Fix typos in comments and documentation (#1072 )

Correct "adressing", "constains", "informations", "Aleady" and "interpretor".
[WRAPPER] Added 1 wrapped function to libxml2 (for #1061 )
2026-11-07 17:43:14 -10:00 · 2026-10-20 15:45:11 +02:00 · 2026-09-27 08:26:24 +02:00 · 2026-09-11 09:45:03 +02:00 · 2026-09-11 09:44:05 +02:00 · 2026-09-11 09:42:51 +02:00
948 changed files with 46278 additions and 18039 deletions
@@ -14,7 +14,7 @@ on:
        - GENERIC_ARM
        - A64
        - GAMESHELL
-        - ODROID
+        - ODROIDXU4
        - PANDORA
        - PHYTIUM
        - PYRA
@@ -23,6 +23,9 @@ on:
        - RPI4
        - RPI4ARM64
        - SD845
+        - SD888
+        - ADLINK
+        - PHYTIUM
      build_type:
        description: 'Build type'
        required: true
@@ -27,28 +27,28 @@ jobs:
    strategy:
      fail-fast: false
      matrix:
-        platform: [I386, OTHER_ARM, PANDORA, PYRA, GAMESHELL, RPI4, RPI4ARM64, ODROID, RK3399, RK3288, A64, SD845, PHYTIUM, ANDROID]
+        platform: [I386, OTHER_ARM, PANDORA, PYRA, GAMESHELL, RPI4, RPI4ARM64, ODROIDXU4, RK3399, RK3288, A64, SD845, SD888, ADLINK, PHYTIUM, ANDROID]
        type: [Release, Trace]
        os: [ubuntu-latest]
        
        include:
          - platform: TEGRAX1
            type: Release
-            os: ubuntu-18.04
+            os: ubuntu-22.04
          - platform: TEGRAX1
            type: Trace
-            os: ubuntu-18.04
+            os: ubuntu-22.04
          - platform: TEGRAX1ARM32
            type: Release
-            os: ubuntu-18.04
+            os: ubuntu-22.04
          - platform: TEGRAX1ARM32
            type: Trace
-            os: ubuntu-18.04
+            os: ubuntu-22.04

    runs-on: ${{ matrix.os }}
    steps:
      - name: "Checkout Box86 Repository"
-        uses: actions/checkout@v2
+        uses: actions/checkout@v3

      - name: "Environment preparation"
        run: |
@@ -57,10 +57,10 @@ jobs:
          if [[ ${{ matrix.platform }} != 'I386' && ${{ matrix.platform }} != 'OTHER_ARM' ]]; then
            sudo apt-get -y install git cmake make python3
            if [[ ${{ matrix.platform }} == 'ANDROID' ]]; then
-              wget https://dl.google.com/android/repository/android-ndk-r23b-linux.zip
-              unzip android-ndk-r23b-linux.zip
-              echo "BOX86_COMPILER=$PWD/android-ndk-r23b/toolchains/llvm/prebuilt/linux-x86_64/bin/armv7a-linux-androideabi31-clang" >> $GITHUB_ENV
-              echo "BOX86_PLATFORM_MARCRO=-DANDROID=1 -DNOLOADADDR=1" >> $GITHUB_ENV
+              wget https://dl.google.com/android/repository/android-ndk-r25b-linux.zip
+              unzip android-ndk-r25b-linux.zip
+              echo "BOX86_COMPILER=$PWD/android-ndk-r25b/toolchains/llvm/prebuilt/linux-x86_64/bin/armv7a-linux-androideabi31-clang" >> $GITHUB_ENV
+              echo "BOX86_PLATFORM_MARCRO=-DANDROID=1 -DARM_DYNAREC=ON -DBAD_SIGNAL=1" >> $GITHUB_ENV
            else
              sudo apt-get -y install gcc-arm-linux-gnueabihf
              echo "BOX86_COMPILER=arm-linux-gnueabihf-gcc" >> $GITHUB_ENV
@@ -111,7 +111,7 @@ jobs:
          ctest -j$(nproc)

      - name: "Upload Artifact"
-        uses: actions/upload-artifact@v2
+        uses: actions/upload-artifact@v4
        with:
          name: box86-${{ matrix.platform }}-${{ matrix.type }}
          path: build/box86
@@ -5,4 +5,5 @@ build*/
 src/git_head.h
 backup/
 box86_update.sh
-restore_backup.sh
+restore_backup.sh
+.cache
@@ -5,33 +5,54 @@ option(PYRA "Set to ON if targeting an OpenPandora device" ${PYRA})
 option(RPI2 "Set to ON if targeting an RaspberryPI2 device" ${RPI2})
 option(RPI3 "Set to ON if targeting an RaspberryPI3 device" ${RPI3})
 option(RPI4 "Set to ON if targeting an RaspberryPI4 device" ${RPI4})
+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(GOA_CLONE "Set to ON if targeting GO Advance clones, like RG351p/v, Gameforce Chi, RGB10..." ${GOA_CLONE})
 option(RK3288 "Set to ON if targeting an Rockchip RK3288 based device" ${RK3288})
 option(RK3399 "Set to ON if targeting an Rockchip RK3399 based device" ${RK3399})
-option(ODROIDN2 "Set to ON if targeting an Odroid-N2 device" ${ODROIDN2})
+option(RK3588 "Set to ON if targeting an Rockchip RK3588 / RK3588S 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 (ARM64) based device" ${TEGRAX1})
 option(TEGRAX1ARM32 "Set to ON if targeting an Tegra X1 (ARM32) based device" ${TEGRAX1ARM32})
+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(GAMESHELL "Set to ON if targeting a GameShell device" ${GAMESHELL})
-option(ODROID "Set to ON if targeting an ODroid device" ${ODROID})
+option(ODROIDXU4 "Set to ON if targeting an ODROID-XU4 device" ${ODROIDXU4})
 option(POWER9 "Set to ON if targeting a POWER9 processor" ${POWER9})
-option(SD845 "Set to ON if targeting a Snapragon 845 based device" ${SD845})
+option(SD845 "Set to ON if targeting a Snapdragon 845 based device" ${SD845})
+option(SD865 "Set to ON if targeting a Snapdragon 865 based device" ${SD865})
+option(SD888 "Set to ON if targeting a Snapdragon 888 based device" ${SD888})
+option(ADLINK "Set to ON if targeting an ADLink AmpereAltra based device" ${ADLINK})
 option(A64 "Set to ON if targeting an Allwinner A64 based device" ${A64})
+option(ARM64 "Set to ON if targeting a generic ARM64 based device" ${ARM64})
+option(BAD_SIGNAL "Set to ON to activate the workaround for incoherent si_info on SIGSEGV" ${BAD_SIGNAL})
 option(LX2160A "Set to ON if targeting an LX2160A based device" ${LX2160A})
 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(NOLOADADDR "Set to ON to avoid fixing the load address of Box86" ${NO_LOADAADR})
 option(NOGETCLOCK "Set to ON to avoid using clock_gettime with CLOCK_MONOTONIC_COARSE for RDTSC opcode (use gltimeofday instead)" ${NOGETCLOCK})
 option(NOGIT "Set to ON if not building from a git clone repo (like when building from a zip download from github)" ${NOGIT})
-if(PANDORA OR PYRA OR RPI2 OR RPI3 OR RPI4 OR GAMESHELL OR ODROID OR GOA_CLONE OR RK3288 OR RK3399 OR ODROIDN2 OR TEGRAX1 OR TEGRAX1ARM32 OR RPI4ARM64 OR PHYTIUM OR SD845 OR A64 OR LX2160A)
+option(NO_LIB_INSTALL "Set ON to not install a few x86 libs that are used by many program" ${NO_LIB_INSTALL})
+option(NO_CONF_INSTALL "Set ON to not install config files" ${NO_CONF_INSTALL})
+
+if(PANDORA OR PYRA OR RPI2 OR RPI3 OR RPI4 OR GAMESHELL OR ODROIDXU4 OR GOA_CLONE OR RK3288 OR RK3399 OR ODROIDN2 OR TEGRAX1 OR TEGRAX1ARM32 OR TEGRA_T194 OR TEGRA_T234 OR RPI3ARM64 OR RPI4ARM64 OR PHYTIUM OR SD845 OR SD865 OR SD888 OR ADLINK OR A64 OR LX2160A OR ARM64 OR RK3588)
    set(LD80BITS OFF CACHE BOOL "")
    set(NOALIGN OFF CACHE BOOL "")
    set(ARM_DYNAREC ON CACHE BOOL "")
+    if(RK3588)
+        set(BAD_SIGNAL ON CACHE BOOL "")
+    endif()
+endif()
+if(TERMUX)
+    set(ANDROID ON CACHE BOOL "")
+    set(TERMUX_PATH "/data/data/com.termux/files")
 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})
@@ -56,21 +77,29 @@ else()
 endif()

 cmake_host_system_information(RESULT CMAKE_HOST_SYSTEM_PROCESSOR QUERY OS_PLATFORM)
-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}" "armhf"  _armhf)

-if(_aarch64 AND (RK3399 OR ODROIDN2 OR TEGRAX1 OR RPI4ARM64 OR PHYTIUM OR SD845 OR A64 OR LX2160A))
-    set(CMAKE_C_COMPILER arm-linux-gnueabihf-gcc)
-    #set(CMAKE_ASM_COMPILER arm-linux-gnueabihf-as) #will use gcc in fact
-    set(CMAKE_C_COMPILER_TARGET arm-linux-gnueabihf)
+if("${CMAKE_HOST_SYSTEM_PROCESSOR}" STREQUAL "aarch64" AND (RK3399 OR ODROIDN2 OR TEGRAX1 OR TEGRA_T194 OR TEGRA_T234 OR RPI3ARM64 OR RPI4ARM64 OR PHYTIUM OR SD845 OR SD888 OR ADLINK OR A64 OR LX2160A OR RK3588 OR ARM64))
+    set(COMPILER_TRIPLE arm-linux-gnueabihf-gcc)
+    find_program(COMPILER_FOUND ${COMPILER_TRIPLE})
+    
+    if(COMPILER_FOUND)
+        set(CMAKE_C_COMPILER ${COMPILER_FOUND})
+    else()
+        message(STATUS "Cross compiler ${COMPILER_TRIPLE} not found, using CC or -DCMAKE_C_COMPILER")
+    endif()
 endif()

 project(box86 C ASM)

 enable_testing()

+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}" "armhf"  _armhf)
+string(COMPARE EQUAL "${CMAKE_SYSTEM_PROCESSOR}" "arm"  _arm)
+string(COMPARE EQUAL "${CMAKE_SYSTEM_PROCESSOR}" "armv8l"  _armv8l)
+
 set(default_build_type "RelwithDebInfo")
 if(NOT CMAKE_BUILD_TYPE AND NOT CMAKE_CONFIGURATION_TYPES)
  message(STATUS "Setting build type to '${default_build_type}' as none was specified.")
@@ -81,6 +110,7 @@ endif()
 if(PANDORA)
    add_definitions(-DPANDORA)
    #add_definitions(-mcpu=cortex-a8 -mfpu=neon -mfloat-abi=softfp -ftree-vectorize -fsingle-precision-constant -ffast-math)
+    set(CMAKE_ASM_FLAGS  "-marm -mcpu=cortex-a8 -mfpu=neon -mfloat-abi=softfp")
 elseif(GAMESHELL)
    add_definitions(-DGAMESHELL)
    add_definitions(-marm -mcpu=cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard)
@@ -104,15 +134,20 @@ elseif(RPI4)
    add_definitions(-DRPI4)
    add_definitions(-marm -march=armv8-a+crc -mtune=cortex-a72 -mfpu=neon-fp-armv8 -mfloat-abi=hard)
    set(CMAKE_ASM_FLAGS  "-marm -march=armv8-a+crc -mtune=cortex-a72 -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(RPI3ARM64)
+    add_definitions(-DRPI)
+    add_definitions(-DRPI3ARM64)
+    add_definitions(-marm -march=armv8-a+crc -mtune=cortex-a53 -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-marm -march=armv8-a+crc -mtune=cortex-a53 -mfpu=neon-fp-armv8 -mfloat-abi=hard")
 elseif(RPI4ARM64)
    add_definitions(-DRPI)
    add_definitions(-DRPI4ARM64)
    add_definitions(-marm -march=armv8-a+crc -mtune=cortex-a72 -mfpu=neon-fp-armv8 -mfloat-abi=hard)
    set(CMAKE_ASM_FLAGS  "-marm -march=armv8-a+crc -mtune=cortex-a72 -mfpu=neon-fp-armv8 -mfloat-abi=hard")
-elseif(ODROID)
-    add_definitions(-DODROID)
-    add_definitions(-mcpu=cortex-a9 -mfpu=neon-vfpv4 -mfloat-abi=hard -marm)
-    set(CMAKE_ASM_FLAGS  "-marm -mcpu=cortex-a9 -mfpu=neon-vfpv4 -mfloat-abi=hard")
+elseif(ODROIDXU4)
+    add_definitions(-DODROIDXU4)
+    add_definitions(-marm -mcpu=cortex-a15.cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-marm -mcpu=cortex-a15.cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard")
 elseif(RK3288)
    add_definitions(-DRK3288)
    add_definitions(-mcpu=cortex-a17 -mfpu=neon -mfloat-abi=hard -marm)
@@ -121,6 +156,19 @@ elseif(SD845)
    add_definitions(-DSD845)
    add_definitions(-marm -march=armv8.2-a+simd+crypto -mtune=cortex-a75.cortex-a55 -mfpu=neon-fp-armv8 -mfloat-abi=hard)
    set(CMAKE_ASM_FLAGS  "-marm -march=armv8.2-a+simd+crypto -mtune=cortex-a75.cortex-a55 -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(SD865)
+    add_definitions(-DSD865)
+    #note that cortex-a77.cortex-a55 is not supported, so fall back to a76 instead
+    add_definitions(-marm -march=armv8.2-a+simd+crypto -mtune=cortex-a76.cortex-a55 -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-marm -march=armv8.2-a+simd+crypto -mtune=cortex-a76.cortex-a55 -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(SD888)
+    add_definitions(-DSD888)
+    add_definitions(-marm -march=armv8.4-a+simd+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-marm -march=armv8.4-a+simd+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(ADLINK)
+    add_definitions(-DADLINK)
+    add_definitions(-marm -mcpu=neoverse-n1 -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-marm -mcpu=neoverse-n1 -mfpu=neon-fp-armv8 -mfloat-abi=hard")
 elseif(A64)
    add_definitions(-DA64)
    add_definitions(-marm -march=armv8-a+crc+simd+crypto -mcpu=cortex-a53+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard)
@@ -133,6 +181,15 @@ elseif(RK3399)
    add_definitions(-DRK3399)
    add_definitions(-pipe -march=armv8-a+crc+simd+crypto -mcpu=cortex-a72+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard)
    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crc+simd+crypto -mcpu=cortex-a72+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(RK3588)
+    add_definitions(-DRK3588)
+    add_definitions(-pipe -mcpu=cortex-a76.cortex-a55+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    add_definitions(-DBAD_SIGNAL)
+    set(CMAKE_ASM_FLAGS  "-pipe -mcpu=cortex-a76.cortex-a55+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(ARM64)
+    add_definitions(-DARM64)
+    add_definitions(-pipe -march=armv8-a+crc+simd+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crc+simd+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard")
 elseif(ODROIDN2)
    add_definitions(-DODROIDN2)
    add_definitions(-pipe -march=armv8-a+crc+simd+crypto -mcpu=cortex-a73.cortex-a53+crypto -mfpu=crypto-neon-fp-armv8 -mfloat-abi=hard)
@@ -145,6 +202,14 @@ elseif(TEGRAX1ARM32)
    add_definitions(-DTEGRAX1ARM32)
    add_definitions(-marm -pipe -march=armv8-a+crc -mcpu=cortex-a57 -mfpu=neon-fp-armv8 -mfloat-abi=hard)
    set(CMAKE_ASM_FLAGS  "-marm -pipe -march=armv8-a+crc -mcpu=cortex-a57 -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(TEGRA_T194)
+    add_definitions(-DTEGRA_T194)
+    add_definitions(-marm -pipe -march=armv8.2-a+fp16+simd+crypto+predres -mfpu=crypto-neon-fp-armv8 -mcpu=cortex-a76+crypto -mtune=cortex-a76 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-marm -pipe -march=armv8.2-a+fp16+simd+crypto+predres -mfpu=crypto-neon-fp-armv8 -mcpu=cortex-a76+crypto -mtune=cortex-a76 -mfloat-abi=hard")
+elseif(TEGRA_T234)
+    add_definitions(-DTEGRA_T234)
+    add_definitions(-marm -pipe -march=armv8.2-a+fp16+simd+crypto+predres -mfpu=crypto-neon-fp-armv8 -mcpu=cortex-a78ae+crypto -mtune=cortex-a78ae -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-marm -pipe -march=armv8.2-a+fp16+simd+crypto+predres -mfpu=crypto-neon-fp-armv8 -mcpu=cortex-a78ae+crypto -mtune=cortex-a78ae -mfloat-abi=hard")
 elseif(PHYTIUM)
    add_definitions(-DPHYTIUM)
    add_definitions(-pipe -march=armv8.1-a+simd+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard)
@@ -153,11 +218,20 @@ elseif(LX2160A)
    add_definitions(-DLX2160A)
    add_definitions(-pipe -march=armv8-a+crypto+crc -mcpu=cortex-a72+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard)
    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crypto+crc -mcpu=cortex-a72+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard")
-elseif(ANDROID)
-    add_definitions(-DANDROID)
 elseif(ARM_DYNAREC)
    #if DYNAREC is selected alone, without any arch
-    set(CMAKE_ASM_FLAGS  "-pipe -march=armv7-a+simd -mfpu=neon")
+    if(NOT ANDROID)
+        set(CMAKE_ASM_FLAGS  "-pipe -march=armv7-a+simd -mfpu=neon")
+    endif()
+endif()
+if(ANDROID)
+    add_definitions(-DANDROID)
+endif()
+if(TERMUX)
+    add_definitions(-DTERMUX)
+endif()
+if(BAD_SIGNAL)
+    add_definitions(-DBAD_SIGNAL)
 endif()

 if(POWER9)
@@ -176,6 +250,12 @@ else()
    set(BOX86 box86)
 endif()

+try_compile(HAVE_ST_INO ${CMAKE_BINARY_DIR}/compile_tests ${CMAKE_SOURCE_DIR}/tests/test_st_ino.c)
+
+if(HAVE_ST_INO)
+  message("stat64 has __st_info field")
+  add_definitions(-DHAVE_ST_INO)
+endif()

 set(BOX86_ELF_ADDRESS "0x62800000")

@@ -260,6 +340,8 @@ set(ELFLOADER_SRC
    "${BOX86_ROOT}/src/tools/box86stack.c"
    "${BOX86_ROOT}/src/tools/my_cpuid.c"
    "${BOX86_ROOT}/src/tools/gtkclass.c"
+    "${BOX86_ROOT}/src/tools/rcfile.c"
+    "${BOX86_ROOT}/src/tools/rbtree.c"
    "${BOX86_ROOT}/src/tools/wine_tools.c"
    "${BOX86_ROOT}/src/elfs/elfloader.c"
    "${BOX86_ROOT}/src/elfs/elfparser.c"
@@ -269,10 +351,8 @@ set(ELFLOADER_SRC
    "${BOX86_ROOT}/src/librarian/dictionnary.c"
    "${BOX86_ROOT}/src/librarian/symbols.c"
    "${BOX86_ROOT}/src/librarian/globalsymbols.c"
+    "${BOX86_ROOT}/src/emu/x86compstrings.c"
    "${BOX86_ROOT}/src/emu/x86emu.c"
-    "${BOX86_ROOT}/src/emu/x86run.c"
-    "${BOX86_ROOT}/src/emu/x86run66.c"
-    "${BOX86_ROOT}/src/emu/x86run660f.c"
    "${BOX86_ROOT}/src/emu/x86run_private.c"
    "${BOX86_ROOT}/src/emu/x86syscall.c"
    "${BOX86_ROOT}/src/emu/x86primop.c"
@@ -280,10 +360,11 @@ set(ELFLOADER_SRC
    "${BOX86_ROOT}/src/emu/x86int3.c"
    "${BOX86_ROOT}/src/emu/x86tls.c"
    "${BOX86_ROOT}/src/emu/x87emu_private.c"
-    "${BOX86_ROOT}/src/emu/x87run.c"
    "${BOX86_ROOT}/src/libtools/auxval.c"
    "${BOX86_ROOT}/src/libtools/myalign.c"
    "${BOX86_ROOT}/src/libtools/myalign64.c"
+    "${BOX86_ROOT}/src/libtools/myalignsys64.c"
+    "${BOX86_ROOT}/src/libtools/stat64_helper.c"
    "${BOX86_ROOT}/src/libtools/myfts.c"
    "${BOX86_ROOT}/src/libtools/sdl1rwops.c"
    "${BOX86_ROOT}/src/libtools/sdl2rwops.c"
@@ -297,12 +378,42 @@ if(NOT ANDROID)
    list(APPEND ELFLOADER_SRC "${BOX86_ROOT}/src/libtools/obstack.c")
 endif()

+set(INTERPRETER
+    "${BOX86_ROOT}/src/emu/x86run.c"
+    "${BOX86_ROOT}/src/emu/x86run0f.c"
+    "${BOX86_ROOT}/src/emu/x86run64.c"
+    "${BOX86_ROOT}/src/emu/x86run640f.c"
+    "${BOX86_ROOT}/src/emu/x86run6466.c"
+    "${BOX86_ROOT}/src/emu/x86run6467.c"
+    "${BOX86_ROOT}/src/emu/x86run66.c"
+    "${BOX86_ROOT}/src/emu/x86run660f.c"
+    "${BOX86_ROOT}/src/emu/x86run66d9.c"
+    "${BOX86_ROOT}/src/emu/x86run66dd.c"
+    "${BOX86_ROOT}/src/emu/x86run66f20f.c"
+    "${BOX86_ROOT}/src/emu/x86run67.c"
+    "${BOX86_ROOT}/src/emu/x86run6766.c"
+    "${BOX86_ROOT}/src/emu/x86rund8.c"
+    "${BOX86_ROOT}/src/emu/x86rund9.c"
+    "${BOX86_ROOT}/src/emu/x86runda.c"
+    "${BOX86_ROOT}/src/emu/x86rundb.c"
+    "${BOX86_ROOT}/src/emu/x86rundc.c"
+    "${BOX86_ROOT}/src/emu/x86rundd.c"
+    "${BOX86_ROOT}/src/emu/x86runde.c"
+    "${BOX86_ROOT}/src/emu/x86rundf.c"
+    "${BOX86_ROOT}/src/emu/x86runf0.c"
+    "${BOX86_ROOT}/src/emu/x86runf066.c"
+    "${BOX86_ROOT}/src/emu/x86runf20f.c"
+    "${BOX86_ROOT}/src/emu/x86runf30f.c"
+)
+
 set(WRAPPEDS
    "${BOX86_ROOT}/src/wrapped/wrappedlibc.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibpthread.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibrt.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibbsd.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibgl.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibglu.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibglx.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibx11.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibasound.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibdl.c"
@@ -327,6 +438,7 @@ set(WRAPPEDS
    "${BOX86_ROOT}/src/wrapped/wrappedpng16.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibxxf86vm.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibxrandr.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibxpresent.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibxext.c"
    "${BOX86_ROOT}/src/wrapped/wrappedxinerama.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibxfixes.c"
@@ -359,6 +471,7 @@ set(WRAPPEDS
    "${BOX86_ROOT}/src/wrapped/wrappedopenal.c"
    "${BOX86_ROOT}/src/wrapped/wrappedalure.c"
    "${BOX86_ROOT}/src/wrapped/wrappedalut.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedopencl.c"
    "${BOX86_ROOT}/src/wrapped/wrappedcurl.c"
    "${BOX86_ROOT}/src/wrapped/wrappedudev0.c"
    "${BOX86_ROOT}/src/wrapped/wrappedudev1.c"
@@ -408,16 +521,19 @@ set(WRAPPEDS
    "${BOX86_ROOT}/src/wrapped/wrappedgsttag.c"
    "${BOX86_ROOT}/src/wrapped/wrappedgstbase.c"
    "${BOX86_ROOT}/src/wrapped/wrappedgstapp.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibibus.c"
    "${BOX86_ROOT}/src/wrapped/wrappeddbusmenugtk.c"
    "${BOX86_ROOT}/src/wrapped/wrappeddbusmenuglib.c"
    "${BOX86_ROOT}/src/wrapped/wrappedappindicator.c"
    "${BOX86_ROOT}/src/wrapped/wrappedappindicator3.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedayatanaappindicator3.c"
    "${BOX86_ROOT}/src/wrapped/wrappedatk.c"
    "${BOX86_ROOT}/src/wrapped/wrappedatkbridge.c"
    "${BOX86_ROOT}/src/wrapped/wrappedpangoft2.c"
    "${BOX86_ROOT}/src/wrapped/wrappedpangocairo.c"
    "${BOX86_ROOT}/src/wrapped/wrappedgthread2.c"
    "${BOX86_ROOT}/src/wrapped/wrappedcairo.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedcairogobject.c"
    "${BOX86_ROOT}/src/wrapped/wrappedpango.c"
    "${BOX86_ROOT}/src/wrapped/wrappeddbusglib1.c"
    "${BOX86_ROOT}/src/wrapped/wrappedgudev1.c"
@@ -462,6 +578,9 @@ set(WRAPPEDS
    "${BOX86_ROOT}/src/wrapped/wrappedpcap.c"
    "${BOX86_ROOT}/src/wrapped/wrappedfaudio.c"
    "${BOX86_ROOT}/src/wrapped/wrappedtbbmallocproxy.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibfuse.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedatomic.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibpcre.c"
 )
 if(ANDROID)
    list(APPEND WRAPPEDS
@@ -491,7 +610,7 @@ add_custom_command(
    OUTPUT "${BOX86_ROOT}/src/wrapped/generated/functions_list.txt"
    COMMAND "${PYTHON_EXECUTABLE}" "${BOX86_ROOT}/rebuild_wrappers.py" 
    "${BOX86_ROOT}" 
-    "PANDORA" "HAVE_LD80BITS" "NOALIGN" "HAVE_TRACE" "POWERPCLE" "ANDROID" "--" 
+    "PANDORA" "HAVE_LD80BITS" "NOALIGN" "HAVE_TRACE" "POWERPCLE" "ANDROID" "TERMUX" "--" 
    ${WRAPPEDS_HEAD}
    MAIN_DEPENDENCY "${BOX86_ROOT}/rebuild_wrappers.py"
    DEPENDS ${WRAPPEDS} ${WRAPPEDS_HEAD}
@@ -515,7 +634,9 @@ if(ARM_DYNAREC)

        "${BOX86_ROOT}/src/dynarec/dynarec_arm.c"
        "${BOX86_ROOT}/src/dynarec/dynarec_arm_functions.c"
+        "${BOX86_ROOT}/src/dynarec/dynarec_arm_jmpnext.c"
        "${BOX86_ROOT}/src/dynarec/arm_printer.c"
+        "${BOX86_ROOT}/src/emu/x86test.c"

        "${BOX86_ROOT}/src/dynarec/arm_prolog.S"
        "${BOX86_ROOT}/src/dynarec/arm_epilog.S"
@@ -563,6 +684,8 @@ if(ARM_DYNAREC)
    set_target_properties(arm_pass2 PROPERTIES COMPILE_FLAGS "-DSTEP=2")
    add_library(arm_pass3 OBJECT ${DYNAREC_PASS})
    set_target_properties(arm_pass3 PROPERTIES COMPILE_FLAGS "-DSTEP=3")
+    add_library(test_interpreter OBJECT ${INTERPRETER})
+    set_target_properties(test_interpreter PROPERTIES COMPILE_FLAGS "-DTEST_INTERPRETER")
    add_dependencies(arm_pass0 WRAPPERS)
    add_dependencies(arm_pass1 WRAPPERS)
    add_dependencies(arm_pass2 WRAPPERS)
@@ -574,6 +697,7 @@ if(ARM_DYNAREC)
        $<TARGET_OBJECTS:arm_pass1>
        $<TARGET_OBJECTS:arm_pass2>
        $<TARGET_OBJECTS:arm_pass3>
+        $<TARGET_OBJECTS:test_interpreter>
    )

 endif()
@@ -583,16 +707,18 @@ if(ARM_DYNAREC)
    add_custom_command(
        OUTPUT "${BOX86_ROOT}/src/git_head.h"
        COMMAND bash -c "echo \\\#define GITREV \\\"$(git rev-parse --short HEAD)\\\">\"${BOX86_ROOT}/src/git_head.h\""
-        DEPENDS dynarec ${ELFLOADER_SRC} ${WRAPPEDS}
+        DEPENDS dynarec ${ELFLOADER_SRC} ${INTERPRETER} ${WRAPPEDS}
        VERBATIM)
 else()
    add_custom_command(
        OUTPUT "${BOX86_ROOT}/src/git_head.h"
        COMMAND bash -c "echo \\\#define GITREV \\\"$(git rev-parse --short HEAD)\\\">\"${BOX86_ROOT}/src/git_head.h\""
-        DEPENDS ${ELFLOADER_SRC} ${WRAPPEDS}
+        DEPENDS ${ELFLOADER_SRC} ${INTERPRETER} ${WRAPPEDS}
        VERBATIM)
 endif()

+add_library(interpreter OBJECT ${INTERPRETER})
+
 if(BOX86LIB)
 #build a static or dynamic lib...
 if(BOX86DYNAMIC)
@@ -602,9 +728,13 @@ else()
 endif()
 if(ARM_DYNAREC)
    if(ANDROID)
-        target_link_libraries(${BOX86} dynarec m dl)
+        if(TERMUX)
+        target_link_libraries(${BOX86} dynarec interpreter m dl android-sysv-semaphore android-spawn)
+        else()
+        target_link_libraries(${BOX86} dynarec interpreter m dl)
+        endif()
    else()
-        target_link_libraries(${BOX86} dynarec m dl rt pthread resolv)
+        target_link_libraries(${BOX86} dynarec interpreter m dl rt pthread resolv)
    endif()
 endif()
 if(${CMAKE_VERSION} VERSION_LESS "3.13")
@@ -624,37 +754,78 @@ add_executable(${BOX86} ${ELFLOADER_SRC} ${WRAPPEDS} "${BOX86_ROOT}/src/git_head
 add_dependencies(${BOX86} WRAPPERS)
 add_dependencies(${BOX86} PRINTER)
 if(ANDROID)
+    if(TERMUX)
+    target_link_libraries(${BOX86} dynarec interpreter m dl android-complex-math android-sysv-semaphore android-spawn )
+    else()
+    target_link_libraries(${BOX86} dynarec interpreter m dl)
+    endif()
    target_link_libraries(${BOX86} c m dl)
 else()
    #target_link_libraries(${BOX86} c m dl rt pthread resolv)
    set_target_properties(${BOX86} PROPERTIES LINK_FLAGS "-Wl,--no-as-needed -lc -lm -ldl -lrt -lpthread -lresolv -Wl,--as-needed -pthread")
 endif()
 if(ARM_DYNAREC)
-    target_link_libraries(${BOX86} dynarec)
+    target_link_libraries(${BOX86} dynarec interpreter)
+else()
+    target_link_libraries(${BOX86} interpreter)
 endif()

 if(${CMAKE_VERSION} VERSION_LESS "3.13")
    if(NOT NOLOADADDR)
-        set_target_properties(${BOX86} PROPERTIES LINK_FLAGS "-rdynamic -Wl,-Ttext-segment,${BOX86_ELF_ADDRESS}")
+        if(CMAKE_C_COMPILER_ID STREQUAL "Clang")
+            set_target_properties(${BOX86} PROPERTIES LINK_FLAGS "-rdynamic -Wl,--image-base,${BOX86_ELF_ADDRESS}")
+        else()
+            set_target_properties(${BOX86} PROPERTIES LINK_FLAGS "-rdynamic -Wl,-Ttext-segment,${BOX86_ELF_ADDRESS}")
+        endif()
    else()
        set_target_properties(${BOX86} PROPERTIES LINK_FLAGS -rdynamic)
    endif()
 else()
    target_link_options(${BOX86} PUBLIC -rdynamic)
    if(NOT NOLOADADDR)
-        target_link_options(${BOX86} PUBLIC -Wl,-Ttext-segment,${BOX86_ELF_ADDRESS})
+        if(CMAKE_C_COMPILER_ID STREQUAL "Clang")
+            target_link_options(${BOX86} PUBLIC -Wl,--image-base,${BOX86_ELF_ADDRESS})
+        else()
+            target_link_options(${BOX86} PUBLIC -Wl,-Ttext-segment,${BOX86_ELF_ADDRESS})
+        endif()
    endif()
 endif()

 if(NOT _x86 AND NOT _x86_64)
-  install(TARGETS ${BOX86}
+  if(NOT TERMUX)
+   install(TARGETS ${BOX86}
    RUNTIME DESTINATION bin)
-  configure_file(system/box86.conf.cmake system/box86.conf)
-  install(FILES ${CMAKE_BINARY_DIR}/system/box86.conf DESTINATION /etc/binfmt.d/)
-  install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libstdc++.so.6 DESTINATION /usr/lib/i386-linux-gnu/)
-  install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libstdc++.so.5 DESTINATION /usr/lib/i386-linux-gnu/)
-  install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libgcc_s.so.1 DESTINATION /usr/lib/i386-linux-gnu/)
-  install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libpng12.so.0 DESTINATION /usr/lib/i386-linux-gnu/)
+  else()
+   install(TARGETS ${BOX86}
+    RUNTIME DESTINATION ${TERMUX_PATH}/usr/bin)
+  endif()
+  if(NOT NO_CONF_INSTALL)
+    configure_file(system/box86.conf.cmake system/box86.conf)
+    	if(NOT TERMUX)
+    		install(FILES ${CMAKE_BINARY_DIR}/system/box86.conf DESTINATION /etc/binfmt.d/)
+		install(FILES ${CMAKE_SOURCE_DIR}/system/box86.box86rc DESTINATION /etc/)
+	else()
+		#install(FILES ${CMAKE_BINARY_DIR}/system/box86.conf DESTINATION ${TERMUX_PATH}/usr/etc/binfmt.d/)
+		install(FILES ${CMAKE_SOURCE_DIR}/system/box86.box86rc DESTINATION ${TERMUX_PATH}/usr/etc/)
+	endif()
+  endif()
+  set(INSTALL_PATH "/usr/lib/box86-i386-linux-gnu/")
+  if(NOT NO_LIB_INSTALL)
+    if(NOT TERMUX)
+    	install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libstdc++.so.6 DESTINATION ${INSTALL_PATH})
+    	install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libstdc++.so.5 DESTINATION ${INSTALL_PATH})
+    	install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libgcc_s.so.1 DESTINATION ${INSTALL_PATH})
+    	install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libpng12.so.0 DESTINATION ${INSTALL_PATH})
+    	install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libunwind.so.8 DESTINATION ${INSTALL_PATH})
+    else()
+
+	    install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libstdc++.so.6 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+	    install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libstdc++.so.5 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+	    install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libgcc_s.so.1 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+	    install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libpng12.so.0 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+	    install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libunwind.so.8 DESTINATION ${TERMUX_PATH}${INSTALL_PATH})
+    endif()
+  endif()
 endif()

 if(NOT TARGET uninstall)
@@ -671,7 +842,11 @@ set(CPACK_GENERATOR "DEB")
 set(CPACK_DEBIAN_PACKAGE_MAINTAINER "ptitSeb")
 set(CPACK_PACKAGE_CONTACT "[email protected]")
 set(CPACK_PACKAGE_DESCRIPTION "Box86 - Linux Userspace x86 Emulator with a twist")
+if(TERMUX)
+set(CPACK_DEBIAN_PACKAGE_DEPENDS "libandroid-sysv-semaphore, libandroid-spawn")
+else()
 set(CPACK_DEBIAN_PACKAGE_DEPENDS "libc6")
+endif()
 set(CPACK_PACKAGE_HOMEPAGE_URL, "https://box86.org")
 file(STRINGS "${BOX86_ROOT}/src/box86version.h" TEMP_STRING REGEX "BOX86_MAJOR\\s*")
 string(REGEX REPLACE "BOX86_MAJOR" "" TEMP_STRING ${TEMP_STRING})
@@ -685,16 +860,29 @@ string(REGEX MATCH "[0-9]" BOX86_REVISION ${TEMP_STRING})
 set(CPACK_PACKAGE_VERSION_MAJOR ${BOX86_MAJOR})
 set(CPACK_PACKAGE_VERSION_MINOR ${BOX86_MINOR})
 set(CPACK_PACKAGE_VERSION_PATCH ${BOX86_REVISION})
+if(NOT TERMUX)
 set(CPACK_DEBIAN_PACKAGE_CONTROL_EXTRA "${CMAKE_CURRENT_SOURCE_DIR}/postinst")
-if(_aarch64 OR _armhf)
+endif()
+if(_aarch64 OR _armhf OR _arm OR _armv8l)
+    if(TERMUX)
+    set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "arm")
+    else()
    set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "armhf")
+    endif()
 elseif(_x86_64 OR _x86)
    set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "i386")
 else()
    #probably wrong...
    execute_process(COMMAND "dpkg --print-architecture" OUTPUT_VARIABLE CPACK_DEBIAN_PACKAGE_ARCHITECTURE)
 endif()
-set(CPACK_DEBIAN_FILE_NAME "${BOX86}-${BOX86_MAJOR}.${BOX86_MINOR}.${BOX86_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 "${BOX86}-${BOX86_MAJOR}.${BOX86_MINOR}.${BOX86_REVISION}_${CMAKE_SYSTEM_NAME}-${CPACK_DEBIAN_PACKAGE_ARCHITECTURE}.deb")
+else()
+set(CPACK_DEBIAN_FILE_NAME "${BOX86}-${BOX86_MAJOR}.${BOX86_MINOR}.${BOX86_REVISION}_Termux-${CPACK_DEBIAN_PACKAGE_ARCHITECTURE}.deb")
+endif()
 INCLUDE(CPack)

 add_test(NAME bootSyscall COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
@@ -784,7 +972,7 @@ add_test(NAME sse COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/$
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref17.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-set_tests_properties(sse PROPERTIES ENVIRONMENT "BOX86_DYNAREC_FASTNAN=0")
+set_tests_properties(sse PROPERTIES ENVIRONMENT "BOX86_DYNAREC_FASTNAN=0;BOX86_DYNAREC_FASTROUND=0")

 add_test(NAME longjumpInSignals COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test18 -D TEST_OUTPUT=tmpfile18.txt 
@@ -801,13 +989,52 @@ add_test(NAME idiv COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref20.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-file(GLOB extension_tests "${CMAKE_SOURCE_DIR}/tests/extensions/*.c")
+add_test(NAME multiple_dlopen COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+    -D TEST_ARGS=${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(NAME aes COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test22 -D TEST_OUTPUT=tmpfile22.txt 
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref22.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+file(GLOB extension_tests "${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}/${BOX86}
        -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/extensions/${testname} -D TEST_OUTPUT=tmpfile-${testname}.txt
        -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/extensions/${testname}.txt
        -P ${CMAKE_SOURCE_DIR}/runTest.cmake)
-endforeach()
+endforeach()        

+add_test(NAME sse_optimized COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test17_o2 -D TEST_OUTPUT=tmpfile17_o2.txt 
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref17_o2.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+set_tests_properties(sse_optimized PROPERTIES ENVIRONMENT "BOX86_DYNAREC_FASTNAN=0;BOX86_DYNAREC_FASTROUND=0")    
+    
+add_test(bswap ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86}
+        -D TEST_ARGS=${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(feround ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86}
+        -D TEST_ARGS=${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(sse4_2 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86}
+        -D TEST_ARGS=${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(fpu_rounding ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86}
+        -D TEST_ARGS=${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(fpu_rounding PROPERTIES ENVIRONMENT "BOX86_DYNAREC_FASTROUND=0")
+    
 endif(BOX86LIB)
@@ -1,3 +1,130 @@
+box86 (0.3.8) unstable; urgency=low
+* Some optimisation of x86 P flags handling
+* Some x87 fixes and improvments
+* Added some of the tim64 wrapped functions to libc and friends
+* Changed the installation folder of x86 libs to avoid conflict with linux distro
+* Some fixes to internal memory tracking
+* Some more Vulkan extensions wrapped
+* Better handling of BOX86_MAXCPU
+* Some changes and improvments on Signal handling
+* More wrapped functions, and fix to some existing ones
+* A few more syscall, and fix to some existing ones
+* New build profile for SD865
+
+-- Sebastien Chevalier <[email protected]>  Fri, 06 Dec 2024 14:06:00 -0100
+
+box86 (0.3.6) unstable; urgency=low
+* Wrapping: More libs and and function wrapping (gtk3, vulkan...)
+* Reworked X11 Callback handling, for better stability
+* Reworked exit and the load/unload of libs
+* Some more profiles in box86rc
+* More SSE3+ opcodes in dynarec
+* Better way to rename executable (including wine ones) so driver like mesa can apply the correct profile
+* Lots of improvments in x87 handling (virtual stack, rounding, ffree, etc.)
+* => should improve games stability
+* Better handling of large block in the dynarec
+* => should speedup dynarec, and allow use of BIGBLOCK 2 or 3 on more software
+* Added suport for BOX86_DYNAREC_CALLRET optimisation.
+* => backported from box64, can give 10% or more speed increase (but can also decrease speed in some cases). Still a bit crashy so use with care.
+
+-- Sebastien Chevalier <[email protected]>  Thu, 31 May 2024 17:30:33 -0100
+
+box86 (0.3.4) unstable; urgency=low
+
+ * X86Libs: Updated libstdc++ & libgcc_s
+ * X86Libs: Use x86 vrsion of libunwind (it's needed by wine)
+ * Dynarec: Fixed some flags propagation issues inside blocks
+ * Dynarec: Improved Strong Memory Model emulation, introducing a 3rd level
+ * CPU: Remoworked how INT, priviliged opcodes and teh Trace flags works
+ * CPU: Added full support for SSE4.2
+ * ElfLoader: rework how elf memory is handled
+ * 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 for ADLink Ampere Altra
+ * OS: Added profile to build for Android and Termux.
+ * => Mostly a stability update with improved elfloader again, and added SSE 4.2 support (still need to work on the Dynarec part of this)
+ * => Some work also to reduce the memory footprint
+
+-- Sebastien Chevalier <[email protected]>  Thu, 15 Dec 2023 17:30:33 -0100
+
+box86 (0.3.2) unstable; urgency=low
+
+ * Better syscall handling
+ * Fixed some wrapped functions, added some others
+ * Introduced another wrapper helper
+ * Added some more opcodes
+ * Added a cosimulation mode (to compare the behaviour of the dynarec against the interpreter)
+ * Added support of FASTNAN for some more opcodes
+ * Reduced memory footprint
+ * Added BOX86_NOSANDBOX (adds an argument to the command line)
+ * Added an ukrainian translation
+ * => Mostly a stability update with improved elfloader, fixed opcodes and refined profiles
+
+-- Sebastien Chevalier <[email protected]>  Thu, 10 Mar 2023 10:20:33 -0500
+
+box86 (0.3.0) unstable; urgency=low
+
+ * Added rcfile handling: box86 now read "/etc/box86.box86rc" and "~/.box86rc" and apply per process parameters
+ * Added BOX86_DYNAREC_FORWARD to allow bigger Dynarec block to be build
+ * Refactored Strong Memory Model emulation (faster and more accurate)
+ * Box86 now return a custom name for the CPU instal of P4, build with Box86 and the cpu name it's running on
+ * Improved memory protection tracking and signal handling, again
+ * More wrapped libs (especially on the gtk familly)
+ * More wrapped functions
+ * Added support for CreateDeviceEx in d3dadapter9, for better gallium nine support
+ * Improved dladdr/dladdr1 wrapped function
+ * Some fixes to flags handling in the Dynarec on some special cases
+ * Added specific suppport for RK3588 board (needed some hack in signal handling, probably due to the use of a non mainlined kernel)
+ * Lots of improvment in the Rounding of float & double (Dynarec and Interpretor), and proper handling of denormals when converting to/from 80 bits long double
+ * Added specific suppport for RK3588 board (needed some hack in signal handling, probably due to the use of a non mainlined kernel)
+
+ -- Sebastien Chevalier <[email protected]>  Thu, 10 Mar 2023 10:20:33 -0500
+
+box86 (0.2.8) unstable; urgency=low
+
+  * Wrapped more Vulkan functions, dxvk2.0 now works
+  * Added support for overridden malloc/free functions
+  * Improved elf loader
+  * Improved dlsym handling with RTLD_NEXT
+  * Added BOX86_DYNAREC_SAFEFLAGS to allow finetuning Dynarec generated code handling of flags accross function calls and returns.
+  * Added BOX86_BASH to setup x86 bash binary, so shell script can be run in x86 world
+  * Added BOX86_ROLLING_LOG to have details log only on crash
+  * Improved TLS Size handling (Unity3D games now works)
+  * Improved execv family of function to have better box86 integration
+  * Added a few more wrapped libs, like FAudio
+  * Improved gtk, SDL2, GL, vulkan (and more) wrapping
+  * Improved Signal Handling and memory protection traking
+  * Added some basic support for statically linked program. Thread are not working yet, but memory allocation and file access does.
+  * add LX2160A config by @lizthegrey in #612
+  * Grammar and formatting COMPILE.md by @Radiicall in #615
+  * Update README.md by @everypizza1 in #617
+  * docs: add instructions for Debian-based Linux by @ryanfortner in #622
+  * Fix typoes in readme by @ryanfortner in #621
+  * Various improvements to install_steam.sh by @ryanfortner in #620
+  * docs: remove unnecessary newline by @ryanfortner in #624
+  * fix: the first parameter of realpath may be null by @ye-yeshun in #630
+  * Add 6 wrapped function to libdrm, and 2 to libexpat. by @cjy880927 in #632
+  * Fix #610 by @rajdakin in #634
+  * some floating point instructions by @ye-yeshun in #637
+  * Fix Typo in readme x86_64 -> x86 by @Jai-JAP in #642
+  * Fix grammar by @Jai-JAP in #645
+  * Test wine preload by @ptitSeb in #649
+  * Align address when call mprotect by @wannacu in #651
+  * Added 66 0F 7D HSUBPD opcode by @wannacu in #652
+  * add TEGRAX1 and TEGRAX1ARM32 targets on ubuntu-18.04 by @theofficialgman in #654
+  * automatically install 32 bit libraries if 64 bit OS is detected in install_steam.sh by @Moldytzu in #659
+  * mention gcc-arm-linux-gnueabihf in compile.md by @Moldytzu in #660
+  * Add libc6:armhf install to RPi4 aarch64 build info by @WheezyE in #685
+  * Updated wine install for RPiOS bullseye 32/64bit by @WheezyE in #689
+  * Add atomic add opcodes by @Trass3r in #695
+  * steam: disable modern login and pass arguments to client by @Moldytzu in #700
+  * Modify sizeof(instsize_t) to 1 byte by @liuli0217 in #721
+  * Add MOV GS:od, AL by @liuli0217 in #722
+
+ -- Sebastien Chevalier <[email protected]>  Thu, 17 Nov 2022 11:20:33 -0500
+
 box86 (0.2.6) unstable; urgency=low

  * Added an option to get generated -NAN on SSE2 operation
@@ -6,7 +6,13 @@
 override_dh_auto_configure:
 	dh_auto_configure -- \
 	      -DCMAKE_BUILD_TYPE=RelWithDebInfo \
-	      -DARM_DYNAREC=1
+	      -DARM_DYNAREC=1 \
+		  -DNOGIT=1

 override_dh_shlibdeps:
-	dh_shlibdeps --exclude=libgcc_s.so.1 --exclude=libpng12.so.0 --exclude=libstdc++.so.5 --exclude=libstdc++.so.6 
+	dh_shlibdeps \
+		--exclude=libgcc_s.so.1 \
+		--exclude=libpng12.so.0 \
+		--exclude=libstdc++.so.5 \
+		--exclude=libstdc++.so.6 \
+		--exclude=libunwind.so.8
@@ -1,3 +1,85 @@
+v0.3.8
+======
+* Some optimisation of x86 P flags handling
+* Some x87 fixes and improvments
+* Added some of the tim64 wrapped functions to libc and friends
+* Changed the installation folder of x86 libs to avoid conflict with linux distro
+* Some fixes to internal memory tracking
+* Some more Vulkan extensions wrapped
+* Better handling of BOX86_MAXCPU
+* Some changes and improvments on Signal handling
+* More wrapped functions, and fix to some existing ones
+* A few more syscall, and fix to some existing ones
+* New build profile for SD865
+
+v0.3.6
+======
+* Wrapping: More libs and function wrapping (gtk3, vulkan...)
+* Reworked X11 Callback handling, for better stability
+* Reworked exit and the load/unload of libs
+* Some more profiles in box86rc
+* More SSE3+ opcodes in dynarec
+* Better way to rename executable (including wine ones) so driver like mesa can apply the correct profile
+* Better handling of the Hardware Counter and TSC emulation
+* Reworked memory managment in the Dynarec, limiting amount of dynamic allocation
+* PageSize detection and handling dynamic. Larger than 4K pagesize will be handled automaticaly. Note that box64 does not emulate 4K page size, so some program might still refuse to work on larger page (like Wine for example)
+=> Linux Games can now load even on a 16K pagesize machine like a mac M1, and games (including Unity3D ones) can be played with no issues.
+* Lots of improvments in x87 handling (virtual stack, rounding, ffree, etc.)
+=> should improve games stability
+* Better handling of large block in the dynarec
+=> should speedup dynarec, and allow use of BIGBLOCK 2 or 3 on more software
+* Added suport for BOX86_DYNAREC_CALLRET optimisation.
+=> backported from box64, can give 10% or more speed increase (but can also decrease speed in some cases). Still a bit crashy so use with care.
+
+v0.3.4
+======
+* X86Libs: Updated libstdc++ & libgcc_s
+* X86Libs: Use x86 vrsion of libunwind (it's needed by wine)
+* Dynarec: Fixed some flags propagation issues inside blocks
+* Dynarec: Improved Strong Memory Model emulation, introducing a 3rd level
+* CPU: Remoworked how INT, priviliged opcodes and teh Trace flags works
+* CPU: Added full support for SSE4.2
+* ElfLoader: rework how elf memory is handled
+* 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 for ADLink Ampere Altra
+* OS: Added profile to build for Android and Termux.
+=> Mostly a stability update with improved elfloader again, and added SSE 4.2 support (still need to work on the Dynarec part of this)
+=> Some work also to reduce the memory footprint
+
+v0.3.2
+======
+* Better syscall handling
+* Fixed some wrapped functions, added some others
+* Introduced another wrapper helper
+* Added some more opcodes
+* Added a cosimulation mode (to compare the behaviour of the dynarec against the interpreter)
+* Added support of FASTNAN for some more opcodes
+* Reduced memory footprint
+* Added BOX86_NOSANDBOX (adds an argument to the command line)
+* Added an ukrainian translation
+=> Mostly a stability update with improved elfloader, fixed opcodes and refined profiles
+
+v0.3.0
+======
+* Added rcfile handling: box86 now read `/etc/box86.box86rc` and `~/.box86rc` and apply per process parameters
+    => Every program can be finetuned, for speed or just to make them work
+* Added BOX86_DYNAREC_FORWARD to allow bigger Dynarec block to be build
+* Refactored Strong Memory Model emulation (faster and more accurate)
+* Box86 now return a custom name for the CPU instal of P4, build with Box86 and the cpu name it's running on
+* Improved memory protection tracking and signal handling, again
+* More wrapped libs (especially on the gtk familly)
+* More wrapped functions
+* Added support for CreateDeviceEx in d3dadapter9, for better gallium nine support
+* Improved dladdr/dladdr1 wrapped function
+* Some fixes to flags handling in the Dynarec on some special cases
+* Added specific suppport for RK3588 board (needed some hack in signal handling, probably due to the use of a non mainlined kernel)
+* Lots of improvment in the Rounding of float & double (Dynarec and Interpretor), and proper handling of denormals when converting to/from 80 bits long double
+* Added specific suppport for RK3588 board (needed some hack in signal handling, probably due to the use of a non mainlined kernel)
+
+
 v0.2.8
 ======
 * Wrapped more vulkan function, dxvk 2.0 now works
@@ -1,16 +1,21 @@
 Compiling/Installing
 ----

-#### Debian-based Linux 
-You can use [@Itai-Nelken](https://github.com/Itai-Nelken)'s apt repository to install precompiled box86 debs, updated weekly. 
+If you don't want to compile box86 yourself and prefer to use third-party pre-build version, go to the [end of the document](#pre-build-packages) for alternatives.

-```
-sudo wget https://itai-nelken.github.io/weekly-box86-debs/debian/box86.list -O /etc/apt/sources.list.d/box86.list
-wget -qO- https://itai-nelken.github.io/weekly-box86-debs/debian/KEY.gpg | sudo apt-key add -
-sudo apt update && sudo apt install box86 -y
+You can also generate your own debian package using the [instructions below](https://github.com/ptitSeb/box86/blob/master/docs/COMPILE.md#debian-packaging). 
+
+Important note, if you are using a 64bits OS, you need to have multi-arch enabled, with armhf support, for box86 to works. On Debian/Ubuntu/Armbian OS, you need to do:
+
+```bash
+sudo dpkg --add-architecture armhf && sudo apt-get update
+sudo apt-get install libc6:armhf -y
 ```

-Alternatively, you can generate your own package using the [instructions below](https://github.com/ptitSeb/box86/blob/master/docs/COMPILE.md#debian-packaging). 
+And of course you will need to install much more library to be able to run software and games (like X11, mesa, SDL, pulseaudio...)
+
+## Per-platform compiling instructions
+----

 #### for Pandora

@@ -52,15 +57,15 @@ _Build box86:armhf on RPiOS 64-bit ARM (aarch64)_
 sudo apt install gcc-arm-linux-gnueabihf # building 32-bit ARM code on aarch64 requires this armhf gcc cross-compiler toolchain 
 git clone https://github.com/ptitSeb/box86
 cd box86
-mkdir build; cd build; cmake .. -DRPI4ARM64=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo # -DRPI4ARM64=1 for Pi4 aarch64 (use `-DRPI2ARM64=1` etc for other models)
+mkdir build; cd build; cmake .. -DRPI4ARM64=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo # -DRPI4ARM64=1 for Pi4 aarch64 (use `-DRPI3ARM64=1` for a PI3 model)
 make -j2
 sudo make install
 sudo systemctl restart systemd-binfmt
 ```

-#### for ODROID
+#### for ODROID-XU3/XU4/MC1/HC1/HC2

-`mkdir build; cd build; cmake .. -DODROID=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+`mkdir build; cd build; cmake .. -DODROIDXU4=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`

 #### for RK3399

@@ -70,6 +75,22 @@ As most RK3399 devices run an AARCH64 OS, you'll need an `armhf` multiarch envir

 Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.

+#### for RK3588 / RK3588S
+
+`mkdir build; cd build; cmake .. -DRK3588=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j4`
+
+As most RK3588 devices run an AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On debian, install it with `sudo apt install gcc-arm-linux-gnueabihf`. 
+
+Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.
+
+#### for ODROID-N2/N2+
+
+`mkdir build; cd build; cmake .. -DODROIDN2=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+
+As most ODROID-N2/N2+ devices run an AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On debian, install it with `sudo apt install gcc-arm-linux-gnueabihf`. 
+
+Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.
+
 #### for Tinker Board (1/1S) or RK3288

 `mkdir build; cd build; cmake .. -DRK3288=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
@@ -82,11 +103,13 @@ As most Allwinner A64 devices run an AARCH64 OS, you'll need an `armhf` multiarc

 Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.

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

 `mkdir build; cd build; cmake .. -DSD845=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j4`
+or
+`mkdir build; cd build; cmake .. -DSD888=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j4`

-As most Snapdragon 845 devices run an AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On mobian, install it with `sudo apt install gcc-arm-linux-gnueabihf`.
+Depending how old/recent you SD is. As most Snapdragon devices run an AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On mobian, install it with `sudo apt install gcc-arm-linux-gnueabihf`.

 #### for Phytium

@@ -94,12 +117,31 @@ As most Snapdragon 845 devices run an AARCH64 OS, you'll need an `armhf` multiar

 As most Phytium (D2000 or FT2000/4) devices run an AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On debian, install it with `sudo apt install gcc-arm-linux-gnueabihf`. 

+
+Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.
+
+#### for ADLink AmpereAltra
+
+`mkdir build; cd build; cmake .. -DADLINK=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j`
+
+As ADLINK AmpereAltra computers run an AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On debian, install it with `sudo apt install gcc-arm-linux-gnueabihf`. 
+
+#### for Other ARM64 64bits Linux platform
+
+`mkdir build; cd build; cmake .. -DARM64=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+
+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'll need an `armhf` multiarch environment, and an armhf gcc: On debian, install it with `sudo apt install gcc-arm-linux-gnueabihf`. 
+
 Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.

 #### for Other ARM Linux platforms

 `mkdir build; cd build; cmake .. -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`

+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
+
 #### for x86 Linux

 `mkdir build; cd build; cmake .. -DCMAKE_C_FLAGS="-m32" -DLD80BITS=1 -DNOALIGN=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
@@ -125,6 +167,7 @@ To have a trace enabled build (***the interpreter will be slightly slower***), a
 #### to have ARM Dynarec

 ###### *Note: VFPv3 and NEON are required for Dynarec.*
+###### You might need to `sudo apt install neon-support` package for CMAKE to correctly detect them.

 ###### *Note: Compiling with `ARM_DYNAREC` without selecting a hardware profile is not advised.*

@@ -132,7 +175,7 @@ Dynarec is only available on the ARM architecture (for the meantime anyways). Ac

 You will most likely need `-marm` in compilation flags (Many compilers default to the Thumb instruction set and Dynarec does not support this).

-###### *Note: If you get error building that "target CPU does not support ARM mode", then try to pick a hardware profile (like ODROID for armv7 or PI4 for armv8).*
+###### *Note: If you get error building that "target CPU does not support ARM mode", then try to pick a hardware profile (like ODROIDXU4 for armv7 or PI4 for armv8).*

 ##### 64bit OS with Dynarec

@@ -145,7 +188,7 @@ If you are not building from a git clone (for example, downloading a release sou

 ----

-Testing
+## Testing
 ----
 A few tests are included with box86.

@@ -161,7 +204,46 @@ NVIDIA doesn't provide armhf libraries for their GPU drivers at this time. There

 ----

-Debian Packaging
+## Debian Packaging
 ----
-Box86 can also be packaged into a .deb file with `DEB_BUILD_OPTIONS=nostrip dpkg-buildpackage -us -uc -nc`.
+Box86 can also be packaged into a .deb file ***using the source code zip from the releases page*** with `DEB_BUILD_OPTIONS=nostrip dpkg-buildpackage -us -uc -nc`. Configure any additional cmake options you might want in `debian/rules`.

+## Pre-build packages
+----
+#### Debian-based Linux 
+You can use the [Pi-Apps-Coders apt repository](https://github.com/Pi-Apps-Coders/box86-debs) to install precompiled box86 debs, updated every 24 hours. 
+
+```
+# check if .list file already exists
+if [ -f /etc/apt/sources.list.d/box86.list ]; then
+  sudo rm -f /etc/apt/sources.list.d/box86.list || exit 1
+fi
+# check if .sources file already exists
+if [ -f /etc/apt/sources.list.d/box86.sources ]; then
+  sudo rm -f /etc/apt/sources.list.d/box86.sources || exit 1
+fi
+# download gpg key from specified url
+if [ -f /usr/share/keyrings/box86-archive-keyring.gpg ]; then
+  sudo rm -f /usr/share/keyrings/box86-archive-keyring.gpg
+fi
+sudo mkdir -p /usr/share/keyrings
+wget -qO- "https://pi-apps-coders.github.io/box86-debs/KEY.gpg" | sudo gpg --dearmor -o /usr/share/keyrings/box86-archive-keyring.gpg
+# create .sources file
+echo "Types: deb
+URIs: https://Pi-Apps-Coders.github.io/box86-debs/debian
+Suites: ./
+Signed-By: /usr/share/keyrings/box86-archive-keyring.gpg" | sudo tee /etc/apt/sources.list.d/box86.sources >/dev/null
+```
+
+On a 32bit OS, run the following additional commands
+```
+sudo apt update
+sudo apt install box86-generic-arm -y
+```
+
+On a 64bit OS, run the following additional commands
+```
+sudo dpkg --add-architecture armhf
+sudo apt update
+sudo apt install box86-generic-arm:armhf -y
+```
@@ -2,7 +2,7 @@

 Linux Userspace x86 Emulator with a twist

-[View changelog](https://github.com/ptitSeb/box86/blob/master/docs/CHANGELOG.md) | [中文](https://github.com/ptitSeb/box86/blob/master/docs/README_CN.md) | [Report an error](https://github.com/ptitSeb/box86/issues/new)
+[View changelog](https://github.com/ptitSeb/box86/blob/master/docs/CHANGELOG.md) | [中文](https://github.com/ptitSeb/box86/blob/master/docs/README_CN.md) | [Українська](https://github.com/ptitSeb/box86/blob/master/docs/README_UK.md) | [Report an error](https://github.com/ptitSeb/box86/issues/new)

 ![build](https://app.travis-ci.com/ptitSeb/box86.svg?branch=master) ![stars](https://img.shields.io/github/stars/ptitSeb/box86) ![forks](https://img.shields.io/github/forks/ptitSeb/box86) ![contributors](https://img.shields.io/github/contributors/ptitSeb/box86) ![prs](https://img.shields.io/github/issues-pr/ptitSeb/box86) ![issues](https://img.shields.io/github/issues/ptitSeb/box86)

@@ -12,9 +12,9 @@ Box86 lets you run x86 Linux programs (such as games) on non-x86 Linux systems,

 You *NEED* a 32-bit subsystem to run and build Box86. Box86 is useless on 64-bit only systems. Also, you *NEED* a 32-bit toolchain to build Box86. A toolchain that only supports 64-bit will not compile Box86, and you'll get errors (typically on aarch64, you get "-marm" not recognized, and you'll need a multiarch or chroot environment).

-Because Box86 uses the native versions of some "system" libraries, like libc, libm, SDL, and OpenGL, it's easy to integrate and use with most applications, and performance can be surprisingly high in many cases. Take a look at those bench analysis for an example [here](https://box86.org/index.php/2021/06/game-performances/). That also means that you will need a 32bits userspace on 64bits OS, like `armhf` on top an ̀`aarch64` 64bits OS.
+Because Box86 uses the native versions of some "system" libraries, like libc, libm, SDL, and OpenGL, it's easy to integrate and use with most applications, and performance can be surprisingly high in many cases. Take a look at those bench analysis for an example [here](https://box86.org/index.php/2021/06/game-performances/). That also means that you will need a 32bits userspace on 64bits OS, like `armhf` on top an `aarch64` 64bits OS.

-Most x86 Games need OpenGL, so on ARM platforms a solution like [gl4es](https://github.com/ptitSeb/gl4es) might be necessary. (Some ARM platforms only support OpenGL ES and/or their OpenGL implementation is dodgy (see OpenGL on Android).)
+Most x86 Games need OpenGL, so on ARM platforms a solution like [gl4es](https://github.com/ptitSeb/gl4es) might be necessary. (Some ARM platforms only support OpenGL ES and/or their OpenGL implementation is dodgy. (see OpenGL on Android))

 Box86 now integrates a DynaRec (dynamic recompiler) for the ARM platform, providing a speed boost between 5 to 10 times faster than only using the interpreter. Some high level information on how the Dynarec work can be found [here](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/).

