {
-                        SET_DFNONE(x2);
                        GETED32(0);
                        MOVw_REG(x3, xRAX);
                        ORRx_REG_LSL(x3, x3, xRDX, 32);
@@ -1026,7 +1174,6 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                           && dyn->insts[ninst-1].x64.addr
                           && *(uint8_t*)(dyn->insts[ninst-1].x64.addr)==0x31
                           && *(uint8_t*)(dyn->insts[ninst-1].x64.addr+1)==0xD2) {
-                            SET_DFNONE(x2);
                            GETED32(0);
                            UDIVx(x2, xRAX, ed);
                            MSUBx(xRDX, x2, ed, xRAX);
@@ -1041,7 +1188,6 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                            UDIVx(x2, xRAX, ed);
                            MSUBx(xRDX, x2, ed, xRAX);
                            MOVx_REG(xRAX, x2);
-                            SET_DFNONE(x2);
                        }
                    }
                    break;
@@ -1049,8 +1195,8 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    INST_NAME("IDIV Ed");
                    NOTEST(x1);
                    SETFLAGS(X_ALL, SF_SET);
+                    SET_DFNONE(x2);
                    if(!rex.w) {
-                        SET_DFNONE(x2)
                        GETSED32w(0);
                        MOVw_REG(x3, xRAX);
                        ORRx_REG_LSL(x3, x3, xRDX, 32);
@@ -1063,7 +1209,6 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                           &&  dyn->insts[ninst-1].x64.addr
                           && *(uint8_t*)(dyn->insts[ninst-1].x64.addr)==0x48
                           && *(uint8_t*)(dyn->insts[ninst-1].x64.addr+1)==0x99) {
-                            SET_DFNONE(x2)
                            GETED32(0);
                            SDIVx(x2, xRAX, ed);
                            MSUBx(xRDX, x2, ed, xRAX);
@@ -1080,12 +1225,58 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                            SDIVx(x2, xRAX, ed);
                            MSUBx(xRDX, x2, ed, xRAX);
                            MOVx_REG(xRAX, x2);
-                            SET_DFNONE(x2)
                        }
                    }
                    break;
            }
            break;
+        case 0xFF:
+            nextop = F8;
+            switch((nextop>>3)&7) {
+                case 2: // CALL Ed
+                    INST_NAME("CALL Ed");
+                    PASS2IF((box64_dynarec_safeflags>1) ||
+                        ((ninst && dyn->insts[ninst-1].x64.set_flags)
+                        || ((ninst>1) && dyn->insts[ninst-2].x64.set_flags)), 1)
+                    {
+                        READFLAGS(X_PEND);          // that's suspicious
+                    } else {
+                        SETFLAGS(X_ALL, SF_SET_NODF);    //Hack to put flag in "don't care" state
+                    }
+                    GETED32(0);
+                    if(box64_dynarec_callret && box64_dynarec_bigblock>1) {
+                        BARRIER(BARRIER_FULL);
+                        BARRIER_NEXT(BARRIER_FULL);
+                    } else {
+                        BARRIER(BARRIER_FLOAT);
+                        *need_epilog = 0;
+                        *ok = 0;
+                    }
+                    GETIP_(addr);
+                    if(box64_dynarec_callret) {
+                        SET_HASCALLRET();
+                        // Push actual return address
+                        if(addr < (dyn->start+dyn->isize)) {
+                            // there is a next...
+                            j64 = (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->native_size)):0;
+                            ADR_S20(x4, j64);
+                            MESSAGE(LOG_NONE, "\tCALLRET set return to +%di\n", j64>>2);
+                        } else {
+                            MESSAGE(LOG_NONE, "\tCALLRET set return to Jmptable(%p)\n", (void*)addr);
+                            j64 = getJumpTableAddress64(addr);
+                            TABLE64(x4, j64);
+                            LDRx_U12(x4, x4, 0);
+                        }
+                        STPx_S7_preindex(x4, xRIP, xSP, -16);
+                    }
+                    PUSH1z(xRIP);
+                    jump_to_next(dyn, 0, ed, ninst, rex.is32bits);
+                    break;
+                default:
+                    DEFAULT;
+                    return addr;
+            }
+            break;

        default:
            DEFAULT;
@@ -0,0 +1,157 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "emu/x64run_private.h"
+#include "x64run.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_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(xRAX+(nextop&7)+(rex.b<<3), gd);
+            } else {                    // mem <= reg
+                grab_segdata(dyn, addr, ninst, x4, seg);
+                addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, NULL, 0, 0, 0);
+                STRw_REG(gd, ed, x4);
+            }
+            break;
+
+        case 0x8B:
+            INST_NAME("MOV Gd, Seg:Ed");
+            nextop=F8;
+            GETGD;
+            if(MODREG) {   // reg => reg
+                MOVxw_REG(gd, xRAX+(nextop&7)+(rex.b<<3));
+            } else {                    // mem => reg
+                grab_segdata(dyn, addr, ninst, x4, seg);
+                addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, NULL, 0, 0, 0);
+                LDRw_REG(gd, ed, x4);
+            }
+            break;
+
+        case 0x8F:
+            INST_NAME("POP Seg:Ed");
+            nextop=F8;
+            if(MODREG) {   // reg <= reg
+                POP1_32(x1);
+                MOVxw_REG(xRAX+(nextop&7)+(rex.b<<3), x1);
+            } else {                    // mem <= reg
+                grab_segdata(dyn, addr, ninst, x4, seg);
+                POP1_32(x1);
+                addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, NULL, 0, 0, 0);
+                STRw_REG(x1, ed, x4);
+            }
+            break;
+
+        case 0xA1:
+            INST_NAME("MOV EAX, Seg:[Od]");
+            i32 = F16;
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            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);
+            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);
+            switch((nextop>>3)&7) {
+                case 6: // Push Ed
+                    INST_NAME("PUSH FS:Ew");
+                    if((nextop&0xC0)==0xC0) {   // reg
+                        DEFAULT;
+                    } else {                    // mem <= i32
+                        SMREAD();
+                        addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0, 0, 0);
+                        LDRw_REG(x3, ed, x4);
+                        PUSH1_32(x3);
+                    }
+                    break;
+                default:
+                    DEFAULT;
+            }
+            break;
+
+
+        default:
+            DEFAULT;
+    }
+    return addr;
+}
@@ -0,0 +1,132 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "emu/x64run_private.h"
+#include "x64run.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_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 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);\
+                    B(i32);                                             \
+                }                                                       \
+            } else {    \
+                /* inside the block */  \
+                i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->native_size);    \
+                Bcond(YES, i32);    \
+            }
+        case 0xE0:
+            INST_NAME("LOOPNZ (16bits)");
+            READFLAGS(X_ZF);
+            i8 = F8S;
+            UXTHw(x1, xRCX);
+            SUBSw_U12(x1, x1, 1);
+            BFIx(xRCX, x1, 0, 16);
+            B_NEXT(cEQ);    // 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);
+            SUBSw_U12(x1, x1, 1);
+            BFIx(xRCX, x1, 0, 16);
+            B_NEXT(cEQ);    // 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;
+}
@@ -38,6 +38,7 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
    MAYUSE(v2);
    MAYUSE(v1);

+    if(MODREG)
    switch(nextop) {
        case 0xC0:
        case 0xC1:
@@ -108,7 +109,7 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                FCMPD(v1, v2);
            }
            FCOM(x1, x2, x3);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xE0:
        case 0xE1:
@@ -178,119 +179,120 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                FDIVD(v1, v2, v1);
            }
            break;
-      
        default:
-            switch((nextop>>3)&7) {
-                case 0:
-                    INST_NAME("FADD ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
-                    s0 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
-                    VLD32(s0, ed, fixedaddress);
-                    if(ST_IS_F(0)) {
-                        FADDS(v1, v1, s0);
-                    } else {
-                        FCVT_D_S(s0, s0);
-                        FADDD(v1, v1, s0);
-                    }
-                    break;
-                case 1:
-                    INST_NAME("FMUL ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
-                    s0 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
-                    VLD32(s0, ed, fixedaddress);
-                    if(ST_IS_F(0)) {
-                        FMULS(v1, v1, s0);
-                    } else {
-                        FCVT_D_S(s0, s0);
-                        FMULD(v1, v1, s0);
-                    }
-                    break;
-                case 2:
-                    INST_NAME("FCOM ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
-                    s0 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
-                    VLD32(s0, ed, fixedaddress);
-                    if(ST_IS_F(0)) {
-                        FCMPS(v1, s0);
-                    } else {
-                        FCVT_D_S(s0, s0);
-                        FCMPD(v1, s0);
-                    }
-                    FCOM(x1, x2, x3);
-                    break;
-                case 3:
-                    INST_NAME("FCOMP ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
-                    s0 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
-                    VLD32(s0, ed, fixedaddress);
-                    if(ST_IS_F(0)) {
-                        FCMPS(v1, s0);
-                    } else {
-                        FCVT_D_S(s0, s0);
-                        FCMPD(v1, s0);
-                    }
-                    FCOM(x1, x2, x3);
-                    x87_do_pop(dyn, ninst, x3);
-                    break;
-                case 4:
-                    INST_NAME("FSUB ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
-                    s0 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
-                    VLD32(s0, ed, fixedaddress);
-                    if(ST_IS_F(0)) {
-                        FSUBS(v1, v1, s0);
-                    } else {
-                        FCVT_D_S(s0, s0);
-                        FSUBD(v1, v1, s0);
-                    }
-                    break;
-                case 5:
-                    INST_NAME("FSUBR ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
-                    s0 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
-                    VLD32(s0, ed, fixedaddress);
-                    if(ST_IS_F(0)) {
-                        FSUBS(v1, s0, v1);
-                    } else {
-                        FCVT_D_S(s0, s0);
-                        FSUBD(v1, s0, v1);
-                    }
-                    break;
-                case 6:
-                    INST_NAME("FDIV ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
-                    s0 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
-                    VLD32(s0, ed, fixedaddress);
-                    if(ST_IS_F(0)) {
-                        FDIVS(v1, v1, s0);
-                    } else {
-                        FCVT_D_S(s0, s0);
-                        FDIVD(v1, v1, s0);
-                    }
-                    break;
-                case 7:
-                    INST_NAME("FDIVR ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
-                    s0 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
-                    VLD32(s0, ed, fixedaddress);
-                    if(ST_IS_F(0)) {
-                        FDIVS(v1, s0, v1);
-                    } else {
-                        FCVT_D_S(s0, s0);
-                        FDIVD(v1, s0, v1);
-                    }
-                    break;
-                default:
-                    DEFAULT;
-            }
-    }
+            DEFAULT;
+            break;
+    } else
+        switch((nextop>>3)&7) {
+            case 0:
+                INST_NAME("FADD ST0, float[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+                s0 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(s0, ed, fixedaddress);
+                if(ST_IS_F(0)) {
+                    FADDS(v1, v1, s0);
+                } else {
+                    FCVT_D_S(s0, s0);
+                    FADDD(v1, v1, s0);
+                }
+                break;
+            case 1:
+                INST_NAME("FMUL ST0, float[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+                s0 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(s0, ed, fixedaddress);
+                if(ST_IS_F(0)) {
+                    FMULS(v1, v1, s0);
+                } else {
+                    FCVT_D_S(s0, s0);
+                    FMULD(v1, v1, s0);
+                }
+                break;
+            case 2:
+                INST_NAME("FCOM ST0, float[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+                s0 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(s0, ed, fixedaddress);
+                if(ST_IS_F(0)) {
+                    FCMPS(v1, s0);
+                } else {
+                    FCVT_D_S(s0, s0);
+                    FCMPD(v1, s0);
+                }
+                FCOM(x1, x2, x3);
+                break;
+            case 3:
+                INST_NAME("FCOMP ST0, float[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+                s0 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(s0, ed, fixedaddress);
+                if(ST_IS_F(0)) {
+                    FCMPS(v1, s0);
+                } else {
+                    FCVT_D_S(s0, s0);
+                    FCMPD(v1, s0);
+                }
+                FCOM(x1, x2, x3);
+                X87_POP_OR_FAIL(dyn, ninst, x3);
+                break;
+            case 4:
+                INST_NAME("FSUB ST0, float[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+                s0 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(s0, ed, fixedaddress);
+                if(ST_IS_F(0)) {
+                    FSUBS(v1, v1, s0);
+                } else {
+                    FCVT_D_S(s0, s0);
+                    FSUBD(v1, v1, s0);
+                }
+                break;
+            case 5:
+                INST_NAME("FSUBR ST0, float[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+                s0 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(s0, ed, fixedaddress);
+                if(ST_IS_F(0)) {
+                    FSUBS(v1, s0, v1);
+                } else {
+                    FCVT_D_S(s0, s0);
+                    FSUBD(v1, s0, v1);
+                }
+                break;
+            case 6:
+                INST_NAME("FDIV ST0, float[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+                s0 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(s0, ed, fixedaddress);
+                if(ST_IS_F(0)) {
+                    FDIVS(v1, v1, s0);
+                } else {
+                    FCVT_D_S(s0, s0);
+                    FDIVD(v1, v1, s0);
+                }
+                break;
+            case 7:
+                INST_NAME("FDIVR ST0, float[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+                s0 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(s0, ed, fixedaddress);
+                if(ST_IS_F(0)) {
+                    FDIVS(v1, s0, v1);
+                } else {
+                    FCVT_D_S(s0, s0);
+                    FDIVD(v1, s0, v1);
+                }
+                break;
+            default:
+                DEFAULT;
+        }
    return addr;
 }
@@ -31,7 +31,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
    uint8_t ed;
    uint8_t wback, wb1;
    uint8_t u8;
-    int64_t fixedaddress;
+    int64_t fixedaddress, j64;
    int unscaled;
    int v1, v2;
    int s0;
@@ -41,6 +41,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
    MAYUSE(v2);
    MAYUSE(v1);

+    if(MODREG)
    switch(nextop) {
        case 0xC0:
        case 0xC1:
@@ -51,7 +52,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
        case 0xC6:
        case 0xC7:
            INST_NAME("FLD STx");
-            v2 = x87_do_push(dyn, ninst, x1, X87_ST(nextop&7));
+            X87_PUSH_OR_FAIL(v2, dyn, ninst, x1, X87_ST(nextop&7));
            v1 = x87_get_st(dyn, ninst, x1, x2, (nextop&7)+1, X87_COMBINE(0, (nextop&7)+1));
            if(ST_IS_F(0)) {
                FMOVS(v2, v1);
@@ -84,7 +85,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin

        case 0xD8:
            INST_NAME("FSTPNCE ST0, ST0");
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xD9:
        case 0xDA:
@@ -98,7 +99,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            x87_get_st_empty(dyn, ninst, x1, x2, nextop&7, X87_ST(nextop&7));
            x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
            x87_swapreg(dyn, ninst, x1, x2, 0, nextop&7);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xE0:
            INST_NAME("FCHS");
@@ -131,98 +132,224 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            break;
        case 0xE5:
            INST_NAME("FXAM");
+            #if 1
+            i1 = x87_get_current_cache(dyn, ninst, 0, NEON_CACHE_ST_D);
+            // value put in x14
+            if(i1==-1) {
+                if(fpu_is_st_freed(dyn, ninst, 0)) {
+                    MOV32w(x4, 0b100000100000000);
+                    B_MARK3_nocond;
+                } else {
+                    // not in cache, so check Empty status and load it
+                    i2 = -dyn->n.x87stack;
+                    LDRw_U12(x3, xEmu, offsetof(x64emu_t, fpu_stack));
+                    if(i2) {
+                        if(i2<0) {
+                            ADDw_U12(x3, x3, -i2);
+                        } else {
+                            SUBw_U12(x3, x3, i2);
+                        }
+                    }
+                    CMPSw_U12(x3, 0);
+                    MOV32w(x4, 0b100000100000000);  // empty: C3,C2,C0 = 101
+                    B_MARK3(cLE);
+                    // x5 will be the actual top
+                    LDRw_U12(x5, xEmu, offsetof(x64emu_t, top));
+                    if(i2) {
+                        if(i2<0) {
+                            SUBw_U12(x5, x5, -i2);
+                        } else {
+                            ADDw_U12(x5, x5, i2);
+                        }
+                        ANDw_mask(x5, x5, 0, 3);    // (emu->top + i)&7
+                    }
+                    // load tag
+                    LDRH_U12(x3, xEmu, offsetof(x64emu_t, fpu_tags));
+                    TSTw_mask(x3, 0, 1);    // 0b11
+                    B_MARK3(cNE);   // empty: C3,C2,C0 = 101
+                    // load x2 with ST0 anyway, for sign extraction
+                    ADDx_REG_LSL(x1, xEmu, x5, 3);
+                    LDRx_U12(x2, x1, offsetof(x64emu_t, x87));
+                }
+            } else {
+                // simply move from cache reg to x2
+                v1 = dyn->n.x87reg[i1];
+                VMOVQDto(x2, v1, 0);
+            }
+            // get exponant in x1
+            LSRx_IMM(x1, x2, 20+32);
+            ANDSx_mask(x1, x1, 1, 0b00000, 0b001010); // 0x7ff
+            B_MARK(cNE); // not zero or denormal
+            ANDSx_mask(x1, x2, 1, 0, 0b111110); // 0x7fffffffffffffff
+            MOV32w(x4, 0b100000000000000); // Zero: C3,C2,C0 = 100
+            MOV32w(x5, 0b100010000000000); // Denormal: C3,C2,C0 = 110
+            CSELx(x4, x4, x5, cEQ);
+            B_MARK3(c__);
+            MARK;
+            CMPSx_U12(x1, 0x7ff);   // infinite/NaN?
+            MOV32w(x5, 0b000010000000000); // normal: C3,C2,C0 = 010
+            CSELx(x4, x5, x4, cNE);
+            B_MARK3(cNE);
+            TSTx_mask(x2, 1, 0b000000, 0b110011); // 0x000fffffffffffff
+            MOV32w(x4, 0b000010100000000); // infinity: C3,C2,C0 = 011
+            MOV32w(x3, 0b000000100000000); // NaN: C3,C2,C0 = 001
+            CSELx(x4, x4, x3, cEQ);
+            MARK3;
+            // Extract signa & Update SW
+            LSRx_IMM(x1, x2, 63);
+            BFIx(x4, x1, 9, 1); //C1
+            LDRH_U12(x1, xEmu, offsetof(x64emu_t, sw));
+            MOV32w(x2, 0b01000111);
+            BICw_REG_LSL(x1, x1, x2, 8);
+            ORRw_REG(x4, x4, x1);
+            STRH_U12(x4, xEmu, offsetof(x64emu_t, sw));
+            #else
            MESSAGE(LOG_DUMP, "Need Optimization\n");
-            x87_refresh(dyn, ninst, x1, x2, 0);
+            i1 = x87_stackcount(dyn, ninst, x1);
+            x87_forget(dyn, ninst, x1, x2, 0);
            CALL(fpu_fxam, -1);  // should be possible inline, but is it worth it?
+            x87_unstackcount(dyn, ninst, x1, i1);
+            #endif
            break;

        case 0xE8:
            INST_NAME("FLD1");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
-            if(ST_IS_F(0)) {
-                FMOVS_8(v1, 0b01110000);
+            if(ninst<dyn->size+2 && (dyn->insts[ninst+1].pred_sz==1) && (dyn->insts[ninst+2].pred_sz==1)
+                && PK(0)==0xD9 && PK(1)==0xE8
+                && PK(2)==0xD9 && PK(3)==0xF3
+            ) {
+                MESSAGE(LOG_DUMP, "Hack for FLD1 FLD1 FPATAN");
+                X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_F);
+                FTABLE64(v1, PI/4.0);
+                addr+=4;
            } else {
-                FMOVD_8(v1, 0b01110000);
+                X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_F);
+                if(ST_IS_F(0)) {
+                    FMOVS_8(v1, 0b01110000);
+                } else {
+                    FMOVD_8(v1, 0b01110000);
+                }
            }
            break;
        case 0xE9:
            INST_NAME("FLDL2T");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
            FTABLE64(v1, L2T);
            break;
-        case 0xEA:     
+        case 0xEA:
            INST_NAME("FLDL2E");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
            FTABLE64(v1, L2E);
            break;
        case 0xEB:
            INST_NAME("FLDPI");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
            FTABLE64(v1, PI);
            break;
        case 0xEC:
            INST_NAME("FLDLG2");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
            FTABLE64(v1, LG2);
            break;
        case 0xED:
            INST_NAME("FLDLN2");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
            FTABLE64(v1, LN2);
            break;
        case 0xEE:
            INST_NAME("FLDZ");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_F);
            VEOR(v1, v1, v1);
            break;

        case 0xF0:
            INST_NAME("F2XM1");
            MESSAGE(LOG_DUMP, "Need Optimization\n");
+            i1 = x87_stackcount(dyn, ninst, x1);
            x87_forget(dyn, ninst, x1, x2, 0);
            CALL(native_f2xm1, -1);
+            x87_unstackcount(dyn, ninst, x1, i1);
            break;
        case 0xF1:
            INST_NAME("FYL2X");
            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_fyl2x, -1);
-            x87_do_pop(dyn, ninst, x3);
+            x87_unstackcount(dyn, ninst, x1, i1);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xF2:
            INST_NAME("FPTAN");
            MESSAGE(LOG_DUMP, "Need Optimization\n");
+            i1 = x87_stackcount(dyn, ninst, x1);
            x87_forget(dyn, ninst, x1, x2, 0);
            CALL(native_ftan, -1);
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
-            if(ST_IS_F(0)) {
-                FMOVS_8(v1, 0b01110000);
+            x87_unstackcount(dyn, ninst, x1, i1);
+            if(PK(0)==0xdd && PK(1)==0xd8) {
+                MESSAGE(LOG_DUMP, "Optimized next DD D8 fstp st0, st0, not emiting 1\n");
+                u8 = F8;
+                u8 = F8;
            } else {
-                FMOVD_8(v1, 0b01110000);
+                X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_F);
+                if(ST_IS_F(0)) {
+                    FMOVS_8(v1, 0b01110000);
+                } else {
+                    FMOVD_8(v1, 0b01110000);
+                }
            }
            break;
        case 0xF3:
            INST_NAME("FPATAN");
            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_fpatan, -1);
-            x87_do_pop(dyn, ninst, x3);
+            x87_unstackcount(dyn, ninst, x1, i1);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xF4:
            INST_NAME("FXTRACT");
            MESSAGE(LOG_DUMP, "Need Optimization\n");
-            x87_do_push_empty(dyn, ninst, 0);
+            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);
+            x87_unstackcount(dyn, ninst, x1, i1);
            break;
        case 0xF5:
            INST_NAME("FPREM1");
            MESSAGE(LOG_DUMP, "Need Optimization\n");
+            #if 0
+            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);
+            FDIVD(s0, v1, v2);
+            FRINTRRD(s0, s0, 0b00); // Nearest == TieToEven?
+            FCVTZSxD(x4, s0);
+            FMULD(s0, s0, v2);
+            FSUBD(v1, v1, s0);
+            LDRw_U12(x1, xEmu, offsetof(x64emu_t, sw));
+            // set C2 = 0
+            BFCw(x1, 10, 1);
+            // set C1 = Q0
+            BFIw(x1, x4, 9, 1);
+            // set C3 = Q1
+            LSRx_IMM(x4, x4, 1);
+            BFIw(x1, x4, 14, 1);
+            // Set C0 = Q2
+            LSRx(x4, x4, 1);
+            BFIw(x1, x4, 8, 1);
+            STRw_U12(x1, xEmu, offsetof(x64emu_t, sw));
+            #endif
            break;
        case 0xF6:
            INST_NAME("FDECSTP");
@@ -242,18 +369,45 @@ 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);
+            FDIVD(s0, v1, v2);
+            FRINTZD(s0, s0);
+            FCVTZSxD(x4, s0);
+            FMULD(s0, s0, v2);
+            FSUBD(v1, v1, s0);
+            LDRw_U12(x1, xEmu, offsetof(x64emu_t, sw));
+            // set C2 = 0
+            BFCw(x1, 10, 1);
+            // set C1 = Q0
+            BFIw(x1, x4, 9, 1);
+            // set C3 = Q1
+            LSRx_IMM(x4, x4, 1);
+            BFIw(x1, x4, 14, 1);
+            // Set C0 = Q2
+            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\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_do_pop(dyn, ninst, x3);
+            x87_unstackcount(dyn, ninst, x1, i1);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xFA:
            INST_NAME("FSQRT");
@@ -267,9 +421,11 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
        case 0xFB:
            INST_NAME("FSINCOS");
            MESSAGE(LOG_DUMP, "Need Optimization\n");
-            x87_do_push_empty(dyn, ninst, 0);
+            X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, 0);
+            i1 = x87_stackcount(dyn, ninst, x1);
            x87_forget(dyn, ninst, x1, x2, 1);
            CALL(native_fsincos, -1);
+            x87_unstackcount(dyn, ninst, x1, i1);
            break;
        case 0xFC:
            INST_NAME("FRNDINT");
@@ -292,108 +448,101 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
        case 0xFD:
            INST_NAME("FSCALE");
            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_fscale, -1);
+            x87_unstackcount(dyn, ninst, x1, i1);
            break;
        case 0xFE:
            INST_NAME("FSIN");
            MESSAGE(LOG_DUMP, "Need Optimization\n");
+            i1 = x87_stackcount(dyn, ninst, x1);
            x87_forget(dyn, ninst, x1, x2, 0);
            CALL(native_fsin, -1);
+            x87_unstackcount(dyn, ninst, x1, i1);
            break;
        case 0xFF:
            INST_NAME("FCOS");
            MESSAGE(LOG_DUMP, "Need Optimization\n");
+            i1 = x87_stackcount(dyn, ninst, x1);
            x87_forget(dyn, ninst, x1, x2, 0);
            CALL(native_fcos, -1);
+            x87_unstackcount(dyn, ninst, x1, i1);
            break;
-
-
-        case 0xD1:
-        case 0xD4:
-        case 0xD5:
-        case 0xD6:
-        case 0xD7:
-        case 0xE2:
-        case 0xE3:
-        case 0xE6:
-        case 0xE7:
-        case 0xEF:
+        default:
            DEFAULT;
            break;
-             
-        default:
-            switch((nextop>>3)&7) {
-                case 0:
-                    INST_NAME("FLD ST0, float[ED]");
-                    v1 = x87_do_push(dyn, ninst, x1, box64_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
-                    VLD32(v1, ed, fixedaddress);
-                    if(!ST_IS_F(0)) {
-                        FCVT_D_S(v1, v1);
-                    }
-                    break;
-                case 2:
-                    INST_NAME("FST float[ED], ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
-                    if(ST_IS_F(0))
-                        s0 = v1;
-                    else {
-                        s0 = fpu_get_scratch(dyn);
-                        FCVT_S_D(s0, v1);
-                    }
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
-                    VST32(s0, ed, fixedaddress);
-                    break;
-                case 3:
-                    INST_NAME("FSTP float[ED], ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
-                    if(!ST_IS_F(0)) {
-                        FCVT_S_D(v1, v1);
-                    }
-                    VST32(v1, ed, fixedaddress);
-                    x87_do_pop(dyn, ninst, x3);
-                    break;
-                case 4:
-                    INST_NAME("FLDENV Ed");
-                    MESSAGE(LOG_DUMP, "Need Optimization\n");
-                    fpu_purgecache(dyn, ninst, 0, x1, x2, x3); // 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);
-                    break;
-                case 5:
-                    INST_NAME("FLDCW Ew");
-                    GETEW(x1, 0);
-                    STRH_U12(x1, xEmu, offsetof(x64emu_t, cw));    // hopefully cw is not too far for an imm8
-                    break;
-                case 6:
-                    INST_NAME("FNSTENV Ed");
-                    MESSAGE(LOG_DUMP, "Need Optimization\n");
-                    fpu_purgecache(dyn, ninst, 0, x1, x2, x3); // 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);
-                    break;
-                case 7:
-                    INST_NAME("FNSTCW Ew");
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
-                    ed = x1;
-                    wb1 = 1;
-                    LDRH_U12(x1, xEmu, offsetof(x64emu_t, cw));
-                    EWBACK;
-                    break;
-                default:
-                    DEFAULT;
-            }
-    }
+    } else
+        switch((nextop>>3)&7) {
+            case 0:
+                INST_NAME("FLD ST0, float[ED]");
+                X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, box64_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(v1, ed, fixedaddress);
+                if(!ST_IS_F(0)) {
+                    FCVT_D_S(v1, v1);
+                }
+                break;
+            case 2:
+                INST_NAME("FST float[ED], ST0");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                if(ST_IS_F(0))
+                    s0 = v1;
+                else {
+                    s0 = fpu_get_scratch(dyn);
+                    FCVT_S_D(s0, v1);
+                }
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VST32(s0, ed, fixedaddress);
+                break;
+            case 3:
+                INST_NAME("FSTP float[ED], ST0");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                if(!ST_IS_F(0)) {
+                    FCVT_S_D(v1, v1);
+                }
+                VST32(v1, ed, fixedaddress);
+                X87_POP_OR_FAIL(dyn, ninst, x3);
+                break;
+            case 4:
+                INST_NAME("FLDENV Ed");
+                MESSAGE(LOG_DUMP, "Need Optimization\n");
+                fpu_purgecache(dyn, ninst, 0, x1, x2, x3); // 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);
+                break;
+            case 5:
+                INST_NAME("FLDCW Ew");
+                GETEW(x1, 0);
+                STRH_U12(x1, xEmu, offsetof(x64emu_t, cw));    // hopefully cw is not too far for an imm8
+                break;
+            case 6:
+                INST_NAME("FNSTENV Ed");
+                MESSAGE(LOG_DUMP, "Need Optimization\n");
+                fpu_purgecache(dyn, ninst, 0, x1, x2, x3); // 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);
+                break;
+            case 7:
+                INST_NAME("FNSTCW Ew");
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
+                ed = x1;
+                wb1 = 1;
+                LDRH_U12(x1, xEmu, offsetof(x64emu_t, cw));
+                EWBACK;
+                break;
+            default:
+                DEFAULT;
+        }
    return addr;
 }
@@ -0,0 +1,229 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "emu/x64run_private.h"
+#include "x64run.h"
+#include "x64emu.h"
+#include "box64stack.h"
+#include "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "emu/x87emu_private.h"
+#include "dynarec_native.h"
+
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
+#include "dynarec_arm64_helper.h"
+#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)
+{
+    uint8_t nextop = F8;
+    int64_t j64;
+    uint8_t ed;
+    uint8_t wback;
+    int v1, v2;
+    int d0;
+    int s0;
+    int64_t fixedaddress;
+    int unscaled;
+
+    MAYUSE(s0);
+    MAYUSE(d0);
+    MAYUSE(v2);
+    MAYUSE(v1);
+    MAYUSE(ed);
+    MAYUSE(j64);
+
+    if(MODREG)
+    switch(nextop) {
+        case 0xC0:
+        case 0xC1:
+        case 0xC2:
+        case 0xC3:
+        case 0xC4:
+        case 0xC5:
+        case 0xC6:
+        case 0xC7:
+            INST_NAME("FCMOVB ST0, STx");
+            READFLAGS(X_CF);
+            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));
+            TSTw_mask(xFlags, 0, 0);    //mask=1<<F_CF
+            if(ST_IS_F(0)) {
+                FCSELS(v1, v2, v1, cNE);   // F_CF==1
+            } else {
+                FCSELD(v1, v2, v1, cNE);   // F_CF==1
+            }
+            break;
+        case 0xC8:
+        case 0xC9:
+        case 0xCA:
+        case 0xCB:
+        case 0xCC:
+        case 0xCD:
+        case 0xCE:
+        case 0xCF:
+            INST_NAME("FCMOVE ST0, STx");
+            READFLAGS(X_ZF);
+            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));
+            TSTw_mask(xFlags, 0b011010, 0); //mask=1<<F_ZF
+            if(ST_IS_F(0)) {
+                FCSELS(v1, v2, v1, cNE);
+            } else {
+                FCSELD(v1, v2, v1, cNE);        // F_ZF==0
+            }
+            break;
+        case 0xD0:
+        case 0xD1:
+        case 0xD2:
+        case 0xD3:
+        case 0xD4:
+        case 0xD5:
+        case 0xD6:
+        case 0xD7:
+            INST_NAME("FCMOVBE ST0, STx");
+            READFLAGS(X_CF|X_ZF);
+            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));
+            MOV32w(x1, (1<<F_CF)|(1<<F_ZF));
+            TSTw_REG(xFlags, x1);
+            if(ST_IS_F(0)) {
+                FCSELS(v1, v2, v1, cNE);
+            } else {
+                FCSELD(v1, v2, v1, cNE);   // F_CF==0 & F_ZF==0
+            }
+            break;
+        case 0xD8:
+        case 0xD9:
+        case 0xDA:
+        case 0xDB:
+        case 0xDC:
+        case 0xDD:
+        case 0xDE:
+        case 0xDF:
+            INST_NAME("FCMOVU ST0, STx");
+            READFLAGS(X_PF);
+            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));
+            TSTw_mask(xFlags, 0b011110, 0); //mask=1<<F_PF
+            if(ST_IS_F(0)) {
+                FCSELS(v1, v2, v1, cNE);
+            } else {
+                FCSELD(v1, v2, v1, cNE);        // F_PF==0
+            }
+            break;
+        case 0xE9:
+            INST_NAME("FUCOMPP ST0, ST1");
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
+            v2 = x87_get_st(dyn, ninst, x1, x2, 1, X87_COMBINE(0, nextop&7));
+            if(ST_IS_F(0)) {
+                FCMPS(v1, v2);
+            } else {
+                FCMPD(v1, v2);
+            }
+            FCOM(x1, x2, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
+            break;
+
+        default:
+            DEFAULT;
+            break;
+    } else
+        switch((nextop>>3)&7) {
+            case 0:
+                INST_NAME("FIADD ST0, Ed");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(v2, ed, fixedaddress);
+                SXTL_32(v2, v2);    // i32 -> i64
+                SCVTFDD(v2, v2);    // i64 -> double
+                FADDD(v1, v1, v2);
+                break;
+            case 1:
+                INST_NAME("FIMUL ST0, Ed");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(v2, ed, fixedaddress);
+                SXTL_32(v2, v2);    // i32 -> i64
+                SCVTFDD(v2, v2);    // i64 -> double
+                FMULD(v1, v1, v2);
+                break;
+            case 2:
+                INST_NAME("FICOM ST0, Ed");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(v2, ed, fixedaddress);
+                SXTL_32(v2, v2);    // i32 -> i64
+                SCVTFDD(v2, v2);    // i64 -> double
+                FCMPD(v1, v2);
+                FCOM(x1, x2, x3);
+                break;
+            case 3:
+                INST_NAME("FICOMP ST0, Ed");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(v2, ed, fixedaddress);
+                SXTL_32(v2, v2);    // i32 -> i64
+                SCVTFDD(v2, v2);    // i64 -> double
+                FCMPD(v1, v2);
+                FCOM(x1, x2, x3);
+                X87_POP_OR_FAIL(dyn, ninst, x3);
+                break;
+            case 4:
+                INST_NAME("FISUB ST0, Ed");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(v2, ed, fixedaddress);
+                SXTL_32(v2, v2);    // i32 -> i64
+                SCVTFDD(v2, v2);    // i64 -> double
+                FSUBD(v1, v1, v2);
+                break;
+            case 5:
+                INST_NAME("FISUBR ST0, Ed");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(v2, ed, fixedaddress);
+                SXTL_32(v2, v2);    // i32 -> i64
+                SCVTFDD(v2, v2);    // i64 -> double
+                FSUBD(v1, v2, v1);
+                break;
+            case 6:
+                INST_NAME("FIDIV ST0, Ed");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(v2, ed, fixedaddress);
+                SXTL_32(v2, v2);    // i32 -> i64
+                SCVTFDD(v2, v2);    // i64 -> double
+                FDIVD(v1, v1, v2);
+                break;
+            case 7:
+                INST_NAME("FIDIVR ST0, Ed");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(v2, ed, fixedaddress);
+                SXTL_32(v2, v2);    // i32 -> i64
+                SCVTFDD(v2, v2);    // i64 -> double
+                FDIVD(v1, v2, v1);
+                break;
+        }
+    return addr;
+}
+
@@ -42,6 +42,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
    MAYUSE(v1);
    MAYUSE(j64);

+    if(MODREG)
    switch(nextop) {
        case 0xC0:
        case 0xC1:
@@ -156,7 +157,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            }
            FCOMI(x1, x2);
            break;
-        case 0xF0:  
+        case 0xF0:
        case 0xF1:
        case 0xF2:
        case 0xF3:
@@ -176,35 +177,29 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            FCOMI(x1, x2);
            break;

-        case 0xE0:
-        case 0xE4:
-        case 0xE5:
-        case 0xE6:
-        case 0xE7:
+        default:
            DEFAULT;
            break;
-
-        default:
-            switch((nextop>>3)&7) {
-                case 0:
-                    INST_NAME("FILD ST0, Ed");
-                    v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
-                    VLD32(v1, ed, fixedaddress);
-                    SXTL_32(v1, v1);    // i32 -> i64
-                    SCVTFDD(v1, v1);    // i64 -> double
-                    break;
-                case 1:
-                    INST_NAME("FISTTP Ed, ST0");
-                    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);
-                    #if 0
-                    ed = x1;
+    } else
+        switch((nextop>>3)&7) {
+            case 0:
+                INST_NAME("FILD ST0, Ed");
+                X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                VLD32(v1, ed, fixedaddress);
+                SXTL_32(v1, v1);    // i32 -> i64
+                SCVTFDD(v1, v1);    // i64 -> double
+                break;
+            case 1:
+                INST_NAME("FISTTP Ed, ST0");
+                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);
+                if(arm64_frintts) {
                    FRINT32ZD(s0, v1);
-                    FCVTZSwD(ed, s0);
-                    WBACK;
-                    #else
+                    FCVTZSwD(x5, s0);
+                    STW(x5, wback, fixedaddress);
+                } else {
                    MRS_fpsr(x5);
                    BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
                    MSR_fpsr(x5);
@@ -217,21 +212,20 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    MOV32w(x5, 0x80000000);
                    STW(x5, wback, fixedaddress);
                    MARK3;
-                    #endif
-                    x87_do_pop(dyn, ninst, x3);
-                    break;
-                case 2:
-                    INST_NAME("FIST Ed, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    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);
-                    #if 0
-                    ed = x1;
+                }
+                X87_POP_OR_FAIL(dyn, ninst, x3);
+                break;
+            case 2:
+                INST_NAME("FIST Ed, ST0");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                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);
+                if(arm64_frintts) {
                    FRINT32XD(s0, v1);
-                    FCVTZSwD(ed, s0);
-                    WBACK;
-                    #else
+                    FCVTZSwD(x5, s0);
+                    STW(x5, wback, fixedaddress);
+                } else {
                    MRS_fpsr(x5);
                    BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
                    MSR_fpsr(x5);
@@ -244,21 +238,20 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    MOV32w(x5, 0x80000000);
                    STW(x5, wback, fixedaddress);
                    MARK3;
-                    #endif
-                    x87_restoreround(dyn, ninst, u8);
-                    break;
-                case 3:
-                    INST_NAME("FISTP Ed, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    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);
-                    #if 0
-                    ed = x1;
+                }
+                x87_restoreround(dyn, ninst, u8);
+                break;
+            case 3:
+                INST_NAME("FISTP Ed, ST0");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                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);
+                if(arm64_frintts) {
                    FRINT32XD(s0, v1);
-                    FCVTZSwD(ed, s0);
-                    WBACK;
-                    #else
+                    FCVTZSwD(x5, s0);
+                    STW(x5, wback, fixedaddress);
+                } else {
                    MRS_fpsr(x5);
                    BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
                    MSR_fpsr(x5);
@@ -271,112 +264,116 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                    MOV32w(x5, 0x80000000);
                    STW(x5, wback, fixedaddress);
                    MARK3;
-                    #endif
-                    x87_restoreround(dyn, ninst, u8);
-                    x87_do_pop(dyn, ninst, x3);
-                    break;
-                case 5:
-                    INST_NAME("FLD tbyte");
+                }
+                x87_restoreround(dyn, ninst, u8);
+                X87_POP_OR_FAIL(dyn, ninst, x3);
+                break;
+            case 5:
+                INST_NAME("FLD tbyte");
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
+                if((PK(0)==0xDB && ((PK(1)>>3)&7)==7) || (!rex.is32bits && PK(0)>=0x40 && PK(0)<=0x4f && PK(1)==0xDB && ((PK(2)>>3)&7)==7)) {
+                    NOTEST(x5);
+                    // the FLD is immediatly followed by an FSTP
+                    LDRx_U12(x5, ed, 0);
+                    LDRH_U12(x6, ed, 8);
+                    // no persistant scratch register, so unrool both instruction here...
+                    MESSAGE(LOG_DUMP, "\tHack: FSTP tbyte\n");
+                    nextop = F8;    // 0xDB or rex
+                    if(!rex.is32bits && nextop>=0x40 && nextop<=0x4f) {
+                        rex.rex = nextop;
+                        nextop = F8;    //0xDB
+                    } else
+                        rex.rex = 0;
+                    nextop = F8;    //modrm
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
-                    if((PK(0)==0xDB && ((PK(1)>>3)&7)==7) || (PK(0)>=0x40 && PK(0)<=0x4f && PK(1)==0xDB && ((PK(2)>>3)&7)==7)) {
-                        // the FLD is immediatly followed by an FSTP
-                        LDRx_U12(x5, ed, 0);
-                        LDRH_U12(x6, ed, 8);
-                        // no persistant scratch register, so unrool both instruction here...
-                        MESSAGE(LOG_DUMP, "\tHack: FSTP tbyte\n");
-                        nextop = F8;    // 0xDB or rex
-                        if(nextop>=0x40 && nextop<=0x4f) {
-                            rex.rex = nextop;
-                            nextop = F8;    //0xDB
-                        } else
-                            rex.rex = 0;
-                        nextop = F8;    //modrm
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
-                        STRx_U12(x5, ed, 0);
-                        STRH_U12(x6, ed, 8);
-                    } else {
-                        if(box64_x87_no80bits) {
-                            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
-                            VLDR64_U12(v1, ed, fixedaddress);
-                        } else {
-                            if(ed!=x1) {
-                                MOVx_REG(x1, ed);
-                            }
-                            x87_do_push_empty(dyn, ninst, x3);
-                            CALL(native_fld, -1);
-                        }
-                    }
-                    break;
-                case 7:
-                    INST_NAME("FSTP tbyte");
+                    STRx_U12(x5, ed, 0);
+                    STRH_U12(x6, ed, 8);
+                } else {
                    if(box64_x87_no80bits) {
-                        v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                        VST64(v1, wback, fixedaddress);
+                        X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
+                        VLDR64_U12(v1, ed, fixedaddress);
                    } else {
-                        #if 0
-                        x87_forget(dyn, ninst, x1, x3, 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(native_fstp, -1);
-                        #else
-                        // Painfully long, straight conversion from the C code, shoud be optimized
-                        v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
-                        FMOVxD(x1, v1);
-                        // do special value first
-                        TSTx_mask(x1, 1, 0b00000, 0b111110);    //0x7fffffffffffffffL
-                        B_MARK(cNE);
-                        // Zero
-                        LSRx(x3, x1, 63-15);    //x3 = sign+exp
-                        MOVZw(x5, 0);           // x5 = mantisse
-                        B_MARK3_nocond;
-                        MARK;
-                        // get sign, in main ouput x5 for sign+exp
-                        ANDx_mask(x5, x1, 1, 1, 0); //0x8000000000000000
-                        LSRx(x5, x5, 63-15);    // x5 = sign
-                        // get exp
-                        LSRx(x3, x1, 52);       // x3 = exp11
-                        ANDw_mask(x3, x3, 0, 0b1010);    //0x7ff
-                        MOV32w(x4, 0x7ff);
-                        CMPSw_REG(x3, x4);
-                        B_MARK2(cNE);
-                        // NaN and Infinite
-                        ORRw_mask(x3, x5, 0, 0b1110);    //x3 = sign | 0x7fff 
-                        TSTx_mask(x1, 1, 0, 0b110011); //0x000fffffffffffffL
-                        ORRx_mask(x5, xZR, 1, 1, 0);    //0x8000000000000000
-                        ORRx_mask(x4, xZR, 1, 0b10, 0b01); //0xc000000000000000
-                        CSELx(x5, x5, x4, cEQ);     // x5 = mantisse
-                        B_MARK3_nocond;
-                        MARK2;
-                        // regular / denormals
-                        ANDx_mask(x1, x1, 1, 0, 0b110011); //0x000fffffffffffffL
-                        LSLx_IMM(x1, x1, 11);   //x1 = mantisse missing "1"
-                        MOVZw(x4, 16383-1023);  //BIAS80 - BIAS64
-                        CBZw(x3, 4+3*4);        // exp11 == 0?
-                        // normals
-                        ADDw_REG(x3, x3, x4);   // x3 = exp16
-                        ORRw_REG(x3, x3, x5);   // x3 = sign | exp
-                        ORRx_mask(x5, x1, 1, 1, 0);    //0x8000000000000000 x5 = mantisse
-                        B_MARK3_nocond;
-                        // denormals
-                        CLZx(x6, x1);
-                        ADDw_U12(x6, x6, 1);    // "one"
-                        SUBw_REG(x3, x4, x6);   // x3 = exp16
-                        ORRw_REG(x3, x3, x5);   // x3 = sign | exp16
-                        LSLx_REG(x5, x1, x6);   // x5 = mantisse
-                        MARK3;
-                        STRx_U12(x5, wback, 0);
-                        STRH_U12(x3, wback, 8);
-                        #endif
+                        X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, 0);
+                        // sync top
+                        s0 = x87_stackcount(dyn, ninst, x3);
+                        CALL(native_fld, -1);
+                        // go back with the top & stack counter
+                        x87_unstackcount(dyn, ninst, x3, s0);
                    }
-                    x87_do_pop(dyn, ninst, x3);
-                    break;
-                default:
-                    DEFAULT;
-            }
-    }
+                }
+                break;
+            case 7:
+                INST_NAME("FSTP tbyte");
+                if(box64_x87_no80bits) {
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
+                    VST64(v1, wback, fixedaddress);
+                } else {
+                    #if 0
+                    x87_forget(dyn, ninst, x1, x3, 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(native_fstp, -1);
+                    #else
+                    // Painfully long, straight conversion from the C code, shoud be optimized
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
+                    FMOVxD(x1, v1);
+                    // do special value first
+                    TSTx_mask(x1, 1, 0b00000, 0b111110);    //0x7fffffffffffffffL
+                    B_MARK(cNE);
+                    // Zero
+                    LSRx(x3, x1, 63-15);    //x3 = sign+exp
+                    MOVZw(x5, 0);           // x5 = mantisse
+                    B_MARK3_nocond;
+                    MARK;
+                    // get sign, in main ouput x5 for sign+exp
+                    ANDx_mask(x5, x1, 1, 1, 0); //0x8000000000000000
+                    LSRx(x5, x5, 63-15);    // x5 = sign
+                    // get exp
+                    LSRx(x3, x1, 52);       // x3 = exp11
+                    ANDw_mask(x3, x3, 0, 0b1010);    //0x7ff
+                    MOV32w(x4, 0x7ff);
+                    CMPSw_REG(x3, x4);
+                    B_MARK2(cNE);
+                    // NaN and Infinite
+                    ORRw_mask(x3, x5, 0, 0b1110);    //x3 = sign | 0x7fff
+                    TSTx_mask(x1, 1, 0, 0b110011); //0x000fffffffffffffL
+                    ORRx_mask(x5, xZR, 1, 1, 0);    //0x8000000000000000
+                    ORRx_mask(x4, xZR, 1, 0b10, 0b01); //0xc000000000000000
+                    CSELx(x5, x5, x4, cEQ);     // x5 = mantisse
+                    B_MARK3_nocond;
+                    MARK2;
+                    // regular / denormals
+                    ANDx_mask(x1, x1, 1, 0, 0b110011); //0x000fffffffffffffL
+                    LSLx_IMM(x1, x1, 11);   //x1 = mantisse missing "1"
+                    MOVZw(x4, 16383-1023);  //BIAS80 - BIAS64
+                    CBZw(x3, 4+3*4);        // exp11 == 0?
+                    // normals
+                    ADDw_REG(x3, x3, x4);   // x3 = exp16
+                    ORRw_REG(x3, x3, x5);   // x3 = sign | exp
+                    ORRx_mask(x5, x1, 1, 1, 0);    //0x8000000000000000 x5 = mantisse
+                    B_MARK3_nocond;
+                    // denormals
+                    CLZx(x6, x1);
+                    ADDw_U12(x6, x6, 1);    // "one"
+                    SUBw_REG(x3, x4, x6);   // x3 = exp16
+                    ORRw_REG(x3, x3, x5);   // x3 = sign | exp16
+                    LSLx_REG(x5, x1, x6);   // x5 = mantisse
+                    MARK3;
+                    STRx_U12(x5, wback, 0);
+                    STRH_U12(x3, wback, 8);
+                    #endif
+                }
+                X87_POP_OR_FAIL(dyn, ninst, x3);
+                break;
+            default:
+                DEFAULT;
+        }
    return addr;
 }
@@ -36,6 +36,7 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
    MAYUSE(v2);
    MAYUSE(v1);