@@ -28,8 +28,8 @@ Some x86 internal opcodes use parts of "Realmode X86 Emulator Library", see [x86
 ----
 Compiling/Installation
 ----
-> Compilation instructions can be found [here](COMPILE.md)  
-> Instructions for installing Wine for Box86 can be found [here](X86WINE.md)  
+> Compilation instructions can be found [here](https://github.com/ptitSeb/box86/blob/master/docs/COMPILE.md)  
+> Instructions for installing Wine for Box86 can be found [here](https://github.com/ptitSeb/box86/blob/master/docs/X86WINE.md)  

 ----

@@ -76,10 +76,19 @@ If you look at a 64bits version of box86, look at [Box64](https://github.com/pti

 ----

+Notes about Box86 configuration
+----
+
+Box86 now have configurations files. There are 2 files loaded. `/etc/box4.box86rc` and `~/.box86rc`. 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. Box86 comes with a default file that should be installed for better stability. The file in in `system/box86.box86rc` and should be installed to `/etc/box86.box86rc` If, for some reasons, you don't want to install that file here, at least copy it to `~/.box86rc` or some game may not function correctly.
+Note that the priority is: `~/.box86rc` > `/etc/box86.box86rc` > command line
+So, your settings in `~/.box86rc` may override the setting from your command line...
+
+----
+
 Notes about Unity game emulation
 ----

-Running Unity games shoudl generaly work now, but you should also note that many Unity3D games require OpenGL 3+ which can be tricky to provide on ARM SBC (single-board computers) for now.
+Running Unity games should generaly work now, but you should also note that many Unity3D games require OpenGL 3+ which can be tricky to provide on ARM SBC (single-board computers) for now.
 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.

 ----
@@ -87,15 +96,15 @@ Hint: on Pi4, use `MESA_GL_VERSION_OVERRIDE=3.2` and with Panfrost use `PAN_MESA
 Notes about Steam
 ----

-Linux Steam's can run now with box86. But it's still a bit unstable, and not everything works:
- First problem is Steam crashing after the sign-in window, if you encounter this issue, you may need to add libappindicator. To install it on Debian, run `sudo apt install libappindicator1`.
- If you select to "Remember password", Steam is crashing on subsequent starts, unless you have libnm installed. To install it on Debian, run `sudo apt install libnm0`.
+Linux Steam's can run now with box86. but you also need box64 for it to completly usable.
+It is advised to run steam in the Small mode, as it use the less memory, but steamwebhelper (a 64bits process) will still be loaded even if not used.
+The login screen cannot run without steamwebhelper, and will just show a blank windows without box64 properly setup in the system.
+File note, Steam will use a lot of memory, and barely fit on system with 4GB of RAM. It will not work anymore on system with less memory (as a workaround, create a swap file, login and check "remember me", the use box64rc to disable steamwebhelper and run only in small mode without swap after the 1st login)
+Final note: the Steam BigPicture will work, but also need steamwebhelper (and so box64), and lots of memory. It will not start on system with only 4GB of RAM without swap.
 - If you have trouble installing Steam, you can find `install_steam.sh` in the root folder of the box86 repo. This simple script will download and install steam in your home folder, and then create a shortcut to steam in `/usr/local/bin` (and for this it will ask for sudo permission). Simply use `steam` to launch once it's installed. Note that the installation, being in the Home folder, will only work for a single user. Don't use this script if you need a multi-user installation.
- Once open, Steam will only work on "Small Mode" and in "Big Picture", not in the regular "Large Mode". This is because some Steam components used in the browser view are only 64-bit now. So go in the "View" menu and switch to "Small view", else the list will stay empty. Alternatively, Steam can be started in small mode directly by using `+open steam://open/minigameslist` command line arguments.
 - To avoid the "libc.so.6 is absent" message, you can use `STEAMOS=1` and `STEAM_RUNTIME=1` as environment variables (it's automatically there if you used the `install_steam.sh` script)
- Some Steam games (most Source engine games, like "Portal" or "Half-Life 2") use libtcmalloc. Box86 will detect it and will try to LD_PRELOAD it, for better compatibility. While it should work without the aforementioned feature, it is safer to add it to your system if you intend to play those games. To install it on Debian, run `sudo apt install libtcmalloc-minimal4`.

-Steam for Windows installs fine but doesn't work yet.
+If you have issue with steam starting, with `steamwebhelper` not starting, you'll need to start steam with `-cef-disable-gpu` or `-cef-disable-gpu-compositor`.

 ----

@@ -120,8 +129,8 @@ I want to thank everyone who has contributed to box86 development.
 There are many ways to contribute: code, financial, hardware and advertisement!
 So, in no particular order, I want to thank:
 * For their major code contribution: rajdakin, icecream95, M-HT
- * For their major financial contribution: FlyingFathead, stormchaser3000, dennis1248, sll00, [libre-computer-project](https://libre.computer/)
- * For their hardware contribution: [Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/), [DragonBox](https://pyra-handheld.com/), [Novaspirit](https://www.youtube.com/channel/UCrjKdwxaQMSV_NDywgKXVmw), [HardKernel](https://www.hardkernel.com/), [TwisterOS team](https://twisteros.com/)
+ * For their major financial contribution: FlyingFathead, stormchaser3000, dennis1248, sll00, [libre-computer-project](https://libre.computer/), [CubeCoders Limited](http://cubecoders.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), [Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/), [DragonBox](https://pyra-handheld.com/), [Novaspirit](https://www.youtube.com/channel/UCrjKdwxaQMSV_NDywgKXVmw), [HardKernel](https://www.hardkernel.com/), [TwisterOS team](https://twisteros.com/), [AYN](https://www.ayntec.com/)
 * For their continuous advertisement of box86 project: salva ([microLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ)), [PILab](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g)/[TwisterOS team](https://twisteros.com/), [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA), [NicoD](https://www.youtube.com/channel/UCpv7NFr0-9AB5xoklh3Snhg)

 And I also thank the many other people who participated even once in the project.
@@ -2,7 +2,7 @@

 Linux用户空间的x86应用模拟器 

-[更新历史](https://github.com/ptitSeb/box86/blob/master/docs/CHANGELOG.md) | [English](https://github.com/ptitSeb/box86/blob/master/docs/README.md) | [报告错误](https://github.com/ptitSeb/box86/issues/new)
+[更新历史](https://github.com/ptitSeb/box86/blob/master/docs/CHANGELOG.md) | [English](https://github.com/ptitSeb/box86/blob/master/docs/README.md) | [Українська](https://github.com/ptitSeb/box86/blob/master/docs/README_UK.md) | [报告错误](https://github.com/ptitSeb/box86/issues/new)

 ![build](https://app.travis-ci.com/ptitSeb/box86.svg?branch=master) ![stars](https://img.shields.io/github/stars/ptitSeb/box86) ![forks](https://img.shields.io/github/forks/ptitSeb/box86) ![contributors](https://img.shields.io/github/contributors/ptitSeb/box86) ![prs](https://img.shields.io/github/issues-pr/ptitSeb/box86) ![issues](https://img.shields.io/github/issues/ptitSeb/box86)

@@ -127,7 +127,7 @@ Box86已经封装了Vulkan。如果设备的系统有32位的Vulkan驱动程序
 以下感谢不分先后： 
 * 代码贡献：rajdakin, icecream95, M-HT
 * 财务帮助：FlyingFathead, stormchaser3000, dennis1248, sll00, [libre-computer-project](https://libre.computer/)
- * 硬件捐赠：[Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/), [DragonBox](https://pyra-handheld.com/), [Novaspirit](https://www.youtube.com/channel/UCrjKdwxaQMSV_NDywgKXVmw), [HardKernel](https://www.hardkernel.com/), [TwisterOS team](https://twisteros.com/)
+ * 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), [Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/), [DragonBox](https://pyra-handheld.com/), [Novaspirit](https://www.youtube.com/channel/UCrjKdwxaQMSV_NDywgKXVmw), [HardKernel](https://www.hardkernel.com/), [TwisterOS team](https://twisteros.com/)
 * 为本项目持续的宣传：([microLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ)), [PILab](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g)/[TwisterOS team](https://twisteros.com/), [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA), [NicoD](https://www.youtube.com/channel/UCpv7NFr0-9AB5xoklh3Snhg)

 我还要感谢很多人，即使他们只参与了一次本项目。
@@ -0,0 +1,138 @@
+![Official logo](img/Box86Logo.png "Official Logo")
+
+Емулятор простіра користувача x86_64 із родзинкою
+
+[Дивитись журнал змін](https://github.com/ptitSeb/box86/blob/master/docs/CHANGELOG.md) | [中文](https://github.com/ptitSeb/box86/blob/master/docs/README_CN.md) | [English](https://github.com/ptitSeb/box86/blob/master/docs/README.md) | [Повідомити про помилку](https://github.com/ptitSeb/box86/issues/new)
+
+![build](https://app.travis-ci.com/ptitSeb/box86.svg?branch=main) ![stars](https://img.shields.io/github/stars/ptitSeb/box86) ![forks](https://img.shields.io/github/forks/ptitSeb/box86) ![contributors](https://img.shields.io/github/contributors/ptitSeb/box86) ![prs](https://img.shields.io/github/issues-pr/ptitSeb/box86) ![issues](https://img.shields.io/github/issues/ptitSeb/box86)
+
+----
+
+Box86 дозволяє запускати x86 Linux програми (наприклад, ігри) на системах Linux відмінних від x86, наприклад ARM (хост-система має мати 32-розрядний порядок редагування).
+
+Вам *ПОТРІБНА* 32-розрядна підсистема для запуску та створення Box86. Box86 марний лише на 64-розрядних системах. Крім того, вам *ПОТРІБЕН* 32-розрядний інструментарій для створення Box86. Ланцюжок інструментів, який підтримує лише 64-розрядні версії, не компілює Box86, і ви отримаєте помилки (зазвичай на aarch64 ви отримуєте що «-marm» не розпізнається, і вам знадобиться середовище з мультиархітектурою або chroot).
+
+Оскільки Box86 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/). Це також означає, що вам знадобиться 32-бітний простір користувача на 64-бітній ОС, наприклад `armhf` поверх `aarch64` 64-бітної ОС.
+
+Більшість x86-ігор потребують OpenGL, тому на платформах ARM може знадобитися таке рішення, як [gl4es](https://github.com/ptitSeb/gl4es). (Деякі платформи ARM підтримують лише OpenGL ES і/або їх реалізація OpenGL хитра. (див. OpenGL на Android))
+
+Box86 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/).
+
+Багато ігор просто працюють без особливих налаштувань, наприклад: WorldOfGoo, Airline Tycoon Deluxe і FTL. Багато ігор GameMaker Linux також працюють нормально. (є довгий список, серед них UNDERTALE, A Risk of Rain і Cook Serve Delicious). Ігри Unity3D також працюють добре, але вимога OpenGL може бути проблемою на деяких платформах ARM.
+
+Якщо ви серйозно ставитеся до розробки Box86, вам слід встановити ccache та створити Box86 з ним. (Використовуйте, наприклад, ccmake.)
+Щоб увімкнути TRACE (тобто скидання в stdout усіх окремих виконаних інструкцій x86 із дампом регістрів), вам також знадобиться [бібліотека Zydis](https://github.com/zyantific/zydis), доступна у вашій системі.
+
+Деякі внутрішні коди операцій x86 використовують частини «Бібліотеки емулятора Realmode X86», дивіться [x86primop.c](../src/emu/x86primop.c) для детальної інформації про авторські права.
+
+----
+Компіляція/Встановлення
+----
+> Інструкцію зі складання Box86 можна знайти [тут](COMPILE.md).\
+> Інструкцію з встановлення Wine для Box86 можна знайти [тут](X86WINE.md).
+
+----
+
+Ось 6 відео, перші 2 відео – це відео "Airline Tycoon Deluxe" і "Heretic 2", які працюють на GigaHertz OpenPandora (друге використовує Dynarec), а наступні 2 відео – це відео "Bit.Trip.Runner" і "Neverwinter Night", які працюють на ODroid XU4 (без dynarec), а останні 2 відео на Pi4: Shovel Knight (відео від @ITotalJustice) і Freedom Planet (відео від @djazz), також без Dynarec.
+
+
+[![Play on Youtube](https://img.youtube.com/vi/bLt0hMoFDLk/3.jpg)](https://www.youtube.com/watch?v=bLt0hMoFDLk) [![Play on Youtube](https://img.youtube.com/vi/MM7kWYts7IA/3.jpg)](https://www.youtube.com/watch?v=MM7kWYts7IA) [![Play on Youtube](https://img.youtube.com/vi/8hr71S029Hg/1.jpg)](https://www.youtube.com/watch?v=8hr71S029Hg) [![Play on Youtube](https://img.youtube.com/vi/B4YN37z3-ws/1.jpg)](https://www.youtube.com/watch?v=B4YN37z3-ws) [![Play on Youtube](https://img.youtube.com/vi/xk8Q30mxqPg/1.jpg)](https://www.youtube.com/watch?v=xk8Q30mxqPg) [![Play on Youtube](https://img.youtube.com/vi/_QMRMVvYrqU/1.jpg)](https://www.youtube.com/watch?v=_QMRMVvYrqU)
+
+Ви можете знайти багато інших відео про Box86 у 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).
+
+Список сумісності доступний тут: https://github.com/ptitSeb/box86-compatibility-list/issues
+
+<img src="img/Box86Icon.png" width="96" height="96">
+
+Логотип і піктограма зроблені @grayduck, дякую!
+
+Зауважте, що цей проект не слід плутати з [86box](https://github.com/86Box/86Box), гарним емулятором "повної системи", що спеціалізується на ранньому (до відносно недавньому) апаратному забезпеченні ПК.
+
+----
+
+Використання
+----
+
+Є кілька змінних середовища для керування поведінкою Box86.
+
+Дивіться [тут](USAGE.md) для перегляду всіх змінних середовища та того, що вони роблять.
+
+Примітка: Dynarec від Box86 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.
+
+----
+
+Історія версій/Журнал змін
+----
+
+Історія версій доступна [тут](docs/CHANGELOG.md).
+
+----
+
+Примітки щодо 64-розрядних платформ
+----
+
+Оскільки Box86 працює шляхом прямого перекладу викликів функцій із x86 на хост-систему, хост-система (на якій працює Box86) повинна мати 32-розрядні бібліотеки. Box86 не містить 32-біт <-> 64-біт перекладу. Отже, щоб запустити Box86, наприклад, на платформі ARM64, вам потрібно буде зібрати Box86 для 32-бітної ARM, а також потрібно мати chroot з 32-бітними бібліотеками.
+
+Якщо ви дивитесь на 64-розрядну версію box86, подивіться на [Box64](https://github.com/ptitSeb/box64): він може запускати двійкові файли x86_64 на 64-розрядних платформах. Але зауважте, що вам все одно потрібен Box86 (і 32-розрядний chroot) для запуску бінарних файлів x86 (як це також трапляється в реальному Linux x86_64, для запуску якого потрібні бібліотеки x86 і двійковий файл в multiarch).
+
+----
+
+Примітки щодо конфігурації Box86
+----
+
+Box86 тепер має конфігураційні файли. Завантажуються 2 файли: `/etc/box4.box86rc` і `~/.box86rc`. Обидва файли мають однаковий синтаксис і в основному є файлами ini. Розділ у квадратних дужках визначає назву процесу, а решта — встановлють змінну середовища. Подивіться на [використання](docs/USAGE.md), щоб дізнатися, які параметри можна використовувати. Box86 поставляється з файлом за замовчуванням, який слід інсталювати для кращої стабільності. Файл знаходиться в `system/box86.box86rc` і має бути встановлено в `/etc/box86.box86rc`. Якщо з певних причин ви не хочете встановлювати цей файл у `/etc`, принаймні скопіюйте його в `~/.box86rc` або гра може не працювати належним чином. Зауважте, що пріоритетом є: `~/.bashrc` > `/etc/box86.box86rc` > 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 Steam
+----
+
+Linux версія Steam тепер може працювати з Box86, але вам також потрібен Box64, щоб його можна було повністю використовувати.
+Рекомендується запускати Steam у small режимі, оскільки він використовує менше пам’яті, але steamwebhelper (64-бітний процес) усе одно завантажуватиметься, навіть якщо він не використовується.
+Екран входу не може працювати без steamwebhelper, і буде показувати лише порожні вікна без належного налаштування box64 у системі.
+Зауважте, що Steam використовуватиме багато пам’яті та ледве поміститься в систему з 4 ГБ оперативної пам’яті. Він більше не працюватиме в системі з меншим об’ємом пам’яті (як обхідний шлях, створіть файл підкачки, увійдіть і поставте прапорець «запам’ятати мене», використовуйте box64rc, щоб вимкнути steamwebhelper і запускати лише в малому режимі без підкачки після першого входу в систему)
+Останнє зауваження: Steam BigPicture працюватиме, але також потрібен steamwebhelper (і box64) і багато пам’яті. Він не запуститься в системі лише з 4 ГБ оперативної пам’яті без файлу підкачки.
+- Якщо у вас виникли проблеми зі встановленням Steam, ви можете знайти `install_steam.sh` у кореневій папці сховища box86. Цей простий скрипт завантажить і встановить Steam у вашу домашню папку, а потім створить ярлик steam у `/usr/local/bin` (для цього він запитає дозвіл sudo). Просто використовуйте `steam` для запуску після встановлення. Зауважте, що інсталяція, яка знаходиться в папці Home, працюватиме лише для одного користувача. Не використовуйте цей скрипт, якщо вам потрібна багатокористувацька інсталяція.
+- Щоб уникнути повідомлення «libc.so.6 відсутній», ви можете використовувати `STEAMOS=1` і `STEAM_RUNTIME=1` як змінні середовища (воно автоматично з’являється, якщо ви використовували сценарій `install_steam.sh`)
+
+Якщо у вас виникла проблема із запуском Steam (не запускається `steamwebhelper`), вам потрібно запустити Steam за допомогою `-cef-disable-gpu` або `-cef-disable-gpu-compositor`.
+
+----
+
+Примітки щодо Wine
+----
+
+Wine тепер підтримується. Інтегровани програми Wine працює як і багато програм та ігор для Windows. Не забувайте, що більшість ігор для Windows використовують Direct3D, для цього може знадобитися повний драйвер OpenGL і якомога більш високий профіль (і gl4es з бекендом ES2 наразі мають проблеми з Wine).
+Примітка: якщо ви плануєте використовувати box86 з Wine на Raspberry Pi 3 або ранішої версії, ці моделі використовують ОС за замовчуванням, яка має ядро з розділенням 2/2 (тобто 2G місця для програми користувача та 2 Г місця для ядра). Це несумісно з програмами Wine, яким потрібен доступ до пам’яті > 2 Гб адреси. Тож вам потрібно буде переналаштувати ядро для розділення 3G/1G.
+
+----
+
+Примітки щодо Vulkan
+----
+
+Box86 вже використовує Vulkan. Якщо у вашій системі є 32-розрядний драйвер Vulkan, Box86 використовуватиме його за потреби. Профіль 1.0, 1.1, 1.2 і 1.3, з деякими розширеннями, має бути в порядку. DXVK, включаючи 2.0, працює. Я знаю, що деякі демонстраційні версії працюють на Pi4 (демонстраційні версії Sascha Willems для x86 працюють так само, як якщо б створювати безпосередньо на armhf). Зауважте, що драйвер Vulkan для Pi4 наразі НЕ підтримує DXVK (обгортка Wine DirectX->Vulkan). Це не проблема Box86, у драйвері відсутні розширення (підтримка апаратного забезпечення) та кілька інших речей, через які DXVK не працює на pi4. З боку Panfrost, PanVK трохи молодий, і я ще не тестував DXVK з ним.
+
+----
+Останнє слово
+----
+
+Я хочу подякувати всім, хто зробив внесок у розвиток box86.
+Є багато способів зробити свій внесок: код, фінанси, апаратне забезпечення та реклама!
+Отже, без особливого порядку, я хочу подякувати:
+ * За основний внесок у код: rajdakin, icecream95, M-HT
+ * За великий фінансовий внесок: FlyingFathead, stormchaser3000, dennis1248, sll00, [libre-computer-project](https://libre.computer/)
+ * 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), [Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/), [DragonBox](https://pyra-handheld.com/), [Novaspirit](https://www.youtube.com/channel/UCrjKdwxaQMSV_NDywgKXVmw), [HardKernel](https://www.hardkernel.com/), [Команда TwisterOS](https://twisteros.com/)
+ * За постійну рекламу Box86: 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)
+
+І я також дякую багатьом іншим людям, які хоч раз брали участь у проекті.
+
+(Якщо ви використовуєте Box86 у своєму проекті, будь ласка, не забудьте згадати про це!)
@@ -3,20 +3,21 @@ Usage
 ----

 There are many environment variables to control Box86 behaviour. 
+Env. var with * can also be put inside box64rc files

-#### BOX86_LOG
+#### BOX86_LOG *
 Controls the Verbosity level of the logs
 * 0: NONE : No message (except some fatal error). (Default.)
 * 1: INFO : Show some minimum log (Example: librairies not found)
 * 2: DEBUG : Details a lot of stuff (Example: relocations or functions called).
 * 3: DUMP : All DEBUG plus DUMP of all ELF Info (so same as `BOX86_LOG=2 BOX86_DUMP=1`)

-#### BOX86_DUMP
+#### BOX86_DUMP *
 Controls the Dump of elf content
 * 0: No dump of elf information (Default)
 * 1: Dump elf sections and relocations and other information (also active if using `BOX86_LOG=3`)

-#### BOX86_ROLLING_LOG
+#### BOX86_ROLLING_LOG *
 Show last few wrapped function call when a Signal is caught
 * 0: No last function call printed (Default.)
 * 1: Last 16 wrapped functions calls printed when a signal is printed. Incompatible with BOX86_LOG>1 (may need BOX86_SHOWSEGV=1 also)
@@ -27,177 +28,263 @@ Disables Box86 printing its version and build
 * 0 : Enable printing its banner. (Default.)
 * 1 : Disable printing its banner. 

-#### BOX86_LD_LIBRARY_PATH
+#### BOX86_LD_LIBRARY_PATH *
 Path to look for x86 libraries. Default is current folder and `lib` in current folder.
-Also, `/usr/lib/i386-linux-gnu` and `/lib/i386-linux-gnu` are added if they exist.
+Also, `/usr/lib/i386-linux-gnu`, `/lib/i386-linux-gnu`, and `/usr/lib/box86-i386-linux-gnu` are added if they exist.

-#### BOX86_PATH
+#### BOX86_PATH *
 Path to look for x86 executable. Default is current folder and `bin` in current folder.

-#### BOX86_DLSYM_ERROR
+#### BOX86_DLSYM_ERROR *
 Enables/Disables the logging of `dlsym` errors.
 * 0 : Don't log `dlsym` errors. (Default.)
 * 1 : Log dlsym errors.

-#### BOX86_TRACE_FILE
+#### BOX86_TRACE_FILE *
 Send all log and trace to a file instead of `stdout`
 Also, if name contains `%pid` then this is replaced by the actual PID of box86 instance
 Use `stderr` to use this instead of default `stdout` 

-#### BOX86_TRACE
+#### BOX86_TRACE *
 Only on build with trace enabled. Trace allow the logging of all instruction executed, along with register dump
 * 0 : No trace. (Default.) 
 * 1 : Trace enabled. Trace start after the initialisation of all depending libraries is done.
 * symbolname : Trace only `symbolname` (trace is disable if the symbol is not found).
 * 0xXXXXXXX-0xYYYYYYY : Trace only between the 2 addresses.

-#### BOX86_TRACE_INIT
+#### BOX86_TRACE_INIT *
 Use BOX86_TRACE_INIT instead of BOX_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.

-#### BOX86_TRACE_START
+#### BOX86_TRACE_START *
 Only on builds with trace enabled.
 * NNNNNNN : Start trace only after NNNNNNNN opcode execute (number is an `uint64_t`).

-#### BOX86_TRACE_XMM
+#### BOX86_TRACE_XMM *
 Only on builds with trace enabled.
 * 0 : The XMM (i.e. SSE/SSE2) register will not be logged with the general and x86 registers. (Default.)
 * 1 : Dump the XMM registers.

-#### BOX86_LOAD_ADDR
+#### BOX86_TRACE_EMM *
+Only on builds with trace enabled.
+ * 0 : The EMM (i.e. MMX) register will not be logged with the general and x86 registers. (Default.)
+ * 1 : Dump the EMM registers.
+
+#### BOX86_LOAD_ADDR *
 Try to load at 0xXXXXXX main binary (if binary is a PIE)
 * 0xXXXXXXXX : The load address . (Only active on PIE programs.)

-#### BOX86_NOSIGSEGV
+#### BOX86_NOSIGSEGV *
 Disable handling of SigSEGV. (Very useful for debugging.)
 * 0 : Let the x86 program set sighandler for SEGV (Default.)
 * 1 : Disable the handling of SigSEGV.

-#### BOX86_NOSIGILL
+#### BOX86_NOSIGILL *
 Disable handling of SigILL (to ease debugging mainly).
 * 0 : Let x86 program set sighandler for Illegal Instruction
 * 1 : Disables the handling of SigILL 

-#### BOX86_X11COLOR16
+#### BOX86_X11COLOR16 *
 PANDORA only: Try converting X11 color from 32 bits to 16 bits (to avoid light green on light cyan windows).
 * 0 : Don't touch X11 colors. (Default.)
 * 1 : Change colors arguments in XSetForeground, XSetBackground and XCreateGC.

-#### BOX86_X11THREADS
+#### BOX86_X11THREADS *
 Call XInitThreads when loading X11. (This is mostly for old Loki games with the Loki_Compat library.)
 * 0 : Don't force call XInitThreads. (Default.)
 * 1 : Call XInitThreads as soon as libX11 is loaded.

-#### BOX86_X11GLX
+#### BOX86_X11GLX *
 Force libX11's GLX extension to be present.
 * 0 : Do not force libX11's GLX extension to be present. 
 * 1 : GLX will always be present when using XQueryExtension. (Default.)

-#### BOX86_DYNAREC_DUMP
+#### BOX86_DYNAREC_DUMP *
 Enables/disables Box86's Dynarec's dump.
 * 0 : Disable Dynarec's blocks dump. (Default.)
 * 1 : Enable Dynarec's blocks dump.
 * 2 : Enable Dynarec's blocks dump with some colors.

-#### BOX86_DYNAREC_LOG
+#### BOX86_DYNAREC_LOG *
 Set the level of DynaRec's logs.
 * 0 : NONE : No Logs for DynaRec. (Default.)
 * 1 :INFO : Minimum Dynarec Logs (only unimplemented OpCode).
 * 2 : DEBUG : Debug Logs for Dynarec (with details on block created / executed).
 * 3 : VERBOSE : All of the above plus more.

-#### BOX86_DYNAREC
+#### BOX86_DYNAREC *
 Enables/Disables Box86's Dynarec.
 * 0 : Disables Dynarec.
 * 1 : Enable Dynarec. (Default.)

-#### BOX86_DYNAREC_TRACE
+#### BOX86_DYNAREC_TRACE *
 Enables/Disables trace for generated code.
 * 0 : Disable trace for generated code. (Default.)
 * 1 : Enable trace for generated code (like regular Trace, this will slow down the program a lot and generate huge logs).

-#### BOX86_NODYNAREC 
+#### BOX86_NODYNAREC *
 Forbid dynablock creation in the interval specified (helpfull for debugging behaviour difference between Dynarec and Interpretor)
 * 0xXXXXXXXX-0xYYYYYYYY : define the interval where dynablock cannot start (inclusive-exclusive)

-#### BOX86_DYNAREC_BIGBLOCK
+#### BOX86_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
+
+#### BOX86_DYNAREC_BIGBLOCK *
 Enables/Disables Box86's Dynarec building BigBlock.
 * 0 : Don't try to build block as big as possible (can help program using lots of thread and a JIT, like C#/Unity) (Default when libmonobdwgc-2.0.so is loaded)
 * 1 : Build Dynarec block as big as possible (Default.)
 * 2 : Build Dynarec block bigger (don't stop when block overlaps)

-#### BOX86_DYNAREC_STRONGMEM
+#### BOX86_DYNAREC_FORWARD *
+Define Box86's Dynarec max allowed forward value when building Block.
+ * 0 : No forward value. When current block end, don't try to go further even if there are previous forward jumps
+ * XXX : Allow up to XXXX bytes of gap when building a Block after the block end to next forward jump (Default: 128)
+ 
+#### BOX86_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

-#### BOX86_DYNAREC_X87DOUBLE
+#### BOX86_DYNAREC_X87DOUBLE *
 Force the use of Double for x87 emulation
 * 0 : Try to use float when possible for x87 emulation (faster, Default)
 * 1 : Only use Double for x87 emulation (slower, may be needed for some specific games, like Crysis were it's enable automatically)

-#### BOX86_DYNAREC_FASTNAN
+#### BOX86_DYNAREC_FASTNAN *
 Enable/Disable generation of -NAN
 * 0 : Generate -NAN like on x86 (slower, more x86 accurate)
 * 1 : Don't do anything special with NAN, to go as fast as possible (Default)

-#### BOX86_DYNAREC_SAFEFLAGS
+#### BOX86_DYNAREC_FASTROUND *
+Enable/Disable generation of precise x86 rounding
+* 0 : Generate float/double -> int rounding and use current rounding mode for float/double compution like on x86
+* 1 : Don't do anything special with edge case Rounding, to go as fast as possible (no INF/NAN/Overflow -> MIN_INT conversion, and no set rounding mode) (faster, Default)
+
+#### BOX86_DYNAREC_SAFEFLAGS *
 Handling of flags on CALL/RET opcodes
 * 0 : Treat CALL/RET as if it never needs any flags (faster but may have side-effects)
 * 1 : most of RET will need flags, most of CALLS will not (Default)
 * 2 : All CALL/RET will need flags (slower, but might be needed. Automatically enabled for Vara.exe)

-#### BOX86_LIBGL
+#### BOX86_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
+
+#### BOX86_DYNAREC_BLEEDING_EDGE *
+Detect MonoBleedingEdge and apply conservative settings
+* 0 : Don't detect MonoBleedingEdge
+* 1 : Detect MonoBleedingEdge, and apply BIGBLOCK=0 STRONGMEM=1 if detected (Default)
+
+#### BOX86_DYNAREC_JVM *
+Detect libjvm and apply conservative settings
+* 0 : Don't detect libjvm
+* 1 : Detect libjvm, and apply BIGBLOCK=0 STRONGMEM=1 if detected (Default)
+
+#### BOX86_DYNAREC_WAIT *
+Behavior with FillBlock is not availble (FillBlock build Dynarec blocks and is not multithreaded)
+* 0 : Dynarec will not wait for FillBlock to ready and use Interpreter instead (might speedup a bit massive multithread or JIT programs)
+* 1 : Dynarec will wait for FillBlock to be ready (Default)
+
+#### BOX86_DYNAREC_MISSING *
+Dynarec print the missing opcodes
+* 0 : not print the missing opcode (Default, unless DYNAREC_LOG>=1 or DYNAREC_DUMP>=1 is used)
+* 1 : Will print the missing opcodes
+
+#### BOX86_SSE_FLUSHTO0 *
+Handling of SSE Flush to 0 flags
+* 0 : Just track the flag (Default)
+* 1 : Direct apply of SSE Flush to 0 flag
+
+#### BOX86_X87_NO80BITS *
+Handling of x87 80bits long double
+* 0 : Try to handle 80bits long double as precise as possible (Default)
+* 1 : Handle them as double
+
+#### BOX86_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)
+
+#### BOX86_LIBCEF *
+Detect libcef and apply malloc_hack settings
+* 0 : Don't detect libcef
+* 1 : Detect libcef, and apply MALLOC_HACK=2 if detected (Default)
+
+#### BOX86_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
+
+#### BOX86_MUTEX_ALIGNED *
+Will mutex are used as-is or wrapped to handle unaligned used
+* 0 : Mutex will be wrapped in case unaligned mutexes are used (Default)
+* 1 : Do no wrap mutex and use them as-is (faster, but might crash with SEGBUS error)
+
+#### BOX86_LIBGL *
 * libXXXX set the name for libGL (defaults to libGL.so.1).
 * /PATH/TO/libGLXXX : Sets the name and path for libGL
 You can also use SDL_VIDEO_GL_DRIVER

-#### BOX86_LD_PRELOAD
+#### BOX86_LD_PRELOAD *
 * XXXX[:YYYYY] force loading XXXX (and YYYY...) libraries with the binary
- PreLoaded libs can be emulated or native, and are treated the same way as if they were comming from the binary
+ PreLoaded libs can be emulated or native, and are treated the same way as if they were comming from the binary (not that using this inside a rcfile might not work if the process needs to be renamed, for example for Wine process)
 
-#### BOX86_EMULATED_LIBS
+#### BOX86_EMULATED_LIBS *
 * XXXX[:YYYYY] force lib XXXX (and YYYY...) to be emulated (and not wrapped)
 Some games uses an old version of some libraries, with an ABI incompatible with native version.
 Note that LittleInferno for example is auto detected, and libvorbis.so.0 is automatical added to emulated libs, and same for Don't Starve (and Together / Server variant) that use an old SDL2 too

-#### BOX86_ALLOWMISSING_LIBS
+#### BOX86_ALLOWMISSING_LIBS *
 Allow Box86 to continue even if a library is missing (renamed from BOX86_ALLOWMISSINGLIBS). 
 * 0 : Box86 will stop if a library cannot be loaded. (Default.)
 * 1 : Continue even if a needed library cannot be loaded. Unadvised, this will, in most cases, crash later on.

-#### BOX86_ALLOWMISSING_SYMBOLS
+#### BOX86_ALLOWMISSING_SYMBOLS *
 Allow dlopen with RTLD_NOW flag to not resolve all symbols. 
 * 0 : Box86 will stop if a library have unresolved symbols. (Default.)
 * 1 : Continue even if a needed library have unresolved symbols (like symbols not yet wrapped).

-#### BOX86_PREFER_WRAPPED
+#### BOX86_PREFER_WRAPPED *
 Box86 will use wrapped libs even if the lib is specified with absolute path
 * 0 : Try to use emulated libs when they are defined with absolute path  (Default.)
 * 1 : Use Wrapped native libs even if path is absolute

-#### BOX86_PREFER_EMULATED
+#### BOX86_PREFER_EMULATED *
 Box86 will prefer emulated libs first (execpt for glibc, alsa, pulse, GL, vulkan and X11
 * 0 : Native libs are prefered (Default.)
 * 1 : Emulated libs are prefered (Default for program running inside pressure-vessel)

-#### BOX86_NOPULSE
+#### BOX86_NOPULSE *
 Disables the load of pulseaudio libraries.
 * 0 : Load pulseaudio libraries if found. (Default.)
 * 1 : Disables the load of pulse audio libraries (libpulse and libpulse-simple), both the native library and the x86 library

-#### BOX86_NOGTK
+#### BOX86_NOGTK *
 Disables the loading of wrapped GTK libraries.
 * 0 : Load wrapped GTK libraries if found. (Default.)
 * 1 : Disables loading wrapped GTK libraries.

-#### BOX86_NOVULKAN
+#### BOX86_NOVULKAN *
 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.)

+#### BOX86_NOCRASHHANDLER *
+Diables crashhandler.so
+* 0 : Regular process for crashhandler.so (Default)
+* 1 : Do not load crashhandler.so, even if present
+
+#### BOX86_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)
+
 #### BOX86_ENV
 * XXX=yyyy
 will add XXX=yyyy env. var.
@@ -206,22 +293,27 @@ Disables the load of vulkan libraries.
 * XXX=yyyy
 will add XXX=yyyy env. var. and continue with BOX86_ENV2 ... until var doesn't exist

-#### BOX86_FIX_64BIT_INODES
+#### BOX86_ARG0
+Define argv[0] of the executable.
+ * yyyy
+Will rewrite argv[0] instead of using program path.
+
+#### BOX86_FIX_64BIT_INODES *
 Fix/Don't fix 64bit inodes
 * 0 : Don't fix 64bit inodes. (Default.)
 * 1 : Fix 64bit inodes. Helps when running on Filesystems with 64bit inodes. Is useful when a program uses API functions which doesn't support it and the program doesn't use inodes information.

-#### BOX86_SHOWSEGV
+#### BOX86_SHOWSEGV *
 Show Segfault signal even if a signal handler is present
 * 0 : Don't force segfault details when a signal handler is install (Default, to limit message spamming)
 * 1 : Show segfault details

-#### BOX86_BACKTRACE
+#### BOX86_SHOWBT *
 Show a backtrace (native and emulated) on Signal
 * 0 : Don't show any backtrace (Default.)
 * 1 : Show native backtrace when a signal is dumped (not available on Andoid build)

-#### BOX86_BASH
+#### BOX86_BASH *
 Define x86 bash to launch script
 * yyyy
 Will use yyyy as x86 bash to launch script. yyyy needs to be a full path to a valid x86 version of bash. Note that is Box64 is available, an x86_64 version of bash can also be used. Also, BOX64_BASH env. var. will also be read if BOX86_BASH is not defined.
@@ -229,8 +321,9 @@ Will use yyyy as x86 bash to launch script. yyyy needs to be a full path to a va
 #### BOX86_BOX64
 Define where to find box64 binary
 * yyyy
-Will use yyyy as x86_64 interpretor, to launch x64_64 binaries
-#### BOX86_JITGDB
+Will use yyyy as x86_64 interpreter, to launch x64_64 binaries
+
+#### BOX86_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.
@@ -241,3 +334,12 @@ Will use yyyy as x86_64 interpretor, to launch x64_64 binaries
 * 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/box86` 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 ` BOX86_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.
+
+#### BOX86_NORCFILES
+If the env var exist, no rc files (like /etc/box86.box86rc and ~/.box86rc) will be loaded
+
+#### BOX86_NOSANDBOX
+ * 0 : Nothing special
+ * 1 : Added "--no-sandbox" to command line arguments (usefull for chrome based programs)
+
@@ -42,7 +42,7 @@ rm ~/wine.tgz # clean up

 # Install shortcuts (make launcher & symlinks. Credits: grayduck, Botspot)
 echo -e '#!/bin/bash\nsetarch linux32 -L '"$HOME/wine/bin/wine "'"$@"' | sudo tee -a /usr/local/bin/wine >/dev/null # Create a script to launch wine programs as 32bit only
-#sudo ln -s ~/wine/bin/wine /usr/local/bin/wine # You could aslo just make a symlink, but box86 only works for 32bit apps at the moment
+#sudo ln -s ~/wine/bin/wine /usr/local/bin/wine # You could also just make a symlink, but box86 only works for 32bit apps at the moment
 sudo ln -s ~/wine/bin/wineboot /usr/local/bin/wineboot
 sudo ln -s ~/wine/bin/winecfg /usr/local/bin/winecfg
 sudo ln -s ~/wine/bin/wineserver /usr/local/bin/wineserver
@@ -112,7 +112,7 @@ if [ "$karch" = "aarch64" ] || [ "$karch" = "aarch64-linux-gnu" ] || [ "$karch"
        libdbus-1-3:armhf libfontconfig1:armhf libfreetype6:armhf libglu1-mesa:armhf libglu1:armhf libgnutls30:armhf \
        libgssapi-krb5-2:armhf libkrb5-3:armhf libodbc1:armhf libosmesa6:armhf libsdl2-2.0-0:armhf libv4l-0:armhf \
        libxcomposite1:armhf libxcursor1:armhf libxfixes3:armhf libxi6:armhf libxinerama1:armhf libxrandr2:armhf \
-        libxrender1:armhf libxxf86vm1 libc6:armhf libcap2-bin:armhf # to run wine-i386 through box86:armhf on aarch64
+        libxrender1:armhf libxxf86vm1:armhf libc6:armhf libcap2-bin:armhf # to run wine-i386 through box86:armhf on aarch64
        # This list found by downloading...
        #	wget https://dl.winehq.org/wine-builds/debian/dists/bullseye/main/binary-i386/wine-devel-i386_7.1~bullseye-1_i386.deb
        #	wget https://dl.winehq.org/wine-builds/debian/dists/bullseye/main/binary-i386/winehq-devel_7.1~bullseye-1_i386.deb
@@ -18,22 +18,33 @@ mv ./usr/* ../
 cd ../ && rm -rf ./tmp/

 # create run script
-echo "#!/bin/bash
+echo '#!/bin/bash
 export STEAMOS=1
 export STEAM_RUNTIME=1
-~/steam/bin/steam -noreactlogin steam://open/minigameslist $@" > steam
+export DBUS_FATAL_WARNINGS=0
+~/steam/bin/steam $@' > steam

 # make script executable and move
 chmod +x steam
 sudo mv steam /usr/local/bin/

-# detect if we're running on 64 bit Debian 
+# detect if we're running on 64 bit Debian (maybe this can be moved in another script then mentioned in documentation for the people that don't use steam but don't want a headache)
 MACHINE_TYPE=`uname -m`
 if [ ${MACHINE_TYPE} == 'aarch64' ] && [ -f '/etc/debian_version' ]; then
 echo "Detected 64 bit ARM Debian. Installing 32 bit libraries"
 sudo dpkg --add-architecture armhf # enable installation of armhf libraries
 sudo apt update # update package lists with the newly added arch
- sudo apt install libc6:armhf libncurses5:armhf libsdl2*:armhf libopenal*:armhf libpng*:armhf libfontconfig*:armhf libXcomposite*:armhf libbz2-dev:armhf libXtst*:armhf # install the libraries that Steam requires
+ # install the libraries that Steam requires
+ sudo apt install libc6:armhf libsdl2-2.0-0:armhf libsdl2-image-2.0-0:armhf libsdl2-mixer-2.0-0:armhf libsdl2-ttf-2.0-0:armhf libopenal1:armhf libpng16-16:armhf libfontconfig1:armhf libxcomposite1:armhf libbz2-1.0:armhf libxtst6:armhf libsm6:armhf libice6:armhf libgl1:armhf libxinerama1:armhf libxdamage1:armhf libibus-1.0-5
+ # this one is not there all the time, so just try it...
+ sudo apt install libncurses5:armhf 
+
+ # install mesa for armhf if already installed
+ if dpkg-query -W libgl1-mesa-dri 2>/dev/null; then
+  echo "Detected mesa driver for 64 bit ARM. Installing complementary 32 bit one"
+  sudo apt install libgl1-mesa-dri:armhf
+ fi
+
 echo "Don't forget to compile/install Box64!"
 fi

@@ -1,7 +1,7 @@
 # Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
 # Author: Sebastien Chevalier <[email protected]>
-pkgname=box86-rk3399-git
-pkgver=r3858.5368285
+pkgname=box86-a64-git
+pkgver=r4581.d698861
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -9,21 +9,28 @@ url="https://github.com/ptitSeb/box86"
 license=('MIT')
 optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
 makedepends=('git' 'cmake' 'make')
-provides=("${pkgname%-git}" "${pkgname%-rk3399-git}-git" "${pkgname%-rk3399-git}")
-conflicts=("${pkgname%-git}" "${pkgname%-rk3399-git}-git" "${pkgname%-rk3399-git}")
+provides=("${pkgname%-git}" "${pkgname%-a64-git}-git" "${pkgname%-a64-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-a64-git}-git" "${pkgname%-a64-git}")
 source=('git+https://github.com/ptitSeb/box86')
 md5sums=('SKIP')

 pkgver() {
-	cd "$srcdir/${pkgname%-rk3399-git}"
+	cd "${srcdir}/${pkgname%-a64-git}"
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }

+prepare() {
+	cd "${srcdir}/${pkgname%-a64-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DA64=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
 build() {
-	cmake -B build -S "${pkgname%-rk3399-git}" -DRK3399=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
-	make -C build
+	cd "${srcdir}/${pkgname%-a64-git}/build"
+	make -j$(nproc)
 }

 package() {
-	make -C build DESTDIR="${pkgdir}/" install
-}
+	cd "${srcdir}/${pkgname%-a64-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}
@@ -1,7 +1,7 @@
 # Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-arm-git
-pkgver=r3858.5368285
+pkgver=r4581.d698861
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -15,15 +15,22 @@ source=('git+https://github.com/ptitSeb/box86')
 md5sums=('SKIP')

 pkgver() {
-	cd "$srcdir/${pkgname%-arm-git}"
+	cd "${srcdir}/${pkgname%-arm-git}"
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }

+prepare() {
+	cd "${srcdir}/${pkgname%-arm-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
 build() {
-	cmake -B build -S "${pkgname%-arm-git}" -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
-	make -C build
+	cd "${srcdir}/${pkgname%-arm-git}/build"
+	make -j$(nproc)
 }

 package() {
-	make -C build DESTDIR="${pkgdir}/" install
-}
+	cd "${srcdir}/${pkgname%-arm-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}
@@ -0,0 +1,36 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-arm64-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('aarch64')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-arm64-git}-git" "${pkgname%-arm64-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-arm64-git}-git" "${pkgname%-arm64-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "${srcdir}/${pkgname%-arm64-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-arm64-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DARM64=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-arm64-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-arm64-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}
@@ -1,7 +1,7 @@
 # Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-gameshell-git
-pkgver=r3858.5368285
+pkgver=r4581.d698861
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -15,15 +15,22 @@ source=('git+https://github.com/ptitSeb/box86')
 md5sums=('SKIP')

 pkgver() {
-	cd "$srcdir/${pkgname%-gameshell-git}"
+	cd "${srcdir}/${pkgname%-gameshell-git}"
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }

+prepare() {
+	cd "${srcdir}/${pkgname%-gameshell-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DGAMESHELL=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
 build() {
-	cmake -B build -S "${pkgname%-gameshell-git}" -DGAMESHELL=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
-	make -C build
+	cd "${srcdir}/${pkgname%-gameshell-git}/build"
+	make -j$(nproc)
 }

 package() {
-	make -C build DESTDIR="${pkgdir}/" install
-}
+	cd "${srcdir}/${pkgname%-gameshell-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}
@@ -0,0 +1,36 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-goaclone-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-goaclone-git}-git" "${pkgname%-goaclone-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-goaclone-git}-git" "${pkgname%-goaclone-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "${srcdir}/${pkgname%-goaclone-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-goaclone-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DGOA_CLONE=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-goaclone-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-goaclone-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}
@@ -0,0 +1,36 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-lx2610a-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-lx2610a-git}-git" "${pkgname%-lx2610a-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-lx2610a-git}-git" "${pkgname%-lx2610a-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "${srcdir}/${pkgname%-lx2610a-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-lx2610a-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DLX2610A=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-lx2610a-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-lx2610a-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}
@@ -1,7 +1,7 @@
 # Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-odroid-git
-pkgver=r3858.5368285
+pkgver=r4581.d698861
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -15,15 +15,22 @@ source=('git+https://github.com/ptitSeb/box86')
 md5sums=('SKIP')

 pkgver() {
-	cd "$srcdir/${pkgname%-odroid-git}"
+	cd "${srcdir}/${pkgname%-odroid-git}"
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }

+prepare() {
+	cd "${srcdir}/${pkgname%-odroid-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DODROID=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
 build() {
-	cmake -B build -S "${pkgname%-odroid-git}" -DODROID=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
-	make -C build
+	cd "${srcdir}/${pkgname%-odroid-git}/build"
+	make -j$(nproc)
 }

 package() {
-	make -C build DESTDIR="${pkgdir}/" install
-}
+	cd "${srcdir}/${pkgname%-odroid-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}
@@ -0,0 +1,36 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-odroidn2-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-odroidn2-git}-git" "${pkgname%-odroidn2-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-odroidn2-git}-git" "${pkgname%-odroidn2-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "${srcdir}/${pkgname%-odroidn2-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-odroidn2-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DODROID=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-odroidn2-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-odroidn2-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}
@@ -1,7 +1,7 @@
 # Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-pandora-git
-pkgver=r3858.5368285
+pkgver=r4581.d698861
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -15,15 +15,22 @@ source=('git+https://github.com/ptitSeb/box86')
 md5sums=('SKIP')

 pkgver() {
-	cd "$srcdir/${pkgname%-pandora-git}"
+	cd "${srcdir}/${pkgname%-pandora-git}"
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }

+prepare() {
+	cd "${srcdir}/${pkgname%-pandora-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DPANDORA=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
 build() {
-	cmake -B build -S "${pkgname%-pandora-git}" -DPANDORA=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
-	make -C build
+	cd "${srcdir}/${pkgname%-pandora-git}/build"
+	make -j$(nproc)
 }

 package() {
-	make -C build DESTDIR="${pkgdir}/" install
-}
+	cd "${srcdir}/${pkgname%-pandora-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}
@@ -0,0 +1,36 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-phytium-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-phytium-git}"-git "${pkgname%-phytium-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-phytium-git}"-git "${pkgname%-phytium-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "${srcdir}/${pkgname%-phytium-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-phytium-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DPHYTIUM=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-phytium-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-phytium-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}
@@ -1,7 +1,7 @@
 # Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-pyra-git
-pkgver=r3858.5368285
+pkgver=r4581.d698861
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -15,15 +15,22 @@ source=('git+https://github.com/ptitSeb/box86')
 md5sums=('SKIP')

 pkgver() {
-	cd "$srcdir/${pkgname%-pyra-git}"
+	cd "${srcdir}/${pkgname%-pyra-git}"
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }

+prepare() {
+	cd "${srcdir}/${pkgname%-pyra-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DPYRA=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
 build() {
-	cmake -B build -S "${pkgname%-pyra-git}" -DPYRA=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
-	make -C build
+	cd "${srcdir}/${pkgname%-pyra-git}/build"
+	make -j$(nproc)
 }

 package() {
-	make -C build DESTDIR="${pkgdir}/" install
-}
+	cd "${srcdir}/${pkgname%-pyra-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}
@@ -1,7 +1,7 @@
 # Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-rpi-git
-pkgver=r3858.5368285
+pkgver=r4581.d698861
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -16,11 +16,12 @@ md5sums=('SKIP')
 _pimodel=`tr -d '\0' < /sys/firmware/devicetree/base/model`

 pkgver() {
-	cd "$srcdir/${pkgname%-rpi-git}"
+	cd "${srcdir}/${pkgname%-rpi-git}"
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }

-build() {
+prepare() {
+	cd "${srcdir}/${pkgname%-rpi-git}"
 	# Check for Pi version number
 	if [[ ${_pimodel} =~ "Raspberry Pi 4" ]]; then
 		_piversion=4
@@ -31,15 +32,20 @@ build() {
 	else
 		# This RPI is not supported or this device is not a pi.
 		# We should cancel then the build.
-		echo -e "\033[01m\033[01;31m==> ERROR:\033[00;01m Your device is not supported! This script is for RPi model 4/3/2 (1/0 isn't supported yet).\033[00m"
+		echo -e "\033[01m\033[01;31m==> ERROR:\033[00;01m Your device is not supported! This script is for RPi model 4/3/2 (1/0 isn't supported).\033[00m"
 		exit 1
 	fi
 	echo -e "\033[01m\033[01;34m==> INFO:\033[00;01m Your detected Pi verion number is: \033[01;31m${_piversion}\033[01m.\033[00m"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DRPI${_piversion}=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}

-	cmake -B build -S "${pkgname%-rpi-git}" -DRPI${_piversion}=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
-	make -C build
+build() {
+	cd "${srcdir}/${pkgname%-rpi-git}/build"
+	make -j$(nproc)
 }

 package() {
-	make -C build DESTDIR="${pkgdir}/" install
+	cd "${srcdir}/${pkgname%-rpi-git}/build"
+	make DESTDIR="${pkgdir}/" install
 }
@@ -0,0 +1,51 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-rk3xxx-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-rk3xxx-git}-git" "${pkgname%-rk3xxx-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-rk3xxx-git}-git" "${pkgname%-rk3xxx-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+_rkmodel=`tr -d '\0' < /proc/device-tree/compatible`
+
+pkgver() {
+	cd "$srcdir/${pkgname%-rk3xxx-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-rk3xxx-git}"
+	# Check for RockChip version number
+	if [[ ${_rkmodel} =~ "rk3288" ]]; then
+		_rksuffix=3288
+	elif [[ ${_rkmodel} =~ "rk3399" ]]; then
+		_rksuffix=3399	
+	elif [[ ${_rkmodel} =~ "rk3588" ]]; then
+		_rksuffix=3588	
+	else
+		# This RKxxxx is not supported or this device is not RockChip based.
+		# We should cancel then the build.
+		echo -e "\033[01m\033[01;31m==> ERROR:\033[00;01m Your device is not supported! This script is for RockChip model 3399/3588.\033[00m"
+		exit 1
+	fi
+	echo -e "\033[01m\033[01;34m==> INFO:\033[00;01m Your detected RockChip verion number is: \033[01;31m${_rksuffix}\033[01m.\033[00m"
+	[[ ! -d ./build ]] && mkdir build
+    cmake -B build -DRK${_rksuffix}=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-rk3xxx-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-rk3xxx-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}
@@ -0,0 +1,36 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-sd845-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-sd845-git}-git" "${pkgname%-sd845-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-sd845-git}-git" "${pkgname%-sd845-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "${srcdir}/${pkgname%-sd845-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-sd845-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DSD845=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-sd845-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-sd845-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}
@@ -0,0 +1,36 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-tegra-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-tegra-git}"-git "${pkgname%-tegra-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-tegra-git}"-git "${pkgname%-tegra-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "${srcdir}/${pkgname%-tegra-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-tegra-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DTEGRAX1ARM32=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-tegra-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-tegra-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}
@@ -1,7 +1,7 @@
 # Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-x86-git
-pkgver=r3858.5368285
+pkgver=r4581.d698861
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('i386')
@@ -14,15 +14,22 @@ source=('git+https://github.com/ptitSeb/box86')
 md5sums=('SKIP')

 pkgver() {
-	cd "$srcdir/${pkgname%-x86-git}"
+	cd "${srcdir}/${pkgname%-x86-git}"
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }

+prepare() {
+	cd "${srcdir}/${pkgname%-x86-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DLD80BITS=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
 build() {
-	cmake -B build -S "${pkgname%-x86-git}" -DLD80BITS=1 -DNOALIGN=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
-	make -C build
+	cd "${srcdir}/${pkgname%-x86-git}/build"
+	make -j$(nproc)
 }

 package() {
-	make -C build DESTDIR="${pkgdir}/" install
-}
+	cd "${srcdir}/${pkgname%-x86-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}
@@ -12,7 +12,7 @@ This will copy the PKGBUILD for your platform (where you type your platform name
 In the future, I might try to do an unified PKGBUILD for all platforms that will recognize your platform automatically.

 ### 2. What platforms are currently supported.
-All specified in [COMPILE.md](../docs/COMPILE.md).
+Most specified in [COMPILE.md](../docs/COMPILE.md). Except PowerPC based.

 ### 3. Any advantages with that sollution?
 This might give you some advantages like those:
@@ -1,4 +1,6 @@
 #!/bin/sh
+# Restart the systemd-binfmt service to load new binary format rules.
+# This ensures the system is immediately ready to run programs with Box86.
 if [ -d '/run/systemd/system' ]; then
    systemctl restart systemd-binfmt || :
 fi
@@ -72,7 +72,7 @@ def main(root, ver, __debug_forceAllDebugging=False):
 		
 		# Get all actual instructions
 		# Ignore white lines and lines beginning with either !, ; or #
-		insts = list(filter(lambda l: not re.match("^\s*$", l) and not re.match("^([!;#])", l), insts.split('\n')))
+		insts = list(filter(lambda l: not re.match("^\\s*$", l) and not re.match("^([!;#])", l), insts.split('\n')))
 		
 		try:
 			# Do not open with `with` to be able to open it in writing mode
@@ -729,7 +729,7 @@ def main(root, ver, __debug_forceAllDebugging=False):
 				
 				variables["u"] = -1 # Unsigned shift
 				
-				variables["w"] = -1 # wback (used in adressing)
+				variables["w"] = -1 # wback (used in addressing)
 				
 				variables["shift"] = False # Shift already detected?
 				
@@ -1834,7 +1834,7 @@ def main(root, ver, __debug_forceAllDebugging=False):
 			start = (start >> 4) + ((start & 0xF) << 1);
 		}
 		
-		char tmp[6];
+		char tmp[7];
 		for (int cur = start; cur < start + size - 1; ++cur) {
 			sprintf(tmp, "%c%d, ", regChr, cur);
 			strcat(ret, tmp);
@@ -74,7 +74,7 @@ For each of them:
  NOTE: Those structure letters are "fake types" that are accepted in the macros.

 `gbl` contains the first list, `redirects` the second and
- `filespec` constains file specific informations (eg, structures, typedefs required...).
+ `filespec` contains file specific information (eg, structures, typedefs required...).

 After sorting the data, it generates:

@@ -631,7 +631,7 @@ class Clauses:
 						combinable[len(last_combined) - 1] = True
 						add = (i | attempt[0], m | (i ^ attempt[0]))
 						if add in combined:
-							continue # Aleady added
+							continue # Already added
 						combined.append(add)
 				last_combined.pop()
 			last_combined = combined
@@ -679,12 +679,12 @@ class Clauses:
 		return isinstance(o, Clauses) and (self.clauses == o.clauses)

 JumbledFunctions     = CustOrderedDictList[Clause, Function]
-FilesSpecific        = Dict[Filename, FileSpec]
+FileSpecifics        = Dict[Filename, FileSpec]

 SortedGlobals        = CustOrderedDictList[Clauses, FunctionType]
 SortedRedirects      = CustOrderedDictList[Clauses, FunctionType]

-def readFiles(files: Iterable[str]) -> Tuple[JumbledFunctions, JumbledFunctions, FilesSpecific]:
+def readFiles(files: Iterable[str]) -> Tuple[JumbledFunctions, JumbledFunctions, FileSpecifics]:
 	"""
 	readFiles
 	
@@ -693,7 +693,7 @@ def readFiles(files: Iterable[str]) -> Tuple[JumbledFunctions, JumbledFunctions,
 	
 	gbls:      JumbledFunctions = CustOrderedDictList()
 	redirects: JumbledFunctions = CustOrderedDictList()
-	filespecs: FilesSpecific    = {}
+	filespecs: FileSpecifics    = {}
 	
 	symbols: Dict[str, Filename] = {}
 	need_halt: bool = False
@@ -751,6 +751,9 @@ def readFiles(files: Iterable[str]) -> Tuple[JumbledFunctions, JumbledFunctions,
 							 or match("gstinterfaces010","gstaudio")\
 							 or match("gstreamer010","gstreamer")	\
 							 or match("appindicator","appindicator3")\
+							 or match("appindicator","ayatanaappindicator3")\
+							 or match("appindicator3","ayatanaappindicator3")\
+							 or match("libgl",       "libglx")      \
 							 \
 							 or match("libc",        "tcmallocminimal") \
 							 or match("libc",        "ldlinux") 	\
@@ -880,7 +883,7 @@ def readFiles(files: Iterable[str]) -> Tuple[JumbledFunctions, JumbledFunctions,
 	
 	return gbls, redirects, filespecs

-def sortArrays(gbl_funcs: JumbledFunctions, red_funcs: JumbledFunctions, filespecs: FilesSpecific) \
+def sortArrays(gbl_funcs: JumbledFunctions, red_funcs: JumbledFunctions, filespecs: FileSpecifics) \
 -> Tuple[SortedGlobals, SortedRedirects]:
 	# First sort file specific stuff
 	for fn in filespecs:
@@ -944,13 +947,13 @@ def sortArrays(gbl_funcs: JumbledFunctions, red_funcs: JumbledFunctions, filespe
 		redirects[red_vals[funtype]].append(funtype)
 	redirects.sort(key=Clauses.splitdef)
 	
-	def fail(): assert(False)
+	def fail(): assert False
 	for clauses in redirects:
 		redirects[clauses].sort(key=lambda v: fail() if v.redirected is None else v.splitchar() + v.redirected.splitchar())
 	
 	return gbls, redirects

-def checkRun(root: str, gbls: SortedGlobals, redirects: SortedRedirects, filesspec: FilesSpecific) -> Optional[str]:
+def checkRun(root: str, gbls: SortedGlobals, redirects: SortedRedirects, filespecs: FileSpecifics) -> Optional[str]:
 	# Check if there was any new functions compared to last run
 	functions_list: str = ""
 	for clauses in gbls:
@@ -960,18 +963,18 @@ def checkRun(root: str, gbls: SortedGlobals, redirects: SortedRedirects, filessp
 		for v in redirects[clauses]:
 			assert(v.redirected is not None)
 			functions_list = functions_list + "#" + str(clauses) + " " + v.orig + " -> " + v.redirected.orig + "\n"
-	for filename in sorted(filesspec.keys()):
+	for filename in sorted(filespecs.keys()):
 		functions_list = functions_list + filename + ":\n"
-		for st in filesspec[filename].structs:
-			struct = filesspec[filename].structs[st]
+		for st in filespecs[filename].structs:
+			struct = filespecs[filename].structs[st]
 			functions_list = functions_list + \
 				"% " + st + " " + struct.name + " " + struct.repl + "\n"
-		for _bare in filesspec[filename].typedefs:
+		for _bare in filespecs[filename].typedefs:
 			functions_list = functions_list + "- " + _bare.orig + ":\n"
-			for fn in filesspec[filename].typedefs[_bare]:
+			for fn in filespecs[filename].typedefs[_bare]:
 				if fn.no_dlsym: continue
 				functions_list = functions_list + "  - " + fn.name + "\n"
-		for funtype in filesspec[filename].structsuses:
+		for funtype in filespecs[filename].structsuses:
 			assert(funtype.redirected is not None)
 			functions_list = functions_list + "% " + funtype.orig + " -> " + funtype.redirected.orig + "\n"
 	
@@ -992,7 +995,7 @@ def checkRun(root: str, gbls: SortedGlobals, redirects: SortedRedirects, filessp
 	return functions_list

 def generate_files(root: str, files: Iterable[str], ver: str, gbls: SortedGlobals, redirects: SortedRedirects, \
- filespecs: FilesSpecific) -> None:
+ filespecs: FileSpecifics) -> None:
 	# Files header and guard
 	files_header = {
 		"wrapper.c": """
@@ -1148,6 +1151,13 @@ def generate_files(root: str, files: Iterable[str], ver: str, gbls: SortedGlobal
 	for fhdr in files_header:
 		files_guard[fhdr] = trim(files_guard[fhdr])
 	
+	# TODO: put the information below in a structure (like in box64, where
+	# typedefs & co. are stored in an ABI-specific structure)
+	
+	return_x87: str = "DKdf"
+	if any(c not in FileSpec.values for c in return_x87):
+		raise NotImplementedError("Invalid character")
+	
 	# Typedefs
 	#           E            v       c         w          i          I          C          W           u           U           f        d         D              K         l           L            p        V        O          S        2                  P        G        N      M      s
 	tdtypes = ["x86emu_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*", "_2uint_struct_t", "void*", "void*", "...", "...", "void*"]
@@ -1249,7 +1259,7 @@ def generate_files(root: str, files: Iterable[str], ver: str, gbls: SortedGlobal
 		f.write("void {0}(x86emu_t *emu, uintptr_t fcn) {2} {1} fn = ({1})fcn; ".format(N.orig, W, "{"))
 		
 		if N.orig == 'vFv':
-			f.write("(void)fcn; ")
+			f.write("(void)emu; ")
 		
 		args = N.orig[2:]
 		if args == 'Ev': args = 'E'
@@ -1309,6 +1319,22 @@ def generate_files(root: str, files: Iterable[str], ver: str, gbls: SortedGlobal
 			if not clauses.empty():
 				file.write("#endif\n")
 		
+		# Finally, write predicate functions
+		
+		# isRetX87Wrapper
+		file.write("\nint isRetX87Wrapper(wrapper_t fun) {\n")
+		for clauses in gbls:
+			empty = True
+			for funtype in gbls[clauses]:
+				if funtype.orig[0] in return_x87: # TODO: put this in a function (functions would request the ABI for more info)
+					if empty and (not clauses.empty()):
+						file.write("#if " + str(clauses) + "\n")
+						empty = False
+					file.write("\tif (fun == &" + funtype.orig + ") return 1;\n")
+			if not empty:
+				file.write("#endif\n")
+		file.write("\treturn 0;\n}\n")
+		
 		file.write(files_guard["wrapper.c"].format(lbr="{", rbr="}", version=ver))
 	
 	# Rewrite the wrapper.h file:
@@ -1398,6 +1424,6 @@ 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:], "2.0.0.10") != 0:
+	if main(sys.argv[1], sys.argv[limit[0]+1:], "2.0.0.11") != 0:
 		exit(2)
 	exit(0)
@@ -22,6 +22,8 @@
 #include <sys/mman.h>
 #include "custommem.h"
 #include "dictionnary.h"
+#include "rcfile.h"
+#include "gltools.h"
 #ifdef DYNAREC
 #include "dynablock.h"
 #include "dynarec/arm_lock_helper.h"
@@ -48,6 +50,7 @@ void finiAllHelpers(box86context_t* context)
    static int finied = 0;
    if(finied)
        return;
+    DeleteParams();
    fini_pthread_helper(context);
    fini_signal_helper();
    fini_bridge_helper();
@@ -76,17 +79,28 @@ 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);
 }

 int unlockMutex()
 {
    int ret = unlockCustommemMutex();
    int i;
+    #ifdef USE_CUSTOM_MUTEX
+    void* tid = (void*)GetTID();
+    #define GO(A, B)                    \
+        i = (arm_lock_storeifref2(&A, NULL, tid)==tid);  \
+        if(i) {                         \
+            ret|=(1<<B);                \
+        }
+    #else
    #define GO(A, B)                    \
        i = checkUnlockMutex(&A);       \
        if(i) {                         \
            ret|=(1<<B);                \
        }
+    #endif

    GO(my_context->mutex_once, 5)
    GO(my_context->mutex_once2, 6)
@@ -100,7 +114,6 @@ int unlockMutex()
    GO(my_context->mutex_thread, 10)
    GO(my_context->mutex_bridge, 11)
    #undef GO
-
    return ret;
 }

@@ -108,7 +121,7 @@ void relockMutex(int locks)
 {
    #define GO(A, B)                    \
        if(locks&(1<<B))                \
-            pthread_mutex_lock(&A);     \
+            mutex_lock(&A);             \

    GO(my_context->mutex_once, 5)
    GO(my_context->mutex_once2, 6)
@@ -127,22 +140,45 @@ void relockMutex(int locks)

 static void init_mutexes(box86context_t* context)
 {
+
+#ifdef DYNAREC
+#ifdef USE_CUSTOM_MUTEX
+    arm_lock_stored(&context->mutex_dyndump, 0);
+    arm_lock_stored(&context->mutex_once, 0);
+    arm_lock_stored(&context->mutex_once2, 0);
+    arm_lock_stored(&context->mutex_trace, 0);
+    arm_lock_stored(&context->mutex_tls, 0);
+    arm_lock_stored(&context->mutex_thread, 0);
+    arm_lock_stored(&context->mutex_bridge, 0);
+#else
+    pthread_mutexattr_t attr;
+    pthread_mutexattr_init(&attr);
+    pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_ERRORCHECK);
+    pthread_mutex_init(&context->mutex_dyndump, &attr);
+    pthread_mutex_init(&context->mutex_once, &attr);
+    pthread_mutex_init(&context->mutex_once2, &attr);
+    pthread_mutex_init(&context->mutex_trace, &attr);
+    pthread_mutex_init(&context->mutex_tls, &attr);
+    pthread_mutex_init(&context->mutex_thread, &attr);
+    pthread_mutex_init(&context->mutex_bridge, &attr);
+
+    pthread_mutexattr_destroy(&attr);
+#endif
+    pthread_mutex_init(&context->mutex_lock, NULL);
+#else
    pthread_mutexattr_t attr;
    pthread_mutexattr_init(&attr);
    pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_ERRORCHECK);
    pthread_mutex_init(&context->mutex_once, &attr);
    pthread_mutex_init(&context->mutex_once2, &attr);
    pthread_mutex_init(&context->mutex_trace, &attr);
-#ifdef DYNAREC
-    pthread_mutex_init(&context->mutex_dyndump, &attr);
-#else
-    pthread_mutex_init(&context->mutex_lock, &attr);
-#endif
    pthread_mutex_init(&context->mutex_tls, &attr);
    pthread_mutex_init(&context->mutex_thread, &attr);
    pthread_mutex_init(&context->mutex_bridge, &attr);
+    pthread_mutex_init(&context->mutex_lock, &attr);

    pthread_mutexattr_destroy(&attr);
+#endif
 }

 static void atfork_child_box64context(void)
@@ -193,7 +229,7 @@ box86context_t *NewBox86Context(int argc)
 #ifdef BUILD_LIB
    context->deferedInit = 0;
 #else
-    context->deferedInit = 1;
+    context->deferredInit = 1;
 #endif
    context->sel_serial = 1;

@@ -203,11 +239,12 @@ box86context_t *NewBox86Context(int argc)
    context->local_maplib = NewLibrarian(context, 1);
    context->versym = NewDictionnary();
    context->system = NewBridge();
-    context->globaldefver = NewDefaultVersion();
-    context->weakdefver = NewDefaultVersion();
+    // Cannot use Bridge name as the map is not initialized yet
    // create vsyscall
    context->vsyscall = AddBridge(context->system, iFEv, x86Syscall, 0, NULL);
    addAlternate((void*)0xffffe400, (void*)context->vsyscall);
+    // create exit bridge
+    context->exit_bridge = AddBridge(context->system, NULL, NULL, 0, NULL);
 #ifdef BUILD_LIB
    context->box86lib = RTLD_DEFAULT;   // not ideal
 #else
@@ -221,7 +258,7 @@ box86context_t *NewBox86Context(int argc)
    init_mutexes(context);
    pthread_atfork(NULL, NULL, atfork_child_box64context);

-    pthread_key_create(&context->tlskey, free_tlsdatasize);
+    pthread_key_create(&context->tlskey, NULL/*free_tlsdatasize*/);

    InitFTSMap(context);

@@ -229,11 +266,13 @@ box86context_t *NewBox86Context(int argc)
    context->canary[/*getrand(4)*/0] = 0;
    printf_log(LOG_DEBUG, "Setting up canary (for Stack protector) at GS:0x14, value:%08X\n", *(uint32_t*)context->canary);

+    context->globdata = NewMapSymbols();
+
    initAllHelpers(context);

    return context;
 }
-
+void freeALProcWrapper(box86context_t* context);
 EXPORTDYN
 void FreeBox86Context(box86context_t** context)
 {
@@ -249,6 +288,7 @@ void FreeBox86Context(box86context_t** context)

    box86context_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)
@@ -271,8 +311,8 @@ void FreeBox86Context(box86context_t** context)
    if(ctx->zydis)
        DeleteX86Trace(ctx);

-    if(ctx->deferedInitList)
-        box_free(ctx->deferedInitList);
+    if(ctx->deferredInitList)
+        box_free(ctx->deferredInitList);

    /*box_free(ctx->argv);*/
    
@@ -319,7 +359,6 @@ void FreeBox86Context(box86context_t** context)
    void* ptr;
    if ((ptr = pthread_getspecific(ctx->tlskey)) != NULL) {
        free_tlsdatasize(ptr);
-        pthread_setspecific(ctx->tlskey, NULL);
    }
    pthread_key_delete(ctx->tlskey);

@@ -333,26 +372,25 @@ void FreeBox86Context(box86context_t** context)
        }
    }

-    free_neededlib(&ctx->neededlibs);
+    free_neededlib(ctx->neededlibs);
+    ctx->neededlibs = NULL;

    if(ctx->emu_sig)
        FreeX86Emu(&ctx->emu_sig);
-    FreeDefaultVersion(&ctx->globaldefver);
-    FreeDefaultVersion(&ctx->weakdefver);
+
+    FreeMapSymbols(&ctx->globdata);

    finiAllHelpers(ctx);

+#ifndef USE_CUSTOM_MUTEX
    pthread_mutex_destroy(&ctx->mutex_once);
    pthread_mutex_destroy(&ctx->mutex_once2);
    pthread_mutex_destroy(&ctx->mutex_trace);
-#ifdef DYNAREC
-    pthread_mutex_destroy(&ctx->mutex_dyndump);
-#else
-    pthread_mutex_destroy(&ctx->mutex_lock);
-#endif
    pthread_mutex_destroy(&ctx->mutex_tls);
    pthread_mutex_destroy(&ctx->mutex_thread);
    pthread_mutex_destroy(&ctx->mutex_bridge);
+#endif
+    pthread_mutex_destroy(&ctx->mutex_lock);

    freeCycleLog(ctx);

@@ -360,20 +398,45 @@ void FreeBox86Context(box86context_t** context)
 }

 int AddElfHeader(box86context_t* ctx, elfheader_t* head) {
-    int idx = ctx->elfsize;
-    if(idx==ctx->elfcap) {
-        // resize...
-        ctx->elfcap += 16;
-        ctx->elfs = (elfheader_t**)box_realloc(ctx->elfs, sizeof(elfheader_t*) * ctx->elfcap);
+    int idx = 0;
+    while(idx<ctx->elfsize && ctx->elfs[idx]) idx++;
+    if(idx == ctx->elfsize) {
+        if(idx==ctx->elfcap) {
+            // resize...
+            ctx->elfcap += 16;
+            ctx->elfs = (elfheader_t**)box_realloc(ctx->elfs, sizeof(elfheader_t*) * ctx->elfcap);
+        }
+        ctx->elfs[idx] = head;
+        ctx->elfsize++;
+    } else {
+        ctx->elfs[idx] = head;
    }
-    ctx->elfs[idx] = head;
-    ctx->elfsize++;
    printf_log(LOG_DEBUG, "Adding \"%s\" as #%d in elf collection\n", ElfName(head), idx);
    return idx;
 }

+void RemoveElfHeader(box86context_t* ctx, elfheader_t* head) {
+    if(GetTLSBase(head)) {
+        // should remove the tls info
+        int tlsbase = GetTLSBase(head);
+        /*if(tlsbase == -ctx->tlssize) {
+            // not really correct, but will do for now
+            ctx->tlssize -= GetTLSSize(head);
+            if(!(++ctx->sel_serial))
+                ++ctx->sel_serial;
+        }*/
+    }
+    for(int i=0; i<ctx->elfsize; ++i)
+        if(ctx->elfs[i] == head) {
+            ctx->elfs[i] = NULL;
+            return;
+        }
+}
+
+
 int AddTLSPartition(box86context_t* context, int tlssize) {
    int oldsize = context->tlssize;
+    // should in fact first try to map a hole, but rewinding all elfs and checking filled space, like with the mapmem utilities
    context->tlssize += tlssize;
    context->tlsdata = box_realloc(context->tlsdata, context->tlssize);
    memmove(context->tlsdata+tlssize, context->tlsdata, oldsize);   // move to the top, using memmove as regions will probably overlap
@@ -2,7 +2,7 @@
 #define __BOX86_VERSION_H_

 #define BOX86_MAJOR 0
-#define BOX86_MINOR 2
-#define BOX86_REVISION 8
+#define BOX86_MINOR 3
+#define BOX86_REVISION 9

 #endif //__BOX86_VERSION_H_
@@ -192,6 +192,9 @@ Op is 20-27
 // and dst, src1, #imm ror rot*2
 #define AND_IMM8_ROR(dst, src, imm8, rot) \
    EMIT(0xe2000000 | ((dst) << 12) | ((src) << 16) | ((rot)<<8) | brIMM(imm8) )
+// and.c dst, src, #(imm8)
+#define AND_IMM8_COND(cond, dst, src, imm8) \
+    EMIT((cond) | 0x02000000 | ((dst) << 12) | ((src) << 16) | brIMM(imm8) )
 // and.s dst, src, #(imm8)
 #define ANDS_IMM8(dst, src, imm8) \
    EMIT(0xe2100000 | ((dst) << 12) | ((src) << 16) | brIMM(imm8) )
@@ -222,6 +225,8 @@ Op is 20-27
 // add dst, src1, src2, lsr #imm
 #define ADD_REG_LSR_IMM5(dst, src1, src2, imm5) \
    EMIT(0xe0800000 | ((dst) << 12) | ((src1) << 16) | brLSR(imm5, src2) )
+#define ADD_REG_LSR_IMM5_COND(cond, dst, src1, src2, imm5) \
+    EMIT((cond) | 0x00800000 | ((dst) << 12) | ((src1) << 16) | brLSR(imm5, src2) )
 #define ADC_IMM8(dst, src, imm8) \
    EMIT(0xe2a00000 | ((dst) << 12) | ((src) << 16) | brIMM(imm8) )
 // add.s dst, src, #(imm8)
@@ -390,6 +395,8 @@ Op is 20-27
 #define LDRD_IMM8(reg, addr, imm8) EMIT(c__ | 0b000<<25 | 1<<24  | (((imm8)<0)?0:1)<<23 | 1<<22 | 0<<21 | 0<<20 | ((reg) << 12) | ((addr) << 16) | ((abs(imm8))&0xf0)<<(8-4) | (0b1101<<4) | ((abs(imm8))&0x0f) )
 // ldrd reg, reg+1, [addr, +rm], reg must be even, reg+1 is implicit
 #define LDRD_REG(reg, addr, rm) EMIT(c__ | 0b000<<25 | 1<<24  | 1<<23 | 0<<22 | 0<<21 | 0<<20 | ((reg) << 12) | ((addr) << 16) | 0b1101<<4 | (rm) )
+// ldr reg, [pc+imm]
+#define LDR_literal(reg, imm12) EMIT(c__ | 0b010<<25 | 1<<24 | (((imm12)>=0)?1:0)<<23 | 0<<21 | 1<<20 | 0b1111<<16 | (reg)<<12 | (abs(imm12)))

 // str reg, [addr, #+/-imm9]
 #define STR_IMM9(reg, addr, imm9) EMIT(0xe5000000 | (((imm9)<0)?0:1)<<23 | ((reg) << 12) | ((addr) << 16) | brIMM(imm9) )
@@ -423,6 +430,8 @@ Op is 20-27

 // blx reg
 #define BLX(reg) EMIT(0xe12fff30 | (reg) )
+// blx cond reg
+#define BLXcond(C, reg) EMIT(C | 0x012fff30 | (reg) )

 // b cond offset
 #define Bcond(C, O) EMIT(C | (0b101<<25) | (0<<24) | (((O)>>2)&0xffffff))
@@ -430,6 +439,10 @@ Op is 20-27
 // bl cond offset
 #define BLcond(C, O) EMIT(C | (0b101<<25) | (1<<24) | (((O)>>2)&0xffffff))

+// adr
+#define ADR_gen(cond, P, Rd, Imm12) (cond | (1<<25) | (((P)?0b0100:0b0010)<<21) | (0b1111<<16) | ((Rd)<<12) | (Imm12))
+#define ADR(cond, Rd, Imm)  EMIT(ADR_gen(cond, (Imm>=0), Rd, ((Imm>=0)?Imm:-Imm)&0b111111111111))
+
 // push reg!, {list}
 //                           all |    const    |pre-index| subs    | no PSR  |writeback| store   |   base    |reg list
 #define PUSH(reg, list) EMIT(c__ | (0b100<<25) | (1<<24) | (0<<23) | (0<<22) | (1<<21) | (0<<20) | ((reg)<<16) | (list))
@@ -589,8 +602,12 @@ Op is 20-27

 // Count leading 0 bit of Rm, store result in Rd
 #define CLZ(Rd, Rm)  EMIT(c__ | 0b00010110<<20 | 0b1111<<16 | (Rd)<<12 | 0b1111<<8 | 0b0001<<4 | (Rm))
+// Count leading 0 bit of Rm, store result in Rd with cond
+#define CLZ_COND(cond, Rd, Rm)  EMIT(cond | 0b00010110<<20 | 0b1111<<16 | (Rd)<<12 | 0b1111<<8 | 0b0001<<4 | (Rm))
 // Reverse bits of Rm, store result in Rd
 #define RBIT(Rd, Rm) EMIT(c__ | 0b01101111<<20 | 0b1111<<16 | (Rd)<<12 | 0b1111<<8 | 0b0011<<4 | (Rm))
+// Reverse bits of Rm, store result in Rd with cond
+#define RBIT_COND(cond, Rd, Rm) EMIT(cond | 0b01101111<<20 | 0b1111<<16 | (Rd)<<12 | 0b1111<<8 | 0b0011<<4 | (Rm))

 #define PLD_gen(U, R, Rn, Imm5, type, Rm) (0b1111<<28 | 0b0111<<24 | (U)<<23 | (R)<<22 | 0b01<<20 | (Rn)<<16 | 0b1111<<12 | (Imm5)<<7 | (type)<<5 | (Rm))
 // Preload Cache Rn+Rm
@@ -600,7 +617,21 @@ Op is 20-27

 #define DMB_gen(opt)    (0b1111<<28 | 0b01010111<<20 | 0b1111<<16 | 0b1111<<12 | 0b0000<<8 | 0b0101<<4 | (opt))
 // Data memory barrier Inner Sharable
+#ifdef PANDORA
+// The Pandora is single core and ordered, DMB is not usefull here so just emit nothing
+#define DMB_ISH()
+#else
 #define DMB_ISH()   EMIT(DMB_gen(0b1011))
+#endif
+
+#define DSB_gen(opt)    (0b1111<<28 | 0b01010111<<20 | 0b1111<<16 | 0b1111<<12 | 0b0000<<8 | 0b0100<<4 | (opt))
+// Data memory barrier Inner Sharable
+#ifdef PANDORA
+// The Pandora is single core and ordered, DSB is not usefull here so just emit nothing
+#define DSB_ISH()
+#else
+#define DSB_ISH()   EMIT(DSB_gen(0b1011))
+#endif

 #define SWP_gen(cond, B, Rn, Rt, Rt2)   (cond | 0b0001<<24 | (B)<<22 | (Rn)<<16 | (Rt)<<12 | 0b1001<<4 | (Rt2))
 // SWAP (atomic) [Rn]->Rt2 / Rt->[Rn], Rt can be same as Rt2
@@ -626,11 +657,15 @@ Op is 20-27
 #define VMRS(Rt)    EMIT(c__ | (0b1110<<24) | (0b1111<<20) | (0b0001<<16) | ((Rt)<<12) | (0b1010<<8) | (0b0001<<4) | (0b0000))
 // Move to FPSCR from Arm register
 #define VMSR(Rt)    EMIT(c__ | (0b1110<<24) | (0b1110<<20) | (0b0001<<16) | ((Rt)<<12) | (0b1010<<8) | (0b0001<<4) | (0b0000))
+// Move to FPSCR from Arm register with cond
+#define VMSR_cond(cond, Rt)    EMIT(cond | (0b1110<<24) | (0b1110<<20) | (0b0001<<16) | ((Rt)<<12) | (0b1010<<8) | (0b0001<<4) | (0b0000))
 // Move to FPSCR from Arm flags APSR
 #define VMRS_APSR()    VMRS(15)

 // Move between Rt to Sm
 #define VMOVtoV(Sm, Rt) EMIT(c__ | (0b1110<<24) | (0b000<<21) | (0<<20) | ((((Sm)&0b11110)>>1)<<16) | ((Rt)<<12) | (0b1010<<8) | (((Sm)&1)<<7) |(0b00<<6) | (1<<4))
+// Move between Rt to Sm with condition
+#define VMOVtoVcond(cond, Sm, Rt) EMIT(cond | (0b1110<<24) | (0b000<<21) | (0<<20) | ((((Sm)&0b11110)>>1)<<16) | ((Rt)<<12) | (0b1010<<8) | (((Sm)&1)<<7) |(0b00<<6) | (1<<4))
 // Move between Sm to Rt
 #define VMOVfrV(Rt, Sm) EMIT(c__ | (0b1110<<24) | (0b000<<21) | (1<<20) | ((((Sm)&0b11110)>>1)<<16) | ((Rt)<<12) | (0b1010<<8) | (((Sm)&1)<<7) |(0b00<<6) | (1<<4))

@@ -820,6 +855,8 @@ Op is 20-27
 #define VLD1_32(Dd, Rn) EMIT(Vxx1gen(1, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b0111, 2, 0, 15))
 // Load [Rn]! => Dd. Align is 4
 #define VLD1_32_W(Dd, Rn) EMIT(Vxx1gen(1, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b0111, 2, 0, 13))
+// Load [Rn]! => Dd. Align is 8
+#define VLD1_64_W(Dd, Rn) EMIT(Vxx1gen(1, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b0111, 3, 0, 13))
 // Load [Rn]! => Dd/Dd+1. Align is 4
 #define VLD1Q_32_W(Dd, Rn) EMIT(Vxx1gen(1, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b1010, 2, 0, 13))
 // Load [Rn, Rm]! => Dd/Dd+1. If Rm==15, no writeback, Rm ignored, else writeback Rn <- Rn+Rm. Align is 4
@@ -841,6 +878,8 @@ Op is 20-27

 // Store [Rn]! => Dd. Align is 4
 #define VST1_32_W(Dd, Rn) EMIT(Vxx1gen(0, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b0111, 2, 0, 13))
+// Store [Rn]! => Dd. Align is 8
+#define VST1_64_W(Dd, Rn) EMIT(Vxx1gen(0, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b0111, 3, 0, 13))
 // Store [Rn] => Dd/Dd+1. Align is 4
 #define VST1Q_32(Dd, Rn) EMIT(Vxx1gen(0, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b1010, 2, 0, 15))
 // Store [Rn]! => Dd/Dd+1. Align is 4
@@ -879,6 +918,8 @@ Op is 20-27
 #define VMOV_igen(i, D, imm3, Vd, cmode, Q, op, imm4)   (0b1111<<28 | 0b001<<25 | (i)<<24 | 1<<23 | (D)<<22 | (imm3)<<16 | (Vd)<<12 | (cmode)<<8 | (Q)<<6 | (op)<<5 | 1<<4 | (imm4))
 #define VMOV_8(Dd, imm)     EMIT(VMOV_igen(((imm)>>7)&1, ((Dd)>>4)&1, ((imm)>>4)&7, (Dd)&15, 0b1110, 0, 0, (imm)&15))
 #define VMOVQ_8(Dd, imm)    EMIT(VMOV_igen(((imm)>>7)&1, ((Dd)>>4)&1, ((imm)>>4)&7, (Dd)&15, 0b1110, 1, 0, (imm)&15))
+// Dd <= imm8 6bits duplicated 1bit -> 1byte
+#define VMOV_D64(Dd, imm)    EMIT(VMOV_igen(((imm)>>7)&1, ((Dd)>>4)&1, ((imm)>>4)&7, (Dd)&15, 0b1110, 0, 1, (imm)&15))
 // Dd <= imm8 in high bits
 #define VMOV_H32(Dd, imm8)  EMIT(VMOV_igen(((imm8)>>7)&1, ((Dd)>>4)&1, ((imm8)>>4)&7, (Dd)&15, 0b0110, 0, 0, (imm8)&15))
 // Dd <= imm8 in high bits
@@ -1089,6 +1130,7 @@ Op is 20-27
 #define VMULL_U32_U16(Dd, Dn, Dm)   EMIT(VMULL_NEON_gen(1, ((Dd)>>4)&1, 1, (Dn)&15, (Dd)&15, 0, ((Dn)>>4)&1, ((Dm)>>4)&1, (Dm)&15))
 #define VMULL_S64_S32(Dd, Dn, Dm)   EMIT(VMULL_NEON_gen(0, ((Dd)>>4)&1, 2, (Dn)&15, (Dd)&15, 0, ((Dn)>>4)&1, ((Dm)>>4)&1, (Dm)&15))
 #define VMULL_U64_U32(Dd, Dn, Dm)   EMIT(VMULL_NEON_gen(1, ((Dd)>>4)&1, 2, (Dn)&15, (Dd)&15, 0, ((Dn)>>4)&1, ((Dm)>>4)&1, (Dm)&15))
+#define VMULL_P64(Dd, Dn, Dm)       EMIT(VMULL_NEON_gen(0, ((Dd)>>4)&1, 2, (Dn)&15, (Dd)&15, 1, ((Dn)>>4)&1, ((Dm)>>4)&1, (Dm)&15))

 #define VMUL_NEON_gen(op, D, size, Vn, Vd, N, Q, M, Vm)    (0b1111<<28 | 0b001<<25 | (op)<<24 | 0<<23 | (D)<<22 | (size)<<20 | (Vn)<<16 | (Vd)<<12 | 0b1001<<8 | (N)<<7 | (Q)<<6 | (M)<<5 | 1<<4 | (Vm))
 #define VMULQ_32(Dd, Dn, Dm)     EMIT(VMUL_NEON_gen(0, ((Dd)>>4)&1, 0b10, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 1, ((Dm)>>4)&1, (Dm)&15))
@@ -1386,4 +1428,17 @@ Op is 20-27
 #define VABSQ_S8(Dd, Dm)        EMIT(VABS_vgen(((Dd)>>4)&1, 0, (Dd)&15, 0, 1, ((Dm)>>4)&1, (Dm)&15))
 #define VABSQ_F(Dd, Dm)         EMIT(VABS_vgen(((Dd)>>4)&1, 2, (Dd)&15, 1, 1, ((Dm)>>4)&1, (Dm)&15))

+// AES
+#define AESD_gen(D, size, Vd, M, Vm)    (0b1111<<28 | 0b0011<<24 | 1<<23 | (D)<<22 | 0b11<<20 | (size)<<18 | (Vd)<<12 | 0b0110<<7 | 1<<6 | (M)<<5 | (Vm))
+#define AESD(Dd, Dm)            EMIT(AESD_gen(((Dd)>>4)&1, 0, (Dd)&15, ((Dm)>>4)&1, (Dm)&15))
+
+#define AESE_gen(D, size, Vd, M, Vm)    (0b1111<<28 | 0b0011<<24 | 1<<23 | (D)<<22 | 0b11<<20 | (size)<<18 | (Vd)<<12 | 0b0110<<7 | 0<<6 | (M)<<5 | (Vm))
+#define AESE(Dd, Dm)            EMIT(AESE_gen(((Dd)>>4)&1, 0, (Dd)&15, ((Dm)>>4)&1, (Dm)&15))
+
+#define AESIMC_gen(D, size, Vd, M, Vm)  (0b1111<<28 | 0b0011<<24 | 1<<23 | (D)<<22 | 0b11<<20 | (size)<<18 | (Vd)<<12 | 0b0111<<7 | 1<<6 | (M)<<5 | (Vm))
+#define AESIMC(Dd, Dm)          EMIT(AESIMC_gen(((Dd)>>4)&1, 0, (Dd)&15, ((Dm)>>4)&1, (Dm)&15))
+
+#define AESMC_gen(D, size, Vd, M, Vm)   (0b1111<<28 | 0b0011<<24 | 1<<23 | (D)<<22 | 0b11<<20 | (size)<<18 | (Vd)<<12 | 0b0111<<7 | 0<<6 | (M)<<5 | (Vm))
+#define AESMC(Dd, Dm)           EMIT(AESMC_gen(((Dd)>>4)&1, 0, (Dd)&15, ((Dm)>>4)&1, (Dm)&15))
+
 #endif  //__ARM_EMITTER_H__
@@ -9,8 +9,11 @@
 .global arm_epilog
 arm_epilog:
    //update register -> emu
-    //pop     {r0}
    stm     r0, {r4-r12,r14}   // put back reg value in emu, including EIP (so r14 must be Flags now)
+    // restore stack pointer
+    ldr     sp, [r0, #(8*4+2*4)]
+    pop     {r4, r5}
+    str     r4, [r0, #(8*4+2*4)]    // put back old value
    //restore all used register
    vpop     {d8-d15}
    pop     {r4-r12, pc}
@@ -20,6 +23,10 @@ arm_epilog:

 .global arm_epilog_fast
 arm_epilog_fast:
+    // restore stack pointer
+    ldr     sp, [r0, #(8*4+2*4)]
+    pop     {r4, r5}
+    str     r4, [r0, #(8*4+2*4)]    // put back old value
    //restore all used register
    vpop     {d8-d15}
    pop     {r4-r12, pc}
@@ -335,9 +335,9 @@ NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 1 @<1> Rm @ param=1 char align[4
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 0 @<4> 1 1 0 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]!
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 0 @<4> 1 1 1 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 0 @<4> Rm @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> 1 1 0 1 @ param2_1=1 char align[4] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\]!
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> 1 1 1 1 @ param2_1=1 char align[4] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\]
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> Rm @ param2_1=1 char align[4] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> 1 1 0 1 @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> 1 1 1 1 @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> Rm @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\], <Rm>
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]!
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
@@ -471,8 +471,10 @@ ARM_ cond 0 0 0 1 1 1 1 0 Rn Rd (1) (1) (1) (1) 1 0 0 1 Rt STREXH<c> <Rd>, <Rt>,
 ARM_ cond 0 0 0 1 1 1 1 1 Rn Rt (1) (1) (1) (1) 1 0 0 1 (1) (1) (1) (1) LDREXH<c> <Rt>, [<Rn>]
 ARM_ cond 0 0 1 0 0 0 0 S Rn Rd imm12 AND{S}<c> <Rd>, <Rn>, #<const>
 ARM_ cond 0 0 1 0 0 0 1 S Rn Rd imm12 EOR{S}<c> <Rd>, <Rn>, #<const>
+ARM_ cond 0 0 1 0 0 1 0 0 1 1 1 1 Rd imm12 ADR<c> <Rd>, #-<const>
 ARM_ cond 0 0 1 0 0 1 0 S Rn Rd imm12 SUB{S}<c> <Rd>, <Rn>, #<const>
 ARM_ cond 0 0 1 0 0 1 1 S Rn Rd imm12 RSB{S}<c> <Rd>, <Rn>, #<const>
+ARM_ cond 0 0 1 0 1 0 0 0 1 1 1 1 Rd imm12 ADR<c> <Rd>, #<const>
 ARM_ cond 0 0 1 0 1 0 0 S Rn Rd imm12 ADD{S}<c> <Rd>, <Rn>, #<const>
 ARM_ cond 0 0 1 0 1 0 1 S Rn Rd imm12 ADC{S}<c> <Rd>, <Rn>, #<const>
 ARM_ cond 0 0 1 0 1 1 0 S Rn Rd imm12 SBC{S}<c> <Rd>, <Rn>, #<const>
@@ -16,7 +16,14 @@
 .global arm_lock_xchg
 .global arm_lock_storeifnull
 .global arm_lock_storeifref
-
+.global arm_lock_storeifref2
+.global arm_lock_decifnot0b
+.global arm_lock_incb
+.global arm_lock_incif0b
+.global arm_lock_stored
+.global arm_lock_storeb
+.global arm_lock_storeif0b
+.global arm_lock_readb

 arm_lock_read_b:
    // address is r0, return is r0
@@ -83,33 +90,120 @@ arm_lock_xchg:
    bx      lr

 arm_lock_storeifnull:
-    // address is r0, value is r1, r1 store to r0 only if [r0] is 0. return new [r0] value (so r1 or old value)
+    // address is r0, value is r1, r1 store to r0 only if [r0] is 0. return old [r0] value
    // r0 needs to be aligned
    dmb     ish
    ldrex   r2, [r0]
    cmp     r2, #0
    bne     arm_lock_storeifnull_exit
-    mov     r2, r1
-    strex   r3, r2, [r0]
+    strex   r3, r1, [r0]
    cmp     r3, #1
    beq     arm_lock_storeifnull
 arm_lock_storeifnull_exit:
    mov     r0, r2
+    dmb     ish
+    bx      lr
+
+arm_lock_storeifref2:
+    dmb     ish
+arm_lock_storeifref2_loop:
+    // address is r0, value is r1, r1 store to r0 only if [r0] is r2. return old [r0] value (so r2 or old value)
+    ldrex   r3, [r0]
+    cmp     r3, r2
+    movne   r0, r3
+    bne     arm_lock_storeifref2_exit
+    strex   r3, r1, [r0]
+    cmp     r3, #1
+    beq     arm_lock_storeifref_loop
+    mov     r0, r2
+arm_lock_storeifref2_exit:
+    dmb     ish
    bx      lr

 arm_lock_storeifref:
-    push    {r4}
    dmb     ish
 arm_lock_storeifref_loop:
    // address is r0, value is r1, r1 store to r0 only if [r0] is r2. return new [r0] value (so r1 or old value)
    ldrex   r3, [r0]
    cmp     r3, r2
+    movne   r0, r3
    bne     arm_lock_storeifref_exit
-    mov     r3, r1
-    strex   r4, r3, [r0]
-    cmp     r4, #1
+    strex   r3, r1, [r0]
+    cmp     r3, #1
    beq     arm_lock_storeifref_loop
+    mov     r0, r1
 arm_lock_storeifref_exit:
-    mov     r0, r3
-    pop     {r4}
+    dmb     ish
+    bx      lr
+
+arm_lock_decifnot0b:
+    dmb     ish
+arm_lock_decifnot0b_0:
+    ldrexb  r1, [r0]
+    cmp     r1, #0
+    subne   r3, r1, #1
+    moveq   r3, r1
+    strexb  r2, r3, [r0]
+    cmp     r2, #1
+    beq     arm_lock_decifnot0b_0
+    dmb     ish
+    mov     r0, r1
+    bx      lr
+
+arm_lock_incb:
+    dmb     ish
+arm_lock_incb_0:
+    ldrexb  r1, [r0]
+    add     r1, r1, #1
+    strexb  r2, r1, [r0]
+    cmp     r2, #1
+    beq     arm_lock_incb_0
+arm_lock_incb_exit:
+    dmb     ish
+    bx      lr
+
+arm_lock_incif0b:
+    dmb     ish
+arm_lock_incif0b_0:
+    ldrexb  r1, [r0]
+    cmp     r1, #0
+    addeq   r3, r1, #1
+    movne   r3, r1
+    strexb  r2, r3, [r0]
+    cmp     r2, #1
+    beq     arm_lock_incif0b_0
+    dmb     ish
+    mov     r0, r1
+    bx      lr
+
+arm_lock_stored:
+    dmb     ish
+    str     r1, [r0]
+    dmb     ish
+    bx      lr
+
+arm_lock_storeb:
+    dmb     ish
+    strb    r1, [r0]
+    dmb     ish
+    bx      lr
+
+arm_lock_storeif0b:
+    dmb     ish
+arm_lock_storeif0b_0:
+    ldrexb  r2, [r0]
+    cmp     r2, #0
+    bne     arm_lock_storeif0b_exit
+    strexb  r3, r1, [r0]
+    cmp     r3, #1
+    beq     arm_lock_storeif0b_0
+    mov     r2, #0
+    dmb     ish
+arm_lock_storeif0b_exit:
+    mov     r0, r2
+    bx      lr
+
+arm_lock_readb:
+    dmb     ish
+    ldrb    r0, [r0]
    bx      lr
@@ -25,10 +25,34 @@ extern int arm_lock_write_dd(uint32_t a, uint32_t b, void* addr);
 // Atomically exchange value at [p] with val, return old p
 extern uintptr_t arm_lock_xchg(void* p, uintptr_t val);

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

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

+// Atomically store value to [p] only if [p] is ref. Return old [p] value
+extern void* arm_lock_storeifref2(void* p, void* val, void* ref);
+
+// decrement atomicaly the byte at [p] (but only if p not 0). Return old [p] value
+extern int arm_lock_decifnot0b(void*p);
+
+// increment atomicaly the byte at [p]
+extern void arm_lock_incb(void*p);
+
+// increment atomicaly the byte at [p] only if it was 0. Return the old value of [p]
+extern int arm_lock_incif0b(void*p);
+
+// atomic store (with memory barrier)
+extern void arm_lock_stored(void*p, uint32_t d);
+
+// atomic store (with memory barrier)
+extern void arm_lock_storeb(void*p, uint8_t b);
+
+// atomic store (with memory barrier) only if it was 0. Return the old value of [p]
+extern int arm_lock_storeif0b(void*p, uint8_t b);
+
+// atomic load (with memory barrier)
+extern int arm_lock_readb(void*p);
+
 #endif  //__ARM_LOCK_HELPER__H__
@@ -9,6 +9,7 @@
 .extern LinkNext

 .global arm_next
+    .4byte  0   // NULL pointer before arm64_next, for getDB
 arm_next:
    // emu is r0
    //stm     r0, {r4-r12,r14}   // don't put put back reg value in emu, faster but more tricky to debug
@@ -159,7 +159,7 @@ const char* print_register_list_fpu(uint8_t start, uint8_t size, uint8_t double_
 		start = (start >> 4) + ((start & 0xF) << 1);
 	}
 	
-	char tmp[6];
+	char tmp[7];
 	for (int cur = start; cur < start + size - 1; ++cur) {
 		sprintf(tmp, "%c%d, ", regChr, cur);
 		strcat(ret, tmp);
@@ -3049,7 +3049,7 @@ const char* arm_print(uint32_t opcode) {
 		int param1_1 = (opcode >> 5) & 0x1;
 		int param2_1 = (opcode >> 4) & 0x1;
 		
-		char align[4];
+		char align[5];
 		if (param2_1 == 1) {
 			sprintf(align, ":%d", 16 << size);
 		} else {
@@ -3064,7 +3064,7 @@ const char* arm_print(uint32_t opcode) {
 		int param1_1 = (opcode >> 5) & 0x1;
 		int param2_1 = (opcode >> 4) & 0x1;
 		
-		char align[4];
+		char align[5];
 		if (param2_1 == 1) {
 			sprintf(align, ":%d", 16 << size);
 		} else {
@@ -3080,7 +3080,7 @@ const char* arm_print(uint32_t opcode) {
 		int param1_1 = (opcode >> 5) & 0x1;
 		int param2_1 = (opcode >> 4) & 0x1;
 		
-		char align[4];
+		char align[5];
 		if (param2_1 == 1) {
 			sprintf(align, ":%d", 16 << size);
 		} else {
@@ -4197,6 +4197,12 @@ const char* arm_print(uint32_t opcode) {
 		uint16_t imm12 = ((opcode >> 0) & 0xFFF);
 		
 		sprintf(ret, "EOR%s%s %s, %s, #0x%x", (s ? "S" : ""), cond, regname[rd], regname[rn], print_modified_imm_ARM(imm12));
+	} else if ((opcode & 0x0FFF0000) == 0x024F0000) {
+		const char* cond = conds[(opcode >> 28) & 0xF];
+		int rd = (opcode >> 12) & 0xF;
+		uint16_t imm12 = ((opcode >> 0) & 0xFFF);
+		
+		sprintf(ret, "ADR%s %s, #-0x%x", cond, regname[rd], print_modified_imm_ARM(imm12));
 	} else if ((opcode & 0x0FE00000) == 0x02400000) {
 		int s = (opcode >> 20) & 1;
 		const char* cond = conds[(opcode >> 28) & 0xF];
@@ -4213,6 +4219,12 @@ const char* arm_print(uint32_t opcode) {
 		uint16_t imm12 = ((opcode >> 0) & 0xFFF);
 		
 		sprintf(ret, "RSB%s%s %s, %s, #0x%x", (s ? "S" : ""), cond, regname[rd], regname[rn], print_modified_imm_ARM(imm12));
+	} else if ((opcode & 0x0FFF0000) == 0x028F0000) {
+		const char* cond = conds[(opcode >> 28) & 0xF];
+		int rd = (opcode >> 12) & 0xF;
+		uint16_t imm12 = ((opcode >> 0) & 0xFFF);
+		
+		sprintf(ret, "ADR%s %s, #0x%x", cond, regname[rd], print_modified_imm_ARM(imm12));
 	} else if ((opcode & 0x0FE00000) == 0x02800000) {
 		int s = (opcode >> 20) & 1;
 		const char* cond = conds[(opcode >> 28) & 0xF];
@@ -11,9 +11,12 @@ arm_prolog:
    //save all used register
    push     {r4-r12, lr}
    vpush     {d8-d15}
+    // save Sp and setup stack for optionnal callret
+    ldr      r4, [r0, #(8*4+2*4)]   // grab old value of xSPSave
+    mov      r5, #0
+    push     {r4-r5}
+    str      sp, [r0, #(8*4+2*4)]
    //setup emu -> register
-    //push     {r0}
    ldm      r0, {r4-r12}    // all 8 register in direct access, plus flags, no EIP (so r14 can be used as scratch)
-    //TODO
    //jump to function
    bx       r1
@@ -41,21 +41,37 @@ uint32_t X31_hash_code(void* addr, int len)
 	return (uint32_t)h;
 }

-dynablocklist_t* NewDynablockList(uintptr_t text, int textsz, int direct)
+dynablock_t* InvalidDynablock(dynablock_t* db, int need_lock)
 {
-    if(!textsz) {
-        printf_log(LOG_NONE, "Error, creating a NULL sized Dynablock\n");
-        return NULL;
+    if(db) {
+        if(db->gone)
+            return NULL; // already in the process of deletion!
+        dynarec_log(LOG_DEBUG, "InvalidDynablock(%p), db->block=%p x86=%p:%p already gone=%d\n", db, db->block, db->x86_addr, db->x86_addr+db->x86_size-1, db->gone);
+        if(need_lock)
+            mutex_lock(&my_context->mutex_dyndump);
+        // remove jumptable
+        setJumpTableDefault(db->x86_addr);
+        db->done = 0;
+        db->gone = 1;
+        if(need_lock)
+            mutex_unlock(&my_context->mutex_dyndump);
    }
-    dynablocklist_t* ret = (dynablocklist_t*)box_calloc(1, sizeof(dynablocklist_t));
-    ret->text = text;
-    ret->textsz = textsz;
-    if(direct && textsz) {
-        ret->direct = (dynablock_t**)box_calloc(textsz, sizeof(dynablock_t*));
-        if(!ret->direct) {printf_log(LOG_NONE, "Warning, fail to create direct block for dynablock @%p\n", (void*)text);}
+    return db;
+}
+
+void FreeInvalidDynablock(dynablock_t* db, int need_lock)
+{
+    if(db) {
+        if(!db->gone)
+            return; // already in the process of deletion!
+        dynarec_log(LOG_DEBUG, "FreeInvalidDynablock(%p), db->block=%p x86=%p:%p already gone=%d\n", db, db->block, db->x86_addr, db->x86_addr+db->x86_size-1, db->gone);
+        if(need_lock)
+            mutex_lock(&my_context->mutex_dyndump);
+        FreeDynarecMap((uintptr_t)db->actual_block);
+        customFree(db);
+        if(need_lock)
+            mutex_unlock(&my_context->mutex_dyndump);
    }
-    dynarec_log(LOG_DEBUG, "New Dynablocklist %p, from %p->%p\n", ret, (void*)text, (void*)(text+textsz));
-    return ret;
 }

 void FreeDynablock(dynablock_t* db, int need_lock)
@@ -63,362 +79,223 @@ void FreeDynablock(dynablock_t* db, int need_lock)
    if(db) {
        if(db->gone)
            return; // already in the process of deletion!
-        dynarec_log(LOG_DEBUG, "FreeDynablock(%p), db->block=%p x86=%p:%p father=%p, with %d son(s) already gone=%d\n", db, db->block, db->x86_addr, db->x86_addr+db->x86_size, db->father, db->sons_size, db->gone);
+        dynarec_log(LOG_DEBUG, "FreeDynablock(%p), db->block=%p x86=%p:%p already gone=%d\n", db, db->block, db->x86_addr, db->x86_addr+db->x86_size-1, db->gone);
        if(need_lock)
-            pthread_mutex_lock(&my_context->mutex_dyndump);
-        db->done = 0;
-        db->gone = 1;
-        // remove from direct if there
-        uintptr_t startdb = db->parent->text;
-        uintptr_t enddb = db->parent->text + db->parent->textsz;
-        if(db->parent->direct) {
-            uintptr_t addr = (uintptr_t)db->x86_addr;
-            if(addr>=startdb && addr<enddb)
-                db->parent->direct[addr-startdb] = NULL;
-        }
+            mutex_lock(&my_context->mutex_dyndump);
        // remove jumptable
        setJumpTableDefault(db->x86_addr);
-        // remove and free the sons
-        for (int i=0; i<db->sons_size; ++i) {
-            dynablock_t *son = (dynablock_t*)arm_lock_xchg(&db->sons[i], 0);
-            FreeDynablock(son, 0);
-        }
-        // only the father free the DynarecMap
-        if(!db->father) {
-            dynarec_log(LOG_DEBUG, " -- FreeDyrecMap(%p, %d)\n", db->block, db->size);
-            FreeDynarecMap(db, (uintptr_t)db->block, db->size);
-            box_free(db->sons);
-            box_free(db->instsize);
-        }
-        box_free(db);
+        dynarec_log(LOG_DEBUG, " -- FreeDyrecMap(%p, %d)\n", db->actual_block, db->size);
+        db->done = 0;
+        db->gone = 1;
+        if(db->previous)
+            FreeInvalidDynablock(db->previous, 0);
+        FreeDynarecMap((uintptr_t)db->actual_block);
+        customFree(db);
        if(need_lock)
-            pthread_mutex_unlock(&my_context->mutex_dyndump);
+            mutex_unlock(&my_context->mutex_dyndump);
    }
 }

-void FreeDynablockList(dynablocklist_t** dynablocks)
-{
-    if(!dynablocks)
-        return;
-    if(!*dynablocks)
-        return;
-    dynarec_log(LOG_DEBUG, "Free Dynablocklist %p, with Direct Blocks %p\n", *dynablocks, (*dynablocks)->direct);
-    if((*dynablocks)->direct) {
-        for (int i=0; i<(*dynablocks)->textsz; ++i) {
-            if((*dynablocks)->direct[i] && !(*dynablocks)->direct[i]->father) 
-                FreeDynablock((*dynablocks)->direct[i], my_context?1:0);
-        }
-        box_free((*dynablocks)->direct);
-    }
-    (*dynablocks)->direct = NULL;

-    box_free(*dynablocks);
-    *dynablocks = NULL;
-}

 void MarkDynablock(dynablock_t* db)
 {
    if(db) {
-        if(db->father)
-            db = db->father;    // mark only father
-        if(db->need_test)
-            return; // already done
-        db->need_test = 1;
-        setJumpTableDefault(db->x86_addr);
-        for(int i=0; i<db->sons_size; ++i)
-            setJumpTableDefault(db->sons[i]->x86_addr);
+        dynarec_log(LOG_DEBUG, "MarkDynablock %p %p-%p\n", db, db->x86_addr, db->x86_addr+db->x86_size-1);
+        if(!setJumpTableIfRef(db->x86_addr, db->jmpnext, db->block)) {
+            dynablock_t* old = db;
+            db = getDB((uintptr_t)old->x86_addr);
+            if(!old->gone && db!=old) {
+                printf_log(LOG_INFO, "Warning, couldn't mark block as dirty for %p, block=%p, current_block=%p\n", old->x86_addr, old, db);
+                // the block is lost, need to invalidate it...
+                old->gone = 1;
+                old->done = 0;
+                if(!db || db->previous)
+                    FreeInvalidDynablock(old, 1);
+                else
+                    db->previous = old;
+            }
+        }
    }
 }

-uintptr_t StartDynablockList(dynablocklist_t* db)
-{
-    if(db)
-        return db->text;
-    return 0;
-}
-uintptr_t EndDynablockList(dynablocklist_t* db)
-{
-    if(db)
-        return db->text+db->textsz-1;
-    return 0;
-}
-
-int IntervalIntersects(uintptr_t start1, uintptr_t end1, uintptr_t start2, uintptr_t end2)
+static int IntervalIntersects(uintptr_t start1, uintptr_t end1, uintptr_t start2, uintptr_t end2)
 {
    if(start1 > end2 || start2 > end1)
        return 0;
    return 1;
 }

-void MarkDirectDynablock(dynablocklist_t* dynablocks, uintptr_t addr, uintptr_t size)
+static int MarkedDynablock(dynablock_t* db)
 {
-    if(!dynablocks)
-        return;
-    if(!dynablocks->direct)
-        return;
-    uintptr_t startdb = dynablocks->text;
-    uintptr_t enddb = startdb + dynablocks->textsz -1;
-    uintptr_t start = addr;
-    uintptr_t end = addr+size-1;
-    if(start<startdb)
-        start = startdb;
-    if(end>enddb)
-        end = enddb;
-    dynablock_t *db;
-    if(end>startdb && start<enddb)
-        for(uintptr_t i = start; i<end; ++i)
-            if((db=dynablocks->direct[i-startdb]))
-                if(IntervalIntersects((uintptr_t)db->x86_addr, (uintptr_t)db->x86_addr+db->x86_size-1, addr, addr+size+1))
-                    MarkDynablock(db);
-}
-
-void FreeRangeDynablock(dynablocklist_t* dynablocks, uintptr_t addr, uintptr_t size)
-{
-    if(!dynablocks)
-        return;
-
-    if(dynablocks->direct) {
-        int need_lock = my_context?1:0;
-        dynablock_t* db;
-        int ret;
-        khint_t k;
-        kh_dynablocks_t *blocks = kh_init(dynablocks);
-        // copy in a temporary list
-        if(dynablocks->direct) {
-            uintptr_t startdb = dynablocks->text;
-            uintptr_t enddb = startdb + dynablocks->textsz;
-            uintptr_t start = addr;
-            uintptr_t end = addr+size;
-            if(start<startdb)
-                start = startdb;
-            if(end>enddb)
-                end = enddb;
-            if(end>startdb && start<enddb)
-                for(uintptr_t i = start; i<end; ++i) {
-                    db = (dynablock_t*)arm_lock_xchg(&dynablocks->direct[i-startdb], 0);
-                    if(db) {
-                        if(db->father)
-                            db = db->father;
-                        if(db->parent==dynablocks) {
-                            k = kh_put(dynablocks, blocks, (uintptr_t)db, &ret);
-                            kh_value(blocks, k) = db;
-                        }
-                    }
-                }
-        }
-        // purge the list
-        kh_foreach_value(blocks, db,
-            FreeDynablock(db, need_lock);
-        );
-        kh_destroy(dynablocks, blocks);
+    if(db) {
+        if(getNeedTest((uintptr_t)db->x86_addr))
+            return 1; // already done
    }
+    return 0;
 }
-void MarkRangeDynablock(dynablocklist_t* dynablocks, uintptr_t addr, uintptr_t size)
+
+void MarkRangeDynablock(dynablock_t* db, uintptr_t addr, uintptr_t size)
 {
-    if(!dynablocks)
+    // Mark will try to find *any* blocks that intersect the range to mark
+    if(!db)
        return;
-    if(dynablocks->direct) {
-        uintptr_t new_addr = addr - dynablocks->maxsz;
-        uintptr_t new_size = size + dynablocks->maxsz;
-        MarkDirectDynablock(dynablocks, new_addr, new_size);
-        // the blocks check before
-        for(int idx=(new_addr)>>DYNAMAP_SHIFT; idx<(addr>>DYNAMAP_SHIFT); ++idx)
-            MarkDirectDynablock(getDB(idx), new_addr, new_size);
+    dynarec_log(LOG_DEBUG, "MarkRangeDynablock %p-%p .. startdb=%p, sizedb=%p\n", (void*)addr, (void*)addr+size-1, (void*)db->x86_addr, (void*)db->x86_size);
+    if(IntervalIntersects((uintptr_t)db->x86_addr, (uintptr_t)db->x86_addr+db->x86_size-1, addr, addr+size+1))
+        MarkDynablock(db);
+}
+
+int FreeRangeDynablock(dynablock_t* db, uintptr_t addr, uintptr_t size)
+{
+    if(!db)
+        return 1;
+
+    int need_lock = my_context?1:0;
+    if(IntervalIntersects((uintptr_t)db->x86_addr, (uintptr_t)db->x86_addr+db->x86_size-1, addr, addr+size+1)) {
+        FreeDynablock(db, need_lock);
+        return 0;
    }
+    return 1;
 }

-dynablock_t* FindDynablockDynablocklist(void* addr, kh_dynablocks_t* dynablocks)
+dynablock_t *AddNewDynablock(uintptr_t addr)
 {
-    if(!dynablocks)
-        return NULL;
-    dynablock_t* db;
-    kh_foreach_value(dynablocks, db, 
-        const uintptr_t s = (uintptr_t)db->block;
-        const uintptr_t e = (uintptr_t)db->block+db->size;
-        if((uintptr_t)addr>=s && (uintptr_t)addr<e)
-            return db->father?db->father:db;
-    )
-    return NULL;
-}
-
-static dynablocklist_t* getDBFromAddress(uintptr_t addr)
-{
-    const int idx = (addr>>DYNAMAP_SHIFT);
-    return getDB(idx);
-}
-
-dynablock_t *AddNewDynablock(dynablocklist_t* dynablocks, uintptr_t addr, int* created)
-{
-    if(!dynablocks) {
-        dynarec_log(LOG_INFO, "Warning: Ask to create a dynablock with a NULL dynablocklist (addr=%p)\n", (void*)addr);
-        *created = 0;
+    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;
    }
-    if((addr<dynablocks->text) || (addr>=(dynablocks->text+dynablocks->textsz))) {
-        // this should be useless
-        //dynarec_log(LOG_INFO, "Warning: Refused to create a Direct Block that is out-of-bound: dynablocks=%p (%p:%p), addr=%p\n", dynablocks, (void*)(dynablocks->text), (void*)(dynablocks->text+dynablocks->textsz), (void*)addr);
-        //*created = 0;
-        //return NULL;
-        return AddNewDynablock(getDBFromAddress(addr), addr, created);
-    }
-    dynablock_t* block = NULL;
-    // first, check if it exist in direct access mode
-    if(dynablocks->direct) {
-        block = dynablocks->direct[addr-dynablocks->text];
-        if(block) {
-            dynarec_log(LOG_VERBOSE, "Block already exist in Direct Map\n");
-            *created = 0;
-            return block;
-        }
-    }
    
-    if (!*created)
-        return block;
-    
-    pthread_mutex_lock(&my_context->mutex_dyndump);
-    if(!dynablocks->direct) {
-        dynablock_t** p = (dynablock_t**)box_calloc(dynablocks->textsz, sizeof(dynablock_t*));
-        if(arm_lock_storeifnull(&dynablocks->direct, p)!=p)
-            box_free(p);    // someone already create the direct array, too late...
-    }
-
+    #endif
    // create and add new block
    dynarec_log(LOG_VERBOSE, "Ask for DynaRec Block creation @%p\n", (void*)addr);
-
-    block = (dynablock_t*)box_calloc(1, sizeof(dynablock_t));
-    block->parent = dynablocks; 
-    dynablock_t* tmp = (dynablock_t*)arm_lock_storeifnull(&dynablocks->direct[addr-dynablocks->text], block);
-    pthread_mutex_unlock(&my_context->mutex_dyndump);
-    if(tmp !=  block) {
-        // a block appeard!
-        box_free(block);
-        *created = 0;
-        return tmp;
-    }
-    *created = 1;
+    block = (dynablock_t*)customCalloc(1, sizeof(dynablock_t));
    return block;
 }

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

 /* 
    return NULL if block is not found / cannot be created. 
    Don't create if create==0
 */
-static dynablock_t* internalDBGetBlock(x86emu_t* emu, uintptr_t addr, uintptr_t filladdr, int create, dynablock_t* current, int need_lock)
+static dynablock_t* internalDBGetBlock(x86emu_t* emu, uintptr_t addr, uintptr_t filladdr, int create, int need_lock)
 {
    if(hasAlternate((void*)addr))
        return NULL;
-    // try the quickest way first: get parent of current and check if ok!
-    dynablocklist_t *dynablocks = NULL;
-    dynablock_t* block = NULL;
-    if(current && current->done) {
-        dynablocks = current->parent;
-        if(dynablocks && !(addr>=dynablocks->text && addr<(dynablocks->text+dynablocks->textsz)))
-            dynablocks = NULL;
-    }
-    // nope, lets do the long way
-    if(!dynablocks) {
-        dynablocks = getDBFromAddress(addr);
-        if(!dynablocks) {
-            dynablocks = GetDynablocksFromAddress(emu->context, addr);
-            if(!dynablocks)
+    dynablock_t* block = getDB(addr);
+    if(block || !create)
+        return block;
+
+    if(need_lock) {
+        if(box86_dynarec_wait) {
+            mutex_lock(&my_context->mutex_dyndump);
+        } else {
+            if(mutex_trylock(&my_context->mutex_dyndump))   // FillBlock not available for now
                return NULL;
        }
-    }
-    // check direct first, without lock
-    if(dynablocks->direct/* && (addr>=dynablocks->text) && (addr<(dynablocks->text+dynablocks->textsz))*/)
-        if((block = dynablocks->direct[addr-dynablocks->text]))
+        block = getDB(addr);    // just in case
+        if(block) {
+            mutex_unlock(&my_context->mutex_dyndump);
            return block;
-
-    int created = create;
-    block = AddNewDynablock(dynablocks, addr, &created);
-    if(!created)
-        return block;   // existing block...
+        }
+    }
+    
+    block = AddNewDynablock(addr);

    // fill the block
    block->x86_addr = (void*)addr;
-    if(need_lock)
-        pthread_mutex_lock(&my_context->mutex_dyndump);
-    if(sigsetjmp(&dynarec_jmpbuf, 1)) {
+    if(sigsetjmp(DYN_JMPBUF, 1)) {
        printf_log(LOG_INFO, "FillBlock at %p triggered a segfault, cancelling\n", (void*)addr);
+        FreeDynablock(block, 0);
        if(need_lock)
-            pthread_mutex_unlock(&my_context->mutex_dyndump);
+            mutex_unlock(&my_context->mutex_dyndump);
        return NULL;
    }
    void* ret = FillBlock(block, filladdr);
-    if(need_lock)
-        pthread_mutex_unlock(&my_context->mutex_dyndump);
    if(!ret) {
        dynarec_log(LOG_DEBUG, "Fillblock of block %p for %p returned an error\n", block, (void*)addr);
-        void* old = (void*)arm_lock_storeifref(&dynablocks->direct[addr-dynablocks->text], 0, block);
-        if(old!=block && old) { // put it back in place, strange things are happening here!
-            dynarec_log(LOG_INFO, "Warning, a wild block appeared at %p: %p\n", (void*)addr, old);
-            // Doing nothing else, the block has not been written
-        }
-        box_free(block);
+        customFree(block);
        block = NULL;
    }
    // check size
-    if(block && block->x86_size) {
+    if(block) {
        int blocksz = block->x86_size;
-        if(dynablocks->maxsz<blocksz) {
-            dynablocks->maxsz = blocksz;
-            for(int idx=(addr>>DYNAMAP_SHIFT)+1; idx<=((addr+blocksz)>>DYNAMAP_SHIFT); ++idx) {
-                dynablocklist_t* dblist;
-                if((dblist = getDB(idx)))
-                    if(dblist->maxsz<blocksz)
-                        dblist->maxsz = blocksz;
-            }
-        }
-        //protectDB((uintptr_t)block->x86_addr, block->x86_size);   // already protected in FillBlock
-        // but check if it already need test
-        if(block->need_test || (block->father && block->father->need_test))
-            return NULL;
+        if(blocksz>my_context->max_db_size)
+            my_context->max_db_size = blocksz;
        // fill-in jumptable
-        addJumpTableIfDefault(block->x86_addr, block->block);
-        for(int i=0; i<block->sons_size; ++i)
-            addJumpTableIfDefault(block->sons[i]->x86_addr, block->sons[i]->block);
+        if(!addJumpTableIfDefault(block->x86_addr, block->dirty?block->jmpnext:block->block)) {
+            FreeDynablock(block, 0);
+            block = getDB(addr);
+            MarkDynablock(block);   // just in case...
+        } else {
+            if(block->x86_size)
+                block->done = 1;    // don't validate the block if the size is null, but keep the block
+        }
    }
+    if(need_lock)
+        mutex_unlock(&my_context->mutex_dyndump);

-    dynarec_log(LOG_DEBUG, " --- DynaRec Block %s @%p:%p (%p, 0x%x bytes, with %d son(s))\n", created?"created":"recycled", (void*)addr, (void*)(addr+((block)?block->x86_size:0)), (block)?block->block:0, (block)?block->size:0, (block)?block->sons_size:0);
+    dynarec_log(LOG_DEBUG, "%04d| --- DynaRec Block created @%p:%p (%p, 0x%x bytes)\n", GetTID(), (void*)addr, (void*)(addr+((block)?block->x86_size:1)-1), (block)?block->block:0, (block)?block->size:0);

    return block;
 }

-dynablock_t* DBGetBlock(x86emu_t* emu, uintptr_t addr, int create, dynablock_t** current)
+dynablock_t* DBGetBlock(x86emu_t* emu, uintptr_t addr, int create)
 {
-    dynablock_t *db = internalDBGetBlock(emu, addr, addr, create, *current, 1);
-    if(db && db->done && db->block && (db->need_test || (db->father && db->father->need_test))) {
-        if(pthread_mutex_trylock(&my_context->mutex_dyndump))
-            return NULL;
-        dynablock_t *father = db->father?db->father:db;
-        uint32_t hash = (getProtection((uintptr_t)father->x86_addr)&PROT_READ)?X31_hash_code(father->x86_addr, father->x86_size):0;
-        if(hash!=father->hash) {
-            dynarec_log(LOG_DEBUG, "Invalidating block %p from %p:%p (hash:%X/%X) with %d son(s) for %p\n", father, father->x86_addr, father->x86_addr+father->x86_size, hash, father->hash, father->sons_size, (void*)addr);
-            // no more current if it gets invalidated too
-            if(*current && father->x86_addr>=(*current)->x86_addr && (father->x86_addr+father->x86_size)<(*current)->x86_addr)
-                *current = NULL;
-            // Free father, it's now invalid!
-            FreeDynablock(father, 0);
+    if(isInHotPage(addr))
+        return NULL;
+    dynablock_t *db = internalDBGetBlock(emu, addr, addr, create, 1);
+    if(db && db->done && db->block && getNeedTest(addr)) {
+        if(db->always_test)
+            sched_yield();  // just calm down...
+        uint32_t hash = X31_hash_code(db->x86_addr, db->x86_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, always_test:%d) for %p\n", db, db->x86_addr, db->x86_addr+db->x86_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)
-            db = internalDBGetBlock(emu, addr, addr, create, *current, 0);
+            db = internalDBGetBlock(emu, addr, addr, create, need_lock);
+            if(db) {
+                if(db->previous)
+                    FreeInvalidDynablock(db->previous, need_lock);
+                db->previous = old;
+            } else
+                FreeInvalidDynablock(old, need_lock);
        } else {
-            father->need_test = 0;
-            protectDB((uintptr_t)father->x86_addr, father->x86_size);
+            dynarec_log(LOG_DEBUG, "Validating block %p from %p:%p (hash:%X, always_test:%d) for %p\n", db, db->x86_addr, db->x86_addr+db->x86_size-1, db->hash, db->always_test, (void*)addr);
+            protectDB((uintptr_t)db->x86_addr, db->x86_size);
            // fill back jumptable
-            addJumpTableIfDefault(father->x86_addr, father->block);
-            for(int i=0; i<father->sons_size; ++i)
-                addJumpTableIfDefault(father->sons[i]->x86_addr, father->sons[i]->block);
+            if(isprotectedDB((uintptr_t)db->x86_addr, db->x86_size) && !db->always_test) {
+                setJumpTableIfRef(db->x86_addr, db->block, db->jmpnext);
+            }
        }
-        pthread_mutex_unlock(&my_context->mutex_dyndump);
+        if(!need_lock)
+            mutex_unlock(&my_context->mutex_dyndump);
    } 
+    if(!db || !db->block || !db->done)
+        emu->test.test = 0;
    return db;
 }