+    if(MODREG)
    switch(nextop) {
        case 0xC0:
        case 0xC1:
@@ -106,7 +107,7 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                FCMPD(v1, v2);
            }
            FCOM(x1, x2, x3);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xE0:
        case 0xE1:
@@ -158,7 +159,7 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            } else {
                FDIVD(v1, v2, v1);
            }
-            break;       
+            break;
        case 0xF8:
        case 0xF9:
        case 0xFA:
@@ -177,75 +178,77 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            }
            break;
        default:
-            switch((nextop>>3)&7) {
-                case 0:
-                    INST_NAME("FADD ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                    VLD64(v2, wback, fixedaddress);
-                    FADDD(v1, v1, v2);
-                    break;
-                case 1:
-                    INST_NAME("FMUL ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                    VLD64(v2, wback, fixedaddress);
-                    FMULD(v1, v1, v2);
-                    break;
-                case 2:
-                    INST_NAME("FCOM ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                    VLD64(v2, wback, fixedaddress);
-                    FCMPD(v1, v2);
-                    FCOM(x1, x2, x3);
-                    break;
-                case 3:
-                    INST_NAME("FCOMP ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                    VLD64(v2, wback, fixedaddress);
-                    FCMPD(v1, v2);
-                    FCOM(x1, x2, x3);
-                    x87_do_pop(dyn, ninst, x3);
-                    break;
-                case 4:
-                    INST_NAME("FSUB ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                    VLD64(v2, wback, fixedaddress);
-                    FSUBD(v1, v1, v2);
-                    break;
-                case 5:
-                    INST_NAME("FSUBR ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                    VLD64(v2, wback, fixedaddress);
-                    FSUBD(v1, v2, v1);
-                    break;
-                case 6:
-                    INST_NAME("FDIV ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                    VLD64(v2, wback, fixedaddress);
-                    FDIVD(v1, v1, v2);
-                    break;
-                case 7:
-                    INST_NAME("FDIVR ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                    VLD64(v2, wback, fixedaddress);
-                    FDIVD(v1, v2, v1);
-                    break;
-            }
-    }
+            DEFAULT;
+            break;
+    } else
+        switch((nextop>>3)&7) {
+            case 0:
+                INST_NAME("FADD ST0, double[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
+                VLD64(v2, wback, fixedaddress);
+                FADDD(v1, v1, v2);
+                break;
+            case 1:
+                INST_NAME("FMUL ST0, double[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
+                VLD64(v2, wback, fixedaddress);
+                FMULD(v1, v1, v2);
+                break;
+            case 2:
+                INST_NAME("FCOM ST0, double[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
+                VLD64(v2, wback, fixedaddress);
+                FCMPD(v1, v2);
+                FCOM(x1, x2, x3);
+                break;
+            case 3:
+                INST_NAME("FCOMP ST0, double[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
+                VLD64(v2, wback, fixedaddress);
+                FCMPD(v1, v2);
+                FCOM(x1, x2, x3);
+                X87_POP_OR_FAIL(dyn, ninst, x3);
+                break;
+            case 4:
+                INST_NAME("FSUB ST0, double[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
+                VLD64(v2, wback, fixedaddress);
+                FSUBD(v1, v1, v2);
+                break;
+            case 5:
+                INST_NAME("FSUBR ST0, double[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
+                VLD64(v2, wback, fixedaddress);
+                FSUBD(v1, v2, v1);
+                break;
+            case 6:
+                INST_NAME("FDIV ST0, double[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
+                VLD64(v2, wback, fixedaddress);
+                FDIVD(v1, v1, v2);
+                break;
+            case 7:
+                INST_NAME("FDIVR ST0, double[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
+                VLD64(v2, wback, fixedaddress);
+                FDIVD(v1, v2, v1);
+                break;
+        }
    return addr;
 }
@@ -40,6 +40,7 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
    MAYUSE(v1);
    MAYUSE(j64);

+    if(MODREG)
    switch(nextop) {
        case 0xC0:
        case 0xC1:
@@ -50,10 +51,20 @@ 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);
+                addr+=2;
+                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:
@@ -74,7 +85,7 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            break;
        case 0xD8:
            INST_NAME("FSTP ST0, ST0");
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xD9:
        case 0xDA:
@@ -88,7 +99,7 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            x87_get_st_empty(dyn, ninst, x1, x2, nextop&7, X87_ST(nextop&7));
            x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
            x87_swapreg(dyn, ninst, x1, x2, 0, nextop&7);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;

        case 0xE0:
@@ -126,119 +137,98 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                FCMPD(v1, v2);
            }
            FCOM(x1, x2, x3);
-            x87_do_pop(dyn, ninst, x3);
-            break;
-
-        case 0xC8:
-        case 0xC9:
-        case 0xCA:
-        case 0xCB:
-        case 0xCC:
-        case 0xCD:
-        case 0xCE:
-        case 0xCF:
-        case 0xF0:
-        case 0xF1:
-        case 0xF2:
-        case 0xF3:
-        case 0xF4:
-        case 0xF5:
-        case 0xF6:
-        case 0xF7:
-        case 0xF8:
-        case 0xF9:
-        case 0xFA:
-        case 0xFB:
-        case 0xFC:
-        case 0xFD:
-        case 0xFE:
-        case 0xFF:
-            DEFAULT;
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;

        default:
-            switch((nextop>>3)&7) {
-                case 0:
-                    INST_NAME("FLD double");
-                    v1 = x87_do_push(dyn, ninst, x3, NEON_CACHE_ST_D);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                    VLD64(v1, ed, fixedaddress);
-                    break;
-                case 1:
-                    INST_NAME("FISTTP i64, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
+            DEFAULT;
+            break;
+    } else
+        switch((nextop>>3)&7) {
+            case 0:
+                INST_NAME("FLD double");
+                X87_PUSH_OR_FAIL(v1, dyn, ninst, x3, NEON_CACHE_ST_D);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
+                VLD64(v1, ed, fixedaddress);
+                break;
+            case 1:
+                INST_NAME("FISTTP i64, ST0");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_I64);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
+                if(ST_IS_I64(0)) {
+                    VST64(v1, ed, fixedaddress);
+                } else {
                    s0 = fpu_get_scratch(dyn);
-                    #if 0
-                    // those are ARM 8.5 opcode!
-                    FRINT64ZD(s0, v1);
-                    FCVTZSxD(x2, s0);
-                    STRx_U12(x2, ed, fixedaddress);
-                    #else
-                    MRS_fpsr(x5);
-                    BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
-                    MSR_fpsr(x5);
-                    FRINTRRD(s0, v1, 3);
-                    FCVTZSxD(x2, s0);
-                    STx(x2, ed, fixedaddress);
-                    MRS_fpsr(x5);   // get back FPSR to check the IOC bit
-                    TBZ_MARK3(x5, FPSR_IOC);
-                    ORRx_mask(x5, xZR, 1, 1, 0);    //0x8000000000000000
-                    STx(x5, ed, fixedaddress);
-                    MARK3;
-                    #endif
-                    x87_do_pop(dyn, ninst, x3);
-                    break;
-                case 2:
-                    INST_NAME("FST double");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                    VST64(v1, ed, fixedaddress);
-                    break;
-                case 3:
-                    INST_NAME("FSTP double");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                    VST64(v1, ed, fixedaddress);
-                    x87_do_pop(dyn, ninst, x3);
-                    break;
-                case 4: 
-                    INST_NAME("FRSTOR m108byte");
-                    MESSAGE(LOG_DUMP, "Need Optimization\n");
-                    fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
-                    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);
-                    break;
-                case 6: 
-                    INST_NAME("FSAVE m108byte");
-                    MESSAGE(LOG_DUMP, "Need Optimization\n");
-                    fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
-                    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);
-                    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));
-                    if(dyn->n.x87stack) {
-                        // update top
-                        if(dyn->n.x87stack>0) {
-                            SUBw_U12(x2, x2, dyn->n.x87stack);
-                        } else {
-                            ADDw_U12(x2, x2, -dyn->n.x87stack);
-                        }
-                        ANDw_mask(x2, x2, 0, 2);
+                    if(arm64_frintts) {
+                        FRINT64ZD(s0, v1);
+                        FCVTZSxD(x2, s0);
+                        STx(x2, ed, fixedaddress);
+                    } else {
+                        MRS_fpsr(x5);
+                        BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
+                        MSR_fpsr(x5);
+                        FRINTRRD(s0, v1, 3);
+                        FCVTZSxD(x2, s0);
+                        MRS_fpsr(x5);   // get back FPSR to check the IOC bit
+                        TBZ_MARK3(x5, FPSR_IOC);
+                        ORRx_mask(x2, xZR, 1, 1, 0);    //0x8000000000000000
+                        MARK3;
+                        STx(x2, ed, fixedaddress);
                    }
-                    BFIw(x3, x2, 11, 3); // inject TOP at bit 11 (3 bits)
-                    STH(x3, ed, fixedaddress);   // store whole sw flags
-                    break;
-                default:
-                    DEFAULT;
-            }
-    }
+                }
+                X87_POP_OR_FAIL(dyn, ninst, x3);
+                break;
+            case 2:
+                INST_NAME("FST double");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
+                VST64(v1, ed, fixedaddress);
+                break;
+            case 3:
+                INST_NAME("FSTP double");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
+                VST64(v1, ed, fixedaddress);
+                X87_POP_OR_FAIL(dyn, ninst, x3);
+                break;
+            case 4:
+                INST_NAME("FRSTOR m108byte");
+                MESSAGE(LOG_DUMP, "Need Optimization\n");
+                fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+                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);
+                break;
+            case 6:
+                INST_NAME("FNSAVE m108byte");
+                MESSAGE(LOG_DUMP, "Need Optimization\n");
+                fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+                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);
+                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));
+                if(dyn->n.x87stack) {
+                    // update top
+                    if(dyn->n.x87stack>0) {
+                        SUBw_U12(x2, x2, dyn->n.x87stack);
+                    } else {
+                        ADDw_U12(x2, x2, -dyn->n.x87stack);
+                    }
+                    ANDw_mask(x2, x2, 0, 2);
+                }
+                BFIw(x3, x2, 11, 3); // inject TOP at bit 11 (3 bits)
+                STRH_U12(x3, xEmu, offsetof(x64emu_t, sw));
+                STH(x3, ed, fixedaddress);   // store whole sw flags
+                break;
+            default:
+                DEFAULT;
+        }
    return addr;
 }
@@ -36,6 +36,7 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
    MAYUSE(v2);
    MAYUSE(v1);

+    if(MODREG)
    switch(nextop) {
        case 0xC0:
        case 0xC1:
@@ -53,7 +54,7 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            } else {
                FADDD(v1, v1, v2);
            }
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xC8:
        case 0xC9:
@@ -71,7 +72,7 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            } else {
                FMULD(v1, v1, v2);
            }
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xD0:
        case 0xD1:
@@ -90,7 +91,7 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                FCMPD(v1, v2);
            }
            FCOM(x1, x2, x3);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xD9:
            INST_NAME("FCOMPP ST0, STx");
@@ -102,8 +103,8 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                FCMPD(v1, v2);
            }
            FCOM(x1, x2, x3);
-            x87_do_pop(dyn, ninst, x3);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xE0:
        case 0xE1:
@@ -121,7 +122,7 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            } else {
                FSUBD(v1, v2, v1);
            }
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xE8:
        case 0xE9:
@@ -139,7 +140,7 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            } else {
                FSUBD(v1, v1, v2);
            }
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xF0:
        case 0xF1:
@@ -157,8 +158,8 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            } else {
                FDIVD(v1, v2, v1);
            }
-            x87_do_pop(dyn, ninst, x3);
-            break;       
+            X87_POP_OR_FAIL(dyn, ninst, x3);
+            break;
        case 0xF8:
        case 0xF9:
        case 0xFA:
@@ -175,110 +176,104 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
            } else {
                FDIVD(v1, v1, v2);
            }
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
-        case 0xD8:
-        case 0xDA:
-        case 0xDB:
-        case 0xDC:
-        case 0xDD:
-        case 0xDE:
-        case 0xDF:
-            return 0;
        default:
-            switch((nextop>>3)&7) {
-                case 0:
-                    INST_NAME("FIADD ST0, word[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
-                    VLD16(v2, wback, fixedaddress);
-                    SXTL_16(v2, v2);
-                    SXTL_32(v2, v2);
-                    SCVTFDD(v2, v2);
-                    FADDD(v1, v1, v2);
-                    break;
-                case 1:
-                    INST_NAME("FIMUL ST0, word[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
-                    VLD16(v2, wback, fixedaddress);
-                    SXTL_16(v2, v2);
-                    SXTL_32(v2, v2);
-                    SCVTFDD(v2, v2);
-                    FMULD(v1, v1, v2);
-                    break;
-                case 2:
-                    INST_NAME("FICOM ST0, word[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
-                    VLD16(v2, wback, fixedaddress);
-                    SXTL_16(v2, v2);
-                    SXTL_32(v2, v2);
-                    SCVTFDD(v2, v2);
-                    FCMPD(v1, v2);
-                    FCOM(x1, x2, x3);
-                    break;
-                case 3:
-                    INST_NAME("FICOMP ST0, word[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
-                    VLD16(v2, wback, fixedaddress);
-                    SXTL_16(v2, v2);
-                    SXTL_32(v2, v2);
-                    SCVTFDD(v2, v2);
-                    FCMPD(v1, v2);
-                    FCOM(x1, x2, x3);
-                    x87_do_pop(dyn, ninst, x3);
-                    break;
-                case 4:
-                    INST_NAME("FISUB ST0, word[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
-                    VLD16(v2, wback, fixedaddress);
-                    SXTL_16(v2, v2);
-                    SXTL_32(v2, v2);
-                    SCVTFDD(v2, v2);
-                    FSUBD(v1, v1, v2);
-                    break;
-                case 5:
-                    INST_NAME("FISUBR ST0, word[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
-                    VLD16(v2, wback, fixedaddress);
-                    SXTL_16(v2, v2);
-                    SXTL_32(v2, v2);
-                    SCVTFDD(v2, v2);
-                    FSUBD(v1, v2, v1);
-                    break;
-                case 6:
-                    INST_NAME("FIDIV ST0, word[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
-                    VLD16(v2, wback, fixedaddress);
-                    SXTL_16(v2, v2);
-                    SXTL_32(v2, v2);
-                    SCVTFDD(v2, v2);
-                    FDIVD(v1, v1, v2);
-                    break;
-                case 7:
-                    INST_NAME("FIDIVR ST0, word[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
-                    VLD16(v2, wback, fixedaddress);
-                    SXTL_16(v2, v2);
-                    SXTL_32(v2, v2);
-                    SCVTFDD(v2, v2);
-                    FDIVD(v1, v2, v1);
-                    break;
-            }
-    }
+            DEFAULT;
+            break;
+    } else
+        switch((nextop>>3)&7) {
+            case 0:
+                INST_NAME("FIADD ST0, word[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
+                VLD16(v2, wback, fixedaddress);
+                SXTL_16(v2, v2);
+                SXTL_32(v2, v2);
+                SCVTFDD(v2, v2);
+                FADDD(v1, v1, v2);
+                break;
+            case 1:
+                INST_NAME("FIMUL ST0, word[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
+                VLD16(v2, wback, fixedaddress);
+                SXTL_16(v2, v2);
+                SXTL_32(v2, v2);
+                SCVTFDD(v2, v2);
+                FMULD(v1, v1, v2);
+                break;
+            case 2:
+                INST_NAME("FICOM ST0, word[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
+                VLD16(v2, wback, fixedaddress);
+                SXTL_16(v2, v2);
+                SXTL_32(v2, v2);
+                SCVTFDD(v2, v2);
+                FCMPD(v1, v2);
+                FCOM(x1, x2, x3);
+                break;
+            case 3:
+                INST_NAME("FICOMP ST0, word[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
+                VLD16(v2, wback, fixedaddress);
+                SXTL_16(v2, v2);
+                SXTL_32(v2, v2);
+                SCVTFDD(v2, v2);
+                FCMPD(v1, v2);
+                FCOM(x1, x2, x3);
+                X87_POP_OR_FAIL(dyn, ninst, x3);
+                break;
+            case 4:
+                INST_NAME("FISUB ST0, word[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
+                VLD16(v2, wback, fixedaddress);
+                SXTL_16(v2, v2);
+                SXTL_32(v2, v2);
+                SCVTFDD(v2, v2);
+                FSUBD(v1, v1, v2);
+                break;
+            case 5:
+                INST_NAME("FISUBR ST0, word[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
+                VLD16(v2, wback, fixedaddress);
+                SXTL_16(v2, v2);
+                SXTL_32(v2, v2);
+                SCVTFDD(v2, v2);
+                FSUBD(v1, v2, v1);
+                break;
+            case 6:
+                INST_NAME("FIDIV ST0, word[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
+                VLD16(v2, wback, fixedaddress);
+                SXTL_16(v2, v2);
+                SXTL_32(v2, v2);
+                SCVTFDD(v2, v2);
+                FDIVD(v1, v1, v2);
+                break;
+            case 7:
+                INST_NAME("FIDIVR ST0, word[ED]");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                v2 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
+                VLD16(v2, wback, fixedaddress);
+                SXTL_16(v2, v2);
+                SXTL_32(v2, v2);
+                SCVTFDD(v2, v2);
+                FDIVD(v1, v2, v1);
+                break;
+        }
    return addr;
 }
@@ -34,12 +34,14 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
    int64_t j64;
    int64_t fixedaddress;
    int unscaled;
+    int i1;

    MAYUSE(s0);
    MAYUSE(v2);
    MAYUSE(v1);
    MAYUSE(j64);

+    if(MODREG)
    switch(nextop) {
        case 0xC0:
        case 0xC1:
@@ -51,12 +53,39 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
        case 0xC7:
            INST_NAME("FFREEP STx");
            // not handling Tag...
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
+            break;
+
+        case 0xD0:
+        case 0xD1:
+        case 0xD2:
+        case 0xD3:
+        case 0xD4:
+        case 0xD5:
+        case 0xD6:
+        case 0xD7:
+            INST_NAME("FSTP STx, ST0");
+            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)) {
+                FMOVS(v2, v1);
+            } else {
+                FMOVD(v2, v1);
+            }
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;

        case 0xE0:
            INST_NAME("FNSTSW AX");
            LDRw_U12(x2, xEmu, offsetof(x64emu_t, top));
+            if(dyn->n.x87stack) {
+                if(dyn->n.x87stack>0) {
+                    SUBw_U12(x2, x2, dyn->n.x87stack);
+                } else {
+                    ADDw_U12(x2, x2, -dyn->n.x87stack);
+                }
+                ANDw_mask(x2, x2, 0, 2);  //mask=7
+            }
            LDRH_U12(x1, xEmu, offsetof(x64emu_t, sw));
            BFIw(x1, x2, 11, 3); // inject top
            STRH_U12(x1, xEmu, offsetof(x64emu_t, sw));
@@ -81,7 +110,7 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                FCMPD(v1, v2);
            }
            FCOMI(x1, x2);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xF0:
        case 0xF1:
@@ -102,284 +131,248 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                FCMPD(v1, v2);
            }
            FCOMI(x1, x2);
-            x87_do_pop(dyn, ninst, x3);
-            break;
-
-        case 0xC8:
-        case 0xC9:
-        case 0xCA:
-        case 0xCB:
-        case 0xCC:
-        case 0xCD:
-        case 0xCE:
-        case 0xCF:
-        case 0xD0:
-        case 0xD1:
-        case 0xD2:
-        case 0xD3:
-        case 0xD4:
-        case 0xD5:
-        case 0xD6:
-        case 0xD7:
-        case 0xD8:
-        case 0xD9:
-        case 0xDA:
-        case 0xDB:
-        case 0xDC:
-        case 0xDD:
-        case 0xDE:
-        case 0xDF:
-        case 0xE1:
-        case 0xE2:
-        case 0xE3:
-        case 0xE4:
-        case 0xE5:
-        case 0xE6:
-        case 0xE7:
-        case 0xF8:
-        case 0xF9:
-        case 0xFA:
-        case 0xFB:
-        case 0xFC:
-        case 0xFD:
-        case 0xFE:
-        case 0xFF:
-            DEFAULT;
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;

        default:
-            switch((nextop>>3)&7) {
-                case 0:
-                    INST_NAME("FILD ST0, Ew");
-                    v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
-                    LDSHw(x1, wback, fixedaddress);
-                    if(ST_IS_F(0)) {
-                        SCVTFSw(v1, x1);
-                    } else {
-                        SCVTFDw(v1, x1);
-                    }
-                    break;
-                case 1:
-                    INST_NAME("FISTTP Ew, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
-                    ed = x1;
-                    s0 = fpu_get_scratch(dyn);
-                    #if 0
-                    // this version needs ARM v8.5, //TODO: add detection of this extension to use it
-                    FRINT32ZD(s0, v1);
-                    // no saturation instruction on Arm, so using NEON
+            DEFAULT;
+            break;
+    } else
+        switch((nextop>>3)&7) {
+            case 0:
+                INST_NAME("FILD ST0, Ew");
+                X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_F);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
+                LDSHw(x1, wback, fixedaddress);
+                if(ST_IS_F(0)) {
+                    SCVTFSw(v1, x1);
+                } else {
+                    SCVTFDw(v1, x1);
+                }
+                break;
+            case 1:
+                INST_NAME("FISTTP Ew, ST0");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
+                ed = x1;
+                s0 = fpu_get_scratch(dyn);
+                #if 0
+                // this version needs ARM v8.5, and doesn't handle saturation for 32bits integer not fitting 16bits
+                FRINT32ZD(s0, v1);
+                // no saturation instruction on Arm, so using NEON
+                VFCVTZSd(s0, s0);
+                SQXTN_S_D(s0, s0);
+                SQXTN_H_S(s0, s0);
+                VST16(s0, wback, fixedaddress);
+                #else
+                MRS_fpsr(x5);
+                BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
+                BFCw(x5, FPSR_QC, 1);   // reset QC bit
+                MSR_fpsr(x5);
+                if(ST_IS_F(0)) {
+                    VFCVTZSs(s0, v1);
+                } else {
+                    VFCVTZSd(s0, v1);
+                    SQXTN_S_D(s0, s0);
+                }
+                VMOVSto(x3, s0, 0);
+                MRS_fpsr(x5);   // get back FPSR to check the IOC bit
+                TBNZ_MARK2(x5, FPSR_IOC);
+                SXTHw(x5, x3);  // check if 16bits value is fine
+                SUBw_REG(x5, x5, x3);
+                CBZw_MARK3(x5);
+                MARK2;
+                MOV32w(x3, 0x8000);
+                MARK3;
+                STH(x3, wback, fixedaddress);
+                #endif
+                X87_POP_OR_FAIL(dyn, ninst, x3);
+                break;
+            case 2:
+                INST_NAME("FIST Ew, ST0");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                u8 = x87_setround(dyn, ninst, x1, x2, x4);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
+                ed = x1;
+                s0 = fpu_get_scratch(dyn);
+                #if 0
+                FRINT32XD(s0, v1);
+                // no saturation instruction on Arm, so using NEON
+                VFCVTZSd(s0, s0);
+                SQXTN_S_D(s0, s0);
+                SQXTN_H_S(s0, s0);
+                VST16(s0, wback, fixedaddress);
+                #else
+                MRS_fpsr(x5);
+                BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
+                BFCw(x5, FPSR_QC, 1);   // reset QC bit
+                MSR_fpsr(x5);
+                if(ST_IS_F(0)) {
+                    FRINTXS(s0, v1);
+                    VFCVTZSs(s0, s0);
+                } else {
+                    FRINTXD(s0, v1);
                    VFCVTZSd(s0, s0);
                    SQXTN_S_D(s0, s0);
-                    SQXTN_H_S(s0, s0);
-                    VST16(s0, wback, fixedaddress);
-                    #else
-                    MRS_fpsr(x5);
-                    BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
-                    BFCw(x5, FPSR_QC, 1);   // reset QC bit
-                    MSR_fpsr(x5);
-                    if(ST_IS_F(0)) {
-                        VFCVTZSs(s0, v1);
-                    } else {
-                        VFCVTZSd(s0, v1);
-                        SQXTN_S_D(s0, s0);
-                    }
-                    VMOVSto(x3, s0, 0);
-                    MRS_fpsr(x5);   // get back FPSR to check the IOC bit
-                    TBNZ_MARK2(x5, FPSR_IOC);
-                    SXTHw(x5, x3);  // check if 16bits value is fine
-                    SUBw_REG(x5, x5, x3);
-                    CBZw_MARK3(x5);
-                    MARK2;
-                    MOV32w(x3, 0x8000);
-                    MARK3;
-                    STH(x3, wback, fixedaddress);
-                    #endif
-                    x87_do_pop(dyn, ninst, x3);
-                    break;
-                case 2:
-                    INST_NAME("FIST Ew, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
-                    u8 = x87_setround(dyn, ninst, x1, x2, x4);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
-                    ed = x1;
-                    s0 = fpu_get_scratch(dyn);
-                    #if 0
-                    FRINT32XD(s0, v1);
-                    // no saturation instruction on Arm, so using NEON
+                }
+                VMOVSto(x3, s0, 0);
+                MRS_fpsr(x5);   // get back FPSR to check the IOC bit
+                TBNZ_MARK2(x5, FPSR_IOC);
+                SXTHw(x5, x3);  // check if 16bits value is fine
+                SUBw_REG(x5, x5, x3);
+                CBZw_MARK3(x5);
+                MARK2;
+                MOV32w(x3, 0x8000);
+                MARK3;
+                STH(x3, wback, fixedaddress);
+                #endif
+                x87_restoreround(dyn, ninst, u8);
+                break;
+            case 3:
+                INST_NAME("FISTP Ew, ST0");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                u8 = x87_setround(dyn, ninst, x1, x2, x4);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
+                ed = x1;
+                s0 = fpu_get_scratch(dyn);
+                #if 0
+                FRINT32XD(s0, v1);
+                // no saturation instruction on Arm, so using NEON
+                VFCVTZSd(s0, s0);
+                SQXTN_S_D(s0, s0);
+                SQXTN_H_S(s0, s0);
+                VST16(s0, wback, fixedaddress);
+                #else
+                MRS_fpsr(x5);
+                BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
+                MSR_fpsr(x5);
+                if(ST_IS_F(0)) {
+                    FRINTXS(s0, v1);
+                    VFCVTZSs(s0, s0);
+                } else {
+                    FRINTXD(s0, v1);
                    VFCVTZSd(s0, s0);
                    SQXTN_S_D(s0, s0);
-                    SQXTN_H_S(s0, s0);
-                    VST16(s0, wback, fixedaddress);
-                    #else
-                    MRS_fpsr(x5);
-                    BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
-                    BFCw(x5, FPSR_QC, 1);   // reset QC bit
-                    MSR_fpsr(x5);
-                    if(ST_IS_F(0)) {
-                        FRINTXS(s0, v1);
-                        VFCVTZSs(s0, s0);
-                    } else {
-                        FRINTXD(s0, v1);
-                        VFCVTZSd(s0, s0);
-                        SQXTN_S_D(s0, s0);
-                    }
-                    VMOVSto(x3, s0, 0);
-                    MRS_fpsr(x5);   // get back FPSR to check the IOC bit
-                    TBNZ_MARK2(x5, FPSR_IOC);
-                    SXTHw(x5, x3);  // check if 16bits value is fine
-                    SUBw_REG(x5, x5, x3);
-                    CBZw_MARK3(x5);
-                    MARK2;
-                    MOV32w(x3, 0x8000);
-                    MARK3;
-                    STH(x3, wback, fixedaddress);
-                    #endif
-                    x87_restoreround(dyn, ninst, u8);
-                    break;
-                case 3:
-                    INST_NAME("FISTP Ew, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
-                    u8 = x87_setround(dyn, ninst, x1, x2, x4);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
-                    ed = x1;
-                    s0 = fpu_get_scratch(dyn);
-                    #if 0
-                    FRINT32XD(s0, v1);
-                    // no saturation instruction on Arm, so using NEON
-                    VFCVTZSd(s0, s0);
-                    SQXTN_S_D(s0, s0);
-                    SQXTN_H_S(s0, s0);
-                    VST16(s0, wback, fixedaddress);
-                    #else
-                    MRS_fpsr(x5);
-                    BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
-                    MSR_fpsr(x5);
-                    if(ST_IS_F(0)) {
-                        FRINTXS(s0, v1);
-                        VFCVTZSs(s0, s0);
-                    } else {
-                        FRINTXD(s0, v1);
-                        VFCVTZSd(s0, s0);
-                        SQXTN_S_D(s0, s0);
-                    }
-                    VMOVSto(x3, s0, 0);
-                    MRS_fpsr(x5);   // get back FPSR to check the IOC bit
-                    TBNZ_MARK2(x5, FPSR_IOC);
-                    SXTHw(x5, x3);  // check if 16bits value is fine
-                    SUBw_REG(x5, x5, x3);
-                    CBZw_MARK3(x5);
-                    MARK2;
-                    MOV32w(x3, 0x8000);
-                    MARK3;
-                    STH(x3, wback, fixedaddress);
-                    #endif
-                    x87_do_pop(dyn, ninst, x3);
-                    x87_restoreround(dyn, ninst, u8);
-                    break;
-                case 4:
-                    INST_NAME("FBLD ST0, tbytes");
-                    x87_do_push_empty(dyn, ninst, x1);
-                    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_fbld, -1);
-                    break;
-                case 5:
-                    INST_NAME("FILD ST0, i64");
-                    v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_I64);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                    VLD64(v1, wback, fixedaddress);
-                    if(!ST_IS_I64(0)) {
-                        if(rex.is32bits) {
-                            // need to also feed the STll stuff...
-                            ADDx_U12(x4, xEmu, offsetof(x64emu_t, fpu_ll));
-                            LDRw_U12(x1, xEmu, offsetof(x64emu_t, top));
-                            int a = 0 - dyn->n.x87stack;
-                            if(a) {
-                                if(a<0) {
-                                    SUBw_U12(x1, x1, -a);
-                                } else {
-                                    ADDw_U12(x1, x1, a);
-                                }
-                                ANDw_mask(x1, x1, 0, 2); //mask=7
+                }
+                VMOVSto(x3, s0, 0);
+                MRS_fpsr(x5);   // get back FPSR to check the IOC bit
+                TBNZ_MARK2(x5, FPSR_IOC);
+                SXTHw(x5, x3);  // check if 16bits value is fine
+                SUBw_REG(x5, x5, x3);
+                CBZw_MARK3(x5);
+                MARK2;
+                MOV32w(x3, 0x8000);
+                MARK3;
+                STH(x3, wback, fixedaddress);
+                #endif
+                X87_POP_OR_FAIL(dyn, ninst, x3);
+                x87_restoreround(dyn, ninst, u8);
+                break;
+            case 4:
+                INST_NAME("FBLD ST0, tbytes");
+                X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, x1);
+                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);
+                x87_unstackcount(dyn, ninst, x3, s0);
+                break;
+            case 5:
+                INST_NAME("FILD ST0, i64");
+                X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_I64);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
+                VLD64(v1, wback, fixedaddress);
+                if(!ST_IS_I64(0)) {
+                    if(rex.is32bits) {
+                        // need to also feed the STll stuff...
+                        ADDx_U12(x4, xEmu, offsetof(x64emu_t, fpu_ll));
+                        LDRw_U12(x1, xEmu, offsetof(x64emu_t, top));
+                        int a = 0 - dyn->n.x87stack;
+                        if(a) {
+                            if(a<0) {
+                                SUBw_U12(x1, x1, -a);
+                            } else {
+                                ADDw_U12(x1, x1, a);
                            }
-                            ADDx_REG_LSL(x1, x4, x1, 4);    // fpu_ll is 2 i64
-                            VSTR64_U12(v1, x1, 8);  // ll
-                        }
-                        SCVTFDD(v1, v1);
-                        if(rex.is32bits) {
-                            VSTR64_U12(v1, x1, 0);  // ref
+                            ANDw_mask(x1, x1, 0, 2); //mask=7
                        }
+                        ADDx_REG_LSL(x1, x4, x1, 4);    // fpu_ll is 2 i64
+                        VSTR64_U12(v1, x1, 8);  // ll
                    }
-                    break;
-                case 6:
-                    INST_NAME("FBSTP tbytes, ST0");
-                    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);
-                    x87_do_pop(dyn, ninst, x3);
-                    break;
-                case 7:
-                    INST_NAME("FISTP i64, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_I64);
-                    if(!ST_IS_I64(0)) {
-                        u8 = x87_setround(dyn, ninst, x1, x2, x4);
+                    SCVTFDD(v1, v1);
+                    if(rex.is32bits) {
+                        VSTR64_U12(v1, x1, 0);  // ref
                    }
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                    ed = x1;
-                    s0 = fpu_get_scratch(dyn);
-                    if(ST_IS_I64(0)) {
-                        VST64(v1, wback, fixedaddress);
-                    } else {
-                        #if 0
-                        FRINT64XD(s0, v1);
-                        VFCVTZSd(s0, s0);
-                        VSTR64_U12(s0, wback, fixedaddress);
-                        #else
-                        if(rex.is32bits) {
-                            // need to check STll first...
-                            ADDx_U12(x5, xEmu, offsetof(x64emu_t, fpu_ll));
-                            LDRw_U12(x1, xEmu, offsetof(x64emu_t, top));
-                            VMOVQDto(x3, v1, 0);
-                            int a = 0 - dyn->n.x87stack;
-                            if(a) {
-                                if(a<0) {
-                                    SUBw_U12(x1, x1, -a);
-                                } else {
-                                    ADDw_U12(x1, x1, a);
-                                }
-                                ANDw_mask(x1, x1, 0, 2); //mask=7
+                }
+                break;
+            case 6:
+                INST_NAME("FBSTP tbytes, ST0");
+                i1 = x87_stackcount(dyn, ninst, x1);
+                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);
+                x87_unstackcount(dyn, ninst, x1, i1);
+                X87_POP_OR_FAIL(dyn, ninst, x3);
+                break;
+            case 7:
+                INST_NAME("FISTP i64, ST0");
+                v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_I64);
+                if(!ST_IS_I64(0)) {
+                    u8 = x87_setround(dyn, ninst, x1, x2, x4);
+                }
+                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
+                ed = x1;
+                s0 = fpu_get_scratch(dyn);
+                if(ST_IS_I64(0)) {
+                    VST64(v1, wback, fixedaddress);
+                } else {
+                    #if 0
+                    FRINT64XD(s0, v1);
+                    VFCVTZSd(s0, s0);
+                    VSTR64_U12(s0, wback, fixedaddress);
+                    #else
+                    if(rex.is32bits) {
+                        // need to check STll first...
+                        ADDx_U12(x5, xEmu, offsetof(x64emu_t, fpu_ll));
+                        LDRw_U12(x1, xEmu, offsetof(x64emu_t, top));
+                        VMOVQDto(x3, v1, 0);
+                        int a = 0 - dyn->n.x87stack;
+                        if(a) {
+                            if(a<0) {
+                                SUBw_U12(x1, x1, -a);
+                            } else {
+                                ADDw_U12(x1, x1, a);
                            }
-                            ADDx_REG_LSL(x1, x5, x1, 4);    // fpu_ll is 2 i64
-                            LDRx_U12(x5, x1, 0);  // ref
-                            SUBx_REG(x5, x5, x3);
-                            CBNZx_MARK2(x5);
-                            LDRx_U12(x5, x1, 8);  // ll
-                            STx(x5, wback, fixedaddress);
-                            B_MARK3(c__);
-                            MARK2;
+                            ANDw_mask(x1, x1, 0, 2); //mask=7
                        }
-                        MRS_fpsr(x5);
-                        BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
-                        MSR_fpsr(x5);
-                        FRINTXD(s0, v1);
-                        VFCVTZSd(s0, s0);
-                        VST64(s0, wback, fixedaddress);
-                        MRS_fpsr(x5);   // get back FPSR to check the IOC bit
-                        TBZ_MARK3(x5, FPSR_IOC);
-                        ORRx_mask(x5, xZR, 1, 1, 0);    //0x8000000000000000
+                        ADDx_REG_LSL(x1, x5, x1, 4);    // fpu_ll is 2 i64
+                        LDRx_U12(x5, x1, 0);  // ref
+                        SUBx_REG(x5, x5, x3);
+                        CBNZx_MARK2(x5);
+                        LDRx_U12(x5, x1, 8);  // ll
                        STx(x5, wback, fixedaddress);
-                        MARK3;
-                        #endif
-                        x87_restoreround(dyn, ninst, u8);
+                        B_MARK3(c__);
+                        MARK2;
                    }
-                    x87_do_pop(dyn, ninst, x3);
-                    break;
-                default:
-                    DEFAULT;
-            }
-    }
+                    MRS_fpsr(x5);
+                    BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
+                    MSR_fpsr(x5);
+                    FRINTXD(s0, v1);
+                    VFCVTZSd(s0, s0);
+                    VST64(s0, wback, fixedaddress);
+                    MRS_fpsr(x5);   // get back FPSR to check the IOC bit
+                    TBZ_MARK3(x5, FPSR_IOC);
+                    ORRx_mask(x5, xZR, 1, 1, 0);    //0x8000000000000000
+                    STx(x5, wback, fixedaddress);
+                    MARK3;
+                    #endif
+                    x87_restoreround(dyn, ninst, u8);
+                }
+                X87_POP_OR_FAIL(dyn, ninst, x3);
+                break;
+            default:
+                DEFAULT;
+        }
    return addr;
 }
@@ -60,8 +60,13 @@ void emit_or32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int64_t c, int
    } else IFX(X_ALL) {
        SET_DFNONE(s4);
    }
-    MOV64xw(s3, c);
-    ORRxw_REG(s1, s1, s3);
+    int mask = convert_bitmask_xw(c);
+    if(mask) {
+        ORRxw_mask(s1, s1, (mask>>12)&1, mask&0x3F, (mask>>6)&0x3F);
+    } else {
+        MOV64xw(s3, c);
+        ORRxw_REG(s1, s1, s3);
+    }
    IFX(X_PEND) {
        STRxw_U12(s1, xEmu, offsetof(x64emu_t, res));
    }
@@ -122,8 +127,13 @@ void emit_xor32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int64_t c, in
    } else IFX(X_ALL) {
        SET_DFNONE(s4);
    }
-    MOV64xw(s3, c);
-    EORxw_REG(s1, s1, s3);
+    int mask = convert_bitmask_xw(c);
+    if(mask) {
+        EORxw_mask(s1, s1, (mask>>12)&1, mask&0x3F, (mask>>6)&0x3F);
+    } else {
+        MOV64xw(s3, c);
+        EORxw_REG(s1, s1, s3);
+    }
    IFX(X_PEND) {
        STRxw_U12(s1, xEmu, offsetof(x64emu_t, res));
    }
@@ -187,11 +197,20 @@ void emit_and32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int64_t c, in
    } else IFX(X_ALL) {
        SET_DFNONE(s4);
    }
-    MOV64xw(s3, c);
-    IFX(X_ALL) {
-        ANDSxw_REG(s1, s1, s3);
+    int mask = convert_bitmask_xw(c);
+    if(mask) {
+        IFX(X_ALL) {
+            ANDSxw_mask(s1, s1, (mask>>12)&1, mask&0x3F, (mask>>6)&0x3F);
+        } else {
+            ANDxw_mask(s1, s1, (mask>>12)&1, mask&0x3F, (mask>>6)&0x3F);
+        }
    } else {
-        ANDxw_REG(s1, s1, s3);
+        MOV64xw(s3, c);
+        IFX(X_ALL) {
+            ANDSxw_REG(s1, s1, s3);
+        } else {
+            ANDxw_REG(s1, s1, s3);
+        }
    }
    IFX(X_PEND) {
        STRxw_U12(s1, xEmu, offsetof(x64emu_t, res));
@@ -230,15 +249,7 @@ void emit_or8(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(xFlags, xFlags, s3);
    }
-    IFX(X_ZF) {
-        TSTw_REG(s1, s1);
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -247,13 +258,18 @@ void emit_or8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 // emit OR8 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
 void emit_or8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
 {
-    MOV32w(s3, c&0xff);
    IFX(X_PEND) {
        SET_DF(s4, d_or8);
    } else IFX(X_ALL) {
        SET_DFNONE(s4);
    }
-    ORRw_REG(s1, s1, s3);
+    int mask = convert_bitmask_w(c);
+    if(mask) {
+        ORRw_mask(s1, s1, mask&0x3F, (mask>>6)&0x3F);
+    } else {
+        MOV32w(s3, c&0xff);
+        ORRw_REG(s1, s1, s3);
+    }
    IFX(X_PEND) {
        STRB_U12(s1, xEmu, offsetof(x64emu_t, res));
    }
@@ -261,15 +277,7 @@ void emit_or8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
        MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
        BICw(xFlags, xFlags, s3);
    }
-    IFX(X_ZF) {
-        TSTw_REG(s1, s1);
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -292,15 +300,7 @@ void emit_xor8(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(xFlags, xFlags, s3);
    }
-    IFX(X_ZF) {
-        TSTw_REG(s1, s1);
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -309,13 +309,18 @@ void emit_xor8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 // emit XOR8 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
 void emit_xor8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
 {
-    MOV32w(s3, c&0xff);
    IFX(X_PEND) {
        SET_DF(s4, d_xor8);
    } else IFX(X_ALL) {
        SET_DFNONE(s4);
    }
-    EORw_REG(s1, s1, s3);
+    int mask = convert_bitmask_w(c);
+    if(mask) {
+        EORw_mask(s1, s1, mask&0x3F, (mask>>6)&0x3F);
+    } else {
+        MOV32w(s3, c&0xff);
+        EORw_REG(s1, s1, s3);
+    }
    IFX(X_PEND) {
        STRB_U12(s1, xEmu, offsetof(x64emu_t, res));
    }
@@ -323,15 +328,7 @@ void emit_xor8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4
        MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
        BICw(xFlags, xFlags, s3);
    }
-    IFX(X_ZF) {
-        TSTw_REG(s1, s1);
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -374,16 +371,25 @@ void emit_and8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 // emit AND8 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
 void emit_and8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
 {
-    MOV32w(s3, c&0xff);
    IFX(X_PEND) {
        SET_DF(s4, d_and8);
    } else IFX(X_ALL) {
        SET_DFNONE(s4);
    }
-    IFX(X_ZF) {
-        ANDSw_REG(s1, s1, s3);
+    int mask = convert_bitmask_w(c);
+    if(mask) {
+        IFX(X_ZF) {
+            ANDSw_mask(s1, s1, mask&0x3F, (mask>>6)&0x3F);
+        } else {
+            ANDw_mask(s1, s1, mask&0x3F, (mask>>6)&0x3F);
+        }
    } else {
-        ANDw_REG(s1, s1, s3);
+        MOV32w(s3, c&0xff);
+        IFX(X_ZF) {
+            ANDSw_REG(s1, s1, s3);
+        } else {
+            ANDw_REG(s1, s1, s3);
+        }
    }
    IFX(X_PEND) {
        STRB_U12(s1, xEmu, offsetof(x64emu_t, res));
@@ -423,15 +429,7 @@ void emit_or16(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(xFlags, xFlags, s3);
    }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 15);  //mask=0xffff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 15);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 16)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -500,15 +498,7 @@ void emit_xor16(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(xFlags, xFlags, s3);
    }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 15);  //mask=0xffff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 15);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 16)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -297,21 +297,12 @@ void emit_add8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        }
    }
    IFX(X_CF) {
-        LSRw(s3, s1, 8);
-        BFIw(xFlags, s3, F_CF, 1);
+        BFXILw(xFlags, s1, 8, 1);
    }
    IFX(X_PEND) {
        STRH_U12(s1, xEmu, offsetof(x64emu_t, res));
    }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 7);   //mask=0xff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -329,7 +320,7 @@ void emit_add8c(dynarec_arm_t* dyn, int ninst, int s1, int c, int s3, int s4)
        SET_DFNONE(s3);
    }
    IFX(X_AF | X_OF) {
-        if(X_PEND) {} else {MOV32w(s4, c&0xff);}
+        IFX(X_PEND) {} else {MOV32w(s4, c&0xff);}
        ORRw_REG(s3, s1, s4);       // s3 = op1 | op2
        ANDw_REG(s4, s1, s4);       // s4 = op1 & op2
    }
@@ -349,21 +340,12 @@ void emit_add8c(dynarec_arm_t* dyn, int ninst, int s1, int c, int s3, int s4)
        }
    }
    IFX(X_CF) {
-        LSRw(s3, s1, 8);
-        BFIw(xFlags, s3, F_CF, 1);
+        BFXILw(xFlags, s1, 8, 1);
    }
    IFX(X_PEND) {
        STRH_U12(s1, xEmu, offsetof(x64emu_t, res));
    }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 0b000111);    //mask=000000ff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -407,15 +389,7 @@ void emit_sub8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
            BFIw(xFlags, s4, F_OF, 1);    // OF: ((bc >> 6) ^ ((bc>>6)>>1)) & 1
        }
    }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 7);   //mask=0xff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -465,15 +439,7 @@ void emit_sub8c(dynarec_arm_t* dyn, int ninst, int s1, int c, int s3, int s4, in
            BFIw(xFlags, s4, F_OF, 1);    // OF: ((bc >> 6) ^ ((bc>>6)>>1)) & 1
        }
    }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 0b000111);    //mask=000000ff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -510,21 +476,12 @@ void emit_add16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        }
    }
    IFX(X_CF) {
-        LSRw(s3, s1, 16);
-        BFIw(xFlags, s3, F_CF, 1);
+        BFXILw(xFlags, s1, 16, 1);
    }
    IFX(X_PEND) {
        STRw_U12(s1, xEmu, offsetof(x64emu_t, res));
    }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 15);  //mask=0xffff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 15);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 16)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -634,15 +591,7 @@ void emit_sub16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
            BFIw(xFlags, s4, F_OF, 1);    // OF: ((bc >> 14) ^ ((bc>>14)>>1)) & 1
        }
    }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 15);  //mask=0xffff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 15);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 16)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -792,16 +741,7 @@ void emit_inc8(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
            BFIw(xFlags, s4, F_OF, 1);    // OF: ((bc >> 6) ^ ((bc>>6)>>1)) & 1
        }
    }
-
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 7);   //mask=0xff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -838,15 +778,7 @@ void emit_inc16(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
            BFIw(xFlags, s4, F_OF, 1);    // OF: ((bc >> 14) ^ ((bc>>14)>>1)) & 1
        }
    }
-    IFX(X_ZF) {
-        TSTw_mask(s1, 0, 0b001111); // mask=0xffff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 15);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 16)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -906,7 +838,7 @@ void emit_dec32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s3, int s4
 void emit_dec8(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
 {
    IFX(X_PEND) {
-        STRB_U12(s3, xEmu, offsetof(x64emu_t, op2));
+        STRB_U12(s1, xEmu, offsetof(x64emu_t, op1));
        SET_DF(s3, d_dec8);
    } else IFX(X_ZF|X_OF|X_AF|X_SF|X_PF) {
        SET_DFNONE(s3);
@@ -1162,15 +1094,7 @@ void emit_adc8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        LSRw(s3, s1, 8);
        BFIw(xFlags, s3, F_CF, 1);
    }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 7);   //mask=0xff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -1224,15 +1148,7 @@ void emit_adc16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        LSRw(s3, s1, 16);
        BFIw(xFlags, s3, F_CF, 1);
    }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 15);   //mask=0xffff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 15);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 16)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -1474,15 +1390,7 @@ void emit_sbb8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
            BFIw(xFlags, s4, F_OF, 1);    // OF: ((bc >> 6) ^ ((bc>>6)>>1)) & 1
        }
    }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 7);   //mask=0xff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -1537,15 +1445,7 @@ void emit_sbb16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
            BFIw(xFlags, s4, F_OF, 1);    // OF: ((bc >> 14) ^ ((bc>>14)>>1)) & 1
        }
    }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 15);  //mask=0xffff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 15);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 16)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -1732,10 +1632,9 @@ void emit_neg8(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
    IFX(X_AF|X_OF) {
        MOVw_REG(s3, s1);
    }
+    NEGw_REG(s1, s1);
    IFX(X_ZF) {
-        NEGSw_REG(s1, s1);
-    } else {
-        NEGw_REG(s1, s1);
+        ANDSw_mask(s1, s1, 0, 7);   // mask 0xff
    }
    IFX(X_PEND) {
        STRB_U12(s1, xEmu, offsetof(x64emu_t, res));
@@ -111,22 +111,11 @@ void emit_cmp16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, i
    } else {
        SET_DFNONE(s3);
    }
-    IFX(X_ZF) {
-        SUBSw_REG(s5, s1, s2);   // res = s1 - s2
-    } else {
-        SUBw_REG(s5, s1, s2);   // res = s1 - s2
-    }
+    SUBw_REG(s5, s1, s2);   // res = s1 - s2
    IFX_PENDOR0 {
        STRH_U12(s5, xEmu, offsetof(x64emu_t, res));
    }
-    IFX(X_ZF) {
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s4, s5, 15);
-        BFIw(xFlags, s4, F_SF, 1);
-    }
+    COMP_ZFSF(s5, 16)
    // bc = (res & (~d | s)) | (~d & s)
    IFX(X_CF|X_AF|X_OF) {
        MVNw_REG(s4, s1);        // s4 = ~d
@@ -170,15 +159,7 @@ void emit_cmp16_0(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
        MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
        BICw(xFlags, xFlags, s3);
    }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 15);  //mask=0xffff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s4, s1, 15);
-        BFIw(xFlags, s4, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 16)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
@@ -194,22 +175,11 @@ void emit_cmp8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, in
    } else {
        SET_DFNONE(s4);
    }
-    IFX(X_ZF) {
-        SUBSw_REG(s5, s1, s2);   // res = s1 - s2
-    } else {
-        SUBw_REG(s5, s1, s2);   // res = s1 - s2
-    }
+    SUBw_REG(s5, s1, s2);   // res = s1 - s2
    IFX_PENDOR0 {
        STRB_U12(s5, xEmu, offsetof(x64emu_t, res));
    }
-    IFX(X_ZF) {
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s5, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s5, 8)
    // bc = (res & (~d | s)) | (~d & s)
    IFX(X_CF|X_AF|X_OF) {
        ORNw_REG(s4, s2, s1);   // s4 = ~d | s
@@ -251,22 +221,14 @@ void emit_cmp8_0(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
        MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
        BICw(xFlags, xFlags, s3);
    }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 0b000111);    //mask=000000ff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
    IFX(X_PF) {
        emit_pf(dyn, ninst, s1, s3, s4);
    }
 }

-// emit TEST32 instruction, from test s1, s2, using s3 and s4 as scratch
-void emit_test32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4)
+// emit TEST32 instruction, from test s1, s2, using s3, s4 and s5 as scratch
+void emit_test32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4, int s5)
 {
    MAYUSE(s1); MAYUSE(s2); MAYUSE(s3); MAYUSE(s4);
    IFX_PENDOR0 {
@@ -292,18 +254,11 @@ void emit_test32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s
    }
    // PF: (((emu->x64emu_parity_tab[(res) / 32] >> ((res) % 32)) & 1) == 0)
    IFX(X_PF) {
-        ANDw_mask(s3, s3, 0b011011, 0b000010); // 0xE0
-        LSRw(s3, s3, 5);
-        MOV64x(s4, (uintptr_t)GetParityTab());
-        LDRw_REG_LSL2(s4, s4, s3);
-        ANDw_mask(s3, s1, 0, 0b000100);   // 0x1f
-        LSRw_REG(s4, s4, s3);
-        MVNx_REG(s4, s4);
-        BFIw(xFlags, s4, F_PF, 1);
+        emit_pf(dyn, ninst, s3, s4, s5);
    }
 }

-// emit TEST16 instruction, from test s1, s2, using s3 and s4 as scratch
+// emit TEST16 instruction, from test s1, s2, using s3, s4 and s5 as scratch
 void emit_test16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, int s5)
 {
    MAYUSE(s1); MAYUSE(s2);
@@ -334,7 +289,7 @@ void emit_test16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4,
    }
 }

-// emit TEST8 instruction, from test s1, s2, using s3 and s4 as scratch
+// emit TEST8 instruction, from test s1, s2, using s3, s4 and s5 as scratch
 void emit_test8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, int s5)
 {
    MAYUSE(s1); MAYUSE(s2);
@@ -28,7 +28,7 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
    uint8_t opcode = F8;
    uint8_t nextop;
    uint8_t gd, ed;
-    uint8_t wback;
+    uint8_t wback, wb1, wb2;
    uint8_t u8;
    uint64_t u64, j64;
    int v0, v1;
@@ -71,6 +71,7 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                d0 = sse_get_reg(dyn, ninst, x1, ed, 1);
                VMOVeD(d0, 0, v0, 0);
            } else {
+                WILLWRITE2();
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
                VST64(v0, ed, fixedaddress);
                SMWRITE2();
@@ -159,30 +160,41 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            switch(opcode) {

                case 0xF0:
-                    INST_NAME("(unsupported) CRC32 Gd, Eb)");
+                    INST_NAME("CRC32 Gd, Eb");
                    nextop = F8;
-                    addr = fakeed(dyn, addr, ninst, nextop);
-                    SETFLAGS(X_ALL, SF_SET);    // Hack to set flags in "don't care" state
-                    GETIP(ip);
-                    STORE_XEMU_CALL(xRIP);
-                    CALL(native_ud, -1);
-                    LOAD_XEMU_CALL(xRIP);
-                    jump_to_epilog(dyn, 0, xRIP, ninst);
-                    *need_epilog = 0;
-                    *ok = 0;
+                    GETEB(x1, 0);
+                    GETGD;
+                    if(arm64_crc32) {
+                        CRC32CB(gd, gd, ed);
+                    } else {
+                        EORw_REG(gd, gd, ed);
+                        MOV32w(x2, 0x82f63b78);
+                        for(int i=0; i<8; ++i) {
+                            LSRw_IMM((i&1)?gd:x4, (i&1)?x4:gd, 1);
+                            TBZ((i&1)?x4:gd, 0, 4+4);
+                            EORw_REG((i&1)?gd:x4, (i&1)?gd:x4, x2);
+                        }
+                    }
                    break;
                case 0xF1:
-                    INST_NAME("(unsupported) CRC32 Gd, Ed)");
+                    INST_NAME("CRC32 Gd, Ed");
                    nextop = F8;
-                    addr = fakeed(dyn, addr, ninst, nextop);
-                    SETFLAGS(X_ALL, SF_SET);    // Hack to set flags in "don't care" state
-                    GETIP(ip);
-                    STORE_XEMU_CALL(xRIP);
-                    CALL(native_ud, -1);
-                    LOAD_XEMU_CALL(xRIP);
-                    jump_to_epilog(dyn, 0, xRIP, ninst);
-                    *need_epilog = 0;
-                    *ok = 0;
+                    GETED(0);
+                    GETGD;
+                    if(arm64_crc32) {
+                        CRC32Cxw(gd, gd, ed);
+                    } else {
+                        MOV32w(x2, 0x82f63b78);
+                        for(int j=0; j<4*(1+rex.w); ++j) {
+                            UBFXxw(x3, ed, 8*j, 8);
+                            EORw_REG(gd, gd, x3);
+                            for(int i=0; i<8; ++i) {
+                                LSRw_IMM((i&1)?gd:x4, (i&1)?x4:gd, 1);
+                                TBZ((i&1)?x4:gd, 0, 4+4);
+                                EORw_REG((i&1)?gd:x4, (i&1)?gd:x4, x2);
+                            }
+                        }
+                    }
                    break;

                default:
@@ -347,17 +359,27 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            GETGX(v0, 1);

            u8 = F8;
-            // only low part need to be suffled. VTBL only handle 8bits value, so the 16bits suffles need to be changed in 8bits
-            u64 = 0;
-            for (int i=0; i<4; ++i) {
-                u64 |= ((uint64_t)((u8>>(i*2))&3)*2+0)<<(i*16+0);
-                u64 |= ((uint64_t)((u8>>(i*2))&3)*2+1)<<(i*16+8);
-            }
-            MOV64x(x2, u64);
-            d0 = fpu_get_scratch(dyn);
-            VMOVQDfrom(d0, 0, x2);
-            VTBL1_8(d0, v1, d0);
-            VMOVeD(v0, 0, d0, 0);
+            if(u8==0b00000000 || u8==0b01010101 || u8==0b10101010 || u8==0b11111111) {
+                if(v0==v1) {
+                    d0 = fpu_get_scratch(dyn);
+                    VMOVQ(d0, v1);
+                }
+                VDUP_16(v0, v1, u8&3);
+                if(v0==v1)
+                    v1 = d0;
+            } else {
+                // only low part need to be suffled. VTBL only handle 8bits value, so the 16bits suffles need to be changed in 8bits
+                u64 = 0;
+                for (int i=0; i<4; ++i) {
+                    u64 |= ((uint64_t)((u8>>(i*2))&3)*2+0)<<(i*16+0);
+                    u64 |= ((uint64_t)((u8>>(i*2))&3)*2+1)<<(i*16+8);
+                }
+                MOV64x(x2, u64);
+                d0 = fpu_get_scratch(dyn);
+                VMOVQDfrom(d0, 0, x2);
+                VTBL1_8(d0, v1, d0);
+                VMOVeD(v0, 0, d0, 0);
+                }
            if(v0!=v1) {
                VMOVeD(v0, 1, v1, 1);
            }
@@ -376,7 +398,22 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            }
            VFADDPQS(v0, v0, v1);
            break;
-            
+        case 0x7D:
+            INST_NAME("HSUBPS Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            if(MODREG) {
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                v1 = fpu_get_scratch(dyn);
+                VLD128(v1, ed, fixedaddress);
+            }
+            d0 = fpu_get_scratch(dyn);
+            VUZP1Q_32(d0, v0, v1);
+            VUZP2Q_32(v0, v0, v1);
+            VFSUBQS(v0, d0, v0);
+            break;
        case 0xC2:
            INST_NAME("CMPSD Gx, Ex, Ib");
            nextop = F8;
@@ -312,7 +312,7 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            VMOVeS(v0, 0, v1, 0);   // to not erase uper part
            #endif
            break;
-            
+
        case 0x6F:
            INST_NAME("MOVDQU Gx,Ex");// no alignment constraint on NEON here, so same as MOVDQA
            nextop = F8;
@@ -333,16 +333,20 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            GETEX(v1, 0, 1) ;
            GETGX(v0, 1);
            u8 = F8;
-            // only high part need to be suffled. VTBL only handle 8bits value, so the 16bits suffles need to be changed in 8bits
-            u64 = 0;
-            for (int i=0; i<4; ++i) {
-                u64 |= ((uint64_t)((u8>>(i*2))&3)*2+8)<<(i*16+0);
-                u64 |= ((uint64_t)((u8>>(i*2))&3)*2+9)<<(i*16+8);
-            }
-            MOV64x(x2, u64);
            d0 = fpu_get_scratch(dyn);
-            VMOVQDfrom(d0, 0, x2);
-            VTBL1_8(d0, v1, d0);
+            if (u8 == 0b00000000 || u8 == 0b01010101 || u8 == 0b10101010 || u8 == 0b11111111) {
+                VDUPQ_16(d0, v1, (u8 & 3) + 4);
+            } else {
+                // only high part need to be suffled. VTBL only handle 8bits value, so the 16bits suffles need to be changed in 8bits
+                u64 = 0;
+                for (int i = 0; i < 4; ++i) {
+                    u64 |= ((uint64_t)((u8 >> (i * 2)) & 3) * 2 + 8) << (i * 16 + 0);
+                    u64 |= ((uint64_t)((u8 >> (i * 2)) & 3) * 2 + 9) << (i * 16 + 8);
+                }
+                MOV64x(x2, u64);
+                VMOVQDfrom(d0, 0, x2);
+                VTBL1_8(d0, v1, d0);
+            }
            VMOVeD(v0, 1, d0, 0);
            if(v0!=v1) {
                VMOVeD(v0, 0, v1, 0);
@@ -30,7 +30,6 @@
 #define XMM8    16
 #define X870    8
 #define EMM0    8
-#define SCRATCH0    24

 // Get a FPU scratch reg
 int fpu_get_scratch(dynarec_arm_t* dyn)
@@ -86,15 +85,28 @@ int fpu_get_reg_xmm(dynarec_arm_t* dyn, int t, int xmm)
    return i;
 }
 // Reset fpu regs counter
-void fpu_reset_reg(dynarec_arm_t* dyn)
+static void fpu_reset_reg_neoncache(neoncache_t* n)
 {
-    dyn->n.fpu_reg = 0;
+    n->fpu_reg = 0;
    for (int i=0; i<24; ++i) {
-        dyn->n.fpuused[i]=0;
-        dyn->n.neoncache[i].v = 0;
+        n->fpuused[i]=0;
+        n->neoncache[i].v = 0;
    }

 }
+void fpu_reset_reg(dynarec_arm_t* dyn)
+{
+    fpu_reset_reg_neoncache(&dyn->n);
+}
+
+int neoncache_no_i64(dynarec_arm_t* dyn, int ninst, int st, int a)
+{
+    if(a==NEON_CACHE_ST_I64) {
+        neoncache_promote_double(dyn, ninst, st);
+        return NEON_CACHE_ST_D;
+    }
+    return a;
+}

 int neoncache_get_st(dynarec_arm_t* dyn, int ninst, int a)
 {
@@ -139,7 +151,7 @@ int neoncache_get_st_f(dynarec_arm_t* dyn, int ninst, int a)
         && dyn->insts[ninst].n.neoncache[i].n==a)
            return i;
    return -1;
-} 
+}
 int neoncache_get_st_f_i64(dynarec_arm_t* dyn, int ninst, int a)
 {
    /*if(a+dyn->insts[ninst].n.stack_next-st<0)
@@ -150,7 +162,7 @@ int neoncache_get_st_f_i64(dynarec_arm_t* dyn, int ninst, int a)
         && dyn->insts[ninst].n.neoncache[i].n==a)
            return i;
    return -1;
-} 
+}
 int neoncache_get_st_f_noback(dynarec_arm_t* dyn, int ninst, int a)
 {
    for(int i=0; i<24; ++i)
@@ -158,7 +170,7 @@ int neoncache_get_st_f_noback(dynarec_arm_t* dyn, int ninst, int a)
         && dyn->insts[ninst].n.neoncache[i].n==a)
            return i;
    return -1;
-} 
+}
 int neoncache_get_st_f_i64_noback(dynarec_arm_t* dyn, int ninst, int a)
 {
    for(int i=0; i<24; ++i)
@@ -166,7 +178,7 @@ int neoncache_get_st_f_i64_noback(dynarec_arm_t* dyn, int ninst, int a)
         && dyn->insts[ninst].n.neoncache[i].n==a)
            return i;
    return -1;
-} 
+}
 int neoncache_get_current_st_f(dynarec_arm_t* dyn, int a)
 {
    for(int i=0; i<24; ++i)
@@ -174,7 +186,7 @@ int neoncache_get_current_st_f(dynarec_arm_t* dyn, int a)
         && dyn->n.neoncache[i].n==a)
            return i;
    return -1;
-} 
+}
 int neoncache_get_current_st_f_i64(dynarec_arm_t* dyn, int a)
 {
    for(int i=0; i<24; ++i)
@@ -182,7 +194,7 @@ int neoncache_get_current_st_f_i64(dynarec_arm_t* dyn, int a)
         && dyn->n.neoncache[i].n==a)
            return i;
    return -1;
-} 
+}
 static void neoncache_promote_double_forward(dynarec_arm_t* dyn, int ninst, int maxinst, int a);
 static void neoncache_promote_double_internal(dynarec_arm_t* dyn, int ninst, int maxinst, int a);
 static void neoncache_promote_double_combined(dynarec_arm_t* dyn, int ninst, int maxinst, int a)
@@ -190,7 +202,7 @@ static void neoncache_promote_double_combined(dynarec_arm_t* dyn, int ninst, int
    if(a == dyn->insts[ninst].n.combined1 || a == dyn->insts[ninst].n.combined2) {
        if(a == dyn->insts[ninst].n.combined1) {
            a = dyn->insts[ninst].n.combined2;
-        } else 
+        } else
            a = dyn->insts[ninst].n.combined1;
        int i = neoncache_get_st_f_i64_noback(dyn, ninst, a);
        //if(box64_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double_combined, ninst=%d combined%c %d i=%d (stack:%d/%d)\n", ninst, (a == dyn->insts[ninst].n.combined2)?'2':'1', a ,i, dyn->insts[ninst].n.stack_push, -dyn->insts[ninst].n.stack_pop);
@@ -200,15 +212,13 @@ static void neoncache_promote_double_combined(dynarec_arm_t* dyn, int ninst, int
                neoncache_promote_double_internal(dyn, ninst-1, maxinst, a-dyn->insts[ninst].n.stack_push);
            // go forward is combined is not pop'd
            if(a-dyn->insts[ninst].n.stack_pop>=0)
-                if(!dyn->insts[ninst+1].n.barrier)
+                if(!((ninst+1<dyn->size) && dyn->insts[ninst+1].n.barrier))
                    neoncache_promote_double_forward(dyn, ninst+1, maxinst, a-dyn->insts[ninst].n.stack_pop);
        }
    }
 }
 static void neoncache_promote_double_internal(dynarec_arm_t* dyn, int ninst, int maxinst, int a)
 {
-    if(dyn->insts[ninst+1].n.barrier)
-        return;
    while(ninst>=0) {
        a+=dyn->insts[ninst].n.stack_pop;    // adjust Stack depth: add pop'd ST (going backward)
        int i = neoncache_get_st_f_i64(dyn, ninst, a);
@@ -314,7 +324,7 @@ static int isCacheEmpty(dynarec_native_t* dyn, int ninst) {
    for(int i=0; i<24; ++i)
        if(dyn->insts[ninst].n.neoncache[i].v) {       // there is something at ninst for i
            if(!(
-            (dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_F 
+            (dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_F
             || dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_D
             || dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_I64)
            && dyn->insts[ninst].n.neoncache[i].n<dyn->insts[ninst].n.stack_pop))
@@ -339,7 +349,7 @@ int fpuCacheNeedsTransform(dynarec_arm_t* dyn, int ninst) {
        for(int i=0; i<24 && !ret; ++i)
            if(dyn->insts[ninst].n.neoncache[i].v) {       // there is something at ninst for i
                if(!(
-                (dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_F 
+                (dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_F
                || dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_D
                || dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_I64)
                && dyn->insts[ninst].n.neoncache[i].n<dyn->insts[ninst].n.stack_pop))
@@ -377,8 +387,9 @@ void neoncacheUnwind(neoncache_t* cache)
 {
    if(cache->swapped) {
        // unswap
-        int a = -1; 
+        int a = -1;
        int b = -1;
+        // in neoncache
        for(int j=0; j<24 && ((a==-1) || (b==-1)); ++j)
            if((cache->neoncache[j].t == NEON_CACHE_ST_D || cache->neoncache[j].t == NEON_CACHE_ST_F || cache->neoncache[j].t == NEON_CACHE_ST_I64)) {
                if(cache->neoncache[j].n == cache->combined1)
@@ -391,11 +402,12 @@ void neoncacheUnwind(neoncache_t* cache)
            cache->neoncache[a].n = cache->neoncache[b].n;
            cache->neoncache[b].n = tmp;
        }
+        // done
        cache->swapped = 0;
        cache->combined1 = cache->combined2 = 0;
    }
    if(cache->news) {
-        // reove the newly created neoncache
+        // remove the newly created neoncache
        for(int i=0; i<24; ++i)
            if(cache->news&(1<<i))
                cache->neoncache[i].v = 0;
@@ -412,11 +424,23 @@ void neoncacheUnwind(neoncache_t* cache)
            }
        }
        cache->x87stack-=cache->stack_push;
+        cache->tags>>=(cache->stack_push*2);
        cache->stack-=cache->stack_push;
+        if(cache->pushed>=cache->stack_push)
+            cache->pushed-=cache->stack_push;
+        else
+            cache->pushed = 0;
        cache->stack_push = 0;
    }
    cache->x87stack+=cache->stack_pop;
    cache->stack_next = cache->stack;
+    if(cache->stack_pop) {
+        if(cache->poped>=cache->stack_pop)
+            cache->poped-=cache->stack_pop;
+        else
+            cache->poped = 0;
+        cache->tags<<=(cache->stack_pop*2);
+    }
    cache->stack_pop = 0;
    cache->barrier = 0;
    // And now, rebuild the x87cache info with neoncache
@@ -529,7 +553,7 @@ void inst_name_pass3(dynarec_native_t* dyn, int ninst, const char* name, rex_t r
 {
    if(box64_dynarec_dump) {
        printf_x64_instruction(rex.is32bits?my_context->dec32:my_context->dec, &dyn->insts[ninst].x64, name);
-        dynarec_log(LOG_NONE, "%s%p: %d emitted opcodes, inst=%d, barrier=%d state=%d/%d(%d), %s=%X/%X, use=%X, need=%X/%X, sm=%d/%d",
+        dynarec_log(LOG_NONE, "%s%p: %d emitted opcodes, inst=%d, barrier=%d state=%d/%d(%d), %s=%X/%X, use=%X, need=%X/%X, sm=%d(%d/%d)",
            (box64_dynarec_dump>1)?"\e[32m":"",
            (void*)(dyn->native_start+dyn->insts[ninst].address),
            dyn->insts[ninst].size/4,
@@ -544,7 +568,7 @@ void inst_name_pass3(dynarec_native_t* dyn, int ninst, const char* name, rex_t r
            dyn->insts[ninst].x64.use_flags,
            dyn->insts[ninst].x64.need_before,
            dyn->insts[ninst].x64.need_after,
-            dyn->smread, dyn->smwrite);
+            dyn->smwrite, dyn->insts[ninst].will_write, dyn->insts[ninst].last_write);
        if(dyn->insts[ninst].pred_sz) {
            dynarec_log(LOG_NONE, ", pred=");
            for(int ii=0; ii<dyn->insts[ninst].pred_sz; ++ii)
@@ -580,4 +604,61 @@ void inst_name_pass3(dynarec_native_t* dyn, int ninst, const char* name, rex_t r
 void print_opcode(dynarec_native_t* dyn, int ninst, uint32_t opcode)
 {
    dynarec_log(LOG_NONE, "\t%08x\t%s\n", opcode, arm64_print(opcode, (uintptr_t)dyn->block));
+}
+
+static void x87_reset(neoncache_t* n)
+{
+    for (int i=0; i<8; ++i)
+        n->x87cache[i] = -1;
+    n->tags = 0;
+    n->x87stack = 0;
+    n->stack = 0;
+    n->stack_next = 0;
+    n->stack_pop = 0;
+    n->stack_push = 0;
+    n->combined1 = n->combined2 = 0;
+    n->swapped = 0;
+    n->barrier = 0;
+    n->pushed = 0;
+    n->poped = 0;
+
+    for(int i=0; i<24; ++i)
+        if(n->neoncache[i].t == NEON_CACHE_ST_F
+         || n->neoncache[i].t == NEON_CACHE_ST_D
+         || n->neoncache[i].t == NEON_CACHE_ST_I64)
+            n->neoncache[i].v = 0;
+}
+
+static void mmx_reset(neoncache_t* n)
+{
+    n->mmxcount = 0;
+    for (int i=0; i<8; ++i)
+        n->mmxcache[i] = -1;
+}
+
+static void sse_reset(neoncache_t* n)
+{
+    for (int i=0; i<16; ++i)
+        n->ssecache[i].v = -1;
+}
+
+void fpu_reset(dynarec_arm_t* dyn)
+{
+    x87_reset(&dyn->n);
+    mmx_reset(&dyn->n);
+    sse_reset(&dyn->n);
+    fpu_reset_reg(dyn);
+}
+
+void fpu_reset_ninst(dynarec_arm_t* dyn, int ninst)
+{
+    x87_reset(&dyn->insts[ninst].n);
+    mmx_reset(&dyn->insts[ninst].n);
+    sse_reset(&dyn->insts[ninst].n);
+    fpu_reset_reg_neoncache(&dyn->insts[ninst].n);
+}
+
+int fpu_is_st_freed(dynarec_native_t* dyn, int ninst, int st)
+{
+    return (dyn->n.tags&(0b11<<(st*2)))?1:0;
 }
@@ -3,6 +3,8 @@

 #include "../dynarec_native_functions.h"

+#define SCRATCH0    24
+
 // Get an FPU scratch reg
 int fpu_get_scratch(dynarec_arm_t* dyn);
 // Reset scratch regs counter
@@ -35,8 +37,11 @@ int neoncache_get_current_st_f_i64(dynarec_arm_t* dyn, int a);
 void neoncache_promote_double(dynarec_arm_t* dyn, int ninst, int a);
 // Combine and propagate if needed (pass 1 only)
 int neoncache_combine_st(dynarec_arm_t* dyn, int ninst, int a, int b);  // with stack current dyn->n_stack*
+// Do not allow i64 type
+int neoncache_no_i64(dynarec_arm_t* dyn, int ninst, int st, int a);

-// FPU Cache transformation (for loops) // Specific, need to be written par backend
+
+// FPU Cache transformation (for loops) // Specific, need to be written by backend
 int fpuCacheNeedsTransform(dynarec_arm_t* dyn, int ninst);

 // Undo the changes of a neoncache to get the status before the instruction
@@ -49,4 +54,11 @@ const char* getCacheName(int t, int n);

 void inst_name_pass3(dynarec_native_t* dyn, int ninst, const char* name, rex_t rex);
 void print_opcode(dynarec_native_t* dyn, int ninst, uint32_t opcode);
+
+// reset the cache
+void fpu_reset(dynarec_native_t* dyn);
+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);
 #endif //__DYNAREC_ARM_FUNCTIONS_H__
@@ -33,28 +33,57 @@
 #define PKip(a)   *(uint8_t*)(ip+a)

 // Strong mem emulation helpers
-// Sequence of Read will trigger a DMB on "first" read if strongmem is 2
-// Squence of Write will trigger a DMB on "last" write if strongmem is 1
+#define SMREAD_VAL  4
+#define SMWRITE2_MIN 1
+#define SMFIRST_MIN 1
+#define SMSEQ_MIN 2
+#define SMSEQ_MAX 3
+#if STEP == 1
+// pass 1 has the jump point available
+#define SMWRITE()   dyn->insts[ninst].will_write = 1; dyn->smwrite = 1
+#define SMREAD()
+#define SMREADLOCK(lock)
+#define SMMIGHTREAD()
+#define WILLWRITE2()   if(box64_dynarec_strongmem>SMWRITE2_MIN) {WILLWRITE();}
+#define SMWRITE2()   if(box64_dynarec_strongmem>SMWRITE2_MIN) {SMWRITE();}
+#define SMWRITELOCK(lock)   SMWRITE()
+#define WILLWRITELOCK(lock)
+#define WILLWRITE()
+#define SMMIGHTWRITE()   if(!MODREG) {SMWRITE();}
+#define SMSTART() dyn->smwrite = 0; dyn->smread = 0;
+#define SMEND() if(dyn->smwrite && (box64_dynarec_strongmem>SMFIRST_MIN)) {int i = ninst; while(i>=0 && !dyn->insts[i].will_write) --i; if(i>=0) {dyn->insts[i].last_write = 1;}} dyn->smwrite = 0
+#define SMDMB()
+#else
+// Sequence of Write will trigger a DMB on "last" write if strongmem is >= 1
+// Block will trigget at 1st and last if strongmem is >= SMFIRST_MIN
+// Read will contribute to trigger a DMB on "first" read if strongmem is >= SMREAD_MIN
 // Opcode will read
-#define SMREAD()    if(!dyn->smread && box64_dynarec_strongmem>1) {SMDMB();}
+#define SMREAD()    if(dyn->insts[ninst].will_write) {WILLWRITE();} else if(box64_dynarec_strongmem==SMREAD_VAL && !dyn->smread) {DSB_SY(); dyn->smread = 1;}
 // Opcode will read with option forced lock
-#define SMREADLOCK(lock)    if(lock || (!dyn->smread && box64_dynarec_strongmem>1)) {SMDMB();}
-// Opcode migh read (depend on nextop)
+#define SMREADLOCK(lock)    if((lock)) {SMWRITELOCK(lock);} else {SMREAD();}
+// Opcode might read (depend on nextop)
 #define SMMIGHTREAD()   if(!MODREG) {SMREAD();}
 // Opcode has wrote
-#define SMWRITE()   dyn->smwrite=1
+#define SMWRITE()   if((box64_dynarec_strongmem>=SMFIRST_MIN) && dyn->smwrite==0 && (box64_dynarec_strongmem!=SMREAD_VAL)) {SMDMB();} if(box64_dynarec_strongmem>SMSEQ_MIN && (box64_dynarec_strongmem!=SMREAD_VAL)) {if(++dyn->smwrite>=SMSEQ_MAX) {SMDMB(); dyn->smwrite=1;}} else dyn->smwrite=1
 // Opcode has wrote (strongmem>1 only)
-#define SMWRITE2()   if(box64_dynarec_strongmem>1) dyn->smwrite=1
+#define WILLWRITE2()   if(box64_dynarec_strongmem>SMWRITE2_MIN) {WILLWRITE();}
+#define SMWRITE2()   if(box64_dynarec_strongmem>SMWRITE2_MIN) {SMWRITE();}
 // Opcode has wrote with option forced lock
-#define SMWRITELOCK(lock)   if(lock) {SMDMB();} else dyn->smwrite=1
-// Opcode migh have wrote (depend on nextop)
+#define SMWRITELOCK(lock)   if(lock) {SMDMB(); dyn->smwrite=1;} else {SMWRITE();}
+// Opcode has wrote with option forced lock
+#define WILLWRITELOCK(lock)   if(lock) {DMB_ISH();} else {WILLWRITE();}
+// Opcode might have wrote (depend on nextop)
 #define SMMIGHTWRITE()   if(!MODREG) {SMWRITE();}
+// Opcode will write (without reading)
+#define WILLWRITE() if((box64_dynarec_strongmem>=SMFIRST_MIN) && dyn->smwrite==0 && (box64_dynarec_strongmem!=SMREAD_VAL)) {SMDMB();} else if(box64_dynarec_strongmem>=SMFIRST_MIN && dyn->insts[ninst].last_write && (box64_dynarec_strongmem!=SMREAD_VAL)) {SMDMB();} dyn->smwrite=1
 // Start of sequence
 #define SMSTART()   SMEND()
 // End of sequence
-#define SMEND()     if(dyn->smwrite && box64_dynarec_strongmem) {DMB_ISH();} dyn->smwrite=0; dyn->smread=0;
+#define SMEND()     if(dyn->smwrite && box64_dynarec_strongmem && (box64_dynarec_strongmem!=SMREAD_VAL)) {DMB_ISH();} dyn->smwrite=0; dyn->smread=0
 // Force a Data memory barrier (for LOCK: prefix)
-#define SMDMB()     DMB_ISH(); dyn->smwrite=0; dyn->smread=1
+#define SMDMB()     if(box64_dynarec_strongmem){DSB_ISH();}else{DMB_ISH();} dyn->smwrite=0; dyn->smread=0
+
+#endif

 //LOCK_* define
 #define LOCK_LOCK   (int*)1
@@ -66,7 +95,7 @@
                    ed = xRAX+(nextop&7)+(rex.b<<3);    \
                    wback = 0;                          \
                } else {                                \
-                    SMREAD()                            \
+                    SMREAD();                           \
                    addr = geted(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;                            \
@@ -85,7 +114,7 @@
                    wback = 0;                          \
                } else {                                \
                    SMREAD();                           \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, D); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<(3-rex.is32bits), rex.is32bits?3:7, rex, NULL, 0, D); \
                    LDz(x1, wback, fixedaddress);       \
                    ed = x1;                            \
                }
@@ -251,6 +280,21 @@
                    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 = xRAX+(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); \
+                    ADDx_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) {           \
                    wback = xRAX+(nextop&7)+(rex.b<<3);\
@@ -360,18 +404,18 @@
                        gb2 = 0;                \
                    } else {                    \
                        gd = (nextop&0x38)>>3;  \
-                        gb2 = ((gd&4)>>2);      \
+                        gb2 = ((gd&4)<<1);      \
                        gb1 = xRAX+(gd&3);      \
                    }                           \
                    gd = i;                     \
-                    UBFXx(gd, gb1, gb2*8, 8);
-//GETSGB signe extend GB, will use i for gd
+                    UBFXx(gd, gb1, gb2, 8);
+//GETSGB sign extend GB, will use i for gd
 #define GETSGB(i)   if(rex.rex) {               \
                        gb1 = xRAX+((nextop&0x38)>>3)+(rex.r<<3);     \
                        gb2 = 0;                \
                    } else {                    \
                        gd = (nextop&0x38)>>3;  \
-                        gb2 = ((gd&4)>>2);      \
+                        gb2 = ((gd&4)<<1);      \
                        gb1 = xRAX+(gd&3);      \
                    }                           \
                    gd = i;                     \
@@ -397,7 +441,7 @@
    if(MODREG) {                                                                                        \
        a = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), w);                                      \
    } else {                                                                                            \
-        SMREAD();                                                                                       \
+        if(w) {WILLWRITE2();} else {SMREAD();}                                                          \
        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, D);  \
        a = fpu_get_scratch(dyn);                                                                       \
        VLD128(a, ed, fixedaddress);                                                                    \
@@ -416,23 +460,40 @@
    if(MODREG) {                                                                                        \
        a = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), w);                                      \
    } else {                                                                                            \
-        SMREAD();                                                                                       \
+        if(w) {WILLWRITE2();} else {SMREAD();}                                                          \
        a = fpu_get_scratch(dyn);                                                                       \
        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, D);   \
        VLD64(a, ed, fixedaddress);                                                                     \
    }

+// Get Ex as 64bits, not a quad (warning, x1 get used)
+#define GETEX64(a, w, D)    GETEXSD(a, w, D)
+
 // Get Ex as a single, not a quad (warning, x1 get used)
 #define GETEXSS(a, w, D)                                                                                \
    if(MODREG) {                                                                                        \
        a = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), w);                                      \
    } else {                                                                                            \
-        SMREAD();                                                                                       \
+        if(w) {WILLWRITE2();} else {SMREAD();}                                                          \
        a = fpu_get_scratch(dyn);                                                                       \
        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, D);   \
        VLD32(a, ed, fixedaddress);                                                                     \
    }

+// Get Ex as 32bits, not a quad (warning, x1 get used)
+#define GETEX32(a, w, D)    GETEXSS(a, w, D)
+
+// Get Ex as 16bits, not a quad (warning, x1 get used)
+#define GETEX16(a, w, D)                                                                                \
+    if(MODREG) {                                                                                        \
+        a = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), w);                                      \
+    } else {                                                                                            \
+        if(w) {WILLWRITE2();} else {SMREAD();}                                                          \
+        a = fpu_get_scratch(dyn);                                                                       \
+        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, D);   \
+        VLD16(a, ed, fixedaddress);                                                                     \
+    }
+
 // Get GM, might use x1, x2 and x3
 #define GETGM(a)                        \
    gd = ((nextop&0x38)>>3);            \
@@ -464,6 +525,9 @@
    TSTw_mask(xFlags, 0b010110, 0); \
    CNEGx(r, r, cNE)

+#define ALIGNED_ATOMICxw ((fixedaddress && !(fixedaddress&(((1<<(2+rex.w))-1)))) || box64_dynarec_aligned_atomics)
+#define ALIGNED_ATOMICH ((fixedaddress && !(fixedaddress&1)) || box64_dynarec_aligned_atomics)
+
 // CALL 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
 #define CALL(F, ret) call_c(dyn, ninst, F, x7, ret, 1, 0)
@@ -627,6 +691,7 @@
    CBNZx(reg, j64)

 #define IFX(A)  if((dyn->insts[ninst].x64.gen_flags&(A)))
+#define IFX2(A, B)  if((dyn->insts[ninst].x64.gen_flags&(A)) B)
 #define IFX_PENDOR0  if((dyn->insts[ninst].x64.gen_flags&(X_PEND) || !dyn->insts[ninst].x64.gen_flags))
 #define IFXX(A) if((dyn->insts[ninst].x64.gen_flags==(A)))
 #define IFX2X(A, B) if((dyn->insts[ninst].x64.gen_flags==(A) || dyn->insts[ninst].x64.gen_flags==(B) || dyn->insts[ninst].x64.gen_flags==((A)|(B))))
@@ -637,12 +702,12 @@
    LDRH_U12(s3, xEmu, offsetof(x64emu_t, sw));   /*offset is 8bits right?*/\
    MOV32w(s1, 0b0100011100000000);                                         \
    BICw_REG(s3, s3, s1);                                                   \
+    /* greater than leave 0 */                                              \
    CSETw(s1, cMI); /* 1 if less than, 0 else */                            \
-    MOV32w(s2, 0b01000101); /* unordered */                                 \
-    CSELw(s1, s2, s1, cVS);                                                 \
    MOV32w(s2, 0b01000000); /* zero */                                      \
    CSELw(s1, s2, s1, cEQ);                                                 \
-    /* greater than leave 0 */                                              \
+    MOV32w(s2, 0b01000101); /* unordered */                                 \
+    CSELw(s1, s2, s1, cVS);                                                 \
    ORRw_REG_LSL(s3, s3, s1, 8);                                            \
    STRH_U12(s3, xEmu, offsetof(x64emu_t, sw))

@@ -742,8 +807,48 @@
    LDP_REGS(R12, R13);     \
    LDP_REGS(R14, R15)

+#if STEP == 0
+#define X87_PUSH_OR_FAIL(var, dyn, ninst, scratch, t)   var = x87_do_push(dyn, ninst, scratch, t)
+#define X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, scratch)     x87_do_push_empty(dyn, ninst, scratch)
+#define X87_POP_OR_FAIL(dyn, ninst, scratch)            x87_do_pop(dyn, ninst, scratch)
+#else
+#define X87_PUSH_OR_FAIL(var, dyn, ninst, scratch, t)   \
+    if ((dyn->n.x87stack==8) || (dyn->n.pushed==8)) {   \
+        if(box64_dynarec_dump) dynarec_log(LOG_NONE, " Warning, suspicious x87 Push, stack=%d/%d on inst %d\n", dyn->n.x87stack, dyn->n.pushed, ninst); \
+        dyn->abort = 1;                                 \
+        return addr;                                    \
+    }                                                   \
+    var = x87_do_push(dyn, ninst, scratch, t)
+
+#define X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, scratch)     \
+    if ((dyn->n.x87stack==8) || (dyn->n.pushed==8)) {   \
+        if(box64_dynarec_dump) dynarec_log(LOG_NONE, " Warning, suspicious x87 Push, stack=%d/%d on inst %d\n", dyn->n.x87stack, dyn->n.pushed, ninst); \
+        dyn->abort = 1;                                 \
+        return addr;                                    \
+    }                                                   \
+    x87_do_push_empty(dyn, ninst, scratch)
+
+#define X87_POP_OR_FAIL(dyn, ninst, scratch)            \
+    if ((dyn->n.x87stack==-8) || (dyn->n.poped==8)) {   \
+        if(box64_dynarec_dump) dynarec_log(LOG_NONE, " Warning, suspicious x87 Pop, stack=%d/%d on inst %d\n", dyn->n.x87stack, dyn->n.poped, ninst); \
+        dyn->abort = 1;                                 \
+        return addr;                                    \
+    }                                                   \
+    x87_do_pop(dyn, ninst, scratch)
+#endif
+
 #define SET_DFNONE(S)    if(!dyn->f.dfnone) {STRw_U12(wZR, xEmu, offsetof(x64emu_t, df)); dyn->f.dfnone=1;}
-#define SET_DF(S, N)     if((N)!=d_none) {MOVZw(S, (N)); STRw_U12(S, xEmu, offsetof(x64emu_t, df)); dyn->f.dfnone=0;} else SET_DFNONE(S)
+#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_(UpdateFlags, -1, 0);              \
+            dyn->f.pending = SF_SET;                \
+            SET_NODF();     \
+        }                   \
+        dyn->f.dfnone=0;    \
+    } else SET_DFNONE(S)
 #define SET_NODF()          dyn->f.dfnone = 0
 #define SET_DFOK()          dyn->f.dfnone = 1

@@ -776,6 +881,8 @@
    if(dyn->insts[ninst].x64.gen_flags) switch(B) {                                             \
        case SF_SUBSET:                                                                         \
        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_SUBSET_PENDING:                                                                 \
        case SF_SET_PENDING:                                                                    \
@@ -784,7 +891,7 @@
    } else dyn->f.pending = SF_SET
 #endif
 #ifndef JUMP
-#define JUMP(A, C)
+#define JUMP(A, C) SMEND()
 #endif
 #ifndef BARRIER
 #define BARRIER(A)
@@ -792,6 +899,9 @@
 #ifndef BARRIER_NEXT
 #define BARRIER_NEXT(A)
 #endif
+#ifndef SET_HASCALLRET
+#define SET_HASCALLRET()
+#endif
 #define UFLAG_OP1(A) if(dyn->insts[ninst].x64.gen_flags) {STRxw_U12(A, xEmu, offsetof(x64emu_t, op1));}
 #define UFLAG_OP2(A) if(dyn->insts[ninst].x64.gen_flags) {STRxw_U12(A, xEmu, offsetof(x64emu_t, op2));}
 #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));}
@@ -811,13 +921,14 @@
 #define FTABLE64(A, V)
 #endif

-#define ARCH_INIT()         \
-    dyn->doublepush = 0;    \
+#define ARCH_INIT()                 \
+    dyn->smread = dyn->smwrite = 0; \
+    dyn->doublepush = 0;            \
    dyn->doublepop = 0;

 #if STEP < 2
-#define GETIP(A)
-#define GETIP_(A)
+#define GETIP(A) TABLE64(0, 0)
+#define GETIP_(A) TABLE64(0, 0)
 #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)                                        \
@@ -874,6 +985,8 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 #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_6764_32  STEPNAME(dynarec64_6764_32)
 #define dynarec64_D8       STEPNAME(dynarec64_D8)
 #define dynarec64_D9       STEPNAME(dynarec64_D9)
 #define dynarec64_DA       STEPNAME(dynarec64_DA)
@@ -884,6 +997,8 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 #define dynarec64_DF       STEPNAME(dynarec64_DF)
 #define dynarec64_F0       STEPNAME(dynarec64_F0)
 #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)
@@ -965,25 +1080,54 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 #define emit_shr32      STEPNAME(emit_shr32)
 #define emit_shr32c     STEPNAME(emit_shr32c)
 #define emit_sar32c     STEPNAME(emit_sar32c)
+#define emit_shl8       STEPNAME(emit_shl8)
+#define emit_shl8c      STEPNAME(emit_shl8c)
+#define emit_shr8       STEPNAME(emit_shr8)
+#define emit_shr8c      STEPNAME(emit_shr8c)
+#define emit_sar8       STEPNAME(emit_sar8)
+#define emit_sar8c      STEPNAME(emit_sar8c)
+#define emit_shl16      STEPNAME(emit_shl16)
+#define emit_shl16c     STEPNAME(emit_shl16c)
+#define emit_shr16      STEPNAME(emit_shr16)
+#define emit_shr16c     STEPNAME(emit_shr16c)
+#define emit_sar16      STEPNAME(emit_sar16)
+#define emit_sar16c     STEPNAME(emit_sar16c)
 #define emit_rol32c     STEPNAME(emit_rol32c)
 #define emit_ror32c     STEPNAME(emit_ror32c)
+#define emit_rol8c      STEPNAME(emit_rol8c)
+#define emit_ror8c      STEPNAME(emit_ror8c)
+#define emit_rol16c     STEPNAME(emit_rol16c)
+#define emit_ror16c     STEPNAME(emit_ror16c)
+#define emit_rcl8c      STEPNAME(emit_rcl8c)
+#define emit_rcr8c      STEPNAME(emit_rcr8c)
+#define emit_rcl16c     STEPNAME(emit_rcl16c)
+#define emit_rcr16c     STEPNAME(emit_rcr16c)
+#define emit_rcl32c     STEPNAME(emit_rcl32c)
+#define emit_rcr32c     STEPNAME(emit_rcr32c)
 #define emit_shrd32c    STEPNAME(emit_shrd32c)
+#define emit_shrd32     STEPNAME(emit_shrd32)
 #define emit_shld32c    STEPNAME(emit_shld32c)
+#define emit_shld32     STEPNAME(emit_shld32)
+#define emit_shrd16c    STEPNAME(emit_shrd16c)
+#define emit_shrd16     STEPNAME(emit_shrd16)
+#define emit_shld16c    STEPNAME(emit_shld16c)
+#define emit_shld16     STEPNAME(emit_shld16)

 #define emit_pf         STEPNAME(emit_pf)

-#define x87_do_push     STEPNAME(x87_do_push)
-#define x87_do_push_empty STEPNAME(x87_do_push_empty)
-#define x87_do_pop      STEPNAME(x87_do_pop)
+#define x87_do_push         STEPNAME(x87_do_push)
+#define x87_do_push_empty   STEPNAME(x87_do_push_empty)
+#define x87_do_pop          STEPNAME(x87_do_pop)
 #define x87_get_current_cache   STEPNAME(x87_get_current_cache)
 #define x87_get_cache   STEPNAME(x87_get_cache)
 #define x87_get_neoncache STEPNAME(x87_get_neoncache)
 #define x87_get_st      STEPNAME(x87_get_st)
 #define x87_get_st_empty  STEPNAME(x87_get_st)
-#define x87_refresh     STEPNAME(x87_refresh)
+#define x87_free        STEPNAME(x87_free)
 #define x87_forget      STEPNAME(x87_forget)
 #define x87_reget_st    STEPNAME(x87_reget_st)
 #define x87_stackcount  STEPNAME(x87_stackcount)
+#define x87_unstackcount  STEPNAME(x87_unstackcount)
 #define x87_swapreg     STEPNAME(x87_swapreg)
 #define x87_setround    STEPNAME(x87_setround)
 #define x87_restoreround STEPNAME(x87_restoreround)
@@ -994,10 +1138,10 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 #define sse_get_reg_empty STEPNAME(sse_get_reg_empty)
 #define sse_forget_reg   STEPNAME(sse_forget_reg)
 #define sse_purge07cache STEPNAME(sse_purge07cache)
+#define sse_reflect_reg  STEPNAME(sse_reflect_reg)

 #define fpu_pushcache   STEPNAME(fpu_pushcache)
 #define fpu_popcache    STEPNAME(fpu_popcache)
-#define fpu_reset       STEPNAME(fpu_reset)
 #define fpu_reset_cache STEPNAME(fpu_reset_cache)
 #define fpu_propagate_stack STEPNAME(fpu_propagate_stack)
 #define fpu_purgecache  STEPNAME(fpu_purgecache)
@@ -1008,6 +1152,9 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);

 #define CacheTransform       STEPNAME(CacheTransform)

+#define arm64_move32        STEPNAME(arm64_move32)
+#define arm64_move64        STEPNAME(arm64_move64)
+
 /* setup r2 to address pointed by */
 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);

@@ -1020,7 +1167,7 @@ uintptr_t geted16(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop,

 // 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);
+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, int ninst, rex_t rex);
 void retn_to_epilog(dynarec_arm_t* dyn, int ninst, rex_t rex, int n);
 void iret_to_epilog(dynarec_arm_t* dyn, int ninst, int is64bits);
@@ -1035,7 +1182,7 @@ void emit_cmp16_0(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4);
 void emit_cmp32_0(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s3, int s4);
 void emit_test8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, int s5);
 void emit_test16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, int s5);
-void emit_test32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4);
+void emit_test32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4, int s5);
 void emit_add32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4);
 void emit_add32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int64_t c, int s3, int s4, int s5);
 void emit_add8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
@@ -1092,16 +1239,46 @@ void emit_shl32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, i
 void emit_shr32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4);
 void emit_shr32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, int s3, int s4);
 void emit_sar32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, int s3, int s4);
+void emit_shl8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
+void emit_shl8c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_shr8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
+void emit_shr8c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_sar8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
+void emit_sar8c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_shl16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
+void emit_shl16c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_shr16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
+void emit_shr16c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_sar16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
+void emit_sar16c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
 void emit_rol32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, int s3, int s4);
 void emit_ror32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, int s3, int s4);
+void emit_rol8c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_ror8c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_rol16c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_ror16c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_rcl8c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_rcr8c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_rcl16c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+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_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);
+void emit_shld32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s5, int s3, int s4);
+void emit_shrd16c(dynarec_arm_t* dyn, int ninst, int s1, int s2, uint32_t c, int s3, int s4);
+void emit_shrd16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s5, int s3, int s4);
+void emit_shld16c(dynarec_arm_t* dyn, int ninst, int s1, int s2, uint32_t c, int s3, int s4);
+void emit_shld16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s5, int s3, int s4);

 void emit_pf(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4);