@@ -426,27 +303,36 @@ dynablock_t* DBAlternateBlock(x86emu_t* emu, uintptr_t addr, uintptr_t filladdr)
 {
    dynarec_log(LOG_DEBUG, "Creating AlternateBlock at %p for %p\n", (void*)addr, (void*)filladdr);
    int create = 1;
-    dynablock_t *db = internalDBGetBlock(emu, addr, filladdr, create, NULL, 1);
-    if(db && db->done && db->block && (db->need_test || (db->father && db->father->need_test))) {
-        if(pthread_mutex_trylock(&my_context->mutex_dyndump))
-            return NULL;
-        dynablock_t *father = db->father?db->father:db;
-        uint32_t hash = (getProtection((uintptr_t)father->x86_addr)&PROT_READ)?X31_hash_code(father->x86_addr, father->x86_size):0;
-        if(hash!=father->hash) {
-            dynarec_log(LOG_DEBUG, "Invalidating alt block %p from %p:%p (hash:%X/%X) with %d son(s) for %p\n", father, father->x86_addr, father->x86_addr+father->x86_size, hash, father->hash, father->sons_size, (void*)addr);
-            // Free father, it's now invalid!
-            FreeDynablock(father, 0);
+    dynablock_t *db = internalDBGetBlock(emu, addr, filladdr, create, 1);
+    if(db && db->done && db->block && getNeedTest(filladdr)) {
+        if(db->always_test)
+            sched_yield();  // just calm down...
+        int need_lock = mutex_trylock(&my_context->mutex_dyndump);
+        uint32_t hash = X31_hash_code(db->x86_addr, db->x86_size);
+        if(hash!=db->hash) {
+            db->done = 0;   // invalidating the block
+            dynarec_log(LOG_DEBUG, "Invalidating alt block %p from %p:%p (hash:%X/%X) for %p\n", db, db->x86_addr, db->x86_addr+db->x86_size, hash, db->hash, (void*)addr);
+            // Free db, it's now invalid!
+            dynablock_t* old = InvalidDynablock(db, need_lock);
            // start again... (will create a new block)
-            db = internalDBGetBlock(emu, addr, filladdr, create, NULL, 0);
+            db = internalDBGetBlock(emu, addr, filladdr, create, need_lock);
+            if(db) {
+                if(db->previous)
+                    FreeInvalidDynablock(db->previous, need_lock);
+                db->previous = old;
+            } else
+                FreeInvalidDynablock(old, need_lock);
        } else {
-            father->need_test = 0;
-            protectDB((uintptr_t)father->x86_addr, father->x86_size);
+            protectDB((uintptr_t)db->x86_addr, db->x86_size);
            // fill back jumptable
-            addJumpTableIfDefault(father->x86_addr, father->block);
-            for(int i=0; i<father->sons_size; ++i)
-                addJumpTableIfDefault(father->sons[i]->x86_addr, father->sons[i]->block);
+            if(isprotectedDB((uintptr_t)db->x86_addr, db->x86_size) && !db->always_test) {
+                setJumpTableIfRef(db->x86_addr, db->block, db->jmpnext);
+            }
        }
-        pthread_mutex_unlock(&my_context->mutex_dyndump);
+        if(!need_lock)
+            mutex_unlock(&my_context->mutex_dyndump);
    } 
+    if(!db || !db->block || !db->done)
+        emu->test.test = 0;
    return db;
 }
@@ -1,36 +1,26 @@
 #ifndef __DYNABLOCK_PRIVATE_H_
 #define __DYNABLOCK_PRIVATE_H_

-typedef struct dynablocklist_s  dynablocklist_t;
-
 typedef struct instsize_s {
    unsigned char x86:4;
    unsigned char nat:4;
 } instsize_t;

 typedef struct dynablock_s {
-    dynablocklist_t* parent;
-    void*           block;
+    void*           block;  // block-sizeof(void*) == self
+    void*           actual_block;   // the actual start of the block (so block-sizeof(void*))
+    struct dynablock_s*    previous;   // a previous block that might need to be freed
    int             size;
    void*           x86_addr;
    uintptr_t       x86_size;
    uint32_t        hash;
-    uint8_t         need_test;
    uint8_t         done;
    uint8_t         gone;
-    uint8_t         dummy;
+    uint8_t         always_test;
+    uint8_t         dirty;      // if need to be tested as soon as it's created
    int             isize;
-    dynablock_t**   sons;   // sons (kind-of dummy dynablock...)
-    int             sons_size;
-    dynablock_t*    father; // set only in the case of a son
    instsize_t*     instsize;
+    void*           jmpnext;    // a branch jmpnext code when block is marked
 } dynablock_t;

-typedef struct dynablocklist_s {
-    uintptr_t           text;
-    int                 textsz;
-    int                 maxsz;     // maxblock size (for this block or previous block)
-    dynablock_t**       direct;    // direct mapping (waste of space, so the array is created at first write)
-} dynablocklist_t;
-
 #endif //__DYNABLOCK_PRIVATE_H_
@@ -20,6 +20,8 @@
 #include "dynablock.h"
 #include "dynablock_private.h"
 #include "bridge.h"
+#include "custommem.h"
+void x86test_check(x86emu_t* ref, uintptr_t ip);
 #endif

 #ifdef ARM
@@ -44,26 +46,31 @@ void* LinkNext(x86emu_t* emu, uintptr_t addr, void* x2)
        printf_log(LOG_NONE, "Warning, jumping to NULL address from %p (db=%p, x86addr=%p/%s)\n", x2, db, x86addr, pcname);
    }
    #endif
-    dynablock_t* current = NULL;
    void * jblock;
-    dynablock_t* block = DBGetBlock(emu, addr, 1, &current);
+    dynablock_t* block = NULL;
+    if(hasAlternate((void*)addr)) {
+        printf_log(LOG_INFO, "Jmp address has alternate: %p", (void*)addr);
+        addr = (uintptr_t)getAlternate((void*)addr);
+        R_EIP = addr;
+        printf_log(LOG_INFO, " -> %p\n", (void*)addr);
+        block = DBGetBlock(emu, addr, 1);
+    } else
+        block = DBGetBlock(emu, addr, 1);
    if(!block) {
        // no block, let link table as is...
-        if(hasAlternate((void*)addr)) {
-            printf_log(LOG_INFO, "Jmp address has alternate: %p", (void*)addr);
-            addr = (uintptr_t)getAlternate((void*)addr);
-            R_EIP = addr;
-            printf_log(LOG_INFO, " -> %p\n", (void*)addr);
-            block = DBGetBlock(emu, addr, 1, &current);
-        }
-        if(!block) {
-            #ifdef HAVE_TRACE
-            dynablock_t* db = FindDynablockFromNativeAddress(x2);
-            dynarec_log(LOG_INFO, "Warning, jumping to a no-block address %p from %p (db=%p, x86addr=%p)\n", (void*)addr, x2, db, db?(void*)getX86Address(db, (uintptr_t)x2):NULL);
-            #endif
-            //tableupdate(arm_epilog, addr, table);
-            return arm_epilog;
+        #ifdef HAVE_TRACE
+        if(LOG_INFO<=box86_dynarec_log) {
+            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, x86addr=%p(elf=%s))\n", (void*)addr, x2-4, db, db?(void*)getX86Address(db, (uintptr_t)x2-4):NULL, h?ElfName(h):"(none)");
+            }
        }
+        #endif
+        //tableupdate(arm_epilog, addr, table);
+        return arm_epilog;
    }
    if(!block->done) {
        // not finished yet... leave linker
@@ -79,166 +86,125 @@ void* LinkNext(x86emu_t* emu, uintptr_t addr, void* x2)
 }
 #endif

+#ifdef ANDROID
+#define JUMPBUFF sigjmp_buf
+#else
+#define JUMPBUFF struct __jmp_buf_tag
+#endif
+
 void DynaCall(x86emu_t* emu, uintptr_t addr)
 {
-    // prepare setjump for signal handling
-    emu_jmpbuf_t *ejb = NULL;
-    int jmpbuf_reset = 0;
-    if(emu->type == EMUTYPE_MAIN) {
-        ejb = GetJmpBuf();
-        if(!ejb->jmpbuf_ok) {
-            ejb->emu = emu;
-            ejb->jmpbuf_ok = 1;
-            jmpbuf_reset = 1;
-            int a;
-            if((a=sigsetjmp((struct __jmp_buf_tag*)ejb->jmpbuf, 1))) {
-                printf_log(LOG_DEBUG, "Setjmp DynaCall %d, fs=0x%x\n", a, ejb->emu->segs[_FS]);
-                addr = R_EIP;   // not sure if it should still be inside DynaCall!
-                if(a==2)
-                    Run(emu, 1);    // "single step" next instruction that is doing auto-smc
-            }
-        }
+    uint32_t old_esp = R_ESP;
+    uint32_t old_ebx = R_EBX;
+    uint32_t old_edi = R_EDI;
+    uint32_t old_esi = R_ESI;
+    uint32_t old_ebp = R_EBP;
+    uint32_t old_eip = R_EIP;
+    // save defered flags
+    defered_flags_t old_df = emu->df;
+    uint32_t old_op1 = emu->op1;
+    uint32_t old_op2 = emu->op2;
+    uint32_t old_res = emu->res;
+    // uc_link
+    i386_ucontext_t* old_uc_link = emu->uc_link;
+    emu->uc_link = NULL;
+
+    PushExit(emu);
+    R_EIP = 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;
+        // and the old registers
+        R_EBX = old_ebx;
+        R_EDI = old_edi;
+        R_ESI = old_esi;
+        R_EBP = old_ebp;
+        R_ESP = old_esp;
+        R_EIP = old_eip;  // and set back instruction pointer
    }
-#ifdef DYNAREC
-    if(!box86_dynarec)
-#endif
-        EmuCall(emu, addr);
-#ifdef DYNAREC
-    else {
-        uint32_t old_esp = R_ESP;
-        uint32_t old_ebx = R_EBX;
-        uint32_t old_edi = R_EDI;
-        uint32_t old_esi = R_ESI;
-        uint32_t old_ebp = R_EBP;
-        uint32_t old_eip = R_EIP;
-        PushExit(emu);
-        R_EIP = addr;
-        emu->df = d_none;
-        dynablock_t* block = NULL;
-        dynablock_t* current = NULL;
-        while(!emu->quit) {
-            block = DBGetBlock(emu, R_EIP, 1, &current);
-            current = block;
-            if(!block || !block->block || !block->done) {
-                // 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 Interpretor @%p, emu=%p\n", GetTID(), (void*)R_EIP, emu);
-                Run(emu, 1);
-            } else {
-                dynarec_log(LOG_DEBUG, "%04d|Calling DynaRec Block @%p (%p) of %d x86 instructions (father=%p) emu=%p\n", GetTID(), (void*)R_EIP, block->block, block->isize ,block->father, emu);
-                CHECK_FLAGS(emu);
-                // block is here, let's run it!
-                #ifdef ARM
-                arm_prolog(emu, block->block);
-                #endif
-            }
-            if(emu->fork) {
-                int forktype = emu->fork;
-                emu->quit = 0;
-                emu->fork = 0;
-                emu = x86emu_fork(emu, forktype);
-                if(emu->type == EMUTYPE_MAIN) {
-                    ejb = GetJmpBuf();
-                    ejb->emu = emu;
-                    ejb->jmpbuf_ok = 1;
-                    jmpbuf_reset = 1;
-                    if(sigsetjmp((struct __jmp_buf_tag*)ejb->jmpbuf, 1)) {
-                        printf_log(LOG_DEBUG, "Setjmp inner DynaCall, fs=0x%x\n", ejb->emu->segs[_FS]);
-                        addr = R_EIP;
-                    }
-                }
-            }
-        }
-        emu->quit = 0;  // reset Quit flags...
-        emu->df = d_none;
-        if(emu->quitonlongjmp && emu->longjmp) {
-            emu->longjmp = 0;   // don't change anything because of the longjmp
-        } else {
-            R_EBX = old_ebx;
-            R_EDI = old_edi;
-            R_ESI = old_esi;
-            R_EBP = old_ebp;
-            R_ESP = old_esp;
-            R_EIP = old_eip;  // and set back instruction pointer
-        }
-    }
-#endif
-    // clear the setjmp
-    if(ejb && jmpbuf_reset)
-        ejb->jmpbuf_ok = 0;
 }

 int my_setcontext(x86emu_t* emu, void* ucp);
 int DynaRun(x86emu_t* emu)
 {
    // prepare setjump for signal handling
-    emu_jmpbuf_t *ejb = NULL;
-#ifdef DYNAREC
-    int jmpbuf_reset = 1;
-#endif
-    if(emu->type == EMUTYPE_MAIN) {
-        ejb = GetJmpBuf();
-        if(!ejb->jmpbuf_ok) {
-            ejb->emu = emu;
-            ejb->jmpbuf_ok = 1;
-            int a;
-#ifdef DYNAREC
-            jmpbuf_reset = 1;
-#endif
-            if((a=sigsetjmp((struct __jmp_buf_tag*)ejb->jmpbuf, 1))) {
-                printf_log(LOG_DEBUG, "Setjmp DynaRun %d, fs=0x%x\n", a, ejb->emu->segs[_FS]);
-                if(a==2)
-                    Run(emu, 1);    // "single step" next instruction that is doing auto-smc
+    JUMPBUFF jmpbuf[1] = {0};
+    int skip = 0;
+    JUMPBUFF *old_jmpbuf = emu->jmpbuf;
+    uintptr_t old_savesp = emu->xSPSave;
+    emu->flags.jmpbuf_ready = 0;

+    while(!(emu->quit)) {
+        if(!emu->jmpbuf || (emu->flags.need_jmpbuf && emu->jmpbuf!=jmpbuf)) {
+            emu->jmpbuf = jmpbuf;
+            emu->old_savedsp = emu->xSPSave;
+            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\n");
+                #ifdef DYNAREC
+                if(box86_dynarec_test) {
+                    if(emu->test.clean)
+                        x86test_check(emu, R_EIP);
+                    emu->test.clean = 0;
+                }
+                #endif
            }
        }
-    }
-#ifdef DYNAREC
-    if(!box86_dynarec)
-#endif
-        return Run(emu, 0);
-#ifdef DYNAREC
-    else {
-        dynablock_t* block = NULL;
-        dynablock_t* current = NULL;
-        while(!emu->quit) {
-            block = DBGetBlock(emu, R_EIP, 1, &current);
-            current = block;
+        if(emu->flags.need_jmpbuf)
+            emu->flags.need_jmpbuf = 0;
+
+#ifndef DYNAREC
+        Run(emu, 0);
+#else
+        if(!box86_dynarec)
+            Run(emu, 0);
+        else {
+            dynablock_t* block = (skip || ACCESS_FLAG(F_TF))?NULL:DBGetBlock(emu, R_EIP, 1);
            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|Running Interpretor @%p, emu=%p\n", GetTID(), (void*)R_EIP, emu);
+                dynarec_log(LOG_DEBUG, "%04d|Running Interpreter @%p, emu=%p\n", GetTID(), (void*)R_EIP, emu);
+                if(box86_dynarec_test)
+                    emu->test.clean = 0;
                Run(emu, 1);
            } else {
-                dynarec_log(LOG_DEBUG, "%04d|Running DynaRec Block @%p (%p) of %d x86 insts (father=%p) emu=%p\n", GetTID(), (void*)R_EIP, block->block, block->isize, block->father, emu);
+                dynarec_log(LOG_DEBUG, "%04d|Running DynaRec Block @%p (%p) of %d x86 insts (hash=0x%x) emu=%p\n", GetTID(), (void*)R_EIP, block->block, block->isize, block->hash, emu);
                // block is here, let's run it!
-                #ifdef ARM
                arm_prolog(emu, block->block);
-                #endif
            }
            if(emu->fork) {
                int forktype = emu->fork;
                emu->quit = 0;
                emu->fork = 0;
                emu = x86emu_fork(emu, forktype);
-                if(emu->type == EMUTYPE_MAIN) {
-                    ejb = GetJmpBuf();
-                    ejb->emu = emu;
-                    ejb->jmpbuf_ok = 1;
-                    jmpbuf_reset = 1;
-                    if(sigsetjmp((struct __jmp_buf_tag*)ejb->jmpbuf, 1))
-                        printf_log(LOG_DEBUG, "Setjmp inner DynaRun, fs=0x%x\n", ejb->emu->segs[_FS]);
-                }
            }
-            else if(emu->quit && emu->uc_link) {
+            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
-    if(ejb && jmpbuf_reset)
-        ejb->jmpbuf_ok = 0;
-    return 0;
-#endif
+    emu->jmpbuf = old_jmpbuf;
+    emu->xSPSave = old_savesp;
 }
@@ -73,8 +73,7 @@ void add_next(dynarec_arm_t *dyn, uintptr_t addr) {
        }
    // add slots
    if(dyn->next_sz == dyn->next_cap) {
-        dyn->next_cap += 16;
-        dyn->next = (uintptr_t*)box_realloc(dyn->next, dyn->next_cap*sizeof(uintptr_t));
+        printf_log(LOG_NONE, "Warning, overallocating next\n");
    }
    dyn->next[dyn->next_sz++] = addr;
 }
@@ -83,16 +82,32 @@ uintptr_t get_closest_next(dynarec_arm_t *dyn, uintptr_t addr) {
    uintptr_t best = 0;
    int i = 0;
    while((i<dyn->next_sz) && (best!=addr)) {
-        if(dyn->next[i]<addr) { // remove the address, it's before current address
-            dyn->next[i] = 0;
-        } else {
-            if((dyn->next[i]<best) || !best)
-                best = dyn->next[i];
+        if(dyn->next[i]) {
+            if(dyn->next[i]<addr) { // remove the address, it's before current address
+                dyn->next[i] = 0;
+            } else {
+                if((dyn->next[i]<best) || !best)
+                    best = dyn->next[i];
+            }
        }
        ++i;
    }
    return best;
 }
+void add_jump(dynarec_arm_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_arm_t *dyn, int next) {
+    for(int i=0; i<dyn->jmp_sz; ++i)
+        if(dyn->insts[dyn->jmps[i]].x86.jmp == next)
+            return dyn->jmps[i];
+    return -2;
+}
+
 #define PK(A) (*((uint8_t*)(addr+(A))))
 int is_nops(dynarec_arm_t *dyn, uintptr_t addr, int n)
 {
@@ -241,54 +256,71 @@ int is_instructions(dynarec_arm_t *dyn, uintptr_t addr, int n)
    return (i==n)?1:0;
 }

-static instsize_t* addInst(instsize_t* insts, size_t* size, size_t* cap, int x86_size, int arm_size)
+void addInst(instsize_t* insts, size_t* size, int x86_size, int native_size)
 {
    // x86 instruction is <16 bytes
    int toadd;
-    if(x86_size>arm_size)
+    if(x86_size>native_size)
        toadd = 1 + x86_size/15;
    else
-        toadd = 1 + arm_size/15;
-    if((*size)+toadd>(*cap)) {
-        *cap = (*size)+toadd;
-        insts = (instsize_t*)box_realloc(insts, (*cap)*sizeof(instsize_t));
-    }
+        toadd = 1 + native_size/15;
    while(toadd) {
        if(x86_size>15)
            insts[*size].x86 = 15;    
        else
            insts[*size].x86 = x86_size;
        x86_size -= insts[*size].x86;
-        if(arm_size>15)
+        if(native_size>15)
            insts[*size].nat = 15;
        else
-            insts[*size].nat = arm_size;
-        arm_size -= insts[*size].nat;
+            insts[*size].nat = native_size;
+        native_size -= insts[*size].nat;
        ++(*size);
        --toadd;
    }
-    return insts;
 }

-static void fillPredecessors(dynarec_arm_t* dyn)
+static void recurse_mark_alive(dynarec_arm_t* dyn, int i)
+{
+    if(dyn->insts[i].x86.alive)
+        return;
+    dyn->insts[i].x86.alive = 1;
+    if(dyn->insts[i].x86.jmp && dyn->insts[i].x86.jmp_insts!=-1)
+        recurse_mark_alive(dyn, dyn->insts[i].x86.jmp_insts);
+    if(i<dyn->size-1 && dyn->insts[i].x86.has_next)
+        recurse_mark_alive(dyn, i+1);
+}
+
+static int sizePredecessors(dynarec_arm_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].x86.jmp && dyn->insts[i].x86.jmp_insts!=-1) {
-            ++pred_sz;
-            ++dyn->insts[dyn->insts[i].x86.jmp_insts].pred_sz;
+        if(dyn->insts[i].x86.alive && dyn->insts[i].x86.jmp && ((jmpto=dyn->insts[i].x86.jmp_insts)!=-1)) {
+            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].x86.has_next && !dyn->insts[i+1].x86.alive)
+            dyn->insts[i].x86.has_next = 0;
+    }
    // second the "has_next"
    for(int i=0; i<dyn->size-1; ++i) {
        if(dyn->insts[i].x86.has_next) {
-            ++pred_sz;
-            ++dyn->insts[i+1].pred_sz;
+            pred_sz++;
+            dyn->insts[i+1].pred_sz++;
        }
    }
-    dyn->predecessor = (int*)box_malloc(pred_sz*sizeof(int));
+    return pred_sz;
+}
+static void fillPredecessors(dynarec_arm_t* dyn)
+{
    // fill pred pointer
    int* p = dyn->predecessor;
    for(int i=0; i<dyn->size; ++i) {
@@ -297,13 +329,14 @@ static void fillPredecessors(dynarec_arm_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) {
-        if(i!=dyn->size-1 && dyn->insts[i].x86.has_next && (!i || dyn->insts[i].pred_sz))
+    for(int i=0; i<dyn->size; ++i) if(dyn->insts[i].x86.alive) {
+        if((i!=dyn->size-1) && dyn->insts[i].x86.has_next)
            dyn->insts[i+1].pred[dyn->insts[i+1].pred_sz++] = i;
-        if(dyn->insts[i].x86.jmp && dyn->insts[i].x86.jmp_insts!=-1)
-            dyn->insts[dyn->insts[i].x86.jmp_insts].pred[dyn->insts[dyn->insts[i].x86.jmp_insts].pred_sz++] = i;
+        if(dyn->insts[i].x86.jmp && (dyn->insts[i].x86.jmp_insts!=-1)) {
+            int j = dyn->insts[i].x86.jmp_insts;
+            dyn->insts[j].pred[dyn->insts[j].pred_sz++] = i;
+        }
    }
-
 }

 // updateNeed goes backward, from last intruction to top
@@ -311,14 +344,10 @@ static int updateNeed(dynarec_arm_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].x86.need_after;
-        if((need&X_PEND) && (dyn->insts[ninst].x86.state_flags==SF_SUBSET)) {
+        if((need&X_PEND) && (dyn->insts[ninst].x86.state_flags==SF_SUBSET || dyn->insts[ninst].x86.state_flags==SF_SET || dyn->insts[ninst].x86.state_flags==SF_SET_NODF)) {
            need &=~X_PEND;
            need |= X_ALL;
        }
-        if((need&X_PEND) && (dyn->insts[ninst].x86.state_flags==SF_SET)) {
-            need &=~X_PEND;
-            need |= dyn->insts[ninst].x86.set_flags;    // SF_SET will compute all flags, it's not SUBSET!
-        }
        if((need&X_PEND) && dyn->insts[ninst].x86.state_flags==SF_SUBSET_PENDING) {
            need |= X_ALL&~(dyn->insts[ninst].x86.set_flags);
        }
@@ -327,11 +356,11 @@ static int updateNeed(dynarec_arm_t* dyn, int ninst, uint8_t need) {
            dyn->insts[ninst].x86.gen_flags |= X_PEND;
        dyn->insts[ninst].x86.need_after = need;
        need = dyn->insts[ninst].x86.need_after&~dyn->insts[ninst].x86.gen_flags;
+
        if(dyn->insts[ninst].x86.may_set)
            need |= dyn->insts[ninst].x86.gen_flags;    // forward the flags
-        // Consume X_PEND if relevant
-        if((need&X_PEND) && (dyn->insts[ninst].x86.set_flags&SF_PENDING))
-            need &=~X_PEND;
+        else if((need&X_PEND) && (dyn->insts[ninst].x86.set_flags&SF_PENDING))
+            need &=~X_PEND;         // Consume X_PEND if relevant
        need |= dyn->insts[ninst].x86.use_flags;
        if(dyn->insts[ninst].x86.need_before == need)
            return ninst - 1;
@@ -346,158 +375,274 @@ static int updateNeed(dynarec_arm_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 instruction_arm_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 CancelBlock(int need_lock)
+{
+    if(need_lock)
+        mutex_lock(&my_context->mutex_dyndump);
+    dynarec_arm_t* helper = (dynarec_arm_t*)current_helper;
+    if(helper) {
+        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);
+}
+
 uintptr_t arm_pass0(dynarec_arm_t* dyn, uintptr_t addr);
 uintptr_t arm_pass1(dynarec_arm_t* dyn, uintptr_t addr);
 uintptr_t arm_pass2(dynarec_arm_t* dyn, uintptr_t addr);
 uintptr_t arm_pass3(dynarec_arm_t* dyn, uintptr_t addr);
+void arm_epilog();
+void arm_next();

-__thread void* current_helper = NULL;
-
-void CancelBlock()
-{
-    dynarec_arm_t* helper = (dynarec_arm_t*)current_helper;
-    current_helper = NULL;
-    if(!helper)
-        return;
-    box_free(helper->next);
-    box_free(helper->insts);
-    box_free(helper->predecessor);
-    if(helper->dynablock && helper->dynablock->block)
-        FreeDynarecMap(helper->dynablock, (uintptr_t)helper->dynablock->block, helper->dynablock->size);
+void* CreateEmptyBlock(dynablock_t* block, uintptr_t addr) {
+    block->isize = 0;
+    block->done = 0;
+    size_t sz = 8*sizeof(void*);
+    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);
+        CancelBlock(0);
+        return NULL;
+    }
+    block->size = sz;
+    block->actual_block = actual_p;
+    block->block = p;
+    block->jmpnext = p;
+    *(dynablock_t**)actual_p = block;
+    *(void**)(p+4*sizeof(void*)) = arm_epilog;
+    CreateJmpNext(block->jmpnext, p+4*sizeof(void*));
+    // all done...
+    __clear_cache(actual_p, actual_p+sz);   // need to clear the cache before execution...
+    return block;
 }

 void* FillBlock(dynablock_t* block, uintptr_t addr) {
+    /*
+        A Block must have this layout:
+
+        0x0000..0x0003  : dynablock_t* : self
+        0x0004..4+4*n   : actual Native instructions, (n is the total number)
+        B ..    B+3     : dynablock_t* : self (as part of JmpNext, that simulate another block)
+        B+4 ..  B+15    : 3 Native code for jmpnext (or jmp epilog in case of empty block)
+        B+16 .. B+19    : jmpnext (or jmp_epilog) address
+        B+20 .. B+31    : empty (in case an architecture needs more than 3 opcodes)
+        B+32 .. B+32+sz : instsize (compressed array with each instruction lenght on x86 and native side)
+
+    */
 dynarec_log(LOG_DEBUG, "Asked to Fill block %p with %p\n", block, (void*)addr);
    if(addr>=box86_nodynarec_start && addr<box86_nodynarec_end) {
        dynarec_log(LOG_DEBUG, "Asked to fill a block in fobidden 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);
        return NULL;
    }
-    if(!isJumpTableDefault((void*)addr)) {
-        dynarec_log(LOG_DEBUG, "Asked to fill a block at %p, but JumpTable is not default\n", (void*)addr);
-        return NULL;
-    }
+    // protect the 1st page
+    protectDB(addr, 1);
    // init the helper
    dynarec_arm_t helper = {0};
    current_helper = &helper;
    helper.dynablock = block;
    helper.start = addr;
    uintptr_t start = addr;
-    helper.cap = 64; // needs epilog handling
-    helper.insts = (instruction_arm_t*)box_calloc(helper.cap, sizeof(instruction_arm_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;
    // pass 0, addresses, x86 jump addresses, overall size of the block
    uintptr_t end = arm_pass0(&helper, addr);
-    // no need for next anymore
-    box_free(helper.next);
-    helper.next_sz = helper.next_cap = 0;
-    helper.next = NULL;
+    if(helper.abort) {
+        if(box86_dynarec_dump || box86_dynarec_log)dynarec_log(LOG_NONE, "Abort dynablock on pass0\n");
+        CancelBlock(0);
+        return NULL;
+    }
    // basic checks
    if(!helper.size) {
        dynarec_log(LOG_DEBUG, "Warning, null-sized dynarec block (%p)\n", (void*)addr);
-        CancelBlock();
-        return (void*)block;
+        CancelBlock(0);
+        return CreateEmptyBlock(block, addr);
+    }
+    if(!isprotectedDB(addr, 1)) {
+        dynarec_log(LOG_INFO, "Warning, write on current page on pass0, aborting dynablock creation (%p)\n", (void*)addr);
+        CancelBlock(0);
+        return NULL;
    }
    // already protect the block and compute hash signature
-    protectDB(addr, end-addr);  //end is 1byte after actual end
+    // protect the block of it goes over the 1st page
+    if((addr&~box86_pagesize)!=(end&~box86_pagesize)) // need to protect some other pages too
+        protectDB(addr, end-addr);  //end is 1byte after actual end
    uint32_t hash = X31_hash_code((void*)addr, end-addr);
    // calculate barriers
-    for(int i=0; i<helper.size; ++i)
-        if(helper.insts[i].x86.jmp) {
-            uintptr_t j = helper.insts[i].x86.jmp;
-            if(j<start || j>=end) {
-                helper.insts[i].x86.jmp_insts = -1;
-                helper.insts[i].x86.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].x86.addr==j)
-                        k=i2;
+    for(int ii=0; ii<helper.jmp_sz; ++ii) {
+        int i = helper.jmps[ii];
+        uintptr_t j = helper.insts[i].x86.jmp;
+        helper.insts[i].x86.jmp_insts = -1;
+        if(j<start || j>=end || j==helper.insts[i].x86.addr) {
+            if(j==helper.insts[i].x86.addr) // if there is a loop on some opcode, make the block "always to tested"
+                helper.always_test = 1;
+            helper.insts[i].x86.need_after |= X_PEND;
+        } else {
+            // find jump address instruction
+            int k=-1;
+            int search = ((j>=helper.insts[0].x86.addr) && j<helper.insts[0].x86.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].x86.addr == j) {
+                    k = i2;
+                    search = 0;
+                } else if(helper.insts[i2].x86.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].x86.addr==j)
+                        k = imin;
+                    else if(helper.insts[imax].x86.addr==j)
+                        k = imax;
+                }
+            }
+            /*for(int i2=0; i2<helper.size && k==-1; ++i2) {
+                if(helper.insts[i2].x86.addr==j)
+                    k=i2;
+            }*/
+            if(k!=-1) {
                if(k!=-1 && !helper.insts[i].barrier_maybe)
                    helper.insts[k].x86.barrier |= BARRIER_FULL;
                helper.insts[i].x86.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);
-    // check for the optionnal barriers now
-    /*for(int i=helper.size-1; i>=0; --i) {
-        if(helper.insts[i].barrier_maybe) {
-            // out-of-block jump
-            if(helper.insts[i].x86.jmp_insts == -1) {
-                // nothing for now
-            } else {
-                // inside block jump
-                int k = helper.insts[i].x86.jmp_insts;
-                if(k>i) {
-                    // jump in the future
-                    if(helper.insts[k].pred_sz>1) {
-                        // with multiple flow, put a barrier
-                        helper.insts[k].x86.barrier|=BARRIER_FLAGS;
-                    }
-                } else {
-                    // jump back
-                    helper.insts[k].x86.barrier|=BARRIER_FLAGS;
-                }
-            }
-	}
-    }*/
-    // check to remove useless barrier, in case of jump when destination doesn't needs flags
-    /*for(int i=helper.size-1; i>=0; --i) {
-        int k;
-        if(helper.insts[i].x86.jmp
-        && ((k=helper.insts[i].x86.jmp_insts)>=0)
-        && helper.insts[k].x86.barrier&BARRIER_FLAGS) {
-            //TODO: optimize FPU barrier too
-            if((!helper.insts[k].x86.need_flags)
-             ||(helper.insts[k].x86.set_flags==X_ALL
-                  && helper.insts[k].x86.state_flags==SF_SET)
-             ||(helper.insts[k].x86.state_flags==SF_SET_PENDING)) {
-                //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "Removed barrier for inst %d\n", k);
-                helper.insts[k].x86.barrier &= ~BARRIER_FLAGS; // remove flag barrier
-             }
-        }
-    }*/
+
    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
    arm_pass1(&helper, addr);
+    if(helper.abort) {
+        if(box86_dynarec_dump || box86_dynarec_log)dynarec_log(LOG_NONE, "Abort dynablock on pass0\n");
+        CancelBlock(0);
+        return NULL;
+    }

    // pass 2, instruction size
    arm_pass2(&helper, addr);
+    if(helper.abort) {
+        if(box86_dynarec_dump || box86_dynarec_log)dynarec_log(LOG_NONE, "Abort dynablock on pass0\n");
+        CancelBlock(0);
+        return NULL;
+    }
    // ok, now allocate mapped memory, with executable flag on
-    int sz = helper.arm_size;
-    void* p = (void*)AllocDynarecMap(block, sz);
-    if(p==NULL) {
-        dynarec_log(LOG_DEBUG, "AllocDynarecMap(%p, %d) failed, cancelling block\n", block, sz);
-        CancelBlock();
+    size_t insts_rsize = (helper.insts_size+2)*sizeof(instsize_t);
+    insts_rsize = (insts_rsize+7)&~7;   // round the size...
+    size_t arm_size = (helper.arm_size+7)&~7;   // round the size...
+    // ok, now allocate mapped memory, with executable flag on
+    size_t sz = sizeof(void*) + arm_size + 8*sizeof(void*) + insts_rsize;
+    //           dynablock_t* block (arm insts) jmpnext code    instsize
+    void* actual_p = (void*)AllocDynarecMap(sz);
+    void* p = (void*)(((uintptr_t)actual_p) + sizeof(void*));
+    void* next = p + arm_size;
+    void* instsize = next + 8*sizeof(void*);
+    if(actual_p==NULL) {
+        dynarec_log(LOG_INFO, "AllocDynarecMap(%p, %zu) failed, cancelling block\n", block, sz);
+        CancelBlock(0);
        return NULL;
    }
    helper.block = p;
+    block->actual_block = actual_p;
    helper.arm_start = (uintptr_t)p;
-    if(helper.sons_size) {
-        helper.sons_x86 = (uintptr_t*)alloca(helper.sons_size*sizeof(uintptr_t));
-        helper.sons_arm = (void**)alloca(helper.sons_size*sizeof(void*));
-    }
-    // pass 3, emit (log emit arm opcode)
+    helper.jmp_next = (uintptr_t)next+sizeof(void*);
+    helper.instsize = (instsize_t*)instsize;
+    *(dynablock_t**)actual_p = block;
+    // pass 3, emit (log emit native opcode)
    if(box86_dynarec_dump) {
        dynarec_log(LOG_NONE, "%s%04d|Emitting %d bytes for %d x86 bytes", (box86_dynarec_dump>1)?"\e[01;36m":"", GetTID(), helper.arm_size, helper.isize); 
        printFunctionAddr(helper.start, " => ");
        dynarec_log(LOG_NONE, "%s\n", (box86_dynarec_dump>1)?"\e[m":"");
    }
+    size_t oldarmsize = helper.arm_size;
+    size_t oldinstsize = helper.insts_size;
    helper.arm_size = 0;
+    helper.insts_size = 0;  // reset
    arm_pass3(&helper, addr);
-    if(sz!=helper.arm_size) {
-        printf_log(LOG_NONE, "BOX86: Warning, size difference in block between pass2 (%d) & pass3 (%d)!\n", sz, helper.arm_size);
+    if(helper.abort) {
+        if(box86_dynarec_dump || box86_dynarec_log)dynarec_log(LOG_NONE, "Abort dynablock on pass0\n");
+        CancelBlock(0);
+        return NULL;
+    }
+    // keep size of instructions for signal handling
+    block->instsize = instsize;
+    // ok, free the helper now
+    helper.insts = NULL;
+    helper.instsize = NULL;
+    helper.predecessor = NULL;
+    block->size = sz;
+    block->isize = helper.size;
+    block->block = p;
+    block->jmpnext = next+sizeof(void*);
+    block->always_test = helper.always_test;
+    block->dirty = block->always_test;
+    *(dynablock_t**)next = block;
+    *(void**)(next+5*sizeof(void*)) = arm_next;
+    CreateJmpNext(block->jmpnext, next+5*sizeof(void*));
+    //block->x86_addr = (void*)start;
+    block->x86_size = end-start;
+    // all done...
+    __clear_cache(actual_p, actual_p+sz);   // need to clear the cache before execution...
+    block->hash = X31_hash_code(block->x86_addr, block->x86_size);
+    // Check if something changed, to abbort if it as
+    if((block->hash != hash)) {
+        dynarec_log(LOG_DEBUG, "Warning, a block changed while being processed hash(%p:%zu)=%x/%x\n", block->x86_addr, block->x86_size, block->hash, hash);
+        CancelBlock(0);
+        return NULL;
+    }
+    if((oldarmsize!=helper.arm_size)) {
+        printf_log(LOG_NONE, "BOx86: Warning, size difference in block between pass2 (%zu) & pass3 (%zu)!\n", sz, helper.arm_size);
        uint8_t *dump = (uint8_t*)helper.start;
        printf_log(LOG_NONE, "Dump of %d x86 opcodes:\n", helper.size);
        for(int i=0; i<helper.size; ++i) {
@@ -507,69 +652,27 @@ dynarec_log(LOG_DEBUG, "Asked to Fill block %p with %p\n", block, (void*)addr);
            printf_log(LOG_NONE, "\t%d -> %d\n", helper.insts[i].size2, helper.insts[i].size);
        }
        printf_log(LOG_NONE, " ------------\n");
-    }
-    // all done...
-    __clear_cache(p, p+sz);   // need to clear the cache before execution...
-    // keep size of instructions for signal handling
-    {
-        size_t cap = 1;
-        for(int i=0; i<helper.size; ++i)
-            cap += 1 + ((helper.insts[i].x86.size>helper.insts[i].size)?helper.insts[i].x86.size:helper.insts[i].size)/15;
-        size_t size = 0;
-        block->instsize = (instsize_t*)box_calloc(cap, sizeof(instsize_t));
-        for(int i=0; i<helper.size; ++i)
-            block->instsize = addInst(block->instsize, &size, &cap, helper.insts[i].x86.size, helper.insts[i].size/4);
-        block->instsize = addInst(block->instsize, &size, &cap, 0, 0);    // add a "end of block" mark, just in case
-    }
-    // ok, free the helper now
-    box_free(helper.insts);
-    helper.insts = NULL;
-    box_free(helper.next);
-    helper.next = NULL;
-    block->size = sz;
-    block->isize = helper.size;
-    block->block = p;
-    block->need_test = 0;
-    //block->x86_addr = (void*)start;
-    block->x86_size = end-start;
-    if(box86_dynarec_largest<block->x86_size)
-        box86_dynarec_largest = block->x86_size;
-    block->hash = X31_hash_code(block->x86_addr, block->x86_size);
-    // Check if something changed, to abbort if it as
-    if(block->hash != hash) {
-        dynarec_log(LOG_INFO, "Warning, a block changed while beeing processed hash(%p:%d)=%x/%x\n", block->x86_addr, block->x86_size, block->hash, hash);
-        CancelBlock();
+        CancelBlock(0);
        return NULL;
    }
-    // fill sons if any
-    dynablock_t** sons = NULL;
-    int sons_size = 0;
-    if(helper.sons_size) {
-        sons = (dynablock_t**)box_calloc(helper.sons_size, sizeof(dynablock_t*));
-        for (int i=0; i<helper.sons_size; ++i) {
-            int created = 1;
-            dynablock_t *son = AddNewDynablock(block->parent, helper.sons_x86[i], &created);
-            if(created) {    // avoid breaking a working block!
-                son->block = helper.sons_arm[i];
-                son->x86_addr = (void*)helper.sons_x86[i];
-                son->x86_size = end-helper.sons_x86[i];
-                if(!son->x86_size) {printf_log(LOG_NONE, "Warning, son with null x86 size! (@%p / ARM=%p)", son->x86_addr, son->block);}
-                son->father = block;
-                son->done = 1;
-                sons[sons_size++] = son;
-                if(!son->parent)
-                    son->parent = block->parent;
-            }
-        }
-        if(sons_size) {
-            block->sons = sons;
-            block->sons_size = sons_size;
-        } else
-            box_free(sons);
+    // ok, free the helper now
+    helper.insts = NULL;
+    if(insts_rsize/sizeof(instsize_t)<helper.insts_size) {
+        printf_log(LOG_NONE, "BOX86: Warning, ists_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:%zu, marking as need_test\n", block->x86_addr, block->x86_size);
+        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->x86_addr, block->x86_size);
    }
-    box_free(helper.predecessor);
-    helper.predecessor = NULL;
    current_helper = NULL;
-    block->done = 1;
+    //block->done = 1;
    return (void*)block;
 }
@@ -18,6 +18,7 @@
 #include "dynarec_arm.h"
 #include "dynarec_arm_private.h"
 #include "arm_printer.h"
+#include "custommem.h"

 #include "dynarec_arm_functions.h"
 #include "dynarec_arm_helper.h"
@@ -29,7 +30,8 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
    int32_t i32, j32;
    uint32_t u32;
    uint8_t gd, ed;
-    uint8_t wback, u8;
+    uint8_t wback, u8, wb1, wb2;
+    uint8_t gb1, gb2, eb1, eb2;
    int fixedaddress;

    MAYUSE(j32);
@@ -46,6 +48,17 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            emit_add32(dyn, ninst, gd, ed, x3, x14);
            break;

+        case 0x20:
+            INST_NAME("AND Eb, Gb");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_fsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGB(x2);
+            GETEBO(x14);
+            emit_and8(dyn, ninst, x1, x2, x14, x2);
+            EBBACK;
+            break;
+
        case 0x2B:
            INST_NAME("SUB Gd, FS:Ed");
            SETFLAGS(X_ALL, SF_SET_PENDING);
@@ -86,6 +99,13 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            emit_cmp32(dyn, ninst, gd, ed, x3, x14);
            break;

+        case 0x64:
+            addr = dynarecFS(dyn, addr, ip, ninst, ok, need_epilog);
+            break;
+        case 0x65:
+            addr = dynarecGS(dyn, addr, ip, ninst, ok, need_epilog);
+            break;
+
        case 0x67:
            // reduced EA size...
            opcode = F8;
@@ -95,7 +115,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    nextop = F8;
                    grab_fsdata(dyn, addr, ninst, x14);
                    GETGD;  // don't need GETGW here
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        ed = xEAX+(nextop&7);
                        if(ed!=gd) {
                            BFI(ed, gd, 0, 16);
@@ -103,6 +123,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    } else {
                        addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
                        STRH_REG(gd, x14, ed);
+                        SMWRITE2();
                    }
                    break;

@@ -111,9 +132,10 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    nextop=F8;
                    grab_fsdata(dyn, addr, ninst, x14);
                    GETGD;
-                    if((nextop&0xC0)==0xC0) {   // reg <= reg
+                    if(MODREG) {   // reg <= reg
                        MOV_REG(gd, xEAX+(nextop&7));
                    } else {                    // mem <= reg
+                        SMREAD();
                        addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
                        LDR_REG_LSL_IMM5(gd, ed, x14, 0);
                    }
@@ -123,12 +145,13 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    INST_NAME("POP FS:Ew16");
                    nextop=F8;
                    grab_fsdata(dyn, addr, ninst, x14);
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        POP1(xEAX+(nextop&7));  // 67 ignored
                    } else {
                        addr = geted16(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
                        POP1(x2);
                        STR_REG_LSL_IMM5(x2, x1, x14, 0);
+                        SMWRITE2();
                    }
                    break;

@@ -140,6 +163,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        MOV32(x2, u32);
                        ADD_REG_LSL_IMM5(x1, x1, x2, 0);
                    }
+                    SMREAD();
                    LDR_IMM9(xEAX, x1, 0);
                    break;

@@ -152,6 +176,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        ADD_REG_LSL_IMM5(x1, x1, x2, 0);
                    }
                    STR_IMM9(xEAX, x1, 0);
+                    SMWRITE2();
                    break;

                case 0xFF:
@@ -160,9 +185,10 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    switch((nextop>>3)&7) {
                        case 6: // Push Ed
                            INST_NAME("PUSH FS:Ew");
-                            if((nextop&0xC0)==0xC0) {   // reg
+                            if(MODREG) {   // reg
                                DEFAULT;
                            } else {                    // mem <= i32
+                                SMREAD();
                                addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
                                LDR_REG_LSL_IMM5(x3, ed, x14, 0);
                                PUSH1(x3);
@@ -176,17 +202,41 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    DEFAULT;
            }
            break;
-
+        case 0x88:
+            INST_NAME("MOV Eb, Gb");
+            grab_fsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            gd = (nextop&0x38)>>3;
+            gb2 = ((gd&4)>>2);
+            gb1 = xEAX+(gd&3);
+            if(gb2) {
+                gd = x3;
+                UXTB(gd, gb1, gb2);
+            } else {
+                gd = gb1;   // no need to extract
+            }
+            if(MODREG) {
+                ed = (nextop&7);
+                eb1 = xEAX+(ed&3);  // Ax, Cx, Dx or Bx
+                eb2 = ((ed&4)<<1);    // L or H
+                BFI(eb1, gd, eb2, 8);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
+                STRB_REG_LSL_IMM5(gd, ed, x14, 0);
+                SMWRITE2();
+            }
+            break;
        case 0x89:
            INST_NAME("MOV FS:Ed, Gd");
            grab_fsdata(dyn, addr, ninst, x14);
            nextop=F8;
            GETGD;
-            if((nextop&0xC0)==0xC0) {   // reg <= reg
+            if(MODREG) {   // reg <= reg
                MOV_REG(xEAX+(nextop&7), gd);
            } else {                    // mem <= reg
                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                STR_REG_LSL_IMM5(gd, ed, x14, 0);
+                SMWRITE2();
            }
            break;

@@ -195,9 +245,10 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            grab_fsdata(dyn, addr, ninst, x14);
            nextop=F8;
            GETGD;
-            if((nextop&0xC0)==0xC0) {   // reg <= reg
+            if(MODREG) {   // reg <= reg
                MOV_REG(gd, xEAX+(nextop&7));
            } else {                    // mem <= reg
+                SMREAD();
                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                LDR_REG_LSL_IMM5(gd, ed, x14, 0);
            }
@@ -207,14 +258,18 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("POP FS:Ed");
            grab_fsdata(dyn, addr, ninst, x14);
            nextop = F8;
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                POP1(xEAX+(nextop&7));
            } else {
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                POP1(x2);
                STR_REG_LSL_IMM5(x2, ed, x14, 0);
+                SMWRITE2();
            }
            break;
+        case 0x90:
+            INST_NAME("NOP");
+            break;

        case 0xA0:
            INST_NAME("MOV AL, FS:Id");
@@ -224,6 +279,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                LDRB_IMM9(xEAX, x1, i32);
            } else {
                MOV32(x2, i32);
+                SMREAD();
                LDRB_REG_LSL_IMM5(xEAX, x1, x2, 0);
            }
            break;
@@ -235,6 +291,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                LDR_IMM9(xEAX, x1, i32);
            } else {
                MOV32(x2, i32);
+                SMREAD();
                LDR_REG_LSL_IMM5(xEAX, x1, x2, 0);
            }
            break;
@@ -247,6 +304,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            } else {
                MOV32(x2, i32);
                STRB_REG_LSL_IMM5(xEAX, x1, x2, 0);
+                SMWRITE2();
            }
            break;
        case 0xA3:
@@ -255,9 +313,11 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            i32 = F32S;
            if(i32>-4096 && i32<4096) {
                STR_IMM9(xEAX, x1, i32);
+                SMWRITE2();
            } else {
                MOV32(x2, i32);
                STR_REG_LSL_IMM5(xEAX, x2, x1, 0);
+                SMWRITE2();
            }
            break;

@@ -274,7 +334,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("MOV FS:Ed, Id");
            grab_fsdata(dyn, addr, ninst, x14);
            nextop=F8;
-            if((nextop&0xC0)==0xC0) {   // reg <= i32
+            if(MODREG) {   // reg <= i32
                i32 = F32S;
                ed = xEAX+(nextop&7);
                MOV32(ed, i32);
@@ -283,6 +343,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                i32 = F32S;
                MOV32(x3, i32);
                STR_REG_LSL_IMM5(x3, ed, x14, 0);
+                SMWRITE2();
            }
            break;
            
@@ -298,29 +359,54 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            switch((nextop>>3)&7) {
                case 2: // CALL Ed
                    INST_NAME("CALL FS:Ed");
-                    PASS2IF(ninst && dyn->insts[ninst-1].x86.set_flags, 1) {
+                    PASS2IF((box86_dynarec_safeflags>1) ||
+                        ((ninst && dyn->insts[ninst-1].x86.set_flags)
+                        || ((ninst>1) && dyn->insts[ninst-2].x86.set_flags)), 1)
+                    {
                        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
                    }
-                    if((nextop&0xC0)==0xC0) {   // reg
+                    if(MODREG) {   // reg
                        MOV_REG(xEIP, xEAX+(nextop&7));
                    } else {                    // mem
                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                        LDR_REG_LSL_IMM5(xEIP, ed, x14, 0);
                    }
-                    BARRIER(BARRIER_FULL);
-                    *need_epilog = 0;
-                    *ok = 0;
+                    if(box86_dynarec_callret && box86_dynarec_bigblock>1) {
+                        BARRIER(BARRIER_FULL);
+                        BARRIER_NEXT(BARRIER_FULL);
+                    } else {
+                        BARRIER(BARRIER_FLOAT);
+                        *need_epilog = 0;
+                        *ok = 0;
+                    }
                    MOV32(x2, addr);
+                    if(box86_dynarec_callret) {
+                        SET_HASCALLRET();
+                        // Push actual return address
+                        if(addr < (dyn->start+dyn->isize)) {
+                            // there is a next...
+                            j32 = (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->arm_size)-8):0;
+                            MESSAGE(LOG_NONE, "\tCALLRET set return to +%di\n", j32>>2);
+                            ADR(c__, x3, j32);
+                        } else {
+                            MESSAGE(LOG_NONE, "\tCALLRET set return to Jmptable(%p)\n", (void*)addr);
+                            j32 = getJumpTableAddress(addr);
+                            MOV32(x3, j32);
+                            LDR_IMM9(x3, x3, 0);
+                        }
+                        PUSH(xSP, (1<<x2)|(1<<x3));
+                    }
                    PUSH1(x2);
                    jump_to_next(dyn, 0, xEIP, ninst);
                    break;
                case 6: // Push Ed
                    INST_NAME("PUSH FS:Ed");
-                    if((nextop&0xC0)==0xC0) {   // reg
+                    if(MODREG) {   // reg
                        DEFAULT;
                    } else {                    // mem <= i32
+                        SMREAD();
                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                        LDR_REG_LSL_IMM5(x3, ed, x14, 0);
                        PUSH1(x3);
@@ -50,7 +50,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            GETGD;
            GETEDO2(x14);
            emit_add32(dyn, ninst, ed, gd, x3, x14);
-            WBACK;
+            WBACK2;
            break;

        case 0x03:
@@ -97,7 +97,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            GETGD;
            GETEDO2(x14);
            emit_and32(dyn, ninst, ed, gd, x3, x14);
-            WBACK;
+            WBACK2;
            break;

        case 0x23:
@@ -118,7 +118,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            GETGD;
            GETEDO2(x14);
            emit_sub32(dyn, ninst, ed, gd, x3, x14);
-            WBACK;
+            WBACK2;
            break;

        case 0x2B:
@@ -139,7 +139,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            GETGD;
            GETEDO2(x14);
            emit_xor32(dyn, ninst, gd, ed, x3, x14);
-            WBACK;
+            WBACK2;
            break;

        case 0x33:
@@ -172,6 +172,13 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            emit_cmp32(dyn, ninst, gd, ed, x3, x14);
            break;

+        case 0x64:
+            addr = dynarecFS(dyn, addr, ip, ninst, ok, need_epilog);
+            break;
+        case 0x65:
+            addr = dynarecGS(dyn, addr, ip, ninst, ok, need_epilog);
+            break;
+
        case 0x69:
            INST_NAME("IMUL Gd, GS:Ed, Id");
            SETFLAGS(X_ALL, SF_PENDING);
@@ -202,7 +209,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    GETEBO(x1);
                    u8 = F8;
                    emit_add8c(dyn, ninst, x1, u8, x2, x14);
-                    EBBACK;
+                    EBBACK2;
                    break;
                case 1: //OR
                    INST_NAME("OR GS:Eb, Ib");
@@ -211,7 +218,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    GETEBO(x1);
                    u8 = F8;
                    emit_or8c(dyn, ninst, x1, u8, x2, x14);
-                    EBBACK;
+                    EBBACK2;
                    break;
                case 2: //ADC
                    INST_NAME("ADC GS:Eb, Ib");
@@ -221,7 +228,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    GETEBO(x1);
                    u8 = F8;
                    emit_adc8c(dyn, ninst, x1, u8, x2, x14);
-                    EBBACK;
+                    EBBACK2;
                    break;
                case 3: //SBB
                    INST_NAME("SBB GS:Eb, Ib");
@@ -231,7 +238,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    GETEBO(x1);
                    u8 = F8;
                    emit_sbb8c(dyn, ninst, x1, u8, x2, x14);
-                    EBBACK;
+                    EBBACK2;
                    break;
                case 4: //AND
                    INST_NAME("AND GS:Eb, Ib");
@@ -240,7 +247,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    GETEBO(x1);
                    u8 = F8;
                    emit_and8c(dyn, ninst, x1, u8, x2, x14);
-                    EBBACK;
+                    EBBACK2;
                    break;
                case 5: //SUB
                    INST_NAME("SUB GS:Eb, Ib");
@@ -249,7 +256,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    GETEBO(x1);
                    u8 = F8;
                    emit_sub8c(dyn, ninst, x1, u8, x2, x14);
-                    EBBACK;
+                    EBBACK2;
                    break;
                case 6: //XOR
                    INST_NAME("XOR GS:Eb, Ib");
@@ -258,7 +265,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    GETEBO(x1);
                    u8 = F8;
                    emit_xor8c(dyn, ninst, x1, u8, x2, x14);
-                    EBBACK;
+                    EBBACK2;
                    break;
                case 7: //CMP
                    INST_NAME("CMP GS:Eb, Ib");
@@ -290,7 +297,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    GETEDO2(x14);
                    if(opcode==0x81) i32 = F32S; else i32 = F8S;
                    emit_add32c(dyn, ninst, ed, i32, x3, x14);
-                    WBACK;
+                    WBACK2;
                    break;
                case 1: //OR
                    if(opcode==0x81) {INST_NAME("OR Ed, Id");} else {INST_NAME("OR Ed, Ib");}
@@ -299,7 +306,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    GETEDO2(x14);
                    if(opcode==0x81) i32 = F32S; else i32 = F8S;
                    emit_or32c(dyn, ninst, ed, i32, x3, x14);
-                    WBACK;
+                    WBACK2;
                    break;
                case 2: //ADC
                    if(opcode==0x81) {INST_NAME("ADC Ed, Id");} else {INST_NAME("ADC Ed, Ib");}
@@ -309,7 +316,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    GETEDO2(x14);
                    if(opcode==0x81) i32 = F32S; else i32 = F8S;
                    emit_adc32c(dyn, ninst, ed, i32, x3, x14);
-                    WBACK;
+                    WBACK2;
                    break;
                case 3: //SBB
                    if(opcode==0x81) {INST_NAME("SBB Ed, Id");} else {INST_NAME("SBB Ed, Ib");}
@@ -319,7 +326,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    GETEDO2(x14);
                    if(opcode==0x81) i32 = F32S; else i32 = F8S;
                    emit_sbb32c(dyn, ninst, ed, i32, x3, x14);
-                    WBACK;
+                    WBACK2;
                    break;
                case 4: //AND
                    if(opcode==0x81) {INST_NAME("AND Ed, Id");} else {INST_NAME("AND Ed, Ib");}
@@ -328,7 +335,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    GETEDO2(x14);
                    if(opcode==0x81) i32 = F32S; else i32 = F8S;
                    emit_and32c(dyn, ninst, ed, i32, x3, x14);
-                    WBACK;
+                    WBACK2;
                    break;
                case 5: //SUB
                    if(opcode==0x81) {INST_NAME("SUB Ed, Id");} else {INST_NAME("SUB Ed, Ib");}
@@ -337,7 +344,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    GETEDO2(x14);
                    if(opcode==0x81) i32 = F32S; else i32 = F8S;
                    emit_sub32c(dyn, ninst, ed, i32, x3, x14);
-                    WBACK;
+                    WBACK2;
                    break;
                case 6: //XOR
                    if(opcode==0x81) {INST_NAME("XOR Ed, Id");} else {INST_NAME("XOR Ed, Ib");}
@@ -346,7 +353,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    GETEDO2(x14);
                    if(opcode==0x81) i32 = F32S; else i32 = F8S;
                    emit_xor32c(dyn, ninst, ed, i32, x3, x14);
-                    WBACK;
+                    WBACK2;
                    break;
                case 7: //CMP
                    if(opcode==0x81) {INST_NAME("CMP Ed, Id");} else {INST_NAME("CMP Ed, Ib");}
@@ -369,11 +376,12 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            grab_tlsdata(dyn, addr, ninst, x14);
            nextop=F8;
            GETGD;
-            if((nextop&0xC0)==0xC0) {   // reg <= reg
+            if(MODREG) {   // reg <= reg
                MOV_REG(xEAX+(nextop&7), gd);
            } else {                    // mem <= reg
                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                STR_REG_LSL_IMM5(gd, ed, x14, 0);
+                SMWRITE2();
            }
            break;

@@ -382,9 +390,10 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            grab_tlsdata(dyn, addr, ninst, x14);
            nextop=F8;
            GETGD;
-            if((nextop&0xC0)==0xC0) {   // reg <= reg
+            if(MODREG) {   // reg <= reg
                MOV_REG(gd, xEAX+(nextop&7));
            } else {                    // mem <= reg
+                SMREAD();
                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                LDR_REG_LSL_IMM5(gd, ed, x14, 0);
            }
@@ -394,19 +403,24 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("POP GS:Ed");
            grab_tlsdata(dyn, addr, ninst, x14);
            nextop = F8;
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                POP1((xEAX+(nextop&7)));
            } else {
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                POP1(x2);
                STR_REG_LSL_IMM5(x2, ed, x14, 0);
+                SMWRITE2();
            }
            break;
+        case 0x90:
+            INST_NAME("NOP");
+            break;

        case 0xA1:
            INST_NAME("MOV EAX, GS:Id");
            grab_tlsdata(dyn, addr, ninst, x1);
            i32 = F32S;
+            SMREAD();
            if(i32>-4096 && i32<4096) {
                LDR_IMM9(xEAX, x1, i32);
            } else {
@@ -423,6 +437,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            MOV32(x2, u32);
            ADD_REG_LSL_IMM5(x2, x1, x2, 0);
            STRB_IMM9(xEAX, x2, 0);
+            SMWRITE2();
            break;

        case 0xA3:
@@ -432,13 +447,14 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            MOV32(x2, u32);
            ADD_REG_LSL_IMM5(x2, x1, x2, 0);
            STR_IMM9(xEAX, x2, 0);
+            SMWRITE2();
            break;

        case 0xC6:
            INST_NAME("MOV GS:Eb, Ib");
            grab_tlsdata(dyn, addr, ninst, x14);
            nextop=F8;
-            if((nextop&0xC0)==0xC0) {   // reg <= u8
+            if(MODREG) {   // reg <= u8
                u8 = F8;
                ed = (nextop&7);
                eb1 = xEAX+(ed&3);  // Ax, Cx, Dx or Bx
@@ -450,13 +466,14 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                u8 = F8;
                MOVW(x3, u8);
                STRB_REG_LSL_IMM5(x3, ed, x14, 0);
+                SMWRITE2();
            }
            break;
        case 0xC7:
            INST_NAME("MOV GS:Ed, Id");
            grab_tlsdata(dyn, addr, ninst, x14);
            nextop=F8;
-            if((nextop&0xC0)==0xC0) {   // reg <= i32
+            if(MODREG) {   // reg <= i32
                i32 = F32S;
                ed = xEAX+(nextop&7);
                MOV32(ed, i32);
@@ -465,6 +482,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                i32 = F32S;
                MOV32(x3, i32);
                STR_REG_LSL_IMM5(x3, ed, x14, 0);
+                SMWRITE2();
            }
            break;

@@ -477,37 +495,35 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
                    GETEDO2(x14);
                    emit_rol32c(dyn, ninst, ed, 1, x3, x14);
-                    WBACK;
+                    WBACK2;
                    break;
                case 1:
                    INST_NAME("ROR Ed, 1");
                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
                    GETEDO2(x14);
                    emit_ror32c(dyn, ninst, ed, 1, x3, x14);
-                    WBACK;
+                    WBACK2;
                    break;
                case 2:
                    INST_NAME("RCL Ed, 1");
                    MESSAGE(LOG_DUMP, "Need Optimization\n");
                    READFLAGS(X_CF);
-                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    SETFLAGS(X_OF|X_CF, SF_SET_DF);
                    MOVW(x2, 1);
                    GETEDO2(x14);
-                    wb = ed;
                    if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
-                    CALL_(rcl32, ed, (1<<x14));
-                    SBACK(wb);
+                    CALL_(rcl32, ed, (1<<x2));
+                    WBACK2;
                    break;
                case 3:
                    INST_NAME("RCR Ed, 1");
                    MESSAGE(LOG_DUMP, "Need Optimization\n");
                    READFLAGS(X_CF);
-                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    SETFLAGS(X_OF|X_CF, SF_SET_DF);
                    MOVW(x2, 1);
-                    wb = ed;
                    if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
-                    CALL_(rcr32, ed, (1<<x14));
-                    SBACK(wb);
+                    CALL_(rcr32, ed, (1<<x2));
+                    WBACK2;
                    break;
                case 4:
                case 6:
@@ -515,21 +531,21 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
                    GETEDO2(x14);
                    emit_shl32c(dyn, ninst, ed, 1, x3, x14);
-                    WBACK;
+                    WBACK2;
                    break;
                case 5:
                    INST_NAME("SHR Ed, 1");
                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
                    GETEDO2(x14);
                    emit_shr32c(dyn, ninst, ed, 1, x3, x14);
-                    WBACK;
+                    WBACK2;
                    break;
                case 7:
                    INST_NAME("SAR Ed, 1");
                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
                    GETEDO2(x14);
                    emit_sar32c(dyn, ninst, ed, 1, x3, x14);
-                    WBACK;
+                    WBACK2;
                    break;
            }
            break;
@@ -546,7 +562,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    RSB_IMM8(x3, x3, 0x20);
                    GETEDO2(x14);
                    MOV_REG_ROR_REG(ed, ed, x3);
-                    WBACK;
+                    WBACK2;
                    UFLAG_IF {  // calculate flags directly
                        CMPS_IMM8(x3, 31);
                        B_MARK(cNE);
@@ -567,7 +583,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    B_MARK2(cEQ);
                    GETEDO2(x14);
                    MOV_REG_ROR_REG(ed, ed, x3);
-                    WBACK;
+                    WBACK2;
                    UFLAG_IF {  // calculate flags directly
                        CMPS_IMM8(x3, 1);
                        B_MARK(cNE);
@@ -585,23 +601,21 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    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);
                    AND_IMM8(x2, xECX, 0x1f);
-                    wb = ed;
                    if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
-                    CALL_(rcl32, ed, (1<<x14));
-                    SBACK(wb);
+                    CALL_(rcl32, ed, (1<<x2));
+                    WBACK2;
                    break;
                case 3:
                    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);
                    AND_IMM8(x2, xECX, 0x1f);
-                    wb = ed;
                    if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
                    CALL_(rcr32, ed, (1<<x14));
-                    SBACK(wb);
+                    WBACK2;
                    break;
                case 4:
                case 6:
@@ -610,7 +624,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    AND_IMM8(x3, xECX, 0x1f);
                    GETEDO2(x14);
                    emit_shl32(dyn, ninst, ed, x3, x3, x14);
-                    WBACK;
+                    WBACK2;
                    break;
                case 5:
                    INST_NAME("SHR Ed, CL");
@@ -618,7 +632,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    AND_IMM8(x3, xECX, 0x1f);
                    GETEDO2(x14);
                    emit_shr32(dyn, ninst, ed, x3, x3, x14);
-                    WBACK;
+                    WBACK2;
                    break;
                case 7:
                    INST_NAME("SAR Ed, CL");
@@ -627,7 +641,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    GETEDO2(x14);
                    UFLAG_OP12(ed, x3);
                    MOV_REG_ASR_REG(ed, ed, x3);
-                    WBACK;
+                    WBACK2;
                    UFLAG_RES(ed);
                    UFLAG_DF(x3, d_sar32);
                    break;
@@ -646,9 +660,10 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            switch((nextop>>3)&7) {
                case 6: // Push Ed
                    INST_NAME("PUSH GS:Ed");
-                    if((nextop&0xC0)==0xC0) {   // reg
+                    if(MODREG) {   // reg
                        DEFAULT;
                    } else {                    // mem <= i32
+                        SMREAD();
                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                        LDR_REG_LSL_IMM5(x3, ed, x14, 0);
                        PUSH1(x3);
@@ -18,6 +18,7 @@
 #include "dynarec_arm.h"
 #include "dynarec_arm_private.h"
 #include "arm_printer.h"
+#include "custommem.h"

 #include "dynarec_arm_helper.h"
 #include "dynarec_arm_functions.h"
@@ -172,9 +173,9 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("SBB AX, Iw");
            READFLAGS(X_CF);
            SETFLAGS(X_ALL, SF_SET_PENDING);
-            i16 = F16;
+            i32 = F16;
            UXTH(x1, xEAX, 0);
-            emit_sbb16c(dyn, ninst, x1, i16, x3, x14);
+            emit_sbb16c(dyn, ninst, x1, i32, x3, x14);
            BFI(xEAX, x1, 0, 16);
            break;
        case 0x1E:
@@ -388,13 +389,13 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("IMUL Gw,Ew,Iw");
            SETFLAGS(X_ALL, SF_PENDING);
            nextop = F8;
-            UFLAG_DF(x1, d_imul16);
            GETEW(x1);
            i16 = F16S;
            MOVW(x2, i16);
            SMULBB(x2, ed, x2);
            UFLAG_RES(x2);
            BFI(xEAX+((nextop&0x38)>>3), x2, 0, 16);
+            UFLAG_DF(x1, d_imul16);
            break;
        case 0x6A:
            INST_NAME("PUSH Ib");
@@ -406,13 +407,13 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("IMUL Gw,Ew,Ib");
            SETFLAGS(X_ALL, SF_PENDING);
            nextop = F8;
-            UFLAG_DF(x1, d_imul16);
            GETEW(x1);
            i16 = F8S;
            MOVW(x2, i16);
            SMULBB(x2, ed, x2);
            UFLAG_RES(x2);
            BFI(xEAX+((nextop&0x38)>>3), x2, 0, 16);
+            UFLAG_DF(x1, d_imul16);
            break;

        case 0x81:
@@ -505,9 +506,9 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            break;

        case 0x87:
-            INST_NAME("XCHG Ew, Gw");
+            INST_NAME("(LOCK) XCHG Ew, Gw");
            nextop = F8;
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                GETGD;
                UXTH(x14, gd, 0);
                ed = xEAX + (nextop&7);
@@ -518,6 +519,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            } else {
                GETGD;
                UXTH(x14, gd, 0);
+                SMDMB();
                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, LOCK_LOCK);
                #if 0
                MARKLOCK;
@@ -531,6 +533,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                LDRH_IMM8(x1, ed, fixedaddress);
                STRH_IMM8(x14, ed, fixedaddress);
                #endif
+                SMDMB();
                BFI(gd, x1, 0, 16);
            }
            break;
@@ -539,7 +542,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("MOV Ew, Gw");
            nextop = F8;
            GETGD;  // don't need GETGW here
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                ed = xEAX+(nextop&7);
                if(ed!=gd) {
                    BFI(ed, gd, 0, 16);
@@ -547,21 +550,21 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            } else {
                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0, &lock);
                STRH_IMM8(gd, ed, fixedaddress);
-                if(lock) {DMB_ISH();}
+                SMWRITELOCK(lock);
            }
            break;
        case 0x8B:
            INST_NAME("MOV Gw, Ew");
            nextop = F8;
            GETGD;  // don't need GETGW neither
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                ed = xEAX+(nextop&7);
                if(ed!=gd) {
                    BFI(gd, ed, 0, 16);
                }
            } else {
                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0, &lock);
-                if(lock) {DMB_ISH();}
+                SMREADLOCK(lock);
                LDRH_IMM8(x1, ed, fixedaddress);
                BFI(gd, x1, 0, 16);
            }
@@ -571,23 +574,25 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            nextop = F8;
            MOV32(x2, offsetof(x86emu_t, segs[(nextop&0x38)>>3]));
            LDRH_REG(x1, xEmu, x2);
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                ed = xEAX+(nextop&7);
                BFI(ed, x1, 0, 16);
            } else {
                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0, NULL);
                STRH_IMM8(x1, ed, fixedaddress);
+                SMWRITE2();
            }
            break;

        case 0x8E:
            INST_NAME("MOV Seg,Ew");
            nextop = F8;
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                ed = xEAX+(nextop&7);
                MOV32(x2, offsetof(x86emu_t, segs[(nextop&0x38)>>3]));
                STRH_REG(ed, xEmu, x2);
            } else {
+                SMREAD();
                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0, NULL);
                LDRH_IMM8(x1, ed, fixedaddress);
                MOV32(x2, offsetof(x86emu_t, segs[(nextop&0x38)>>3]));
@@ -632,7 +637,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            break;
        case 0x9D:
            INST_NAME("POPF (16b)");
-            SETFLAGS(X_ALL, SF_SET);    // lower 16bits is all flags handled in dynarec
+            SETFLAGS(X_ALL, SF_SET_NODF);    // lower 16bits is all flags handled in dynarec
            LDRHA_IMM8(x2, xESP, 2);
            MOV32(x1, 0x7FD7);
            AND_REG_LSL_IMM5(x2, x2, x1, 0);
@@ -645,6 +650,8 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("MOV, AX, Od");
            u32 = F32;
            MOV32(x2, u32);
+            if(isLockAddress(u32)) lock=1; else lock = 0;
+            SMREADLOCK(lock);
            LDRH_IMM8(x2, x2, 0);
            BFI(xEAX, x2, 0, 16);
            break;
@@ -652,7 +659,9 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("MOV Od, AX");
            u32 = F32;
            MOV32(x2, u32);
+            if(isLockAddress(u32)) lock=1; else lock = 0;
            STRH_IMM8(xEAX, x2, 0);
+            SMWRITELOCK(lock);
            break;
        case 0xA5:
            INST_NAME("MOVSW");
@@ -717,26 +726,34 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            switch((nextop>>3)&7) {
                case 0:
                    INST_NAME("ROL Ew, Ib");
-                    SETFLAGS(X_OF|X_CF, SF_SET);
-                    GETEW(x1);
-                    u8 = F8;
-                    MOVW(x2, u8);
-                    CALL_(rol16, x1, (1<<x3));
-                    EWBACK;
+                    if(geted_ib(dyn, addr, ninst, nextop)&31) {
+                        SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
+                        GETEW(x1);
+                        u8 = F8;
+                        emit_rol16c(dyn, ninst, x1, u8&15, x2, x14);
+                        EWBACK;
+                    } else {
+                        FAKEED;
+                        F8;
+                    }
                    break;
                case 1:
                    INST_NAME("ROR Ew, Ib");
-                    SETFLAGS(X_OF|X_CF, SF_SET);
-                    GETEW(x1);
-                    u8 = F8;
-                    MOVW(x2, u8);
-                    CALL_(ror16, x1, (1<<x3));
-                    EWBACK;
+                    if(geted_ib(dyn, addr, ninst, nextop)&31) {
+                        SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
+                        GETEW(x1);
+                        u8 = F8;
+                        emit_ror16c(dyn, ninst, x1, u8&15, x2, x14);
+                        EWBACK;
+                    } else {
+                        FAKEED;
+                        F8;
+                    }
                    break;
                case 2:
                    INST_NAME("RCL Ew, Ib");
                    READFLAGS(X_CF);
-                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    SETFLAGS(X_OF|X_CF, SF_SET_DF);
                    GETEW(x1);
                    u8 = F8;
                    MOVW(x2, u8);
@@ -746,7 +763,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                case 3:
                    INST_NAME("RCR Ew, Ib");
                    READFLAGS(X_CF);
-                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    SETFLAGS(X_OF|X_CF, SF_SET_DF);
                    GETEW(x1);
                    u8 = F8;
                    MOVW(x2, u8);
@@ -756,39 +773,42 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                case 4:
                case 6:
                    INST_NAME("SHL Ew, Ib");
-                    SETFLAGS(X_ALL, SF_PENDING);
-                    GETEW(x1);
-                    u8 = F8;
-                    MOVW(x2, (u8&0x1f));
-                    UFLAG_OP12(ed, x2)
-                    MOV_REG_LSL_REG(ed, ed, x2);
-                    EWBACK;
-                    UFLAG_RES(ed);
-                    UFLAG_DF(x3, d_shl16);
+                    if(geted_ib(dyn, addr, ninst, nextop)&31) {
+                        SETFLAGS(X_ALL, SF_PENDING);
+                        GETEW(x1);
+                        u8 = F8;
+                        emit_shl16c(dyn, ninst, x1, u8&31, x2, x14);
+                        EWBACK;
+                    } else {
+                        FAKEED;
+                        F8;
+                    }
                    break;
                case 5:
                    INST_NAME("SHR Ed, Ib");
-                    SETFLAGS(X_ALL, SF_PENDING);
-                    GETEW(x1);
-                    u8 = F8;
-                    MOVW(x2, (u8&0x1f));
-                    UFLAG_OP12(ed, x2)
-                    MOV_REG_LSR_REG(ed, ed, x2);
-                    EWBACK;
-                    UFLAG_RES(ed);
-                    UFLAG_DF(x3, d_shr16);
+                    if(geted_ib(dyn, addr, ninst, nextop)&31) {
+                        SETFLAGS(X_ALL, SF_PENDING);
+                        GETEW(x1);
+                        u8 = F8;
+                        emit_shr16c(dyn, ninst, x1, u8&31, x2, x14);
+                        EWBACK;
+                    } else {
+                        FAKEED;
+                        F8;
+                    }
                    break;
                case 7:
                    INST_NAME("SAR Ed, Ib");
-                    SETFLAGS(X_ALL, SF_PENDING);
-                    GETSEW(x1);
-                    u8 = F8;
-                    MOVW(x2, (u8&0x1f));
-                    UFLAG_OP12(ed, x2)
-                    MOV_REG_ASR_REG(ed, ed, x2);
-                    EWBACK;
-                    UFLAG_RES(ed);
-                    UFLAG_DF(x3, d_sar16);
+                    if(geted_ib(dyn, addr, ninst, nextop)&31) {
+                        SETFLAGS(X_ALL, SF_PENDING);
+                        GETSEW(x1);
+                        u8 = F8;
+                        emit_sar16c(dyn, ninst, x1, u8&31, x2, x14);
+                        EWBACK;
+                    } else {
+                        FAKEED;
+                        F8;
+                    }
                    break;
            }
            break;
@@ -796,7 +816,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xC7:
            INST_NAME("MOV Ew, Iw");
            nextop = F8;
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                ed = xEAX+(nextop&7);
                u16 = F16;
                MOVW(x1, u16);
@@ -810,101 +830,178 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            break;

        case 0xD1:
-        case 0xD3:
            nextop = F8;
            switch((nextop>>3)&7) {
                case 0:
-                    if(opcode==0xD1) {
-                        INST_NAME("ROL Ew, 1");
-                        MOVW(x2, 1);
-                    } else {
-                        INST_NAME("ROL Ew, CL");
-                        AND_IMM8(x2, xECX, 0x1f);
-                    }
-                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    INST_NAME("ROL Ew, 1");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
                    GETEW(x1);
-                    CALL_(rol16, x1, (1<<x3));
+                    emit_rol16c(dyn, ninst, x1, 1, x2, x14);
                    EWBACK;
                    break;
                case 1:
-                    if(opcode==0xD1) {
-                        INST_NAME("ROR Ew, 1");
-                        MOVW(x2, 1);
-                    } else {
-                        INST_NAME("ROR Ew, CL");
-                        AND_IMM8(x2, xECX, 0x1f);
-                    }
-                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    INST_NAME("ROR Ew, 1");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
                    GETEW(x1);
-                    CALL_(ror16, x1, (1<<x3));
+                    emit_ror16c(dyn, ninst, x1, 1, x2, x14);
                    EWBACK;
                    break;
                case 2:
-                    if(opcode==0xD1) {INST_NAME("RCL Ew, 1"); } else { INST_NAME("RCL Ew, CL");}
+                    INST_NAME("RCL Ew, 1");
                    READFLAGS(X_CF);
-                    SETFLAGS(X_OF|X_CF, SF_SET);
-                    if(opcode==0xD1) {MOVW(x2, 1);} else {AND_IMM8(x2, xECX, 0x1f);}
+                    SETFLAGS(X_OF|X_CF, SF_SET_DF);
+                    MOVW(x2, 1);
                    GETEW(x1);
                    CALL_(rcl16, x1, (1<<x3));
                    EWBACK;
                    break;
                case 3:
-                    if(opcode==0xD1) {INST_NAME("RCR Ew, 1");} else {INST_NAME("RCR Ew, CL");}
+                    INST_NAME("RCR Ew, 1");
                    READFLAGS(X_CF);
-                    SETFLAGS(X_OF|X_CF, SF_SET);
-                    if(opcode==0xD1) {MOVW(x2, 1);} else {AND_IMM8(x2, xECX, 0x1f);}
+                    SETFLAGS(X_OF|X_CF, SF_SET_DF);
+                    MOVW(x2, 1);
                    GETEW(x1);
                    CALL_(rcr16, x1, (1<<x3));
                    EWBACK;
                    break;
                case 4:
                case 6:
-                    if(opcode==0xD1) {
-                        INST_NAME("SHL Ew, 1");
-                        MOVW(x14, 1);
-                    } else {
-                        INST_NAME("SHL Ew, CL");
-                        AND_IMM8(x14, xECX, 0x1f);
-                    }
-                    SETFLAGS(X_ALL, SF_PENDING);
+                    INST_NAME("SHL Ew, 1");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
                    GETEW(x1);
-                    UFLAG_OP12(ed, x14)
-                    MOV_REG_LSL_REG(ed, ed, x14);
+                    emit_shl16c(dyn, ninst, x1, 1, x2, x14);
                    EWBACK;
-                    UFLAG_RES(ed);
-                    UFLAG_DF(x3, d_shl16);
                    break;
                case 5:
-                    if(opcode==0xD1) {
-                        INST_NAME("SHR Ew, 1");
-                        MOVW(x14, 1);
-                    } else {
-                        INST_NAME("SHR Ew, CL");
-                        AND_IMM8(x14, xECX, 0x1f);
-                    }
-                    SETFLAGS(X_ALL, SF_PENDING);
+                    INST_NAME("SHR Ew, 1");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
                    GETEW(x1);
-                    UFLAG_OP12(ed, x14)
-                    MOV_REG_LSR_REG(ed, ed, x14);
+                    emit_shr16c(dyn, ninst, x1, 1, x2, x14);
                    EWBACK;
-                    UFLAG_RES(ed);
-                    UFLAG_DF(x3, d_shr16);
                    break;
                case 7:
-                    if(opcode==0xD1) {
-                        INST_NAME("SAR Ew, 1");
-                        MOVW(x14, 1);
-                    } else {
-                        INST_NAME("SAR Ew, CL");
-                        AND_IMM8(x14, xECX, 0x1f);
-                    }
-                    SETFLAGS(X_ALL, SF_PENDING);
+                    INST_NAME("SAR Ew, 1");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
                    GETSEW(x1);
-                    UFLAG_OP12(ed, x14)
-                    MOV_REG_ASR_REG(ed, ed, x14);
+                    emit_sar16c(dyn, ninst, x1, 1, x2, x14);
+                    EWBACK;
+                    break;
+            }
+            break;
+            
+        case 0xD3:
+            nextop = F8;
+            switch((nextop>>3)&7) {
+                case 0:
+                    INST_NAME("ROL Ew, CL");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    if(box86_dynarec_safeflags>1)
+                        MAYSETFLAGS();
+                    UFLAG_IF {
+                        TSTS_IMM8(xECX, 0x1f);
+                        B_NEXT(cEQ);
+                    }
+                    AND_IMM8(x2, xECX, 0x0f);
+                    RSB_IMM8(x2, x2, 16);
+                    GETEW(x1);
+                    ORR_REG_LSL_IMM5(ed, ed, ed, 16);
+                    MOV_REG_LSR_REG(ed, ed, x2);
+                    EWBACK;
+                    UFLAG_IF {  // calculate flags directly
+                        CMPS_IMM8(x2, 15);
+                            ADD_REG_LSR_IMM5_COND(cEQ, x3, ed, ed, 15);
+                            BFI_COND(cEQ, xFlags, x3, F_OF, 1);
+                        BFI(xFlags, ed, F_CF, 1);
+                        UFLAG_DF(x2, d_none);
+                    }
+                    break;
+                case 1:
+                    INST_NAME("ROR Ew, CL");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    if(box86_dynarec_safeflags>1)
+                        MAYSETFLAGS();
+                    UFLAG_IF {
+                        TSTS_IMM8(xECX, 0x1f);
+                        B_NEXT(cEQ);
+                    }
+                    AND_IMM8(x2, xECX, 0x0f);
+                    GETEW(x1);
+                    ORR_REG_LSL_IMM5(ed, ed, ed, 16);
+                    MOV_REG_LSR_REG(ed, ed, x2);
+                    EWBACK;
+                    UFLAG_IF {  // calculate flags directly
+                        CMPS_IMM8(x2, 1);
+                            MOV_REG_LSR_IMM5_COND(cEQ, x2, ed, 14); // x2 = d>>14
+                            XOR_REG_LSR_IMM8_COND(cEQ, x2, x2, x2, 1); // x2 = ((d>>14) ^ ((d>>14)>>1))
+                            BFI_COND(cEQ, xFlags, x2, F_OF, 1);
+                        MOV_REG_LSR_IMM5(x2, ed, 15);
+                        BFI(xFlags, x2, F_CF, 1);
+                        UFLAG_DF(x2, d_none);
+                    }
+                    break;
+                case 2:
+                    INST_NAME("RCL Ew, CL");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_OF|X_CF, SF_SET_DF);
+                    AND_IMM8(x2, xECX, 0x1f);
+                    GETEW(x1);
+                    CALL_(rcl16, x1, (1<<x3));
+                    EWBACK;
+                    break;
+                case 3:
+                    INST_NAME("RCR Ew, CL");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_OF|X_CF, SF_SET_DF);
+                    AND_IMM8(x2, xECX, 0x1f);
+                    GETEW(x1);
+                    CALL_(rcr16, x1, (1<<x3));
+                    EWBACK;
+                    break;
+                case 4:
+                case 6:
+                    INST_NAME("SHL Ew, CL");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                    if(box86_dynarec_safeflags>1)
+                        MAYSETFLAGS();
+                    UFLAG_IF {
+                        ANDS_IMM8(x2, xECX, 0x1f);
+                        B_NEXT(cEQ);
+                    } else {
+                        AND_IMM8(x2, xECX, 0x1f);
+                    }
+                    GETEW(x1);
+                    emit_shl16(dyn, ninst, x1, x2, x2, x14);
+                    EWBACK;
+                    break;
+                case 5:
+                    INST_NAME("SHR Ew, CL");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                    if(box86_dynarec_safeflags>1)
+                        MAYSETFLAGS();
+                    UFLAG_IF {
+                        ANDS_IMM8(x2, xECX, 0x1f);
+                        B_NEXT(cEQ);
+                    } else {
+                        AND_IMM8(x2, xECX, 0x1f);
+                    }
+                    GETEW(x1);
+                    emit_shr16(dyn, ninst, x1, x2, x2, x14);
+                    EWBACK;
+                    break;
+                case 7:
+                    INST_NAME("SAR Ew, CL");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                    if(box86_dynarec_safeflags>1)
+                        MAYSETFLAGS();
+                    UFLAG_IF {
+                        ANDS_IMM8(x2, xECX, 0x1f);
+                        B_NEXT(cEQ);
+                    } else {
+                        AND_IMM8(x2, xECX, 0x1f);
+                    }
+                    GETSEW(x1);
+                    emit_sar16(dyn, ninst, x1, x2, x2, x14);
                    EWBACK;
-                    UFLAG_RES(ed);
-                    UFLAG_DF(x3, d_sar16);
                    break;
            }
            break;
@@ -1052,19 +1149,20 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                case 5:
                    INST_NAME("IMUL AX, Ew");
                    SETFLAGS(X_ALL, SF_PENDING);
-                    UFLAG_DF(x1, d_imul16);
                    GETEW(x1);
                    SMULBB(x2, xEAX, ed);
                    UFLAG_RES(x2);
                    BFI(xEAX, x2, 0, 16);
                    MOV_REG_LSR_IMM5(x2, x2, 16);
                    BFI(xEDX, x2, 0, 16);
+                    UFLAG_DF(x1, d_imul16);
                    break;
                case 6:
                    INST_NAME("DIV Ew");
                    if(arm_div) {
                        GETEW(x1);
-                        SETFLAGS(X_ALL, SF_SET);
+                        SETFLAGS(X_ALL, SF_SET_NODF);
+                        SET_DFNONE(x2);
                        UXTH(x2, xEAX, 0);
                        ORR_REG_LSL_IMM5(x2, x2, xEDX, 16);
                        UDIV(x3, x2, ed);
@@ -1073,7 +1171,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        BFI(xEDX, x14, 0, 16);
                    } else {
                        MESSAGE(LOG_DUMP, "Need Optimization\n");
-                        SETFLAGS(X_ALL, SF_SET);
+                        SETFLAGS(X_ALL, SF_SET_DF);
                        GETEW(x1);
                        STM(xEmu, (1<<xEAX) | (1<<xECX) | (1<<xEDX));
                        CALL(div16, -1, 0);
@@ -1083,7 +1181,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                case 7:
                    INST_NAME("IDIV Ew");
                    MESSAGE(LOG_DUMP, "Need Optimization\n");
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_DF);
                    GETEW(x1);
                    STM(xEmu, (1<<xEAX) | (1<<xECX) | (1<<xEDX));
                    CALL(idiv16, -1, 0);
@@ -1122,6 +1220,11 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    emit_dec16(dyn, ninst, x1, x2, x14);
                    EWBACK;
                    break;
+                case 6:
+                    INST_NAME("PUSH Ew");
+                    GETEW(x1);
+                    PUSH16(x1, xESP);
+                    break;
                default:
                    DEFAULT;
            }
@@ -41,7 +41,6 @@ uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
        case 0x81:
        case 0x83:
            nextop = F8;
-            DMB_ISH();
            switch((nextop>>3)&7) {
                case 0: //ADD
                    if(opcode==0x81) {
@@ -50,8 +49,7 @@ uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                        INST_NAME("LOCK ADD Ew, Ib");
                    }
                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    DMB_ISH();
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
                        ed = xEAX+(nextop&7);
                        emit_add16c(dyn, ninst, ed, i16, x3, x14);
@@ -69,9 +67,9 @@ uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                            STREXH(x3, x1, wback);
                            CMPS_IMM8(x3, 0);
                            B_MARKLOCK(cNE);
+                            SMDMB();
                        }
                        if(!fixedaddress) {
-                            DMB_ISH();
                            B_NEXT(c__);
                        }
                        if(!fixedaddress || (fixedaddress && (fixedaddress&1))) {
@@ -84,24 +82,24 @@ uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                            CMPS_IMM8(x3, 0);
                            B_MARK(cNE);
                            STRH_IMM8(x1, wback, 0); // put the whole value
+                            SMDMB();
                        }
-                        DMB_ISH();
                    }
                    break;
                default:
                  DEFAULT;
            }
-            DMB_ISH();
+            SMDMB();
            break;
        case 0xFF:
            nextop = F8;
-            DMB_ISH();
+            SMDMB();
            switch((nextop>>3)&7)
            {
                case 0: // INC Ew
                    INST_NAME("INC Ew");
                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        GETEW(x1);
                        emit_inc16(dyn, ninst, ed, x3, x14);
                        EWBACK(x1);
@@ -118,9 +116,9 @@ uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                            STREXH(x3, x1, wback);
                            CMPS_IMM8(x3, 0);
                            B_MARKLOCK(cNE);
+                            SMDMB();
                        }
                        if(!fixedaddress) {
-                            DMB_ISH();
                            B_NEXT(c__);
                        }
                        if(!fixedaddress || (fixedaddress && (fixedaddress&1))) {
@@ -133,13 +131,14 @@ uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                            CMPS_IMM8(x3, 0);
                            B_MARK(cNE);
                            STRH_IMM8(x1, wback, 0);
+                            SMDMB();
                        }
                    }
                    break;
                case 1: //DEC Ew
                    INST_NAME("DEC Ew");
                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        GETEW(x1);
                        emit_dec32(dyn, ninst, ed, x3, x14);
                        EWBACK(x1);
@@ -156,9 +155,9 @@ uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                            STREXH(x3, x1, wback);
                            CMPS_IMM8(x3, 0);
                            B_MARKLOCK(cNE);
+                            SMDMB();
                        }
                        if(!fixedaddress) {
-                            DMB_ISH();
                            B_NEXT(c__);
                        }
                        if(!fixedaddress || (fixedaddress && (fixedaddress&3))) {
@@ -171,6 +170,7 @@ uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                            CMPS_IMM8(x3, 0);
                            B_MARK(cNE);
                            STRH_IMM8(x1, wback, 0);
+                            SMDMB();
                        }
                    }
                    break;
@@ -178,7 +178,7 @@ uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                    // default to NO LOCK
                    addr-=2;
            }
-            DMB_ISH();
+            SMDMB();
            break;
        default:
            DEFAULT;
@@ -50,7 +50,7 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    nextop = F8;
                    grab_fsdata(dyn, addr, ninst, x14);
                    GETGD;  // don't need GETGW here
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        ed = xEAX+(nextop&7);
                        if(ed!=gd) {
                            BFI(ed, gd, 0, 16);
@@ -58,6 +58,7 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    } else {
                        addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
                        STRH_REG(gd, x14, ed);
+                        SMWRITE2();
                    }
                    break;

@@ -66,12 +67,13 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    nextop=F8;
                    grab_fsdata(dyn, addr, ninst, x14);
                    GETGD;
-                    if((nextop&0xC0)==0xC0) {   // reg <= reg
+                    if(MODREG) {   // reg <= reg
                        ed = xEAX+(nextop&7);
                        if(ed!=gd) {
                            BFI(gd, ed, 0, 16);
                        }
                    } else {                    // mem <= reg
+                        SMREAD();
                        addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
                        LDR_REG_LSL_IMM5(gd, x14, ed, 0);
                    }
@@ -81,12 +83,13 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    INST_NAME("POP FS:Ew16");
                    nextop=F8;
                    grab_fsdata(dyn, addr, ninst, x14);
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        POP1(xEAX+(nextop&7));  // 67 ignored
                    } else {
                        addr = geted16(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
                        POP1(x2);
                        STR_REG_LSL_IMM5(x2, x14, ed, 0);
+                        SMWRITE2();
                    }
                    break;

@@ -98,6 +101,7 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        MOV32(x2, u32);
                        ADD_REG_LSL_IMM5(x1, x1, x2, 0);
                    }
+                    SMREAD();
                    LDR_IMM9(xEAX, x1, 0);
                    break;

@@ -110,6 +114,7 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        ADD_REG_LSL_IMM5(x1, x1, x2, 0);
                    }
                    STR_IMM9(xEAX, x1, 0);
+                    SMWRITE2();
                    break;

                case 0xFF:
@@ -118,9 +123,10 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    switch((nextop>>3)&7) {
                        case 6: // Push Ed
                            INST_NAME("PUSH FS:Ew");
-                            if((nextop&0xC0)==0xC0) {   // reg
+                            if(MODREG) {   // reg
                                DEFAULT;
                            } else {                    // mem <= i32
+                                SMREAD();
                                addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
                                LDR_REG_LSL_IMM5(x3, ed, x14, 0);
                                PUSH1(x3);
@@ -33,6 +33,7 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
    int d1;
    int fixedaddress;
    int parity;
+    uint8_t u8;

    MAYUSE(d1);
    MAYUSE(s0);
@@ -52,11 +53,15 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FADD ST0, STx");
            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(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VADD_F32(v1, v1, v2);
            } else {
                VADD_F64(v1, v1, v2);
            }
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            break;
        case 0xC8:
        case 0xC9:
@@ -69,11 +74,15 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FMUL ST0, STx");
            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(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VMUL_F32(v1, v1, v2);
            } else {
                VMUL_F64(v1, v1, v2);
            }
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            break;
        case 0xD0:
        case 0xD1:
@@ -110,7 +119,7 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                VCMP_F64(v1, v2);
            }
            FCOM(x1, x2);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xE0:
        case 0xE1:
@@ -123,11 +132,15 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FSUB ST0, STx");
            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(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VSUB_F32(v1, v1, v2);
            } else {
                VSUB_F64(v1, v1, v2);
            }
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            break;
        case 0xE8:
        case 0xE9:
@@ -140,11 +153,15 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FSUBR ST0, STx");
            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(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VSUB_F32(v1, v2, v1);
            } else {
                VSUB_F64(v1, v2, v1);
            }
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            break;
        case 0xF0:
        case 0xF1:
@@ -157,11 +174,15 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FDIV ST0, STx");
            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(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VDIV_F32(v1, v1, v2);
            } else {
                VDIV_F64(v1, v1, v2);
            }
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            break;
        case 0xF8:
        case 0xF9:
@@ -174,11 +195,15 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FDIVR ST0, STx");
            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(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VDIV_F32(v1, v2, v1);
            } else {
                VDIV_F64(v1, v2, v1);
            }
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            break;
      
        default:
@@ -196,12 +221,16 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        GETED;
                        VMOVtoV(s0, ed);
                    }
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    if(ST_IS_F(0)) {
                        VADD_F32(v1, v1, s0);
                    } else {
                        VCVT_F64_F32(d1, s0);
                        VADD_F64(v1, v1, d1);
                    }
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
                case 1:
                    INST_NAME("FMUL ST0, float[ED]");
@@ -216,12 +245,16 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        GETED;
                        VMOVtoV(s0, ed);
                    }
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    if(ST_IS_F(0)) {
                        VMUL_F32(v1, v1, s0);
                    } else {
                        VCVT_F64_F32(d1, s0);
                        VMUL_F64(v1, v1, d1);
                    }
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
                case 2:
                    INST_NAME("FCOM ST0, float[ED]");
@@ -264,7 +297,7 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        VCMP_F64(v1, d1);
                    }
                    FCOM(x1, x2);
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                    break;
                case 4:
                    INST_NAME("FSUB ST0, float[ED]");
@@ -279,12 +312,16 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        GETED;
                        VMOVtoV(s0, ed);
                    }
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    if(ST_IS_F(0)) {
                        VSUB_F32(v1, v1, s0);
                    } else {
                        VCVT_F64_F32(d1, s0);
                        VSUB_F64(v1, v1, d1);
                    }
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
                case 5:
                    INST_NAME("FSUBR ST0, float[ED]");
@@ -299,12 +336,16 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        GETED;
                        VMOVtoV(s0, ed);
                    }
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    if(ST_IS_F(0)) {
                        VSUB_F32(v1, s0, v1);
                    } else {
                        VCVT_F64_F32(d1, s0);
                        VSUB_F64(v1, d1, v1);
                    }
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
                case 6:
                    INST_NAME("FDIV ST0, float[ED]");
@@ -319,12 +360,16 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        GETED;
                        VMOVtoV(s0, ed);
                    }
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    if(ST_IS_F(0)) {
                        VDIV_F32(v1, v1, s0);
                    } else {
                        VCVT_F64_F32(d1, s0);
                        VDIV_F64(v1, v1, d1);
                    }
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
                case 7:
                    INST_NAME("FDIVR ST0, float[ED]");
@@ -339,12 +384,16 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        GETED;
                        VMOVtoV(s0, ed);
                    }
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    if(ST_IS_F(0)) {
                        VDIV_F32(v1, s0, v1);
                    } else {
                        VCVT_F64_F32(d1, s0);
                        VDIV_F64(v1, d1, v1);
                    }
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
                default:
                    DEFAULT;
@@ -67,7 +67,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xC6:
        case 0xC7:
            INST_NAME("FLD STx");
-            v2 = x87_do_push(dyn, ninst, x3, X87_ST(nextop&7));
+            X87_PUSH_OR_FAIL(v2, dyn, ninst, x3, 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)) {
                VMOV_32(v2, v1);
@@ -134,25 +134,49 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            i1 = x87_get_current_cache(dyn, ninst, 0, NEON_CACHE_ST_D);
            // value put in x14
            if(i1==-1) {
-                // not in cache, so check Empty status and load it
-                // x14 will be the actual top
-                LDR_IMM9(x14, xEmu, offsetof(x86emu_t, top));
-                i2 = -dyn->n.x87stack;
-                if(i2) {
-                    if(i2<0) {
-                        SUB_IMM8(x14, x14, -i2);
-                    } else {
-                        ADD_IMM8(x14, x14, i2);
+                if(fpu_is_st_freed(dyn, ninst, 0)) {
+                    MOV32(x14, 0b100000100000000);
+                    B_MARK3(c__);
+                } else {                // not in cache, so check Empty status and load it
+                    // load tag
+                    LDRH_IMM8(x3, xEmu, offsetof(x86emu_t, fpu_tags));
+                    if(i2) {
+                        if(i2<0) {
+                            MOV_REG_LSL_IMM5(x3, x3, -i2*2);
+                        } else {
+                            MOV32(x14, 0xffff);
+                            ORR_REG_LSL_IMM5(x3, x3, x14, 16);
+                            MOV_REG_LSR_IMM5(x3, x3, i2*2);
+                        }
                    }
-                    AND_IMM8(x14, x14, 7);    // (emu->top + i)&7
+                    TSTS_IMM8(x3, 0b11);
+                    MOVW(x14, 0b100000100000000);   // empty
+                    B_MARK3(cNE);
+                    // check stack  TODO: this test should not be needed, something is wrong with tag handling
+                    LDR_IMM9(x3, xEmu, offsetof(x86emu_t, fpu_stack));
+                    if(i2) {
+                        if(i2<0) {
+                            ADD_IMM8(x3, x3, -i2);
+                        } else {
+                            SUB_IMM8(x3, x3, i2);
+                        }
+                    }
+                    CMPS_IMM8(x3, 0);
+                    B_MARK3(cLE);
+                    // x14 will be the actual top
+                    LDR_IMM9(x14, xEmu, offsetof(x86emu_t, top));
+                    i2 = -dyn->n.x87stack;
+                    if(i2) {
+                        if(i2<0) {
+                            SUB_IMM8(x14, x14, -i2);
+                        } else {
+                            ADD_IMM8(x14, x14, i2);
+                        }
+                        AND_IMM8(x14, x14, 7);    // (emu->top + i)&7
+                    }
+                    ADD_REG_LSL_IMM5(x1, xEmu, x14, 3);
+                    LDRD_IMM8(x2, x1, offsetof(x86emu_t, x87)); // load r2/r3 with ST0 anyway, for sign extraction
                }
-                ADD_REG_LSL_IMM5(x1, xEmu, x14, 3);
-                LDRD_IMM8(x2, x1, offsetof(x86emu_t, x87)); // load r2/r3 with ST0 anyway, for sign extraction
-                ADD_REG_LSL_IMM5(x1, xEmu, x14, 2);
-                LDR_IMM9(x1, x1, offsetof(x86emu_t, p_regs));
-                CMPS_IMM8(x1, 0b11);
-                MOVW_COND(cEQ, x14, 0b100000100000000); // empty: C3,C2,C0 = 101
-                B_MARK3(cEQ);
            } else {
                // simply move from cache reg to r2/r3
                v1 = dyn->n.x87reg[i1];
@@ -194,7 +218,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,

        case 0xE8:
            INST_NAME("FLD1");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
            if(ST_IS_F(0)) {
                VMOV_i_32(v1, 0b01110000);
            } else {
@@ -204,7 +228,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            break;
        case 0xE9:
            INST_NAME("FLDL2T");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
            if(ST_IS_F(0)) {
                MOV32(x2, (&f_l2t));
                VLDR_32(v1, x2, 0);
@@ -216,7 +240,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            break;
        case 0xEA:     
            INST_NAME("FLDL2E");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
            if(ST_IS_F(0)) {
                MOV32(x2, (&f_l2e));
                VLDR_32(v1, x2, 0);
@@ -228,7 +252,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            break;
        case 0xEB:
            INST_NAME("FLDPI");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
            if(ST_IS_F(0)) {
                MOV32(x2, (&f_pi));
                VLDR_32(v1, x2, 0);
@@ -240,7 +264,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            break;
        case 0xEC:
            INST_NAME("FLDLG2");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
            if(ST_IS_F(0)) {
                MOV32(x2, (&f_lg2));
                VLDR_32(v1, x2, 0);
@@ -252,7 +276,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            break;
        case 0xED:
            INST_NAME("FLDLN2");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
            if(ST_IS_F(0)) {
                MOV32(x2, (&f_ln2));
                VLDR_32(v1, x2, 0);
@@ -264,7 +288,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            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, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
            if(ST_IS_F(0)) {
                VMOV_8(v1/2, 0);  // float is *2...
            } else {
@@ -281,18 +305,20 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            CALL(arm_f2xm1, -1, 0);
            #else
            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-            //if(ST0.d!=0.0)
-            //    ST0.d = exp2(ST0.d)-1.0;
-            VMOV_64(0, v1);
-            CALL_1D(exp2, 0);   // return is d0
            if((PK(0)==0xD9 && PK(1)==0xE8) && // next inst is FLD1
            (PK(2)==0xDE && PK(3)==0xC1)) {
                MESSAGE(LOG_DUMP, "Hack for fld1 / faddp st1, st0\n");
+                VMOV_64(0, v1);
+                CALL_1D(exp2, 0);   // return is d0
                VMOV_64(v1, 0);
                addr+=4;
            } else {
-                VMOV_i_64(v1, 0b01110000);  // 1.0
-                VSUB_F64(v1, 0, v1);
+                //ST0.d = expm1(LN2 * ST0.d);
+                MOV32(x2, (&d_ln2));
+                VLDR_64(0, x2, 0);
+                VMUL_F64(0, 0, v1);
+                CALL_1D(expm1, 0);   // return is d0
+                VMOV_64(v1, 0);
            }
            #endif
            // should set C1 to 0
@@ -301,44 +327,58 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FYL2X");
            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);
+
            VMOV_64(0, v1);    // prepare call to log2
            CALL_1D(log2, 0);
            VMUL_F64(v2, v2, 0);    //ST(1).d = log2(ST0.d)*ST(1).d
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            // should set C1 to 0
            break;
        case 0xF2:
            INST_NAME("FPTAN");
-            v2 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            X87_PUSH_OR_FAIL(v2, dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
            v1 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
+            // seems that tan of glib doesn't follow the rounding direction mode
+            if(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            VMOV_64(0, v1);    // prepare call to tan
-            CALL_1D(tan, 0);
+            CALL_1D(tan, box86_dynarec_fastround ? 0 : (1 << u8));
            VMOV_64(v1, 0);
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            //emu->sw.f.F87_C2 = 0; 
            //emu->sw.f.F87_C1 = 0; 
            LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
            BFC(x1, 9, 2); //C1 C2 = 0 0
            STRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
-            // so here: F64: Imm8 = abcd efgh that gives => aBbbbbbb bbcdefgh 0000000 00000000 00000000...
-            // and want 1.0 = 0x3ff0000000000000
-            // so 00111111 11110000 00000000 00000000....
-            // a = 0, b = 1, c = 1, d = 1, efgh=0
-            // 0b01110000
-            if(ST_IS_F(0)) {
-                VMOV_i_32(v2, 0b01110000);
+            if(PK(0)==0xdd && PK(1)==0xd8) {
+                MESSAGE(LOG_DUMP, "Optimized next DD D8 fstp st0, st0, not emiting 1\n");
            } else {
-                VMOV_i_64(v2, 0b01110000);
+                // so here: F64: Imm8 = abcd efgh that gives => aBbbbbbb bbcdefgh 0000000 00000000 00000000...
+                // and want 1.0 = 0x3ff0000000000000
+                // so 00111111 11110000 00000000 00000000....
+                // a = 0, b = 1, c = 1, d = 1, efgh=0
+                // 0b01110000
+                if(ST_IS_F(0)) {
+                    VMOV_i_32(v2, 0b01110000);
+                } else {
+                    VMOV_i_64(v2, 0b01110000);
+                }
            }
            break;
        case 0xF3:
            INST_NAME("FPATAN");
            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
            v2 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
+            if(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            VMOV_64(0, v2);    // prepare call to atan2
            VMOV_64(1, v1);
-            CALL_2D(atan2, 0);
+            CALL_2D(atan2, box86_dynarec_fastround ? 0 : (1 << u8));
            VMOV_64(v2, 0);    //ST(1).d = atan2(ST1.d, ST0.d);
-            x87_do_pop(dyn, ninst, x3);
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            // should set C1 to 0
            break;
        case 0xF4:
@@ -346,7 +386,9 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            MESSAGE(LOG_DUMP, "Need Optimization\n");
            x87_do_push_empty(dyn, ninst, 0);
            x87_forget(dyn, ninst, x1, x2, 1);
+            s0 = x87_stackcount(dyn, ninst, x3);
            CALL(arm_fxtract, -1, 0);
+            x87_unstackcount(dyn, ninst, x3, s0);
            // C1 set only if stack under/overflow occurs
            break;
        case 0xF5:
@@ -445,32 +487,45 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FYL2XP1");
            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);
-            VMOV_i_64(0, 0b01110000);   // D0 = 1.0
-            VADD_F64(0, 0, v1);    // prepare call to log2
-            CALL_1D(log2, 0);
-            VMUL_F64(v2, v2, 0);    //ST(1).d = log2(ST0.d + 1.0)*ST(1).d;
-            x87_do_pop(dyn, ninst, x3);
+
+            //ST(1).d = (ST(1).d * log1p(ST0.d)) / M_LN2;
+            VMOV_64(0, v1);    // prepare call to log1p
+            CALL_1D(log1p, 0);
+            VMUL_F64(v2, v2, 0);
+            MOV32(x2, (&d_ln2));
+            VLDR_64(0, x2, 0);
+            VDIV_F64(v2, v2, 0);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            // should set C1 to 0
            break;
        case 0xFA:
            INST_NAME("FSQRT");
            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+            if(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VSQRT_F32(v1, v1);
            } else {
                VSQRT_F64(v1, v1);
            }
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            // should set C1 to 0
            break;
        case 0xFB:
            INST_NAME("FSINCOS");
-            v2 = x87_do_push(dyn, ninst, x3, NEON_CACHE_ST_D);
+            X87_PUSH_OR_FAIL(v2, dyn, ninst, x3, NEON_CACHE_ST_D);
            v1 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
+            // seems that sin and cos function of glibc don't follow the rounding mode
+            if(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            VMOV_64(0, v1);
-            CALL_1D(sin, 0);
+            CALL_1D(sin, box86_dynarec_fastround ? 0 : (1 << u8));
            VSWP(v1, 0);
-            CALL_1D(cos, 0);    // would it be faster to do sqrt(1-sin()²) ???
+            CALL_1D(cos, box86_dynarec_fastround ? 0 : (1 << u8));    // would it be faster to do sqrt(1-sin()²) ???
            VMOV_64(v2, 0);
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            //emu->sw.f.F87_C2 = 0; C1 too
            LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
            BFC(x1, 9, 2); //C2 C1 = 0 0
@@ -505,26 +560,32 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
            v2 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
            //if(ST0.d!=0.0)
-            //    ST0.d *= exp2(trunc(ST1.d));
-            VCMP_F64_0(v1);
-            VMRS_APSR();
-            B_NEXT(cEQ);
-            if(!arm_v8) {
-                VMOV_64(0, v2);
-                CALL_1DD(trunc, exp2, 0);
-            } else {
-                VRINTZ_F64(0, v2);
-                CALL_1D(exp2, 0);
-            }
-            VMUL_F64(v1, v1, 0);
+            //    ST0.d = ldexp(ST0.d, trunc(ST1.d));
+            if(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
+            s0 = fpu_get_scratch_single(dyn);
+            // value of s0 =
+            // 2^31-1 (ST1 >= 2^31), -2^31 (ST1 < -2^31) or int(ST1) (other situations)
+            VCVT_S32_F64(s0 , v2);
+            VMOVfrV(x2, s0);
+            VMOV_64(0, v1);
+            CALL_1DDR(ldexp, x2, x3, box86_dynarec_fastround ? 0 : (1 << u8));
+            VMOV_64(v1, 0);
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            // should set C1 to 0
            break;
        case 0xFE:
            INST_NAME("FSIN");
            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            // seems that sin of glib doesn't follow the rounding direction mode
+            if(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            VMOV_64(0, v1);    // prepare call to sin
-            CALL_1D(sin, 0);
+            CALL_1D(sin, box86_dynarec_fastround ? 0 : (1 << u8));
            VMOV_64(v1, 0);
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            //emu->sw.f.F87_C2 = 0; C1 too
            LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
            BFC(x1, 9, 2); //C2 C1 = 0 0
@@ -533,9 +594,14 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xFF:
            INST_NAME("FCOS");
            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            // seems that cos of glib doesn't follow the rounding direction mode
+            if(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            VMOV_64(0, v1);    // prepare call to cos
-            CALL_1D(cos, 0);
+            CALL_1D(cos, box86_dynarec_fastround ? 0 : (1 << u8));
            VMOV_64(v1, 0);
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            //emu->sw.f.F87_C2 = 0; C1 too
            LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
            BFC(x1, 9, 2); //C2 C1 = 0 0
@@ -568,7 +634,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            switch((nextop>>3)&7) {
                case 0:
                    INST_NAME("FLD ST0, float[ED]");
-                    v1 = x87_do_push(dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+                    X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
                    if(ST_IS_F(0))
                        s0 = v1;
                    else
@@ -589,12 +655,16 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 2:
                    INST_NAME("FST float[ED], ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
                    if(ST_IS_F(0))
                        s0 = v1;
                    else {
                        s0 = fpu_get_scratch_single(dyn);
+                        if(!box86_dynarec_fastround)
+                            u8 = x87_setround(dyn, ninst, x1, x2, x14);
                        VCVT_F32_F64(s0, v1);
+                        if(!box86_dynarec_fastround)
+                            x87_restoreround(dyn, ninst, u8);
                    }
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
@@ -608,12 +678,16 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 3:
                    INST_NAME("FSTP float[ED], ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
                    if(ST_IS_F(0))
                        s0 = v1;
                    else {
                        s0 = fpu_get_scratch_single(dyn);
+                        if(!box86_dynarec_fastround)
+                            u8 = x87_setround(dyn, ninst, x1, x2, x14);
                        VCVT_F32_F64(s0, v1);
+                        if(!box86_dynarec_fastround)
+                            x87_restoreround(dyn, ninst, u8);
                    }
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
@@ -624,7 +698,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        VMOVfrV(x2, s0);
                        STR_IMM9(x2, ed, fixedaddress);
                    }
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                    break;
                case 4:
                    INST_NAME("FLDENV Ed");
@@ -33,6 +33,7 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
    int d0;
    int s0;
    int fixedaddress;
+    uint8_t u8;

    MAYUSE(s0);
    MAYUSE(d0);
@@ -128,8 +129,8 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                VCMP_F64(v1, v2);
            }
            FCOM(x1, x2);
-            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 0xE4:
@@ -155,7 +156,11 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    s0 = fpu_get_scratch_single(dyn);
                    VMOVtoV(s0, ed);
                    VCVT_F64_S32(d0, s0);
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    VADD_F64(v1, v1, d0);
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
                case 1:
                    INST_NAME("FIMUL ST0, Ed");
@@ -165,7 +170,11 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    s0 = fpu_get_scratch_single(dyn);
                    VMOVtoV(s0, ed);
                    VCVT_F64_S32(d0, s0);
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    VMUL_F64(v1, v1, d0);
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
                case 2:
                    INST_NAME("FICOM ST0, Ed");
@@ -188,7 +197,7 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    VCVT_F64_S32(d0, s0);
                    VCMP_F64(v1, d0);
                    FCOM(x1, x2);
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                    break;
                case 4:
                    INST_NAME("FISUB ST0, Ed");
@@ -198,7 +207,11 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    s0 = fpu_get_scratch_single(dyn);
                    VMOVtoV(s0, ed);
                    VCVT_F64_S32(d0, s0);
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    VSUB_F64(v1, v1, d0);
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
                case 5:
                    INST_NAME("FISUBR ST0, Ed");
@@ -208,7 +221,11 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    s0 = fpu_get_scratch_single(dyn);
                    VMOVtoV(s0, ed);
                    VCVT_F64_S32(d0, s0);
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    VSUB_F64(v1, d0, v1);
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
                case 6:
                    INST_NAME("FIDIV ST0, Ed");
@@ -218,7 +235,11 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    s0 = fpu_get_scratch_single(dyn);
                    VMOVtoV(s0, ed);
                    VCVT_F64_S32(d0, s0);
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    VDIV_F64(v1, v1, d0);
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
                case 7:
                    INST_NAME("FIDIVR ST0, Ed");
@@ -228,7 +249,11 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    s0 = fpu_get_scratch_single(dyn);
                    VMOVtoV(s0, ed);
                    VCVT_F64_S32(d0, s0);
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    VDIV_F64(v1, d0, v1);
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
            }
    }
@@ -141,7 +141,7 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xEE:
        case 0xEF:
            INST_NAME("FUCOMI ST0, STx");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_NODF);
            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)) {
@@ -160,7 +160,7 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xF6:
        case 0xF7:
            INST_NAME("FCOMI ST0, STx");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_NODF);
            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)) {
@@ -183,7 +183,7 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            switch((nextop>>3)&7) {
                case 0:
                    INST_NAME("FILD ST0, Ed");
-                    v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
+                    X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
                    s0 = fpu_get_scratch_single(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
@@ -218,12 +218,12 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    MOV_IMM_COND(cNE, ed, 0b10, 1);   // 0x80000000
                    WBACK;
                    VMSR(x14);  // put back values
-                    x87_do_pop(dyn, ninst, x3);
+                    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, x14); // x1 have the modified RPSCR reg
+                    u8 = x87_setround_reset(dyn, ninst, x1, x2, x14); // x1 have the modified RPSCR reg
                    if((nextop&0xC0)==0xC0) {
                        ed = xEAX+(nextop&7);
                        wback = 0;
@@ -233,10 +233,6 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    }
                    s0 = fpu_get_scratch_single(dyn);
                    MSR_nzcvq_0();
-                    // x1 already have FPCSR reg to clear exceptions flags
-                    ORR_IMM8(x3, x1, 0b010, 9); // enable exceptions
-                    BIC_IMM8(x3, x3, 0b10011111, 0);
-                    VMSR(x3);
                    VCVTR_S32_F64(s0, v1);
                    VMRS(x3);   // get the FPCSR reg and test FPU execption (invalid operation only)
                    VMOVfrV(ed, s0);
@@ -248,7 +244,7 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                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, x14); // x1 have the modified RPSCR reg
+                    u8 = x87_setround_reset(dyn, ninst, x1, x2, x14); // x1 have the modified RPSCR reg
                    if((nextop&0xC0)==0xC0) {
                        ed = xEAX+(nextop&7);
                        wback = 0;
@@ -258,17 +254,13 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    }
                    s0 = fpu_get_scratch_single(dyn);
                    MSR_nzcvq_0();
-                    // The FPCSR reg to clear exceptions flags
-                    ORR_IMM8(x3, x1, 0b010, 9); // enable exceptions
-                    BIC_IMM8(x3, x3, 0b10011111, 0);
-                    VMSR(x3);
                    VCVTR_S32_F64(s0, v1);
                    VMRS(x3);   // get the FPCSR reg and test FPU execption (denormal and overflow only)
                    VMOVfrV(ed, s0);
                    TSTS_IMM8_ROR(x3, 0b00000001, 0);
                    MOV_IMM_COND(cNE, ed, 0b10, 1);   // 0x80000000
                    WBACK;
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                    x87_restoreround(dyn, ninst, u8);
                    break;
                case 5:
@@ -276,28 +268,98 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                    if(PK(0)==0xDB && ((PK(1)>>3)&7)==7) {
                        // the FLD is immediatly followed by an FSTP
-                        LDR_IMM9(x2, ed, 0);
-                        LDR_IMM9(x3, ed, 4);
+                        LDM(ed, (1<<x2)|(1<<x3));
                        LDRH_IMM8(x14, ed, 8);
                        // no persistant scratch register, so unrool both instruction here...
                        MESSAGE(LOG_DUMP, "\tHack: FSTP tbyte\n");
                        nextop = F8;    //0xDB
                        nextop = F8;    //modrm
                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
-                        STR_IMM9(x2, ed, 0);
-                        STR_IMM9(x3, ed, 4);
+                        STM(ed, (1<<x2)|(1<<x3));
                        STRH_IMM8(x14, ed, 8);
+                    } else {
+                        if(box86_x87_no80bits) {
+                            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
+                            parity = getedparity(dyn, ninst, addr, nextop, 3);
+                            if (parity) {
+                                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
+                                VLDR_64(v1, ed, fixedaddress);
+                            } else {
+                                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
+                                LDR_IMM9(x2, ed, fixedaddress);
+                                LDR_IMM9(x3, ed, fixedaddress + 4);
+                                VMOVtoV_D(v1, x2, x3);
+                            }
+                        } else {
+                            #if 0
+                            MESSAGE(LOG_DUMP, "Need Optimization\n");
+                            if(ed!=x1) {
+                                MOV_REG(x1, ed);
+                            }
+X87_PUSH_OR_FAIL_empty(                   , dyn, ninst, x3);
+                            CALL(arm_fld, -1, 0);
+                            #else
+                            X87_PUSH_OR_FAIL(v1, dyn, ninst, x2, NEON_CACHE_ST_D);
+                            // copy 10bytes of *ED to STld(0)
+                            LDR_IMM9(x3, xEmu, offsetof(x86emu_t, top));
+                            int a = -dyn->n.x87stack;
+                            if(a) {
+                                if(a<0) {
+                                    SUB_IMM8(x3, x3, -a);
+                                } else {
+                                    ADD_IMM8(x3, x3, a);
+                                }
+                                AND_IMM8(x3, x3, 7);
+                            }
+                            ADD_IMM8(x2, xEmu, offsetof(x86emu_t, fpu_ld)+offsetof(fpu_ld_t, ld));
+                            MOVW(x14, sizeof(fpu_ld_t));
+                            MLA(x2, x3, x14, x2);
+                            LDR_IMM9(x14, ed, 0);
+                            STR_IMM9(x14, x2, 0);
+                            LDR_IMM9(x14, ed, 4);
+                            STR_IMM9(x14, x2, 4);
+                            LDRH_IMM8(x14, ed, 8);
+                            STRH_IMM8(x14, x2, 8);
+                            // convert 10bytes -> double
+                            CALL_1RD(FromLD, ed, x2, (1<<x3)|(1<<xFlags)|(1<<x2));
+                            // store ref to STld(0).uref, but can be unaligned...
+                            VMOVfrV_D(x3, x14, 0);
+                            // also, store result to ST0
+                            VMOVD(v1, 0);
+                            STR_IMM9(x3, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld));
+                            STR_IMM9(x14, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld)+4);
+                            #endif
+                        }
+                    }
+                    break;
+                case 7:
+                    INST_NAME("FSTP tbyte");
+                        if(box86_x87_no80bits) {
+                        v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                        parity = getedparity(dyn, ninst, addr, nextop, 3);
+                        if(parity) {
+                            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
+                            VSTR_64(v1, ed, fixedaddress);
+                        } else {
+                            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
+                            VMOVfrV_D(x2, x3, v1);
+                            STR_IMM9(x2, ed, fixedaddress);
+                            STR_IMM9(x3, ed, fixedaddress+4);
+                        }
                    } else {
                        #if 0
                        MESSAGE(LOG_DUMP, "Need Optimization\n");
+                        x87_forget(dyn, ninst, x1, x3, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                        if(ed!=x1) {
                            MOV_REG(x1, ed);
                        }
-                        x87_do_push_empty(dyn, ninst, x3);
-                        CALL(arm_fld, -1, 0);
+                        CALL(arm_fstp, -1, 0);
                        #else
-                        v1 = x87_do_push(dyn, ninst, x2, NEON_CACHE_ST_D);
-                        // copy 10bytes of *ED to STld(0)
+                        v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                        v2 = fpu_get_scratch_double(dyn);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
+                        // get top in x3
                        LDR_IMM9(x3, xEmu, offsetof(x86emu_t, top));
                        int a = -dyn->n.x87stack;
                        if(a) {
@@ -308,86 +370,42 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                            }
                            AND_IMM8(x3, x3, 7);
                        }
+                        // get STld(0)
                        ADD_IMM8(x2, xEmu, offsetof(x86emu_t, fpu_ld)+offsetof(fpu_ld_t, ld));
                        MOVW(x14, sizeof(fpu_ld_t));
                        MLA(x2, x3, x14, x2);
-                        LDR_IMM9(x14, ed, 0);
-                        STR_IMM9(x14, x2, 0);
-                        LDR_IMM9(x14, ed, 4);
-                        STR_IMM9(x14, x2, 4);
-                        LDRH_IMM8(x14, ed, 8);
-                        STRH_IMM8(x14, x2, 8);
-                        // convert 10bytes -> double
-                        CALL_1RD(FromLD, ed, x2, (1<<x3)|(1<<xFlags)|(1<<x2));
-                        // store ref to STld(0).uref, but can be unaligned...
-                        VMOVfrV_D(x3, x14, 0);
-                        // also, store result to ST0
-                        VMOVD(v1, 0);
-                        STR_IMM9(x3, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld));
-                        STR_IMM9(x14, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld)+4);
+                        // compare uref with actual value, top is not used anymore, so x3 is free
+                        LDR_IMM9(x14, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld));
+                        LDR_IMM9(x3, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld)+4);
+                        VMOVtoV_D(v2, x14, x3);
+                        VCEQ_32(v2, v2, v1);    // same?
+                        VMOVfrV_D(x14, x3, v2);
+                        TSTS_REG_LSL_IMM5(x14, x3, 0);
+                        B_MARK(cEQ);    // different, do the conversion
+                        // same... copy STld(0).ld
+                        LDR_IMM9(x14, x2, 0);
+                        LDR_IMM9(x3, x2, 4);
+                        STR_IMM9(x14, ed, 0);
+                        STR_IMM9(x3, ed, 4);
+                        LDRH_IMM8(x14, x2, 8);
+                        STRH_IMM8(x14, ed, 8);
+                        B_MARK2(c__);
+                        MARK;
+                        // different, call D2LD
+                        SUB_IMM8(xSP, xSP, 8);
+                        VSTR_64(v1, xSP, 0);
+                        PUSH(xSP, (1<<xEmu)|(1<<xFlags));
+                        ADD_IMM8(0, xSP, 8);    //1st arg is &d
+                        if(ed!=x1) {
+                            MOV_REG_LSL_IMM5(x1, ed, 0);    //2nd arg is ed
+                        }
+                        CALL_S(D2LD, -2, 0);
+                        POP(xSP, (1<<xEmu)|(1<<xFlags));
+                        ADD_IMM8(xSP, xSP, 8);
+                        MARK2;
                        #endif
                    }
-                    break;
-                case 7:
-                    INST_NAME("FSTP tbyte");
-                    #if 0
-                    MESSAGE(LOG_DUMP, "Need Optimization\n");
-                    x87_forget(dyn, ninst, x1, x3, 0);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
-                    if(ed!=x1) {
-                        MOV_REG(x1, ed);
-                    }
-                    CALL(arm_fstp, -1, 0);
-                    #else
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
-                    v2 = fpu_get_scratch_double(dyn);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
-                    // get top in x3
-                    LDR_IMM9(x3, xEmu, offsetof(x86emu_t, top));
-                    int a = -dyn->n.x87stack;
-                    if(a) {
-                        if(a<0) {
-                            SUB_IMM8(x3, x3, -a);
-                        } else {
-                            ADD_IMM8(x3, x3, a);
-                        }
-                        AND_IMM8(x3, x3, 7);
-                    }
-                    // get STld(0)
-                    ADD_IMM8(x2, xEmu, offsetof(x86emu_t, fpu_ld)+offsetof(fpu_ld_t, ld));
-                    MOVW(x14, sizeof(fpu_ld_t));
-                    MLA(x2, x3, x14, x2);
-                    // compare uref with actual value, top is not used anymore, so x3 is free
-                    LDR_IMM9(x14, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld));
-                    LDR_IMM9(x3, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld)+4);
-                    VMOVtoV_D(v2, x14, x3);
-                    VCEQ_32(v2, v2, v1);    // same?
-                    VMOVfrV_D(x14, x3, v2);
-                    TSTS_REG_LSL_IMM5(x14, x3, 0);
-                    B_MARK(cEQ);    // different, do the conversion
-                    // same... copy STld(0).ld
-                    LDR_IMM9(x14, x2, 0);
-                    LDR_IMM9(x3, x2, 4);
-                    STR_IMM9(x14, ed, 0);
-                    STR_IMM9(x3, ed, 4);
-                    LDRH_IMM8(x14, x2, 8);
-                    STRH_IMM8(x14, ed, 8);
-                    B_MARK2(c__);
-                    MARK;
-                    // different, call D2LD
-                    SUB_IMM8(xSP, xSP, 8);
-                    VSTR_64(v1, xSP, 0);
-                    PUSH(xSP, (1<<xEmu)|(1<<xFlags));
-                    ADD_IMM8(0, xSP, 8);    //1st arg is &d
-                    if(ed!=x1) {
-                        MOV_REG_LSL_IMM5(x1, ed, 0);    //2nd arg is ed
-                    }
-                    CALL_S(D2LD, -2, 0);
-                    POP(xSP, (1<<xEmu)|(1<<xFlags));
-                    ADD_IMM8(xSP, xSP, 8);
-                    MARK2;
-                    #endif
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                    break;
                default:
                    DEFAULT;
@@ -31,6 +31,7 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
    int d1;
    int fixedaddress;
    int parity;
+    uint8_t u8;

    MAYUSE(d1);
    MAYUSE(v2);
@@ -48,11 +49,15 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FADD STx, ST0");
            v2 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
            v1 = x87_get_st(dyn, ninst, x1, x2, nextop&7, X87_COMBINE(0, nextop&7));
+            if(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VADD_F32(v1, v1, v2);
            } else {
                VADD_F64(v1, v1, v2);
            }
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            break;
        case 0xC8:
        case 0xC9:
@@ -65,11 +70,15 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FMUL STx, ST0");
            v2 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
            v1 = x87_get_st(dyn, ninst, x1, x2, nextop&7, X87_COMBINE(0, nextop&7));
+            if(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VMUL_F32(v1, v1, v2);
            } else {
                VMUL_F64(v1, v1, v2);
            }
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            break;
        case 0xD0:
        case 0xD1:
@@ -106,7 +115,7 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                VCMP_F64(v1, v2);
            }
            FCOM(x1, x2);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xE0:
        case 0xE1:
@@ -119,11 +128,15 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FSUBR STx, ST0");
            v2 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
            v1 = x87_get_st(dyn, ninst, x1, x2, nextop&7, X87_COMBINE(0, nextop&7));
+            if(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VSUB_F32(v1, v2, v1);
            } else {
                VSUB_F64(v1, v2, v1);
            }
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            break;
        case 0xE8:
        case 0xE9:
@@ -136,11 +149,15 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FSUB STx, ST0");
            v2 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
            v1 = x87_get_st(dyn, ninst, x1, x2, nextop&7, X87_COMBINE(0, nextop&7));
+            if(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VSUB_F32(v1, v1, v2);
            } else {
                VSUB_F64(v1, v1, v2);
            }
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            break;
        case 0xF0:
        case 0xF1:
@@ -153,11 +170,15 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FDIVR STx, ST0");
            v2 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
            v1 = x87_get_st(dyn, ninst, x1, x2, nextop&7, X87_COMBINE(0, nextop&7));
+            if(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VDIV_F32(v1, v2, v1);
            } else {
                VDIV_F64(v1, v2, v1);
            }
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            break;       
        case 0xF8:
        case 0xF9:
@@ -170,11 +191,15 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FDIV STx, ST0");
            v2 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
            v1 = x87_get_st(dyn, ninst, x1, x2, nextop&7, X87_COMBINE(0, nextop&7));
+            if(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VDIV_F32(v1, v1, v2);
            } else {
                VDIV_F64(v1, v1, v2);
            }
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
            break;
        default:
            switch((nextop>>3)&7) {
@@ -192,7 +217,11 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        LDR_IMM9(x3, wback, fixedaddress+4);
                        VMOVtoV_D(d1, x2, x3);
                    }
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    VADD_F64(v1, v1, d1);
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
                case 1:
                    INST_NAME("FMUL ST0, double[ED]");
@@ -208,7 +237,11 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        LDR_IMM9(x3, wback, fixedaddress+4);
                        VMOVtoV_D(d1, x2, x3);
                    }
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    VMUL_F64(v1, v1, d1);
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
                case 2:
                    INST_NAME("FCOM ST0, double[ED]");
@@ -243,7 +276,7 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    }
                    VCMP_F64(v1, d1);
                    FCOM(x1, x2);
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                    break;
                case 4:
                    INST_NAME("FSUB ST0, double[ED]");
@@ -259,7 +292,11 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        LDR_IMM9(x3, wback, fixedaddress+4);
                        VMOVtoV_D(d1, x2, x3);
                    }
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    VSUB_F64(v1, v1, d1);
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
                case 5:
                    INST_NAME("FSUBR ST0, double[ED]");
@@ -275,7 +312,11 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        LDR_IMM9(x3, wback, fixedaddress+4);
                        VMOVtoV_D(d1, x2, x3);
                    }
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    VSUB_F64(v1, d1, v1);
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
                case 6:
                    INST_NAME("FDIV ST0, double[ED]");
@@ -291,7 +332,11 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        LDR_IMM9(x3, wback, fixedaddress+4);
                        VMOVtoV_D(d1, x2, x3);
                    }
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    VDIV_F64(v1, v1, d1);
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
                case 7:
                    INST_NAME("FDIVR ST0, double[ED]");
@@ -307,7 +352,11 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        LDR_IMM9(x3, wback, fixedaddress+4);
                        VMOVtoV_D(d1, x2, x3);
                    }
+                    if(!box86_dynarec_fastround)
+                        u8 = x87_setround(dyn, ninst, x1, x2, x14);
                    VDIV_F64(v1, d1, v1);
+                    if(!box86_dynarec_fastround)
+                        x87_restoreround(dyn, ninst, u8);
                    break;
            }
    }
@@ -32,6 +32,7 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
    uint8_t ed;
    int v1, v2;
    int i1, i2, i3;
+    int j32;