 // x87 helper
-// cache of the local stack counter, to avoid upadte at every call
-void x87_stackcount(dynarec_arm_t* dyn, int ninst, int scratch);
+// cache of the local stack counter, to avoid update at every call, return old internal stack counter
+int x87_stackcount(dynarec_arm_t* dyn, int ninst, int scratch);
+// revert local stack counter to previous version (return from x87_stackcount)
+void x87_unstackcount(dynarec_arm_t* dyn, int ninst, int scratch, int count);
 // fpu push. Return the Dd value to be used
 int x87_do_push(dynarec_arm_t* dyn, int ninst, int s1, int t);
 // fpu push. Do not allocate a cache register. Needs a scratch register to do x87stack synch (or 0 to not do it)
@@ -1118,8 +1295,8 @@ int x87_get_neoncache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a);
 int x87_get_st(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a, int t);
 // get vfpu register for a x87 reg, create the entry if needed. Do not fetch the Stx if not already in cache
 int x87_get_st_empty(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a, int t);
-// refresh a value from the cache ->emu (nothing done if value is not cached)
-void x87_refresh(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st);
+// Free st, using the FFREE opcode (so it's freed but stack is not moved)
+void x87_free(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int st);
 // refresh a value from the cache ->emu and then forget the cache (nothing done if value is not cached)
 void x87_forget(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st);
 // refresh the cache value from emu
@@ -1135,6 +1312,9 @@ int sse_setround(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);

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

+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
 #else
@@ -1143,6 +1323,8 @@ void CacheTransform(dynarec_arm_t* dyn, int ninst, int cacheupd, int s1, int s2,

 #define neoncache_st_coherency STEPNAME(neoncache_st_coherency)
 int neoncache_st_coherency(dynarec_arm_t* dyn, int ninst, int a, int b);
+// scratch fpu regs for convertions
+#define SCRATCH 31

 #if STEP == 0
 #define ST_IS_F(A)          0
@@ -1154,8 +1336,8 @@ int neoncache_st_coherency(dynarec_arm_t* dyn, int ninst, int a, int b);
 #define ST_IS_F(A) (neoncache_get_current_st(dyn, ninst, A)==NEON_CACHE_ST_F)
 #define ST_IS_I64(A) (neoncache_get_current_st(dyn, ninst, A)==NEON_CACHE_ST_I64)
 #define X87_COMBINE(A, B) neoncache_combine_st(dyn, ninst, A, B)
-#define X87_ST0     neoncache_get_current_st(dyn, ninst, 0)
-#define X87_ST(A)   neoncache_get_current_st(dyn, ninst, A)
+#define X87_ST0     neoncache_no_i64(dyn, ninst, 0, neoncache_get_current_st(dyn, ninst, 0))
+#define X87_ST(A)   neoncache_no_i64(dyn, ninst, A, neoncache_get_current_st(dyn, ninst, A))
 #else
 #define ST_IS_F(A) (neoncache_get_st(dyn, ninst, A)==NEON_CACHE_ST_F)
 #define ST_IS_I64(A) (neoncache_get_st(dyn, ninst, A)==NEON_CACHE_ST_I64)
@@ -1183,10 +1365,9 @@ int sse_get_reg_empty(dynarec_arm_t* dyn, int ninst, int s1, int a);
 void sse_forget_reg(dynarec_arm_t* dyn, int ninst, int a);
 // purge the XMM0..XMM7 cache (before function call)
 void sse_purge07cache(dynarec_arm_t* dyn, int ninst, int s1);
-
+// Push current value to the cache
+void sse_reflect_reg(dynarec_arm_t* dyn, int ninst, int a);
 // common coproc helpers
-// reset the cache
-void fpu_reset(dynarec_arm_t* dyn);
 // reset the cache with n
 void fpu_reset_cache(dynarec_arm_t* dyn, int ninst, int reset_n);
 // propagate stack state
@@ -1208,16 +1389,20 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
 //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_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_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_660F(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_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_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
@@ -1354,4 +1539,30 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
            opcode = F8;                        \
        }

+#define COMP_ZFSF(s1, A)                        \
+    IFX(X_ZF|X_SF) {                            \
+        if(arm64_flagm) {                       \
+            SETF##A(s1);                        \
+            IFX(X_ZF) {                         \
+                CSETw(s3, cEQ);                 \
+                BFIw(xFlags, s3, F_ZF, 1);      \
+            }                                   \
+            IFX(X_SF) {                         \
+                CSETw(s3, cMI);                 \
+                BFIw(xFlags, s3, F_SF, 1);      \
+            }                                   \
+        } else {                                \
+            IFX(X_ZF) {                         \
+                ANDSw_mask(s1, s1, 0, (A)-1);   \
+                CSETw(s3, cEQ);                 \
+                BFIw(xFlags, s3, F_ZF, 1);      \
+            }                                   \
+            IFX(X_SF) {                         \
+                LSRw(s3, s1, (A)-1);            \
+                BFIw(xFlags, s3, F_SF, 1);      \
+            }                                   \
+        }                                       \
+    }
+
+
 #endif //__DYNAREC_ARM64_HELPER_H__
@@ -12,20 +12,16 @@
        dyn->f.pending=SF_SET
 #define SETFLAGS(A,B)   \
        dyn->insts[ninst].x64.set_flags = A;    \
-        dyn->insts[ninst].x64.state_flags = B;  \
+        dyn->insts[ninst].x64.state_flags = (B)&~SF_DF;  \
        dyn->f.pending=(B)&SF_SET_PENDING;      \
-        dyn->f.dfnone=((B)&SF_SET)?1:0;
-#define EMIT(A)     
-#define JUMP(A, C)         add_next(dyn, (uintptr_t)A); dyn->insts[ninst].x64.jmp = A; dyn->insts[ninst].x64.jmp_cond = C
+        dyn->f.dfnone=((B)&SF_SET)?(((B)==SF_SET_NODF)?0:1):0;
+#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_NEXT(A) dyn->insts[ninst+1].x64.barrier = A
+#define BARRIER_NEXT(A) dyn->insts[ninst].x64.barrier_next = A
+#define SET_HASCALLRET()    dyn->insts[ninst].x64.has_callret = 1
 #define NEW_INST \
        ++dyn->size;                            \
-        if(dyn->size+3>=dyn->cap) {             \
-                dyn->insts = (instruction_native_t*)dynaRealloc(dyn->insts, sizeof(instruction_native_t)*dyn->cap*2);\
-                memset(&dyn->insts[dyn->cap], 0, sizeof(instruction_native_t)*dyn->cap);   \
-                dyn->cap *= 2;                  \
-        }                                       \
        dyn->insts[ninst].x64.addr = ip;        \
        dyn->n.combined1 = dyn->n.combined2 = 0;\
        dyn->n.swapped = 0; dyn->n.barrier = 0; \
@@ -42,7 +38,7 @@
        *ok = -1;                       \
        if(box64_dynarec_log>=LOG_INFO || box64_dynarec_dump || box64_dynarec_missing) {\
        dynarec_log(LOG_NONE, "%p: Dynarec stopped because of %sOpcode %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X", \
-        (void*)ip, rex.is32bits?"32bits ":"",\
+        (void*)ip, rex.is32bits?"x86 ":"x64 ",\
        PKip(0),                        \
        PKip(1), PKip(2), PKip(3),      \
        PKip(4), PKip(5), PKip(6),      \
@@ -42,10 +42,13 @@ typedef struct neoncache_s {
    uint8_t             combined2;
    uint8_t             swapped;        // the combined reg were swapped
    uint8_t             barrier;        // is there a barrier at instruction epilog?
+    uint8_t             pushed;         // positive pushed value (to check for overflow)
+    uint8_t             poped;          // positive poped value (to check for underflow)
    uint32_t            news;           // bitmask, wich neoncache are new for this opcode
    // fpu cache
    int8_t              x87cache[8];    // cache status for the 8 x87 register behind the fpu stack
    int8_t              x87reg[8];      // reg used for x87cache entry
+    int16_t             tags;           // similar to fpu_tags
    int8_t              mmxcache[8];    // cache status for the 8 MMX registers
    sse_cache_t         ssecache[16];   // cache status for the 16 SSE(2) registers
    int8_t              fpuused[24];    // all 0..24 double reg from fpu, used by x87, sse and mmx
@@ -76,9 +79,11 @@ typedef struct instruction_arm64_s {
    int                 pass2choice;// value for choices that are fixed on pass2 for pass3
    uintptr_t           natcall;
    int                 retn;
-    int                 barrier_maybe;
-    flagcache_t         f_exit;     // flags status at end of intruction
-    neoncache_t         n;          // neoncache at end of intruction (but before poping)
+    uint8_t             barrier_maybe;
+    uint8_t             will_write;
+    uint8_t             last_write;
+    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
 } instruction_arm64_t;

@@ -87,12 +92,12 @@ typedef struct dynarec_arm_s {
    int32_t             size;
    int32_t             cap;
    uintptr_t           start;      // start of the block
-    uint32_t            isize;      // size in byte of x64 instructions included
+    uint32_t            isize;      // size in bytes of x64 instructions included
    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
    uintptr_t           last_ip;    // last set IP in RIP (or NULL if unclean state) TODO: move to a cache something
-    uint64_t*           table64;   // table of 64bits value
+    uint64_t*           table64;    // table of 64bits values
    int                 table64size;// size of table (will be appended at end of executable code)
    int                 table64cap;
    uintptr_t           tablestart;
@@ -102,31 +107,37 @@ typedef struct dynarec_arm_s {
    uintptr_t*          next;       // variable array of "next" jump address
    int                 next_sz;
    int                 next_cap;
+    int*                jmps;       // variable array of jump instructions
+    int                 jmp_sz;
+    int                 jmp_cap;
    int*                predecessor;// single array of all predecessor
    dynablock_t*        dynablock;
    instsize_t*         instsize;
    size_t              insts_size; // size of the instruction size array (calculated)
-    uint8_t             smread;    // for strongmem model emulation
-    uint8_t             smwrite;    // for strongmem model emulation
    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
    int                 forward_ninst;  // ninst at the forward point
+    uint8_t             smwrite;    // for strongmem model emulation
+    uint8_t             smread;
    uint8_t             doublepush;
    uint8_t             doublepop;
    uint8_t             always_test;
+    uint8_t             abort;      // abort the creation of the block
 } dynarec_arm_t;

 void add_next(dynarec_arm_t *dyn, uintptr_t addr);
 uintptr_t get_closest_next(dynarec_arm_t *dyn, uintptr_t addr);
+void add_jump(dynarec_arm_t *dyn, int ninst);
+int get_first_jump(dynarec_arm_t *dyn, int 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 Table64(dynarec_arm_t *dyn, uint64_t val, int pass);  // add a value to etable64 (if needed) and gives back the imm19 to use in LDR_literal
+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)      \
+#define GO_TRACE(A, B, s0)  \
    GETIP(addr);            \
    MOVx_REG(x1, xRIP);     \
    STORE_XEMU_CALL(xRIP);  \
@@ -25,6 +25,7 @@

 #include "custommem.h"
 #include "khash.h"
+#include "rbtree.h"

 uint32_t X31_hash_code(void* addr, int len)
 {
@@ -41,12 +42,24 @@ dynablock_t* InvalidDynablock(dynablock_t* db, int need_lock)
        if(db->gone)
            return NULL; // already in the process of deletion!
        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);
        if(need_lock)
            mutex_lock(&my_context->mutex_dyndump);
-        // remove jumptable
-        setJumpTableDefault64(db->x64_addr);
        db->done = 0;
        db->gone = 1;
+        int db_size = db->x64_size;
+        if(db_size && my_context) {
+            uint32_t n = rb_get(my_context->db_sizes, db_size);
+            if(n>1)
+                rb_set(my_context->db_sizes, db_size, db_size+1, n-1);
+            else
+                rb_unset(my_context->db_sizes, db_size, db_size+1);
+            if(db_size == my_context->max_db_size) {
+                my_context->max_db_size = rb_get_righter(my_context->db_sizes);
+                dynarec_log(LOG_INFO, "BOX64 Dynarec: lower max_db=%d\n", my_context->max_db_size);
+            }
+        }
        if(need_lock)
            mutex_unlock(&my_context->mutex_dyndump);
    }
@@ -74,13 +87,25 @@ void FreeDynablock(dynablock_t* db, int need_lock)
        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_lock)
            mutex_lock(&my_context->mutex_dyndump);
-        // remove jumptable
-        setJumpTableDefault64(db->x64_addr);
        dynarec_log(LOG_DEBUG, " -- FreeDyrecMap(%p, %d)\n", db->actual_block, db->size);
        db->done = 0;
        db->gone = 1;
+        int db_size = db->x64_size;
+        if(db_size && my_context) {
+            uint32_t n = rb_get(my_context->db_sizes, db_size);
+            if(n>1)
+                rb_set(my_context->db_sizes, db_size, db_size+1, n-1);
+            else
+                rb_unset(my_context->db_sizes, db_size, db_size+1);
+            if(db_size == my_context->max_db_size) {
+                my_context->max_db_size = rb_get_righter(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);
@@ -120,15 +145,6 @@ static int IntervalIntersects(uintptr_t start1, uintptr_t end1, uintptr_t start2
    return 1;
 }

-static int MarkedDynablock(dynablock_t* db)
-{
-    if(db) {
-        if(getNeedTest((uintptr_t)db->x64_addr))
-            return 1; // already done
-    }
-    return 0;
-}
-
 void MarkRangeDynablock(dynablock_t* db, uintptr_t addr, uintptr_t size)
 {
    // Mark will try to find *any* blocks that intersect the range to mark
@@ -155,15 +171,6 @@ int FreeRangeDynablock(dynablock_t* db, uintptr_t addr, uintptr_t size)
 dynablock_t *AddNewDynablock(uintptr_t addr)
 {
    dynablock_t* block;
-    #if 0
-    // check if memory as the correct flags
-    int prot = getProtection(addr);
-    if(!(prot&(PROT_EXEC|PROT_DYNAREC|PROT_DYNAREC_R))) {
-        dynarec_log(LOG_VERBOSE, "Block asked on a memory with no execution flags 0x%02X\n", prot);
-        return NULL;
-    }
-    
-    #endif
    // 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));
@@ -171,11 +178,16 @@ dynablock_t *AddNewDynablock(uintptr_t addr)
 }

 //TODO: move this to dynrec_arm.c and track allocated structure to avoid memory leak
-static __thread struct __jmp_buf_tag dynarec_jmpbuf;
+static __thread JUMPBUFF dynarec_jmpbuf;
+#ifdef ANDROID
+#define DYN_JMPBUF dynarec_jmpbuf
+#else
+#define DYN_JMPBUF &dynarec_jmpbuf
+#endif

 void cancelFillBlock()
 {
-    longjmp(&dynarec_jmpbuf, 1);
+    longjmp(DYN_JMPBUF, 1);
 }

 /* 
@@ -208,8 +220,8 @@ static dynablock_t* internalDBGetBlock(x64emu_t* emu, uintptr_t addr, uintptr_t

    // fill the block
    block->x64_addr = (void*)addr;
-    if(sigsetjmp(&dynarec_jmpbuf, 1)) {
-        printf_log(LOG_INFO, "FillBlock at %p triggered a segfault, cancelling\n", (void*)addr);
+    if(sigsetjmp(DYN_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);
@@ -223,17 +235,20 @@ static dynablock_t* internalDBGetBlock(x64emu_t* emu, uintptr_t addr, uintptr_t
    }
    // check size
    if(block) {
-        int blocksz = block->x64_size;
-        if(blocksz>my_context->max_db_size)
-            my_context->max_db_size = blocksz;
        // fill-in jumptable
        if(!addJumpTableIfDefault64(block->x64_addr, block->dirty?block->jmpnext:block->block)) {
            FreeDynablock(block, 0);
            block = getDB(addr);
            MarkDynablock(block);   // just in case...
        } else {
-            if(block->x64_size)
+            if(block->x64_size) {
+                if(block->x64_size>my_context->max_db_size) {
+                    my_context->max_db_size = block->x64_size;
+                    dynarec_log(LOG_INFO, "BOX64 Dynarec: higher max_db=%d\n", my_context->max_db_size);
+                }
                block->done = 1;    // don't validate the block if the size is null, but keep the block
+                rb_set(my_context->db_sizes, block->x64_size, block->x64_size+1, rb_get(my_context->db_sizes, block->x64_size)+1);
+            }
        }
    }
    if(need_lock)
@@ -246,33 +261,17 @@ static dynablock_t* internalDBGetBlock(x64emu_t* emu, uintptr_t addr, uintptr_t

 dynablock_t* DBGetBlock(x64emu_t* emu, uintptr_t addr, int create, int is32bits)
 {
+    if(isInHotPage(addr))
+        return NULL;
    dynablock_t *db = internalDBGetBlock(emu, addr, addr, create, 1, is32bits);
    if(db && db->done && db->block && getNeedTest(addr)) {
        if(db->always_test)
            sched_yield();  // just calm down...
-        if(AreaInHotPage((uintptr_t)db->x64_addr, (uintptr_t)db->x64_addr + db->x64_size - 1)) {
-            emu->test.test = 0;
-            if(box64_dynarec_fastpage) {
-                uint32_t hash = X31_hash_code(db->x64_addr, db->x64_size);
-                if(hash==db->hash) { // seems ok, run it without reprotecting it
-                    setJumpTableIfRef64(db->x64_addr, db->block, db->jmpnext);
-                    return db;
-                }
-                db->done = 0;   // invalidating the block, it's already not good
-                dynarec_log(LOG_DEBUG, "Invalidating block %p from %p:%p (hash:%X/%X) for %p\n", db, db->x64_addr, db->x64_addr+db->x64_size-1, hash, db->hash, (void*)addr);
-                // Free db, it's now invalid!
-                FreeDynablock(db, 1);
-                return NULL;    // not building a new one, it's still a hotpage
-            } else {
-                dynarec_log(LOG_INFO, "Not running block %p from %p:%p with for %p because it's in a hotpage\n", db, db->x64_addr, db->x64_addr+db->x64_size-1, (void*)addr);
-                return NULL;
-            }
-        }
        uint32_t hash = X31_hash_code(db->x64_addr, db->x64_size);
        int need_lock = mutex_trylock(&my_context->mutex_dyndump);
        if(hash!=db->hash) {
            db->done = 0;   // invalidating the block
-            dynarec_log(LOG_DEBUG, "Invalidating block %p from %p:%p (hash:%X/%X) for %p\n", db, db->x64_addr, db->x64_addr+db->x64_size-1, hash, db->hash, (void*)addr);
+            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);
            // Free db, it's now invalid!
            dynablock_t* old = InvalidDynablock(db, need_lock);
            // start again... (will create a new block)
@@ -284,12 +283,11 @@ dynablock_t* DBGetBlock(x64emu_t* emu, uintptr_t addr, int create, int is32bits)
            } else
                FreeInvalidDynablock(old, need_lock);
        } else {
-            dynarec_log(LOG_DEBUG, "Validating block %p from %p:%p (hash:%X) for %p\n", db, db->x64_addr, db->x64_addr+db->x64_size-1, db->hash, (void*)addr);
-            protectDB((uintptr_t)db->x64_addr, db->x64_size);
-            // fill back jumptable
-            if(isprotectedDB((uintptr_t)db->x64_addr, db->x64_size) && !db->always_test) {
-                setJumpTableIfRef64(db->x64_addr, db->block, db->jmpnext);
-            }
+            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
+                protectDBJumpTable((uintptr_t)db->x64_addr, db->x64_size, db->block, db->jmpnext);
        }
        if(!need_lock)
            mutex_unlock(&my_context->mutex_dyndump);
@@ -323,11 +321,10 @@ dynablock_t* DBAlternateBlock(x64emu_t* emu, uintptr_t addr, uintptr_t filladdr,
            } else
                FreeInvalidDynablock(old, need_lock);
        } else {
-            protectDB((uintptr_t)db->x64_addr, db->x64_size);
-            // fill back jumptable
-            if(isprotectedDB((uintptr_t)db->x64_addr, db->x64_size) && !db->always_test) {
-                setJumpTableIfRef64(db->x64_addr, db->block, db->jmpnext);
-            }
+            if(db->always_test)
+                protectDB((uintptr_t)db->x64_addr, db->x64_size);
+            else
+                protectDBJumpTable((uintptr_t)db->x64_addr, db->x64_size, db->block, db->jmpnext);
        }
        if(!need_lock)
            mutex_unlock(&my_context->mutex_dyndump);
@@ -19,6 +19,7 @@
 #include "dynablock_private.h"
 #include "bridge.h"
 #include "dynarec_next.h"
+#include "custommem.h"
 #endif
 #ifdef HAVE_TRACE
 #include "elfloader.h"
@@ -34,6 +35,9 @@ void* LinkNext(x64emu_t* emu, uintptr_t addr, void* x2, uintptr_t* x3)
    if(!addr) {
        dynablock_t* db = FindDynablockFromNativeAddress(x2-4);
        printf_log(LOG_NONE, "Warning, jumping to NULL address from %p (db=%p, x64addr=%p/%s)\n", x2-4, db, db?(void*)getX64Address(db, (uintptr_t)x2-4):NULL, db?getAddrFunctionName(getX64Address(db, (uintptr_t)x2-4)):"(nil)");
+    } else if(addr<0x10000) {
+        dynablock_t* db = FindDynablockFromNativeAddress(x2-4);
+        printf_log(LOG_NONE, "Warning, jumping to low 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
    void * jblock;
@@ -52,9 +56,13 @@ void* LinkNext(x64emu_t* emu, uintptr_t addr, void* x2, uintptr_t* x3)
    if(!block) {
        #ifdef HAVE_TRACE
        if(LOG_INFO<=box64_dynarec_log) {
-            dynablock_t* db = FindDynablockFromNativeAddress(x2-4);
-            elfheader_t* h = FindElfAddress(my_context, (uintptr_t)x2-4);
-            dynarec_log(LOG_INFO, "Warning, jumping to a no-block address %p from %p (db=%p, x64addr=%p(elf=%s))\n", (void*)addr, x2-4, db, db?(void*)getX64Address(db, (uintptr_t)x2-4):NULL, h?ElfName(h):"(none)");
+            if(checkInHotPage(addr)) {
+                dynarec_log(LOG_INFO, "Not trying to run a block from a Hotpage at %p\n", (void*)addr);
+            } else {
+                dynablock_t* db = FindDynablockFromNativeAddress(x2-4);
+                elfheader_t* h = FindElfAddress(my_context, (uintptr_t)x2-4);
+                dynarec_log(LOG_INFO, "Warning, jumping to a no-block address %p from %p (db=%p, x64addr=%p(elf=%s), RIP=%p)\n", (void*)addr, x2-4, db, db?(void*)getX64Address(db, (uintptr_t)x2-4):NULL, h?ElfName(h):"(none)", (void*)*x3);
+            }
        }
        #endif
        //tableupdate(native_epilog, addr, table);
@@ -62,6 +70,12 @@ void* LinkNext(x64emu_t* emu, uintptr_t addr, void* x2, uintptr_t* x3)
    }
    if(!block->done) {
        // not finished yet... leave linker
+        #ifdef HAVE_TRACE
+        if(box64_dynarec_log && !block->isize) {
+            dynablock_t* db = FindDynablockFromNativeAddress(x2-4);
+            printf_log(LOG_NONE, "Warning, NULL block at %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
        return native_epilog;
    }
    if(!(jblock=block->block)) {
@@ -73,117 +87,102 @@ void* LinkNext(x64emu_t* emu, uintptr_t addr, void* x2, uintptr_t* x3)
 }
 #endif

-#ifdef __GNUC__
-// Disable "clobbered" warnings
-#pragma GCC diagnostic push
-#pragma GCC diagnostic ignored "-Wclobbered"
-#endif
 void DynaCall(x64emu_t* emu, uintptr_t addr)
 {
-    // prepare setjump for signal handling
-    struct __jmp_buf_tag jmpbuf[1] = {0};
-    int skip = 0;
-    struct __jmp_buf_tag *old_jmpbuf = emu->jmpbuf;
-    emu->jmpbuf = jmpbuf;
+    uint64_t old_rsp = R_RSP;
+    uint64_t old_rbx = R_RBX;
+    uint64_t old_rdi = R_RDI;
+    uint64_t old_rsi = R_RSI;
+    uint64_t old_rbp = R_RBP;
+    uint64_t old_rip = R_RIP;
+    // save defered flags
+    deferred_flags_t old_df = emu->df;
+    multiuint_t old_op1 = emu->op1;
+    multiuint_t old_op2 = emu->op2;
+    multiuint_t old_res = emu->res;
+    multiuint_t old_op1_sav= emu->op1_sav;
+    multiuint_t old_res_sav= emu->res_sav;
+    deferred_flags_t old_df_sav= emu->df_sav;
+    // uc_link
+    x64_ucontext_t* old_uc_link = emu->uc_link;
+    emu->uc_link = NULL;

-    if((skip = sigsetjmp(emu->jmpbuf, 1))) {
-        printf_log(LOG_DEBUG, "Setjmp DynaCall, fs=0x%x\n", emu->segs[_FS]);
-        addr = R_RIP;   // not sure if it should still be inside DynaCall!
-        #ifdef DYNAREC
-        if(box64_dynarec_test) {
-            if(emu->test.clean)
-                x64test_check(emu, R_RIP);
-            emu->test.clean = 0;
-        }
-        #endif
+    PushExit(emu);
+    R_RIP = addr;
+    emu->df = d_none;
+    DynaRun(emu);
+    emu->quit = 0;  // reset Quit flags...
+    emu->df = d_none;
+    emu->uc_link = old_uc_link;
+    if(emu->flags.quitonlongjmp && emu->flags.longjmp) {
+        if(emu->flags.quitonlongjmp==1)
+            emu->flags.longjmp = 0;   // don't change anything because of the longjmp
+    } else {
+        // restore defered flags
+        emu->df = old_df;
+        emu->op1 = old_op1;
+        emu->op2 = old_op2;
+        emu->res = old_res;
+        emu->op1_sav = old_op1_sav;
+        emu->res_sav = old_res_sav;
+        emu->df_sav = old_df_sav;
+        // and the old registers
+        R_RBX = old_rbx;
+        R_RDI = old_rdi;
+        R_RSI = old_rsi;
+        R_RBP = old_rbp;
+        R_RSP = old_rsp;
+        R_RIP = old_rip;  // and set back instruction pointer
    }
-#ifdef DYNAREC
-    if(!box64_dynarec)
-#endif
-        EmuCall(emu, addr);
-#ifdef DYNAREC
-    else {
-        uint64_t old_rsp = R_RSP;
-        uint64_t old_rbx = R_RBX;
-        uint64_t old_rdi = R_RDI;
-        uint64_t old_rsi = R_RSI;
-        uint64_t old_rbp = R_RBP;
-        uint64_t old_rip = R_RIP;
-        PushExit(emu);
-        R_RIP = addr;
-        emu->df = d_none;
-        while(!emu->quit) {
-            int is32bits = (emu->segs[_CS]==0x23);
-            dynablock_t* block = (skip)?NULL:DBGetBlock(emu, R_RIP, 1, is32bits);
-            if(!block || !block->block || !block->done) {
-                skip = 0;
-                // no block, of block doesn't have DynaRec content (yet, temp is not null)
-                // Use interpreter (should use single instruction step...)
-                dynarec_log(LOG_DEBUG, "%04d|Calling Interpreter @%p, emu=%p\n", GetTID(), (void*)R_RIP, emu);
-                if(box64_dynarec_test)
-                    emu->test.clean = 0;
-                Run(emu, 1);
-            } else {
-                dynarec_log(LOG_DEBUG, "%04d|Calling DynaRec Block @%p (%p) of %d x64 instructions (hash=0x%x) emu=%p\n", GetTID(), (void*)R_RIP, block->block, block->isize ,block->hash, emu);
-                CHECK_FLAGS(emu);
-                // block is here, let's run it!
-                native_prolog(emu, block->block);
-            }
-            if(emu->fork) {
-                int forktype = emu->fork;
-                emu->quit = 0;
-                emu->fork = 0;
-                emu = x64emu_fork(emu, forktype);
-            }
-        }
-        emu->quit = 0;  // reset Quit flags...
-        emu->df = d_none;
-        if(emu->quitonlongjmp && emu->longjmp) {
-            if(emu->quitonlongjmp==1)
-                emu->longjmp = 0;   // don't change anything because of the longjmp
-        } else {
-            R_RBX = old_rbx;
-            R_RDI = old_rdi;
-            R_RSI = old_rsi;
-            R_RBP = old_rbp;
-            R_RSP = old_rsp;
-            R_RIP = old_rip;  // and set back instruction pointer
-        }
-    }
-#endif
-    // clear the setjmp
-    emu->jmpbuf = old_jmpbuf;
 }

-int DynaRun(x64emu_t* emu)
+int my_setcontext(x64emu_t* emu, void* ucp);
+void DynaRun(x64emu_t* emu)
 {
    // prepare setjump for signal handling
-    struct __jmp_buf_tag jmpbuf[1] = {0};
+    JUMPBUFF jmpbuf[1] = {0};
    int skip = 0;
-    struct __jmp_buf_tag *old_jmpbuf = emu->jmpbuf;
-    emu->jmpbuf = jmpbuf;
+    JUMPBUFF *old_jmpbuf = emu->jmpbuf;
+    #ifdef RV64
+    uintptr_t old_savesp = emu->xSPSave;
+    #endif
+    emu->flags.jmpbuf_ready = 0;

-    if((skip=sigsetjmp(emu->jmpbuf, 1))) {
-        printf_log(LOG_DEBUG, "Setjmp DynaRun, fs=0x%x\n", emu->segs[_FS]);
-        #ifdef DYNAREC
-        if(box64_dynarec_test) {
-            if(emu->test.clean)
-                x64test_check(emu, R_RIP);
-            emu->test.clean = 0;
+    while(!(emu->quit)) {
+        if(!emu->jmpbuf || (emu->flags.need_jmpbuf && emu->jmpbuf!=jmpbuf)) {
+            emu->jmpbuf = jmpbuf;
+            #ifdef RV64
+            emu->old_savedsp = emu->xSPSave;
+            #endif
+            emu->flags.jmpbuf_ready = 1;
+            #ifdef ANDROID
+            if((skip=sigsetjmp(*(JUMPBUFF*)emu->jmpbuf, 1)))
+            #else
+            if((skip=sigsetjmp(emu->jmpbuf, 1)))
+            #endif
+            {
+                printf_log(LOG_DEBUG, "Setjmp DynaRun, fs=0x%x\n", emu->segs[_FS]);
+                #ifdef DYNAREC
+                if(box64_dynarec_test) {
+                    if(emu->test.clean)
+                        x64test_check(emu, R_RIP);
+                    emu->test.clean = 0;
+                }
+                #endif
+            }
        }
-        #endif
-    }
+        if(emu->flags.need_jmpbuf)
+            emu->flags.need_jmpbuf = 0;

 #ifdef DYNAREC
-    if(!box64_dynarec)
+        if(!box64_dynarec)
 #endif
-        return Run(emu, 0);
+            Run(emu, 0);
 #ifdef DYNAREC
-    else {
-        while(!emu->quit) {
+        else {
            int is32bits = (emu->segs[_CS]==0x23);
            dynablock_t* block = (skip)?NULL:DBGetBlock(emu, R_RIP, 1, is32bits);
-            if(!block || !block->block || !block->done) {
+            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)
                // Use interpreter (should use single instruction step...)
@@ -195,6 +194,9 @@ int DynaRun(x64emu_t* emu)
                dynarec_log(LOG_DEBUG, "%04d|Running DynaRec Block @%p (%p) of %d x64 insts (hash=0x%x) emu=%p\n", GetTID(), (void*)R_RIP, block->block, block->isize, block->hash, emu);
                // block is here, let's run it!
                native_prolog(emu, block->block);
+                extern int running32bits;
+                if(emu->segs[_CS]==0x23)
+                    running32bits = 1;
            }
            if(emu->fork) {
                int forktype = emu->fork;
@@ -202,13 +204,18 @@ int DynaRun(x64emu_t* emu)
                emu->fork = 0;
                emu = x64emu_fork(emu, forktype);
            }
+            if(emu->quit && emu->uc_link) {
+                emu->quit = 0;
+                my_setcontext(emu, emu->uc_link);
+            }
        }
+#endif
+        if(emu->flags.need_jmpbuf)
+            emu->quit = 0;
    }
    // clear the setjmp
    emu->jmpbuf = old_jmpbuf;
-    return 0;
-#endif
+    #ifdef RV64
+    emu->xSPSave = old_savesp;
+    #endif
 }
-#ifdef __GNUC__
-#pragma GCC diagnostic pop
-#endif
@@ -15,18 +15,24 @@
 #include "arm64/arm64_printer.h"
 #include "arm64/dynarec_arm64_private.h"
 #include "arm64/dynarec_arm64_functions.h"
-#elif defined(LA464)
+// Limit here is defined by LD litteral, that is 19bits
+#define MAXBLOCK_SIZE ((1<<19)-200)
+#elif defined(LA64)

-#define instruction_native_t        instruction_la464_t
-#define dynarec_native_t            dynarec_la464_t
+#define instruction_native_t        instruction_la64_t
+#define dynarec_native_t            dynarec_la64_t

-#define ADDITIONNAL_DEFINITION()
+#define ADDITIONNAL_DEFINITION() \
+    int fpuCacheNeedsTransform(dynarec_native_t* dyn, int ninst);

-#define OTHER_CACHE()
+#define OTHER_CACHE() \
+    if (fpuCacheNeedsTransform(dyn, ninst)) ret |= 2;

-#include "la464/la464_printer.h"
-#include "la464/dynarec_la464_private.h"
-#include "la464/dynarec_la464_functions.h"
+#include "la64/la64_printer.h"
+#include "la64/dynarec_la64_private.h"
+#include "la64/dynarec_la64_functions.h"
+// Limit here is unconditionnal jump, that is signed 28bits
+#define MAXBLOCK_SIZE ((1 << 27) - 200)
 #elif defined(RV64)

 #define instruction_native_t        instruction_rv64_t
@@ -41,6 +47,8 @@
 #include "rv64/rv64_printer.h"
 #include "rv64/dynarec_rv64_private.h"
 #include "rv64/dynarec_rv64_functions.h"
+// Limit here is unconditionnal jump, that is signed 21bits
+#define MAXBLOCK_SIZE ((1<<20)-200)
 #else
 #error Unsupported platform
 #endif
@@ -3,8 +3,8 @@

 #ifdef ARM64
 #include "arm64/dynarec_arm64_helper.h"
-#elif defined(LA464)
-#include "la464/dynarec_la464_helper.h"
+#elif defined(LA64)
+#include "la64/dynarec_la64_helper.h"
 #elif defined(RV64)
 #include "rv64/dynarec_rv64_helper.h"
 #else
@@ -72,13 +72,12 @@ void add_next(dynarec_native_t *dyn, uintptr_t addr) {
        }
    // add slots
    if(dyn->next_sz == dyn->next_cap) {
-        dyn->next_cap += 64;
-        dyn->next = (uintptr_t*)dynaRealloc(dyn->next, dyn->next_cap*sizeof(uintptr_t));
+        printf_log(LOG_NONE, "Warning, overallocating next\n");
    }
    dyn->next[dyn->next_sz++] = addr;
 }
 uintptr_t get_closest_next(dynarec_native_t *dyn, uintptr_t addr) {
-    // get closest, but no addresses befores
+    // get closest, but no addresses before
    uintptr_t best = 0;
    int i = 0;
    while((i<dyn->next_sz) && (best!=addr)) {
@@ -94,6 +93,20 @@ uintptr_t get_closest_next(dynarec_native_t *dyn, uintptr_t addr) {
    }
    return best;
 }
+void add_jump(dynarec_native_t *dyn, int ninst) {
+    // add slots
+    if(dyn->jmp_sz == dyn->jmp_cap) {
+        printf_log(LOG_NONE, "Warning, overallocating jmps\n");
+    }
+    dyn->jmps[dyn->jmp_sz++] = ninst;
+}
+int get_first_jump(dynarec_native_t *dyn, int next) {
+    for(int i=0; i<dyn->jmp_sz; ++i)
+        if(dyn->insts[dyn->jmps[i]].x64.jmp == next)
+            return dyn->jmps[i];
+    return -2;
+}
+
 #define PK(A) (*((uint8_t*)(addr+(A))))
 int is_nops(dynarec_native_t *dyn, uintptr_t addr, int n)
 {
@@ -124,8 +137,8 @@ int is_nops(dynarec_native_t *dyn, uintptr_t addr, int n)
    return 0;
 }

-// return size of next instuciton, -1 is unknown
-// not all instrction are setup
+// return size of next instruction, -1 is unknown
+// not all instructions are setup
 int next_instruction(dynarec_native_t *dyn, uintptr_t addr)
 {
    uint8_t opcode = PK(0);
@@ -274,15 +287,10 @@ int Table64(dynarec_native_t *dyn, uint64_t val, int pass)
            idx = i;
    // not found, add it
    if(idx==-1) {
-        if(dyn->table64size == dyn->table64cap) {
-            dyn->table64cap+=16;
-            if(pass<3)  // do not resize on pass3, it's not the same type of memory anymore
-                dyn->table64 = (uint64_t*)dynaRealloc(dyn->table64, dyn->table64cap * sizeof(uint64_t));
-        }
        idx = dyn->table64size++;
        if(idx < dyn->table64cap)
            dyn->table64[idx] = val;
-        else
+        else if(pass==3)
            printf_log(LOG_NONE, "Warning, table64 bigger than expected %d vs %d\n", idx, dyn->table64cap);
    }
    // calculate offset
@@ -290,17 +298,36 @@ int Table64(dynarec_native_t *dyn, uint64_t val, int pass)
    return delta;
 }

-static void fillPredecessors(dynarec_native_t* dyn)
+static void recurse_mark_alive(dynarec_native_t* dyn, int i)
+{
+    if(dyn->insts[i].x64.alive)
+        return;
+    dyn->insts[i].x64.alive = 1;
+    if(dyn->insts[i].x64.jmp && dyn->insts[i].x64.jmp_insts!=-1)
+        recurse_mark_alive(dyn, dyn->insts[i].x64.jmp_insts);
+    if(i<dyn->size-1 && dyn->insts[i].x64.has_next)
+        recurse_mark_alive(dyn, i+1);
+}
+
+static int sizePredecessors(dynarec_native_t* dyn)
 {
    int pred_sz = 1;    // to be safe
-    // compute total size of predecessor to alocate the array
+    // compute total size of predecessor to allocate the array
+    // mark alive...
+    recurse_mark_alive(dyn, 0);
    // first compute the jumps
+    int jmpto;
    for(int i=0; i<dyn->size; ++i) {
-        if(dyn->insts[i].x64.jmp && (dyn->insts[i].x64.jmp_insts!=-1)) {
+        if(dyn->insts[i].x64.alive && dyn->insts[i].x64.jmp && ((jmpto=dyn->insts[i].x64.jmp_insts)!=-1)) {
            pred_sz++;
-            dyn->insts[dyn->insts[i].x64.jmp_insts].pred_sz++;
+            dyn->insts[jmpto].pred_sz++;
        }
    }
+    // remove "has_next" from orphan branch
+    for(int i=0; i<dyn->size-1; ++i) {
+        if(dyn->insts[i].x64.has_next && !dyn->insts[i+1].x64.alive)
+            dyn->insts[i].x64.has_next = 0;
+    }
    // second the "has_next"
    for(int i=0; i<dyn->size-1; ++i) {
        if(dyn->insts[i].x64.has_next) {
@@ -308,8 +335,10 @@ static void fillPredecessors(dynarec_native_t* dyn)
            dyn->insts[i+1].pred_sz++;
        }
    }
-    int alloc_size = pred_sz;
-    dyn->predecessor = (int*)dynaMalloc(pred_sz*sizeof(int));
+    return pred_sz;
+}
+static void fillPredecessors(dynarec_native_t* dyn)
+{
    // fill pred pointer
    int* p = dyn->predecessor;
    for(int i=0; i<dyn->size; ++i) {
@@ -318,7 +347,7 @@ static void fillPredecessors(dynarec_native_t* dyn)
        dyn->insts[i].pred_sz=0;  // reset size, it's reused to actually fill pred[]
    }
    // fill pred
-    for(int i=0; i<dyn->size; ++i) {
+    for(int i=0; i<dyn->size; ++i) if(dyn->insts[i].x64.alive) {
        if((i!=dyn->size-1) && dyn->insts[i].x64.has_next)
            dyn->insts[i+1].pred[dyn->insts[i+1].pred_sz++] = i;
        if(dyn->insts[i].x64.jmp && (dyn->insts[i].x64.jmp_insts!=-1)) {
@@ -328,19 +357,15 @@ static void fillPredecessors(dynarec_native_t* dyn)
    }
 }

-// updateNeed goes backward, from last intruction to top
+// updateNeed goes backward, from last instruction to top
 static int updateNeed(dynarec_native_t* dyn, int ninst, uint8_t need) {
    while (ninst>=0) {
        // need pending but instruction is only a subset: remove pend and use an X_ALL instead
        need |= dyn->insts[ninst].x64.need_after;
-        if((need&X_PEND) && (dyn->insts[ninst].x64.state_flags==SF_SUBSET)) {
+        if((need&X_PEND) && (dyn->insts[ninst].x64.state_flags==SF_SUBSET || dyn->insts[ninst].x64.state_flags==SF_SET || dyn->insts[ninst].x64.state_flags==SF_SET_NODF)) {
            need &=~X_PEND;
            need |= X_ALL;
        }
-        if((need&X_PEND) && (dyn->insts[ninst].x64.state_flags==SF_SET)) {
-            need &=~X_PEND;
-            need |= dyn->insts[ninst].x64.set_flags;    // SF_SET will compute all flags, it's not SUBSET!
-        }
        if((need&X_PEND) && dyn->insts[ninst].x64.state_flags==SF_SUBSET_PENDING) {
            need |= X_ALL&~(dyn->insts[ninst].x64.set_flags);
        }
@@ -349,11 +374,11 @@ static int updateNeed(dynarec_native_t* dyn, int ninst, uint8_t need) {
            dyn->insts[ninst].x64.gen_flags |= X_PEND;
        dyn->insts[ninst].x64.need_after = need;
        need = dyn->insts[ninst].x64.need_after&~dyn->insts[ninst].x64.gen_flags;
+
        if(dyn->insts[ninst].x64.may_set)
            need |= dyn->insts[ninst].x64.gen_flags;    // forward the flags
-        // Consume X_PEND if relevant
-        if((need&X_PEND) && (dyn->insts[ninst].x64.set_flags&SF_PENDING))
-            need &=~X_PEND;
+        else if((need&X_PEND) && (dyn->insts[ninst].x64.set_flags&SF_PENDING))
+            need &=~X_PEND;         // Consume X_PEND if relevant
        need |= dyn->insts[ninst].x64.use_flags;
        if(dyn->insts[ninst].x64.need_before == need)
            return ninst - 1;
@@ -368,32 +393,32 @@ static int updateNeed(dynarec_native_t* dyn, int ninst, uint8_t need) {
            else
                updateNeed(dyn, dyn->insts[ninst].pred[i], need);
        }
-        if(!ok)
-            return ninst - 1;
        --ninst;
+        if(!ok)
+            return ninst;
    }
    return ninst;
 }

 void* current_helper = NULL;
+static int static_jmps[MAX_INSTS+2];
+static uintptr_t static_next[MAX_INSTS+2];
+static uint64_t static_table64[(MAX_INSTS+3)/4];
+static instruction_native_t static_insts[MAX_INSTS+2] = {0};
+// TODO: ninst could be a uint16_t instead of an int, that could same some temp. memory

 void CancelBlock64(int need_lock)
 {
    if(need_lock)
        mutex_lock(&my_context->mutex_dyndump);
    dynarec_native_t* helper = (dynarec_native_t*)current_helper;
-    current_helper = NULL;
    if(helper) {
-        dynaFree(helper->next);
-        dynaFree(helper->insts);
-        dynaFree(helper->predecessor);
-        if(helper->table64 && (helper->table64!=(uint64_t*)helper->tablestart))
-            dynaFree(helper->table64);
        if(helper->dynablock && helper->dynablock->actual_block) {
            FreeDynarecMap((uintptr_t)helper->dynablock->actual_block);
            helper->dynablock->actual_block = NULL;
        }
    }
+    current_helper = NULL;
    if(need_lock)
        mutex_unlock(&my_context->mutex_dyndump);
 }
@@ -410,7 +435,7 @@ void* CreateEmptyBlock(dynablock_t* block, uintptr_t addr) {
    void* actual_p = (void*)AllocDynarecMap(sz);
    void* p = actual_p + sizeof(void*);
    if(actual_p==NULL) {
-        dynarec_log(LOG_INFO, "AllocDynarecMap(%p, %zu) failed, cancelling block\n", block, sz);
+        dynarec_log(LOG_INFO, "AllocDynarecMap(%p, %zu) failed, canceling block\n", block, sz);
        CancelBlock64(0);
        return NULL;
    }
@@ -435,21 +460,17 @@ void* FillBlock64(dynablock_t* block, uintptr_t addr, int alternate, int is32bit
        A ..    A+8*n   : Table64: n 64bits values
        B ..    B+7     : dynablock_t* : self (as part of JmpNext, that simulate another block)
        B+8 ..  B+15    : 2 Native code for jmpnext (or jmp epilog in case of empty block)
-        B+16 .. B+23    : jmpnext (or jmp_epilog) address
+        B+16 .. B+23    : jmpnext (or jmp_epilog) address. jumpnext is used when the block needs testing
        B+24 .. B+31    : empty (in case an architecture needs more than 2 opcodes)
-        B+32 .. B+32+sz : instsize (compressed array with each instruction lenght on x64 and native side)
+        B+32 .. B+32+sz : instsize (compressed array with each instruction length on x64 and native side)

    */
-    if(IsInHotPage(addr)) {
-        dynarec_log(LOG_DEBUG, "Cancelling dynarec FillBlock on hotpage for %p\n", (void*)addr);
-        return NULL;
-    }
    if(addr>=box64_nodynarec_start && addr<box64_nodynarec_end) {
        dynarec_log(LOG_INFO, "Create empty block in no-dynarec zone\n");
        return CreateEmptyBlock(block, addr);
    }
    if(current_helper) {
-        dynarec_log(LOG_DEBUG, "Cancelling dynarec FillBlock at %p as anothor one is going on\n", (void*)addr);
+        dynarec_log(LOG_DEBUG, "Canceling dynarec FillBlock at %p as another one is going on\n", (void*)addr);
        return NULL;
    }
    // protect the 1st page
@@ -460,14 +481,21 @@ void* FillBlock64(dynablock_t* block, uintptr_t addr, int alternate, int is32bit
    helper.dynablock = block;
    helper.start = addr;
    uintptr_t start = addr;
-    helper.cap = 128;
-    helper.insts = (instruction_native_t*)dynaCalloc(helper.cap, sizeof(instruction_native_t));
+    helper.cap = MAX_INSTS;
+    helper.insts = static_insts;
+    helper.jmps = static_jmps;
+    helper.jmp_cap = MAX_INSTS;
+    helper.next = static_next;
+    helper.next_cap = MAX_INSTS;
+    helper.table64 = static_table64;
+    helper.table64cap = sizeof(static_table64)/sizeof(uint64_t);
    // pass 0, addresses, x64 jump addresses, overall size of the block
    uintptr_t end = native_pass0(&helper, addr, alternate, is32bits);
-    // no need for next anymore
-    dynaFree(helper.next);
-    helper.next_sz = helper.next_cap = 0;
-    helper.next = NULL;
+    if(helper.abort) {
+        if(box64_dynarec_dump || box64_dynarec_log)dynarec_log(LOG_NONE, "Abort dynablock on pass0\n");
+        CancelBlock64(0);
+        return NULL;
+    }
    // basic checks
    if(!helper.size) {
        dynarec_log(LOG_INFO, "Warning, null-sized dynarec block (%p)\n", (void*)addr);
@@ -476,48 +504,99 @@ void* FillBlock64(dynablock_t* block, uintptr_t addr, int alternate, int is32bit
    }
    if(!isprotectedDB(addr, 1)) {
        dynarec_log(LOG_INFO, "Warning, write on current page on pass0, aborting dynablock creation (%p)\n", (void*)addr);
-        AddHotPage(addr);
        CancelBlock64(0);
        return NULL;
    }
    // protect the block of it goes over the 1st page
-    if((addr&~box64_pagesize)!=(end&~box64_pagesize)) // need to protect some other pages too
+    if((addr&~(box64_pagesize-1))!=(end&~(box64_pagesize-1))) // need to protect some other pages too
        protectDB(addr, end-addr);  //end is 1byte after actual end
    // compute hash signature
    uint32_t hash = X31_hash_code((void*)addr, end-addr);
    // calculate barriers
-    for(int i=0; i<helper.size; ++i)
-        if(helper.insts[i].x64.jmp) {
-            uintptr_t j = helper.insts[i].x64.jmp;
-            if(j<start || j>=end || j==helper.insts[i].x64.addr) {
-                if(j==helper.insts[i].x64.addr) // if there is a loop on some opcode, make the block "always to tested"
-                    helper.always_test = 1;
-                helper.insts[i].x64.jmp_insts = -1;
-                helper.insts[i].x64.need_after |= X_PEND;
-            } else {
-                // find jump address instruction
-                int k=-1;
-                for(int i2=0; i2<helper.size && k==-1; ++i2) {
-                    if(helper.insts[i2].x64.addr==j)
-                        k=i2;
+    for(int ii=0; ii<helper.jmp_sz; ++ii) {
+        int i = helper.jmps[ii];
+        uintptr_t j = helper.insts[i].x64.jmp;
+        helper.insts[i].x64.jmp_insts = -1;
+        if(j<start || j>=end || j==helper.insts[i].x64.addr) {
+            if(j==helper.insts[i].x64.addr) // if there is a loop on some opcode, make the block "always to tested"
+                helper.always_test = 1;
+            helper.insts[i].x64.need_after |= X_PEND;
+        } else {
+            // find jump address instruction
+            int k=-1;
+            int search = ((j>=helper.insts[0].x64.addr) && j<helper.insts[0].x64.addr+helper.isize)?1:0;
+            int imin = 0;
+            int imax = helper.size-1;
+            int i2 = helper.size/2;
+            // dichotomy search
+            while(search) {
+                if(helper.insts[i2].x64.addr == j) {
+                    k = i2;
+                    search = 0;
+                } else if(helper.insts[i2].x64.addr>j) {
+                    imax = i2;
+                    i2 = (imax+imin)/2;
+                } else {
+                    imin = i2;
+                    i2 = (imax+imin)/2;
                }
+                if(search && (imax-imin)<2) {
+                    search = 0;
+                    if(helper.insts[imin].x64.addr==j)
+                        k = imin;
+                    else if(helper.insts[imax].x64.addr==j)
+                        k = imax;
+                }
+            }
+            /*for(int i2=0; i2<helper.size && k==-1; ++i2) {
+                if(helper.insts[i2].x64.addr==j)
+                    k=i2;
+            }*/
+            if(k!=-1) {
                if(k!=-1 && !helper.insts[i].barrier_maybe)
                    helper.insts[k].x64.barrier |= BARRIER_FULL;
                helper.insts[i].x64.jmp_insts = k;
            }
        }
+    }
+    // no need for next and jmps anymore
+    helper.next_sz = helper.next_cap = 0;
+    helper.next = NULL;
+    helper.jmp_sz = helper.jmp_cap = 0;
+    helper.jmps = NULL;
    // fill predecessors with the jump address
+    int alloc_size = sizePredecessors(&helper);
+    helper.predecessor = (int*)alloca(alloc_size*sizeof(int));
    fillPredecessors(&helper);

    int pos = helper.size;
    while (pos>=0)
        pos = updateNeed(&helper, pos, 0);
+    // remove fpu stuff on non-executed code
+    for(int i=1; i<helper.size-1; ++i)
+        if(!helper.insts[i].pred_sz) {
+            int ii = i;
+            while(ii<helper.size && !helper.insts[ii].pred_sz)
+                fpu_reset_ninst(&helper, ii++);
+            i = ii;
+        }