    MAYUSE(v2);
    MAYUSE(v1);
@@ -45,10 +46,21 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xC5:
        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_POP_OR_FAIL(dyn, ninst, x1);
+                addr+=2;
+                SKIPTEST(x1, x14);
+            } else
+                x87_free(dyn, ninst, x1, x2, x3, nextop&7);
+            #else
            INST_NAME("FFREE STx");
            x87_purgecache(dyn, ninst, 0, x1, x2, x3);
            MOVW(x1, nextop&7);
            CALL(fpu_do_free, -1, 0);
+            #endif
            break;
        case 0xD0:
        case 0xD1:
@@ -69,7 +81,7 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            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:
@@ -81,7 +93,7 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FSTP ST0, STx");
            // copy the cache value for st0 to stx
            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:
@@ -119,7 +131,7 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                VCMP_F64(v1, v2);
            }
            FCOM(x1, x2);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;

        case 0xC8:
@@ -153,7 +165,7 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            switch((nextop>>3)&7) {
                case 0:
                    INST_NAME("FLD double");
-                    v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
+                    X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
 		            parity = getedparity(dyn, ninst, addr, nextop, 3);
 		            if (parity) {
                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
@@ -167,11 +179,54 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 1:
                    INST_NAME("FISTTP i64, ST0");
-                    x87_forget(dyn, ninst, x1, x2, 0);
+                    parity = getedparity(dyn, ninst, addr, nextop, 3);
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
-                    if(ed!=x1) {MOV_REG(x1, ed);}
-                    CALL(arm_fistt64, -1, 0);
-                    x87_do_pop(dyn, ninst, x3);
+                    if(ninst
+                      && dyn->insts[ninst-1].x86.addr
+                      && *(uint8_t*)dyn->insts[ninst-1].x86.addr==0xDF
+                      && (((*(uint8_t*)(dyn->insts[ninst-1].x86.addr+1))>>3)&7)==5)
+                    {
+                        if(parity) {
+                            STRD_IMM8(x2, ed, 0);    // x2/x3 is 64bits
+                        } else {
+                            STR_IMM9(x2, ed, 0);
+                            STR_IMM9(x3, ed, 4);
+                        }
+                    } else {
+                        v1 = x87_get_st(dyn, ninst, x2, x3, 0, NEON_CACHE_ST_D);
+                        //addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
+                        fpu_get_scratch_double(dyn); // to alocate v0
+                        v2 = fpu_get_scratch_double(dyn);
+                        //  get TOP
+                        LDR_IMM9(x14, xEmu, offsetof(x86emu_t, top));
+                        int a = 0 - dyn->n.x87stack;
+                        if(a<0) {
+                            SUB_IMM8(x14, x14, -a);
+                            AND_IMM8(x14, x14, 7);    // (emu->top + i)&7
+                        } else if(a>0) {
+                            ADD_IMM8(x14, x14, a);
+                            AND_IMM8(x14, x14, 7);    // (emu->top + i)&7
+                        }
+                        ADD_REG_LSL_IMM5(x14, xEmu, x14, 4);  // each fpu_ll is 2 int64: ref than ll
+                        MOVW(x2, offsetof(x86emu_t, fpu_ll));   //can be optimized?
+                        ADD_REG_LSL_IMM5(x14, x14, x2, offsetof(fpu_ll_t, sref));
+                        VLDR_64(v2, x14, 0);
+                        VCEQ_32(v2, v2, v1);    // compare
+                        VMOVfrV_D(x2, x3, v2);
+                        ANDS_REG_LSL_IMM5(x2, x2, x3, 0);   // if NE then values are the same!
+                        B_MARK(cEQ);    // do the i64 conversion
+                        // memcpy(ed, &STll(0).ll, sizeof(int64_t));
+                        LDRD_IMM8(x2, x14, offsetof(fpu_ll_t, sq));  // load ll
+                        B_MARK3(c__);
+                        MARK;
+                        MOV32(x2, arm_fist64_3);
+                        VMOV_64(0, v1);    // prepare call to log2
+                        CALL_1DR_U64(x2, x2, x3, x14, (1<<x1));
+                        MARK3;
+                        STR_IMM9(x2, ed, 0);
+                        STR_IMM9(x3, ed, 4);
+                    }
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                    break;
                case 2:
                    INST_NAME("FST double");
@@ -200,7 +255,7 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        STR_IMM9(x2, ed, fixedaddress);
                        STR_IMM9(x3, ed, fixedaddress+4);
                    }
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                    break;
                case 4: 
                    INST_NAME("FRSTOR m108byte");
@@ -215,6 +270,7 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                    if(ed!=x1) {MOV_REG(x1, ed);}
                    CALL(arm_fsave, -1, 0);
+                    CALL(reset_fpu, -1, 0);
                    break;
                case 7:
                    INST_NAME("FNSTSW m2byte");
@@ -7,6 +7,7 @@
 #include "debug.h"
 #include "box86context.h"
 #include "dynarec.h"
+#include "dynarec/arm_emitter.h"
 #include "emu/x86emu_private.h"
 #include "emu/x86run_private.h"
 #include "x86run.h"
@@ -27,6 +28,7 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
 {
    uint8_t nextop = F8;
    int v1, v2;
+    uint8_t u8;

    MAYUSE(v2);
    MAYUSE(v1);
@@ -43,12 +45,16 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FADDP 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(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VADD_F32(v2, v2, v1);
            } else {
                VADD_F64(v2, v2, v1);
            }
-            x87_do_pop(dyn, ninst, x3);
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xC8:
        case 0xC9:
@@ -61,12 +67,16 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FMULP 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(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VMUL_F32(v2, v2, v1);
            } else {
                VMUL_F64(v2, v2, v1);
            }
-            x87_do_pop(dyn, ninst, x3);
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xD0:
        case 0xD1:
@@ -85,7 +95,7 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                VCMP_F64(v1, v2);
            }
            FCOM(x1, x2);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;

        case 0xD9: 
@@ -98,8 +108,8 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                VCMP_F64(v1, v2);
            }
            FCOM(x1, x2);
-            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:
@@ -112,12 +122,16 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FSUBRP 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(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VSUB_F32(v2, v1, v2);
            } else {
                VSUB_F64(v2, v1, v2);
            }
-            x87_do_pop(dyn, ninst, x3);
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xE8:
        case 0xE9:
@@ -130,12 +144,16 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FSUBP 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(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VSUB_F32(v2, v2, v1);
            } else {
                VSUB_F64(v2, v2, v1);
            }
-            x87_do_pop(dyn, ninst, x3);
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xF0:
        case 0xF1:
@@ -148,12 +166,16 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FDIVRP 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(!box86_dynarec_fastround)
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
            if(ST_IS_F(0)) {
                VDIV_F32(v2, v1, v2);
            } else {
                VDIV_F64(v2, v1, v2);
            }
-            x87_do_pop(dyn, ninst, x3);
+            if(!box86_dynarec_fastround)
+                x87_restoreround(dyn, ninst, u8);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xF8:
        case 0xF9:
@@ -166,11 +188,21 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("FDIVP 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(!box86_dynarec_fastnan) {
+            if(!box86_dynarec_fastround || !box86_dynarec_fastnan) {
                VMRS(x14);               // get fpscr
-                ORR_IMM8(x3, x14, 0b010, 9); // enable exceptions
-                BIC_IMM8(x3, x3, 0b10011111, 0);
-                VMSR(x3);
+                if(!box86_dynarec_fastnan) {
+                    ORR_IMM8(x3, x14, 0b010, 9); // enable exceptions
+                    BIC_IMM8(x3, x3, 0b10011111, 0);
+                } else if(!box86_dynarec_fastround)
+                    MOV_REG(x3, x14);
+                if(!box86_dynarec_fastround){
+                    LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, cw));    // hopefully cw is not too far for an imm8
+                    UBFX(x1, x1, 10, 2);    // extract round...
+                    UBFX(x2, x1, 1, 1);     // swap bits 0 and 1
+                    BFI(x2, x1, 1, 1);
+                    BFI(x3, x2, 22, 2);     // inject new round
+                }
+                VMSR(x3);               // put new fpscr
            }
            if(ST_IS_F(0)) {
                VDIV_F32(v2, v2, v1);
@@ -185,9 +217,10 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                } else {
                    VNEG_F64_cond(cNE, v2, v2);
                }
-                VMSR(x14);  // restore fpscr
            }
-            x87_do_pop(dyn, ninst, x3);
+            if(!box86_dynarec_fastround || !box86_dynarec_fastnan)
+                VMSR(x14);  // restore fpscr
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;

        case 0xD8:
@@ -53,7 +53,7 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xC7:
            INST_NAME("FFREEP STx");
            // not handling Tag...
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;

        case 0xE0:
@@ -81,7 +81,7 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xEE:
        case 0xEF:
            INST_NAME("FUCOMIP ST0, STx");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_NODF);
            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))
@@ -91,7 +91,7 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                VCMP_F64(v1, v2);
            }
            FCOMI(x1, x2);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;
        case 0xF0:
        case 0xF1:
@@ -102,7 +102,7 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xF6:
        case 0xF7:
            INST_NAME("FCOMIP ST0, STx");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_NODF);
            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))
@@ -112,7 +112,7 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                VCMP_F64(v1, v2);
            }
            FCOMI(x1, x2);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
            break;

        case 0xC8:
@@ -161,7 +161,7 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            switch((nextop>>3)&7) {
                case 0:
                    INST_NAME("FILD ST0, Ew");
-                    v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+                    X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL);
                    LDRSH_IMM8(x1, wback, fixedaddress);
                    if(ST_IS_F(0)) {
@@ -175,7 +175,7 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 1:
                    INST_NAME("FISTTP Ew, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0, 0, NULL);
                    ed = x1;
                    s0 = fpu_get_scratch_single(dyn);
@@ -186,9 +186,9 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    BIC_IMM8(x3, x3, 0b10011111, 0);
                    VMSR(x3);
                    if(ST_IS_F(0)) {
-                        VCVTR_S32_F32(s0, v1);
+                        VCVT_S32_F32(s0, v1);
                    } else {
-                        VCVTR_S32_F64(s0, v1);
+                        VCVT_S32_F64(s0, v1);
                    }
                    VMRS(x3);   // get the FPCSR reg and test FPU execption (invalid operation only)
                    VMOVfrV(ed, s0);
@@ -197,21 +197,17 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    CMPS_REG_LSL_IMM5_COND(cEQ, ed, x3, 0);
                    MOVW_COND(cNE, x3, 0x8000); // saturated
                    STRH_IMM8(x3, wback, fixedaddress);
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                    VMSR(x14);
                    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, x14);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+                    u8 = x87_setround_reset(dyn, ninst, x1, x2, x14);
                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0, 0, NULL);
                    ed = x1;
                    s0 = fpu_get_scratch_single(dyn);
                    MSR_nzcvq_0();
-                    // x1 already have FPCSR reg to clear exceptions flags
-                    ORR_IMM8(x3, x1, 0b010, 9); // enable exceptions
-                    BIC_IMM8(x3, x3, 0b10011111, 0);
-                    VMSR(x3);
                    if(ST_IS_F(0)) {
                        VCVTR_S32_F32(s0, v1);
                    } else {
@@ -228,16 +224,12 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    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, x14);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+                    u8 = x87_setround_reset(dyn, ninst, x1, x2, x14);
                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0, 0, NULL);
                    ed = x1;
                    s0 = fpu_get_scratch_single(dyn);
                    MSR_nzcvq_0();
-                    // x1 already have FPCSR reg to clear exceptions flags
-                    ORR_IMM8(x3, x1, 0b010, 9); // enable exceptions
-                    BIC_IMM8(x3, x3, 0b10011111, 0);
-                    VMSR(x3);
                    if(ST_IS_F(0)) {
                        VCVTR_S32_F32(s0, v1);
                    } else {
@@ -250,20 +242,22 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    CMPS_REG_LSL_IMM5_COND(cEQ, ed, x3, 0);
                    MOVW_COND(cNE, x3, 0x8000); // saturated
                    STRH_IMM8(x3, wback, fixedaddress);
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                    x87_restoreround(dyn, ninst, u8);
                    break;
                case 4:
                    INST_NAME("FBLD ST0, tbytes");
                    MESSAGE(LOG_DUMP, "Need Optimization\n");
-                    x87_do_push_empty(dyn, ninst, x1);
+                    X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, x1);
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                    if(ed!=x1) {MOV_REG(x1, ed);}
+                    s0 = x87_stackcount(dyn, ninst, x3);
                    CALL(fpu_fbld, -1, 0);
+                    x87_unstackcount(dyn, ninst, x3, s0);
                    break;
                case 5:
                    INST_NAME("FILD ST0, i64");
-                    v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
+                    X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
                    v2 = fpu_get_scratch_double(dyn);
                    s0 = fpu_get_scratch_single(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 3);
@@ -322,8 +316,10 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    x87_forget(dyn, ninst, x1, x2, 0);
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                    if(ed!=x1) {MOV_REG(x1, ed);}
+                    s0 = x87_stackcount(dyn, ninst, x3);
                    CALL(fpu_fbst, -1, 0);
-                    x87_do_pop(dyn, ninst, x3);
+                    x87_unstackcount(dyn, ninst, x3, s0);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                    break;
                case 7: // could be inlined for most thing, but is it usefull?
                    INST_NAME("FISTP i64, ST0");
@@ -341,12 +337,10 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                            STR_IMM9(x3, ed, 4);
                        }
                    } else {
-                        #if 1
                        v1 = x87_get_st(dyn, ninst, x2, x3, 0, NEON_CACHE_ST_D);
                        //addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                        fpu_get_scratch_double(dyn); // to alocate v0
                        v2 = fpu_get_scratch_double(dyn);
-                        #if 1
                        //  get TOP
                        LDR_IMM9(x14, xEmu, offsetof(x86emu_t, top));
                        int a = 0 - dyn->n.x87stack;
@@ -379,76 +373,8 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        MARK3;
                        STR_IMM9(x2, ed, 0);
                        STR_IMM9(x3, ed, 4);
-                        #else
-                        v0 = fpu_get_scratch_double(dyn);
-                        s0 = fpu_get_scratch_single(dyn);
-                        // check STll(0).ref==ST(0).q so emu->fpu_ll[emu->top].ref == emu->mmx87[emu->top]
-                        //  get TOP
-                        LDR_IMM9(x14, xEmu, offsetof(x86emu_t, top));
-                        int a = 0 - dyn->n.x87stack;
-                        if(a<0) {
-                            SUB_IMM8(x14, x14, -a);
-                            AND_IMM8(x14, x14, 7);    // (emu->top + i)&7
-                        } else if(a>0) {
-                            ADD_IMM8(x14, x14, a);
-                            AND_IMM8(x14, x14, 7);    // (emu->top + i)&7
-                        }
-                        ADD_REG_LSL_IMM5(x14, xEmu, x14, 4);  // each fpu_ll is 2 int64: ref than ll
-                        MOVW(x2, offsetof(x86emu_t, fpu_ll));   //can be optimized?
-                        ADD_REG_LSL_IMM5(x14, x14, x2, 0);
-                        VLDR_64(v2, x14, 0);
-                        VCEQ_32(v2, v2, v1);    // compare
-                        VMOVfrV_D(x2, x3, v2);
-                        ANDS_REG_LSL_IMM5(x2, x2, x3, 0);   // if NE then values are the same!
-                        B_MARK(cEQ);    // do the i64 conversion
-                        // memcpy(ed, &STll(0).ll, sizeof(int64_t));
-                        LDRD_IMM8(x2, x14, 8);  // load ll
-                        B_MARK3(c__);
-                        MARK;
-                        VEOR(v0, v0, v0);
-                        MOVW(x2, 0x41F0);
-                        VMOVtoDx_16(v0, 3, x2); // V0 = (1<<32) as double
-                        VMOVfrDx_32(x14, v1, 1);    // get high part to extract sign
-                        VABS_F64(v1 ,v1);    //ST0 will be poped, so lost...
-                        VDIV_F64(v2, v1, v0);   // v2 = abs(ST0)/(1<<32) : so 32 bits high part
-                        MSR_nzcvq_0();
-                        VMRS(x2);               // get fpscr
-                        ORR_IMM8(x3, x2, 0b010, 9); // enable exceptions
-                        BIC_IMM8(x3, x3, 0b10011111, 0);
-                        VMSR(x3);
-                        VCVT_U32_F64(s0, v2);   // convert high part to U32
-                        VMRS(x3);   // get the FPCSR reg and test FPU execption (invalid operation only)
-                        VMSR(x2);               // put back fpscr
-                        TSTS_IMM8_ROR(x3, 0b00000001, 0);
-                        B_MARK2(cEQ);   // not overflow...
-                        MARKLOCK;
-                        MOV_IMM(x3, 0b10, 1);   // 0x80000000
-                        MOVW(x2, 0);
-                        B_MARK3(c__);
-                        MARK2;  // continue conversion, it fits an int64!
-                        VCVT_F64_U32(v2, s0);   // int part now
-                        VMLS_F64(v1, v2, v0);   // compute low part
-                        VMOVfrV(x3, s0);        // transfert high path
-                        TSTS_IMM8_ROR(x14, 0b10, 1);    // test high part with 0x800000000
-                        B_MARKLOCK(cNE);        // int overflow...
-                        VCVT_U32_F64(s0, v1);   // convert low part
-                        VMOVfrV(x2, s0);        // transfert low part
-                        TSTS_IMM8_ROR(x14, 0b10, 1);    // 0x800000000
-                        B_MARK3(cEQ);
-                        RSBS_IMM8(x2, x2, 0);   // NEG(i64)
-                        RSC_IMM8(x3, x3, 0);
-                        MARK3;
-                        STR_IMM9(x2, x1, 0);
-                        STR_IMM9(x3, x1, 4);
-                        #endif
-                        #else
-                        MESSAGE(LOG_DUMP, "Need Optimization\n");
-                        x87_forget(dyn, ninst, x2, x3, 0);
-                        if(ed!=x1) {MOV_REG(x1, ed);}
-                        CALL(arm_fistp64, -1, 0);
-                        #endif
                    }
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                    break;
                default:
                    DEFAULT;
@@ -53,7 +53,7 @@ void emit_or32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -96,7 +96,7 @@ void emit_or32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -130,7 +130,7 @@ void emit_xor32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -173,7 +173,7 @@ void emit_xor32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -207,7 +207,7 @@ void emit_and32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -258,7 +258,7 @@ void emit_and32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -292,7 +292,7 @@ void emit_or8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -326,7 +326,7 @@ void emit_or8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -360,7 +360,7 @@ void emit_xor8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -394,7 +394,7 @@ void emit_xor8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -428,7 +428,7 @@ void emit_and8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -462,7 +462,7 @@ void emit_and8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -497,7 +497,7 @@ void emit_or16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -540,7 +540,7 @@ void emit_or16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -574,7 +574,7 @@ void emit_xor16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -617,7 +617,7 @@ void emit_xor16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -651,7 +651,7 @@ void emit_and16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -694,6 +694,6 @@ void emit_and16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }
@@ -69,7 +69,7 @@ void emit_add32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -151,7 +151,7 @@ void emit_add32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -203,7 +203,7 @@ void emit_sub32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -288,7 +288,7 @@ void emit_sub32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -338,7 +338,7 @@ void emit_add8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, in
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
    IFX(X_AF|X_OF|X_PF){if(save_s4) {POP(xSP, 1<<s4);}}
 }
@@ -390,7 +390,7 @@ void emit_add8c(dynarec_arm_t* dyn, int ninst, int s1, int c, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -442,7 +442,7 @@ void emit_sub8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, in
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
    IFX(X_AF|X_OF|X_CF|X_PF){if(save_s4) {POP(xSP, 1<<s4);}}
 }
@@ -495,7 +495,7 @@ void emit_sub8c(dynarec_arm_t* dyn, int ninst, int s1, int c, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -547,7 +547,7 @@ void emit_add16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, i
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
    IFX(X_AF|X_OF|X_PF){if(save_s4) {POP(xSP, 1<<s4);}}
 }
@@ -615,7 +615,7 @@ void emit_add16c(dynarec_arm_t* dyn, int ninst, int s1, int c, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -668,7 +668,7 @@ void emit_sub16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, i
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
    IFX(X_AF|X_OF|X_CF|X_PF){if(save_s4) {POP(xSP, 1<<s4);}}
 }
@@ -734,7 +734,7 @@ void emit_sub16c(dynarec_arm_t* dyn, int ninst, int s1, int c, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -778,7 +778,7 @@ void emit_inc32(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -823,7 +823,7 @@ void emit_inc8(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -870,7 +870,7 @@ void emit_inc16(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -915,7 +915,7 @@ void emit_dec32(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -923,7 +923,7 @@ void emit_dec32(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
 void emit_dec8(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
 {
    IFX(X_PEND) {
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
        SET_DF(s3, d_dec8);
    } else IFX(X_ZF|X_OF|X_AF|X_SF|X_PF) {
        SET_DFNONE(s3);
@@ -960,7 +960,7 @@ void emit_dec8(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -1006,7 +1006,7 @@ void emit_dec16(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -1057,7 +1057,7 @@ void emit_adc32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -1127,7 +1127,7 @@ void emit_adc32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -1179,7 +1179,7 @@ void emit_adc8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, in
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
    IFX(X_AF|X_OF|X_PF){if(save_s4) {POP(xSP, 1<<s4);}}
 }
@@ -1232,7 +1232,7 @@ void emit_adc8c(dynarec_arm_t* dyn, int ninst, int s1, int c, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -1285,7 +1285,7 @@ void emit_adc16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, i
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
    IFX(X_AF|X_OF|X_PF){if(save_s4) {POP(xSP, 1<<s4);}}
 }
@@ -1353,7 +1353,7 @@ void emit_adc16c(dynarec_arm_t* dyn, int ninst, int s1, int c, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -1407,7 +1407,7 @@ void emit_sbb32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -1479,7 +1479,7 @@ void emit_sbb32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -1533,7 +1533,7 @@ void emit_sbb8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, in
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
    IFX(X_AF|X_OF|X_CF|X_PF){if(save_s4) {POP(xSP, 1<<s4);}}
 }
@@ -1587,7 +1587,7 @@ void emit_sbb8c(dynarec_arm_t* dyn, int ninst, int s1, int c, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -1642,7 +1642,7 @@ void emit_sbb16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, i
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
    IFX(X_AF|X_OF|X_CF|X_PF){if(save_s4) {POP(xSP, 1<<s4);}}
 }
@@ -1710,7 +1710,7 @@ void emit_sbb16c(dynarec_arm_t* dyn, int ninst, int s1, int c, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -1758,7 +1758,7 @@ void emit_neg32(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -1807,7 +1807,7 @@ void emit_neg16(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -1855,6 +1855,6 @@ void emit_neg8(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }
@@ -76,7 +76,7 @@ void emit_shl32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
            BFC_COND(cNE, xFlags, F_OF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -130,7 +130,7 @@ void emit_shl32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        }
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -154,7 +154,17 @@ void emit_shr32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
            }
            B_NEXT(cEQ);
    }
-    IFX(X_ZF|X_CF) {
+    IFX(X_CF) {
+        SUB_IMM8(s4, s2, 1);
+        MOV_REG_LSR_REG(s4, s1, s4);
+        BFI(xFlags, s4, F_CF, 1);
+    }
+    IFX(X_OF) {
+        CMPS_IMM8(s2, 1);   // if s2==1
+            MOV_REG_LSR_IMM5_COND(cEQ, s4, s1, 31);
+            BFI_COND(cEQ, xFlags, s4, F_OF, 1);
+    }
+    IFX(X_ZF) {
        MOVS_REG_LSR_REG(s1, s1, s2);
    } else {
        MOV_REG_LSR_REG(s1, s1, s2);
@@ -162,27 +172,16 @@ void emit_shr32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
    IFX(X_PEND) {
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
    }
-    IFX(X_ZF|X_CF) {
-        BIC_IMM8(xFlags, xFlags, (1<<F_ZF)|(1<<F_CF), 0);
-    }
    IFX(X_ZF) {
+        BIC_IMM8(xFlags, xFlags, (1<<F_ZF), 0);
        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
    }
-    IFX(X_CF) {
-        ORR_IMM8_COND(cCS, xFlags, xFlags, 1<<F_CF, 0);
-    }
    IFX(X_SF) {
        MOV_REG_LSR_IMM5(s4, s1, 31);
        BFI(xFlags, s4, F_SF, 1);
    }
-    IFX(X_OF) {
-        CMPS_IMM8(s2, 1);
-            MOV_REG_LSR_IMM5_COND(cEQ, s4, s1, 30);
-            XOR_REG_LSR_IMM8_COND(cEQ, s4, s4, s4, 1);
-            BFI_COND(cEQ, xFlags, s4, F_OF, 1);
-    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -203,7 +202,21 @@ void emit_shr32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        }
        return;
    }
-    IFX(X_ZF|X_CF) {
+    IFX(X_CF) {
+        if(c==1) {
+            BFI(xFlags, s1, F_CF, 1);
+        } else {
+            MOV_REG_LSR_IMM5(s3, s1, c-1);
+            BFI(xFlags, s3, F_CF, 1);
+        }
+    }
+    IFX(X_OF) {
+        if(c==1) {
+            MOV_REG_LSR_IMM5(s4, s1, 31);
+            BFI(xFlags, s4, F_OF, 1);
+        }
+    }
+    IFX(X_ZF) {
        MOVS_REG_LSR_IMM5(s1, s1, c);
    } else {
        MOV_REG_LSR_IMM5(s1, s1, c);
@@ -211,28 +224,16 @@ void emit_shr32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
    IFX(X_PEND) {
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
    }
-    IFX(X_ZF|X_CF) {
-        BIC_IMM8(xFlags, xFlags, (1<<F_ZF)|(1<<F_CF), 0);
-    }
    IFX(X_ZF) {
+        BIC_IMM8(xFlags, xFlags, (1<<F_ZF), 0);
        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
    }
-    IFX(X_CF) {
-        ORR_IMM8_COND(cCS, xFlags, xFlags, 1<<F_CF, 0);
-    }
    IFX(X_SF) {
        MOV_REG_LSR_IMM5(s3, s1, 31);
        BFI(xFlags, s3, F_SF, 1);
    }
-    IFX(X_OF) {
-        if(c==1) {
-            MOV_REG_LSR_IMM5(s4, s1, 30);
-            XOR_REG_LSR_IMM8(s4, s4, s4, 1);
-            BFI(xFlags, s4, F_OF, 1);
-        }
-    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -274,8 +275,567 @@ void emit_sar32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        MOV_REG_LSR_IMM5(s3, s1, 31);
        BFI(xFlags, s3, F_SF, 1);
    }
+    IFX(X_OF)
+        if(c==1) {
+            BFC(xFlags, F_OF, 1);
+    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
+    }
+}
+
+// emit SHL8 instruction, from s1 , shift s2, store result in s1 using s3 and s4 as scratch. s3 can be same as s2
+void emit_shl8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
+{
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_shl8);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    IFX(X_CF | X_OF) {
+        RSB_IMM8(s4, s2, 8);
+        MOV_REG_LSR_REG(s4, s1, s4);
+        BFI(xFlags, s4, F_CF, 1);
+    }
+    MOV_REG_LSL_REG(s1, s1, s2);
+    IFX(X_ZF) {
+        ANDS_IMM8(s1, s1, 0xff);
+    } else IFX(X_PEND) {
+        AND_IMM8(s1, s1, 0xff);
+    }
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_ZF) {
+        MOV32(s4, 1);
+        BFI_COND(cEQ, xFlags, s4, F_ZF, 1);
+    }
+    IFX(X_SF) {
+        MOV_REG_LSR_IMM5(s4, s1, 7);
+        BFI(xFlags, s4, F_SF, 1);
+    }
+    IFX(X_OF) {
+        CMPS_IMM8(s2, 1);   // if s2==1
+            IFX(X_SF) {} else {MOV_REG_LSR_IMM5_COND(cEQ, s4, s1, 7);}
+            XOR_REG_LSL_IMM5_COND(cEQ, s4, s4, xFlags, 0);  // CF is set if OF is asked
+            BFI_COND(cEQ, xFlags, s4, F_OF, 1);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s4);
+    }
+}
+
+// emit SHL8 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_shl8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
+{
+    if(!c)
+        return;
+    IFX(X_PEND) {
+        MOV32(s3, c);
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_shl8);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    if(c<8) {
+        IFX(X_CF|X_OF) {
+            MOV_REG_LSR_IMM5(s3, s1, 8-c);
+            BFI(xFlags, s3, F_CF, 1);
+        }
+        MOV_REG_LSL_IMM5(s1, s1, c);
+
+        IFX(X_PEND) {
+            AND_IMM8(s1, s1, 0xff);
+            STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+        }
+        IFX(X_ZF) {
+            TSTS_IMM8(s1, 0xff);
+            MOVW_COND(cNE, s4, 0);
+            MOVW_COND(cEQ, s4, 1);
+            BFI(xFlags, s4, F_ZF, 1);
+        }
+        IFX(X_SF) {
+            MOV_REG_LSR_IMM5(s4, s1, 7);
+            BFI(xFlags, s4, F_SF, 1);
+        }
+        IFX(X_OF) {
+            if(c==1) {
+                IFX(X_SF) {} else {MOV_REG_LSR_IMM5(s4, s1, 7);}
+                XOR_REG_LSL_IMM5(s4, s4, xFlags, 0);  // CF is set if OF is asked
+                BFI(xFlags, s4, F_OF, 1);
+            } else {
+                BFC(xFlags, F_OF, 1);
+            }
+        }
+        IFX(X_PF) {
+            emit_pf(dyn, ninst, s1, s4);
+        }
+    } else {
+        IFX(X_CF) {
+            MOV_REG_LSL_IMM5(s3, s1, c-1);
+            MOV_REG_LSR_IMM5(s4, s3, 7);
+            BFI(xFlags, s4, F_CF, 1);
+        }
+        MOVW(s1, 0);
+        IFX(X_OF) {
+            BFC(xFlags, F_OF, 1);
+        }
+        IFX(X_SF) {
+            BFC(xFlags, F_SF, 1);
+        }
+        IFX(X_PF | X_ZF) {
+            MOVW(s3, 1);
+            IFX(X_ZF) {
+                BFI(xFlags, s3, F_ZF, 1);
+            }
+            IFX(X_PF) {
+                BFI(xFlags, s3, F_PF, 1);
+            }
+        }
+    }
+}
+
+// emit SHR8 instruction, from s1 , s2, store result in s1 using s3 and s4 as scratch, s2 can be same as s3
+void emit_shr8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
+{
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_shr8);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    IFX(X_CF) {
+        SUB_IMM8(s4, s2, 1);
+        MOV_REG_LSR_REG(s4, s1, s4);
+        BFI(xFlags, s4, 0, 1);
+    }
+    IFX(X_OF) {
+        CMPS_IMM8(s2, 1);   // if s2==1
+            MOV_REG_LSR_IMM5_COND(cEQ, s4, s1, 7);
+            BFI_COND(cEQ, xFlags, s4, F_OF, 1);
+    }
+    MOV_REG_LSR_REG(s1, s1, s2);
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_ZF) {
+        TSTS_IMM8(s1, 0xff);
+        MOVW_COND(cNE, s4, 0);
+        MOVW_COND(cEQ, s4, 1);
+        BFI(xFlags, s4, F_ZF, 1);
+    }
+    IFX(X_SF) {
+        MOV_REG_LSR_IMM5(s4, s1, 7);
+        BFI(xFlags, s4, F_SF, 1);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s4);
+    }
+}
+
+// emit SHR8 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_shr8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
+{
+    if(!c)
+        return;
+    IFX(X_PEND) {
+        MOV32(s3, c);
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_shr8);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    IFX(X_CF) {
+        if(c==1) {
+            BFI(xFlags, s1, 0, 1);
+        } else {
+            MOV_REG_LSR_IMM5(s3, s1, c-1);
+            BFI(xFlags, s3, 0, 1);
+        }
+    }
+    IFX(X_OF) {
+        if(c==1) {
+            MOV_REG_LSR_IMM5(s4, s1, 7);
+            BFI(xFlags, s4, F_OF, 1);
+        }
+    }
+    MOV_REG_LSR_IMM5(s1, s1, c);
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_ZF) {
+        TSTS_IMM8(s1, 0xff);
+        MOVW_COND(cNE, s4, 0);
+        MOVW_COND(cEQ, s4, 1);
+        BFI(xFlags, s4, F_ZF, 1);
+    }
+    IFX(X_SF) {
+        MOV_REG_LSR_IMM5(s4, s1, 7);
+        BFI(xFlags, s4, F_SF, 1);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s4);
+    }
+}
+
+// emit SAR8 instruction, from s1 , shift s2, store result in s1 using s3 and s4 as scratch, s2 can be same as s3
+void emit_sar8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
+{
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_sar8);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    IFX(X_CF) {
+        SUB_IMM8(s4, s2, 1);
+        MOV_REG_ASR_REG(s4, s1, s4);
+        BFI(xFlags, s4, 0, 1);
+    }
+    MOV_REG_ASR_REG(s1, s1, s2);
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_ZF) {
+        TSTS_IMM8(s1, 0xff);
+        MOVW_COND(cNE, s4, 0);
+        MOVW_COND(cEQ, s4, 1);
+        BFI(xFlags, s4, F_ZF, 1);
+    }
+    IFX(X_SF) {
+        MOV_REG_LSR_IMM5(s4, s1, 7);
+        BFI(xFlags, s4, F_SF, 1);
+    }
+    IFX(X_OF) {
+        CMPS_IMM8(s2, 1);
+            BFC_COND(cEQ, xFlags, F_OF, 1);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s4);
+    }
+}
+
+// emit SAR8 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_sar8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
+{
+    if(!c)
+        return;
+    IFX(X_PEND) {
+        MOV32(s3, c);
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_sar8);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    IFX(X_CF) {
+        MOV_REG_ASR_IMM5(s3, s1, c-1);
+        BFI(xFlags, s3, 0, 1);
+    }
+    MOV_REG_ASR_IMM5(s1, s1, c);
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_ZF) {
+        TSTS_IMM8(s1, 0xff);
+        MOVW_COND(cNE, s4, 0);
+        MOVW_COND(cEQ, s4, 1);
+        BFI(xFlags, s4, F_ZF, 1);
+    }
+    IFX(X_SF) {
+        MOV_REG_LSR_IMM5(s4, s1, 7);
+        BFI(xFlags, s4, F_SF, 1);
+    }
+    IFX(X_OF)
+        if(c==1) {
+            BFC(xFlags, F_OF, 1);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s4);
+    }
+}
+
+// emit SHL16 instruction, from s1 , shift s2, store result in s1 using s3 and s4 as scratch. s3 can be same as s2
+void emit_shl16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
+{
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_shl16);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    IFX(X_CF | X_OF) {
+        RSB_IMM8(s4, s2, 16);
+        MOV_REG_LSR_REG(s4, s1, s4);
+        BFI(xFlags, s4, F_CF, 1);
+    }
+    MOV_REG_LSL_REG(s1, s1, s2);
+    UBFX(s1, s1, 0, 16);
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_ZF) {
+        TSTS_REG_LSL_IMM5(s1, s1, 0);
+        MOV32(s4, 1);
+        BFI_COND(cEQ, xFlags, s4, F_ZF, 1);
+    }
+    IFX(X_SF) {
+        MOV_REG_LSR_IMM5(s4, s1, 15);
+        BFI(xFlags, s4, F_SF, 1);
+    }
+    IFX(X_OF) {
+        CMPS_IMM8(s2, 1);   // if s2==1
+            IFX(X_SF) {} else {MOV_REG_LSR_IMM5_COND(cEQ, s4, s1, 15);}
+            XOR_REG_LSL_IMM5_COND(cEQ, s4, s4, xFlags, 0);  // CF is set if OF is asked
+            BFI_COND(cEQ, xFlags, s4, F_OF, 1);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s4);
+    }
+}
+
+// emit SHL16 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_shl16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
+{
+    if(!c)
+        return;
+    IFX(X_PEND) {
+        MOV32(s3, c);
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_shl16);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    if(c<16) {
+        IFX(X_CF|X_OF) {
+            MOV_REG_LSR_IMM5(s3, s1, 16-c);
+            BFI(xFlags, s3, F_CF, 1);
+        }
+        MOV_REG_LSL_IMM5(s1, s1, c);
+
+        UBFX(s1, s1, 0, 16);
+        IFX(X_PEND) {
+            STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+        }
+        IFX(X_ZF) {
+            TSTS_REG_LSL_IMM5(s1, s1, 0);
+            MOVW_COND(cNE, s4, 0);
+            MOVW_COND(cEQ, s4, 1);
+            BFI(xFlags, s4, F_ZF, 1);
+        }
+        IFX(X_SF) {
+            MOV_REG_LSR_IMM5(s4, s1, 15);
+            BFI(xFlags, s4, F_SF, 1);
+        }
+        IFX(X_OF) {
+            if(c==1) {
+                IFX(X_SF) {} else {MOV_REG_LSR_IMM5(s4, s1, 15);}
+                XOR_REG_LSL_IMM5(s4, s4, xFlags, 0);  // CF is set if OF is asked
+                BFI(xFlags, s4, F_OF, 1);
+            } else {
+                BFC(xFlags, F_OF, 1);
+            }
+        }
+        IFX(X_PF) {
+            emit_pf(dyn, ninst, s1, s4);
+        }
+    } else {
+        IFX(X_CF) {
+            MOV_REG_LSL_IMM5(s3, s1, c-1);
+            MOV_REG_LSR_IMM5(s4, s3, 15);
+            BFI(xFlags, s4, F_CF, 1);
+        }
+        MOVW(s1, 0);
+        IFX(X_OF) {
+            BFC(xFlags, F_OF, 1);
+        }
+        IFX(X_SF) {
+            BFC(xFlags, F_SF, 1);
+        }
+        IFX(X_PF | X_ZF) {
+            MOVW(s3, 1);
+            IFX(X_ZF) {
+                BFI(xFlags, s3, F_ZF, 1);
+            }
+            IFX(X_PF) {
+                BFI(xFlags, s3, F_PF, 1);
+            }
+        }
+    }
+}
+
+// emit SHR16 instruction, from s1 , s2, store result in s1 using s3 and s4 as scratch, s2 can be same as s3
+void emit_shr16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
+{
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_shr16);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    IFX(X_CF) {
+        SUB_IMM8(s4, s2, 1);
+        MOV_REG_LSR_REG(s4, s1, s4);
+        BFI(xFlags, s4, 0, 1);
+    }
+    IFX(X_OF) {
+        CMPS_IMM8(s2, 1);   // if s2==1
+            MOV_REG_LSR_IMM5_COND(cEQ, s4, s1, 16);
+            BFI_COND(cEQ, xFlags, s4, F_OF, 1);
+    }
+    MOV_REG_LSR_REG(s1, s1, s2);
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_ZF) {
+        TSTS_REG_LSL_IMM5(s1, s1, 0);
+        MOVW_COND(cNE, s4, 0);
+        MOVW_COND(cEQ, s4, 1);
+        BFI(xFlags, s4, F_ZF, 1);
+    }
+    IFX(X_SF) {
+        MOV_REG_LSR_IMM5(s4, s1, 15);
+        BFI(xFlags, s4, F_SF, 1);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s4);
+    }
+}
+
+// emit SHR16 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_shr16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
+{
+    if(!c)
+        return;
+    IFX(X_PEND) {
+        MOV32(s3, c);
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_shr16);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    IFX(X_CF) {
+        if(c==1) {
+            BFI(xFlags, s1, 0, 1);
+        } else {
+            MOV_REG_LSR_IMM5(s3, s1, c-1);
+            BFI(xFlags, s3, 0, 1);
+        }
+    }
+    IFX(X_OF) {
+        if(c==1) {
+            MOV_REG_LSR_IMM5(s4, s1, 15);
+            BFI(xFlags, s4, F_OF, 1);
+        }
+    }
+    MOV_REG_LSR_IMM5(s1, s1, c);
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_ZF) {
+        TSTS_REG_LSL_IMM5(s1, s1, 0);
+        MOVW_COND(cNE, s4, 0);
+        MOVW_COND(cEQ, s4, 1);
+        BFI(xFlags, s4, F_ZF, 1);
+    }
+    IFX(X_SF) {
+        MOV_REG_LSR_IMM5(s4, s1, 15);
+        BFI(xFlags, s4, F_SF, 1);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s4);
+    }
+}
+
+// emit SAR16 instruction, from s1 , shift s2, store result in s1 using s3 and s4 as scratch, s2 can be same as s3
+void emit_sar16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
+{
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_sar16);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    IFX(X_CF) {
+        SUB_IMM8(s4, s2, 1);
+        MOV_REG_ASR_REG(s4, s1, s4);
+        BFI(xFlags, s4, 0, 1);
+    }
+    MOV_REG_ASR_REG(s1, s1, s2);
+    UBFX(s1, s1, 0, 16);
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_ZF) {
+        TSTS_REG_LSL_IMM5(s1, s1, 0);
+        MOVW_COND(cNE, s4, 0);
+        MOVW_COND(cEQ, s4, 1);
+        BFI(xFlags, s4, F_ZF, 1);
+    }
+    IFX(X_SF) {
+        MOV_REG_LSR_IMM5(s4, s1, 15);
+        BFI(xFlags, s4, F_SF, 1);
+    }
+    IFX(X_OF) {
+        CMPS_IMM8(s2, 1);
+            BFC_COND(cEQ, xFlags, F_OF, 1);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s4);
+    }
+}
+
+// emit SAR16 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_sar16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
+{
+    if(!c)
+        return;
+    IFX(X_PEND) {
+        MOV32(s3, c);
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_sar16);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    IFX(X_CF) {
+        MOV_REG_ASR_IMM5(s3, s1, c-1);
+        BFI(xFlags, s3, 0, 1);
+    }
+    MOV_REG_ASR_IMM5(s1, s1, c);
+    IFX(X_PEND | X_ZF) {
+        UXTH(s1, s1, 0);
+    }
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_ZF) {
+        TSTS_REG_LSL_IMM5(s1, s1, 0);
+        MOVW_COND(cNE, s4, 0);
+        MOVW_COND(cEQ, s4, 1);
+        BFI(xFlags, s4, F_ZF, 1);
+    }
+    IFX(X_SF) {
+        MOV_REG_LSR_IMM5(s4, s1, 15);
+        BFI(xFlags, s4, F_SF, 1);
+    }
+    IFX(X_OF)
+        if(c==1) {
+            BFC(xFlags, F_OF, 1);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -366,6 +926,12 @@ void emit_shrd32c(dynarec_arm_t* dyn, int ninst, int s1, int s2, int32_t c, int
        }
        return;
    }
+    IFX(X_OF) {
+        if(c==1) {
+            MOV_REG_LSR_IMM5(s4, s1, 31);
+            BFI(xFlags, s4, F_OF, 1);   // store sign for later use
+        }
+    }
    IFX(X_CF) {
        MOVS_REG_LSR_IMM5(s1, s1, c);
    } else {
@@ -394,13 +960,15 @@ void emit_shrd32c(dynarec_arm_t* dyn, int ninst, int s1, int s2, int32_t c, int
    }
    IFX(X_OF) {
        if(c==1) {
-            MOV_REG_LSR_IMM5(s4, s1, 30);
-            XOR_REG_LSR_IMM8(s4, s4, s4, 1);
+            MOV_REG_LSR_IMM5(s4, s1, 31);
+            XOR_REG_LSR_IMM8(s4, s4, xFlags, F_OF); // set if sign changed
            BFI(xFlags, s4, F_OF, 1);
+        } else {
+            BFC(xFlags, F_OF, 1);
        }
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -425,6 +993,12 @@ void emit_shld32c(dynarec_arm_t* dyn, int ninst, int s1, int s2, int32_t c, int
        }
        return;
    }
+    IFX(X_OF) {
+        if(c==1) {
+            MOV_REG_LSR_IMM5(s4, s1, 31);
+            BFI(xFlags, s4, F_OF, 1);   // store sign for later use
+        }
+    }
    IFX(X_CF) {
        MOV_REG_LSR_IMM5(s3, s1, 32-c);
        BFI(xFlags, s3, F_CF, 1);
@@ -436,9 +1010,9 @@ void emit_shld32c(dynarec_arm_t* dyn, int ninst, int s1, int s2, int32_t c, int
    }
    IFX(X_OF) {
        if(c==1) {
-            UBFX(s3, s1, (31-c), 1);
-            XOR_IMM8_COND(cCS, s3, s3, 1);
-            BFI(xFlags, s3, F_OF, 1);
+            MOV_REG_LSR_IMM5(s4, s1, 31);
+            XOR_REG_LSR_IMM8(s4, s4, xFlags, F_OF); // set if sign changed
+            BFI(xFlags, s4, F_OF, 1);
        } else {
            BFC(xFlags, F_OF, 1);
        }
@@ -461,7 +1035,7 @@ void emit_shld32c(dynarec_arm_t* dyn, int ninst, int s1, int s2, int32_t c, int
        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -474,7 +1048,7 @@ void emit_shrd32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
        // same flags calc as shr32
-        SET_DF(s4, d_shr32);
+        SET_DF(s4, d_shrd32);
    } else IFX(X_ALL) {
        SET_DFNONE(s4);
    }
@@ -484,6 +1058,10 @@ void emit_shrd32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        }
        B_NEXT(cEQ);

+    IFX(X_OF) {
+        MOV_REG_LSR_IMM5(s4, s1, 31);
+        BFI(xFlags, s4, F_OF, 1);   // store sign for later use
+    }
    IFX(X_CF) {
        MOVS_REG_LSR_REG(s1, s1, s3);
    } else {
@@ -513,12 +1091,12 @@ void emit_shrd32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
    }
    IFX(X_OF) {
        CMPS_IMM8(s3, 31);  // 32-c
-            MOV_REG_LSR_IMM5_COND(cEQ, s4, s1, 30);
-            XOR_REG_LSR_IMM8_COND(cEQ, s4, s4, s4, 1);
+            MOV_REG_LSR_IMM5_COND(cEQ, s4, s1, 31);
+            XOR_REG_LSR_IMM8_COND(cEQ, s4, s4, xFlags, F_OF);
            BFI_COND(cEQ, xFlags, s4, F_OF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -531,7 +1109,7 @@ void emit_shld32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
        // same flags calc as shl32
-        SET_DF(s4, d_shl32);
+        SET_DF(s4, d_shld32);
    } else IFX(X_ALL) {
        SET_DFNONE(s4);
    }
@@ -544,6 +1122,10 @@ void emit_shld32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        }
        B_NEXT(cEQ);

+    IFX(X_OF) {
+        MOV_REG_LSR_IMM5(s4, s1, 31);
+        BFI(xFlags, s4, F_OF, 1);   // store sign for later use
+    }
    IFX(X_CF) {
        RSB_IMM8(s4, s3, 32);
        MOV_REG_LSR_REG(s4, s1, s4);
@@ -555,13 +1137,10 @@ void emit_shld32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        MOV_REG_LSL_REG(s1, s1, s3);
    }
    IFX(X_OF) {
-        MOVW_COND(cCS, s4, 1);
-        MOVW_COND(cCC, s4, 0);
-        CMPS_IMM8(s3, 1);
-            ORR_REG_LSR_IMM5_COND(cEQ, s4, s4, s1, 30);
+        CMPS_IMM8(s3, 31);  // 32-c
+            MOV_REG_LSR_IMM5_COND(cEQ, s4, s1, 31);
+            XOR_REG_LSR_IMM8_COND(cEQ, s4, s4, xFlags, F_OF);
            BFI_COND(cEQ, xFlags, s4, F_OF, 1);
-        //else
-            BFC_COND(cNE, xFlags, F_OF, 1);
    }
    RSB_IMM8(s3, s3, 32);
    IFX(X_ZF) {
@@ -581,7 +1160,7 @@ void emit_shld32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BFI(xFlags, s4, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -602,8 +1181,8 @@ void emit_rol8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4
        return;
    }
    if(c&7) {
-        BFI(s1, s1, 24, 8);  // duplicate the value for rotate left
-        MOV_REG_ROR_IMM5(s1, s1, 32-(c&7));
+        BFI(s1, s1, 8, 8);
+        MOV_REG_LSR_IMM5(s1, s1, 8-(c&7));
    }
    //AND_IMM8(s1, s1, 0xff);
    IFX(X_PEND) {
@@ -614,7 +1193,7 @@ void emit_rol8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4
    }
    IFX(X_OF) {
        if(c==1) {
-            ADD_REG_LSR_IMM5(s3, s1, s1, 7);
+            XOR_REG_LSR_IMM8(s3, s1, s1, 7);
            BFI(xFlags, s3, F_OF, 1);
        }
    }
@@ -638,7 +1217,7 @@ void emit_ror8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4
    }
    if(c&7) {
        BFI(s1, s1, 8, 8);  // duplicate the value for Rotate right
-        MOV_REG_ROR_IMM5(s1, s1, c&7);
+        MOV_REG_LSR_IMM5(s1, s1, c&7);
    }
    IFX(X_PEND) {
        AND_IMM8(s1, s1, 0xff);
@@ -656,3 +1235,75 @@ void emit_ror8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4
        }
    }
 }
+
+// emit ROL16 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_rol16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
+{
+    IFX(X_PEND) {
+        MOVW(s3, c);
+        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_rol16);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    if(!c) {
+        IFX(X_PEND) {
+            STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+        }
+        return;
+    }
+    if(c&15) {
+        BFI(s1, s1, 16, 16);
+        MOV_REG_LSR_IMM5(s1, s1, 16-(c&15));
+    }
+    IFX(X_PEND) {
+        UBFX(s1, s1, 0, 16);
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_CF) {
+        BFI(xFlags, s1, F_CF, 1);
+    }
+    IFX(X_OF) {
+        if(c==1) {
+            XOR_REG_LSR_IMM8(s3, s1, s1, 15);
+            BFI(xFlags, s3, F_OF, 1);
+        }
+    }
+}
+
+// emit ROR16 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_ror16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
+{
+    IFX(X_PEND) {
+        MOVW(s3, c);
+        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_ror16);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    if(!c) {
+        IFX(X_PEND) {
+            STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+        }
+        return;
+    }
+    if(c&15) {
+        BFI(s1, s1, 16, 16);  // duplicate the value for Rotate right
+        MOV_REG_LSR_IMM5(s1, s1, c&15);
+    }
+    IFX(X_PEND) {
+        UBFX(s1, s1, 0, 16);
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_CF) {
+        MOV_REG_LSR_IMM5(s3, s1, 15);
+        BFI(xFlags, s3, F_CF, 1);
+    }
+    IFX(X_OF) {
+        if(c==1) {
+            MOV_REG_LSR_IMM5(s4, s1, 14);
+            XOR_REG_LSR_IMM8(s4, s4, s4, 1);
+            BFI(xFlags, s4, F_OF, 1);
+        }
+    }
+}
@@ -55,7 +55,10 @@ void emit_cmp32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        MOV_REG_LSR_IMM5(s4, s3, 31);
        BFI(xFlags, s4, F_SF, 1);
    }
-    // and now the tricky ones (and mostly unused), PF and AF
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s3, s4);
+    }
+    // and now the tricky ones (and mostly unused), AF
    IFX(X_AF) {
        // bc = (res & (~d | s)) | (~d & s)
        MVN_REG_LSL_IMM5(s4, s1, 0);        // s4 = ~d
@@ -68,19 +71,6 @@ void emit_cmp32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
            BFI(xFlags, s4, F_AF, 1);    // AF: bc & 0x08
        }
    }
-    IFX(X_PF) {
-        // PF: (((emu->x86emu_parity_tab[(res) / 32] >> ((res) % 32)) & 1) == 0)
-        IFX(X_CF|X_AF) {
-            SUB_REG_LSL_IMM5(s3, s1, s2, 0);
-        }
-        AND_IMM8(s3, s3, 0xE0); // lsr 5 masking pre-applied
-        MOV32(s4, GetParityTab());
-        LDR_REG_LSR_IMM5(s4, s4, s3, 5-2);   // x/32 and then *4 because array is integer
-        SUB_REG_LSL_IMM5(s3, s1, s2, 0);
-        AND_IMM8(s3, s3, 31);
-        MVN_REG_LSR_REG(s4, s4, s3);
-        BFI(xFlags, s4, F_PF, 1);
-    }
 }

 // emit CMP32 instruction, from cmp s1 , 0, using s3 and s4 as scratch
@@ -112,7 +102,7 @@ void emit_cmp32_0(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
        BFI(xFlags, s4, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -140,6 +130,9 @@ void emit_cmp16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        MOV_REG_LSR_IMM5(s4, s3, 15);
        BFI(xFlags, s4, F_SF, 1);
    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s3, s4);
+    }
    // bc = (res & (~d | s)) | (~d & s)
    IFX(X_CF|X_AF|X_OF) {
        MVN_REG_LSL_IMM5(s4, s1, 0);        // s4 = ~d
@@ -161,18 +154,6 @@ void emit_cmp16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
            BFI(xFlags, s4, F_OF, 1);    // OF: ((bc >> 14) ^ ((bc>>14)>>1)) & 1
        }
    }
-    IFX(X_PF) {
-        IFX(X_CF|X_AF|X_OF) {
-            SUB_REG_LSL_IMM5(s3, s1, s2, 0);
-        }
-        AND_IMM8(s3, s3, 0xE0); // lsr 5 masking pre-applied
-        MOV32(s4, GetParityTab());
-        LDR_REG_LSR_IMM5(s4, s4, s3, 5-2);   // x/32 and then *4 because array is integer
-        SUB_REG_LSL_IMM5(s3, s1, s2, 0);
-        AND_IMM8(s3, s3, 31);
-        MVN_REG_LSR_REG(s4, s4, s3);
-        BFI(xFlags, s4, F_PF, 1);
-    }
 }

 // emit CMP16 instruction, from cmp s1 , #0, using s3 and s4 as scratch
@@ -204,7 +185,7 @@ void emit_cmp16_0(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
        BFI(xFlags, s4, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }
 // emit CMP8 instruction, from cmp s1 , s2, using s3 and s4 as scratch
@@ -230,6 +211,9 @@ void emit_cmp8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        MOV_REG_LSR_IMM5(s4, s3, 7);
        BFI(xFlags, s4, F_SF, 1);
    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s3, s4);
+    }
    // bc = (res & (~d | s)) | (~d & s)
    IFX(X_CF|X_AF|X_OF) {
        MVN_REG_LSL_IMM5(s4, s1, 0);        // s4 = ~d
@@ -251,18 +235,6 @@ void emit_cmp8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
            BFI(xFlags, s4, F_OF, 1);    // OF: ((bc >> 6) ^ ((bc>>6)>>1)) & 1
        }
    }
-    IFX(X_PF) {
-        IFX(X_CF|X_AF|X_OF) {
-            SUB_REG_LSL_IMM5(s3, s1, s2, 0);
-        }
-        AND_IMM8(s3, s3, 0xE0); // lsr 5 masking pre-applied
-        MOV32(s4, GetParityTab());
-        LDR_REG_LSR_IMM5(s4, s4, s3, 5-2);   // x/32 and then *4 because array is integer
-        SUB_REG_LSL_IMM5(s3, s1, s2, 0);
-        AND_IMM8(s3, s3, 31);
-        MVN_REG_LSR_REG(s4, s4, s3);
-        BFI(xFlags, s4, F_PF, 1);
-    }
 }
 // emit CMP8 instruction, from cmp s1 , 0, using s3 and s4 as scratch
 void emit_cmp8_0(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
@@ -292,7 +264,7 @@ void emit_cmp8_0(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
        BFI(xFlags, s4, F_SF, 1);
    }
    IFX(X_PF) {
-        emit_pf(dyn, ninst, s1, s3, s4);
+        emit_pf(dyn, ninst, s1, s4);
    }
 }

@@ -321,19 +293,8 @@ void emit_test32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        MOV_REG_LSR_IMM5(s4, s3, 31);
        BFI(xFlags, s4, F_SF, 1);
    }
-    // PF: (((emu->x86emu_parity_tab[(res) / 32] >> ((res) % 32)) & 1) == 0)
    IFX(X_PF) {
-        AND_IMM8(s3, s3, 0xE0); // lsr 5 masking pre-applied
-        MOV32(s4, GetParityTab());
-        LDR_REG_LSR_IMM5(s4, s4, s3, 5-2);   // x/32 and then *4 because array is integer
-        if(s1==s2) {
-            AND_IMM8(s3, s1, 31);
-        } else {
-            AND_REG_LSL_IMM5(s3, s1, s2, 0);
-            AND_IMM8(s3, s3, 31);
-        }
-        MVN_REG_LSR_REG(s4, s4, s3);
-        BFI(xFlags, s4, F_PF, 1);
+        emit_pf(dyn, ninst, s3, s4);
    }
 }

@@ -362,15 +323,8 @@ void emit_test16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        MOV_REG_LSR_IMM5(s4, s3, 15);
        BFI(xFlags, s4, F_SF, 1);
    }
-    // PF: (((emu->x86emu_parity_tab[(res) / 32] >> ((res) % 32)) & 1) == 0)
    IFX(X_PF) {
-        AND_IMM8(s3, s3, 0xE0); // lsr 5 masking pre-applied
-        MOV32(s4, GetParityTab());
-        LDR_REG_LSR_IMM5(s4, s4, s3, 5-2);   // x/32 and then *4 because array is integer
-        AND_REG_LSL_IMM5(s3, s1, s2, 0);
-        AND_IMM8(s3, s3, 31);
-        MVN_REG_LSR_REG(s4, s4, s3);
-        BFI(xFlags, s4, F_PF, 1);
+        emit_pf(dyn, ninst, s3, s4);
    }
 }

@@ -399,13 +353,7 @@ void emit_test8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        MOV_REG_LSR_IMM5(s4, s3, 7);
        BFI(xFlags, s4, F_SF, 1);
    }
-    // PF: (((emu->x86emu_parity_tab[(res) / 32] >> ((res) % 32)) & 1) == 0)
    IFX(X_PF) {
-        AND_IMM8(s2, s3, 0xE0); // lsr 5 masking pre-applied
-        MOV32(s4, GetParityTab());
-        LDR_REG_LSR_IMM5(s4, s4, s2, 5-2);   // x/32 and then *4 because array is integer
-        AND_IMM8(s3, s3, 31);
-        MVN_REG_LSR_REG(s4, s4, s3);
-        BFI(xFlags, s4, F_PF, 1);
+        emit_pf(dyn, ninst, s3, s4);
    }
 }
@@ -42,8 +42,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGB(x2);
-            DMB_ISH();
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                wback = (nextop&7);    
                wb2 = (wback>>2);      
                wback = xEAX+(wback&3);
@@ -59,15 +58,14 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                CMPS_IMM8(x14, 0);
                B_MARKLOCK(cNE);    // write failed, try again
            }
-            DMB_ISH();
+            SMDMB();
            break;
        case 0x01:
            INST_NAME("LOCK ADD Ed, Gd");
            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGD;
-            DMB_ISH();
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                ed = xEAX+(nextop&7);
                emit_add32(dyn, ninst, ed, gd, x3, x14);
            } else {
@@ -79,14 +77,13 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                CMPS_IMM8(x3, 0);
                B_MARKLOCK(cNE);
            }
-            DMB_ISH();
+            SMDMB();
            break;
        case 0x08:
            INST_NAME("LOCK OR Eb, Gb");
            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
-            DMB_ISH();
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                GETGB(x2);
                wback = (nextop&7);    
                wb2 = (wback>>2);      
@@ -104,15 +101,14 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                CMPS_IMM8(x14, 0);
                B_MARKLOCK(cNE);    // write failed, try again
            }
-            DMB_ISH();
+            SMDMB();
            break;
        case 0x09:
            INST_NAME("LOCK OR Ed, Gd");
            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGD;
-            DMB_ISH();
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                ed = xEAX+(nextop&7);
                emit_or32(dyn, ninst, ed, gd, x3, x14);
            } else {
@@ -124,7 +120,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                CMPS_IMM8(x3, 0);
                B_MARKLOCK(cNE);
            }
-            DMB_ISH();
+            SMDMB();
            break;
        case 0x10:
            INST_NAME("LOCK ADC Eb, Gb");
@@ -132,8 +128,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGB(x2);
-            DMB_ISH();
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                wback = (nextop&7);    
                wb2 = (wback>>2);      
                wback = xEAX+(wback&3);
@@ -149,7 +144,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                CMPS_IMM8(x14, 0);
                B_MARKLOCK(cNE);    // write failed, try again
            }
-            DMB_ISH();
+            SMDMB();
            break;
        case 0x11:
            INST_NAME("LOCK ADC Ed, Gd");
@@ -157,8 +152,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGD;
-            DMB_ISH();
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                ed = xEAX+(nextop&7);
                emit_adc32(dyn, ninst, ed, gd, x3, x14);
            } else {
@@ -170,7 +164,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                CMPS_IMM8(x3, 0);
                B_MARKLOCK(cNE);
            }
-            DMB_ISH();
+            SMDMB();
            break;
        case 0x18:
            INST_NAME("LOCK SBB Eb, Gb");
@@ -178,8 +172,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGB(x2);
-            DMB_ISH();
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                wback = (nextop&7);    
                wb2 = (wback>>2);      
                wback = xEAX+(wback&3);
@@ -195,7 +188,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                CMPS_IMM8(x14, 0);
                B_MARKLOCK(cNE);    // write failed, try again
            }
-            DMB_ISH();
+            SMDMB();
            break;
        case 0x19:
            INST_NAME("LOCK SBB Ed, Gd");
@@ -203,8 +196,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGD;
-            DMB_ISH();
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                ed = xEAX+(nextop&7);
                emit_sbb32(dyn, ninst, ed, gd, x3, x14);
            } else {
@@ -216,14 +208,13 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                CMPS_IMM8(x3, 0);
                B_MARKLOCK(cNE);
            }
-            DMB_ISH();
+            SMDMB();
            break;
        case 0x20:
            INST_NAME("LOCK AND Eb, Gb");
            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
-            DMB_ISH();
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                GETGB(x2);
                wback = (nextop&7);    
                wb2 = (wback>>2);      
@@ -241,15 +232,14 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                CMPS_IMM8(x14, 0);
                B_MARKLOCK(cNE);    // write failed, try again
            }
-            DMB_ISH();
+            SMDMB();
            break;
        case 0x21:
            INST_NAME("LOCK AND Ed, Gd");
            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGD;
-            DMB_ISH();
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                ed = xEAX+(nextop&7);
                emit_and32(dyn, ninst, ed, gd, x3, x14);
            } else {
@@ -261,15 +251,14 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                CMPS_IMM8(x3, 0);
                B_MARKLOCK(cNE);
            }
-            DMB_ISH();
+            SMDMB();
            break;
        case 0x28:
            INST_NAME("LOCK SUB Eb, Gb");
            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGB(x2);
-            DMB_ISH();
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                wback = (nextop&7);    
                wb2 = (wback>>2);      
                wback = xEAX+(wback&3);
@@ -285,15 +274,14 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                CMPS_IMM8(x14, 0);
                B_MARKLOCK(cNE);    // write failed, try again
            }
-            DMB_ISH();
+            SMDMB();
            break;
        case 0x29:
            INST_NAME("LOCK SUB Ed, Gd");
            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGD;
-            DMB_ISH();
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                ed = xEAX+(nextop&7);
                emit_sub32(dyn, ninst, ed, gd, x3, x14);
            } else {
@@ -305,14 +293,13 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                CMPS_IMM8(x3, 0);
                B_MARKLOCK(cNE);
            }
-            DMB_ISH();
+            SMDMB();
            break;
        case 0x38:
            INST_NAME("LOCK XOR Eb, Gb");
            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
-            DMB_ISH();
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                GETGB(x2);
                wback = (nextop&7);    
                wb2 = (wback>>2);      
@@ -330,15 +317,14 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                CMPS_IMM8(x14, 0);
                B_MARKLOCK(cNE);    // write failed, try again
            }
-            DMB_ISH();
+            SMDMB();
            break;
        case 0x39:
            INST_NAME("LOCK XOR Ed, Gd");
            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGD;
-            DMB_ISH();
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                ed = xEAX+(nextop&7);
                emit_xor32(dyn, ninst, ed, gd, x3, x14);
            } else {
@@ -350,7 +336,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                CMPS_IMM8(x3, 0);
                B_MARKLOCK(cNE);
            }
-            DMB_ISH();
+            SMDMB();
            break;
        
        case 0x66:
@@ -368,8 +354,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        INST_NAME("LOCK ADD Ed, Ib");
                    }
                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    DMB_ISH();
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        if(opcode==0x81) i32 = F32S; else i32 = F8S;
                        ed = xEAX+(nextop&7);
                        emit_add32c(dyn, ninst, ed, i32, x3, x14);
@@ -389,7 +374,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                            B_MARKLOCK(cNE);
                        }
                        if(!fixedaddress) {
-                            DMB_ISH();
+                            SMDMB();
                            B_NEXT(c__);
                        }
                        if(!fixedaddress || (fixedaddress && (fixedaddress&3))) {
@@ -403,14 +388,13 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                            B_MARK(cNE);
                            STR_IMM9(x1, wback, 0);    // put the whole value
                        }
-                        DMB_ISH();
+                        SMDMB();
                    }
                    break;
                case 1: //OR
                    if(opcode==0x81) {INST_NAME("LOCK OR Ed, Id");} else {INST_NAME("LOCK OR Ed, Ib");}
                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    DMB_ISH();
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        if(opcode==0x81) i32 = F32S; else i32 = F8S;
                        ed = xEAX+(nextop&7);
                        emit_or32c(dyn, ninst, ed, i32, x3, x14);
@@ -424,14 +408,13 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        CMPS_IMM8(x3, 0);
                        B_MARKLOCK(cNE);
                    }
-                    DMB_ISH();
+                    SMDMB();
                    break;
                case 2: //ADC
                    if(opcode==0x81) {INST_NAME("LOCK ADC Ed, Id");} else {INST_NAME("LOCK ADC Ed, Ib");}
                    READFLAGS(X_CF);
                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    DMB_ISH();
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        if(opcode==0x81) i32 = F32S; else i32 = F8S;
                        ed = xEAX+(nextop&7);
                        emit_adc32c(dyn, ninst, ed, i32, x3, x14);
@@ -445,14 +428,13 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        CMPS_IMM8(x3, 0);
                        B_MARKLOCK(cNE);
                    }
-                    DMB_ISH();
+                    SMDMB();
                    break;
                case 3: //SBB
                    if(opcode==0x81) {INST_NAME("LOCK SBB Ed, Id");} else {INST_NAME("LOCK SBB Ed, Ib");}
                    READFLAGS(X_CF);
                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    DMB_ISH();
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        if(opcode==0x81) i32 = F32S; else i32 = F8S;
                        ed = xEAX+(nextop&7);
                        emit_sbb32c(dyn, ninst, ed, i32, x3, x14);
@@ -466,13 +448,12 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        CMPS_IMM8(x3, 0);
                        B_MARKLOCK(cNE);
                    }
-                    DMB_ISH();
+                    SMDMB();
                    break;
                case 4: //AND
                    if(opcode==0x81) {INST_NAME("LOCK AND Ed, Id");} else {INST_NAME("LOCK AND Ed, Ib");}
                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    DMB_ISH();
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        if(opcode==0x81) i32 = F32S; else i32 = F8S;
                        ed = xEAX+(nextop&7);
                        emit_and32c(dyn, ninst, ed, i32, x3, x14);
@@ -486,17 +467,16 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        CMPS_IMM8(x3, 0);
                        B_MARKLOCK(cNE);
                    }
-                    DMB_ISH();
+                    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((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        if(opcode==0x81) i32 = F32S; else i32 = F8S;
                        ed = xEAX+(nextop&7);
                        emit_sub32c(dyn, ninst, ed, i32, x3, x14);
                    } else {
-                        DMB_ISH();
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 1, LOCK_LOCK);
                        if(opcode==0x81) i32 = F32S; else i32 = F8S;
                        if(!fixedaddress) {
@@ -512,7 +492,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                            B_MARKLOCK(cNE);
                        }
                        if(!fixedaddress) {
-                            DMB_ISH();
+                            SMDMB();
                            B_NEXT(c__);
                        }
                        if(!fixedaddress || (fixedaddress && (fixedaddress&3))) {
@@ -526,18 +506,17 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                            B_MARK(cNE);
                            STR_IMM9(x1, wback, 0);    // put the whole value
                        }
-                        DMB_ISH();
+                        SMDMB();
                    }
                    break;
                case 6: //XOR
                    if(opcode==0x81) {INST_NAME("LOCK XOR Ed, Id");} else {INST_NAME("LOCK XOR Ed, Ib");}
                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        if(opcode==0x81) i32 = F32S; else i32 = F8S;
                        ed = xEAX+(nextop&7);
                        emit_xor32c(dyn, ninst, ed, i32, x3, x14);
                    } else {
-                        DMB_ISH();
                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 0, LOCK_LOCK);
                        if(opcode==0x81) i32 = F32S; else i32 = F8S;
                        MARKLOCK;
@@ -546,13 +525,12 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        STREX(x3, x1, wback);
                        CMPS_IMM8(x3, 0);
                        B_MARKLOCK(cNE);
-                        DMB_ISH();
+                        SMDMB();
                    }
                    break;
                case 7: //CMP
                    if(opcode==0x81) {INST_NAME("(LOCK) CMP Ed, Id");} else {INST_NAME("(LOCK) CMP Ed, Ib");}
                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    DMB_ISH();
                    GETEDH(x1);
                    // No need to LOCK, this is readonly
                    if(opcode==0x81) i32 = F32S; else i32 = F8S;
@@ -566,7 +544,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            }
            break;

-        // generic case, no lock needed, the value is only read (note that on x86 locked read is always followed wy a locked write)
+        // generic case, no lock needed, the value is only read (note that on x86 locked read is always followed by a locked write)
        #define GO(A)   \
        case A+0x02:    \
        case A+0x03:    \
@@ -582,18 +560,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        #undef GO
        case 0x86:  // for this two, the lock is already done by the opcode, so just ignoring it
        case 0x87:
-            DMB_ISH();
-            --addr;
-            break;
-        case 0x90:  // This is xchg reg, reg, so let's ignore the LOCK prefix...
-        case 0x91:
-        case 0x92:
-        case 0x93:
-        case 0x94:
-        case 0x95:
-        case 0x96:
-        case 0x97:
-            DMB_ISH();
+            SMDMB();
            --addr;
            break;
    
@@ -603,9 +570,8 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                case 0xB0:
                    INST_NAME("LOCK CMPXCHG Eb, Gb");
                    SETFLAGS(X_ALL, SF_SET_PENDING);
-                    DMB_ISH();
                    nextop = F8;
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        wback = (nextop&7);    
                        wb2 = (wback>>2);      
                        wback = xEAX+(wback&3);
@@ -640,7 +606,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    // AL != Eb
                    BFI(xEAX, ed, 0, 8);
                    MARK3;
-                    DMB_ISH();
+                    SMDMB();
                    // done, do the cmp now
                    UFLAG_IF {emit_cmp8(dyn, ninst, x1, x2, x3, x14);}
                    break;
@@ -649,8 +615,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    nextop = F8;
                    GETGD;
-                    DMB_ISH();
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        ed = xEAX+(nextop&7);
                        // same as no lock, all in regs...
                        CMPS_REG_LSL_IMM5(xEAX, ed, 0);
@@ -679,7 +644,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                                B_MARK(cNE);
                                // EAX == Ed
                                STREX(x14, gd, wback);
-                                DMB_ISH();
+                                SMDMB();
                                CMPS_IMM8(x14, 0);
                                B_MARKLOCK(cNE);
                                B_MARK(c__);
@@ -695,7 +660,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                            CMPS_IMM8(x14, 0);
                            B_MARK2(cNE);
                            STR_IMM9(gd, wback, 0);
-                            DMB_ISH();
+                            SMDMB();
                        } else {
                            MARKLOCK;
                            // alignment is good
@@ -706,7 +671,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                            STREX(x14, gd, wback);
                            CMPS_IMM8(x14, 0);
                            B_MARKLOCK(cNE);
-                            DMB_ISH();
+                            SMDMB();
                        }
                        // end of EAX == Ed or EAX != Ed
                        MARK;
@@ -721,8 +686,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    SET_DFNONE(x1);
                    nextop = F8;
                    GETGD;
-                    DMB_ISH();
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        ed = xEAX+(nextop&7);
                        wback = 0;
                    } else {
@@ -736,30 +700,26 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    }
                    AND_IMM8(x2, gd, 0x1f);
                    MOV_REG_LSR_REG(x14, ed, x2);
-                    ANDS_IMM8(x14, x14, 1);
                    BFI(xFlags, x14, F_CF, 1);
-                    B_MARK3(cEQ); // bit already clear, jump to end of instruction
                    MOVW(x14, 1);
-                    XOR_REG_LSL_REG(ed, ed, x14, x2);
+                    BIC_REG_LSL_REG(ed, ed, x14, x2);
                    if(wback) {
                        STREX(x14, ed, wback);
                        CMPS_IMM8(x14, 0);
                        B_MARKLOCK(cNE);
                    }
-                    MARK3;
-                    DMB_ISH();
+                    SMDMB();
                    break;

                case 0xBA:
                    nextop = F8;
-                    DMB_ISH();
                    switch((nextop>>3)&7) {
                        case 4:
                            INST_NAME("(LOCK) BT Ed, Ib");
                            SETFLAGS(X_CF, SF_SUBSET);
                            SET_DFNONE(x1);
                            gd = x2;
-                            if((nextop&0xC0)==0xC0) {
+                            if(MODREG) {
                                ed = xEAX+(nextop&7);
                            } else {
                                addr = geted(dyn, addr, ninst, nextop, &ed, x3, &fixedaddress, 4095-32, 0, 0, LOCK_LOCK);
@@ -778,26 +738,21 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                            INST_NAME("LOCK BTS Ed, Ib");
                            SETFLAGS(X_CF, SF_SUBSET);
                            SET_DFNONE(x1);
-                            gd = x2;
-                            if((nextop&0xC0)==0xC0) {
+                            if(MODREG) {
                                ed = xEAX+(nextop&7);
-                                u8 = F8;
-                                MOVW(gd, u8&0x1f);
+                                u8 = F8&0x1f;
                                wback = 0;
                            } else {
                                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0, 0, 0, LOCK_LOCK);
-                                u8 = F8;
+                                u8 = F8&0x1f;
                                ed = x1;
-                                MOVW(gd, u8&0x1f);
                                MARKLOCK;
                                LDREX(ed, wback);
                            }
-                            MOV_REG_LSR_REG(x14, ed, x2);
-                            ANDS_IMM8(x14, x14, 1);
+                            MOV_REG_LSR_IMM5(x14, ed, u8);
                            BFI(xFlags, x14, F_CF, 1);
-                            B_MARK3(cNE); // bit already set, jump to next instruction
                            MOVW(x14, 1);
-                            XOR_REG_LSL_REG(ed, ed, x14, x2);
+                            BFI(ed, x14, u8, 1);
                            if(wback) {
                                STREX(x14, ed, wback);
                                CMPS_IMM8(x14, 0);
@@ -809,26 +764,46 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                            INST_NAME("LOCK BTR Ed, Ib");
                            SETFLAGS(X_CF, SF_SUBSET);
                            SET_DFNONE(x1);
-                            gd = x2;
-                            if((nextop&0xC0)==0xC0) {
+                            if(MODREG) {
                                ed = xEAX+(nextop&7);
-                                u8 = F8;
-                                MOVW(gd, u8&0x1f);
+                                u8 = F8&0x1f;
                                wback = 0;
                            } else {
                                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0, 0, 0, LOCK_LOCK);
-                                u8 = F8;
+                                u8 = F8&0x1f;
                                ed = x1;
-                                MOVW(gd, u8&0x1f);
                                MARKLOCK;
                                LDREX(ed, wback);
                            }
-                            MOV_REG_LSR_REG(x14, ed, x2);
-                            ANDS_IMM8(x14, x14, 1);
+                            MOV_REG_LSR_IMM5(x14, ed, u8);
                            BFI(xFlags, x14, F_CF, 1);
-                            B_MARK3(cEQ); // bit already clear, jump to next instruction
-                            //MOVW(x14, 1); // already 0x01
-                            XOR_REG_LSL_REG(ed, ed, x14, x2);
+                            BFC(ed, u8, 1);
+                            if(wback) {
+                                STREX(x14, ed, wback);
+                                CMPS_IMM8(x14, 0);
+                                B_MARKLOCK(cNE);
+                            }
+                            MARK3;
+                            break;
+                        case 7:
+                            INST_NAME("LOCK BTC Ed, Ib");
+                            SETFLAGS(X_CF, SF_SUBSET);
+                            SET_DFNONE(x1);
+                            if(MODREG) {
+                                ed = xEAX+(nextop&7);
+                                u8 = F8&0x1f;
+                                wback = 0;
+                            } else {
+                                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0, 0, 0, LOCK_LOCK);
+                                u8 = F8&0x1f;
+                                ed = x1;
+                                MARKLOCK;
+                                LDREX(ed, wback);
+                            }
+                            MOV_REG_LSR_IMM5(x14, ed, u8);
+                            BFI(xFlags, x14, F_CF, 1);
+                            MOV32(x14, 1);
+                            XOR_REG_LSL_IMM5(ed, ed, x14, u8);
                            if(wback) {
                                STREX(x14, ed, wback);
                                CMPS_IMM8(x14, 0);
@@ -839,7 +814,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        default:
                            DEFAULT;
                    }
-                    DMB_ISH();
+                    SMDMB();
                    break;
                case 0xBB:
                    INST_NAME("LOCK BTC Ed, Gd");
@@ -847,8 +822,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    SET_DFNONE(x1);
                    nextop = F8;
                    GETGD;
-                    DMB_ISH();
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        ed = xEAX+(nextop&7);
                        wback = 0;
                    } else {
@@ -870,16 +844,16 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        CMPS_IMM8(x14, 0);
                        B_MARKLOCK(cNE);
                    }
-                    DMB_ISH();
+                    SMDMB();
                    break;
            
                case 0xC0:
                    INST_NAME("LOCK XADD Gb, Eb");
                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    nextop = F8;
-                    DMB_ISH();
+                    SMDMB();
                    GETGB(x1);
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        wback = (nextop&7); 
                        wb2 = (wback>>2); 
                        wback = xEAX+(wback&3); 
@@ -902,15 +876,14 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    } else {
                        BFI(wback, ed, wb2*8, 8);
                    }
-                    DMB_ISH();
+                    SMDMB();
                    break;
                case 0xC1:
                    INST_NAME("LOCK XADD Gd, Ed");
                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    nextop = F8;
                    GETGD;
-                    DMB_ISH();
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        ed = xEAX+(nextop&7);
                        if(gd!=ed) {
                            XOR_REG_LSL_IMM5(gd, gd, ed, 0);    // swap gd, ed
@@ -955,7 +928,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        }
                        MOV_REG(gd, x1);
                    }
-                    DMB_ISH();
+                    SMDMB();
                    break;

                case 0xC7:
@@ -964,7 +937,6 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                        case 1:
                            INST_NAME("LOCK CMPXCHG8B Gq, Eq");
                            SETFLAGS(X_ZF, SF_SUBSET);
-                            DMB_ISH();
                            addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 0, 0, 1, LOCK_LOCK);
                            MARKLOCK;
                            if(!fixedaddress || (fixedaddress && (fixedaddress&7))) {
@@ -1001,7 +973,7 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                            MOV_REG(xEDX, x3);
                            MOVW(x1, 0);
                            MARK3;
-                            DMB_ISH();
+                            SMDMB();
                            BFI(xFlags, x1, F_ZF, 1);
                            break;
                        default:
@@ -1010,20 +982,65 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;

                default:
-                    // dafault to NO LOCK
-                    addr-=2;
+                    DEFAULT;
            }
            break;

+        case 0xFE:
+            nextop = F8;
+            SMDMB();
+            switch((nextop>>3)&7)
+            {
+                case 0: // INC Eb
+                    INST_NAME("INC Eb");
+                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
+                    if(MODREG) {
+                        GETEB(x1);
+                        emit_inc8(dyn, ninst, x1, x2, x14);
+                        EBBACK;
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 1, LOCK_LOCK);
+                        MARKLOCK;
+                        LDREXB(x1, wback);
+                        emit_inc8(dyn, ninst, x1, x3, x14);
+                        STREXB(x3, x1, wback);
+                        CMPS_IMM8(x3, 0);
+                        B_MARKLOCK(cNE);
+                        SMDMB();
+                    }
+                    break;
+                case 1: //DEC Ed
+                    INST_NAME("DEC Eb");
+                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
+                    if(MODREG) {
+                        GETEB(x1);
+                        emit_dec8(dyn, ninst, x1, x2, x14);
+                        EBBACK;
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 1, LOCK_LOCK);
+                        MARKLOCK;
+                        LDREXB(x1, wback);
+                        emit_dec8(dyn, ninst, x1, x3, x14);
+                        STREXB(x3, x1, wback);
+                        CMPS_IMM8(x3, 0);
+                        B_MARKLOCK(cNE);
+                        SMDMB();
+                    }
+                    break;
+                default:
+                    DEFAULT;
+            }
+            SMDMB();
+            break;
        case 0xFF:
            nextop = F8;
-            DMB_ISH();
+            SMDMB();
            switch((nextop>>3)&7)
            {
                case 0: // INC Ed
                    INST_NAME("INC Ed");
                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        ed = xEAX+(nextop&7);
                        emit_inc32(dyn, ninst, ed, x3, x14);
                    } else {
@@ -1039,9 +1056,9 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                            STREX(x3, x1, wback);
                            CMPS_IMM8(x3, 0);
                            B_MARKLOCK(cNE);
+                            SMDMB();
                        }
                        if(!fixedaddress) {
-                            DMB_ISH();
                            B_NEXT(c__);
                        }
                        if(!fixedaddress || (fixedaddress && (fixedaddress&3))) {
@@ -1054,13 +1071,14 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                            CMPS_IMM8(x3, 0);
                            B_MARK(cNE);
                            STR_IMM9(x1, wback, 0);
+                            SMDMB();
                        }
                    }
                    break;
                case 1: //DEC Ed
                    INST_NAME("DEC Ed");
                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
-                    if((nextop&0xC0)==0xC0) {
+                    if(MODREG) {
                        ed = xEAX+(nextop&7);
                        emit_dec32(dyn, ninst, ed, x3, x14);
                    } else {
@@ -1076,9 +1094,9 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                            STREX(x3, x1, wback);
                            CMPS_IMM8(x3, 0);
                            B_MARKLOCK(cNE);
+                            SMDMB();
                        }
                        if(!fixedaddress) {
-                            DMB_ISH();
                            B_NEXT(c__);
                        }
                        if(!fixedaddress || (fixedaddress && (fixedaddress&3))) {
@@ -1091,19 +1109,18 @@ uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                            CMPS_IMM8(x3, 0);
                            B_MARK(cNE);
                            STR_IMM9(x1, wback, 0);
+                            SMDMB();
                        }
                    }
                    break;
                default:
-                    // default to NO LOCK
-                    addr-=2;
+                    DEFAULT;
            }
-            DMB_ISH();
+            SMDMB();
            break;
       
        default:
-            // default to NO LOCK
-            --addr;
+            DEFAULT;
    }

    return addr;
@@ -24,11 +24,12 @@

 // Get Ex as a double, not a quad (warning, x2 and x3 may get used)
 #define GETEX(a, w)                                 \
-    if((nextop&0xC0)==0xC0) {                       \
+    if(MODREG) {                       \
        a = sse_get_reg(dyn, ninst, x1, nextop&7,w);\
    } else {    \
        parity = getedparity(dyn, ninst, addr, nextop, 3);  \
        a = fpu_get_scratch_double(dyn);            \
+        SMREAD();                                   \
        if(parity) {                                \
            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL); \
            VLDR_64(a, ed, fixedaddress);           \
@@ -67,11 +68,12 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("MOVSD Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
                d0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                VMOVD(v0, d0);
            } else {
+                SMREAD();
                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
                LDR_IMM9(x2, ed, fixedaddress+0);
@@ -85,7 +87,7 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            nextop = F8;
            gd = (nextop&0x38)>>3;
            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                d0 = sse_get_reg(dyn, ninst, x1, nextop&7, 1);
                VMOVD(d0, v0);
            } else {
@@ -93,17 +95,19 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
                STR_IMM9(x2, ed, fixedaddress+0);
                STR_IMM9(x3, ed, fixedaddress+4);
+                SMWRITE2();
            }
            break;
        case 0x12:
            INST_NAME("MOVDDUP Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                d0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                VMOVD(v0, d0);
            } else {
+                SMREAD();
                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
                LDR_IMM9(x2, ed, fixedaddress+0);
@@ -129,31 +133,53 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            nextop = F8;
            GETGD;
            s0 = fpu_get_scratch_single(dyn);
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                d0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
            } else {
+                SMREAD();
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                d0 = fpu_get_scratch_double(dyn);
                VLDR_64(d0, ed, fixedaddress);
            }
+            if(!box86_dynarec_fastround) {
+                VMRS(x14);   // Get FPCSR reg to clear exceptions flags
+                ORR_IMM8(x3, x14, 0b010, 9); // enable exceptions
+                BIC_IMM8(x3, x3, 0b10011111, 0);
+                VMSR(x3);
+            }
            VCVT_S32_F64(s0, d0);
            VMOVfrV(gd, s0);
+            if(!box86_dynarec_fastround) {
+                VMRS(x3);   // get the FPCSR reg and test FPU exception (IO only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                MOV_IMM_COND(cNE, gd, 0b10, 1);   // 0x80000000
+                VMSR(x14);  // put back values
+            }
            break;
        case 0x2D:
            INST_NAME("CVTSD2SI Gd, Ex");
-            u8 = sse_setround(dyn, ninst, x1, x2, x14);
+            if(!box86_dynarec_fastround)
+                u8 = sse_setround_reset(dyn, ninst, x1, x2, x14);
+            else
+                u8 = sse_setround(dyn, ninst, x1, x2, x14);
            nextop = F8;
            GETGD;
            s0 = fpu_get_scratch_single(dyn);
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                d0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
            } else {
+                SMREAD();
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                d0 = fpu_get_scratch_double(dyn);
                VLDR_64(d0, ed, fixedaddress);
            }
            VCVTR_S32_F64(s0, d0);
            VMOVfrV(gd, s0);
+            if(!box86_dynarec_fastround) {
+                VMRS(x3);   // get the FPCSR reg and test FPU exception (IO only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                MOV_IMM_COND(cNE, gd, 0b10, 1);   // 0x80000000
+            }
            x87_restoreround(dyn, ninst, u8);
            break;

@@ -302,9 +328,10 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            nextop = F8;
            gd = (nextop&0x38)>>3;
            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
            } else {
+                SMREAD();
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                v1 = fpu_get_scratch_quad(dyn);
                VLD1Q_64(v1, ed);
@@ -337,9 +364,10 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            nextop = F8;
            gd = (nextop&0x38)>>3;
            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
            } else {
+                SMREAD();
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                v1 = fpu_get_scratch_quad(dyn);
                VLD1Q_64(v1, ed);
@@ -401,15 +429,43 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            VMOV_64(v0, v1);
            break;

+        case 0xE6:
+            INST_NAME("CVTPD2DQ Gx, Ex");
+            nextop = F8;
+            gd = (nextop&0x38)>>3;
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+            if(MODREG) {
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
+                v1 = fpu_get_scratch_quad(dyn);
+                VLD1Q_8(v1, ed);
+            }
+            u8 = sse_setround(dyn, ninst, x1, x2, x14);
+            if(v0<16) {
+                VCVTR_S32_F64(v0*2, v1);
+                VCVTR_S32_F64(v0*2+1, v1+1);
+            } else {
+                d0 = fpu_get_scratch_double(dyn);
+                VCVTR_S32_F64(d0*2, v1);
+                VCVTR_S32_F64(d0*2+1, v1+1);
+                VMOVD(v0, d0);
+            }
+            x87_restoreround(dyn, ninst, u8);
+            VEOR(v0+1, v0+1, v0+1);
+            break;
+
        case 0xF0:
            INST_NAME("LDDQ Gx,Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                VMOVQ(v0, v1);
            } else {
+                SMREAD();
                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-12, 0, 0, NULL);
                LDR_IMM9(x2, ed, fixedaddress+0);
@@ -24,7 +24,7 @@

 // Get Ex as a single, not a quad, with b a intermediary double
 #define GETEX(a, b)                                 \
-    if((nextop&0xC0)==0xC0) {                       \
+    if(MODREG) {                       \
        b = sse_get_reg(dyn, ninst, x1, nextop&7,0);\
        if(b<16)                                    \
            a = b;                                  \
@@ -37,6 +37,7 @@
        b = fpu_get_scratch_double(dyn);            \
        a = b*2;                                    \
        parity = getedparity(dyn, ninst, addr, nextop, 2);  \
+        SMREAD();                                   \
        if(parity) {                                \
            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 0, 0, NULL); \
            VLDR_32(a, ed, fixedaddress);           \
@@ -80,7 +81,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("MOVSS Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
                q0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                if(v0<16)
@@ -102,6 +103,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            } else {
                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                parity = getedparity(dyn, ninst, addr, nextop, 2);
+                SMREAD();
                if(parity && (v0<16)) {
                    VEORQ(v0, v0, v0);
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 0, 0, NULL);
@@ -119,7 +121,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            nextop = F8;
            gd = (nextop&0x38)>>3;
            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                q0 = sse_get_reg(dyn, ninst, x1, nextop&7, 1);
                if(v0<16)
                    d0 = v0;
@@ -147,15 +149,17 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0, 0, NULL);
                    STR_IMM9(x2, ed, fixedaddress);
                }
+                SMWRITE2();
            }
            break;
        case 0x12:
            INST_NAME("MOVSLDUP Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                q1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
            } else {
+                SMREAD();
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                q1 = fpu_get_scratch_quad(dyn);
                VLD1Q_64(q1, ed);
@@ -169,9 +173,10 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("MOVSHDUP Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                q1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
            } else {
+                SMREAD();
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                q1 = fpu_get_scratch_quad(dyn);
                VLD1Q_64(q1, ed);
@@ -218,7 +223,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            nextop = F8;
            GETGD;
            s0 = fpu_get_scratch_single(dyn);
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                v0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                if(v0<16)
                    d0 = v0;
@@ -229,6 +234,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                VMOV_32(s0, d0*2);
            } else {
                parity = getedparity(dyn, ninst, addr, nextop, 3);
+                SMREAD();
                if(parity) {
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                    VLDR_32(s0, ed, fixedaddress);
@@ -238,16 +244,31 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                    VMOVtoV(s0, x2);
                }
            }
+            if(!box86_dynarec_fastround) {
+                VMRS(x14);   // Get FPCSR reg to clear exceptions flags
+                ORR_IMM8(x3, x14, 0b010, 9); // enable exceptions
+                BIC_IMM8(x3, x3, 0b10011111, 0);
+                VMSR(x3);
+            }
            VCVT_S32_F32(s0, s0);
            VMOVfrV(gd, s0);
+            if(!box86_dynarec_fastround) {
+                VMRS(x3);   // get the FPCSR reg and test FPU exception (IO only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                MOV_IMM_COND(cNE, gd, 0b10, 1);   // 0x80000000
+                VMSR(x14);  // put back values
+            }
            break;
        case 0x2D:
            INST_NAME("CVTSS2SI Gd, Ex");
-            u8 = sse_setround(dyn, ninst, x1, x2, x14);
+            if(!box86_dynarec_fastround)
+                u8 = sse_setround_reset(dyn, ninst, x1, x2, x14);
+            else
+                u8 = sse_setround(dyn, ninst, x1, x2, x14);
            nextop = F8;
            GETGD;
            s0 = fpu_get_scratch_single(dyn);
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                v0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                if(v0<16)
                    d0 = v0;
@@ -258,6 +279,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                VMOV_32(s0, d0*2);
            } else {
                parity = getedparity(dyn, ninst, addr, nextop, 3);
+                SMREAD();
                if(parity) {
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                    VLDR_32(s0, ed, fixedaddress);
@@ -269,6 +291,11 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            }
            VCVTR_S32_F32(s0, s0);
            VMOVfrV(gd, s0);
+            if(!box86_dynarec_fastround) {
+                VMRS(x3);   // get the FPCSR reg and test FPU exception (IO only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                MOV_IMM_COND(cNE, gd, 0b10, 1);   // 0x80000000
+            }
            x87_restoreround(dyn, ninst, u8);
            break;

@@ -393,9 +420,10 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
        case 0x5B:
            INST_NAME("CVTTPS2DQ Gx, Ex");
            nextop = F8;
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
            } else {
+                SMREAD();
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                v1 = fpu_get_scratch_quad(dyn);
                VLD1Q_64(v1, ed);
@@ -495,11 +523,12 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("MOVDQU Gx,Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                VMOVQ(v0, v1);
            } else {
+                SMREAD();
                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-12, 0, 0, NULL);
                LDR_IMM9(x2, ed, fixedaddress+0);
@@ -514,14 +543,15 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("PSHUFHW Gx, Ex, Ib");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
            } else {
+                SMREAD();
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                v1 = fpu_get_scratch_quad(dyn);
                VLD1Q_64(v1, ed);
            }
+            v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
            u8 = F8;
            // only high part need to be shuffled. VTBL only handle 8bits value, so the 16bits shuffles need to be changed in 8bits
            u32 = 0;
@@ -549,11 +579,12 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("MOVQ Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                VMOVD(v0, v1);
            } else {
+                SMREAD();
                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                VLD1_64(v0, ed);
@@ -565,7 +596,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            nextop = F8;
            gd = (nextop&0x38)>>3;
            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                v1 = sse_get_reg_empty(dyn, ninst, x1, nextop&7);
                VMOVQ(v1, v0);
            } else {
@@ -576,6 +607,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                VMOVfrV_D(x2, x3, v0+1);
                STR_IMM9(x2, ed, fixedaddress+8);
                STR_IMM9(x3, ed, fixedaddress+12);
+                SMWRITE2();
            }
            break;

@@ -634,12 +666,13 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("MOVQ2DQ Gx, Em");
            nextop = F8;
            GETGX_empty(v0);
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                v1 = mmx_get_reg(dyn, ninst, x1, x2, x3, (nextop&7));
                VEORQ(v0, v0, v0);  // usefull?
                VMOV_64(v0, v1);
            } else {
                parity = getedparity(dyn, ninst, addr, nextop, 3);
+                SMREAD();
                if(parity) {
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                    VLDR_64(v0, ed, fixedaddress);    // vfpu opcode here
@@ -655,7 +688,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
        case 0xE6:
            INST_NAME("CVTDQ2PD Gx, Ex");
            nextop = F8;
-            if((nextop&0xC0)==0xC0) {
+            if(MODREG) {
                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                if(v1<16)
                    v0 = v1;
@@ -666,6 +699,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            } else {
                v0 = fpu_get_scratch_double(dyn);
                parity = getedparity(dyn, ninst, addr, nextop, 3);
+                SMREAD();
                if(parity) {
                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                    VLDR_64(v0, ed, fixedaddress);    // vfpu opcode here
@@ -27,6 +27,8 @@
 #include "dynarec_arm_private.h"
 #include "dynarec_arm_functions.h"
 #include "bridge.h"
+#include "arm_printer.h"
+#include "custommem.h"

 void arm_fstp(x86emu_t* emu, void* p)
 {
@@ -60,19 +62,41 @@ void arm_fpatan(x86emu_t* emu)
 }
 void arm_fxtract(x86emu_t* emu)
 {
-    int32_t tmp32s = (ST1.q&0x7ff0000000000000LL)>>52;
-    tmp32s -= 1023;
-    ST1.d /= exp2(tmp32s);
-    ST0.d = tmp32s;
+    int tmp32s;
+    if(isnan(ST1.d)) {
+        ST0.d = ST1.d;
+    } else if(isinf(ST1.d)) {
+        ST0.d = ST1.d;
+        ST1.d = INFINITY;
+    } else if(ST1.d==0.0) {
+        ST0.d = ST1.d;
+        ST1.d = -INFINITY;
+    } else {
+        // LD80bits doesn't have implicit "1" bit, so need to adjust for that
+        ST0.d = frexp(ST1.d, &tmp32s)*2;
+        ST1.d = tmp32s-1;
+    }
 }
 void arm_fprem(x86emu_t* emu)
 {
-    int64_t tmp64s = ST0.d / ST1.d;
-    ST0.d -= ST1.d * tmp64s;
-    emu->sw.f.F87_C2 = 0;
-    emu->sw.f.F87_C1 = (tmp64s&1);
-    emu->sw.f.F87_C3 = ((tmp64s>>1)&1);
-    emu->sw.f.F87_C0 = ((tmp64s>>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 arm_frndint(x86emu_t* emu)
 {
@@ -102,25 +126,10 @@ void arm_fbstp(x86emu_t* emu, uint8_t* ed)
    fpu_fbst(emu, ed);
 }

-void arm_fistp64(x86emu_t* emu, int64_t* ed)
-{
-    // used of memcpy to avoid aligments issues
-    if(STll(0).sref==ST(0).sq) {
-        memcpy(ed, &STll(0).sq, sizeof(int64_t));
-    } else {
-        int64_t tmp;
-        if(isgreater(ST0.d, (double)(int64_t)0x7fffffffffffffffLL) || isless(ST0.d, (double)(int64_t)0x8000000000000000LL) || !isfinite(ST0.d))
-            tmp = 0x8000000000000000LL;
-        else
-            tmp = fpu_round(emu, ST0.d);
-        memcpy(ed, &tmp, sizeof(tmp));
-    }
-}
-
 int64_t arm_fist64_0(double d)
 {
    int64_t tmp;
-    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, (double)(int64_t)0x8000000000000000LL) || !isfinite(d))
+    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, -(double)(uint64_t)0x8000000000000000LL) || !isfinite(d))
        tmp = 0x8000000000000000LL;
    else
        tmp = nearbyint(d);
@@ -129,7 +138,7 @@ int64_t arm_fist64_0(double d)
 int64_t arm_fist64_1(double d)
 {
    int64_t tmp;
-    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, (double)(int64_t)0x8000000000000000LL) || !isfinite(d))
+    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, -(double)(uint64_t)0x8000000000000000LL) || !isfinite(d))
        tmp = 0x8000000000000000LL;
    else
        tmp = floor(d);
@@ -138,7 +147,7 @@ int64_t arm_fist64_1(double d)
 int64_t arm_fist64_2(double d)
 {
    int64_t tmp;
-    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, (double)(int64_t)0x8000000000000000LL) || !isfinite(d))
+    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, -(double)(uint64_t)0x8000000000000000LL) || !isfinite(d))
        tmp = 0x8000000000000000LL;
    else
        tmp = ceil(d);
@@ -147,20 +156,13 @@ int64_t arm_fist64_2(double d)
 int64_t arm_fist64_3(double d)
 {
    int64_t tmp;
-    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, (double)(int64_t)0x8000000000000000LL) || !isfinite(d))
+    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, -(double)(uint64_t)0x8000000000000000LL) || !isfinite(d))
        tmp = 0x8000000000000000LL;
    else
        tmp = trunc(d);
    return tmp;
 }

-void arm_fistt64(x86emu_t* emu, int64_t* ed)
-{
-    // used of memcpy to avoid aligments issues
-    int64_t tmp = ST0.d;
-    memcpy(ed, &tmp, sizeof(tmp));
-}
-
 void arm_fld(x86emu_t* emu, uint8_t* ed)
 {
    memcpy(&STld(0).ld, ed, 10);
@@ -170,9 +172,22 @@ void arm_fld(x86emu_t* emu, uint8_t* ed)

 void arm_ud(x86emu_t* emu)
 {
+    emu->test.test = 0;
    emit_signal(emu, SIGILL, (void*)R_EIP, 0);
 }

+void arm_priv(x86emu_t* emu)
+{
+    emu->test.test = 0;
+    emit_signal(emu, SIGSEGV, (void*)R_EIP, 128);
+}
+
+void arm_singlestep(x86emu_t* emu)
+{
+    emu->test.test = 0;
+    emit_signal(emu, SIGTRAP, (void*)R_EIP, 2);
+}
+
 void arm_fsave(x86emu_t* emu, uint8_t* ed)
 {
    fpu_savenv(emu, (char*)ed, 0);
@@ -199,15 +214,185 @@ void arm_frstor(x86emu_t* emu, uint8_t* ed)

 void arm_fprem1(x86emu_t* emu)
 {
-    // simplified version
-    int64_t tmp64s = lrint(ST0.d / ST1.d);
-    ST0.d -= ST1.d*tmp64s;
-    emu->sw.f.F87_C2 = 0;
-    emu->sw.f.F87_C1 = (tmp64s&1);
-    emu->sw.f.F87_C3 = ((tmp64s>>1)&1);
-    emu->sw.f.F87_C0 = ((tmp64s>>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)
+{
+	int retval = 0;
+
+	for(int i = 0; i < 8; i++) {
+		if((b & 1) == 1)
+			retval ^= a;
+
+		if((a & 0x80)) {
+			a <<= 1;
+			a  ^= 0x1b;
+		} else {
+			a <<= 1;
+		}
+
+		b >>= 1;
+	}
+
+	return retval;
+}
+
+void arm_aesimc(x86emu_t* emu, int xmm)
+{
+    sse_regs_t eax1 = emu->xmm[xmm];
+
+    for(int j=0; j<4; ++j) {
+        emu->xmm[xmm].ub[0+j*4] = ff_mult(0x0E, eax1.ub[0+j*4]) ^ ff_mult(0x0B, eax1.ub[1+j*4]) ^ ff_mult(0x0D, eax1.ub[2+j*4]) ^ ff_mult(0x09, eax1.ub[3+j*4]);
+        emu->xmm[xmm].ub[1+j*4] = ff_mult(0x09, eax1.ub[0+j*4]) ^ ff_mult(0x0E, eax1.ub[1+j*4]) ^ ff_mult(0x0B, eax1.ub[2+j*4]) ^ ff_mult(0x0D, eax1.ub[3+j*4]);
+        emu->xmm[xmm].ub[2+j*4] = ff_mult(0x0D, eax1.ub[0+j*4]) ^ ff_mult(0x09, eax1.ub[1+j*4]) ^ ff_mult(0x0E, eax1.ub[2+j*4]) ^ ff_mult(0x0B, eax1.ub[3+j*4]);
+        emu->xmm[xmm].ub[3+j*4] = ff_mult(0x0B, eax1.ub[0+j*4]) ^ ff_mult(0x0D, eax1.ub[1+j*4]) ^ ff_mult(0x09, eax1.ub[2+j*4]) ^ ff_mult(0x0E, eax1.ub[3+j*4]);
+    }
+}
+void arm_aesmc(x86emu_t* emu, int xmm)
+{
+    sse_regs_t eax1 = emu->xmm[xmm];
+
+    for(int j=0; j<4; ++j) {
+        emu->xmm[xmm].ub[0+j*4] = ff_mult(0x02, eax1.ub[0+j*4]) ^ ff_mult(0x03, eax1.ub[1+j*4]) ^               eax1.ub[2+j*4]  ^               eax1.ub[3+j*4] ;
+        emu->xmm[xmm].ub[1+j*4] =               eax1.ub[0+j*4]  ^ ff_mult(0x02, eax1.ub[1+j*4]) ^ ff_mult(0x03, eax1.ub[2+j*4]) ^               eax1.ub[3+j*4] ;
+        emu->xmm[xmm].ub[2+j*4] =               eax1.ub[0+j*4]  ^               eax1.ub[1+j*4]  ^ ff_mult(0x02, eax1.ub[2+j*4]) ^ ff_mult(0x03, eax1.ub[3+j*4]);
+        emu->xmm[xmm].ub[3+j*4] = ff_mult(0x03, eax1.ub[0+j*4]) ^               eax1.ub[1+j*4]  ^               eax1.ub[2+j*4]  ^ ff_mult(0x02, eax1.ub[3+j*4]);
+    }
+}
+void arm_aesdlast(x86emu_t* emu, int xmm)
+{
+                            //   A0 B1 C2 D3 E4 F5 G6 H7 I8 J9 Ka Lb Mc Nd Oe Pf
+                            //   A  N  K  H  E  B  O  L  I  F  C  P  M  J  G  D
+    const uint8_t invshiftrows[] = {0,13,10, 7, 4, 1,14,11, 8, 5, 2,15,12, 9, 6, 3};
+    const uint8_t invsubbytes[256] = {
+        0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38, 0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb,
+        0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87, 0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb,
+        0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d, 0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e,
+        0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2, 0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25,
+        0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16, 0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92,
+        0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda, 0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84,
+        0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a, 0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06,
+        0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02, 0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b,
+        0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea, 0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73,
+        0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85, 0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e,
+        0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89, 0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b,
+        0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20, 0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4,
+        0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31, 0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f,
+        0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d, 0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef,
+        0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0, 0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61,
+        0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26, 0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d,
+    };
+
+    sse_regs_t eax1;
+    for(int i=0; i<16; ++i)
+        eax1.ub[i] = emu->xmm[xmm].ub[invshiftrows[i]];
+    //STATE ← InvSubBytes( STATE );
+    for(int i=0; i<16; ++i)
+        emu->xmm[xmm].ub[i] = invsubbytes[eax1.ub[i]];
+
+}
+static const uint8_t shiftrows[] = {0, 5,10,15, 4, 9,14, 3, 8,13, 2, 7,12, 1, 6,11};
+static const uint8_t subbytes[256] = {
+    0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5, 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76,
+    0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0, 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0,
+    0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc, 0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15,
+    0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a, 0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75,
+    0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0, 0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84,
+    0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b, 0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf,
+    0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85, 0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8,
+    0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5, 0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2,
+    0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17, 0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73,
+    0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88, 0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb,
+    0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c, 0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79,
+    0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9, 0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08,
+    0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6, 0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a,
+    0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e, 0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e,
+    0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94, 0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf,
+    0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68, 0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16,
+};
+void arm_aeselast(x86emu_t* emu, int xmm)
+{
+                            //   A0 B1 C2 D3 E4 F5 G6 H7 I8 J9 Ka Lb Mc Nd Oe Pf
+                            //   A  F  K  P  E  J  O  D  I  N  C  H  M  B  G  L
+    sse_regs_t eax1;
+    for(int i=0; i<16; ++i)
+        eax1.ub[i] = emu->xmm[xmm].ub[shiftrows[i]];
+    //STATE ← SubBytes( STATE );
+    for(int i=0; i<16; ++i)
+        emu->xmm[xmm].ub[i] = subbytes[eax1.ub[i]];
+}
+void arm_aesd(x86emu_t* emu, int xmm)
+{
+    arm_aesdlast(emu, xmm);
+    arm_aesimc(emu, xmm);
+}
+void arm_aese(x86emu_t* emu, int xmm)
+{
+    arm_aeselast(emu, xmm);
+    arm_aesmc(emu, xmm);
+}
+void arm_aeskeygenassist(x86emu_t* emu, int gx, void* p, uint32_t u8)
+{
+    sse_regs_t *EX = ((uintptr_t)p>8)?((sse_regs_t*)p):&emu->xmm[(uintptr_t)p];
+    sse_regs_t *GX = &emu->xmm[gx];
+    for (int i = 4; i < 8; ++i)
+        GX->ub[i] = subbytes[EX->ub[i]];
+    for (int i = 12; i < 16; ++i)
+        GX->ub[i] = subbytes[EX->ub[i]];
+    GX->ud[0] = GX->ud[1];
+    uint8_t tmp8u = GX->ub[4];
+    GX->ud[1] = GX->ud[1] >> 8;
+    GX->ub[7] = tmp8u;
+    GX->ud[1] ^= u8;
+    GX->ud[2] = GX->ud[3];
+    tmp8u = GX->ub[12];
+    GX->ud[3] = GX->ud[3] >> 8;
+    GX->ub[15] = tmp8u;
+    GX->ud[3] ^= u8;
+}
+
+void arm_pclmul(x86emu_t* emu, int gx, void* p, uint32_t u8)
+{
+    int ex = (((uintptr_t)p)<8)?((uintptr_t)p):-1;
+    sse_regs_t *EX = (ex==-1)?((sse_regs_t*)p):&emu->xmm[ex];
+    sse_regs_t *GX = &emu->xmm[gx];
+    int g = (u8&1)?1:0;
+    int e = (u8&0b10000)?1:0;
+    uint64_t result[2] = {0};
+    uint64_t op2 = EX->q[e];
+    for (int i=0; i<64; ++i)
+        if(GX->q[g]&(1LL<<i)) {
+            result[0] ^= (EX->q[e]<<i);
+            if(i)
+                result[1] ^= (EX->q[e]>>(64-i));
+
+        }
+    GX->q[0] = result[0];
+    GX->q[1] = result[1];
+}
+
+void arm_clflush(x86emu_t* emu, void* p)
+{
+    cleanDBFromAddressRange((uintptr_t)p, 8, 0);
+}

 // Get a FPU single scratch reg
 int fpu_get_scratch_single(dynarec_arm_t* dyn)
@@ -290,13 +475,18 @@ void fpu_free_reg_quad(dynarec_arm_t* dyn, int reg)
    dyn->n.neoncache[i+1].v = 0;
 }
 // Reset fpu regs counter
+static void fpu_reset_reg_neoncache(neoncache_t* n)
+{
+    n->fpu_reg = 0;
+    for (int i=0; i<24; ++i) {
+        n->fpuused[i]=0;
+        n->neoncache[i].v = 0;
+    }
+
+}
 void fpu_reset_reg(dynarec_arm_t* dyn)
 {
-    dyn->n.fpu_reg = 0;
-    for (int i=0; i<24; ++i) {
-        dyn->n.fpuused[i]=0;
-        dyn->n.neoncache[i].v = 0;
-    }
+    fpu_reset_reg_neoncache(&dyn->n);
 }

 int neoncache_get_st(dynarec_arm_t* dyn, int ninst, int a)
@@ -543,6 +733,7 @@ void neoncacheUnwind(neoncache_t* cache)
        // unswap
        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)) {
                if(cache->neoncache[j].n == cache->combined1)
@@ -559,7 +750,7 @@ void neoncacheUnwind(neoncache_t* cache)
        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;
@@ -576,11 +767,23 @@ 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
@@ -608,6 +811,7 @@ void neoncacheUnwind(neoncache_t* cache)
                case NEON_CACHE_XMMW:
                    cache->ssecache[cache->neoncache[i].n].reg = i+FPUFIRST;
                    cache->ssecache[cache->neoncache[i].n].write = (cache->neoncache[i].t==NEON_CACHE_XMMW)?1:0;
+                    cache->neoncache[i] = cache->neoncache[i+1];
                    ++cache->fpu_reg;
                    ++i;    // next next
                    break;
@@ -694,6 +898,12 @@ uintptr_t fakeed(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop)
    }
    return addr;
 }
+// return Ib on a mod/rm opcode without emiting anything
+uint8_t geted_ib(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop)
+{
+    addr = fakeed(dyn, addr, ninst, nextop);
+    return F8;
+}
 #undef F8
 #undef F32

@@ -702,13 +912,13 @@ int isNativeCall(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t* calladdress, int
 #define PK(a)       *(uint8_t*)(addr+a)
 #define PK32(a)     *(uint32_t*)(addr+a)

-    if(!addr)
+    if(!addr || !getProtection(addr))
        return 0;
    if(PK(0)==0xff && PK(1)==0x25) {  // absolute jump, maybe the GOT
        uintptr_t a1 = (PK32(2));   // need to add a check to see if the address is from the GOT !
        addr = (uintptr_t)getAlternate(*(void**)a1); 
    }
-    if(addr<0x10000)    // too low, that is suspicious
+    if(addr<0x10000 || !getProtection(addr))    // too low, that is suspicious
        return 0;
    onebridge_t *b = (onebridge_t*)(addr);
    if(b->CC==0xCC && b->S=='S' && b->C=='C' && b->w!=(wrapper_t)0 && b->f!=(uintptr_t)PltResolver) {
@@ -737,62 +947,143 @@ const char* getCacheName(int t, int n)
    return buff;
 }

-// is inst clean for a son branch?
-int isInstClean(dynarec_arm_t* dyn, int ninst)
-{
-    // check flags cache
-    if(dyn->insts[ninst].f_entry.dfnone || dyn->insts[ninst].f_entry.pending)
-        return 0;
-    if(dyn->insts[ninst].x86.state_flags)
-        return 0;
-    // check neoncache
-    neoncache_t* n = &dyn->insts[ninst].n;
-    if(n->news || n->stack || n->stack_next)
-        return 0;
-    for(int i=0; i<24; ++i)
-        if(n->neoncache[i].v)
-            return 0;
-    return 1;
-}
-
 static int flagsCacheNeedsTransform(dynarec_arm_t* dyn, int ninst) {
    int jmp = dyn->insts[ninst].x86.jmp_insts;
    if(jmp<0)
        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) {*/
-    switch (dyn->insts[jmp].f_entry.pending) {
-        case SF_UNKNOWN: return 0;
-        case SF_SET: 
-            if(dyn->insts[ninst].f_exit.pending!=SF_SET && dyn->insts[ninst].f_exit.pending!=SF_SET_PENDING) 
-                return 1; 
-            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 
-                return 0;
-        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[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:
+            if(dyn->insts[ninst].f_exit.pending!=SF_SET && dyn->insts[ninst].f_exit.pending!=SF_SET_PENDING)
+                return 1;
+            else
+                return 0;
+        case SF_SET_PENDING:
+            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_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;
+    }
    return 0;
 }
+
 int CacheNeedsTransform(dynarec_arm_t* dyn, int ninst) {
    int ret = 0;
    if (fpuCacheNeedsTransform(dyn, ninst)) ret|=1;
    if (flagsCacheNeedsTransform(dyn, ninst)) ret|=2;
    return ret;
+}
+
+void inst_name_pass3(dynarec_arm_t* dyn, int ninst, const char* name)
+{
+    if(box86_dynarec_dump) {
+        printf_x86_instruction(my_context->dec, &dyn->insts[ninst].x86, name);
+        dynarec_log(LOG_NONE, "%s%p: %d emited opcodes, inst=%d, barrier=%d state=%d/%d(%d), %s=%X/%X, use=%X, need=%X/%X sm=%d/%d",
+            (box86_dynarec_dump>1)?"\e[32m":"",
+            (void*)(dyn->arm_start+dyn->insts[ninst].address),
+            dyn->insts[ninst].size/4,
+            ninst,
+            dyn->insts[ninst].x86.barrier,
+            dyn->insts[ninst].x86.state_flags,
+            dyn->f.pending,
+            dyn->f.dfnone,
+            dyn->insts[ninst].x86.may_set?"may":"set",
+            dyn->insts[ninst].x86.set_flags,
+            dyn->insts[ninst].x86.gen_flags,
+            dyn->insts[ninst].x86.use_flags,
+            dyn->insts[ninst].x86.need_before,
+            dyn->insts[ninst].x86.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].x86.jmp && dyn->insts[ninst].x86.jmp_insts>=0)
+            dynarec_log(LOG_NONE, ", jmp=%d", dyn->insts[ninst].x86.jmp_insts);
+        if(dyn->insts[ninst].x86.jmp && dyn->insts[ninst].x86.jmp_insts==-1)
+            dynarec_log(LOG_NONE, ", jmp=out");
+        for(int ii=0; ii<24; ++ii) {
+            switch(dyn->insts[ninst].n.neoncache[ii].t) {
+                case NEON_CACHE_ST_D: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;
+                case NEON_CACHE_ST_F: dynarec_log(LOG_NONE, " S%d:%s", (ii+8)*2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;
+                case NEON_CACHE_MM: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;
+                case NEON_CACHE_XMMW: dynarec_log(LOG_NONE, " Q%d:%s", (ii+8)/2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); ++ii; break;
+                case NEON_CACHE_XMMR: dynarec_log(LOG_NONE, " Q%d:%s", (ii+8)/2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); ++ii; break;
+                case NEON_CACHE_SCR: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;
+                case NEON_CACHE_NONE:
+                default:    break;
+            }
+        }
+        if(dyn->n.stack || dyn->insts[ninst].n.stack_next || dyn->insts[ninst].n.x87stack)
+            dynarec_log(LOG_NONE, " X87:%d/%d(+%d/-%d)%d", dyn->n.stack, dyn->insts[ninst].n.stack_next, dyn->insts[ninst].n.stack_push, dyn->insts[ninst].n.stack_pop, dyn->insts[ninst].n.x87stack);
+        if(dyn->insts[ninst].n.combined1 || dyn->insts[ninst].n.combined2)
+            dynarec_log(LOG_NONE, " %s:%d/%d", dyn->insts[ninst].n.swapped?"SWP":"CMB", dyn->insts[ninst].n.combined1, dyn->insts[ninst].n.combined2);
+        dynarec_log(LOG_NONE, "%s\n", (box86_dynarec_dump>1)?"\e[m":"");
+    }
+}
+
+void print_opcode(dynarec_arm_t* dyn, int ninst, uint32_t opcode)
+{
+    dynarec_log(LOG_NONE, "\t%08x\t%s\n", opcode, arm_print(opcode));
+}
+
+static void x87_reset(neoncache_t* n)
+{
+    for (int i=0; i<8; ++i)
+        n->x87cache[i] = -1;
+    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].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<8; ++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_arm_t* dyn, int ninst, int st)
+{
+    return (dyn->n.tags&(0b11<<(st*2)))?1:0;
 }
@@ -16,19 +16,29 @@ void arm_fscale(x86emu_t* emu);
 void arm_fbld(x86emu_t* emu, uint8_t* ed);
 void arm_fild64(x86emu_t* emu, int64_t* ed);
 void arm_fbst(x86emu_t* emu, uint8_t* ed);
-void arm_fistp64(x86emu_t* emu, int64_t* ed);
-void arm_fistt64(x86emu_t* emu, int64_t* ed);
 void arm_fld(x86emu_t* emu, uint8_t* ed);
 void arm_fsave(x86emu_t* emu, uint8_t* ed);
 void arm_frstor(x86emu_t* emu, uint8_t* ed);
 void arm_fprem1(x86emu_t* emu);

+void arm_aesd(x86emu_t* emu, int xmm);
+void arm_aese(x86emu_t* emu, int xmm);
+void arm_aesdlast(x86emu_t* emu, int xmm);
+void arm_aeselast(x86emu_t* emu, int xmm);
+void arm_aesimc(x86emu_t* emu, int xmm);
+void arm_aeskeygenassist(x86emu_t* emu, int gx, void* p, uint32_t u8);
+void arm_pclmul(x86emu_t* emu, int gx, void* p, uint32_t u8);
+
+void arm_clflush(x86emu_t* emu, void* p);
+
 int64_t arm_fist64_0(double d);
 int64_t arm_fist64_1(double d);
 int64_t arm_fist64_2(double d);
 int64_t arm_fist64_3(double d);

 void arm_ud(x86emu_t* emu);
+void arm_priv(x86emu_t* emu);
+void arm_singlestep(x86emu_t* emu);

 #define FPUFIRST    8

@@ -71,9 +81,6 @@ int CacheNeedsTransform(dynarec_arm_t* dyn, int i1);
 // Undo the changes of a neoncache to get the status before the instruction
 void neoncacheUnwind(neoncache_t* cache);

-// is inst clean for a son branch?
-int isInstClean(dynarec_arm_t* dyn, int ninst);
-
 // predecessor access
 int isPred(dynarec_arm_t* dyn, int ninst, int pred);
 int getNominalPred(dynarec_arm_t* dyn, int ninst);
@@ -82,10 +89,20 @@ int getNominalPred(dynarec_arm_t* dyn, int ninst);
 int getedparity(dynarec_arm_t* dyn, int ninst, uintptr_t addr, uint8_t nextop, int parity);
 // Do the GETED, but don't emit anything...
 uintptr_t fakeed(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop);
+uint8_t geted_ib(dynarec_arm_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_arm_t* dyn, uintptr_t addr, uintptr_t* calladdress, int* retn);

 const char* getCacheName(int t, int n);

+void inst_name_pass3(dynarec_arm_t* dyn, int ninst, const char* name);
+void print_opcode(dynarec_arm_t* dyn, int ninst, uint32_t opcode);
+
+// reset the cache
+void fpu_reset(dynarec_arm_t* dyn);
+void fpu_reset_ninst(dynarec_arm_t* dyn, int ninst);
+
+// is st freed
+int fpu_is_st_freed(dynarec_arm_t* dyn, int ninst, int st);
 #endif //__DYNAREC_ARM_FUNCTIONS_H__
@@ -27,6 +27,35 @@
 #define PKip(a) *(uint8_t*)(ip+a)
 #define PKa(a)  *(uint8_t*)(a)

+#define MODREG  ((nextop&0xC0)==0xC0)
+
+// Strong mem emulation helpers
+#define SMREAD_MIN  2
+#define SMWRITE_MIN 1
+// Sequence of Read will trigger a DMB on "first" read if strongmem is >= SMREAD_MIN
+// Sequence of Write will trigger a DMB on "last" write if strongmem is >= 1
+// All Write operation that might use a lock all have a memory barrier if strongmem is >= SMWRITE_MIN
+// Opcode will read
+#define SMREAD()    if((dyn->smread==0) && (box86_dynarec_strongmem>SMREAD_MIN)) {SMDMB();} else dyn->smread=1
+// Opcode will read with option forced lock
+#define SMREADLOCK(lock)    if((lock) || ((dyn->smread==0) && (box86_dynarec_strongmem>SMREAD_MIN))) {SMDMB();}
+// Opcode might read (depend on nextop)
+#define SMMIGHTREAD()   if(!MODREG) {SMREAD();}
+// Opcode has wrote
+#define SMWRITE()   dyn->smwrite=1
+// Opcode has wrote (strongmem>1 only)
+#define SMWRITE2()   if(box86_dynarec_strongmem>SMREAD_MIN) dyn->smwrite=1
+// Opcode has wrote with option forced lock
+#define SMWRITELOCK(lock)   if(lock || (box86_dynarec_strongmem>SMWRITE_MIN)) {SMDMB();} else dyn->smwrite=1
+// Opcode might have wrote (depend on nextop)
+#define SMMIGHTWRITE()   if(!MODREG) {SMWRITE();}
+// Start of sequence
+#define SMSTART()   SMEND()
+// End of sequence
+#define SMEND()     if(dyn->smwrite && box86_dynarec_strongmem) {if(box86_dynarec_strongmem){DSB_ISH();}else{DMB_ISH();}} dyn->smwrite=0; dyn->smread=0;
+// Force a Data memory barrier (for LOCK: prefix)
+#define SMDMB()     if(box86_dynarec_strongmem){DSB_ISH();}else{DMB_ISH();} dyn->smwrite=0; dyn->smread=1
+
 //LOCK_* define
 #define LOCK_LOCK   (int*)1