-    // pass 1, float optimisations, first pass for flags
+
+    // pass 1, float optimizations, first pass for flags
    native_pass1(&helper, addr, alternate, is32bits);
+    if(helper.abort) {
+        if(box64_dynarec_dump || box64_dynarec_log)dynarec_log(LOG_NONE, "Abort dynablock on pass1\n");
+        CancelBlock64(0);
+        return NULL;
+    }
    
    // pass 2, instruction size
    native_pass2(&helper, addr, alternate, is32bits);
+    if(helper.abort) {
+        if(box64_dynarec_dump || box64_dynarec_log)dynarec_log(LOG_NONE, "Abort dynablock on pass2\n");
+        CancelBlock64(0);
+        return NULL;
+    }
    // keep size of instructions for signal handling
    size_t insts_rsize = (helper.insts_size+2)*sizeof(instsize_t);
    insts_rsize = (insts_rsize+7)&~7;   // round the size...
@@ -531,7 +610,7 @@ void* FillBlock64(dynablock_t* block, uintptr_t addr, int alternate, int is32bit
    void* next = tablestart + helper.table64size*sizeof(uint64_t);
    void* instsize = next + 4*sizeof(void*);
    if(actual_p==NULL) {
-        dynarec_log(LOG_INFO, "AllocDynarecMap(%p, %zu) failed, cancelling block\n", block, sz);
+        dynarec_log(LOG_INFO, "AllocDynarecMap(%p, %zu) failed, canceling block\n", block, sz);
        CancelBlock64(0);
        return NULL;
    }
@@ -543,7 +622,6 @@ void* FillBlock64(dynablock_t* block, uintptr_t addr, int alternate, int is32bit
    helper.instsize = (instsize_t*)instsize;
    *(dynablock_t**)actual_p = block;
    helper.table64cap = helper.table64size;
-    dynaFree(helper.table64);
    helper.table64 = (uint64_t*)helper.tablestart;
    // pass 3, emit (log emit native opcode)
    if(box64_dynarec_dump) {
@@ -558,14 +636,15 @@ void* FillBlock64(dynablock_t* block, uintptr_t addr, int alternate, int is32bit
    helper.table64size = 0; // reset table64 (but not the cap)
    helper.insts_size = 0;  // reset
    native_pass3(&helper, addr, alternate, is32bits);
+    if(helper.abort) {
+        if(box64_dynarec_dump || box64_dynarec_log)dynarec_log(LOG_NONE, "Abort dynablock on pass1\n");
+        CancelBlock64(0);
+        return NULL;
+    }
    // keep size of instructions for signal handling
    block->instsize = instsize;
-    // ok, free the helper now
-    dynaFree(helper.insts);
-    helper.insts = NULL;
    helper.table64 = NULL;
    helper.instsize = NULL;
-    dynaFree(helper.predecessor);
    helper.predecessor = NULL;
    block->size = sz;
    block->isize = helper.size;
@@ -581,10 +660,9 @@ void* FillBlock64(dynablock_t* block, uintptr_t addr, int alternate, int is32bit
    // all done...
    __clear_cache(actual_p, actual_p+sz);   // need to clear the cache before execution...
    block->hash = X31_hash_code(block->x64_addr, block->x64_size);
-    // Check if something changed, to abbort if it as
+    // Check if something changed, to abort if it is
    if((block->hash != hash)) {
        dynarec_log(LOG_DEBUG, "Warning, a block changed while being processed hash(%p:%ld)=%x/%x\n", block->x64_addr, block->x64_size, block->hash, hash);
-        AddHotPage(addr);
        CancelBlock64(0);
        return NULL;
    }
@@ -603,15 +681,24 @@ void* FillBlock64(dynablock_t* block, uintptr_t addr, int alternate, int is32bit
        CancelBlock64(0);
        return NULL;
    }
+    // ok, free the helper now
+    //dynaFree(helper.insts);
+    helper.insts = NULL;
    if(insts_rsize/sizeof(instsize_t)<helper.insts_size) {
-        printf_log(LOG_NONE, "BOX64: Warning, ists_size difference in block between pass2 (%zu) and pass3 (%zu), allocated: %zu\n", oldinstsize, helper.insts_size, insts_rsize/sizeof(instsize_t));
+        printf_log(LOG_NONE, "BOX64: Warning, insts_size difference in block between pass2 (%zu) and pass3 (%zu), allocated: %zu\n", oldinstsize, helper.insts_size, insts_rsize/sizeof(instsize_t));
    }
    if(!isprotectedDB(addr, end-addr)) {
        dynarec_log(LOG_DEBUG, "Warning, block unprotected while being processed %p:%ld, marking as need_test\n", block->x64_addr, block->x64_size);
-        AddHotPage(addr);
        block->dirty = 1;
        //protectDB(addr, end-addr);
    }
+    if(getProtection(addr)&PROT_NEVERCLEAN) {
+        block->dirty = 1;
+        block->always_test = 1;
+    }
+    if(block->always_test) {
+        dynarec_log(LOG_DEBUG, "Note: block marked as always dirty %p:%ld\n", block->x64_addr, block->x64_size);
+    }
    current_helper = NULL;
    //block->done = 1;
    return (void*)block;
@@ -66,12 +66,24 @@ void native_fxtract(x64emu_t* emu)
 }
 void native_fprem(x64emu_t* emu)
 {
-    int32_t tmp32s = ST0.d / ST1.d;
-    ST0.d -= ST1.d * tmp32s;
-    emu->sw.f.F87_C2 = 0;
-    emu->sw.f.F87_C1 = (tmp32s&1);
-    emu->sw.f.F87_C3 = ((tmp32s>>1)&1);
-    emu->sw.f.F87_C0 = ((tmp32s>>2)&1);
+    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;
+    } 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;
+    }
 }
 void native_fyl2xp1(x64emu_t* emu)
 {
@@ -138,7 +150,7 @@ void native_fistp64(x64emu_t* emu, int64_t* ed)

 void native_fistt64(x64emu_t* emu, int64_t* ed)
 {
-    // used of memcpy to avoid aligments issues
+    // used of memcpy to avoid alignments issues
    int64_t tmp = ST0.d;
    memcpy(ed, &tmp, sizeof(tmp));
 }
@@ -163,11 +175,28 @@ void native_priv(x64emu_t* emu)
    emit_signal(emu, SIGSEGV, (void*)R_RIP, 0);
 }

+void native_int(x64emu_t* emu, int num)
+{
+    emu->test.test = 0;
+    emit_interruption(emu, num, (void*)R_RIP);
+}
+
 void native_singlestep(x64emu_t* emu)
 {
    emit_signal(emu, SIGTRAP, (void*)R_RIP, 1);
 }

+void native_int3(x64emu_t* emu)
+{
+    emit_signal(emu, SIGTRAP, (void*)R_RIP, 128);
+}
+
+void native_div0(x64emu_t* emu)
+{
+    emu->test.test = 0;
+    emit_div0(emu,  (void*)R_RIP, 0);
+}
+
 void native_fsave(x64emu_t* emu, uint8_t* ed)
 {
    fpu_savenv(emu, (char*)ed, 0);
@@ -194,13 +223,24 @@ void native_frstor(x64emu_t* emu, uint8_t* ed)

 void native_fprem1(x64emu_t* emu)
 {
-    // simplified version
-    int32_t tmp32s = round(ST0.d / ST1.d);
-    ST0.d -= ST1.d*tmp32s;
-    emu->sw.f.F87_C2 = 0;
-    emu->sw.f.F87_C1 = (tmp32s&1);
-    emu->sw.f.F87_C3 = ((tmp32s>>1)&1);
-    emu->sw.f.F87_C0 = ((tmp32s>>2)&1);
+    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;
+    } 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;
+    }
 }

 static uint8_t ff_mult(uint8_t a, uint8_t b)
@@ -365,9 +405,8 @@ static int flagsCacheNeedsTransform(dynarec_native_t* dyn, int ninst) {
        return 0;
    if(dyn->insts[ninst].f_exit.dfnone)  // flags are fully known, nothing we can do more
        return 0;
-/*    if((dyn->f.pending!=SF_SET)
-    && (dyn->f.pending!=SF_SET_PENDING)) {
-        if(dyn->f.pending!=SF_PENDING) {*/
+    if(dyn->insts[jmp].f_entry.dfnone && !dyn->insts[ninst].f_exit.dfnone)
+        return 1;
    switch (dyn->insts[jmp].f_entry.pending) {
        case SF_UNKNOWN: return 0;
        case SF_SET:
@@ -376,22 +415,14 @@ static int flagsCacheNeedsTransform(dynarec_native_t* dyn, int ninst) {
            else
                return 0;
        case SF_SET_PENDING:
-            if(dyn->insts[ninst].f_exit.pending!=SF_SET
-            && dyn->insts[ninst].f_exit.pending!=SF_SET_PENDING
-            && dyn->insts[ninst].f_exit.pending!=SF_PENDING)
-                return 1;
-            else
+            if(dyn->insts[ninst].f_exit.pending==SF_SET_PENDING)
                return 0;
+            return 1;
        case SF_PENDING:
-            if(dyn->insts[ninst].f_exit.pending!=SF_SET
-            && dyn->insts[ninst].f_exit.pending!=SF_SET_PENDING
-            && dyn->insts[ninst].f_exit.pending!=SF_PENDING)
-                return 1;
-            else
-                return (dyn->insts[jmp].f_entry.dfnone  == dyn->insts[ninst].f_exit.dfnone)?0:1;
+            if(dyn->insts[ninst].f_exit.pending==SF_PENDING || dyn->insts[ninst].f_exit.pending==SF_SET_PENDING)
+                return 0;
+            return (dyn->insts[jmp].f_entry.dfnone  == dyn->insts[ninst].f_exit.dfnone)?0:1;
    }
-    if(dyn->insts[jmp].f_entry.dfnone && !dyn->insts[ninst].f_exit.dfnone)
-        return 1;
    return 0;
 }

@@ -445,6 +476,12 @@ uintptr_t fakeed(dynarec_native_t* dyn, uintptr_t addr, int ninst, uint8_t nexto
    }
    return addr;
 }
+// return Ib on a mod/rm opcode without emiting anything
+uint8_t geted_ib(dynarec_native_t* dyn, uintptr_t addr, int ninst, uint8_t nextop)
+{
+    addr = fakeed(dyn, addr, ninst, nextop);
+    return F8;
+}
 #undef F8

 int isNativeCall(dynarec_native_t* dyn, uintptr_t addr, uintptr_t* calladdress, int* retn)
@@ -46,6 +46,9 @@ void native_clflush(x64emu_t* emu, void* p);
 void native_ud(x64emu_t* emu);
 void native_priv(x64emu_t* emu);
 void native_singlestep(x64emu_t* emu);
+void native_int3(x64emu_t* emu);
+void native_int(x64emu_t* emu, int num);
+void native_div0(x64emu_t* emu);

 // Caches transformation (for loops) // Specific, need to be written par backend
 int CacheNeedsTransform(dynarec_native_t* dyn, int i1);
@@ -56,6 +59,8 @@ int getNominalPred(dynarec_native_t* dyn, int ninst);

 // Do the GETED, but don't emit anything...
 uintptr_t fakeed(dynarec_native_t* dyn, uintptr_t addr, int ninst, uint8_t nextop);
+// return Ib on a mod/rm opcode without emiting anything
+uint8_t geted_ib(dynarec_native_t* dyn, uintptr_t addr, int ninst, uint8_t nextop);

 // Is what pointed at addr a native call? And if yes, to what function?
 int isNativeCall(dynarec_native_t* dyn, uintptr_t addr, uintptr_t* calladdress, int* retn);
@@ -26,6 +26,12 @@
 #error No STEP defined
 #endif

+#if STEP == 0
+#ifndef PROT_READ
+#define PROT_READ 0x1
+#endif
+#endif
+
 uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int is32bits)
 {
    int ok = 1;
@@ -34,7 +40,7 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
    uintptr_t ip = addr;
    uintptr_t init_addr = addr;
    rex_t rex;
-    int rep;    // 0 none, 1=F2 prefix, 2=F3 prefix
+    int rep = 0;    // 0 none, 1=F2 prefix, 2=F3 prefix
    int need_epilog = 1;
    // Clean up (because there are multiple passes)
    dyn->f.pending = 0;
@@ -43,14 +49,37 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
    dyn->forward_to = 0;
    dyn->forward_size = 0;
    dyn->forward_ninst = 0;
+    #if STEP == 0
+    memset(&dyn->insts[ninst], 0, sizeof(instruction_native_t));
+    #endif
    fpu_reset(dyn);
    ARCH_INIT();
    int reset_n = -1;
    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+(FindElfAddress(my_context, addr)?0:1); // if block is in elf_memory, it can be extended with bligblocks==2, else it needs 3
+    int stopblock = 2+(FindElfAddress(my_context, addr)?0:1); // if block is in elf_memory, it can be extended with box64_dynarec_bigblock==2, else it needs 3
    // ok, go now
    INIT;
+    #if STEP == 0
+    uintptr_t cur_page = (addr)&~box64_pagesize;
+    #endif
    while(ok) {
+        #if STEP == 0
+        if(cur_page != ((addr)&~box64_pagesize)) {
+            cur_page = (addr)&~box64_pagesize;
+            if(!(getProtection(addr)&PROT_READ)) {
+                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
+        if((dyn->native_size >= MAXBLOCK_SIZE) || (ninst >= MAX_INSTS)) {
+            need_epilog = 1;
+            break;
+        }
+        #endif
+        fpu_propagate_stack(dyn, ninst);
        ip = addr;
        if (reset_n!=-1) {
            dyn->last_ip = 0;
@@ -73,10 +102,17 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
                }
            }
            reset_n = -1;
-        } 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 comming here
-        fpu_propagate_stack(dyn, ninst);
+        }
+        #if STEP > 0
+        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
        NEW_INST;
+        #if STEP == 0
+        if(ninst && dyn->insts[ninst-1].x64.barrier_next) {
+            BARRIER(dyn->insts[ninst-1].x64.barrier_next);
+        }
+        #endif
        if(!ninst) {
            GOTEST(x1, x2);
        }
@@ -86,19 +122,19 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
            READFLAGS(dyn->insts[ninst].x64.need_before&~X_PEND);
        }
        if(box64_dynarec_test) {
-            MESSAGE(LOG_DUMP, "TEST INIT ----\n");
+            MESSAGE(LOG_DUMP, "TEST STEP ----\n");
            fpu_reflectcache(dyn, ninst, x1, x2, x3);
-            GO_TRACE(x64test_init, 1);
+            GO_TRACE(x64test_step, 1, x5);
            fpu_unreflectcache(dyn, ninst, x1, x2, x3);
            MESSAGE(LOG_DUMP, "----------\n");
        }
 #ifdef HAVE_TRACE
        else if(my_context->dec && box64_dynarec_trace) {
-        if((trace_end == 0) 
+        if((trace_end == 0)
            || ((ip >= trace_start) && (ip < trace_end)))  {
                MESSAGE(LOG_DUMP, "TRACE ----\n");
                fpu_reflectcache(dyn, ninst, x1, x2, x3);
-                GO_TRACE(PrintTrace, 1);
+                GO_TRACE(PrintTrace, 1, x5);
                fpu_unreflectcache(dyn, ninst, x1, x2, x3);
                MESSAGE(LOG_DUMP, "----------\n");
            }
@@ -107,12 +143,13 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int

        rep = 0;
        uint8_t pk = PK(0);
-        while((pk==0xF2) || (pk==0xF3)) {
-            rep = pk-0xF1;
-            ++addr;
-            pk = PK(0);
-        }
-        while(pk==0x3E || pk==0x26) {   //Branch Taken Hint ignored, same for ES: prefix
+        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;
+            }
            ++addr;
            pk = PK(0);
        }
@@ -126,23 +163,25 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
            }

        addr = dynarec64_00(dyn, addr, ip, ninst, rex, rep, &ok, &need_epilog);
-
+        if(dyn->abort)
+            return ip;
        INST_EPILOG;

        int next = ninst+1;
        #if STEP > 0
        if(!dyn->insts[ninst].x64.has_next && dyn->insts[ninst].x64.jmp && dyn->insts[ninst].x64.jmp_insts!=-1)
            next = dyn->insts[ninst].x64.jmp_insts;
-        #endif
        if(dyn->insts[ninst].x64.has_next && dyn->insts[next].x64.barrier) {
-            if(dyn->insts[next].x64.barrier&BARRIER_FLOAT)
+            if(dyn->insts[next].x64.barrier&BARRIER_FLOAT) {
                fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+            }
            if(dyn->insts[next].x64.barrier&BARRIER_FLAGS) {
                dyn->f.pending = 0;
                dyn->f.dfnone = 0;
                dyn->last_ip = 0;
            }
        }
+        #endif
        #ifndef PROT_READ
        #define PROT_READ 1
        #endif
@@ -151,8 +190,8 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
            ok = 1;
            // we use the 1st predecessor here
            int ii = ninst+1;
-            if(ii<dyn->size && !dyn->insts[ii].pred_sz) {
-                while(ii<dyn->size && (!dyn->insts[ii].pred_sz || (dyn->insts[ii].pred_sz==1 && dyn->insts[ii].pred[0]==ii-1))) {
+            if(ii<dyn->size && !dyn->insts[ii].x64.alive) {
+                while(ii<dyn->size && !dyn->insts[ii].x64.alive) {
                    // may need to skip opcodes to advance
                    ++ninst;
                    NEW_INST;
@@ -166,7 +205,7 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
            if((dyn->insts[ii].x64.barrier&BARRIER_FULL)==BARRIER_FULL)
                reset_n = -2;    // hack to say Barrier!
            else {
-                reset_n = getNominalPred(dyn, ii);  // may get -1 if no predecessor are availble
+                reset_n = getNominalPred(dyn, ii);  // may get -1 if no predecessor are available
                if(reset_n==-1) {
                    reset_n = -2;
                    MESSAGE(LOG_DEBUG, "Warning, Reset Caches mark not found\n");
@@ -174,16 +213,25 @@ 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 && !(*(uint8_t*)addr) && !(*(uint8_t*)(addr+1))) {
+            if(box64_dynarec_dump) dynarec_log(LOG_NONE, "Stopping block at %p reason: %s\n", (void*)addr, "Next opcode is 00 00");
+            ok = 0;
+            need_epilog = 1;
+        }
        if(dyn->forward) {
-            if(dyn->forward_to == addr && !need_epilog) {
+            if(dyn->forward_to == addr && !need_epilog && ok>=0) {
                // we made it!
-                if(box64_dynarec_dump) dynarec_log(LOG_NONE, "Forward extend block for %d bytes %p -> %p\n", dyn->forward_to-dyn->forward, (void*)dyn->forward, (void*)dyn->forward_to);
+                reset_n = get_first_jump(dyn, addr);
+                if(box64_dynarec_dump) dynarec_log(LOG_NONE, "Forward extend block for %d bytes %s%p -> %p (ninst %d - %d)\n", dyn->forward_to-dyn->forward, dyn->insts[dyn->forward_ninst].x64.has_callret?"(opt. call) ":"", (void*)dyn->forward, (void*)dyn->forward_to, reset_n, ninst);
+                if(dyn->insts[dyn->forward_ninst].x64.has_callret && !dyn->insts[dyn->forward_ninst].x64.has_next)
+                    dyn->insts[dyn->forward_ninst].x64.has_next = 1;  // this block actually continue
                dyn->forward = 0;
                dyn->forward_to = 0;
                dyn->forward_size = 0;
                dyn->forward_ninst = 0;
                ok = 1; // in case it was 0
-            } else if ((dyn->forward_to < addr) || !ok) {
+            } else if ((dyn->forward_to < addr) || ok<=0) {
                // something when wrong! rollback
                if(box64_dynarec_dump) dynarec_log(LOG_NONE, "Could not forward extend block for %d bytes %p -> %p\n", dyn->forward_to-dyn->forward, (void*)dyn->forward, (void*)dyn->forward_to);
                ok = 0;
@@ -200,32 +248,39 @@ 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)<15) && is_nops(dyn, addr, next-addr))
                    /*||(((next-addr)<30) && is_instructions(dyn, addr, next-addr))*/ ))
                {
                    ok = 1;
-                    // need to find back that instruction to copy the caches, as previous version cannot be used anymore
-                    reset_n = -2;
-                    for(int ii=0; ii<ninst; ++ii)
-                        if(dyn->insts[ii].x64.jmp == next) {
-                            reset_n = ii;
-                            ii=ninst;
-                        }
-                    if(box64_dynarec_dump) dynarec_log(LOG_NONE, "Extend block %p, %p -> %p (ninst=%d, jump from %d)\n", dyn, (void*)addr, (void*)next, ninst, reset_n);
+                    if(dyn->insts[ninst].x64.has_callret && !dyn->insts[ninst].x64.has_next) {
+                        dyn->insts[ninst].x64.has_next = 1;  // this block actually continue
+                    } 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);
+                    }
+                    if(box64_dynarec_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 && (next-addr)<box64_dynarec_forward && (getProtection(next)&PROT_READ)/*box64_dynarec_bigblock>=stopblock*/) {
                    if(!((box64_dynarec_bigblock<stopblock) && !isJumpTableDefault64((void*)next))) {
-                        dyn->forward = addr;
-                        dyn->forward_to = next;
-                        dyn->forward_size = dyn->size;
-                        dyn->forward_ninst = ninst;
-                        reset_n = -2;
-                        ok = 1;
+                        if(dyn->forward) {
+                            if(next<dyn->forward_to)
+                                dyn->forward_to = next;
+                            reset_n = -2;
+                            ok = 1;
+                        } else {
+                            dyn->forward = addr;
+                            dyn->forward_to = next;
+                            dyn->forward_size = dyn->size;
+                            dyn->forward_ninst = ninst;
+                            reset_n = -2;
+                            ok = 1;
+                        }
                    }
                }
            }
        #endif
        if(ok<0)  {
-            ok = 0; need_epilog=1; 
+            ok = 0; need_epilog=1;
            #if STEP == 0
            if(ninst) {
                --ninst;
@@ -235,14 +290,27 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
                dyn->insts[ninst].x64.need_after |= X_PEND;
                ++ninst;
            }
+            if(dyn->forward) {
+                // stopping too soon
+                dyn->size = dyn->forward_size;
+                ninst = dyn->forward_ninst+1;
+                addr = dyn->forward;
+                dyn->forward = 0;
+                dyn->forward_to = 0;
+                dyn->forward_size = 0;
+                dyn->forward_ninst = 0;
+            }
            #endif
        }
+        if((ok>0) && dyn->insts[ninst].x64.has_callret)
+            reset_n = -2;
        ++ninst;
        #if STEP == 0
-        if(ok && (((box64_dynarec_bigblock<stopblock) && !isJumpTableDefault64((void*)addr)) 
+        memset(&dyn->insts[ninst], 0, sizeof(instruction_native_t));
+        if((ok>0) && (((box64_dynarec_bigblock<stopblock) && !isJumpTableDefault64((void*)addr))
            || (addr>=box64_nodynarec_start && addr<box64_nodynarec_end)))
        #else
-        if(ok && (ninst==dyn->size))
+        if((ok>0) && (ninst==dyn->size))
        #endif
        {
            #if STEP == 0
@@ -272,14 +340,14 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
            ++ninst;
            NOTEST(x3);
            fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
-            jump_to_next(dyn, addr, 0, ninst);
+            jump_to_next(dyn, addr, 0, ninst, rex.is32bits);
            ok=0; need_epilog=0;
        }
    }
    if(need_epilog) {
        NOTEST(x3);
        fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
-        jump_to_epilog(dyn, ip, 0, ninst);  // no linker here, it's an unknow instruction
+        jump_to_epilog(dyn, ip, 0, ninst);  // no linker here, it's an unknown instruction
    }
    FINI;
    MESSAGE(LOG_DUMP, "---- END OF BLOCK ---- (%d)\n", dyn->size);
@@ -8,13 +8,13 @@ void arm64_epilog(void) EXPORTDYN;
 #define native_next         arm64_next
 #define native_prolog       arm64_prolog
 #define native_epilog       arm64_epilog
-#elif defined(LA464)
-void la464_next(void) EXPORTDYN;
-void la464_prolog(x64emu_t* emu, void* addr) EXPORTDYN;
-void la464_epilog(void) EXPORTDYN;
-#define native_next         la464_next
-#define native_prolog       la464_prolog
-#define native_epilog       la464_epilog
+#elif defined(LA64)
+void la64_next(void) EXPORTDYN;
+void la64_prolog(x64emu_t* emu, void* addr) EXPORTDYN;
+void la64_epilog(void) EXPORTDYN;
+#define native_next         la64_next
+#define native_prolog       la64_prolog
+#define native_epilog       la64_epilog
 #elif defined(RV64)
 void rv64_next(void) EXPORTDYN;
 void rv64_prolog(x64emu_t* emu, void* addr) EXPORTDYN;
@@ -24,15 +24,22 @@
 #define SF_SUB      4
 #define SF_SUBSET   (SF_SUB|SF_SET)
 #define SF_SUBSET_PENDING   (SF_SUBSET|SF_PENDING)
+#define SF_DF       8
+#define SF_SET_DF   (SF_SET|SF_DF)
+#define SF_NODF     16
+#define SF_SET_NODF (SF_SET|SF_NODF)

 typedef struct instruction_x64_s {
    uintptr_t   addr;       //address of the instruction
    int32_t     size;       // size of the instruction
    uintptr_t   jmp;        // offset to jump to, even if conditionnal (0 if not), no relative offset here
    int         jmp_insts;  // instuction to jump to (-1 if out of the block)
-    uint8_t     jmp_cond;   // 1 of conditionnal jump
-    uint8_t     has_next;   // does this opcode can continue to the next?
+    uint8_t     jmp_cond:1;   // 1 of conditionnal jump
+    uint8_t     has_next:1;   // does this opcode can continue to the next?
+    uint8_t     has_callret:1;    // this instruction have an optimised call setup
+    uint8_t     alive:1;    // this opcode gets executed (0 if dead code in that block)
    uint8_t     barrier;    // next instruction is a jump point, so no optim allowed
+    uint8_t     barrier_next;   // next instruction needs a barrier
    uint8_t     state_flags;// One of SF_XXX state
    uint8_t     use_flags;  // 0 or combination of X_?F
    uint8_t     set_flags;  // 0 or combination of X_?F
@@ -0,0 +1,814 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "emu/x64run_private.h"
+#include "la64_emitter.h"
+#include "x64run.h"
+#include "x64emu.h"
+#include "box64stack.h"
+#include "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynarec_native.h"
+#include "my_cpuid.h"
+#include "emu/x87emu_private.h"
+#include "emu/x64shaext.h"
+#include "bitutils.h"
+
+#include "la64_printer.h"
+#include "dynarec_la64_private.h"
+#include "dynarec_la64_functions.h"
+#include "dynarec_la64_helper.h"
+
+uintptr_t dynarec64_0F(dynarec_la64_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
+{
+    (void)ip;
+    (void)need_epilog;
+
+    uint8_t opcode = F8;
+    uint8_t nextop, u8;
+    uint8_t gd, ed;
+    uint8_t wb1, wback, wb2, gback;
+    uint8_t eb1, eb2;
+    uint8_t gb1, gb2;
+    int32_t i32, i32_;
+    int cacheupd = 0;
+    int v0, v1;
+    int q0, q1;
+    int d0, d1;
+    int s0, s1;
+    uint64_t tmp64u;
+    int64_t j64;
+    int64_t fixedaddress, gdoffset;
+    int unscaled;
+    MAYUSE(wb2);
+    MAYUSE(gback);
+    MAYUSE(eb1);
+    MAYUSE(eb2);
+    MAYUSE(q0);
+    MAYUSE(q1);
+    MAYUSE(d0);
+    MAYUSE(d1);
+    MAYUSE(s0);
+    MAYUSE(j64);
+    MAYUSE(cacheupd);
+
+    switch (opcode) {
+        case 0x01:
+            // TODO:, /0 is SGDT. While 0F 01 D0 is XGETBV, etc...
+            nextop = F8;
+            if(MODREG) {
+                switch(nextop) {
+                    case 0xD0:
+                        INST_NAME("FAKE xgetbv");
+                        nextop = F8;
+                        addr = fakeed(dyn, addr, ninst, nextop);
+                        SETFLAGS(X_ALL, SF_SET); // Hack to set flags in "don't care" state
+                        GETIP(ip);
+                        STORE_XEMU_CALL();
+                        CALL(native_ud, -1);
+                        LOAD_XEMU_CALL();
+                        jump_to_epilog(dyn, 0, xRIP, ninst);
+                        *need_epilog = 0;
+                        *ok = 0;
+                        break;
+                    default:
+                        DEFAULT;
+                }
+            } else {
+                switch((nextop>>3)&7) {
+                    default:
+                        DEFAULT;
+                }
+            }
+            break;
+        case 0x05:
+            INST_NAME("SYSCALL");
+            NOTEST(x1);
+            SMEND();
+            GETIP(addr);
+            STORE_XEMU_CALL();
+            CALL_S(x64Syscall, -1);
+            LOAD_XEMU_CALL();
+            TABLE64(x3, addr); // expected return address
+            BNE_MARK(xRIP, x3);
+            LD_W(w1, xEmu, offsetof(x64emu_t, quit));
+            CBZ_NEXT(w1);
+            MARK;
+            LOAD_XEMU_REM();
+            jump_to_epilog(dyn, 0, xRIP, ninst);
+            break;
+        case 0x10:
+            INST_NAME("MOVUPS Gx,Ex");
+            nextop = F8;
+            GETG;
+            if (MODREG) {
+                ed = (nextop & 7) + (rex.b << 3);
+                v1 = sse_get_reg(dyn, ninst, x1, ed, 0);
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                VOR_V(v0, v1, v1);
+            } else {
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x3, &fixedaddress, rex, NULL, 1, 0);
+                VLD(v0, ed, fixedaddress);
+            }
+            break;
+        case 0x11:
+            INST_NAME("MOVUPS 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);
+                VOR_V(v1, v0, v0);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x3, &fixedaddress, rex, NULL, 1, 0);
+                VST(v0, ed, fixedaddress);
+                SMWRITE2();
+            }
+            break;
+        case 0x12:
+            nextop = F8;
+            if (MODREG) {
+                INST_NAME("MOVHLPS Gx,Ex");
+                GETGX(v0, 1);
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop & 7) + (rex.b << 3), 0);
+                VEXTRINS_D(v0, v1, 1);
+            } else {
+                INST_NAME("MOVLPS Gx,Ex");
+                GETGX(v0, 1);
+                SMREAD();
+                GETEX(q1, 0, 0);
+                VEXTRINS_D(v0, q1, 0);
+            }
+            break;
+        case 0x13:
+            nextop = F8;
+            INST_NAME("MOVLPS Ex,Gx");
+            GETGX(v0, 0);
+            if(MODREG) {
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop & 7) + (rex.b << 3), 1);
+                VEXTRINS_D(v1, v0, 0);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x3, &fixedaddress, rex, NULL, 1, 0);
+                FST_D(v0, ed, fixedaddress);
+                SMWRITE2();
+            }
+            break;
+        case 0x14:
+            INST_NAME("UNPCKLPS Gx, Ex");
+            nextop = F8;
+            SMREAD();
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
+            VILVL_W(v0, q0, v0);
+            break;
+        case 0x15:
+            INST_NAME("UNPCKHPS Gx, Ex");
+            nextop = F8;
+            SMREAD();
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
+            VILVH_W(v0, q0, v0);
+            break;
+        case 0x16:
+            nextop = F8;
+            if (MODREG) {
+                INST_NAME("MOVLHPS Gx,Ex");
+                GETGX(v0, 1);
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop & 7) + (rex.b << 3), 0);
+            } else {
+                INST_NAME("MOVHPS Gx,Ex");
+                SMREAD();
+                GETGX(v0, 1);
+                v1 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x3, &fixedaddress, rex, NULL, 1, 0);
+                FLD_D(v1, ed, fixedaddress);
+            }
+            VILVL_D(v0, v1, v0); // v0[127:64] = v1[63:0]
+            break;
+        case 0x17:
+            nextop = F8;
+            INST_NAME("MOVHPS Ex,Gx");
+            GETGX(v0, 0);
+            if(MODREG) {
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop & 7) + (rex.b << 3), 1);
+                VEXTRINS_D(v1, v0, 0x01);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x3, &fixedaddress, rex, NULL, 1, 0);
+                v1 = fpu_get_scratch(dyn);
+                VEXTRINS_D(v1, v0, 0x01);
+                FST_D(v1, ed, fixedaddress);
+                SMWRITE2();
+            }
+            break;
+        case 0x18:
+            nextop = F8;
+            if ((nextop & 0xC0) == 0xC0) {
+                INST_NAME("NOP (multibyte)");
+            } else
+                switch ((nextop >> 3) & 7) {
+                    case 0:
+                    case 1:
+                    case 2:
+                    case 3:
+                        INST_NAME("PREFETCHh Ed");
+                        FAKEED;
+                        break;
+                    default:
+                        INST_NAME("NOP (multibyte)");
+                        FAKEED;
+                }
+            break;
+        case 0x1F:
+            INST_NAME("NOP (multibyte)");
+            nextop = F8;
+            FAKEED;
+            break;
+        case 0x28:
+            INST_NAME("MOVAPS Gx,Ex");
+            nextop = F8;
+            GETG;
+            if (MODREG) {
+                ed = (nextop & 7) + (rex.b << 3);
+                v1 = sse_get_reg(dyn, ninst, x1, ed, 0);
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                VOR_V(v0, v1, v1);
+            } else {
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x3, &fixedaddress, rex, NULL, 1, 0);
+                VLD(v0, ed, fixedaddress);
+            }
+            break;
+        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);
+                VOR_V(v1, v0, v0);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x3, &fixedaddress, rex, NULL, 1, 0);
+                VST(v0, ed, fixedaddress);
+                SMWRITE2();
+            }
+            break;
+        case 0x2E:
+            // no special check...
+        case 0x2F:
+            if (opcode == 0x2F) {
+                INST_NAME("COMISS Gx, Ex");
+            } else {
+                INST_NAME("UCOMISS Gx, Ex");
+            }
+            SETFLAGS(X_ALL, SF_SET);
+            SET_DFNONE();
+            nextop = F8;
+            GETGX(d0, 0);
+            GETEXSS(v0, 0, 0);
+            CLEAR_FLAGS(x2);
+            // if isnan(d0) || isnan(v0)
+            IFX (X_ZF | X_PF | X_CF) {
+                FCMP_S(fcc0, d0, v0, cUN);
+                BCEQZ_MARK(fcc0);
+                ORI(xFlags, xFlags, (1 << F_ZF) | (1 << F_PF) | (1 << F_CF));
+                B_MARK3_nocond;
+            }
+            MARK;
+            // else if isless(d0, v0)
+            IFX (X_CF) {
+                FCMP_S(fcc1, d0, v0, cLT);
+                BCEQZ_MARK2(fcc1);
+                ORI(xFlags, xFlags, 1 << F_CF);
+                B_MARK3_nocond;
+            }
+            MARK2;
+            // else if d0 == v0
+            IFX (X_ZF) {
+                FCMP_S(fcc2, d0, v0, cEQ);
+                BCEQZ_MARK3(fcc2);
+                ORI(xFlags, xFlags, 1 << F_ZF);
+            }
+            MARK3;
+            IFX (X_ALL) {
+                SPILL_EFLAGS();
+            }
+            break;
+        #define GO(GETFLAGS, NO, YES, F, I)                                                          \
+            READFLAGS(F);                                                                            \
+            if (la64_lbt) {                                                                          \
+                X64_SETJ(x1, I);                                                                     \
+            } else {                                                                                 \
+                GETFLAGS;                                                                            \
+            }                                                                                        \
+            nextop = F8;                                                                             \
+            GETGD;                                                                                   \
+            if (MODREG) {                                                                            \
+                ed = TO_LA64((nextop & 7) + (rex.b << 3));                                           \
+                if (la64_lbt)                                                                        \
+                    BEQZ(x1, 8);                                                                     \
+                else                                                                                 \
+                    B##NO(x1, 8);                                                                    \
+                MV(gd, ed);                                                                          \
+            } else {                                                                                 \
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x4, &fixedaddress, rex, NULL, 1, 0); \
+                if (la64_lbt)                                                                        \
+                    BEQZ(x1, 8);                                                                     \
+                else                                                                                 \
+                    B##NO(x1, 8);                                                                    \
+                LDxw(gd, ed, fixedaddress);                                                          \
+            }                                                                                        \
+            if (!rex.w) ZEROUP(gd);
+
+            GOCOND(0x40, "CMOV", "Gd, Ed");
+
+        #undef GO
+
+        case 0x51:
+            INST_NAME("SQRTPS Gx, Ex");
+            nextop = F8;
+            GETEX(q0, 0, 0);
+            GETGX_empty(v0);
+            VFSQRT_S(v0, q0);
+            break;
+        case 0x52:
+            INST_NAME("RSQRTPS Gx, Ex");
+            nextop = F8;
+            SKIPTEST(x1);
+            GETEX(q0, 0, 0);
+            GETGX_empty(q1);
+            VFRSQRT_S(q1, q0);
+            break;
+        case 0x54:
+            INST_NAME("ANDPS Gx, Ex");
+            nextop = F8;
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
+            VAND_V(v0, v0, q0);
+            break;
+        case 0x55:
+            INST_NAME("ANDNPS Gx, Ex");
+            nextop = F8;
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
+            VANDN_V(v0, v0, q0);
+            break;
+        case 0x56:
+            INST_NAME("ORPS Gx, Ex");
+            nextop = F8;
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
+            VOR_V(v0, v0, q0);
+            break;
+        case 0x57:
+            INST_NAME("XORPS Gx, Ex");
+            nextop = F8;
+            GETG;
+            if (MODREG && ((nextop & 7) + (rex.b << 3) == gd)) {
+                // special case for XORPS Gx, Gx
+                q0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                VXOR_V(q0, q0, q0);
+            } else {
+                q0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+                GETEX(q1, 0, 0);
+                VXOR_V(q0, q0, q1);
+            }
+            break;
+        case 0x58:
+            INST_NAME("ADDPS Gx, Ex");
+            nextop = F8;
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
+            VFADD_S(v0, v0, q0);
+            break;
+        case 0x59:
+            INST_NAME("MULPS Gx, Ex");
+            nextop = F8;
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
+            VFMUL_S(v0, v0, q0);
+            break;
+        case 0x5A:
+            INST_NAME("CVTPS2PD Gx, Ex");
+            nextop = F8;
+            GETEX(q0, 0, 0);
+            GETGX(q1, 1);
+            VFCVTL_D_S(q1, q0);
+            break;
+        case 0x5C:
+            INST_NAME("SUBPS Gx, Ex");
+            nextop = F8;
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
+            VFSUB_S(v0, v0, q0);
+            break;
+
+        #define GO(GETFLAGS, NO, YES, F, I)                                                         \
+            if (box64_dynarec_test == 2) { NOTEST(x1); }                                            \
+            READFLAGS(F);                                                                           \
+            i32_ = F32S;                                                                            \
+            BARRIER(BARRIER_MAYBE);                                                                 \
+            JUMP(addr + i32_, 1);                                                                   \
+            if (la64_lbt) {                                                                         \
+                X64_SETJ(x1, I);                                                                    \
+            } else {                                                                                \
+                GETFLAGS;                                                                           \
+            }                                                                                       \
+            if (dyn->insts[ninst].x64.jmp_insts == -1 || CHECK_CACHE()) {                           \
+                /* out of the block */                                                              \
+                i32 = dyn->insts[ninst].epilog - (dyn->native_size);                                \
+                if (la64_lbt)                                                                       \
+                    BEQZ_safe(x1, i32);                                                             \
+                else                                                                                \
+                    B##NO##_safe(x1, 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 + i32_, 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); \
+                    B(i32);                                                                         \
+                }                                                                                   \
+            } else {                                                                                \
+                /* inside the block */                                                              \
+                i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address - (dyn->native_size);     \
+                if (la64_lbt)                                                                       \
+                    BNEZ_safe(x1, i32);                                                             \
+                else                                                                                \
+                    B##YES##_safe(x1, i32);                                                         \
+            }
+
+            GOCOND(0x80, "J", "Id");
+
+        #undef GO
+
+
+        #define GO(GETFLAGS, NO, YES, F, I)                                                          \
+            READFLAGS(F);                                                                            \
+            if (la64_lbt) {                                                                          \
+                X64_SETJ(x3, I);                                                                     \
+            } else {                                                                                 \
+                GETFLAGS;                                                                            \
+                S##YES(x3, x1);                                                                      \
+            }                                                                                        \
+            nextop = F8;                                                                             \
+            if (MODREG) {                                                                            \
+                if (rex.rex) {                                                                       \
+                    eb1 = TO_LA64((nextop & 7) + (rex.b << 3));                                      \
+                    eb2 = 0;                                                                         \
+                } else {                                                                             \
+                    ed = (nextop & 7);                                                               \
+                    eb2 = (ed >> 2) * 8;                                                             \
+                    eb1 = TO_LA64(ed & 3);                                                           \
+                }                                                                                    \
+                BSTRINS_D(eb1, x3, eb2 + 7, eb2);                                                    \
+            } else {                                                                                 \
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, NULL, 1, 0); \
+                ST_B(x3, ed, fixedaddress);                                                          \
+                SMWRITE();                                                                           \
+            }
+
+            GOCOND(0x90, "SET", "Eb");
+        #undef GO
+
+        case 0xA2:
+            INST_NAME("CPUID");
+            NOTEST(x1);
+            MV(A1, xRAX);
+            CALL_(my_cpuid, -1, 0);
+            // BX and DX are not synchronized durring the call, so need to force the update
+            LD_D(xRDX, xEmu, offsetof(x64emu_t, regs[_DX]));
+            LD_D(xRBX, xEmu, offsetof(x64emu_t, regs[_BX]));
+            break;
+        case 0xA3:
+            INST_NAME("BT Ed, Gd");
+            SETFLAGS(X_CF, SF_SUBSET);
+            SET_DFNONE();
+            nextop = F8;
+            GETGD;
+            if (MODREG) {
+                ed = TO_LA64((nextop & 7) + (rex.b << 3));
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, x1, &fixedaddress, rex, NULL, 1, 0);
+                SRAIxw(x1, gd, 5 + rex.w);        // r1 = (gd>>5)
+                ALSL_D(x3, wback, x1, 2 + rex.w); // (&ed) += r1*4;
+                LDxw(x1, x3, fixedaddress);
+                ed = x1;
+            }
+            ANDI(x2, gd, rex.w ? 0x3f : 0x1f);
+            SRLxw(x4, ed, x2);
+            if (la64_lbt)
+                X64_SET_EFLAGS(x4, X_CF);
+            else
+                BSTRINS_D(xFlags, x4, F_CF, F_CF);
+            break;
+        case 0xAF:
+            INST_NAME("IMUL Gd, Ed");
+            SETFLAGS(X_ALL, SF_PENDING);
+            nextop = F8;
+            GETGD;
+            GETED(0);
+            if (box64_dynarec_test) {
+                // avoid noise during test
+                CLEAR_FLAGS(x3);
+            }
+            if (rex.w) {
+                // 64bits imul
+                UFLAG_IF {
+                    if (la64_lbt) {
+                        X64_MUL_D(gd, ed);
+                    }
+                    MULH_D(x3, gd, ed);
+                    MUL_D(gd, gd, ed);
+                    IFX (X_PEND) {
+                        UFLAG_OP1(x3);
+                        UFLAG_RES(gd);
+                        UFLAG_DF(x3, d_imul64);
+                    } else {
+                        SET_DFNONE();
+                    }
+                    IFXA (X_CF | X_OF, !la64_lbt) {
+                        SRAI_D(x4, gd, 63);
+                        XOR(x3, x3, x4);
+                        SNEZ(x3, x3);
+                        IFX (X_CF) {
+                            BSTRINS_D(xFlags, x3, F_CF, F_CF);
+                        }
+                        IFX (X_OF) {
+                            BSTRINS_D(xFlags, x3, F_OF, F_OF);
+                        }
+                    }
+                } else {
+                    MULxw(gd, gd, ed);
+                }
+            } else {
+                // 32bits imul
+                UFLAG_IF {
+                    if (la64_lbt) {
+                        X64_MUL_W(gd, ed);
+                    }
+                    MUL_D(gd, gd, ed);
+                    SRLI_D(x3, gd, 32);
+                    SLLI_W(gd, gd, 0);
+                    IFX (X_PEND) {
+                        UFLAG_RES(gd);
+                        UFLAG_OP1(x3);
+                        UFLAG_DF(x4, d_imul32);
+                    } else IFX (X_CF | X_OF) {
+                        SET_DFNONE();
+                    }
+                    IFXA (X_CF | X_OF, !la64_lbt) {
+                        SRAI_W(x4, gd, 31);
+                        SUB_D(x3, x3, x4);
+                        SNEZ(x3, x3);
+                        IFX (X_CF) {
+                            BSTRINS_D(xFlags, x3, F_CF, F_CF);
+                        }
+                        IFX (X_OF) {
+                            BSTRINS_D(xFlags, x3, F_OF, F_OF);
+                        }
+                    }
+                } else {
+                    MULxw(gd, gd, ed);
+                }
+                ZEROUP(gd);
+            }
+            break;
+        case 0xB6:
+            INST_NAME("MOVZX Gd, Eb");
+            nextop = F8;
+            GETGD;
+            if (MODREG) {
+                if (rex.rex) {
+                    eb1 = TO_LA64((nextop & 7) + (rex.b << 3));
+                    eb2 = 0;
+                } else {
+                    ed = (nextop & 7);
+                    eb1 = TO_LA64(ed & 3); // Ax, Cx, Dx or Bx
+                    eb2 = (ed & 4) >> 2;   // L or H
+                }
+                BSTRPICK_D(gd, eb1, eb2 * 8 + 7, eb2 * 8);
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, NULL, 1, 0);
+                LD_BU(gd, ed, fixedaddress);
+            }
+            break;
+        case 0xB7:
+            INST_NAME("MOVZX Gd, Ew");
+            nextop = F8;
+            GETGD;
+            if (MODREG) {
+                ed = TO_LA64((nextop & 7) + (rex.b << 3));
+                BSTRPICK_D(gd, ed, 15, 0);
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, NULL, 1, 0);
+                LD_HU(gd, ed, fixedaddress);
+            }
+            break;
+        case 0xBA:
+            nextop = F8;
+            switch ((nextop >> 3) & 7) {
+                case 4:
+                    INST_NAME("BT Ed, Ib");
+                    SETFLAGS(X_CF, SF_SUBSET);
+                    SET_DFNONE();
+                    GETED(1);
+                    u8 = F8;
+                    u8 &= rex.w ? 0x3f : 0x1f;
+                    BSTRPICK_D(x4, ed, u8, u8);
+                    if (la64_lbt)
+                        X64_SET_EFLAGS(x4, X_CF);
+                    else
+                        BSTRINS_D(xFlags, x4, 0, 0);
+                    break;
+                case 5:
+                    INST_NAME("BTS Ed, Ib");
+                    SETFLAGS(X_CF, SF_SUBSET);
+                    SET_DFNONE();
+                    GETED(1);
+                    u8 = F8;
+                    u8 &= (rex.w ? 0x3f : 0x1f);
+                    BSTRPICK_D(x4, ed, u8, u8);
+                    if (la64_lbt)
+                        X64_SET_EFLAGS(x4, X_CF);
+                    else
+                        BSTRINS_D(xFlags, x4, 0, 0);
+                    ADDI_D(x4, xZR, -1);
+                    BSTRINS_D(ed, x4, u8, u8);
+                    if (wback) {
+                        SDxw(ed, wback, fixedaddress);
+                        SMWRITE();
+                    }
+                    MARK;
+                    break;
+                case 6:
+                    INST_NAME("BTR Ed, Ib");
+                    SETFLAGS(X_CF, SF_SUBSET);
+                    SET_DFNONE();
+                    GETED(1);
+                    u8 = F8;
+                    u8 &= (rex.w ? 0x3f : 0x1f);
+                    BSTRPICK_D(x4, ed, u8, u8);
+                    if (la64_lbt)
+                        X64_SET_EFLAGS(x4, X_CF);
+                    else
+                        BSTRINS_D(xFlags, x4, 0, 0);
+                    BSTRINS_D(ed, xZR, u8, u8);
+                    if (wback) {
+                        SDxw(ed, wback, fixedaddress);
+                        SMWRITE();
+                    }
+                    MARK;
+                    break;
+                default:
+                    DEFAULT;
+            }
+            break;
+        case 0xBC:
+            INST_NAME("BSF Gd, Ed");
+            SETFLAGS(X_ZF, SF_SUBSET);
+            SET_DFNONE();
+            nextop = F8;
+            GETED(0);
+            GETGD;
+            if (!rex.w && MODREG) {
+                AND(x4, ed, xMASK);
+                ed = x4;
+            }
+            BNE_MARK(ed, xZR);
+            if (la64_lbt) {
+                ADDI_D(x3, xZR, 1 << F_ZF);
+                X64_SET_EFLAGS(x3, X_ZF);
+            } else {
+                ORI(xFlags, xFlags, 1 << F_ZF);
+            }
+            B_NEXT_nocond;
+            MARK;
+            // gd is undefined if ed is all zeros, don't worry.
+            if (rex.w)
+                CTZ_D(gd, ed);
+            else
+                CTZ_W(gd, ed);
+            if (la64_lbt) {
+                X64_SET_EFLAGS(xZR, X_ZF);
+            } else {
+                ADDI_D(x3, xZR, ~(1 << F_ZF));
+                AND(xFlags, xFlags, x3);
+            }
+            break;
+        case 0xBD:
+            INST_NAME("BSR Gd, Ed");
+            SETFLAGS(X_ZF, SF_SUBSET);
+            SET_DFNONE();
+            nextop = F8;
+            GETED(0);
+            GETGD;
+            if (!rex.w && MODREG) {
+                AND(x4, ed, xMASK);
+                ed = x4;
+            }
+            BNE_MARK(ed, xZR);
+            if (la64_lbt) {
+                ADDI_D(x3, xZR, 1 << F_ZF);
+                X64_SET_EFLAGS(x3, X_ZF);
+            } else {
+                ORI(xFlags, xFlags, 1 << F_ZF);
+            }
+            B_NEXT_nocond;
+            MARK;
+            if (la64_lbt) {
+                X64_SET_EFLAGS(xZR, X_ZF);
+            } else {
+                ADDI_D(x3, xZR, ~(1 << F_ZF));
+                AND(xFlags, xFlags, x3);
+            }
+            if (rex.w)
+                CLZ_D(gd, ed);
+            else
+                CLZ_W(gd, ed);
+            ADDI_D(x1, xZR, rex.w ? 63 : 31);
+            SUB_D(gd, x1, gd);
+            break;
+        case 0xBE:
+            INST_NAME("MOVSX Gd, Eb");
+            nextop = F8;
+            GETGD;
+            if (MODREG) {
+                if (rex.rex) {
+                    wback = TO_LA64((nextop & 7) + (rex.b << 3));
+                    wb2 = 0;
+                } else {
+                    wback = (nextop & 7);
+                    wb2 = (wback >> 2) * 8;
+                    wback = TO_LA64(wback & 3);
+                }
+                BSTRPICK_D(gd, wback, wb2 + 7, wb2);
+                EXT_W_B(gd, gd);
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x3, x1, &fixedaddress, rex, NULL, 1, 0);
+                LD_B(gd, ed, fixedaddress);
+            }
+            if (!rex.w) ZEROUP(gd);
+            break;
+        case 0xBF:
+            INST_NAME("MOVSX Gd, Ew");
+            nextop = F8;
+            GETGD;
+            if (MODREG) {
+                ed = TO_LA64((nextop & 7) + (rex.b << 3));
+                EXT_W_H(gd, ed);
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x3, x1, &fixedaddress, rex, NULL, 1, 0);
+                LD_H(gd, ed, fixedaddress);
+            }
+            if (!rex.w) ZEROUP(gd);
+            break;
+        case 0xC6:
+            INST_NAME("SHUFPS Gx, Ex, Ib");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(v1, 0, 1);
+            u8 = F8;
+            if (v0 == v1) {
+                VSHUF4I_W(v0, v0, u8);
+            } else {
+                q1 = fpu_get_scratch(dyn);
+                VSHUF4I_W(v0, v0, u8);
+                VSHUF4I_W(q1, v1, u8);
+                VEXTRINS_D(v0, q1, 0x11); // v0[127:64] = q1[127:64]
+            }
+            break;
+        case 0xC8:
+        case 0xC9:
+        case 0xCA:
+        case 0xCB:
+        case 0xCC:
+        case 0xCD:
+        case 0xCE:
+        case 0xCF:
+            INST_NAME("BSWAP Reg");
+            gd = TO_LA64((opcode & 7) + (rex.b << 3));
+            REVBxw(gd, gd);
+            break;
+        default:
+            DEFAULT;
+    }
+    return addr;
+}
@@ -0,0 +1,90 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "emu/x64run_private.h"
+#include "x64run.h"
+#include "x64emu.h"
+#include "box64stack.h"
+#include "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynarec_native.h"
+#include "custommem.h"
+
+#include "la64_printer.h"
+#include "dynarec_la64_private.h"
+#include "dynarec_la64_helper.h"
+#include "dynarec_la64_functions.h"
+
+uintptr_t dynarec64_64(dynarec_la64_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)rep;
+    (void)need_epilog;
+
+    uint8_t opcode = F8;
+    uint8_t nextop;
+    uint8_t u8;
+    uint8_t gd, ed, eb1, eb2, gb1, gb2;
+    uint8_t gback, wback, wb1, wb2, wb;
+    int64_t i64, j64;
+    uint64_t u64;
+    int v0, v1;
+    int q0;
+    int d0;
+    int64_t fixedaddress, gdoffset;
+    int unscaled;
+    MAYUSE(eb1);
+    MAYUSE(eb2);
+    MAYUSE(wb1);
+    MAYUSE(wb2);
+    MAYUSE(gb1);
+    MAYUSE(gb2);
+    MAYUSE(j64);
+    MAYUSE(d0);
+    MAYUSE(q0);
+    MAYUSE(v0);
+    MAYUSE(v1);
+
+    while ((opcode == 0xF2) || (opcode == 0xF3)) {
+        rep = opcode - 0xF1;
+        opcode = F8;
+    }
+
+    GETREX();
+
+    switch (opcode) {
+        case 0x33:
+            INST_NAME("XOR Gd, Seg:Ed");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            GETGD;
+            GETEDO(x4, 0);
+            emit_xor32(dyn, ninst, rex, gd, ed, x3, x4);
+            break;
+        case 0x8B:
+            INST_NAME("MOV Gd, Seg:Ed");
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            GETGD;
+            if (MODREG) { // reg <= reg
+                MVxw(gd, TO_LA64((nextop & 7) + (rex.b << 3)));
+            } else { // mem <= reg
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, NULL, 1, 0);
+                ADD_D(x4, ed, x4);
+                LDxw(gd, x4, fixedaddress);
+            }
+            break;
+        default:
+            DEFAULT;
+    }
+    return addr;
+}
@@ -0,0 +1,534 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "emu/x64run_private.h"
+#include "la64_emitter.h"
+#include "x64run.h"
+#include "x64emu.h"
+#include "box64stack.h"
+#include "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynarec_native.h"
+#include "custommem.h"
+
+#include "la64_printer.h"
+#include "dynarec_la64_private.h"
+#include "dynarec_la64_helper.h"
+#include "dynarec_la64_functions.h"
+
+
+uintptr_t dynarec64_66(dynarec_la64_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+{
+    uint8_t opcode = F8;
+    uint8_t nextop, u8;
+    int16_t i16;
+    uint16_t u16;
+    uint64_t u64;
+    int32_t i32;
+    int64_t j64;
+    uint8_t gd, ed;
+    uint8_t wback, wb1;
+    int64_t fixedaddress;
+    int unscaled;
+    int lock;
+    MAYUSE(u8);
+    MAYUSE(u16);
+    MAYUSE(u64);
+    MAYUSE(j64);
+    MAYUSE(lock);
+
+    while ((opcode == 0x2E) || (opcode == 0x36) || (opcode == 0x66)) // ignoring CS:, SS: 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
+
+    switch (opcode) {
+        case 0x01:
+            INST_NAME("ADD Ew, Gw");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETGW(x2);
+            GETEW(x1, 0);
+            emit_add16(dyn, ninst, x1, x2, x4, x5, x6);
+            EWBACK;
+            break;
+        case 0x03:
+            INST_NAME("ADD Gw, Ew");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETGW(x1);
+            GETEW(x2, 0);
+            emit_add16(dyn, ninst, x1, x2, x3, x4, x6);
+            GWBACK;
+            break;
+        case 0x05:
+            INST_NAME("ADD AX, Iw");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            i32 = F16;
+            BSTRPICK_D(x1, xRAX, 15, 0);
+            MOV32w(x2, i32);
+            emit_add16(dyn, ninst, x1, x2, x3, x4, x6);
+            BSTRINS_D(xRAX, x1, 15, 0);
+            break;
+        case 0x09:
+            INST_NAME("OR Ew, Gw");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETGW(x2);
+            GETEW(x1, 0);
+            emit_or16(dyn, ninst, x1, x2, x4, x2);
+            EWBACK;
+            break;
+        case 0x0B:
+            INST_NAME("OR Gw, Ew");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETGW(x1);
+            GETEW(x2, 0);
+            emit_or16(dyn, ninst, x1, x2, x4, x5);
+            GWBACK;
+            break;
+        case 0x0F:
+            switch (rep) {
+                case 0: addr = dynarec64_660F(dyn, addr, ip, ninst, rex, ok, need_epilog); break;
+                default:
+                    DEFAULT;
+            }
+            break;
+        case 0x19:
+            INST_NAME("SBB Ew, Gw");
+            READFLAGS(X_CF);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETGW(x2);
+            GETEW(x1, 0);
+            emit_sbb16(dyn, ninst, x1, x2, x4, x5, x6);
+            EWBACK;
+            break;
+        case 0x21:
+            INST_NAME("AND Ew, Gw");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETGW(x2);
+            GETEW(x1, 0);
+            emit_and16(dyn, ninst, x1, x2, x4, x5);
+            EWBACK;
+            break;
+        case 0x23:
+            INST_NAME("AND Gw, Ew");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETGW(x1);
+            GETEW(x2, 0);
+            emit_and16(dyn, ninst, x1, x2, x3, x4);
+            GWBACK;
+            break;
+        case 0x25:
+            INST_NAME("AND AX, Iw");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            i32 = F16;
+            BSTRPICK_D(x1, xRAX, 15, 0);
+            MOV32w(x2, i32);
+            emit_and16(dyn, ninst, x1, x2, x3, x4);
+            BSTRINS_D(xRAX, x1, 15, 0);
+            break;
+        case 0x29:
+            INST_NAME("SUB Ew, Gw");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETGW(x1);
+            GETEW(x2, 0);
+            emit_sub16(dyn, ninst, x2, x1, x4, x5, x6);
+            EWBACK;
+            break;
+        case 0x2B:
+            INST_NAME("SUB Gw, Ew");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETGW(x1);
+            GETEW(x2, 0);
+            emit_sub16(dyn, ninst, x1, x2, x3, x4, x5);
+            GWBACK;
+            break;
+        case 0x2D:
+            INST_NAME("SUB AX, Iw");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            i32 = F16;
+            BSTRPICK_D(x1, xRAX, 15, 0);
+            MOV32w(x2, i32);
+            emit_sub16(dyn, ninst, x1, x2, x3, x4, x5);
+            BSTRINS_D(xRAX, x1, 15, 0);
+            break;
+        case 0x31:
+            INST_NAME("XOR Ew, Gw");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETGW(x2);
+            GETEW(x1, 0);
+            emit_xor16(dyn, ninst, x1, x2, x4, x5, x6);
+            EWBACK;
+            break;
+        case 0x39:
+            INST_NAME("CMP Ew, Gw");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETGW(x2);
+            GETEW(x1, 0);
+            emit_cmp16(dyn, ninst, x1, x2, x3, x4, x5, x6);
+            break;
+        case 0x3B:
+            INST_NAME("CMP Gw, Ew");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETGW(x1);
+            GETEW(x2, 0);
+            emit_cmp16(dyn, ninst, x1, x2, x3, x4, x5, x6);
+            break;
+        case 0x3D:
+            INST_NAME("CMP AX, Iw");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            i32 = F16;
+            BSTRPICK_D(x1, xRAX, 15, 0);
+            if (i32) {
+                MOV32w(x2, i32);
+                emit_cmp16(dyn, ninst, x1, x2, x3, x4, x5, x6);
+            } else {
+                emit_cmp16_0(dyn, ninst, x1, x3, x4);
+            }
+            break;
+        case 0x81:
+        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);
+                    GETEW(x1, (opcode == 0x81) ? 2 : 1);
+                    if (opcode == 0x81)
+                        i16 = F16S;
+                    else
+                        i16 = F8S;
+                    MOV64x(x5, i16);
+                    emit_add16(dyn, ninst, ed, x5, x2, x4, x6);
+                    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);
+                    GETEW(x1, (opcode == 0x81) ? 2 : 1);
+                    if (opcode == 0x81)
+                        i16 = F16S;
+                    else
+                        i16 = F8S;
+                    MOV64x(x5, i16);
+                    emit_or16(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);
+                    GETEW(x1, (opcode == 0x81) ? 2 : 1);
+                    if (opcode == 0x81)
+                        i16 = F16S;
+                    else
+                        i16 = F8S;
+                    MOV64x(x5, i16);
+                    emit_and16(dyn, ninst, x1, x5, 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);
+                    GETEW(x1, (opcode == 0x81) ? 2 : 1);
+                    if (opcode == 0x81)
+                        i16 = F16S;
+                    else
+                        i16 = F8S;
+                    MOV32w(x5, i16);
+                    emit_sub16(dyn, ninst, x1, x5, x2, x4, x6);
+                    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);
+                    GETEW(x1, (opcode == 0x81) ? 2 : 1);
+                    if (opcode == 0x81)
+                        i16 = F16S;
+                    else
+                        i16 = F8S;
+                    if (i16) {
+                        MOV64x(x2, i16);
+                        emit_cmp16(dyn, ninst, x1, x2, x3, x4, x5, x6);
+                    } else
+                        emit_cmp16_0(dyn, ninst, x1, x3, x4);
+                    break;
+                default:
+                    DEFAULT;
+            }
+            break;
+        case 0x85:
+            INST_NAME("TEST Ew, Gw");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETEW(x1, 0);
+            GETGW(x2);
+            emit_test16(dyn, ninst, x1, x2, x3, x4, x5);
+            break;
+        case 0x89:
+            INST_NAME("MOV Ew, Gw");
+            nextop = F8;
+            GETGD;
+            if (MODREG) {
+                ed = TO_LA64((nextop & 7) + (rex.b << 3));
+                if (ed != gd) {
+                    BSTRINS_D(ed, gd, 15, 0);
+                }
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, &lock, 1, 0);
+                ST_H(gd, ed, fixedaddress);
+                SMWRITELOCK(lock);
+            }
+            break;
+        case 0x90:
+        case 0x91:
+        case 0x92:
+        case 0x93:
+        case 0x94:
+        case 0x95:
+        case 0x96:
+        case 0x97:
+            gd = TO_LA64((opcode & 0x07) + (rex.b << 3));
+            if (gd == xRAX) {
+                INST_NAME("NOP");
+            } else {
+                INST_NAME("XCHG AX, Reg");
+                MV(x2, xRAX);
+                BSTRINS_D(xRAX, gd, 15, 0);
+                BSTRINS_D(gd, x2, 15, 0);
+            }
+            break;
+        case 0xAB:
+            if (rep) {
+                INST_NAME("REP STOSW");
+                CBZ_NEXT(xRCX);
+                ANDI(x1, xFlags, 1 << F_DF);
+                BNEZ_MARK2(x1);
+                MARK; // Part with DF==0
+                ST_H(xRAX, xRDI, 0);
+                ADDI_D(xRDI, xRDI, 2);
+                ADDI_D(xRCX, xRCX, -1);
+                BNEZ_MARK(xRCX);
+                B_NEXT_nocond;
+                MARK2; // Part with DF==1
+                ST_H(xRAX, xRDI, 0);
+                ADDI_D(xRDI, xRDI, -2);
+                ADDI_D(xRCX, xRCX, -1);
+                BNEZ_MARK2(xRCX);
+                // done
+            } else {
+                INST_NAME("STOSW");
+                GETDIR(x3, x1, 2);
+                ST_H(xRAX, xRDI, 0);
+                ADD_D(xRDI, xRDI, x3);
+            }
+            break;
+        case 0xB8:
+        case 0xB9:
+        case 0xBA:
+        case 0xBB:
+        case 0xBC:
+        case 0xBD:
+        case 0xBE:
+        case 0xBF:
+            INST_NAME("MOV Reg16, Iw");
+            u16 = F16;
+            MOV32w(x1, u16);
+            gd = TO_LA64((opcode & 7) + (rex.b << 3));
+            BSTRINS_D(gd, x1, 15, 0);
+            break;
+        case 0xC1:
+            nextop = F8;
+            switch ((nextop >> 3) & 7) {
+                case 0:
+                    INST_NAME("ROL Ew, Ib");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    SETFLAGS(X_OF | X_CF, SF_SET_DF);
+                    GETEW(x1, 1);
+                    u8 = F8;
+                    MOV32w(x2, u8);
+                    CALL_(rol16, x1, x3);
+                    EWBACK;
+                    break;
+                case 1:
+                    INST_NAME("ROR Ew, Ib");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    SETFLAGS(X_OF | X_CF, SF_SET_DF);
+                    GETEW(x1, 1);
+                    u8 = F8;
+                    MOV32w(x2, u8);
+                    CALL_(ror16, x1, x3);
+                    EWBACK;
+                    break;
+                case 4:
+                case 6:
+                    INST_NAME("SHL Ew, Ib");
+                    UFLAG_IF { MESSAGE(LOG_DUMP, "Need Optimization for flags\n"); }
+                    SETFLAGS(X_ALL, SF_PENDING);
+                    GETEW(x1, 1);
+                    u8 = F8;
+                    UFLAG_IF { MOV32w(x2, (u8 & 15)); }
+                    UFLAG_OP12(ed, x2)
+                    if (MODREG) {
+                        SLLI_D(ed, ed, 48 + (u8 & 15));
+                        SRLI_D(ed, ed, 48);
+                    } else {
+                        SLLI_D(ed, ed, u8 & 15);
+                    }
+                    EWBACK;
+                    UFLAG_RES(ed);
+                    UFLAG_DF(x3, d_shl16);
+                    break;
+                case 5:
+                    INST_NAME("SHR Ew, Ib");
+                    UFLAG_IF { MESSAGE(LOG_DUMP, "Need Optimization for flags\n"); }
+                    SETFLAGS(X_ALL, SF_PENDING);
+                    GETEW(x1, 1);
+                    u8 = F8;
+                    UFLAG_IF { MOV32w(x2, (u8 & 15)); }
+                    UFLAG_OP12(ed, x2)
+                    SRLI_D(ed, ed, u8 & 15);
+                    EWBACK;
+                    UFLAG_RES(ed);
+                    UFLAG_DF(x3, d_shr16);
+                    break;
+                default:
+                    DEFAULT;
+            }
+            break;
+        case 0xC7:
+            INST_NAME("MOV Ew, Iw");
+            nextop = F8;
+            if (MODREG) {
+                ed = TO_LA64((nextop & 7) + (rex.b << 3));
+                u16 = F16;
+                MOV32w(x1, u16);
+                BSTRINS_D(ed, x1, 15, 0);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, &lock, 1, 2);
+                u16 = F16;
+                MOV32w(x1, u16);
+                ST_H(x1, ed, fixedaddress);
+                SMWRITELOCK(lock);
+            }
+            break;
+        case 0xD1:
+        case 0xD3:
+            nextop = F8;
+            switch ((nextop >> 3) & 7) {
+                case 5:
+                    if (opcode == 0xD1) {
+                        INST_NAME("SHR Ew, 1");
+                        MOV32w(x2, 1);
+                    } else {
+                        INST_NAME("SHR Ew, CL");
+                        ANDI(x2, xRCX, 0x1f);
+                        BEQ_NEXT(x2, xZR);
+                    }
+                    SETFLAGS(X_ALL, SF_SET_PENDING); // some flags are left undefined
+                    if (box64_dynarec_safeflags > 1) MAYSETFLAGS();
+                    GETEW(x1, 0);
+                    emit_shr16(dyn, ninst, x1, x2, x5, x4, x6);
+                    EWBACK;
+                    break;
+                case 4:
+                case 6:
+                    if (opcode == 0xD1) {
+                        INST_NAME("SHL Ew, 1");
+                        MOV32w(x2, 1);
+                    } else {
+                        INST_NAME("SHL Ew, CL");
+                        ANDI(x2, xRCX, 0x1f);
+                        BEQ_NEXT(x2, xZR);
+                    }
+                    SETFLAGS(X_ALL, SF_SET_PENDING); // some flags are left undefined
+                    if (box64_dynarec_safeflags > 1)
+                        MAYSETFLAGS();
+                    GETEW(x1, 0);
+                    emit_shl16(dyn, ninst, x1, x2, x5, x4, x6);
+                    EWBACK;
+                    break;
+                case 7:
+                    if (opcode == 0xD1) {
+                        INST_NAME("SAR Ew, 1");
+                        MOV32w(x2, 1);
+                    } else {
+                        INST_NAME("SAR Ew, CL");
+                        ANDI(x2, xRCX, 0x1f);
+                        BEQ_NEXT(x2, xZR);
+                    }
+                    SETFLAGS(X_ALL, SF_SET_PENDING); // some flags are left undefined
+                    if (box64_dynarec_safeflags > 1)
+                        MAYSETFLAGS();
+                    GETSEW(x1, 0);
+                    emit_sar16(dyn, ninst, x1, x2, x5, x4, x6);
+                    EWBACK;
+                    break;
+                default:
+                    DEFAULT;
+            }
+            break;
+        case 0xF7:
+            nextop = F8;
+            switch ((nextop >> 3) & 7) {
+                case 0:
+                case 1:
+                    INST_NAME("TEST Ew, Iw");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    GETEW(x1, 2);
+                    u16 = F16;
+                    MOV32w(x2, u16);
+                    emit_test16(dyn, ninst, x1, x2, x3, x4, x5);
+                    break;
+                default:
+                    DEFAULT;
+            }
+        default:
+            DEFAULT;
+    }
+    return addr;
+}
@@ -0,0 +1,806 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "emu/x64run_private.h"
+#include "la64_emitter.h"
+#include "x64run.h"
+#include "x64emu.h"
+#include "box64stack.h"
+#include "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynarec_native.h"
+#include "bitutils.h"
+
+#include "la64_printer.h"
+#include "dynarec_la64_private.h"
+#include "dynarec_la64_functions.h"
+#include "dynarec_la64_helper.h"
+
+uintptr_t dynarec64_660F(dynarec_la64_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
+{
+    (void)ip;
+    (void)need_epilog;
+
+    uint8_t opcode = F8;
+    uint8_t nextop, u8, s8;
+    int32_t i32;
+    uint8_t gd, ed;
+    uint8_t wback, wb1, wb2, gback;
+    uint8_t eb1, eb2;
+    int64_t j64;
+    uint64_t tmp64u, tmp64u2;
+    int v0, v1;
+    int q0, q1;
+    int d0, d1, d2;
+    int64_t fixedaddress, gdoffset;
+    int unscaled;
+
+    MAYUSE(d0);
+    MAYUSE(d1);
+    MAYUSE(q0);
+    MAYUSE(q1);
+    MAYUSE(eb1);
+    MAYUSE(eb2);
+    MAYUSE(j64);
+
+    switch (opcode) {
+        case 0x14:
+            INST_NAME("UNPCKLPD Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            if (MODREG) {
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop & 7) + (rex.b << 3), 0);
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x3, &fixedaddress, rex, NULL, 1, 0);
+                v1 = fpu_get_scratch(dyn);
+                FLD_D(v1, ed, fixedaddress);
+            }
+            VEXTRINS_D(v0, v1, 0x10);
+            break;
+        case 0x1F:
+            INST_NAME("NOP (multibyte)");
+            nextop = F8;
+            FAKEED;
+            break;
+        case 0x28:
+            INST_NAME("MOVAPD Gx,Ex");
+            nextop = F8;
+            GETG;
+            if (MODREG) {
+                ed = (nextop & 7) + (rex.b << 3);
+                v1 = sse_get_reg(dyn, ninst, x1, ed, 0);
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                VOR_V(v0, v1, v1);
+            } else {
+                SMREAD();
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x3, &fixedaddress, rex, NULL, 1, 0);
+                VLD(v0, ed, fixedaddress);
+            }
+            break;
+        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);
+            SET_DFNONE();
+            nextop = F8;
+            GETGX(d0, 0);
+            GETEXSD(v0, 0, 0);
+
+            CLEAR_FLAGS(x3);
+            // if isnan(d0) || isnan(v0)
+            IFX (X_ZF | X_PF | X_CF) {
+                FCMP_D(fcc0, d0, v0, cUN);
+                BCEQZ_MARK(fcc0);
+                ORI(xFlags, xFlags, (1 << F_ZF) | (1 << F_PF) | (1 << F_CF));
+                B_MARK3_nocond;
+            }
+            MARK;
+            // else if isless(d0, v0)
+            IFX (X_CF) {
+                FCMP_D(fcc1, d0, v0, cLT);
+                BCEQZ_MARK2(fcc1);
+                ORI(xFlags, xFlags, 1 << F_CF);
+                B_MARK3_nocond;
+            }
+            MARK2;
+            // else if d0 == v0
+            IFX (X_ZF) {
+                FCMP_D(fcc2, d0, v0, cEQ);
+                BCEQZ_MARK3(fcc2);
+                ORI(xFlags, xFlags, 1 << F_ZF);
+            }
+            MARK3;
+            SPILL_EFLAGS();
+            break;
+        case 0x38: // SSSE3 opcodes
+            nextop = F8;
+            switch (nextop) {
+                case 0x00:
+                    INST_NAME("PSHUFB Gx, Ex");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    GETEX(q1, 0, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    v1 = fpu_get_scratch(dyn);
+                    VLDI(v0, 0b0000010001111); // broadcast 0b10001111 as byte
+                    VAND_V(v0, v0, q1);
+                    VMINI_BU(v0, v0, 0x1f);
+                    VXOR_V(v1, v1, v1);
+                    VSHUF_B(q0, v1, q0, v0);
+                    break;
+                case 0x01:
+                    INST_NAME("PHADDW Gx, Ex");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    GETEX(q1, 0, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    v1 = fpu_get_scratch(dyn);
+                    VPICKEV_H(v0, q1, q0);
+                    VPICKOD_H(v1, q1, q0);
+                    VADD_H(q0, v0, v1);
+                    break;
+                case 0x02:
+                    INST_NAME("PHADDD Gx, Ex");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    GETEX(q1, 0, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    v1 = fpu_get_scratch(dyn);
+                    VPICKEV_W(v0, q1, q0);
+                    VPICKOD_W(v1, q1, q0);
+                    VADD_W(q0, v0, v1);
+                    break;
+                case 0x03:
+                    INST_NAME("PHADDSW Gx, Ex");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    GETEX(q1, 0, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    v1 = fpu_get_scratch(dyn);
+                    VPICKEV_H(v0, q1, q0);
+                    VPICKOD_H(v1, q1, q0);
+                    VSADD_H(q0, v0, v1);
+                    break;
+                case 0x04:
+                    INST_NAME("PMADDUBSW Gx,Ex");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    GETEX(q1, 0, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    v1 = fpu_get_scratch(dyn);
+                    VEXT2XV_HU_BU(v0, q0);
+                    VEXT2XV_H_B(v1, q1);
+                    XVMUL_H(v0, v0, v1);
+                    XVPERMI_Q(v1, v0, 1); // v1[127:0] = v0[255:128];
+                    VPICKEV_H(q0, v1, v0);
+                    VPICKOD_H(v0, v1, v0);
+                    VSADD_H(q0, v0, q0);
+                    break;
+                case 0x05:
+                    INST_NAME("PHSUBW Gx, Ex");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    GETEX(q1, 0, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    v1 = fpu_get_scratch(dyn);
+                    VPICKEV_H(v0, q1, q0);
+                    VPICKOD_H(v1, q1, q0);
+                    VSUB_H(q0, v0, v1);
+                    break;
+                case 0x06:
+                    INST_NAME("PHSUBD Gx, Ex");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    GETEX(q1, 0, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    v1 = fpu_get_scratch(dyn);
+                    VPICKEV_W(v0, q1, q0);
+                    VPICKOD_W(v1, q1, q0);
+                    VSUB_D(q0, v0, v1);
+                    break;
+                case 0x07:
+                    INST_NAME("PHSUBSW Gx, Ex");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    GETEX(q1, 0, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    v1 = fpu_get_scratch(dyn);
+                    VPICKEV_H(v0, q1, q0);
+                    VPICKOD_H(v1, q1, q0);
+                    VSSUB_H(q0, v0, v1);
+                    break;
+                case 0x08:
+                    INST_NAME("PSIGNB Gx, Ex");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    GETEX(q1, 0, 0);
+                    VSIGNCOV_B(q0, q1, q0);
+                    break;
+                case 0x09:
+                    INST_NAME("PSIGNW Gx, Ex");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    GETEX(q1, 0, 0);
+                    VSIGNCOV_H(q0, q1, q0);
+                    break;
+                case 0x0A:
+                    INST_NAME("PSIGND Gx, Ex");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    GETEX(q1, 0, 0);
+                    VSIGNCOV_W(q0, q1, q0);
+                    break;
+                default:
+                    DEFAULT;
+            }
+            break;
+        case 0x54:
+            INST_NAME("ANDPD Gx, Ex");
+            nextop = F8;
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
+            VAND_V(v0, v0, q0);
+            break;
+        case 0x55:
+            INST_NAME("ANDNPD Gx, Ex");
+            nextop = F8;
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
+            VANDN_V(v0, v0, q0);
+            break;
+        case 0x56:
+            INST_NAME("ORPD Gx, Ex");
+            nextop = F8;
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
+            VOR_V(v0, v0, q0);
+            break;
+        case 0x57:
+            INST_NAME("XORPD Gx, Ex");
+            nextop = F8;
+            GETG;
+            if (MODREG && ((nextop & 7) + (rex.b << 3) == gd)) {
+                // special case for XORPD Gx, Gx
+                q0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                VXOR_V(q0, q0, q0);
+            } else {
+                q0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+                GETEX(q1, 0, 0);
+                VXOR_V(q0, q0, q1);
+            }
+            break;
+        case 0x5A:
+            INST_NAME("CVTPD2PS Gx, Ex");
+            nextop = F8;
+            GETEX(v1, 0, 0);
+            GETGX_empty(v0);
+            // TODO: is there any way to support !box64_dynarec_fastround?
+            q0 = fpu_get_scratch(dyn);
+            VFCVT_S_D(q0, v1, v1);
+            VXOR_V(v0, v0, v0);
+            VEXTRINS_D(v0, q0, 0);
+            break;
+        case 0x60:
+            INST_NAME("PUNPCKLBW Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(q0, 0, 0);
+            VILVL_B(v0, q0, v0);
+            break;
+        case 0x61:
+            INST_NAME("PUNPCKLWD Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(q0, 0, 0);
+            VILVL_H(v0, q0, v0);
+            break;
+        case 0x62:
+            INST_NAME("PUNPCKLDQ Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(q0, 0, 0);
+            VILVL_W(v0, q0, v0);
+            break;
+        case 0x64:
+            INST_NAME("PCMPGTB Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(v1, 0, 0);
+            VSLT_B(v0, v1, v0);
+            break;
+        case 0x65:
+            INST_NAME("PCMPGTW Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(v1, 0, 0);
+            VSLT_H(v0, v1, v0);
+            break;
+        case 0x66:
+            INST_NAME("PCMPGTD Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(v1, 0, 0);
+            VSLT_W(v0, v1, v0);
+            break;
+        case 0x67:
+            INST_NAME("PACKUSWB Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(v1, 0, 0);
+            q0 = fpu_get_scratch(dyn);
+            VLDI(q0, 0b0010011111111); // broadcast 0xff as 16-bit elements to all lanes
+            if (v0 == v1) {
+                VMAXI_H(v0, v0, 0);
+                VMIN_H(v0, v0, q0);
+                VPICKEV_B(v0, v0, v0);
+            } else {
+                q1 = fpu_get_scratch(dyn);
+                VMAXI_H(v0, v0, 0);
+                VMAXI_H(q1, v1, 0);
+                VMIN_H(v0, v0, q0);
+                VMIN_H(q1, q1, q0);
+                VPICKEV_B(v0, q1, v0);
+            }
+            break;
+        case 0x68:
+            INST_NAME("PUNPCKHBW Gx,Ex");
+            nextop = F8;
+            GETGX(q0, 1);
+            GETEX(q1, 0, 0);
+            VILVH_B(q0, q1, q0);
+            break;
+        case 0x69:
+            INST_NAME("PUNPCKHWD Gx,Ex");
+            nextop = F8;
+            GETGX(q0, 1);
+            GETEX(q1, 0, 0);
+            VILVH_H(q0, q1, q0);
+            break;
+        case 0x6A:
+            INST_NAME("PUNPCKHDQ Gx,Ex");
+            nextop = F8;
+            GETGX(q0, 1);
+            GETEX(q1, 0, 0);
+            VILVH_W(q0, q1, q0);
+            break;
+        case 0x6C:
+            INST_NAME("PUNPCKLQDQ Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            if (MODREG) {
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop & 7) + (rex.b << 3), 0);
+            } else {
+                v1 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x3, &fixedaddress, rex, NULL, 1, 0);
+                FLD_D(v1, ed, fixedaddress);
+            }
+            VILVL_D(v0, v1, v0); // v0[127:64] = v1[63:0]
+            break;
+        case 0x6D:
+            INST_NAME("PUNPCKHQDQ Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            if (MODREG) {
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop & 7) + (rex.b << 3), 0);
+            } else {
+                v1 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x3, &fixedaddress, rex, NULL, 8, 0);
+                FLD_D(v1, ed, fixedaddress + 8);
+                VEXTRINS_D(v1, v1, 0x10);
+            }
+            VILVH_D(v0, v1, v0);
+            break;
+        case 0x6E:
+            INST_NAME("MOVD Gx, Ed");
+            nextop = F8;
+            GETGX_empty(v0);
+            v1 = fpu_get_scratch(dyn);
+            if (MODREG) {
+                ed = TO_LA64((nextop & 7) + (rex.b << 3));
+                if (rex.w) {
+                    MOVGR2FR_D(v1, ed);
+                } else {
+                    MOVGR2FR_W(v1, ed);
+                }
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, NULL, 1, 0);
+                FLDxw(v1, ed, fixedaddress);
+            }
+            VXOR_V(v0, v0, v0);
+            if (rex.w) {
+                VEXTRINS_D(v0, v1, 0); // v0[63:0] = v1[63:0]
+            } else {
+                VEXTRINS_W(v0, v1, 0); // v0[31:0] = v1[31:0]
+            }
+            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);
+                VOR_V(v0, v1, v1);
+            } else {
+                GETGX_empty(v0);
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, NULL, 1, 0);
+                VLD(v0, ed, fixedaddress);
+            }
+            break;
+        case 0x70:
+            INST_NAME("PSHUFD Gx,Ex,Ib");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(v1, 0, 1);
+            u8 = F8;
+            VSHUF4I_W(v0, v1, u8);
+            break;
+        case 0x71:
+            nextop = F8;
+            switch ((nextop >> 3) & 7) {
+                case 2:
+                    INST_NAME("PSRLW Ex, Ib");
+                    GETEX(q0, 1, 1);
+                    u8 = F8;
+                    if (u8) {
+                        if (u8 > 15) {
+                            VXOR_V(q0, q0, q0);
+                        } else if (u8) {
+                            VSRLI_H(q0, q0, u8);
+                        }
+                        PUTEX(q0);
+                    }
+                    break;
+                case 4:
+                    INST_NAME("PSRAW Ex, Ib");
+                    GETEX(q0, 1, 1);
+                    u8 = F8;
+                    if (u8 > 15) u8 = 15;
+                    if (u8) {
+                        VSRAI_H(q0, q0, u8);
+                    }
+                    PUTEX(q0);
+                    break;
+                case 6:
+                    INST_NAME("PSLLW Ex, Ib");
+                    GETEX(q0, 1, 1);
+                    u8 = F8;
+                    if (u8) {
+                        if (u8 > 15) {
+                            VXOR_V(q0, q0, q0);
+                        } else {
+                            VSLLI_H(q0, q0, u8);
+                        }
+                        PUTEX(q0);
+                    }
+                    break;
+                default:
+                    *ok = 0;
+                    DEFAULT;
+            }
+            break;
+        case 0x72:
+            nextop = F8;
+            switch ((nextop >> 3) & 7) {
+                case 2:
+                    INST_NAME("PSRLD Ex, Ib");
+                    GETEX(q0, 1, 1);
+                    u8 = F8;
+                    if (u8) {
+                        if (u8 > 31) {
+                            VXOR_V(q0, q0, q0);
+                        } else if (u8) {
+                            VSRLI_W(q0, q0, u8);
+                        }
+                        PUTEX(q0);
+                    }
+                    break;
+                case 4:
+                    INST_NAME("PSRAD Ex, Ib");
+                    GETEX(q0, 1, 1);
+                    u8 = F8;
+                    if (u8 > 31) u8 = 31;
+                    if (u8) {
+                        VSRAI_W(q0, q0, u8);
+                    }
+                    PUTEX(q0);
+                    break;
+                case 6:
+                    INST_NAME("PSLLD Ex, Ib");
+                    GETEX(q0, 1, 1);
+                    u8 = F8;
+                    if (u8) {
+                        if (u8 > 31) {
+                            VXOR_V(q0, q0, q0);
+                        } else {
+                            VSLLI_W(q0, q0, u8);
+                        }
+                        PUTEX(q0);
+                    }
+                    break;
+                default:
+                    DEFAULT;
+            }
+            break;
+        case 0x73:
+            nextop = F8;
+            switch ((nextop >> 3) & 7) {
+                case 2:
+                    INST_NAME("PSRLQ Ex, Ib");
+                    GETEX(q0, 1, 1);
+                    u8 = F8;
+                    if (u8) {
+                        if (u8 > 63) {
+                            VXOR_V(q0, q0, q0);
+                        } else {
+                            VSRLI_D(q0, q0, u8);
+                        }
+                        PUTEX(q0);
+                    }
+                    break;
+                case 3:
+                    INST_NAME("PSRLDQ Ex, Ib");
+                    GETEX(q0, 1, 1);
+                    u8 = F8;
+                    if (u8) {
+                        if (u8 > 15) {
+                            XVOR_V(q0, q0, q0);
+                        } else {
+                            VBSRL_V(q0, q0, u8);
+                        }
+                        PUTEX(q0);
+                    }
+                    break;
+                case 6:
+                    INST_NAME("PSLLQ Ex, Ib");
+                    GETEX(q0, 1, 1);
+                    u8 = F8;
+                    if (u8) {
+                        if (u8 > 63) {
+                            XVOR_V(q0, q0, q0);
+                        } else {
+                            VSLLI_D(q0, q0, u8);
+                        }
+                        PUTEX(q0);
+                    }
+                    break;
+                case 7:
+                    INST_NAME("PSLLDQ Ex, Ib");
+                    GETEX(q0, 1, 1);
+                    u8 = F8;
+                    if (u8) {
+                        if (u8 > 15) {
+                            VXOR_V(q0, q0, q0);
+                        } else if (u8 > 0) {
+                            VBSLL_V(q0, q0, u8);
+                        }
+                        PUTEX(q0);
+                    }
+                    break;
+                default:
+                    DEFAULT;
+            }
+            break;
+        case 0x76:
+            INST_NAME("PCMPEQD Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(q0, 0, 0);
+            VSEQ_W(v0, v0, q0);
+            break;
+        case 0x7E:
+            INST_NAME("MOVD Ed,Gx");
+            nextop = F8;
+            GETGX(v0, 0);
+            if (rex.w) {
+                if (MODREG) {
+                    ed = TO_LA64((nextop & 7) + (rex.b << 3));
+                    MOVFR2GR_D(ed, v0);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, x3, &fixedaddress, rex, NULL, 1, 0);
+                    FST_D(v0, ed, fixedaddress);
+                    SMWRITE2();
+                }
+            } else {
+                if (MODREG) {
+                    ed = TO_LA64((nextop & 7) + (rex.b << 3));
+                    MOVFR2GR_S(ed, v0);
+                    ZEROUP(ed);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, x3, &fixedaddress, rex, NULL, 1, 0);
+                    FST_S(v0, ed, fixedaddress);
+                    SMWRITE2();
+                }
+            }
+            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);
+                VOR_V(v1, v0, v0);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x3, &fixedaddress, rex, NULL, 1, 0);
+                VST(v0, ed, fixedaddress);
+                SMWRITE2();
+            }
+            break;
+        case 0xAF:
+            INST_NAME("IMUL Gw,Ew");
+            SETFLAGS(X_ALL, SF_PENDING);
+            nextop = F8;
+            UFLAG_DF(x1, d_imul16);
+            GETSEW(x1, 0);
+            GETSGW(x2);
+            MUL_W(x2, x2, x1);
+            UFLAG_RES(x2);
+            BSTRPICK_D(x2, x2, 15, 0);
+            GWBACK;
+            break;
+        case 0xBE:
+            INST_NAME("MOVSX Gw, Eb");
+            nextop = F8;
+            GETGD;
+            if (MODREG) {
+                if (rex.rex) {
+                    ed = TO_LA64((nextop & 7) + (rex.b << 3));
+                    eb1 = ed;
+                    eb2 = 0;
+                } else {
+                    ed = (nextop & 7);
+                    eb1 = TO_LA64(ed & 3); // Ax, Cx, Dx or Bx
+                    eb2 = (ed & 4) >> 2;   // L or H
+                }
+                if (eb2) {
+                    SRLI_D(x1, eb1, eb2 * 8);
+                    EXT_W_B(x1, x1);
+                } else {
+                    EXT_W_B(x1, eb1);
+                }
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x4, &fixedaddress, rex, NULL, 1, 0);
+                LD_B(x1, ed, fixedaddress);
+            }
+            BSTRINS_D(gd, x1, 15, 0);
+            break;
+        case 0xD4:
+            INST_NAME("PADDQ Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(q0, 0, 0);
+            VADD_D(v0, v0, q0);
+            break;
+        case 0xD5:
+            INST_NAME("PMULLW Gx,Ex");
+            nextop = F8;
+            GETGX(q0, 1);
+            GETEX(q1, 0, 0);
+            VMUL_H(q0, q0, q1);
+            break;
+        case 0xD6:
+            INST_NAME("MOVQ Ex, Gx");
+            nextop = F8;
+            GETGX(v0, 0);
+            if (MODREG) {
+                v1 = sse_get_reg_empty(dyn, ninst, x1, (nextop & 7) + (rex.b << 3));
+                VXOR_V(v1, v1, v1);
+                VEXTRINS_D(v1, v0, 0);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x3, &fixedaddress, rex, NULL, 1, 0);
+                FST_D(v0, ed, fixedaddress);
+                SMWRITE2();
+            }
+            break;
+        case 0xDB:
+            INST_NAME("PAND Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(q0, 0, 0);
+            VAND_V(v0, v0, q0);
+            break;
+        case 0xDF:
+            INST_NAME("PANDN Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(q0, 0, 0);
+            VANDN_V(v0, v0, q0);
+            break;
+        case 0xE4:
+            INST_NAME("PMULHUW Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(v1, 0, 0);
+            VMUH_HU(v0, v0, v1);
+            break;
+        case 0xEB:
+            INST_NAME("POR Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(q0, 0, 0);
+            VOR_V(v0, v0, q0);
+            break;
+        case 0xEF:
+            INST_NAME("PXOR Gx,Ex");
+            nextop = F8;
+            GETG;
+            if (MODREG && ((nextop & 7) + (rex.b << 3) == gd)) {
+                // special case for PXOR Gx, Gx
+                q0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                VXOR_V(q0, q0, q0);
+            } else {
+                q0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+                GETEX(q1, 0, 0);
+                VXOR_V(q0, q0, q1);
+            }
+            break;
+        case 0xF8:
+            INST_NAME("PSUBB Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(q0, 0, 0);
+            VSUB_B(v0, v0, q0);
+            break;
+        case 0xF9:
+            INST_NAME("PSUBW Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(q0, 0, 0);
+            VSUB_H(v0, v0, q0);
+            break;
+        case 0xFA:
+            INST_NAME("PSUBD Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(q0, 0, 0);
+            VSUB_W(v0, v0, q0);
+            break;
+        case 0xFB:
+            INST_NAME("PSUBQ Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(q0, 0, 0);
+            VSUB_D(v0, v0, q0);
+            break;
+        case 0xFC:
+            INST_NAME("PADDB Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(q0, 0, 0);
+            VADD_B(v0, v0, q0);
+            break;
+        case 0xFD:
+            INST_NAME("PADDW Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(q0, 0, 0);
+            VADD_H(v0, v0, q0);
+            break;
+        case 0xFE:
+            INST_NAME("PADDD Gx,Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEX(q0, 0, 0);
+            VADD_W(v0, v0, q0);
+            break;
+        default:
+            DEFAULT;
+    }
+    return addr;
+}
@@ -0,0 +1,631 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "emu/x64run_private.h"
+#include "la64_emitter.h"
+#include "x64run.h"
+#include "x64emu.h"
+#include "box64stack.h"
+#include "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynarec_native.h"
+
+#include "la64_printer.h"
+#include "dynarec_la64_private.h"
+#include "dynarec_la64_functions.h"
+#include "dynarec_la64_helper.h"
+
+// emit XOR8 instruction, from s1, s2, store result in s1 using s3 and s4 as scratch, s4 can be same as s2 (and so s2 destroyed)
+void emit_xor8(dynarec_la64_t* dyn, int ninst, int s1, int s2, int s3, int s4)
+{
+    IFX (X_PEND) {
+        SET_DF(s4, d_xor8);
+    } else IFX (X_ALL) {
+        SET_DFNONE();
+    }
+
+    IFXA (X_ALL, la64_lbt) {
+        X64_XOR_B(s1, s2);
+    }
+
+    XOR(s1, s1, s2);
+    ANDI(s1, s1, 0xff);
+
+    IFX (X_PEND) {
+        ST_B(s1, xEmu, offsetof(x64emu_t, res));
+    }
+
+    if (la64_lbt) return;
+
+    CLEAR_FLAGS(s3);
+    IFX (X_SF) {
+        SRLI_D(s3, s1, 7);
+        BEQZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    IFX (X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX (X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit XOR8 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_xor8c(dynarec_la64_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
+{
+    IFX (X_PEND) {
+        SET_DF(s4, d_xor8);
+    } else IFX (X_ALL) {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX (X_ALL) {
+            ADDI_D(s3, xZR, c & 0xff);
+            X64_XOR_B(s1, s3);
+        }
+        XORI(s1, s1, c & 0xff);
+        IFX (X_PEND)
+            ST_B(s1, xEmu, offsetof(x64emu_t, res));
+        return;
+    }
+
+    XORI(s1, s1, c & 0xff);
+    ANDI(s1, s1, 0xff);
+    CLEAR_FLAGS(s3);
+    IFX (X_SF) {
+        SRLI_D(s3, s1, 7);
+        BEQZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    IFX (X_PEND) {
+        ST_B(s1, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX (X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX (X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit XOR16 instruction, from s1, s2, store result in s1 using s3 and s4 as scratch, s4 can be same as s2 (and so s2 destroyed)
+void emit_xor16(dynarec_la64_t* dyn, int ninst, int s1, int s2, int s3, int s4, int s5)
+{
+    IFX (X_PEND) {
+        SET_DF(s4, d_xor16);
+    } else IFX (X_ALL) {
+        SET_DFNONE();
+    }
+
+
+    IFXA (X_ALL, la64_lbt) {
+        X64_XOR_W(s1, s2);
+    }
+
+    XOR(s1, s1, s2);
+    BSTRINS_D(s1, s1, 15, 0);
+
+    IFX (X_PEND) {
+        ST_H(s1, xEmu, offsetof(x64emu_t, res));
+    }
+
+    if (la64_lbt) return;
+
+    CLEAR_FLAGS(s3);
+    IFX (X_ZF | X_SF) {
+        IFX (X_ZF) {
+            BNEZ(s1, 8);
+            ORI(xFlags, xFlags, 1 << F_ZF);
+        }
+        IFX (X_SF) {
+            SRLI_D(s3, s1, 15);
+            BEQZ(s3, 8);
+            ORI(xFlags, xFlags, 1 << F_SF);
+        }
+    }
+    IFX (X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit XOR32 instruction, from s1, s2, store result in s1 using s3 and s4 as scratch
+void emit_xor32(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4)
+{
+    IFX(X_PEND) {
+        SET_DF(s4, rex.w ? d_xor64 : d_xor32);
+    } else IFX(X_ALL) {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX(X_ALL) {
+            if (rex.w)
+                X64_XOR_D(s1, s2);
+            else
+                X64_XOR_W(s1, s2);
+        }
+        XOR(s1, s1, s2);
+        if (!rex.w && s1 != s2) ZEROUP(s1);
+
+        IFX(X_PEND)
+            SDxw(s1, xEmu, offsetof(x64emu_t, res));
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    XOR(s1, s1, s2);
+
+    // test sign bit before zeroup.
+    IFX(X_SF) {
+        if (!rex.w) SEXT_W(s1, s1);
+        BGE(s1, xZR, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    if (!rex.w && s1 != s2) {
+        ZEROUP(s1);
+    }
+
+    IFX(X_PEND) {
+        SDxw(s1, xEmu, offsetof(x64emu_t, res));
+    }
+
+    IFX(X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+
+// emit XOR32 instruction, from s1, c, store result in s1 using s3 and s4 as scratch
+void emit_xor32c(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, int64_t c, int s3, int s4)
+{
+    IFX (X_PEND) {
+        SET_DF(s4, rex.w ? d_xor64 : d_xor32);
+    } else IFX (X_ALL) {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX (X_ALL) {
+            MOV64xw(s3, c);
+            if (rex.w)
+                X64_XOR_D(s1, s3);
+            else
+                X64_XOR_W(s1, s3);
+        }
+
+        if (c >= 0 && c <= 4095) {
+            XORI(s1, s1, c);
+        } else {
+            IFX (X_ALL) {
+            } else {
+                MOV64xw(s3, c);
+            }
+            XOR(s1, s1, s3);
+        }
+        if (!rex.w) {
+            ZEROUP(s1);
+        }
+        IFX (X_PEND)
+            SDxw(s1, xEmu, offsetof(x64emu_t, res));
+        return;
+    }
+
+    if (c >= 0 && c <= 4095) {
+        XORI(s1, s1, c);
+    } else {
+        MOV64xw(s3, c);
+        XOR(s1, s1, s3);
+    }
+
+    CLEAR_FLAGS(s3);
+    // test sign bit before zeroup.
+    IFX (X_SF) {
+        if (!rex.w) SEXT_W(s1, s1);
+        BGE(s1, xZR, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    if (!rex.w) {
+        ZEROUP(s1);
+    }
+
+    IFX (X_PEND) {
+        SDxw(s1, xEmu, offsetof(x64emu_t, res));
+    }
+
+    IFX (X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX (X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit AND8 instruction, from s1, s2, store result in s1 using s3 and s4 as scratch, s4 can be same as s2 (and so s2 destroyed)
+void emit_and8(dynarec_la64_t* dyn, int ninst, int s1, int s2, int s3, int s4)
+{
+    IFX(X_PEND) {
+        SET_DF(s3, d_and8);
+    } else IFX(X_ALL) {
+        SET_DFNONE();
+    }
+
+    IFXA(X_ALL, la64_lbt) {
+        X64_AND_B(s1, s2);
+    }
+
+    AND(s1, s1, s2);
+
+    IFX(X_PEND) {
+        ST_B(s1, xEmu, offsetof(x64emu_t, res));
+    }
+
+    if (la64_lbt) return;
+
+    CLEAR_FLAGS(s3);
+    IFX(X_SF) {
+        SRLI_D(s3, s1, 7);
+        BEQZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    IFX(X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+
+// emit AND8 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_and8c(dynarec_la64_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
+{
+    IFX(X_PEND) {
+        SET_DF(s3, d_and8);
+    } else IFX(X_ALL) {
+        SET_DFNONE();
+    }
+
+
+    IFXA(X_ALL, la64_lbt) {
+        MOV32w(s3, c);
+        X64_AND_B(s1, s3);
+    }
+
+    ANDI(s1, s1, c&0xff);
+
+    IFX(X_PEND) {
+        ST_D(s1, xEmu, offsetof(x64emu_t, res));
+    }
+
+    if (la64_lbt) return;
+
+    CLEAR_FLAGS(s3);
+    IFX(X_SF) {
+        SRLI_D(s3, s1, 7);
+        BEQZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    IFX(X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+void emit_and16(dynarec_la64_t* dyn, int ninst, int s1, int s2, int s3, int s4)
+{
+    IFX (X_PEND) {
+        SET_DF(s3, d_tst16);
+    } else IFX (X_ALL) {
+        SET_DFNONE();
+    }
+
+    IFXA (X_ALL, la64_lbt) {
+        X64_AND_W(s1, s2);
+    }
+
+    AND(s1, s1, s2); // res = s1 & s2
+
+
+    IFX (X_PEND) {
+        ST_H(s1, xEmu, offsetof(x64emu_t, res));
+    }
+
+    if (la64_lbt) return;
+
+    CLEAR_FLAGS(s3);
+    IFX (X_SF) {
+        SRLI_D(s3, s1, 15);
+        BEQZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    IFX (X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX (X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit AND32 instruction, from s1, s2, store result in s1 using s3 and s4 as scratch
+void emit_and32(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4)
+{
+    IFX(X_PEND) {
+        SET_DF(s3, rex.w ? d_tst64 : d_tst32);
+    } else IFX(X_ALL) {
+        SET_DFNONE();
+    }
+
+
+    IFXA(X_ALL, la64_lbt) {
+        if (rex.w)
+            X64_AND_D(s1, s2);
+        else
+            X64_AND_W(s1, s2);
+    }
+
+    AND(s1, s1, s2); // res = s1 & s2
+    if (!rex.w) ZEROUP(s1);
+
+    IFX(X_PEND) {
+        SDxw(s1, xEmu, offsetof(x64emu_t, res));
+    }
+
+    if (la64_lbt) return;
+
+    CLEAR_FLAGS(s3);
+    IFX(X_SF) {
+        SRLI_D(s3, s1, rex.w ? 63 : 31);
+        BEQZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    IFX(X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF)
+    {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit AND32 instruction, from s1, c, store result in s1 using s3 and s4 as scratch
+void emit_and32c(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, int64_t c, int s3, int s4)
+{
+    IFX(X_PEND) {
+        SET_DF(s3, rex.w ? d_tst64 : d_tst32);
+    } else IFX(X_ALL) {
+        SET_DFNONE();
+    }
+
+    IFXA(X_ALL, la64_lbt) {
+        MOV64xw(s3, c);
+        if (rex.w)
+            X64_AND_D(s1, s3);
+        else
+            X64_AND_W(s1, s3);
+    }
+
+    if (c >= 0 && c <= 4095) {
+        ANDI(s1, s1, c);
+    } else {
+        IFXA(X_ALL, la64_lbt) { } else MOV64xw(s3, c);
+        AND(s1, s1, s3); // res = s1 & s2
+    }
+
+    IFX(X_PEND)  {
+        SDxw(s1, xEmu, offsetof(x64emu_t, res));
+    }
+
+    if (la64_lbt) return;
+
+    CLEAR_FLAGS(s3);
+    IFX(X_SF) {
+        SRLI_D(s3, s1, rex.w ? 63 : 31);
+        BEQZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    IFX(X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+
+// emit OR16 instruction, from s1, s2, store result in s1 using s3 and s4 as scratch, s4 can be same as s2 (and so s2 destroyed)
+void emit_or16(dynarec_la64_t* dyn, int ninst, int s1, int s2, int s3, int s4)
+{
+    IFX (X_PEND) {
+        SET_DF(s3, d_or16);
+    } else IFX (X_ALL) {
+        SET_DFNONE();
+    }
+
+    IFXA (X_ALL, la64_lbt) {
+        X64_OR_W(s1, s2);
+    }
+
+    OR(s1, s1, s2);
+    BSTRPICK_D(s1, s1, 15, 0);
+    IFX (X_PEND) {
+        ST_D(s1, xEmu, offsetof(x64emu_t, res));
+    }
+
+    if (la64_lbt) return;
+
+    CLEAR_FLAGS(s3);
+    IFX (X_SF) {
+        SRLI_D(s3, s1, 15);
+        BEQZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+
+    IFX (X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX (X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit OR32 instruction, from s1, s2, store result in s1 using s3 and s4 as scratch
+void emit_or32(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4)
+{
+    IFX(X_PEND) {
+        SET_DF(s4, rex.w?d_or64:d_or32);
+    } else IFX(X_ALL) {
+        SET_DFNONE();
+    }
+
+    IFXA (X_ALL, la64_lbt) {
+        if (rex.w)
+            X64_OR_D(s1, s2);
+        else
+            X64_OR_W(s1, s2);
+    }
+
+    OR(s1, s1, s2);
+    if (!rex.w) ZEROUP(s1);
+
+    IFX(X_PEND) {
+        SDxw(s1, xEmu, offsetof(x64emu_t, res));
+    }
+
+    if(la64_lbt) return;
+
+    CLEAR_FLAGS(s3);
+    // test sign bit before zeroup.
+    IFX(X_SF) {
+        if (!rex.w) SEXT_W(s1, s1);
+        BGE(s1, xZR, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    IFX(X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit OR32 instruction, from s1, c, store result in s1 using s3 and s4 as scratch
+void emit_or32c(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, int64_t c, int s3, int s4)
+{
+    IFX(X_PEND) {
+        SET_DF(s4, rex.w ? d_or64 : d_or32);
+    } else IFX(X_ALL) {
+        SET_DFNONE();
+    }
+
+    IFXA(X_ALL, la64_lbt) {
+        MOV64xw(s3, c);
+        if (rex.w)
+            X64_OR_D(s1, s3);
+        else
+            X64_OR_W(s1, s3);
+    }
+
+    if (c >= 0 && c <= 4095) {
+        ORI(s1, s1, c);
+    } else {
+        IFXA(X_ALL, la64_lbt) { } else MOV64xw(s3, c);
+        OR(s1, s1, s3);
+    }
+
+    IFX(X_PEND) {
+        SDxw(s1, xEmu, offsetof(x64emu_t, res));
+    }
+
+    if (la64_lbt) {
+        if (!rex.w) ZEROUP(s1);
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    // test sign bit before zeroup.
+    IFX(X_SF) {
+        if (!rex.w) SEXT_W(s1, s1);
+        BGE(s1, xZR, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    if (!rex.w) {
+        ZEROUP(s1);
+    }
+
+    IFX(X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+
+// emit OR8 instruction, from s1, s2, store result in s1 using s3 and s4 as scratch, s4 can be same as s2 (and so s2 destroyed)
+void emit_or8(dynarec_la64_t* dyn, int ninst, int s1, int s2, int s3, int s4)
+{
+    IFX (X_PEND) {
+        SET_DF(s3, d_or8);
+    } else IFX (X_ALL) {
+        SET_DFNONE();
+    }
+
+    IFXA (X_ALL, la64_lbt) {
+        X64_OR_B(s1, s2);
+    }
+
+    OR(s1, s1, s2);
+
+    IFX (X_PEND) {
+        ST_B(s1, xEmu, offsetof(x64emu_t, res));
+    }
+
+    if (la64_lbt) return;
+
+    CLEAR_FLAGS(s3);
+    IFX (X_SF) {
+        SRLI_D(s3, s1, 7);
+        BEQZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    IFX (X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX (X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit OR8 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_or8c(dynarec_la64_t* dyn, int ninst, int s1, int32_t c, int s2, int s3, int s4)
+{
+    MOV32w(s2, c & 0xff);
+    emit_or8(dyn, ninst, s1, s2, s3, s4);
+}
@@ -0,0 +1,750 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "emu/x64run_private.h"
+#include "la64_emitter.h"
+#include "x64run.h"
+#include "x64emu.h"
+#include "box64stack.h"
+#include "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynarec_native.h"
+
+#include "la64_printer.h"
+#include "dynarec_la64_private.h"
+#include "dynarec_la64_functions.h"
+#include "dynarec_la64_helper.h"
+
+// emit SHL16 instruction, from s1 , shift s2, store result in s1 using s3, s4 and s5 as scratch
+void emit_shl16(dynarec_la64_t* dyn, int ninst, int s1, int s2, int s3, int s4, int s5)
+{
+    // s2 is not 0 here and is 1..1f/3f
+    IFX (X_PEND) {
+        ST_H(s1, xEmu, offsetof(x64emu_t, op1));
+        ST_H(s2, xEmu, offsetof(x64emu_t, op2));
+        SET_DF(s4, d_shl16);
+    } else IFX (X_ALL) {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX (X_ALL) {
+            X64_SLL_H(s1, s2);
+        }
+        SLL_D(s1, s1, s2);
+        BSTRPICK_D(s1, s1, 15, 0);
+        IFX (X_PEND) {
+            ST_H(s1, xEmu, offsetof(x64emu_t, res));
+        }
+
+        return;
+    }
+
+    SLL_D(s1, s1, s2);
+
+    CLEAR_FLAGS(s3);
+    IFX (X_CF | X_OF) {
+        SRLI_D(s5, s1, 16);
+        ANDI(s5, s5, 1); // LSB == F_CF
+        IFX (X_CF) {
+            OR(xFlags, xFlags, s5);
+        }
+    }
+
+    SLLI_D(s1, s1, 48);
+    IFX (X_SF) {
+        BGE(s1, xZR, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    SRLI_D(s1, s1, 48);
+
+    IFX (X_PEND) {
+        ST_H(s1, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX (X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX (X_OF) {
+        // OF flag is affected only on 1-bit shifts
+        ADDI_D(s3, s2, -1);
+        BNEZ(s3, 4 + 4 * 4);
+        SRLI_D(s3, s1, 15);
+        XOR(s3, s3, s5);
+        SLLI_D(s3, s3, F_OF);
+        OR(xFlags, xFlags, s3);
+    }
+    IFX (X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit SHL32 instruction, from s1 , shift s2, store result in s1 using s3, s4 and s5 as scratch
+void emit_shl32(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4, int s5)
+{
+    // s2 is not 0 here and is 1..1f/3f
+    IFX(X_PEND) {
+        SDxw(s1, xEmu, offsetof(x64emu_t, op1));
+        SDxw(s2, xEmu, offsetof(x64emu_t, op2));
+        SET_DF(s4, rex.w?d_shl64:d_shl32);
+    } else IFX(X_ALL) {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX(X_ALL) {
+            if (rex.w)
+                X64_SLL_D(s1, s2);
+            else
+                X64_SLL_W(s1, s2);
+        }
+        SLL_D(s1, s1, s2);
+        if (!rex.w) { ZEROUP(s1); }
+        IFX(X_PEND) {
+            SDxw(s1, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    IFX(X_CF | X_OF) {
+        ADDI_D(s5, s2, rex.w?-64:-32);
+        SUB_D(s5, xZR, s5);
+        SRL_W(s3, s1, s5);
+        ANDI(s5, s3, 1); // LSB == F_CF
+        IFX(X_CF) {
+            OR(xFlags, xFlags, s5);
+        }
+    }
+
+    SLL_D(s1, s1, s2);
+
+    IFX(X_SF) {
+        BGE(s1, xZR, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    if (!rex.w) {
+        ZEROUP(s1);
+    }
+    IFX(X_PEND) {
+        SDxw(s1, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX(X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_OF) {
+        // OF flag is affected only on 1-bit shifts
+        ADDI_D(s3, s2, -1);
+        BNEZ(s3, 4 + 4 * 4);
+        SRLIxw(s3, s1, rex.w?63:31);
+        XOR(s3, s3, s5);
+        SLLI_D(s3, s3, F_OF);
+        OR(xFlags, xFlags, s3);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit SHL32 instruction, from s1 , constant c, store result in s1 using s3, s4 and s5 as scratch
+void emit_shl32c(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, int s3, int s4, int s5)
+{
+    if (!c) return;
+
+    IFX(X_PEND) {
+        if (c) {
+            MOV64x(s3, c);
+            SDxw(s3, xEmu, offsetof(x64emu_t, op2));
+        } else {
+            SDxw(xZR, xEmu, offsetof(x64emu_t, op2));
+        }
+        SDxw(s1, xEmu, offsetof(x64emu_t, op1));
+        SET_DF(s4, rex.w?d_shl64:d_shl32);
+    } else IFX(X_ALL) {
+        SET_DFNONE();
+    }
+
+    if(la64_lbt) {
+        IFX(X_PEND) {} else { MOV64x(s3, c); }
+        IFX(X_ALL) {
+            if (rex.w)
+                X64_SLL_D(s1, s3);
+            else
+                X64_SLL_W(s1, s3);
+        }
+
+        SLLIxw(s1, s1, c);
+
+        IFX(X_PEND) {
+            SDxw(s1, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    IFX(X_CF | X_OF) {
+        if (c > 0) {
+            SRLI_D(s3, s1, (rex.w ? 64 : 32) - c);
+            ANDI(s5, s3, 1); // LSB == F_CF
+            IFX(X_CF) {
+                OR(xFlags, xFlags, s5);
+            }
+        }
+    }
+
+    if (rex.w) {
+        SLLI_D(s1, s1, c);
+    } else {
+        SLLI_W(s1, s1, c);
+    }
+
+    IFX(X_SF) {
+        BGE(s1, xZR, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    if (!rex.w) {
+        ZEROUP(s1);
+    }
+    IFX(X_PEND) {
+        SDxw(s1, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX(X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_OF) {
+        // OF flag is affected only on 1-bit shifts
+        if (c == 1) {
+            SRLIxw(s3, s1, rex.w ? 63 : 31);
+            XOR(s3, s3, s5);
+            SLLI_D(s3, s3, F_OF);
+            OR(xFlags, xFlags, s3);
+        }
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit SHR8 instruction, from s1 , shift s2 (!0 and and'd already), store result in s1 using s3 and s4 as scratch
+void emit_shr8(dynarec_la64_t* dyn, int ninst, int s1, int s2, int s3, int s4, int s5)
+{
+    int64_t j64;
+
+
+    IFX (X_PEND) {
+        ST_B(s2, xEmu, offsetof(x64emu_t, op2));
+        ST_B(s1, xEmu, offsetof(x64emu_t, op1));
+        SET_DF(s4, d_shr8);
+    } else IFX (X_ALL) {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX (X_ALL) {
+            X64_SRL_B(s1, s2);
+        }
+        SRL_D(s1, s1, s2);
+        ANDI(s1, s1, 0xff);
+
+        IFX (X_PEND) {
+            ST_B(s1, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    IFX (X_CF) {
+        ADDI_D(s3, s2, -1);
+        SRA_D(s3, s1, s3);
+        ANDI(s3, s3, 1); // LSB == F_CF
+        OR(xFlags, xFlags, s3);
+    }
+    IFX (X_OF) {
+        // OF flag is affected only on 1-bit shifts
+        // OF flag is set to the most-significant bit of the original operand
+        ADDI_D(s3, xZR, 1);
+        BNE(s2, s3, 4 + 3 * 4);
+        SRLI_D(s3, s1, 7);
+        SLLI_D(s3, s3, F_OF);
+        OR(xFlags, xFlags, s3);
+    }
+
+    SRL_D(s1, s1, s2);
+    ANDI(s1, s1, 0xff);
+
+    // SF should be unset
+    IFX (X_PEND) {
+        ST_B(s1, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX (X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX (X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit SHR16 instruction, from s1 , shift s2 (!0 and and'd already), store result in s1 using s3 and s4 as scratch
+void emit_shr16(dynarec_la64_t* dyn, int ninst, int s1, int s2, int s3, int s4, int s5)
+{
+    int64_t j64;
+
+    IFX (X_PEND) {
+        ST_H(s2, xEmu, offsetof(x64emu_t, op2));
+        ST_H(s1, xEmu, offsetof(x64emu_t, op1));
+        SET_DF(s4, d_shr16);
+    } else IFX (X_ALL) {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX (X_ALL) {
+            X64_SRL_H(s1, s2);
+        }
+
+        SRL_D(s1, s1, s2);
+        BSTRPICK_D(s1, s1, 15, 0);
+
+        IFX (X_PEND) {
+            ST_H(s1, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    IFX (X_CF) {
+        ADDI_D(s3, s2, -1);
+        SRA_D(s3, s1, s3);
+        ANDI(s3, s3, 1); // LSB == F_CF
+        OR(xFlags, xFlags, s3);
+    }
+    IFX (X_OF) {
+        // OF flag is affected only on 1-bit shifts
+        // OF flag is set to the most-significant bit of the original operand
+        ADDI_D(s3, xZR, 1);
+        BNE(s2, s3, 4 + 3 * 4);
+        SRLI_D(s3, s1, 15);
+        SLLI_D(s3, s3, F_OF);
+        OR(xFlags, xFlags, s3);
+    }
+
+    SRL_D(s1, s1, s2);
+    BSTRPICK_D(s1, s1, 15, 0);
+
+    // SF should be unset
+    IFX (X_PEND) {
+        ST_H(s1, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX (X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX (X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit SHR32 instruction, from s1 , shift s2 (!0 and and'd already), store result in s1 using s3 and s4 as scratch
+void emit_shr32(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4)
+{
+    int64_t j64;
+
+
+    IFX (X_PEND) {
+        SDxw(s2, xEmu, offsetof(x64emu_t, op2));
+        SDxw(s1, xEmu, offsetof(x64emu_t, op1));
+        SET_DF(s4, rex.w ? d_shr64 : d_shr32);
+    } else IFX (X_ALL) {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX (X_ALL) {
+            if (rex.w)
+                X64_SRL_D(s1, s2);
+            else
+                X64_SRL_W(s1, s2);
+        }
+        SRL_D(s1, s1, s2);
+        if (!rex.w) { ZEROUP(s1); }
+        IFX (X_PEND) {
+            SDxw(s1, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    IFX (X_CF) {
+        ADDI_D(s3, s2, -1);
+        SRA_D(s3, s1, s3);
+        ANDI(s3, s3, 1); // LSB == F_CF
+        OR(xFlags, xFlags, s3);
+    }
+    IFX (X_OF) {
+        // OF flag is affected only on 1-bit shifts
+        // OF flag is set to the most-significant bit of the original operand
+        ADDI_D(s3, xZR, 1);
+        BNE(s2, s3, 4 + 4 * 4);
+        SRLIxw(s3, s1, rex.w ? 63 : 31);
+        SLLI_D(s3, s3, F_OF);
+        OR(xFlags, xFlags, s3);
+    }
+
+    SRL_D(s1, s1, s2);
+
+    IFX (X_SF) {
+        BGE(s1, xZR, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    if (!rex.w) {
+        ZEROUP(s1);
+    }
+    IFX (X_PEND) {
+        SDxw(s1, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX (X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX (X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit SHR32 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_shr32c(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, int s3, int s4)
+{
+    IFX(X_PEND) {
+        if (c) {
+            MOV64x(s3, c);
+            SDxw(s3, xEmu, offsetof(x64emu_t, op2));
+        } else
+            SDxw(xZR, xEmu, offsetof(x64emu_t, op2));
+        SDxw(s1, xEmu, offsetof(x64emu_t, op1));
+        SET_DF(s4, rex.w ? d_shr64 : d_shr32);
+    } else IFX(X_ALL) {
+        SET_DFNONE();
+    }
+    if (!c) {
+        IFX(X_PEND) {
+            SDxw(s1, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    if (la64_lbt) {
+        IFX(X_PEND) {} else { MOV64x(s3, c); }
+        IFX(X_ALL) {
+            if (rex.w)
+                X64_SRL_D(s1, s3);
+            else
+                X64_SRL_W(s1, s3);
+        }
+
+        SRLIxw(s1, s1, c);
+
+        IFX(X_PEND) {
+            SDxw(s1, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    IFX(X_CF) {
+        if (c > 1) {
+            SRAI_D(s3, s1, c - 1);
+            ANDI(s3, s3, 1); // LSB == F_CF
+        } else {
+            // no need to shift
+            ANDI(s3, s1, 1); // LSB == F_CF
+        }
+        OR(xFlags, xFlags, s3);
+    }
+    IFX(X_OF) {
+        // OF flag is affected only on 1-bit shifts
+        // OF flag is set to the most-significant bit of the original operand
+        if (c == 1) {
+            SRLIxw(s3, s1, rex.w ? 63 : 31);
+            SLLI_D(s3, s3, F_OF);
+            OR(xFlags, xFlags, s3);
+        }
+    }
+
+    if (rex.w) {
+        SRLI_D(s1, s1, c);
+    } else {
+        SRLI_W(s1, s1, c);
+    }
+
+    IFX(X_SF) {
+        BGE(s1, xZR, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    if (!rex.w) {
+        ZEROUP(s1);
+    }
+    IFX(X_PEND) {
+        SDxw(s1, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX(X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+
+// emit SAR16 instruction, from s1, shift s2 (!0 and and'd already), store result in s1 using s3, s4 and s5 as scratch
+void emit_sar16(dynarec_la64_t* dyn, int ninst, int s1, int s2, int s3, int s4, int s5)
+{
+    int64_t j64;
+
+    IFX (X_PEND) {
+        ST_H(s2, xEmu, offsetof(x64emu_t, op2));
+        ST_H(s1, xEmu, offsetof(x64emu_t, op1));
+        SET_DF(s4, d_sar8);
+    } else IFX (X_ALL) {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX (X_ALL) {
+            X64_SRA_H(s1, s2);
+        }
+
+        SRL_D(s1, s1, s2);
+        BSTRPICK_D(s1, s1, 15, 0);
+
+        IFX (X_PEND) {
+            ST_H(s1, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    IFX (X_CF) {
+        ADDI_D(s3, s2, -1);
+        SRA_D(s3, s1, s3);
+        ANDI(s3, s3, 1); // LSB == F_CF
+        OR(xFlags, xFlags, s3);
+    }
+    // For the SAR instruction, the OF flag is cleared for all 1-bit shifts.
+    // OF nop
+    IFX (X_SF) {
+        // SF is the same as the original operand
+        BGE(s1, xZR, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+
+    SRL_D(s1, s1, s2);
+    BSTRPICK_D(s1, s1, 15, 0);
+
+    IFX (X_PEND) {
+        ST_H(s1, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX (X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX (X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit SAR32 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_sar32c(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, int s3, int s4)
+{
+    IFX(X_PEND) {
+        if (c) {
+            MOV64x(s3, c);
+            SDxw(s3, xEmu, offsetof(x64emu_t, op2));
+        } else
+            SDxw(xZR, xEmu, offsetof(x64emu_t, op2));
+        SDxw(s1, xEmu, offsetof(x64emu_t, op1));
+        SET_DF(s4, rex.w ? d_sar64 : d_sar32);
+    } else IFX(X_ALL) {
+        SET_DFNONE();
+    }
+    if (!c) {
+        IFX(X_PEND) {
+            SDxw(s1, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    if (la64_lbt) {
+        IFX(X_PEND) {} else { MOV64x(s3, c); }
+        IFX(X_ALL) {
+            if (rex.w)
+                X64_SRA_D(s1, s3);
+            else
+                X64_SRA_W(s1, s3);
+        }
+
+        SRAIxw(s1, s1, c);
+
+        IFX(X_PEND) {
+            SDxw(s1, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    IFX(X_CF) {
+        if (c > 1) {
+            SRAI_D(s3, s1, c - 1);
+            ANDI(s3, s3, 1); // LSB == F_CF
+        } else {
+            // no need to shift
+            ANDI(s3, s1, 1); // LSB == F_CF
+        }
+        OR(xFlags, xFlags, s3);
+    }
+
+    if (rex.w) {
+        SRAI_D(s1, s1, c);
+    } else {
+        SRAI_W(s1, s1, c);
+    }
+
+    // SRAI_W sign-extends, so test sign bit before clearing upper bits
+    IFX(X_SF) {
+        BGE(s1, xZR, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    if (!rex.w) {
+        ZEROUP(s1);
+    }
+    IFX(X_PEND) {
+        SDxw(s1, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX(X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+
+// emit ROR32 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_ror32c(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, int s3, int s4)
+{
+    if (!c) return;
+
+    IFX (X_PEND) {
+        MOV32w(s3, c);
+        SDxw(s3, xEmu, offsetof(x64emu_t, op2));
+        SET_DF(s4, rex.w ? d_ror64 : d_ror32);
+    } else IFX (X_ALL) {
+        SET_DFNONE();
+    }
+    if (!c) {
+        IFX (X_PEND) {
+            SDxw(s1, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    if (la64_lbt) {
+        IFX (X_ALL) {
+            if (rex.w)
+                X64_ROTRI_D(s1, c);
+            else
+                X64_ROTRI_W(s1, c);
+        }
+    }
+
+    ROTRIxw(s1, s1, c);
+
+    IFX (X_PEND) {
+        SDxw(s1, xEmu, offsetof(x64emu_t, res));
+    }
+
+    if (la64_lbt) return;
+
+    CLEAR_FLAGS(s3);
+    IFX (X_CF) {
+        SRLIxw(s3, s1, rex.w ? 63 : 31);
+        OR(xFlags, xFlags, s3);
+    }
+    IFX (X_OF) {
+        // the OF flag is set to the exclusive OR of the two most-significant bits of the result
+        if (c == 1) {
+            SRLI_D(s3, s1, rex.w ? 62 : 30);
+            SRLI_D(s4, s3, 1);
+            XOR(s3, s3, s4);
+            ANDI(s3, s3, 1);
+            SLLI_D(s3, s3, F_OF);
+            OR(xFlags, xFlags, s3);
+        }
+    }
+}
+
+// emit ROL32 instruction, from s1, s2, store result in s1 using s3 and s4 as scratch
+void emit_rol32(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4)
+{
+    int64_t j64;
+
+    IFX (X_PEND) {
+        SDxw(s2, xEmu, offsetof(x64emu_t, op2));
+        SET_DF(s4, rex.w ? d_rol64 : d_rol32);
+    } else IFX (X_ALL) {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX (X_ALL) {
+            if (rex.w)
+                X64_ROTL_D(s1, s2);
+            else
+                X64_ROTL_W(s1, s2);
+        }
+    }
+
+    if (rex.w) {
+        ANDI(s4, s2, 0x3f);
+    } else {
+        ANDI(s4, s2, 0x1f);
+    }
+
+    SLLxw(s3, s1, s4);
+    NEG_D(s4, s4);
+    ADDI_D(s4, s4, rex.w ? 64 : 32);
+    SRLxw(s1, s1, s4);
+    OR(s1, s3, s1);
+
+    IFX (X_PEND) {
+        SDxw(s1, xEmu, offsetof(x64emu_t, res));
+    }
+
+    if (la64_lbt) return;
+
+    CLEAR_FLAGS(s3);
+    IFX (X_CF | X_OF) {
+        ANDI(s4, s1, 1); // LSB == F_CF
+        IFX (X_CF) OR(xFlags, xFlags, s4);
+    }
+    IFX (X_OF) {
+        // the OF flag is set to the exclusive OR of the CF bit (after the rotate) and the most-significant bit of the result.
+        ADDI_D(s3, xZR, 1);
+        BNE_NEXT(s2, s3);
+        SRLIxw(s3, s1, rex.w ? 63 : 31);
+        XOR(s3, s3, s4); // s3: MSB, s4: CF bit
+        SLLI_D(s3, s3, F_OF);
+        OR(xFlags, xFlags, s3);
+    }
+}
@@ -0,0 +1,474 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "emu/x64run_private.h"
+#include "x64run.h"
+#include "x64emu.h"
+#include "box64stack.h"
+#include "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynarec_native.h"
+
+#include "la64_printer.h"
+#include "dynarec_la64_private.h"
+#include "dynarec_la64_functions.h"
+#include "dynarec_la64_helper.h"
+
+
+// emit CMP8 instruction, from cmp s1, s2, using s3, s4, s5 and s6 as scratch
+void emit_cmp8(dynarec_la64_t* dyn, int ninst, int s1, int s2, int s3, int s4, int s5, int s6)
+{
+    IFX_PENDOR0 {
+        ST_B(s1, xEmu, offsetof(x64emu_t, op1));
+        ST_B(s2, xEmu, offsetof(x64emu_t, op2));
+        SET_DF(s4, d_cmp8);
+    } else {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX(X_ALL) {
+            X64_SUB_B(s1, s2);
+        }
+
+        IFX_PENDOR0 {
+            SUB_D(s6, s1, s2);
+            ST_B(s6, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    IFX(X_AF | X_CF | X_OF) {
+        // for later flag calculation
+        NOR(s5, xZR, s1);
+    }
+
+    // It's a cmp, we can't store the result back to s1.
+    SUB_D(s6, s1, s2);
+    ANDI(s6, s6, 0xff);
+    IFX_PENDOR0 {
+        ST_B(s6, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX(X_SF) {
+        SRLI_D(s3, s6, 7);
+        BEQZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    CALC_SUB_FLAGS(s5, s2, s6, s3, s4, 8);
+    IFX(X_ZF) {
+        BNEZ(s6, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s6, s3, s4);
+    }
+}
+
+// emit CMP8 instruction, from cmp s1 , 0, using s3 and s4 as scratch
+void emit_cmp8_0(dynarec_la64_t* dyn, int ninst, int s1, int s3, int s4)
+{
+    IFX_PENDOR0 {
+        ST_B(s1, xEmu, offsetof(x64emu_t, op1));
+        ST_B(xZR, xEmu, offsetof(x64emu_t, op2));
+        ST_B(s1, xEmu, offsetof(x64emu_t, res));
+        SET_DF(s3, d_cmp8);
+    } else {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX(X_ALL) {
+            X64_SUB_B(s1, xZR);
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    IFX(X_SF) {
+        SRLI_D(s3, s1, 7);
+        BEQZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    IFX(X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit CMP16 instruction, from cmp s1, s2, using s3 and s4 as scratch
+void emit_cmp16(dynarec_la64_t* dyn, int ninst, int s1, int s2, int s3, int s4, int s5, int s6)
+{
+    IFX_PENDOR0 {
+        ST_H(s1, xEmu, offsetof(x64emu_t, op1));
+        ST_H(s2, xEmu, offsetof(x64emu_t, op2));
+        SET_DF(s4, d_cmp16);
+    } else {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX(X_ALL) {
+            X64_SUB_H(s1, s2);
+        }
+
+        IFX_PENDOR0 {
+            SUB_D(s6, s1, s2);
+            ST_H(s6, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    IFX(X_AF | X_CF | X_OF) {
+        // for later flag calculation
+        NOR(s5, xZR, s1);
+    }
+
+    // It's a cmp, we can't store the result back to s1.
+    SUB_D(s6, s1, s2);
+    IFX(X_ALL) {
+        BSTRPICK_D(s6, s6, 15, 0);
+    }
+    IFX_PENDOR0 {
+        ST_H(s6, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX(X_SF) {
+        SRLI_D(s3, s6, 15);
+        BEQZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    CALC_SUB_FLAGS(s5, s2, s6, s3, s4, 16);
+    IFX(X_ZF) {
+        BNEZ(s6, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s6, s3, s4);
+    }
+}
+
+// emit CMP16 instruction, from cmp s1 , #0, using s3 and s4 as scratch
+void emit_cmp16_0(dynarec_la64_t* dyn, int ninst, int s1, int s3, int s4)
+{
+    IFX_PENDOR0 {
+        ST_H(s1, xEmu, offsetof(x64emu_t, op1));
+        ST_H(xZR, xEmu, offsetof(x64emu_t, op2));
+        ST_H(s1, xEmu, offsetof(x64emu_t, res));
+        SET_DF(s3, d_cmp16);
+    } else {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX(X_ALL) {
+            X64_SUB_H(s1, xZR);
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    IFX(X_SF) {
+        SRLI_D(s3, s1, 15);
+        BEQZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    IFX(X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit CMP32 instruction, from cmp s1, s2, using s3 and s4 as scratch
+void emit_cmp32(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4, int s5, int s6)
+{
+    IFX_PENDOR0 {
+        SDxw(s1, xEmu, offsetof(x64emu_t, op1));
+        SDxw(s2, xEmu, offsetof(x64emu_t, op2));
+        SET_DF(s4, rex.w?d_cmp64:d_cmp32);
+    } else {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX(X_ALL) {
+            if (rex.w)
+                X64_SUB_D(s1, s2);
+            else
+                X64_SUB_W(s1, s2);
+        }
+
+        IFX_PENDOR0 {
+            SUBxw(s6, s1, s2);
+            SDxw(s6, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    IFX(X_AF | X_CF | X_OF) {
+        // for later flag calculation
+        NOR(s5, xZR, s1);
+    }
+
+    // It's a cmp, we can't store the result back to s1.
+    SUBxw(s6, s1, s2);
+    IFX_PENDOR0 {
+        SDxw(s6, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX(X_SF) {
+        BGE(s6, xZR, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    if (!rex.w) {
+        ZEROUP(s6);
+    }
+    CALC_SUB_FLAGS(s5, s2, s6, s3, s4, rex.w?64:32);
+    IFX(X_ZF) {
+        BNEZ(s6, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s6, s3, s4);
+    }
+}
+
+// emit CMP32 instruction, from cmp s1, 0, using s3 and s4 as scratch
+void emit_cmp32_0(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, int s3, int s4)
+{
+    IFX_PENDOR0 {
+        ST_D(s1, xEmu, offsetof(x64emu_t, op1));
+        ST_D(xZR, xEmu, offsetof(x64emu_t, op2));
+        ST_D(s1, xEmu, offsetof(x64emu_t, res));
+        SET_DF(s4, rex.w?d_cmp64:d_cmp32);
+    } else {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX(X_ALL) {
+            if (rex.w)
+                X64_SUB_D(s1, xZR);
+            else
+                X64_SUB_W(s1, xZR);
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    IFX(X_SF) {
+        if (rex.w) {
+            BGE(s1, xZR, 8);
+        } else {
+            SRLI_D(s3, s1, 31);
+            BEQZ(s3, 8);
+        }
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    IFX(X_ZF) {
+        BNEZ(s1, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+
+// emit TEST8 instruction, from test s1, s2, using s3, s4 and s5 as scratch
+void emit_test8(dynarec_la64_t* dyn, int ninst, int s1, int s2, int s3, int s4, int s5)
+{
+    IFX_PENDOR0 {
+        SET_DF(s3, d_tst8);
+    } else {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX(X_ALL) {
+            X64_AND_B(s1, s2);
+        }
+
+        IFX_PENDOR0 {
+            AND(s3, s1, s2);
+            ST_D(s3, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    AND(s3, s1, s2); // res = s1 & s2
+
+    IFX_PENDOR0 {
+        ST_D(s3, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX(X_SF) {
+        SRLI_D(s4, s3, 7);
+        BEQZ(s4, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    IFX(X_ZF) {
+        BNEZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s3, s4, s5);
+    }
+}
+
+// emit TEST16 instruction, from test s1, s2, using s3, s4 and s5 as scratch
+void emit_test16(dynarec_la64_t* dyn, int ninst, int s1, int s2, int s3, int s4, int s5)
+{
+    IFX_PENDOR0 {
+        SET_DF(s3, d_tst16);
+    } else {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX (X_ALL) {
+            X64_AND_H(s1, s2);
+        }
+
+        IFX_PENDOR0 {
+            AND(s3, s1, s2);
+            ST_H(s3, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+    CLEAR_FLAGS(s3);
+    AND(s3, s1, s2); // res = s1 & s2
+
+    IFX_PENDOR0 {
+        ST_H(s3, xEmu, offsetof(x64emu_t, res));
+    }
+
+    IFX (X_SF) {
+        SRLI_D(s4, s3, 15);
+        BEQZ(s4, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    IFX (X_ZF) {
+        BNEZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX (X_PF) {
+        emit_pf(dyn, ninst, s3, s4, s5);
+    }
+}
+
+// emit TEST32 instruction, from test s1, s2, using s3 and s4 as scratch
+void emit_test32(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4, int s5)
+{
+    IFX_PENDOR0 {
+        SET_DF(s3, rex.w?d_tst64:d_tst32);
+    } else {
+        SET_DFNONE();
+    }
+
+    if (la64_lbt) {
+        IFX(X_ALL) {
+            if (rex.w)
+                X64_AND_D(s1, s2);
+            else
+                X64_AND_W(s1, s2);
+        }
+
+        IFX_PENDOR0 {
+            AND(s3, s1, s2);
+            SDxw(s3, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    AND(s3, s1, s2); // res = s1 & s2
+
+    IFX_PENDOR0 {
+        SDxw(s3, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX(X_SF | X_ZF) {
+        if (!rex.w) ZEROUP(s3);
+    }
+    IFX(X_SF) {
+        SRLI_D(s4, s3, rex.w?63:31);
+        BEQZ(s4, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    IFX(X_ZF) {
+        BNEZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s3, s4, s5);
+    }
+}
+
+// emit TEST32 instruction, from test s1, s2, using s3 and s4 as scratch
+void emit_test32c(dynarec_la64_t* dyn, int ninst, rex_t rex, int s1, int64_t c, int s3, int s4, int s5)
+{
+    IFX_PENDOR0 {
+        SET_DF(s3, rex.w ? d_tst64 : d_tst32);
+    } else {
+        SET_DFNONE();
+    }
+
+
+    if (la64_lbt) {
+        IFX(X_ALL) {
+            MOV64xw(s3, c);
+            if (rex.w)
+                X64_AND_D(s1, s3);
+            else
+                X64_AND_W(s1, s3);
+        }
+
+        IFX_PENDOR0 {
+            if (c >= 0 && c <= 4095) {
+                ANDI(s3, s1, c);
+            } else {
+                IFX(X_ALL) { } else MOV64xw(s3, c);
+                AND(s3, s1, s3);
+            }
+            SDxw(s3, xEmu, offsetof(x64emu_t, res));
+        }
+        return;
+    }
+
+    CLEAR_FLAGS(s3);
+    if (c >= 0 && c <= 4095) {
+        ANDI(s3, s1, c);
+    } else {
+        IFXA(X_ALL, la64_lbt) { } else MOV64xw(s3, c);
+        AND(s3, s1, s3);
+    }
+
+    IFX_PENDOR0 {
+        SDxw(s3, xEmu, offsetof(x64emu_t, res));
+    }
+    IFX(X_SF) {
+        SRLI_D(s4, s3, rex.w ? 63 : 31);
+        BEQZ(s4, 8);
+        ORI(xFlags, xFlags, 1 << F_SF);
+    }
+    IFX(X_ZF) {
+        BNEZ(s3, 8);
+        ORI(xFlags, xFlags, 1 << F_ZF);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s3, s4, s5);
+    }
+}
@@ -0,0 +1,286 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "emu/x64run_private.h"
+#include "la64_emitter.h"
+#include "x64run.h"
+#include "x64emu.h"
+#include "box64stack.h"
+#include "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynarec_native.h"
+
+#include "la64_printer.h"
+#include "dynarec_la64_private.h"
+#include "dynarec_la64_helper.h"
+#include "dynarec_la64_functions.h"
+
+
+uintptr_t dynarec64_F0(dynarec_la64_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+{
+    (void)ip;
+    (void)rep;
+    (void)need_epilog;
+
+    uint8_t opcode = F8;
+    uint8_t nextop;
+    uint8_t gd, ed, u8;
+    uint8_t wback, wb1, wb2, eb1, eb2, gb1, gb2;
+    int32_t i32;
+    int64_t i64, j64;
+    int64_t fixedaddress;
+    int unscaled;
+    MAYUSE(eb1);
+    MAYUSE(eb2);
+    MAYUSE(gb1);
+    MAYUSE(gb2);
+    MAYUSE(wb1);
+    MAYUSE(wb2);
+    MAYUSE(j64);
+
+    while ((opcode == 0xF2) || (opcode == 0xF3)) {
+        rep = opcode - 0xF1;
+        opcode = F8;
+    }
+
+    GETREX();
+
+    switch (opcode) {
+        case 0x01:
+            INST_NAME("LOCK ADD Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETGD;
+            SMDMB();
+            if ((nextop & 0xC0) == 0xC0) {
+                ed = TO_LA64((nextop & 7) + (rex.b << 3));
+                emit_add32(dyn, ninst, rex, ed, gd, x3, x4, x5);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, LOCK_LOCK, 0, 0);
+                MARKLOCK;
+                LLxw(x1, wback, 0);
+                ADDxw(x4, x1, gd);
+                SCxw(x4, wback, 0);
+                BEQZ_MARKLOCK(x4);
+                IFX (X_ALL | X_PEND) {
+                    emit_add32(dyn, ninst, rex, x1, gd, x3, x4, x5);
+                }
+            }
+            SMDMB();
+            break;
+        case 0x0F:
+            nextop = F8;
+            switch (nextop) {
+                case 0xB1:
+                    switch (rep) {
+                        case 0:
+                            INST_NAME("LOCK CMPXCHG Ed, Gd");
+                            SETFLAGS(X_ALL, SF_SET_PENDING);
+                            nextop = F8;
+                            GETGD;
+                            if (MODREG) {
+                                ed = TO_LA64((nextop & 7) + (rex.b << 3));
+                                wback = 0;
+                                UFLAG_IF { emit_cmp32(dyn, ninst, rex, xRAX, ed, x3, x4, x5, x6); }
+                                MV(x1, ed); // save value
+                                SUB_D(x2, x1, xRAX);
+                                BNE_MARK2(x2, xZR);
+                                MV(ed, gd);
+                                MARK2;
+                                MVxw(xRAX, x1);
+                            } else {
+                                SMDMB();
+                                addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, LOCK_LOCK, 0, 0);
+                                MARKLOCK;
+                                MV(x4, gd);
+                                LLxw(x1, wback, 0);
+                                SUBxw(x3, x1, xRAX);
+                                BNEZ_MARK(x3);
+                                // EAX == Ed
+                                SCxw(x4, wback, 0);
+                                BEQZ_MARKLOCK(x4);
+                                MARK;
+                                UFLAG_IF { emit_cmp32(dyn, ninst, rex, xRAX, x1, x3, x4, x5, x6); }
+                                MVxw(xRAX, x1);
+                                SMDMB();
+                            }
+                            break;
+                        default:
+                            DEFAULT;
+                    }
+                    break;
+                case 0xC1:
+                    switch (rep) {
+                        case 0:
+                            INST_NAME("LOCK XADD Gd, Ed");
+                            SETFLAGS(X_ALL, SF_SET_PENDING);
+                            nextop = F8;
+                            GETGD;
+                            SMDMB();
+                            if (MODREG) {
+                                ed = TO_LA64((nextop & 7) + (rex.b << 3));
+                                MVxw(x1, ed);
+                                MVxw(ed, gd);
+                                MVxw(gd, x1);
+                                emit_add32(dyn, ninst, rex, ed, gd, x3, x4, x5);
+                            } else {
+                                addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, LOCK_LOCK, 0, 0);
+                                MARKLOCK;
+                                LLxw(x1, wback, 0);
+                                ADDxw(x4, x1, gd);
+                                SCxw(x4, wback, 0);
+                                BEQZ_MARKLOCK(x4);
+                                IFX(X_ALL | X_PEND) {
+                                    MVxw(x2, x1);
+                                    emit_add32(dyn, ninst, rex, x2, gd, x3, x4, x5);
+                                }
+                                MVxw(gd, x1);
+                            }
+                            SMDMB();
+                            break;
+                        default:
+                            DEFAULT;
+                    }
+                    break;
+                default:
+                    DEFAULT;
+            }
+            break;
+        case 0x29:
+            INST_NAME("LOCK SUB Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETGD;
+            SMDMB();
+            if (MODREG) {
+                ed = TO_LA64((nextop & 7) + (rex.b << 3));
+                emit_sub32(dyn, ninst, rex, ed, gd, x3, x4, x5);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, LOCK_LOCK, 0, 0);
+                MARKLOCK;
+                LLxw(x1, wback, 0);
+                SUB_D(x4, x1, gd);
+                SCxw(x4, wback, 0);
+                BEQZ_MARKLOCK(x4);
+                IFX (X_ALL | X_PEND)
+                    emit_sub32(dyn, ninst, rex, x1, gd, x3, x4, x5);
+            }
+            SMDMB();
+            break;
+        case 0x81:
+        case 0x83:
+            nextop = F8;
+            SMDMB();
+            switch ((nextop >> 3) & 7) {
+                case 0: // ADD
+                    if (opcode == 0x81) {
+                        INST_NAME("LOCK ADD Ed, Id");
+                    } else {
+                        INST_NAME("LOCK ADD Ed, Ib");
+                    }
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    if (MODREG) {
+                        if (opcode == 0x81)
+                            i64 = F32S;
+                        else
+                            i64 = F8S;
+                        ed = TO_LA64((nextop & 7) + (rex.b << 3));
+                        emit_add32c(dyn, ninst, rex, ed, i64, x3, x4, x5, x6);
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, LOCK_LOCK, 0, (opcode == 0x81) ? 4 : 1);
+                        if (opcode == 0x81)
+                            i64 = F32S;
+                        else
+                            i64 = F8S;
+                        if (i64 < -2048 || i64 >= 2048) {
+                            MOV64xw(x3, i64);
+                        }
+                        MARKLOCK;
+                        LLxw(x1, wback, 0);
+                        if (i64 >= -2048 && i64 < 2048) {
+                            ADDIxw(x4, x1, i64);
+                        } else {
+                            ADDxw(x4, x1, x3);
+                        }
+                        SCxw(x4, wback, 0);
+                        BEQZ_MARKLOCK(x4);
+                        IFX(X_ALL | X_PEND) {
+                            emit_add32c(dyn, ninst, rex, x1, i64, x3, x4, x5, x6);
+                        }
+                        SMDMB();
+                    }
+                    break;
+                case 5: // SUB
+                    if (opcode == 0x81) {
+                        INST_NAME("LOCK SUB Ed, Id");
+                    } else {
+                        INST_NAME("LOCK SUB Ed, Ib");
+                    }
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    if (MODREG) {
+                        if (opcode == 0x81)
+                            i64 = F32S;
+                        else
+                            i64 = F8S;
+                        ed = xRAX + (nextop & 7) + (rex.b << 3);
+                        emit_sub32c(dyn, ninst, rex, ed, i64, x3, x4, x5, x6);
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, LOCK_LOCK, 0, (opcode == 0x81) ? 4 : 1);
+                        if (opcode == 0x81)
+                            i64 = F32S;
+                        else
+                            i64 = F8S;
+                        if (i64 <= -2048 || i64 > 2048) {
+                            MOV64xw(x3, i64);
+                        }
+                        MARKLOCK;
+                        LLxw(x1, wback, 0);
+                        if (i64 > -2048 && i64 <= 2048) {
+                            ADDIxw(x4, x1, -i64);
+                        } else {
+                            SUBxw(x4, x1, x3);
+                        }
+                        SCxw(x4, wback, 0);
+                        BEQZ_MARKLOCK(x4);
+                        IFX (X_ALL | X_PEND)
+                            emit_sub32c(dyn, ninst, rex, x1, i64, x3, x4, x5, x6);
+                    }
+                    break;
+                default:
+                    DEFAULT;
+            }
+            break;
+        case 0x87:
+            INST_NAME("LOCK XCHG Ed, Gd");
+            nextop = F8;
+            if (MODREG) {
+                GETGD;
+                GETED(0);
+                MV(x1, gd);
+                MV(gd, ed);
+                MV(ed, x1);
+            } else {
+                SMDMB();
+                GETGD;
+                addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, LOCK_LOCK, 0, 0);
+                MARKLOCK;
+                LLxw(x1, wback, 0);
+                MV(x4, gd);
+                SCxw(x4, wback, 0);
+                BEQZ_MARKLOCK(x4);
+                MVxw(gd, x1);
+                SMDMB();
+            }
+            break;
+        default:
+            DEFAULT;
+    }
+    return addr;
+}
@@ -0,0 +1,271 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "emu/x64run_private.h"
+#include "la64_emitter.h"
+#include "x64run.h"
+#include "x64emu.h"
+#include "box64stack.h"
+#include "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynarec_native.h"
+
+#include "la64_printer.h"
+#include "dynarec_la64_private.h"
+#include "dynarec_la64_functions.h"
+#include "dynarec_la64_helper.h"
+
+uintptr_t dynarec64_F20F(dynarec_la64_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
+{
+    (void)ip; (void)need_epilog;
+
+    uint8_t opcode = F8;
+    uint8_t nextop;
+    uint8_t gd, ed;
+    uint8_t wback, wb1, wb2;
+    uint8_t u8;
+    uint64_t u64, j64;
+    int v0, v1;
+    int q0, q1;
+    int d0, d1;
+    int64_t fixedaddress;
+    int unscaled;
+
+    MAYUSE(d0);
+    MAYUSE(d1);
+    MAYUSE(q0);
+    MAYUSE(v0);
+    MAYUSE(v1);
+
+    switch(opcode) {
+        case 0x10:
+            INST_NAME("MOVSD Gx, Ex");
+            nextop = F8;
+            GETG;
+            if(MODREG) {
+                ed = (nextop & 7) + (rex.b << 3);
+                v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+                v1 = sse_get_reg(dyn, ninst, x1, ed, 0);
+            } else {
+                SMREAD();
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                v1 = fpu_get_scratch(dyn);
+                VXOR_V(v0, v0, v0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, x2, &fixedaddress, rex, NULL, 8, 0);
+                FLD_D(v1, ed, fixedaddress);
+            }
+            VEXTRINS_D(v0, v1, 0); // v0[63:0] = v1[63:0]
+            break;
+        case 0x11:
+            INST_NAME("MOVSD Ex, Gx");
+            nextop = F8;
+            GETG;
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
+            if(MODREG) {
+                ed = (nextop & 7) + (rex.b << 3);
+                d0 = sse_get_reg(dyn, ninst, x1, ed, 0);
+                VEXTRINS_D(d0, v0, 0); // d0[63:0] = v0[63:0]
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, x2, &fixedaddress, rex, NULL, 1, 0);
+                FST_D(v0, ed, fixedaddress);
+                SMWRITE2();
+            }
+            break;
+        case 0x2A:
+            INST_NAME("CVTSI2SD Gx, Ed");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETED(0);
+            d1 = fpu_get_scratch(dyn);
+            if (rex.w) {
+                MOVGR2FR_D(d1, ed);
+                FFINT_D_L(d1, d1);
+            } else {
+                MOVGR2FR_W(d1, ed);
+                FFINT_D_W(d1, d1);
+            }
+            VEXTRINS_D(v0, d1, 0);
+            break;
+        case 0x2C:
+            INST_NAME("CVTTSD2SI Gd, Ex");
+            nextop = F8;
+            GETGD;
+            GETEXSD(q0, 0, 0);
+            if (!box64_dynarec_fastround) {
+                MOVGR2FCSR(FCSR2, xZR); // reset all bits
+            }
+            d1 = fpu_get_scratch(dyn);
+            if (rex.w) {
+                FTINTRZ_L_D(d1, q0);
+                MOVFR2GR_D(gd, d1);
+            } else {
+                FTINTRZ_W_D(d1, q0);
+                MOVFR2GR_S(gd, d1);
+            }
+            if (!rex.w) ZEROUP(gd);
+            if (!box64_dynarec_fastround) {
+                MOVFCSR2GR(x5, FCSR2); // get back FPSR to check
+                MOV32w(x3, (1 << FR_V) | (1 << FR_O));
+                AND(x5, x5, x3);
+                CBZ_NEXT(x5);
+                if (rex.w) {
+                    MOV64x(gd, 0x8000000000000000LL);
+                } else {
+                    MOV32w(gd, 0x80000000);
+                }
+            }
+            break;
+        case 0x2D:
+            INST_NAME("CVTSD2SI Gd, Ex");
+            nextop = F8;
+            GETGD;
+            GETEXSD(q0, 0, 0);
+            if (!box64_dynarec_fastround) {
+                MOVGR2FCSR(FCSR2, xZR); // reset all bits
+            }
+            d1 = fpu_get_scratch(dyn);
+            u8 = sse_setround(dyn, ninst, x2, x3);
+            if (rex.w) {
+                FTINT_L_D(d1, q0);
+                MOVFR2GR_D(gd, d1);
+            } else {
+                FTINT_W_D(d1, q0);
+                MOVFR2GR_S(gd, d1);
+            }
+            x87_restoreround(dyn, ninst, u8);
+            if (!box64_dynarec_fastround) {
+                MOVFCSR2GR(x5, FCSR2); // get back FPSR to check
+                MOV32w(x3, (1 << FR_V) | (1 << FR_O));
+                AND(x5, x5, x3);
+                CBZ_NEXT(x5);
+                if (rex.w) {
+                    MOV64x(gd, 0x8000000000000000LL);
+                } else {
+                    MOV32w(gd, 0x80000000);
+                }
+            }
+            break;
+        case 0x58:
+            INST_NAME("ADDSD Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEXSD(v1, 0, 0);
+            d0 = fpu_get_scratch(dyn);
+            FADD_D(d0, v0, v1);
+            if (!box64_dynarec_fastnan) {
+                FCMP_D(fcc0, v0, v1, cUN);
+                BCNEZ_MARK(fcc0);
+                FCMP_D(fcc1, d0, d0, cOR);
+                BCNEZ_MARK(fcc1);
+                FNEG_D(d0, d0);
+            }
+            MARK;
+            VEXTRINS_D(v0, d0, 0); // v0[63:0] = d0[63:0]
+            break;
+        case 0x59:
+            INST_NAME("MULSD Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEXSD(v1, 0, 0);
+            d0 = fpu_get_scratch(dyn);
+            FMUL_D(d0, v0, v1);
+            if (!box64_dynarec_fastnan) {
+                FCMP_D(fcc0, v0, v1, cUN);
+                BCNEZ_MARK(fcc0);
+                FCMP_D(fcc1, d0, d0, cOR);
+                BCNEZ_MARK(fcc1);
+                FNEG_D(d0, d0);
+            }
+            MARK;
+            VEXTRINS_D(v0, d0, 0); // v0[63:0] = d0[63:0]
+            break;
+        case 0x5A:
+            INST_NAME("CVTSD2SS Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEXSD(d0, 0, 0);
+            d1 = fpu_get_scratch(dyn);
+            FCVT_S_D(d1, d0);
+            VEXTRINS_W(v0, d1, 0);
+            break;
+        case 0x5C:
+            INST_NAME("SUBSD Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEXSD(v1, 0, 0);
+            d0 = fpu_get_scratch(dyn);
+            FSUB_D(d0, v0, v1);
+            if (!box64_dynarec_fastnan) {
+                FCMP_D(fcc0, v0, v1, cUN);
+                BCNEZ_MARK(fcc0);
+                FCMP_D(fcc1, d0, d0, cOR);
+                BCNEZ_MARK(fcc1);
+                FNEG_D(d0, d0);
+            }
+            MARK;
+            VEXTRINS_D(v0, d0, 0); // v0[63:0] = d0[63:0]
+            break;
+        case 0x5E:
+            INST_NAME("DIVSD Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEXSD(v1, 0, 0);
+            d0 = fpu_get_scratch(dyn);
+            FDIV_D(d0, v0, v1);
+            if (!box64_dynarec_fastnan) {
+                FCMP_D(fcc0, v0, v1, cUN);
+                BCNEZ_MARK(fcc0);
+                FCMP_D(fcc1, d0, d0, cOR);
+                BCNEZ_MARK(fcc1);
+                FNEG_D(d0, d0);
+            }
+            MARK;
+            VEXTRINS_D(v0, d0, 0); // v0[63:0] = d0[63:0]
+            break;
+        case 0x5F:
+            INST_NAME("MAXSD Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEXSD(v1, 0, 0);
+            FCMP_D(fcc0, v0, v1, cUN);
+            BCNEZ_MARK(fcc0);
+            FCMP_D(fcc1, v0, v1, cLE);
+            BCEQZ_MARK2(fcc1);
+            MARK;
+            VEXTRINS_D(v0, v1, 0);
+            MARK2;
+            break;
+        case 0xC2:
+            INST_NAME("CMPSD Gx, Ex, Ib");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEXSS(v1, 0, 1);
+            u8 = F8;
+            switch (u8 & 7) {
+                case 0: FCMP_D(fcc0, v0, v1, cEQ); break;  // Equal
+                case 1: FCMP_D(fcc0, v0, v1, cLT); break;  // Less than
+                case 2: FCMP_D(fcc0, v0, v1, cLE); break;  // Less or equal
+                case 3: FCMP_D(fcc0, v0, v1, cUN); break;  // NaN
+                case 4: FCMP_D(fcc0, v0, v1, cUNE); break; // Not Equal or unordered
+                case 5: FCMP_D(fcc0, v1, v0, cULE); break; // Greater or equal or unordered
+                case 6: FCMP_D(fcc0, v1, v0, cULT); break; // Greater or unordered, test inverted, N!=V so unordered or less than (inverted)
+                case 7: FCMP_D(fcc0, v0, v1, cOR); break;  // not NaN
+            }
+            MOVCF2GR(x2, fcc0);
+            NEG_D(x2, x2);
+            q1 = fpu_get_scratch(dyn);
+            MOVGR2FR_D(q1, x2);
+            VEXTRINS_D(v0, q1, 0);
+            break;
+        default:
+            DEFAULT;
+    }
+    return addr;
+}
@@ -0,0 +1,283 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "emu/x64run_private.h"
+#include "la64_emitter.h"
+#include "x64run.h"
+#include "x64emu.h"
+#include "box64stack.h"
+#include "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynarec_native.h"
+#include "bitutils.h"
+
+#include "la64_printer.h"
+#include "dynarec_la64_private.h"
+#include "dynarec_la64_functions.h"
+#include "dynarec_la64_helper.h"
+
+uintptr_t dynarec64_F30F(dynarec_la64_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
+{
+    (void)ip;
+    (void)need_epilog;
+
+    uint8_t opcode = F8;
+    uint8_t nextop, u8;
+    uint8_t gd, ed;
+    uint8_t wback, gback;
+    uint64_t u64;
+    int v0, v1;
+    int q0, q1;
+    int d0, d1;
+    int64_t fixedaddress, gdoffset;
+    int unscaled;
+    int64_t j64;
+
+    MAYUSE(d0);
+    MAYUSE(d1);
+    MAYUSE(q0);
+    MAYUSE(q1);
+    MAYUSE(v0);
+    MAYUSE(v1);
+    MAYUSE(j64);
+
+    switch (opcode) {
+        case 0x10:
+            INST_NAME("MOVSS Gx, Ex");
+            nextop = F8;
+            GETG;
+            if(MODREG) {
+                v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop & 7) + (rex.b << 3), 0);
+            } else {
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                v1 = fpu_get_scratch(dyn);
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, x2, &fixedaddress, rex, NULL, 1, 0);
+                VXOR_V(v0, v0, v0);
+                FLD_S(v1, ed, fixedaddress);
+            }
+            VEXTRINS_W(v0, v1, 0);
+            break;
+        case 0x11:
+            INST_NAME("MOVSS Ex, Gx");
+            nextop = F8;
+            GETG;
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
+            if (MODREG) {
+                q0 = sse_get_reg(dyn, ninst, x1, (nextop & 7) + (rex.b << 3), 1);
+                VEXTRINS_W(q0, v0, 0);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, x2, &fixedaddress, rex, NULL, 1, 0);
+                FST_S(v0, ed, fixedaddress);
+                SMWRITE2();
+            }
+            break;
+        case 0x1E:
+            INST_NAME("NOP / ENDBR32 / ENDBR64");
+            nextop = F8;
+            FAKEED;
+            break;
+        case 0x2A:
+            INST_NAME("CVTSI2SS Gx, Ed");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETED(0);
+            d1 = fpu_get_scratch(dyn);
+            MOVGR2FR_D(d1, ed);
+            if(rex.w) {
+                FFINT_S_L(d1, d1);
+            } else {
+                FFINT_S_W(d1, d1);
+            }
+            VEXTRINS_W(v0, d1, 0);
+            break;
+        case 0x2C:
+            INST_NAME("CVTTSS2SI Gd, Ex");
+            nextop = F8;
+            GETGD;
+            GETEXSS(d0, 0, 0);
+            if (!box64_dynarec_fastround) {
+                MOVGR2FCSR(FCSR2, xZR); // reset all bits
+            }
+            d1 = fpu_get_scratch(dyn);
+            if (rex.w) {
+                FTINTRZ_L_S(d1, d0);
+                MOVFR2GR_D(gd, d1);
+            } else {
+                FTINTRZ_W_S(d1, d0);
+                MOVFR2GR_S(gd, d1);
+            }
+            if (!rex.w) ZEROUP(gd);
+            if (!box64_dynarec_fastround) {
+                MOVFCSR2GR(x5, FCSR2); // get back FPSR to check
+                MOV32w(x3, (1 << FR_V) | (1 << FR_O));
+                AND(x5, x5, x3);
+                CBZ_NEXT(x5);
+                if (rex.w) {
+                    MOV64x(gd, 0x8000000000000000LL);
+                } else {
+                    MOV32w(gd, 0x80000000);
+                }
+            }
+            break;
+        case 0x51:
+            INST_NAME("SQRTSS Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            d1 = fpu_get_scratch(dyn);
+            GETEXSS(d0, 0, 0);
+            FSQRT_S(d1, d0);
+            VEXTRINS_W(v0, d1, 0);
+            break;
+        case 0x58:
+            INST_NAME("ADDSS Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            d1 = fpu_get_scratch(dyn);
+            GETEXSS(d0, 0, 0);
+            FADD_S(d1, v0, d0);
+            VEXTRINS_W(v0, d1, 0);
+            break;
+        case 0x59:
+            INST_NAME("MULSS Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            d1 = fpu_get_scratch(dyn);
+            GETEXSS(d0, 0, 0);
+            FMUL_S(d1, v0, d0);
+            VEXTRINS_W(v0, d1, 0);
+            break;
+        case 0x5A:
+            INST_NAME("CVTSS2SD Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEXSS(v1, 0, 0);
+            d1 = fpu_get_scratch(dyn);
+            FCVT_D_S(d1, v1);
+            VEXTRINS_D(v0, d1, 0);
+            break;
+        case 0x5C:
+            INST_NAME("SUBSS Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            d1 = fpu_get_scratch(dyn);
+            GETEXSS(d0, 0, 0);
+            FSUB_S(d1, v0, d0);
+            VEXTRINS_W(v0, d1, 0);
+            break;
+        case 0x5D:
+            INST_NAME("MINSS Gx, Ex");
+            nextop = F8;
+            GETGX(d0, 1);
+            GETEXSS(d1, 0, 0);
+            FCMP_S(fcc0, d0, d1, cUN);
+            BCNEZ_MARK(fcc0);
+            FCMP_S(fcc1, d1, d0, cLT);
+            BCEQZ_MARK2(fcc1);
+            MARK;
+            v1 = fpu_get_scratch(dyn);
+            FMOV_S(v1, d1);
+            VEXTRINS_W(d0, v1, 0);
+            MARK2;
+            break;
+        case 0x5E:
+            INST_NAME("DIVSS Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            d1 = fpu_get_scratch(dyn);
+            GETEXSS(d0, 0, 0);
+            FDIV_S(d1, v0, d0);
+            VEXTRINS_W(v0, d1, 0);
+            break;
+        case 0x5F:
+            INST_NAME("MAXSS Gx, Ex");
+            nextop = F8;
+            GETGX(d0, 1);
+            GETEXSS(d1, 0, 0);
+            FCMP_S(fcc0, d0, d1, cUN);
+            BCNEZ_MARK(fcc0);
+            FCMP_S(fcc1, d0, d1, cLT);
+            BCEQZ_MARK2(fcc1);
+            MARK;
+            v1 = fpu_get_scratch(dyn);
+            FMOV_S(v1, d1);
+            VEXTRINS_W(d0, v1, 0);
+            MARK2;
+            break;
+        case 0x6F:
+            INST_NAME("MOVDQU Gx, Ex");
+            nextop = F8;
+            if (MODREG) {
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop & 7) + (rex.b << 3), 0);
+                GETGX_empty(v0);
+                VOR_V(v0, v1, v1);
+            } else {
+                GETGX_empty(v0);
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x3, &fixedaddress, rex, NULL, 0, 0);
+                VLD(v0, ed, fixedaddress);
+            }
+            break;
+        case 0x7E:
+            INST_NAME("MOVQ Gx, Ex");
+            nextop = F8;
+            GETGX_empty(v0);
+            VXOR_V(v0, v0, v0);
+            if (MODREG) {
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop & 7) + (rex.b << 3), 0);
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, x2, &fixedaddress, rex, NULL, 1, 0);
+                v1 = fpu_get_scratch(dyn);
+                FLD_D(v1, ed, fixedaddress);
+            }
+            VEXTRINS_D(v0, v1, 0); // v0[63:0] = v1[63:0]
+            break;
+        case 0x7F:
+            INST_NAME("MOVDQU Ex,Gx");
+            nextop = F8;
+            GETGX(v0, 0);
+            if(MODREG) {
+                v1 = sse_get_reg_empty(dyn, ninst, x1, (nextop & 7) + (rex.b << 3));
+                VOR_V(v1, v0, v0);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, x2, &fixedaddress, rex, NULL, 1, 0);
+                VST(v0, ed, fixedaddress);
+                SMWRITE2();
+            }
+            break;
+        case 0xC2:
+            INST_NAME("CMPSS Gx, Ex, Ib");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEXSS(v1, 0, 1);
+            u8 = F8;
+            switch (u8 & 7) {
+                case 0: FCMP_S(fcc0, v0, v1, cEQ); break;  // Equal
+                case 1: FCMP_S(fcc0, v0, v1, cLT); break;  // Less than
+                case 2: FCMP_S(fcc0, v0, v1, cLE); break;  // Less or equal
+                case 3: FCMP_S(fcc0, v0, v1, cUN); break;  // NaN
+                case 4: FCMP_S(fcc0, v0, v1, cUNE); break; // Not Equal or unordered
+                case 5: FCMP_S(fcc0, v1, v0, cULE); break; // Greater or equal or unordered
+                case 6: FCMP_S(fcc0, v1, v0, cULT); break; // Greater or unordered, test inverted, N!=V so unordered or less than (inverted)
+                case 7: FCMP_S(fcc0, v0, v1, cOR); break;  // not NaN
+            }
+            MOVCF2GR(x2, fcc0);
+            NEG_D(x2, x2);
+            q1 = fpu_get_scratch(dyn);
+            MOVGR2FR_W(q1, x2);
+            VEXTRINS_W(v0, q1, 0);
+            break;
+        default:
+            DEFAULT;
+    }
+    return addr;
+}
@@ -0,0 +1,387 @@
+#define _GNU_SOURCE
+#include <stdio.h>
+#include <stdlib.h>
+#include <errno.h>
+#include <string.h>
+#include <math.h>
+#include <signal.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "x64run.h"
+#include "x64emu.h"
+#include "box64stack.h"
+#include "callback.h"
+#include "emu/x64run_private.h"
+#include "emu/x87emu_private.h"
+#include "x64trace.h"
+#include "signals.h"
+#include "dynarec_native.h"
+#include "dynarec_la64_private.h"
+#include "dynarec_la64_functions.h"
+#include "custommem.h"
+#include "bridge.h"
+
+#define XMM0 0
+#define XMM8 16
+#define X870 8
+#define EMM0 8
+
+// Get a FPU scratch reg
+int fpu_get_scratch(dynarec_la64_t* dyn)
+{
+    return SCRATCH0 + dyn->lsx.fpu_scratch++; // return an Sx
+}
+// Reset scratch regs counter
+void fpu_reset_scratch(dynarec_la64_t* dyn)
+{
+    dyn->lsx.fpu_scratch = 0;
+}
+
+// Free a FPU double reg
+void fpu_free_reg(dynarec_la64_t* dyn, int reg)
+{
+    // TODO: check upper limit?
+    dyn->lsx.fpuused[reg] = 0;
+    if (dyn->lsx.lsxcache[reg].t != LSX_CACHE_ST_F && dyn->lsx.lsxcache[reg].t != LSX_CACHE_ST_D && dyn->lsx.lsxcache[reg].t != LSX_CACHE_ST_I64)
+        dyn->lsx.lsxcache[reg].v = 0;
+}
+
+// Get an XMM quad reg
+int fpu_get_reg_xmm(dynarec_la64_t* dyn, int t, int xmm)
+{
+    int i;
+    if (xmm > 7) {
+        i = XMM8 + xmm - 8;
+    } else {
+        i = XMM0 + xmm;
+    }
+    dyn->lsx.fpuused[i] = 1;
+    dyn->lsx.lsxcache[i].t = t;
+    dyn->lsx.lsxcache[i].n = xmm;
+    dyn->lsx.news |= (1 << i);
+    return i;
+}
+
+// Reset fpu regs counter
+static void fpu_reset_reg_lsxcache(lsxcache_t* lsx)
+{
+    lsx->fpu_reg = 0;
+    for (int i = 0; i < 24; ++i) {
+        lsx->fpuused[i] = 0;
+        lsx->lsxcache[i].v = 0;
+    }
+}
+
+void fpu_reset_reg(dynarec_la64_t* dyn)
+{
+    fpu_reset_reg_lsxcache(&dyn->lsx);
+}
+
+static int isCacheEmpty(dynarec_native_t* dyn, int ninst)
+{
+    if (dyn->insts[ninst].lsx.stack_next) {
+        return 0;
+    }
+    for (int i = 0; i < 24; ++i)
+        if (dyn->insts[ninst].lsx.lsxcache[i].v) { // there is something at ninst for i
+            if (!(
+                    (dyn->insts[ninst].lsx.lsxcache[i].t == LSX_CACHE_ST_F
+                        || dyn->insts[ninst].lsx.lsxcache[i].t == LSX_CACHE_ST_D
+                        || dyn->insts[ninst].lsx.lsxcache[i].t == LSX_CACHE_ST_I64)
+                    && dyn->insts[ninst].lsx.lsxcache[i].n < dyn->insts[ninst].lsx.stack_pop))
+                return 0;
+        }
+    return 1;
+}
+
+int fpuCacheNeedsTransform(dynarec_la64_t* dyn, int ninst)
+{
+    int i2 = dyn->insts[ninst].x64.jmp_insts;
+    if (i2 < 0)
+        return 1;
+    if ((dyn->insts[i2].x64.barrier & BARRIER_FLOAT))
+        // if the barrier as already been apply, no transform needed
+        return ((dyn->insts[ninst].x64.barrier & BARRIER_FLOAT)) ? 0 : (isCacheEmpty(dyn, ninst) ? 0 : 1);
+    int ret = 0;
+    if (!i2) { // just purge
+        if (dyn->insts[ninst].lsx.stack_next) {
+            return 1;
+        }
+        for (int i = 0; i < 24 && !ret; ++i)
+            if (dyn->insts[ninst].lsx.lsxcache[i].v) { // there is something at ninst for i
+                if (!(
+                        (dyn->insts[ninst].lsx.lsxcache[i].t == LSX_CACHE_ST_F
+                            || dyn->insts[ninst].lsx.lsxcache[i].t == LSX_CACHE_ST_D
+                            || dyn->insts[ninst].lsx.lsxcache[i].t == LSX_CACHE_ST_I64)
+                        && dyn->insts[ninst].lsx.lsxcache[i].n < dyn->insts[ninst].lsx.stack_pop))
+                    ret = 1;
+            }
+        return ret;
+    }
+    // Check if ninst can be compatible to i2
+    if (dyn->insts[ninst].lsx.stack_next != dyn->insts[i2].lsx.stack - dyn->insts[i2].lsx.stack_push) {
+        return 1;
+    }
+    lsxcache_t cache_i2 = dyn->insts[i2].lsx;
+    lsxcacheUnwind(&cache_i2);
+
+    for (int i = 0; i < 24; ++i) {
+        if (dyn->insts[ninst].lsx.lsxcache[i].v) { // there is something at ninst for i
+            if (!cache_i2.lsxcache[i].v) {         // but there is nothing at i2 for i
+                ret = 1;
+            } else if (dyn->insts[ninst].lsx.lsxcache[i].v != cache_i2.lsxcache[i].v) { // there is something different
+                if (dyn->insts[ninst].lsx.lsxcache[i].n != cache_i2.lsxcache[i].n) {    // not the same x64 reg
+                    ret = 1;
+                } else if (dyn->insts[ninst].lsx.lsxcache[i].t == LSX_CACHE_XMMR && cache_i2.lsxcache[i].t == LSX_CACHE_XMMW) { /* nothing */
+                } else
+                    ret = 1;
+            }
+        } else if (cache_i2.lsxcache[i].v)
+            ret = 1;
+    }
+    return ret;
+}
+
+void lsxcacheUnwind(lsxcache_t* cache)
+{
+    if (cache->swapped) {
+        // unswap
+        int a = -1;
+        int b = -1;
+        for (int j = 0; j < 24 && ((a == -1) || (b == -1)); ++j)
+            if ((cache->lsxcache[j].t == LSX_CACHE_ST_D || cache->lsxcache[j].t == LSX_CACHE_ST_F || cache->lsxcache[j].t == LSX_CACHE_ST_I64)) {
+                if (cache->lsxcache[j].n == cache->combined1)
+                    a = j;
+                else if (cache->lsxcache[j].n == cache->combined2)
+                    b = j;
+            }
+        if (a != -1 && b != -1) {
+            int tmp = cache->lsxcache[a].n;
+            cache->lsxcache[a].n = cache->lsxcache[b].n;
+            cache->lsxcache[b].n = tmp;
+        }
+        cache->swapped = 0;
+        cache->combined1 = cache->combined2 = 0;
+    }
+    if (cache->news) {
+        // reove the newly created lsxcache
+        for (int i = 0; i < 24; ++i)
+            if (cache->news & (1 << i))
+                cache->lsxcache[i].v = 0;
+        cache->news = 0;
+    }
+    if (cache->stack_push) {
+        // unpush
+        for (int j = 0; j < 24; ++j) {
+            if ((cache->lsxcache[j].t == LSX_CACHE_ST_D || cache->lsxcache[j].t == LSX_CACHE_ST_F || cache->lsxcache[j].t == LSX_CACHE_ST_I64)) {
+                if (cache->lsxcache[j].n < cache->stack_push)
+                    cache->lsxcache[j].v = 0;
+                else
+                    cache->lsxcache[j].n -= cache->stack_push;
+            }
+        }
+        cache->x87stack -= cache->stack_push;
+        cache->stack -= cache->stack_push;
+        cache->stack_push = 0;
+    }
+    cache->x87stack += cache->stack_pop;
+    cache->stack_next = cache->stack;
+    cache->stack_pop = 0;
+    cache->barrier = 0;
+    // And now, rebuild the x87cache info with lsxcache
+    cache->mmxcount = 0;
+    cache->fpu_scratch = 0;
+    cache->fpu_extra_qscratch = 0;
+    cache->fpu_reg = 0;
+    for (int i = 0; i < 8; ++i) {
+        cache->x87cache[i] = -1;
+        cache->mmxcache[i] = -1;
+        cache->x87reg[i] = 0;
+        cache->ssecache[i * 2].v = -1;
+        cache->ssecache[i * 2 + 1].v = -1;
+    }
+    int x87reg = 0;
+    for (int i = 0; i < 24; ++i) {
+        if (cache->lsxcache[i].v) {
+            cache->fpuused[i] = 1;
+            switch (cache->lsxcache[i].t) {
+                case LSX_CACHE_MM:
+                    cache->mmxcache[cache->lsxcache[i].n] = i;
+                    ++cache->mmxcount;
+                    ++cache->fpu_reg;
+                    break;
+                case LSX_CACHE_XMMR:
+                case LSX_CACHE_XMMW:
+                    cache->ssecache[cache->lsxcache[i].n].reg = i;
+                    cache->ssecache[cache->lsxcache[i].n].write = (cache->lsxcache[i].t == LSX_CACHE_XMMW) ? 1 : 0;
+                    ++cache->fpu_reg;
+                    break;
+                case LSX_CACHE_ST_F:
+                case LSX_CACHE_ST_D:
+                case LSX_CACHE_ST_I64:
+                    cache->x87cache[x87reg] = cache->lsxcache[i].n;
+                    cache->x87reg[x87reg] = i;
+                    ++x87reg;
+                    ++cache->fpu_reg;
+                    break;
+                case LSX_CACHE_SCR:
+                    cache->fpuused[i] = 0;
+                    cache->lsxcache[i].v = 0;
+                    break;
+            }
+        } else {
+            cache->fpuused[i] = 0;
+        }
+    }
+}
+
+const char* getCacheName(int t, int n)
+{
+    static char buff[20];
+    switch (t) {
+        case LSX_CACHE_ST_D: sprintf(buff, "ST%d", n); break;
+        case LSX_CACHE_ST_F: sprintf(buff, "st%d", n); break;
+        case LSX_CACHE_ST_I64: sprintf(buff, "STi%d", n); break;
+        case LSX_CACHE_MM: sprintf(buff, "MM%d", n); break;
+        case LSX_CACHE_XMMW: sprintf(buff, "XMM%d", n); break;
+        case LSX_CACHE_XMMR: sprintf(buff, "xmm%d", n); break;
+        case LSX_CACHE_SCR: sprintf(buff, "Scratch"); break;
+        case LSX_CACHE_NONE: buff[0] = '\0'; break;
+    }
+    return buff;
+}
+
+void inst_name_pass3(dynarec_native_t* dyn, int ninst, const char* name, rex_t rex)
+{
+    if (box64_dynarec_dump) {
+        printf_x64_instruction(rex.is32bits ? my_context->dec32 : my_context->dec, &dyn->insts[ninst].x64, name);
+        dynarec_log(LOG_NONE, "%s%p: %d emitted opcodes, inst=%d, barrier=%d state=%d/%d(%d), %s=%X/%X, use=%X, need=%X/%X, sm=%d/%d",
+            (box64_dynarec_dump > 1) ? "\e[32m" : "",
+            (void*)(dyn->native_start + dyn->insts[ninst].address),
+            dyn->insts[ninst].size / 4,
+            ninst,
+            dyn->insts[ninst].x64.barrier,
+            dyn->insts[ninst].x64.state_flags,
+            dyn->f.pending,
+            dyn->f.dfnone,
+            dyn->insts[ninst].x64.may_set ? "may" : "set",
+            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->smread, dyn->smwrite);
+        if (dyn->insts[ninst].pred_sz) {
+            dynarec_log(LOG_NONE, ", pred=");
+            for (int ii = 0; ii < dyn->insts[ninst].pred_sz; ++ii)
+                dynarec_log(LOG_NONE, "%s%d", ii ? "/" : "", dyn->insts[ninst].pred[ii]);
+        }
+        if (dyn->insts[ninst].x64.jmp && dyn->insts[ninst].x64.jmp_insts >= 0)
+            dynarec_log(LOG_NONE, ", jmp=%d", dyn->insts[ninst].x64.jmp_insts);
+        if (dyn->insts[ninst].x64.jmp && dyn->insts[ninst].x64.jmp_insts == -1)
+            dynarec_log(LOG_NONE, ", jmp=out");
+        if (dyn->last_ip)
+            dynarec_log(LOG_NONE, ", last_ip=%p", (void*)dyn->last_ip);
+        for (int ii = 0; ii < 24; ++ii) {
+            switch (dyn->insts[ninst].lsx.lsxcache[ii].t) {
+                case LSX_CACHE_ST_D: dynarec_log(LOG_NONE, " D%d:%s", ii, getCacheName(dyn->insts[ninst].lsx.lsxcache[ii].t, dyn->insts[ninst].lsx.lsxcache[ii].n)); break;
+                case LSX_CACHE_ST_F: dynarec_log(LOG_NONE, " S%d:%s", ii, getCacheName(dyn->insts[ninst].lsx.lsxcache[ii].t, dyn->insts[ninst].lsx.lsxcache[ii].n)); break;
+                case LSX_CACHE_ST_I64: dynarec_log(LOG_NONE, " D%d:%s", ii, getCacheName(dyn->insts[ninst].lsx.lsxcache[ii].t, dyn->insts[ninst].lsx.lsxcache[ii].n)); break;
+                case LSX_CACHE_MM: dynarec_log(LOG_NONE, " D%d:%s", ii, getCacheName(dyn->insts[ninst].lsx.lsxcache[ii].t, dyn->insts[ninst].lsx.lsxcache[ii].n)); break;
+                case LSX_CACHE_XMMW: dynarec_log(LOG_NONE, " Q%d:%s", ii, getCacheName(dyn->insts[ninst].lsx.lsxcache[ii].t, dyn->insts[ninst].lsx.lsxcache[ii].n)); break;
+                case LSX_CACHE_XMMR: dynarec_log(LOG_NONE, " Q%d:%s", ii, getCacheName(dyn->insts[ninst].lsx.lsxcache[ii].t, dyn->insts[ninst].lsx.lsxcache[ii].n)); break;
+                case LSX_CACHE_SCR: dynarec_log(LOG_NONE, " D%d:%s", ii, getCacheName(dyn->insts[ninst].lsx.lsxcache[ii].t, dyn->insts[ninst].lsx.lsxcache[ii].n)); break;
+                case LSX_CACHE_NONE:
+                default: break;
+            }
+        }
+        if (dyn->lsx.stack || dyn->insts[ninst].lsx.stack_next || dyn->insts[ninst].lsx.x87stack)
+            dynarec_log(LOG_NONE, " X87:%d/%d(+%d/-%d)%d", dyn->lsx.stack, dyn->insts[ninst].lsx.stack_next, dyn->insts[ninst].lsx.stack_push, dyn->insts[ninst].lsx.stack_pop, dyn->insts[ninst].lsx.x87stack);
+        if (dyn->insts[ninst].lsx.combined1 || dyn->insts[ninst].lsx.combined2)
+            dynarec_log(LOG_NONE, " %s:%d/%d", dyn->insts[ninst].lsx.swapped ? "SWP" : "CMB", dyn->insts[ninst].lsx.combined1, dyn->insts[ninst].lsx.combined2);
+        dynarec_log(LOG_NONE, "%s\n", (box64_dynarec_dump > 1) ? "\e[m" : "");
+    }
+}
+
+// will go badly if address is unaligned
+static uint8_t extract_byte(uint32_t val, void* address)
+{
+    int idx = (((uintptr_t)address)&3)*8;
+    return (val>>idx)&0xff;
+}
+
+static uint32_t insert_byte(uint32_t val, uint8_t b, void* address)
+{
+    int idx = (((uintptr_t)address)&3)*8;
+    val&=~(0xff<<idx);
+    val|=(((uint32_t)b)<<idx);
+    return val;
+}
+
+static uint16_t extract_half(uint32_t val, void* address)
+{
+    int idx = (((uintptr_t)address)&3)*8;
+    return (val>>idx)&0xffff;
+}
+
+static uint32_t insert_half(uint32_t val, uint16_t h, void* address)
+{
+    int idx = (((uintptr_t)address)&3)*8;
+    val&=~(0xffff<<idx);
+    val|=(((uint32_t)h)<<idx);
+    return val;
+}
+
+uint8_t la64_lock_xchg_b_slow(void* addr, uint8_t val)
+{
+    uint32_t ret;
+    uint32_t* aligned = (uint32_t*)(((uintptr_t)addr)&~3);
+    do {
+        ret = *aligned;
+    } while(la64_lock_cas_d(aligned, ret, insert_byte(ret, val, addr)));
+    return extract_byte(ret, addr);
+}
+
+int la64_lock_cas_b_slow(void* addr, uint8_t ref, uint8_t val)
+{
+    uint32_t* aligned = (uint32_t*)(((uintptr_t)addr)&~3);
+    uint32_t tmp = *aligned;
+    return la64_lock_cas_d(aligned, ref, insert_byte(tmp, val, addr));
+}
+
+int la64_lock_cas_h_slow(void* addr, uint16_t ref, uint16_t val)
+{
+    uint32_t* aligned = (uint32_t*)(((uintptr_t)addr)&~3);
+    uint32_t tmp = *aligned;
+    return la64_lock_cas_d(aligned, ref, insert_half(tmp, val, addr));
+}
+
+void print_opcode(dynarec_native_t* dyn, int ninst, uint32_t opcode)
+{
+    dynarec_log(LOG_NONE, "\t%08x\t%s\n", opcode, la64_print(opcode, (uintptr_t)dyn->block));
+}
+
+static void sse_reset(lsxcache_t* lsx)
+{
+    for (int i = 0; i < 16; ++i)
+        lsx->ssecache[i].v = -1;
+}
+
+void fpu_reset(dynarec_la64_t* dyn)
+{
+    // TODO: x87 and mmx
+    sse_reset(&dyn->lsx);
+    fpu_reset_reg(dyn);
+}
+
+void fpu_reset_ninst(dynarec_la64_t* dyn, int ninst)
+{
+    // TODO: x87 and mmx
+    sse_reset(&dyn->insts[ninst].lsx);
+    fpu_reset_reg_lsxcache(&dyn->insts[ninst].lsx);
+}
@@ -0,0 +1,37 @@
+#ifndef __DYNAREC_LA64_FUNCTIONS_H__
+#define __DYNAREC_LA64_FUNCTIONS_H__
+
+#include "../dynarec_native_functions.h"
+
+#define SCRATCH0 24
+
+typedef struct x64emu_s x64emu_t;
+typedef struct dynarec_la64_s dynarec_la64_t;
+
+// Get an FPU scratch reg
+int fpu_get_scratch(dynarec_la64_t* dyn);
+// Reset scratch regs counter
+void fpu_reset_scratch(dynarec_la64_t* dyn);
+// Get an XMM quad reg
+int fpu_get_reg_xmm(dynarec_la64_t* dyn, int t, int xmm);
+// Free a FPU/MMX/XMM reg
+void fpu_free_reg(dynarec_la64_t* dyn, int reg);
+// Reset fpu regs counter
+void fpu_reset_reg(dynarec_la64_t* dyn);
+
+// FPU Cache transformation (for loops) // Specific, need to be written by backend
+int fpuCacheNeedsTransform(dynarec_la64_t* dyn, int ninst);
+
+// Undo the changes of a lsxcache to get the status before the instruction
+void lsxcacheUnwind(lsxcache_t* cache);
+
+const char* getCacheName(int t, int n);
+
+void inst_name_pass3(dynarec_native_t* dyn, int ninst, const char* name, rex_t rex);
+void print_opcode(dynarec_native_t* dyn, int ninst, uint32_t opcode);
+
+// reset the cache
+void fpu_reset(dynarec_native_t* dyn);
+void fpu_reset_ninst(dynarec_native_t* dyn, int ninst);
+
+#endif //__DYNAREC_LA64_FUNCTIONS_H__
@@ -0,0 +1,13 @@
+#include <stdint.h>
+
+#include "la64_emitter.h"
+
+#define EMIT(A) *block = (A); ++block;
+void CreateJmpNext(void* addr, void* next)
+{
+    uint32_t* block = (uint32_t*)addr;
+    uintptr_t diff = (intptr_t)next - (intptr_t)addr;
+    PCADDU12I(x2, SPLIT20(diff));
+    LD_D(x2, x2, SPLIT12(diff));
+    BR(x2);
+}
@@ -0,0 +1,54 @@
+#define INIT uintptr_t sav_addr = addr
+#define FINI                           \
+    dyn->isize = addr - sav_addr;      \
+    dyn->insts[ninst].x64.addr = addr; \
+    if (ninst) dyn->insts[ninst - 1].x64.size = dyn->insts[ninst].x64.addr - dyn->insts[ninst - 1].x64.addr
+
+#define MESSAGE(A, ...)
+#define MAYSETFLAGS() dyn->insts[ninst].x64.may_set = 1
+#define READFLAGS(A)                     \
+    dyn->insts[ninst].x64.use_flags = A; \
+    dyn->f.dfnone = 1;                   \
+    dyn->f.pending = SF_SET
+#define SETFLAGS(A, B)                     \
+    dyn->insts[ninst].x64.set_flags = A;   \
+    dyn->insts[ninst].x64.state_flags = B; \
+    dyn->f.pending = (B) & SF_SET_PENDING; \
+    dyn->f.dfnone = ((B) & SF_SET) ? 1 : 0;
+#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 SET_HASCALLRET() dyn->insts[ninst].x64.has_callret = 1
+#define NEW_INST                                 \
+    ++dyn->size;                                 \
+    dyn->insts[ninst].x64.addr = ip;             \
+    dyn->lsx.combined1 = dyn->lsx.combined2 = 0; \
+    dyn->lsx.swapped = 0;                        \
+    dyn->lsx.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; }
+#define INST_EPILOG                    \
+    dyn->insts[ninst].f_exit = dyn->f; \
+    dyn->insts[ninst].lsx = dyn->lsx;  \
+    dyn->insts[ninst].x64.has_next = (ok > 0) ? 1 : 0;
+#define INST_NAME(name)
+#define DEFAULT                                                                                                                                     \
+    --dyn->size;                                                                                                                                    \
+    *ok = -1;                                                                                                                                       \
+    if (box64_dynarec_log >= LOG_INFO || box64_dynarec_dump || box64_dynarec_missing) {                                                             \
+        dynarec_log(LOG_NONE, "%p: Dynarec stopped because of %sOpcode %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X", \
+            (void*)ip, rex.is32bits ? "32bits " : "",                                                                                               \
+            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(LOG_NONE, "\n");                                                                                                                \
+    }
@@ -0,0 +1,15 @@
+#define INIT
+#define FINI
+#define MESSAGE(A, ...)
+#define EMIT(A)
+#define NEW_INST                                 \
+    dyn->insts[ninst].f_entry = dyn->f;          \
+    dyn->lsx.combined1 = dyn->lsx.combined2 = 0; \
+    dyn->lsx.swapped = 0;                        \
+    dyn->lsx.barrier = 0
+
+#define INST_EPILOG                   \
+    dyn->insts[ninst].lsx = dyn->lsx; \
+    dyn->insts[ninst].f_exit = dyn->f
+
+#define INST_NAME(name)
@@ -0,0 +1,26 @@
+#define INIT dyn->native_size = 0
+#define FINI                                                                                                                                                   \
+    if (ninst) {                                                                                                                                               \
+        dyn->insts[ninst].address = (dyn->insts[ninst - 1].address + dyn->insts[ninst - 1].size);                                                              \
+        dyn->insts_size += 1 + ((dyn->insts[ninst].x64.size > (dyn->insts[ninst].size / 4)) ? dyn->insts[ninst].x64.size : (dyn->insts[ninst].size / 4)) / 15; \
+    }
+
+#define MESSAGE(A, ...)
+#define EMIT(A)                      \
+    do {                             \
+        dyn->insts[ninst].size += 4; \
+        dyn->native_size += 4;       \
+    } while (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; \
+    }
+#define INST_EPILOG dyn->insts[ninst].epilog = dyn->native_size;
+#define INST_NAME(name)
+#define TABLE64(A, V)         \
+    {                         \
+        Table64(dyn, (V), 2); \
+        EMIT(0);              \
+        EMIT(0);              \
+    }
@@ -0,0 +1,29 @@
+#define INIT
+#define FINI                                                                                              \
+    if (ninst)                                                                                            \
+        addInst(dyn->instsize, &dyn->insts_size, dyn->insts[ninst].x64.size, dyn->insts[ninst].size / 4); \
+    addInst(dyn->instsize, &dyn->insts_size, 0, 0);
+#define EMIT(A)                                                          \
+    do {                                                                 \
+        if (box64_dynarec_dump) print_opcode(dyn, ninst, (uint32_t)(A)); \
+        if ((uintptr_t)dyn->block < dyn->tablestart)                     \
+            *(uint32_t*)(dyn->block) = (uint32_t)(A);                    \
+        dyn->block += 4;                                                 \
+        dyn->native_size += 4;                                           \
+        dyn->insts[ninst].size2 += 4;                                    \
+    } while (0)
+
+#define MESSAGE(A, ...) \
+    if (box64_dynarec_dump) dynarec_log(LOG_NONE, __VA_ARGS__)
+#define NEW_INST \
+    if (ninst)   \
+        addInst(dyn->instsize, &dyn->insts_size, dyn->insts[ninst - 1].x64.size, dyn->insts[ninst - 1].size / 4);
+#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)); \
+        PCADDU12I(A, SPLIT20(val64offset));           \
+        LD_D(A, A, SPLIT12(val64offset));             \
+    }