@@ -37,6 +66,7 @@
                    ed = xEAX+(nextop&7);   \
                    wback = 0;              \
                } else {                    \
+                    SMREAD();               \
                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0, 0, NULL); \
                    LDR_IMM9(x1, wback, fixedaddress); \
                    ed = x1;                \
@@ -46,6 +76,7 @@
                    ed = xEAX+(nextop&7);   \
                    wback = 0;              \
                } else {                    \
+                    SMREAD();               \
                    addr = geted(dyn, addr, ninst, nextop, &wback, (hint==x2)?x1:x2, &fixedaddress, 4095, 0, 0, NULL); \
                    LDR_IMM9(hint, wback, fixedaddress); \
                    ed = hint;              \
@@ -56,19 +87,21 @@
                    MOV_REG(ret, ed);       \
                    wback = 0;              \
                } else {                    \
+                    SMREAD();               \
                    addr = geted(dyn, addr, ninst, nextop, &wback, hint, &fixedaddress, 4095, 0, 0, NULL); \
                    ed = ret;               \
                    LDR_IMM9(ed, wback, fixedaddress); \
                }
 // Write back ed in wback (if wback not 0)
-#define WBACK       if(wback) {STR_IMM9(ed, wback, fixedaddress);}
-// Send back wb to either ed or wback
-#define SBACK(wb)   if(wback) {STR_IMM9(wb, wback, fixedaddress);} else {MOV_REG(ed, wb);}
+#define WBACK       if(wback) {STR_IMM9(ed, wback, fixedaddress); SMWRITE();}
+// Write back ed in wback (if wback not 0) (SMWRITE2 version)
+#define WBACK2      if(wback) {STR_IMM9(ed, wback, fixedaddress); SMWRITE2();}
 //GETEDO can use r1 for ed, and r2 for wback. wback is 0 if ed is xEAX..xEDI
 #define GETEDO(O)   if((nextop&0xC0)==0xC0) {   \
                    ed = xEAX+(nextop&7);   \
                    wback = 0;              \
                } else {                    \
+                    SMREAD();               \
                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 0, NULL); \
                    LDR_REG_LSL_IMM5(x1, wback, O, 0);  \
                    ed = x1;                 \
@@ -78,19 +111,22 @@
                    ed = xEAX+(nextop&7);               \
                    wback = 0;                          \
                } else {                                \
+                    SMREAD();                           \
                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 0, NULL); \
                    ADD_REG_LSL_IMM5(x2, wback, O, 0);  \
                    if(wback != x2) wback = x2;         \
                    LDR_IMM9(x1, wback, 0);             \
                    ed = x1;                            \
                }
-#define WBACKO(O)   if(wback) {STR_REG_LSL_IMM5(ed, wback, O, 0);}
+#define WBACKO(O)   if(wback) {STR_REG_LSL_IMM5(ed, wback, O, 0); SMWRITE();}
 //FAKEELike GETED, but doesn't get anything
 #define FAKEED  if((nextop&0xC0)!=0xC0) {   \
                    addr = fakeed(dyn, addr, ninst, nextop); \
                }
 // GETGW extract x86 register in gd, that is i
 #define GETGW(i) gd = xEAX+((nextop&0x38)>>3); UXTH(i, gd, 0); gd = i;
+// GETSGW extract x86 signed register in gd, that is i
+#define GETSGW(i) gd = xEAX+((nextop&0x38)>>3); SXTH(i, gd, 0); gd = i;
 //GETEWW will use i for ed, and can use w for wback.
 #define GETEWW(w, i) if((nextop&0xC0)==0xC0) {  \
                    wback = xEAX+(nextop&7);\
@@ -98,6 +134,7 @@
                    ed = i;                 \
                    wb1 = 0;                \
                } else {                    \
+                    SMREAD();               \
                    addr = geted(dyn, addr, ninst, nextop, &wback, w, &fixedaddress, 255, 0, 0, NULL); \
                    LDRH_IMM8(i, wback, fixedaddress); \
                    ed = i;                 \
@@ -110,6 +147,7 @@
                    ed = i;                 \
                    wb1 = 0;                \
                } else {                    \
+                    SMREAD();               \
                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL); \
                    LDRH_IMM8(i, wback, fixedaddress); \
                    ed = i;                 \
@@ -122,15 +160,16 @@
                    ed = i;                 \
                    wb1 = 0;                \
                } else {                    \
+                    SMREAD();               \
                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL); \
                    LDRSH_IMM8(i, wback, fixedaddress);\
                    ed = i;                 \
                    wb1 = 1;                \
                }
 // Write ed back to original register / memory
-#define EWBACK   if(wb1) {STRH_IMM8(ed, wback, fixedaddress);} else {BFI(wback, ed, 0, 16);}
+#define EWBACK   if(wb1) {STRH_IMM8(ed, wback, fixedaddress); SMWRITE();} else {BFI(wback, ed, 0, 16);}
 // Write w back to original register / memory
-#define EWBACKW(w)   if(wb1) {STRH_IMM8(w, wback, fixedaddress);} else {BFI(wback, w, 0, 16);}
+#define EWBACKW(w)   if(wb1) {STRH_IMM8(w, wback, fixedaddress); SMWRITE();} else {BFI(wback, w, 0, 16);}
 // Write back gd in correct register
 #define GWBACK       BFI((xEAX+((nextop&0x38)>>3)), gd, 0, 16);
 //GETEB will use i for ed, and can use r3 for wback.
@@ -142,6 +181,7 @@
                    wb1 = 0;                \
                    ed = i;                 \
                } else {                    \
+                    SMREAD();               \
                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095, 0, 0, NULL); \
                    LDRB_IMM9(i, wback, fixedaddress); \
                    wb1 = 1;                \
@@ -156,6 +196,7 @@
                    wb1 = 0;                \
                    ed = i;                 \
                } else {                    \
+                    SMREAD();               \
                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0, 0, 0, NULL); \
                    ADD_REG_LSL_IMM5(x3, wback, i, 0);  \
                    if(wback!=x3) wback = x3;           \
@@ -172,13 +213,16 @@
                    wb1 = 0;                \
                    ed = i;                 \
                } else {                    \
+                    SMREAD();               \
                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL); \
                    LDRSB_IMM8(i, wback, fixedaddress);\
                    wb1 = 1;                \
                    ed = i;                 \
                }
 // Write eb (ed) back to original register / memory
-#define EBBACK   if(wb1) {STRB_IMM9(ed, wback, fixedaddress);} else {BFI(wback, ed, wb2*8, 8);}
+#define EBBACK   if(wb1) {STRB_IMM9(ed, wback, fixedaddress); SMWRITE();} else {BFI(wback, ed, wb2*8, 8);}
+// Write eb (ed) back to original register / memory (SMWRITE2 version)
+#define EBBACK2   if(wb1) {STRB_IMM9(ed, wback, fixedaddress); SMWRITE2();} else {BFI(wback, ed, wb2*8, 8);}
 //GETGB will use i for gd
 #define GETGB(i)    gd = (nextop&0x38)>>3;  \
                    gb2 = ((gd&4)>>2);      \
@@ -218,22 +262,26 @@
 #define CALL_1DD(F, F2, M) call_d(dyn, ninst, F, F2, 1, x3, -1, M, 0)
 // CALL_1D_U64 will use S as scratch. Return value in ret/ret2, 1 ARG in D0
 #define CALL_1DR_U64(R, ret, ret2, S, M) call_dr(dyn, ninst, R, 1, S, ret, ret2, M, 1)
-// CALL_1D will use S as scratch. Return value in D0, 1 ARG in D0
+// CALL_1DR will use S as scratch. Return value in D0, 1 ARG in D0
 #define CALL_1DR(R, S, M) call_dr(dyn, ninst, R, 1, S, -1, -1, M, 0)
+// CALL_1DI will use S as scratch. Return value in D0, 1 ARG in D0, 1 ARG in R0
+#define CALL_1DDR(F, R, S, M) call_ddr(dyn, ninst, F, NULL, R, S, -1, M, 0)
 // CALL_1RD will use S as scratch. Return value in D0, 1 ARG in R
 #define CALL_1RD(F, R, S, M) call_rd(dyn, ninst, F, R, S, M, 0);

-#define MARK    if(dyn->insts) {dyn->insts[ninst].mark = (uintptr_t)dyn->arm_size;}
-#define GETMARK ((dyn->insts)?dyn->insts[ninst].mark:(dyn->arm_size+4))
-#define MARK2   if(dyn->insts) {dyn->insts[ninst].mark2 = (uintptr_t)dyn->arm_size;}
-#define GETMARK2 ((dyn->insts)?dyn->insts[ninst].mark2:(dyn->arm_size+4))
-#define MARK3   if(dyn->insts) {dyn->insts[ninst].mark3 = (uintptr_t)dyn->arm_size;}
-#define GETMARK3 ((dyn->insts)?dyn->insts[ninst].mark3:(dyn->arm_size+4))
-#define MARKF   if(dyn->insts) {dyn->insts[ninst].markf = (uintptr_t)dyn->arm_size;}
-#define GETMARKF ((dyn->insts)?dyn->insts[ninst].markf:(dyn->arm_size+4))
-#define MARKSEG if(dyn->insts) {dyn->insts[ninst].markseg = (uintptr_t)dyn->arm_size;}
-#define GETMARKSEG ((dyn->insts)?dyn->insts[ninst].markseg:(dyn->arm_size+4))
-#define MARKLOCK if(dyn->insts) {dyn->insts[ninst].marklock = (uintptr_t)dyn->arm_size;}
+#define MARK        if(dyn->insts) {dyn->insts[ninst].mark = (uintptr_t)dyn->arm_size;}
+#define GETMARK     ((dyn->insts)?dyn->insts[ninst].mark:(dyn->arm_size+4))
+#define MARK2       if(dyn->insts) {dyn->insts[ninst].mark2 = (uintptr_t)dyn->arm_size;}
+#define GETMARK2    ((dyn->insts)?dyn->insts[ninst].mark2:(dyn->arm_size+4))
+#define MARK3       if(dyn->insts) {dyn->insts[ninst].mark3 = (uintptr_t)dyn->arm_size;}
+#define GETMARK3    ((dyn->insts)?dyn->insts[ninst].mark3:(dyn->arm_size+4))
+#define MARKF       if(dyn->insts) {dyn->insts[ninst].markf = (uintptr_t)dyn->arm_size;}
+#define GETMARKF    ((dyn->insts)?dyn->insts[ninst].markf:(dyn->arm_size+4))
+#define MARKF2      if(dyn->insts) {dyn->insts[ninst].markf2 = (uintptr_t)dyn->arm_size;}
+#define GETMARKF2   ((dyn->insts)?dyn->insts[ninst].markf2:(dyn->arm_size+4))
+#define MARKSEG     if(dyn->insts) {dyn->insts[ninst].markseg = (uintptr_t)dyn->arm_size;}
+#define GETMARKSEG  ((dyn->insts)?dyn->insts[ninst].markseg:(dyn->arm_size+4))
+#define MARKLOCK    if(dyn->insts) {dyn->insts[ninst].marklock = (uintptr_t)dyn->arm_size;}
 #define GETMARKLOCK ((dyn->insts)?dyn->insts[ninst].marklock:(dyn->arm_size+4))

 // Branch to MARK if cond (use j32)
@@ -268,7 +316,7 @@
 #define IFXN(A, B)  if((dyn->insts[ninst].x86.gen_flags&(A) && !(dyn->insts[ninst].x86.gen_flags&(B))))

 // Generate FCOM with s1 and s2 scratch regs (the VCMP is already done)
-#define FCOM(s1, s2)    \
+#define FCOM(s1, s2)                                                            \
    VMRS_APSR();    /* 0b0100011100000000 */                                    \
    LDRH_IMM8(s1, xEmu, offsetof(x86emu_t, sw));   /*offset is 8bits right?*/   \
    BIC_IMM8(s1, s1, 0b01000111, 12);                                           \
@@ -279,7 +327,7 @@
    STRH_IMM8(s1, xEmu, offsetof(x86emu_t, sw))

 // Generate FCOMI with s1 and s2 scratch regs (the VCMP is already done)
-#define FCOMI(s1, s2)    \
+#define FCOMI(s1, s2)                                                       \
    IFX(X_CF|X_PF|X_ZF|X_PEND) {                                            \
        VMRS_APSR();    /* 0b111 */                                         \
        BIC_IMM8(xFlags, xFlags, 0b1000101, 0);                             \
@@ -287,16 +335,55 @@
        ORR_IMM8_COND(cEQ, xFlags, xFlags, 0b01000000, 0); /* zero */       \
        ORR_IMM8_COND(cMI, xFlags, xFlags, 0b00000001, 0); /* less than */  \
        /* greater than leave 0 */                                          \
-    }                                                                       \
-    SET_DFNONE(s1);                                                         \
-    IFX(X_OF|X_AF|X_SF|X_PEND) {                                            \
        /* clear 0b100010010000 */                                          \
        BIC_IMM8(xFlags, xFlags, 0b10001001, 14);                           \
    }                                                                       \
+    SET_DFNONE(s1);                                                         \
+
+
+#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(box86_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(box86_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(box86_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) {MOVW(S, d_none); STR_IMM9(S, xEmu, offsetof(x86emu_t, df)); dyn->f.dfnone=1;}
-#define SET_DF(S, N)     if(N) {MOVW(S, N); STR_IMM9(S, xEmu, offsetof(x86emu_t, df)); dyn->f.dfnone=0;} else SET_DFNONE(S)
+#define SET_DF(S, N)     \
+    if(N) {             \
+        MOVW(S, N);\
+        STR_IMM9(S, xEmu, offsetof(x86emu_t, df));  \
+        if(dyn->f.pending==SF_PENDING && dyn->insts[ninst].x86.need_after && !(dyn->insts[ninst].x86.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

@@ -306,7 +393,7 @@

 #ifndef READFLAGS
 #define READFLAGS(A) \
-    if((dyn->f.pending!=SF_SET)                         \
+    if(((A)!=X_PEND && dyn->f.pending!=SF_SET)          \
    && (dyn->f.pending!=SF_SET_PENDING)) {              \
        if(dyn->f.pending!=SF_PENDING) {                \
            LDR_IMM9(x3, xEmu, offsetof(x86emu_t, df)); \
@@ -330,6 +417,8 @@
    if(dyn->insts[ninst].x86.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:                                                                    \
@@ -338,7 +427,7 @@
    } else dyn->f.pending = SF_SET
 #endif
 #ifndef JUMP
-#define JUMP(A, C) 
+#define JUMP(A, C) SMEND()
 #endif
 #ifndef BARRIER
 #define BARRIER(A) 
@@ -346,6 +435,9 @@
 #ifndef BARRIER_NEXT
 #define BARRIER_NEXT(A)
 #endif
+#ifndef SET_HASCALLRET
+#define SET_HASCALLRET()
+#endif
 #define UFLAG_OP1(A) if(dyn->insts[ninst].x86.gen_flags) {STR_IMM9(A, xEmu, offsetof(x86emu_t, op1));}
 #define UFLAG_OP2(A) if(dyn->insts[ninst].x86.gen_flags) {STR_IMM9(A, xEmu, offsetof(x86emu_t, op2));}
 #define UFLAG_OP12(A1, A2) if(dyn->insts[ninst].x86.gen_flags) {STR_IMM9(A1, xEmu, offsetof(x86emu_t, op1));STR_IMM9(A2, 0, offsetof(x86emu_t, op2));}
@@ -404,6 +496,7 @@ void* arm_next(x86emu_t* emu, uintptr_t addr);
 #define iret_to_epilog  STEPNAME(iret_to_epilog_)
 #define call_c          STEPNAME(call_c_)
 #define call_d          STEPNAME(call_d_)
+#define call_ddr        STEPNAME(call_ddr_)
 #define call_dr         STEPNAME(call_dr_)
 #define call_rd         STEPNAME(call_rd_)
 #define grab_fsdata     STEPNAME(grab_fsdata_)
@@ -473,10 +566,24 @@ void* arm_next(x86emu_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_rol8c      STEPNAME(emit_rol8c)
 #define emit_ror32c     STEPNAME(emit_ror32c)
 #define emit_ror8c      STEPNAME(emit_ror8c)
+#define emit_rol16c     STEPNAME(emit_rol16c)
+#define emit_ror16c     STEPNAME(emit_ror16c)
 #define emit_shrd32c    STEPNAME(emit_shrd32c)
 #define emit_shld32c    STEPNAME(emit_shld32c)
 #define emit_shrd32     STEPNAME(emit_shrd32)
@@ -492,28 +599,34 @@ void* arm_next(x86emu_t* emu, uintptr_t addr);
 #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_free        STEPNAME(x87_free)
 #define x87_refresh     STEPNAME(x87_refresh)
 #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_setround    STEPNAME(x87_setround)
+#define x87_setround_reset    STEPNAME(x87_setround_reset)
 #define x87_restoreround STEPNAME(x87_restoreround)
 #define x87_swapreg     STEPNAME(x87_swapreg)
 #define sse_setround    STEPNAME(sse_setround)
+#define sse_setround_reset    STEPNAME(sse_setround_reset)
 #define mmx_get_reg     STEPNAME(mmx_get_reg)
 #define mmx_get_reg_empty STEPNAME(mmx_get_reg_empty)
 #define sse_get_reg     STEPNAME(sse_get_reg)
 #define sse_get_reg_empty STEPNAME(sse_get_reg_empty)
+#define sse_forget_reg  STEPNAME(sse_forget_reg)
+#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)
 #define x87_purgecache  STEPNAME(x87_purgecache)
 #define mmx_purgecache  STEPNAME(mmx_purgecache)
-#ifdef HAVE_TRACE
 #define fpu_reflectcache STEPNAME(fpu_reflectcache)
-#endif
+#define fpu_unreflectcache STEPNAME(fpu_unreflectcache)

 // get the single reg that from the double "reg" (so Dx[idx])
 #define fpu_get_single_reg      STEPNAME(fpu_get_single_reg)
@@ -538,6 +651,7 @@ void iret_to_epilog(dynarec_arm_t* dyn, int ninst);
 void call_c(dynarec_arm_t* dyn, int ninst, void* fnc, int reg, int ret, uint32_t mask, int saveflags);
 void call_d(dynarec_arm_t* dyn, int ninst, void* fnc, void* fnc2, int n, int reg, int ret, uint32_t mask, int saveflags);
 void call_dr(dynarec_arm_t* dyn, int ninst, int reg, int n, int s1, int ret, int ret2, uint32_t mask, int saveflags);
+void call_ddr(dynarec_arm_t* dyn, int ninst, void* fnc, void* fnc2, int arg, int reg, int ret, uint32_t mask, int saveflags);
 void call_rd(dynarec_arm_t* dyn, int ninst, void* fnc, int reg, int s1, uint32_t mask, int saveflags);
 void grab_fsdata(dynarec_arm_t* dyn, uintptr_t addr, int ninst, int reg);
 void grab_tlsdata(dynarec_arm_t* dyn, uintptr_t addr, int ninst, int reg);
@@ -606,20 +720,36 @@ void emit_shl32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
 void emit_shr32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
 void emit_shr32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
 void emit_sar32c(dynarec_arm_t* dyn, int ninst, int s1, int32_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, int32_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, int32_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, int32_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, int32_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, int32_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, int32_t c, int s3, int s4);
 void emit_rol32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
 void emit_rol8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
 void emit_ror32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
 void emit_ror8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
+void emit_rol16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
+void emit_ror16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
 void emit_shrd32c(dynarec_arm_t* dyn, int ninst, int s1, int s2, int32_t c, int s3, int s4);
 void emit_shld32c(dynarec_arm_t* dyn, int ninst, int s1, int s2, int32_t c, int s3, int s4);
 void emit_shrd32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
 void emit_shld32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);

-void emit_pf(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4);
+void emit_pf(dynarec_arm_t* dyn, int ninst, int s1, 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);
+int x87_stackcount(dynarec_arm_t* dyn, int ninst, int scratch);
+// restore the local stack counter
+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)
@@ -636,6 +766,8 @@ 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);
+// 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 (nothing done if value is not cached)
 void x87_refresh(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st);
 // refresh a value from the cache ->emu and then forget the cache (nothing done if value is not cached)
@@ -644,12 +776,16 @@ void x87_forget(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st);
 void x87_reget_st(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st);
 // Set rounding according to cw flags, return reg to restore flags
 int x87_setround(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
+// Set rounding according to cw flags, return reg to restore flags, also enable exceptions and reset counters
+int x87_setround_reset(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
 // Restore round flag
 void x87_restoreround(dynarec_arm_t* dyn, int ninst, int s1);
 // swap 2 x87 regs
 void x87_swapreg(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a, int b);
 // Set rounding according to mxcsr flags, return reg to restore flags
 int sse_setround(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
+// Set rounding according to mxcsr flags, return reg to restore flags, also enable exceptions and reset counters
+int sse_setround_reset(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);

@@ -694,17 +830,22 @@ int mmx_get_reg_empty(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int
 int sse_get_reg(dynarec_arm_t* dyn, int ninst, int s1, int a, int forwrite);
 // get neon register for a SSE reg, but don't try to synch it if it needed to be created
 int sse_get_reg_empty(dynarec_arm_t* dyn, int ninst, int s1, int a);
+// forget an SSE reg
+void sse_forget_reg(dynarec_arm_t* dyn, int ninst, int a, int s1);
+// sync reg with emu for sse reg (return 1 if reflect was done, s1 = &xmm[a] then)
+int sse_reflect_reg(dynarec_arm_t* dyn, int ninst, int a, int s1);

 // 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
+void fpu_propagate_stack(dynarec_arm_t* dyn, int ninst);
 // purge the FPU cache (needs 3 scratch registers) next=1 if for a conditionnal branch jumping out of block (no tracking updated)
 void fpu_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1, int s2, int s3);
 void x87_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1, int s2, int s3);
 void mmx_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1);
-#ifdef HAVE_TRACE
 void fpu_reflectcache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
-#endif
+void fpu_unreflectcache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
 void fpu_pushcache(dynarec_arm_t* dyn, int ninst, int s1);
 void fpu_popcache(dynarec_arm_t* dyn, int ninst, int s1);

@@ -751,4 +892,25 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
 #define MAYUSE(A)   
 #endif

+#define NOTEST(s1, s2)                                      \
+    if(box86_dynarec_test) {                                \
+        MOVW(s1, 0);                                        \
+        MOVW(s2, offsetof(x86emu_t, test.test));            \
+        STR_REG_LSL_IMM5(s1, xEmu, s2, 0);                  \
+        ADD_IMM8(s2, s2, offsetof(x86emu_t, test.clean)-offsetof(x86emu_t, test.test));\
+        STR_REG_LSL_IMM5(s1, xEmu, s2, 0);                  \
+    }
+#define SKIPTEST(s1, s2)                                    \
+    if(box86_dynarec_test) {                                \
+        MOVW(s1, 0);                                        \
+        MOVW(s2, offsetof(x86emu_t, test.clean));           \
+        STR_REG_LSL_IMM5(s1, xEmu, s2, 0);                  \
+    }
+#define GOTEST(s1, s2)                                      \
+    if(box86_dynarec_test) {                                \
+        MOVW(s2, 1);                                        \
+        MOVW(s1, offsetof(x86emu_t, test.test));            \
+        STR_REG_LSL_IMM5(s2, xEmu, s1, 0);                  \
+    }
+
 #endif //__DYNAREC_ARM_HELPER_H__
@@ -0,0 +1,12 @@
+#include <stdint.h>
+#include <stdlib.h>
+
+#include "arm_emitter.h"
+
+#define EMIT(A) *block = (A); ++block
+void CreateJmpNext(void* addr, void* next)
+{
+    uint32_t* block = (uint32_t*)addr;
+    LDR_literal(x2, ((intptr_t)next - (intptr_t)addr)-8);
+    BX(x2);
+}
@@ -21,11 +21,18 @@
 #include "dynarec_arm_functions.h"
 #include "dynarec_arm_helper.h"
 #include "custommem.h"
+#include "elfloader.h"

 #ifndef STEP
 #error No STEP defined
 #endif

+#if STEP == 0
+#ifndef PROT_READ
+#define PROT_READ 0x1
+#endif
+#endif
+
 uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
 {
    int ok = 1;
@@ -35,16 +42,35 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
    uintptr_t init_addr = addr;
    MAYUSE(init_addr);
    int need_epilog = 1;
-    dyn->sons_size = 0;
    // Clean up (because there are multiple passes)
    dyn->f.pending = 0;
    dyn->f.dfnone = 0;
+    dyn->forward = 0;
+    dyn->forward_to = 0;
+    dyn->forward_size = 0;
+    dyn->forward_ninst = 0;
+    #if STEP == 0
+    memset(&dyn->insts[ninst], 0, sizeof(instruction_arm_t));
+    #endif
    fpu_reset(dyn);
    int reset_n = -1;
-    // ok, go now
+    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    // ok, go now
    INIT;
+    #if STEP == 0
+    uintptr_t cur_page = (addr)&~box86_pagesize;
+    #endif
    while(ok) {
+        #if STEP == 0
+        if(cur_page != (addr)&~box86_pagesize) {
+            cur_page = (addr)&~box86_pagesize;
+            if(!(getProtection(addr)&PROT_READ)) {
+                need_epilog = 1;
+                break;
+            }
+        }
+        #endif
        ip = addr;
+        fpu_propagate_stack(dyn, ninst);
        if (reset_n!=-1) {
            if(reset_n==-2) {
                MESSAGE(LOG_DEBUG, "Reset Caches to zero\n");
@@ -52,44 +78,7 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
                dyn->f.pending = 0;
                fpu_reset(dyn);
            } else {
-                MESSAGE(LOG_DEBUG, "Reset Caches with %d\n",reset_n);
-                #if STEP > 1
-                // for STEP 2 & 3, just need to refrest with current, and undo the changes (push & swap)
-                dyn->n = dyn->insts[ninst].n;
-                neoncacheUnwind(&dyn->n);
-                #ifdef HAVE_TRACE
-                if(box86_dynarec_dump)
-                    if(memcmp(&dyn->n, &dyn->insts[reset_n].n, sizeof(neon_cache_t))) {
-                        MESSAGE(LOG_DEBUG, "Warning, difference in neoncache: reset=");
-                        for(int i=0; i<24; ++i)
-                            if(dyn->insts[reset_n].n.neoncache[i].v)
-                                MESSAGE(LOG_DEBUG, " %02d:%s", i, getCacheName(dyn->insts[reset_n].n.neoncache[i].t, dyn->insts[reset_n].n.neoncache[i].n));
-                        if(dyn->insts[reset_n].n.combined1 || dyn->insts[reset_n].n.combined2)
-                            MESSAGE(LOG_DEBUG, " %s:%02d/%02d", dyn->insts[reset_n].n.swapped?"SWP":"CMB", dyn->insts[reset_n].n.combined1, dyn->insts[reset_n].n.combined2);
-                        if(dyn->insts[reset_n].n.stack_push || dyn->insts[reset_n].n.stack_pop)
-                            MESSAGE(LOG_DEBUG, " (%d:%d)", dyn->insts[reset_n].n.stack_push, -dyn->insts[reset_n].n.stack_pop);
-                        MESSAGE(LOG_DEBUG, " ==> ");
-                        for(int i=0; i<24; ++i)
-                            if(dyn->insts[ninst].n.neoncache[i].v)
-                                MESSAGE(LOG_DEBUG, " %02d:%s", i, getCacheName(dyn->insts[ninst].n.neoncache[i].t, dyn->insts[ninst].n.neoncache[i].n));
-                        if(dyn->insts[ninst].n.combined1 || dyn->insts[ninst].n.combined2)
-                            MESSAGE(LOG_DEBUG, " %s:%02d/%02d", dyn->insts[ninst].n.swapped?"SWP":"CMB", dyn->insts[ninst].n.combined1, dyn->insts[ninst].n.combined2);
-                        if(dyn->insts[ninst].n.stack_push || dyn->insts[ninst].n.stack_pop)
-                            MESSAGE(LOG_DEBUG, " (%d:%d)", dyn->insts[ninst].n.stack_push, -dyn->insts[ninst].n.stack_pop);
-                        MESSAGE(LOG_DEBUG, " -> ");
-                        for(int i=0; i<24; ++i)
-                            if(dyn->n.neoncache[i].v)
-                                MESSAGE(LOG_DEBUG, " %02d:%s", i, getCacheName(dyn->n.neoncache[i].t, dyn->n.neoncache[i].n));
-                        if(dyn->n.combined1 || dyn->n.combined2)
-                            MESSAGE(LOG_DEBUG, " %s:%02d/%02d", dyn->n.swapped?"SWP":"CMB", dyn->n.combined1, dyn->n.combined2);
-                        if(dyn->n.stack_push || dyn->n.stack_pop)
-                            MESSAGE(LOG_DEBUG, " (%d:%d)", dyn->n.stack_push, -dyn->n.stack_pop);
-                        MESSAGE(LOG_DEBUG, "\n");
-                    }
-                #endif //HAVE_TRACE
-                #else
-                dyn->n = dyn->insts[reset_n].n;
-                #endif
+                fpu_reset_cache(dyn, ninst, reset_n);
                dyn->f = dyn->insts[reset_n].f_exit;
                if(dyn->insts[ninst].x86.barrier&BARRIER_FLOAT) {
                    MESSAGE(LOG_DEBUG, "Apply Barrier Float\n");
@@ -103,28 +92,33 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
            }
            reset_n = -1;
        }
-        // propagate ST stack state, especial stack pop that are defered
-        if(dyn->n.stack_pop) {
-            for(int j=0; j<24; ++j)
-                if((dyn->n.neoncache[j].t == NEON_CACHE_ST_D || dyn->n.neoncache[j].t == NEON_CACHE_ST_F)) {
-                    if(dyn->n.neoncache[j].n<dyn->n.stack_pop)
-                        dyn->n.neoncache[j].v = 0;
-                    else
-                        dyn->n.neoncache[j].n-=dyn->n.stack_pop;
-                }
-            dyn->n.stack_pop = 0;
-        }
-        dyn->n.stack = dyn->n.stack_next;
-        dyn->n.news = 0;
-        dyn->n.stack_push = 0;
-        dyn->n.swapped = 0;
        NEW_INST;
+        #if STEP == 0
+        if(ninst && dyn->insts[ninst-1].x86.barrier_next) {
+            BARRIER(dyn->insts[ninst-1].x86.barrier_next);
+        }
+        #endif
+        if(!ninst) {
+            GOTEST(x1, x2);
+        }
+        if(dyn->insts[ninst].pred_sz>1) {SMSTART();}
        fpu_reset_scratch(dyn);
        if((dyn->insts[ninst].x86.need_before&~X_PEND) && !dyn->insts[ninst].pred_sz) {
            READFLAGS(dyn->insts[ninst].x86.need_before&~X_PEND);
        }
+        if(box86_dynarec_test)  {
+            MESSAGE(LOG_DUMP, "TEST INIT  ----\n");
+            fpu_reflectcache(dyn, ninst, x1, x2, x3);
+            MOV32(x1, ip);
+            STM(xEmu, (1<<xEAX)|(1<<xEBX)|(1<<xECX)|(1<<xEDX)|(1<<xESI)|(1<<xEDI)|(1<<xESP)|(1<<xEBP));
+            STR_IMM9(x1, xEmu, offsetof(x86emu_t, ip));
+            MOVW(x2, 1);
+            CALL(x86test_init, -1, 0);
+            fpu_unreflectcache(dyn, ninst, x1, x2, x3);
+            MESSAGE(LOG_DUMP, "----------\n");
+        }
 #ifdef HAVE_TRACE
-        if(my_context->dec && box86_dynarec_trace) {
+        else if(my_context->dec && box86_dynarec_trace) {
        if((trace_end == 0) 
            || ((ip >= trace_start) && (ip < trace_end)))  {
                MESSAGE(LOG_DUMP, "TRACE ----\n");
@@ -134,6 +128,7 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
                STR_IMM9(x1, xEmu, offsetof(x86emu_t, ip));
                MOVW(x2, 1);
                CALL(PrintTrace, -1, 0);
+                fpu_unreflectcache(dyn, ninst, x1, x2, x3);
                MESSAGE(LOG_DUMP, "----------\n");
            }
        }
@@ -147,7 +142,6 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
        #if STEP > 0
        if(!dyn->insts[ninst].x86.has_next && dyn->insts[ninst].x86.jmp && dyn->insts[ninst].x86.jmp_insts!=-1)
            next = dyn->insts[ninst].x86.jmp_insts;
-        #endif
        if(dyn->insts[ninst].x86.has_next && dyn->insts[next].x86.barrier) {
            if(dyn->insts[next].x86.barrier&BARRIER_FLOAT)
                fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
@@ -156,13 +150,27 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
                dyn->f.dfnone = 0;
            }
        }
+        #endif
+        #ifndef PROT_READ
+        #define PROT_READ 1
+        #endif
        #if STEP != 0
        if(!ok && !need_epilog && (addr < (dyn->start+dyn->isize))) {
            ok = 1;
            // we use the 1st predecessor here
            int ii = ninst+1;
-            while(ii<dyn->size && !dyn->insts[ii].pred_sz) 
-                ++ii;
+            if(ii<dyn->size && !dyn->insts[ii].x86.alive) {
+                while(ii<dyn->size && !dyn->insts[ii].x86.alive) {
+                    // may need to skip opcodes to advance
+                    ++ninst;
+                    NEW_INST;
+                    MESSAGE(LOG_DEBUG, "Skipping unused opcode\n");
+                    INST_NAME("Skipped opcode");
+                    INST_EPILOG;
+                    addr += dyn->insts[ii].x86.size;
+                    ++ii;
+                }
+            }
            if((dyn->insts[ii].x86.barrier&BARRIER_FULL)==BARRIER_FULL)
                reset_n = -2;    // hack to say Barrier!
            else {
@@ -174,43 +182,123 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
            }
        }
        #else
-        #ifndef PROT_READ
-        #define PROT_READ 1
-        #endif
-        if(!ok && !need_epilog && box86_dynarec_bigblock && (getProtection(addr+3)&PROT_READ))
+        // 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(box86_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 && ok>=0) {
+                // we made it!
+                reset_n = get_first_jump(dyn, addr);
+                if(box86_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].x86.has_callret?"(opt. call) ":"", (void*)dyn->forward, (void*)dyn->forward_to, reset_n, ninst);
+                if(dyn->insts[dyn->forward_ninst].x86.has_callret && !dyn->insts[dyn->forward_ninst].x86.has_next)
+                    dyn->insts[dyn->forward_ninst].x86.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<=0) {
+                // something when wrong! rollback
+                if(box86_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;
+                dyn->size = dyn->forward_size;
+                ninst = dyn->forward_ninst;
+                addr = dyn->forward;
+                dyn->forward = 0;
+                dyn->forward_to = 0;
+                dyn->forward_size = 0;
+                dyn->forward_ninst = 0;
+            }
+            // else just continue
+        } else if(!ok && !need_epilog && box86_dynarec_bigblock && (getProtection(addr+3)&~PROT_READ))
            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)) 
-                    && (((box86_dynarec_bigblock<2) && isJumpTableDefault((void*)next)) || (box86_dynarec_bigblock>1))
                    /*||(((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].x86.jmp == next) {
-                            reset_n = ii;
-                            ii=ninst;
+                    if(dyn->insts[ninst].x86.has_callret && !dyn->insts[ninst].x86.has_next) {
+                        dyn->insts[ninst].x86.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(box86_dynarec_dump) dynarec_log(LOG_NONE, "Extend block %p, %s%p -> %p (ninst=%d, jump from %d)\n", dyn, dyn->insts[ninst].x86.has_callret?"(opt. call) ":"", (void*)addr, (void*)next, ninst, dyn->insts[ninst].x86.has_callret?ninst:reset_n);
+                } else if(next && (next-addr)<box86_dynarec_forward && (getProtection(next)&PROT_READ)/*box86_dynarec_bigblock>=stopblock*/) {
+                    if(!((box86_dynarec_bigblock<stopblock) && !isJumpTableDefault((void*)next))) {
+                        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;
                        }
-                    if(box86_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);
-                } else if(next && (next-addr)<30) {
-                    if(box86_dynarec_dump) dynarec_log(LOG_NONE, "Cannot extend block %p -> %p (%02X %02X %02X %02X %02X %02X %02X %02x)\n", (void*)addr, (void*)next, PK(0), PK(1), PK(2), PK(3), PK(4), PK(5), PK(6), PK(7));
+                    }
                }
            }
        #endif
-        if(ok<0)  {ok = 0; need_epilog=1;}
+        if(ok<0)  {
+            ok = 0; need_epilog=1; 
+            #if STEP == 0
+            if(ninst) {
+                --ninst;
+                if(!dyn->insts[ninst].x86.barrier) {
+                    BARRIER(BARRIER_FLOAT);
+                }
+                dyn->insts[ninst].x86.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].x86.has_callret)
+            reset_n = -2;
        ++ninst;
        #if STEP == 0
-        if((ok && !isJumpTableDefault((void*)addr) && (box86_dynarec_bigblock<2)) ||
-            (ok && dyn->size >= 2000))
+        memset(&dyn->insts[ninst], 0, sizeof(instruction_arm_t));
+        if((ok>0) && (((box86_dynarec_bigblock<stopblock) && !isJumpTableDefault((void*)addr)) 
+            || (addr>=box86_nodynarec_start && addr<box86_nodynarec_end)))
        #else
-        if(ok && (ninst==dyn->size))
+        if((ok>0) && (ninst==dyn->size))
        #endif
        {
+            #if STEP == 0
+            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
            int j32;
            MAYUSE(j32);
-            MESSAGE(LOG_DEBUG, "Stopping block %p (%d / %d)\n",(void*)init_addr, ninst, dyn->size); 
+            MESSAGE(LOG_DEBUG, "Stopping block %p (%d / %d)\n",(void*)init_addr, ninst, dyn->size);
+            if(!box86_dynarec_dump && addr>=box86_nodynarec_start && addr<box86_nodynarec_end)
+                dynarec_log(LOG_INFO, "Stopping block in no-dynarec zone\n");
            --ninst;
            if(!dyn->insts[ninst].x86.barrier) {
                BARRIER(BARRIER_FLOAT);
@@ -219,16 +307,18 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
            dyn->insts[ninst].x86.need_after |= X_PEND;
            #endif
            ++ninst;
+            NOTEST(x1, x3);
            fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
            jump_to_next(dyn, addr, 0, ninst);
            ok=0; need_epilog=0;
        }
    }
    if(need_epilog) {
+        NOTEST(x1, x3);
        fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
        jump_to_epilog(dyn, ip, 0, ninst);  // no linker here, it's an unknow instruction
    }
    FINI;
-    MESSAGE(LOG_DUMP, "---- END OF BLOCK ---- (%d, %d sons)\n", dyn->size, dyn->sons_size);
+    MESSAGE(LOG_DUMP, "---- END OF BLOCK ---- (%d)\n", dyn->size);
    return addr;
 }
@@ -4,6 +4,7 @@
    dyn->isize = addr-sav_addr;\
    dyn->insts[ninst].x86.addr = addr;\
    if(ninst) dyn->insts[ninst-1].x86.size = dyn->insts[ninst].x86.addr - dyn->insts[ninst-1].x86.addr
+
 #define MESSAGE(A, ...)  
 #define MAYSETFLAGS()   dyn->insts[ninst].x86.may_set = 1
 #define READFLAGS(A)    \
@@ -11,20 +12,16 @@
        dyn->f.pending=SF_SET
 #define SETFLAGS(A,B)   \
        dyn->insts[ninst].x86.set_flags = A;    \
-        dyn->insts[ninst].x86.state_flags = B;  \
+        dyn->insts[ninst].x86.state_flags = (B)&~SF_DF;  \
        dyn->f.pending=(B)&SF_SET_PENDING;      \
-        dyn->f.dfnone=((B)&SF_SET)?1:0;
+        dyn->f.dfnone=((B)&SF_SET)?(((B)==SF_SET_NODF)?0:1):0;
 #define EMIT(A)     
-#define JUMP(A, C)      if((A)>addr) add_next(dyn, (uintptr_t)(A)); dyn->insts[ninst].x86.jmp = A; dyn->insts[ninst].x86.jmp_cond = C
+#define JUMP(A, C)      add_jump(dyn, ninst); add_next(dyn, (uintptr_t)A); SMEND(); dyn->insts[ninst].x86.jmp = A; dyn->insts[ninst].x86.jmp_cond = C; dyn->insts[ninst].x86.jmp_insts = 0
 #define BARRIER(A)      if(A!=BARRIER_MAYBE) {fpu_purgecache(dyn, ninst, 0, x1, x2, x3); dyn->insts[ninst].x86.barrier = A;} else dyn->insts[ninst].barrier_maybe = 1
-#define BARRIER_NEXT(A) dyn->insts[ninst+1].x86.barrier = A
+#define BARRIER_NEXT(A) dyn->insts[ninst].x86.barrier_next = A
+#define SET_HASCALLRET()    dyn->insts[ninst].x86.has_callret = 1
 #define NEW_INST \
        ++dyn->size;                            \
-        if(dyn->size+3>=dyn->cap) {             \
-                dyn->insts = (instruction_arm_t*)box_realloc(dyn->insts, sizeof(instruction_arm_t)*dyn->cap*2); \
-                memset(&dyn->insts[dyn->cap], 0, sizeof(instruction_arm_t)*dyn->cap);   \
-                dyn->cap *= 2;                  \
-        }                                       \
        dyn->insts[ninst].x86.addr = ip;        \
        dyn->n.combined1 = dyn->n.combined2 = 0;\
        dyn->n.swapped = 0; dyn->n.barrier = 0; \
@@ -34,12 +31,12 @@
 #define INST_EPILOG                             \
        dyn->insts[ninst].f_exit = dyn->f;      \
        dyn->insts[ninst].n = dyn->n;           \
-        dyn->insts[ninst].x86.has_next = ok;
+        dyn->insts[ninst].x86.has_next = (ok>0)?1:0;
 #define INST_NAME(name) 
 #define DEFAULT                         \
        --dyn->size;                    \
        *ok = -1;                       \
-        if(box86_dynarec_log>=LOG_INFO || box86_dynarec_dump) {\
+        if(box86_dynarec_log>=LOG_INFO || box86_dynarec_dump || box86_dynarec_missing) {\
        dynarec_log(LOG_NONE, "%p: Dynarec stopped @ %s (%s:%u) because of Opcode %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X", \
        (void*)ip, __FUNCTION__, __FILE__, __LINE__, PKip(0), \
        PKip(1), PKip(2), PKip(3),      \
@@ -2,7 +2,6 @@
 #define FINI
 #define MESSAGE(A, ...)  
 #define EMIT(A)     
-
 #define NEW_INST                                \
        dyn->insts[ninst].f_entry = dyn->f;     \
        dyn->n.combined1 = dyn->n.combined2 = 0;\
@@ -1,13 +1,16 @@
 #define INIT        dyn->arm_size = 0
-#define FINI        if(ninst) {dyn->insts[ninst].address = (dyn->insts[ninst-1].address+dyn->insts[ninst-1].size);}
+#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].x86.size>dyn->insts[ninst].size)?dyn->insts[ninst].x86.size:dyn->insts[ninst].size)/15; \
+        }

 #define MESSAGE(A, ...)  
-#define EMIT(A)     dyn->insts[ninst].size+=4; dyn->arm_size+=4
+#define EMIT(A)     do{dyn->insts[ninst].size+=4; dyn->arm_size+=4;}while(0)
 #define NEW_INST                                                                                        \
        if(ninst) {                                                                                     \
                dyn->insts[ninst].address = (dyn->insts[ninst-1].address+dyn->insts[ninst-1].size);     \
-                if(ninst && isInstClean(dyn, ninst))                                                    \
-                        ++dyn->sons_size;                                                               \
+                dyn->insts_size += 1+((dyn->insts[ninst-1].x86.size>dyn->insts[ninst-1].size)?dyn->insts[ninst-1].x86.size:dyn->insts[ninst-1].size)/15; \
        }
 #define INST_EPILOG dyn->insts[ninst].epilog = dyn->arm_size; 
 #define INST_NAME(name) 
@@ -1,64 +1,20 @@
 #define INIT    
-#define FINI
-#define EMIT(A)     \
-    if(box86_dynarec_dump) {dynarec_log(LOG_NONE, "\t%08x\t%s\n", (A), arm_print(A));} \
-    *(uint32_t*)(dyn->block) = A;       \
-    dyn->block += 4; dyn->arm_size += 4;\
-    dyn->insts[ninst].size2 += 4
+#define FINI        \
+    if(ninst)       \
+        addInst(dyn->instsize, &dyn->insts_size, dyn->insts[ninst].x86.size, dyn->insts[ninst].size/4); \
+    addInst(dyn->instsize, &dyn->insts_size, 0, 0);
+#define EMIT(A)                                         \
+    do{                                                 \
+        if(box86_dynarec_dump) print_opcode(dyn, ninst, (uint32_t)(A)); \
+        if((uintptr_t)dyn->block<(uintptr_t)dyn->next-sizeof(void*))\
+            *(uint32_t*)(dyn->block) = (uint32_t)(A);   \
+        dyn->block += 4; dyn->arm_size += 4;            \
+        dyn->insts[ninst].size2 += 4;                   \
+    }while(0)

 #define MESSAGE(A, ...)  if(box86_dynarec_dump) dynarec_log(LOG_NONE, __VA_ARGS__)
 #define NEW_INST        \
-    if(ninst && isInstClean(dyn, ninst)) {                      \
-        dyn->sons_x86[dyn->sons_size] = (uintptr_t)ip;          \
-        dyn->sons_arm[dyn->sons_size] = dyn->block;             \
-        MESSAGE(LOG_DUMP, "----> potential Son here %p/%p\n", (void*)ip, dyn->block);  \
-        ++dyn->sons_size;                                       \
-    }
-
+    if(ninst)                                                   \
+        addInst(dyn->instsize, &dyn->insts_size, dyn->insts[ninst-1].x86.size, dyn->insts[ninst-1].size/4);
 #define INST_EPILOG     
-#define INST_NAME(name) \
-    if(box86_dynarec_dump) {\
-        printf_x86_instruction(my_context->dec, &dyn->insts[ninst].x86, name); \
-        dynarec_log(LOG_NONE, "%s%p: %d emited opcodes, inst=%d, barrier=%d state=%d/%d(%d), %s=%X/%X, use=%X, need=%X/%X", \
-            (box86_dynarec_dump>1)?"\e[32m":"", \
-            (void*)(dyn->arm_start+dyn->insts[ninst].address),  \
-            dyn->insts[ninst].size/4,           \
-            ninst,                              \
-            dyn->insts[ninst].x86.barrier,      \
-            dyn->insts[ninst].x86.state_flags,  \
-            dyn->f.pending,                     \
-            dyn->f.dfnone,                      \
-            dyn->insts[ninst].x86.may_set?"may":"set",          \
-            dyn->insts[ninst].x86.set_flags,    \
-            dyn->insts[ninst].x86.gen_flags,    \
-            dyn->insts[ninst].x86.use_flags,    \
-            dyn->insts[ninst].x86.need_before,  \
-            dyn->insts[ninst].x86.need_after);  \
-        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].x86.jmp && dyn->insts[ninst].x86.jmp_insts>=0)\
-            dynarec_log(LOG_NONE, ", jmp=%d", dyn->insts[ninst].x86.jmp_insts);\
-        if(dyn->insts[ninst].x86.jmp && dyn->insts[ninst].x86.jmp_insts==-1)\
-            dynarec_log(LOG_NONE, ", jmp=out");\
-        for(int ii=0; ii<24; ++ii) {            \
-            switch(dyn->insts[ninst].n.neoncache[ii].t) {    \
-                case NEON_CACHE_ST_D: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;            \
-                case NEON_CACHE_ST_F: dynarec_log(LOG_NONE, " S%d:%s", (ii+8)*2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;        \
-                case NEON_CACHE_MM: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;              \
-                case NEON_CACHE_XMMW: dynarec_log(LOG_NONE, " Q%d:%s", (ii+8)/2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); ++ii; break;  \
-                case NEON_CACHE_XMMR: dynarec_log(LOG_NONE, " Q%d:%s", (ii+8)/2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); ++ii; break;  \
-                case NEON_CACHE_SCR: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;             \
-                case NEON_CACHE_NONE:           \
-                default:    break;              \
-            }                                   \
-        }                                       \
-        if(dyn->n.stack || dyn->insts[ninst].n.stack_next || dyn->insts[ninst].n.x87stack)     \
-            dynarec_log(LOG_NONE, " X87:%d/%d(+%d/-%d)%d", dyn->n.stack, dyn->insts[ninst].n.stack_next, dyn->insts[ninst].n.stack_push, dyn->insts[ninst].n.stack_pop, dyn->insts[ninst].n.x87stack); \
-        if(dyn->insts[ninst].n.combined1 || dyn->insts[ninst].n.combined2)                     \
-            dynarec_log(LOG_NONE, " %s:%d/%d", dyn->insts[ninst].n.swapped?"SWP":"CMB", dyn->insts[ninst].n.combined1, dyn->insts[ninst].n.combined2);   \
-        dynarec_log(LOG_NONE, "%s\n", (box86_dynarec_dump>1)?"\e[m":"");                       \
-    }
-
+#define INST_NAME(name) inst_name_pass3(dyn, ninst, name)
@@ -5,6 +5,7 @@

 typedef struct x86emu_s x86emu_t;
 typedef struct dynablock_s dynablock_t;
+typedef struct instsize_s instsize_t;

 #define BARRIER_MAYBE   8

@@ -40,10 +41,13 @@ 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[8];// cache status for the 8 SSE(2) registers
    int8_t              fpuused[24];// all 8..31 double reg from fpu, used by x87, sse and mmx
@@ -68,7 +72,7 @@ typedef struct instruction_arm_s {
    int                 pred_sz;    // size of predecessor list
    int                 *pred;      // predecessor array
    uintptr_t           mark, mark2, mark3;
-    uintptr_t           markf;
+    uintptr_t           markf, markf2;
    uintptr_t           markseg;
    uintptr_t           marklock;
    int                 pass2choice;// value for choices that are fixed on pass2 for pass3
@@ -90,21 +94,36 @@ typedef struct dynarec_arm_s {
    void*               block;      // memory pointer where next instruction is emited
    uintptr_t           arm_start;  // start of the arm code
    int                 arm_size;   // size of emitted arm code
+    uintptr_t           jmp_next;   // address of the jump_next address
    flagcache_t         f;
    neoncache_t         n;          // cache for the 8..31 double reg from fpu, plus x87 stack delta
    uintptr_t*          next;       // variable array of "next" jump address
    int                 next_sz;
    int                 next_cap;
-    uintptr_t*          sons_x86;   // the x86 address of potential dynablock sons
-    void**              sons_arm;   // the arm address of potential dynablock sons
-    int                 sons_size;  // number of potential dynablock sons
+    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             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);

+void CreateJmpNext(void* addr, void* next);
+
 #endif //__DYNAREC_ARM_PRIVATE_H_
@@ -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_x86_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
@@ -44,4 +51,17 @@ typedef struct instruction_x86_s {

 void printf_x86_instruction(zydis_dec_t* dec, instruction_x86_t* inst, const char* name);

+//#define USE_CUSTOM_MEM
+#ifdef USE_CUSTOM_MEM
+#define dynaMalloc customMalloc
+#define dynaCalloc customCalloc
+#define dynaRealloc customRealloc
+#define dynaFree customFree
+#else
+#define dynaMalloc box_malloc
+#define dynaCalloc box_calloc
+#define dynaRealloc box_realloc
+#define dynaFree box_free
+#endif
+
 #endif //__DYNAREC_PRIVATE_H_
@@ -233,9 +233,9 @@ NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 1 @<1> Rm @ param=1 char align[4
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 0 @<4> 1 1 0 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]!
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 0 @<4> 1 1 1 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 0 @<4> Rm @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> 1 1 0 1 @ param2_1=1 char align[4] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\]!
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> 1 1 1 1 @ param2_1=1 char align[4] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\]
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> Rm @ param2_1=1 char align[4] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> 1 1 0 1 @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> 1 1 1 1 @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> Rm @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\], <Rm>
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]!
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
@@ -353,8 +353,10 @@ ARM_ cond 0 0 0 1 1 1 1 0 Rn Rd (1) (1) (1) (1) 1 0 0 1 Rt STREXH<c> <Rd>, <Rt>,
 ARM_ cond 0 0 0 1 1 1 1 1 Rn Rt (1) (1) (1) (1) 1 0 0 1 (1) (1) (1) (1) LDREXH<c> <Rt>, [<Rn>]
 ARM_ cond 0 0 1 0 0 0 0 S Rn Rd imm12 AND{S}<c> <Rd>, <Rn>, #<const>
 ARM_ cond 0 0 1 0 0 0 1 S Rn Rd imm12 EOR{S}<c> <Rd>, <Rn>, #<const>
+ARM_ cond 0 0 1 0 0 1 0 0 1 1 1 1 Rd imm12 ADR<c> <Rd>, #-<const>
 ARM_ cond 0 0 1 0 0 1 0 S Rn Rd imm12 SUB{S}<c> <Rd>, <Rn>, #<const>
 ARM_ cond 0 0 1 0 0 1 1 S Rn Rd imm12 RSB{S}<c> <Rd>, <Rn>, #<const>
+ARM_ cond 0 0 1 0 1 0 0 0 1 1 1 1 Rd imm12 ADR<c> <Rd>, #<const>
 ARM_ cond 0 0 1 0 1 0 0 S Rn Rd imm12 ADD{S}<c> <Rd>, <Rn>, #<const>
 ARM_ cond 0 0 1 0 1 0 1 S Rn Rd imm12 ADC{S}<c> <Rd>, <Rn>, #<const>
 ARM_ cond 0 0 1 0 1 1 0 S Rn Rd imm12 SBC{S}<c> <Rd>, <Rn>, #<const>
@@ -45,7 +45,9 @@ const char* DumpSection(Elf32_Shdr *s, char* SST) {
        GO(SHT_GNU_ATTRIBUTES);
        GO(SHT_GNU_HASH);
        GO(SHT_GNU_LIBLIST);
-        GO(SHT_CHECKSUM);
+        #ifndef TERMUX
+          GO(SHT_CHECKSUM);
+        #endif
        GO(SHT_LOSUNW);
        //GO(SHT_SUNW_move);
        GO(SHT_SUNW_COMDAT);
@@ -235,12 +237,12 @@ void DumpMainHeader(Elf32_Ehdr *header, elfheader_t *h)
        printf_dump(LOG_NEVER, "ELF Dump ==========\n");

        printf_dump(LOG_NEVER, "ELF Dump PEntries (%d)\n", h->numSHEntries);
-        for (int i=0; i<h->numPHEntries; ++i)
+        for (uint16_t i=0; i<h->numPHEntries; ++i)
            printf_dump(LOG_NEVER, "  PHEntry %04d : %s\n", i, DumpPHEntry(h->PHEntries+i));
        printf_dump(LOG_NEVER, "ELF Dump PEntries ====\n");

        printf_dump(LOG_NEVER, "ELF Dump Sections (%d)\n", h->numSHEntries);
-        for (int i=0; i<h->numSHEntries; ++i)
+        for (uint16_t i=0; i<h->numSHEntries; ++i)
            printf_dump(LOG_NEVER, "  Section %04d : %s\n", i, DumpSection(h->SHEntries+i, h->SHStrTab));
        printf_dump(LOG_NEVER, "ELF Dump Sections ====\n");
    }
@@ -251,7 +253,7 @@ void DumpSymTab(elfheader_t *h)
    if(box86_dump && h->SymTab) {
        const char* name = ElfName(h);
        printf_dump(LOG_NEVER, "ELF Dump SymTab(%d)=\n", h->numSymTab);
-        for (int i=0; i<h->numSymTab; ++i)
+        for (uint32_t i=0; i<h->numSymTab; ++i)
            printf_dump(LOG_NEVER, "  %s:SymTab[%d] = \"%s\", value=%p, size=%d, info/other=%d/%d index=%d\n", name, 
                i, h->StrTab+h->SymTab[i].st_name, (void*)h->SymTab[i].st_value, h->SymTab[i].st_size,
                h->SymTab[i].st_info, h->SymTab[i].st_other, h->SymTab[i].st_shndx);
@@ -263,7 +265,7 @@ void DumpDynamicSections(elfheader_t *h)
 {
    if(box86_dump && h->Dynamic) {
        printf_dump(LOG_NEVER, "ELF Dump Dynamic(%d)=\n", h->numDynamic);
-        for (int i=0; i<h->numDynamic; ++i)
+        for (uint32_t i=0; i<h->numDynamic; ++i)
            printf_dump(LOG_NEVER, "  Dynamic %04d : %s\n", i, DumpDynamic(h->Dynamic+i));
        printf_dump(LOG_NEVER, "ELF Dump Dynamic=====\n");
    }
@@ -274,7 +276,7 @@ void DumpDynSym(elfheader_t *h)
    if(box86_dump && h->DynSym) {
        const char* name = ElfName(h);
        printf_dump(LOG_NEVER, "ELF Dump DynSym(%d)=\n", h->numDynSym);
-        for (int i=0; i<h->numDynSym; ++i) {
+        for (uint32_t i=0; i<h->numDynSym; ++i) {
            int version = h->VerSym?((Elf32_Half*)((uintptr_t)h->VerSym+h->delta))[i]:-1;
            printf_dump(LOG_NEVER, "  %s:DynSym[%d] = %s\n", name, i, DumpSym(h, h->DynSym+i, version));
        }
@@ -286,7 +288,7 @@ void DumpDynamicNeeded(elfheader_t *h)
 {
    if(box86_dump && h->DynStrTab) {
        printf_dump(LOG_NEVER, "ELF Dump DT_NEEDED=====\n");
-        for (int i=0; i<h->numDynamic; ++i)
+        for (uint32_t i=0; i<h->numDynamic; ++i)
            if(h->Dynamic[i].d_tag==DT_NEEDED) {
                printf_dump(LOG_NEVER, "  Needed : %s\n", h->DynStrTab+h->Dynamic[i].d_un.d_val + h->delta);
            }
@@ -298,7 +300,7 @@ void DumpDynamicRPath(elfheader_t *h)
 {
    if(box86_dump && h->DynStrTab) {
        printf_dump(LOG_NEVER, "ELF Dump DT_RPATH/DT_RUNPATH=====\n");
-        for (int i=0; i<h->numDynamic; ++i) {
+        for (uint32_t i=0; i<h->numDynamic; ++i) {
            if(h->Dynamic[i].d_tag==DT_RPATH) {
                printf_dump(LOG_NEVER, "   RPATH : %s\n", h->DynStrTab+h->Dynamic[i].d_un.d_val + h->delta);
            }
@@ -1,10 +1,6 @@
 #ifndef __ELFLOADER_PRIVATE_H_
 #define __ELFLOADER_PRIVATE_H_

-#ifdef DYNAREC
-typedef struct dynablocklist_s dynablocklist_t;
-#endif
-
 typedef struct library_s library_t;
 typedef struct needed_libs_s needed_libs_t;
 typedef struct kh_mapsymbols_s kh_mapsymbols_t;
@@ -12,24 +8,36 @@ typedef struct kh_mapsymbols_s kh_mapsymbols_t;
 #include <elf.h>
 #include "elfloader.h"

-struct elfheader_s {
+
+typedef struct multiblock_s {
+    void*       p;
+    uintptr_t   offs;
+    uintptr_t   paddr;
+    uintptr_t   align;
+    size_t      size;
+    size_t      asize;
+    uint8_t     flags;
+} multiblock_t;
+
+typedef struct elfheader_s {
    char*       name;
    char*       path;   // Resolved path to file
-    int         numPHEntries;
+    char*       soname; // soname of the elf
+    uint16_t    numPHEntries;
    Elf32_Phdr  *PHEntries;
-    int         numSHEntries;
+    uint16_t    numSHEntries;
    Elf32_Shdr  *SHEntries;
    int         SHIdx;
    int         numSST;
    char*       SHStrTab;
    char*       StrTab;
    Elf32_Sym*  SymTab;
-    int         numSymTab;
+    uint32_t    numSymTab;
    char*       DynStr;
    Elf32_Sym*  DynSym;
-    int         numDynSym;
+    uint32_t    numDynSym;
    Elf32_Dyn*  Dynamic;
-    int         numDynamic;
+    uint32_t    numDynamic;
    char*       DynStrTab;
    int         szDynStrTab;
    Elf32_Half* VerSym;
@@ -38,8 +46,12 @@ struct elfheader_s {
    Elf32_Verdef*   VerDef;
    int         szVerDef;
    int         e_type;
+    uint32_t    flags;

    intptr_t    delta;  // should be 0
+    void*       image;
+    void*       raw;    // raw pointer (might be unaligned vs image and max_align of elf)
+    size_t      raw_size;

    uintptr_t   entrypoint;
    uintptr_t   initentry;
@@ -70,37 +82,45 @@ struct elfheader_s {
    int         textsz;
    uintptr_t   bss;
    int         bsssz;
+    uintptr_t   ehframe;
+    uintptr_t   ehframe_end;
+    uintptr_t   ehframehdr;

    uintptr_t   paddr;
    uintptr_t   vaddr;
-    int         align;
+    uint32_t    align;
    uint32_t    memsz;
    uint32_t    reserve;
    uint32_t    stacksz;
-    int         stackalign;
+    uint32_t    stackalign;
    uintptr_t   tlsaddr;
    uint32_t    tlssize;
    uint32_t    tlsfilesize;
-    int         tlsalign;
+    uint32_t    tlsalign;

    int32_t     tlsbase;    // the base of the tlsdata in the global tlsdata (always negative)

    int         init_done;
    int         fini_done;
+    int         refcnt;     // ref count for the elf
+    int         malloc_hook_2;  // this elf hook malloc, hacking it

    char*       memory; // char* and not void* to allow math on memory pointer
-    void**      multiblock;
-    uintptr_t*  multiblock_offs;
-    uint32_t*   multiblock_size;
+    multiblock_t*  multiblocks;
    int         multiblock_n;

    library_t   *lib;
-    needed_libs_t *neededlibs;
+    needed_libs_t *needed;
+
+    FILE*       file;
+    int         fileno;

    kh_mapsymbols_t   *mapsymbols;
    kh_mapsymbols_t   *weaksymbols;
    kh_mapsymbols_t   *localsymbols;
-};
+    kh_defaultversion_t *globaldefver;  // the global default version for symbols (the XXX@@vvvv of symbols)
+    kh_defaultversion_t *weakdefver;    // the weak default version for symbols (the XXX@@vvvv of symbols)
+} elfheader_t;

 #define R_386_NONE	0
 #define R_386_32	1
@@ -116,4 +136,4 @@ struct elfheader_s {

 elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec);

-#endif //__ELFLOADER_PRIVATE_H_
+#endif //__ELFLOADER_PRIVATE_H_
@@ -43,7 +43,7 @@ int FindSection(Elf32_Shdr *s, int n, char* SHStrTab, const char* name)
    return 0;
 }

-void LoadNamedSection(FILE *f, Elf32_Shdr *s, int size, char* SHStrTab, const char* name, const char* clearname, uint32_t type, void** what, int* num)
+void LoadNamedSection(FILE *f, Elf32_Shdr *s, int size, char* SHStrTab, const char* name, const char* clearname, uint32_t type, void** what, uint32_t* num)
 {
    int n = FindSection(s, size, SHStrTab, name);
    printf_dump(LOG_DEBUG, "Loading %s (idx = %d)\n", clearname, n);
@@ -136,7 +136,7 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
            h->numSHEntries = section.sh_size;
        }
        // now read all section headers
-        printf_dump(LOG_DEBUG, "Read %d Section header\n", h->numSHEntries);
+        printf_dump(LOG_DEBUG, "Read %u Section header\n", h->numSHEntries);
        h->SHEntries = (Elf32_Shdr*)box_calloc(h->numSHEntries, sizeof(Elf32_Shdr));
        fseeko64(f, header.e_shoff ,SEEK_SET);
        if(fread(h->SHEntries, sizeof(Elf32_Shdr), h->numSHEntries, f)!=h->numSHEntries) {
@@ -188,7 +188,7 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
        // grab DT_REL & DT_RELA stuffs
        // also grab the DT_STRTAB string table
        {
-            for (int i=0; i<h->numDynamic; ++i) {
+            for (uint32_t i=0; i<h->numDynamic; ++i) {
                Elf32_Dyn d = h->Dynamic[i];
                Elf32_Word val = d.d_un.d_val;
                Elf32_Addr ptr = d.d_un.d_ptr;
@@ -242,8 +242,10 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
                    printf_dump(LOG_DEBUG, "The DT_INIT_ARRAYSZ is %d\n", h->initarray_sz);
                    break;
                case DT_PREINIT_ARRAYSZ:
-                    if(val)
-                        printf_dump(LOG_NONE, "Warning, PreInit Array (size=%d) present and ignored!\n", val);
+                    #ifndef ANDROID
+                        if(val)
+                            printf_dump(LOG_NONE, "Warning, PreInit Array (size=%d) present and ignored!\n", val);
+                    #endif
                    break;
                case DT_FINI: // Exit hook
                    h->finientry = ptr;
@@ -273,6 +275,10 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
                    h->VerDef = (Elf32_Verdef*)ptr;
                    printf_dump(LOG_DEBUG, "The DT_VERDEF is at address %p\n", h->VerDef);
                    break;
+                case DT_FLAGS:
+                    h->flags = val;
+                    printf_dump(LOG_DEBUG, "The DT_FLAGS is 0x%x\n", h->flags);
+                    break;
                }
            }
            if(h->rel) {
@@ -351,6 +357,17 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
            h->bsssz = h->SHEntries[ii].sh_size;
            printf_dump(LOG_DEBUG, "The .bss is at address %p, and is %d big\n", (void*)h->bss, h->bsssz);
        }
+        ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".eh_frame");
+        if(ii) {
+            h->ehframe = (uintptr_t)(h->SHEntries[ii].sh_addr);
+            h->ehframe_end = h->ehframe + h->SHEntries[ii].sh_size;
+            printf_dump(LOG_DEBUG, "The .eh_frame section is at address %p..%p\n", (void*)h->ehframe, (void*)h->ehframe_end);
+        }
+        ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".eh_frame_hdr");
+        if(ii) {
+            h->ehframehdr = (uintptr_t)(h->SHEntries[ii].sh_addr);
+            printf_dump(LOG_DEBUG, "The .eh_frame_hdr section is at address %p\n", (void*)h->ehframehdr);
+        }

        LoadNamedSection(f, h->SHEntries, h->numSHEntries, h->SHStrTab, ".dynstr", "DynSym Strings", SHT_STRTAB, (void**)&h->DynStr, NULL);
        LoadNamedSection(f, h->SHEntries, h->numSHEntries, h->SHStrTab, ".dynsym", "DynSym", SHT_DYNSYM, (void**)&h->DynSym, &h->numDynSym);
@@ -399,3 +416,74 @@ const char* GetParentSymbolVersion(elfheader_t* h, int index)
    }
    return NULL;
 }
+
+int GetVersionIndice(elfheader_t* h, const char* vername)
+{
+    if(!vername)
+        return 0;
+    if(h->VerDef) {
+        Elf32_Verdef *def = (Elf32_Verdef*)((uintptr_t)h->VerDef + h->delta);
+        while(def) {
+            Elf32_Verdaux *aux = (Elf32_Verdaux*)((uintptr_t)def + def->vd_aux);
+            if(!strcmp(h->DynStr+aux->vda_name, vername))
+                return def->vd_ndx;
+            def = def->vd_next?((Elf32_Verdef*)((uintptr_t)def + def->vd_next)):NULL;
+        }
+    }
+    return 0;
+}
+
+int GetNeededVersionCnt(elfheader_t* h, const char* libname)
+{
+    if(!libname)
+        return 0;
+    if(h->VerNeed) {
+        Elf32_Verneed *ver = (Elf32_Verneed*)((uintptr_t)h->VerNeed + h->delta);
+        while(ver) {
+            char *filename = h->DynStr + ver->vn_file;
+            Elf32_Vernaux *aux = (Elf32_Vernaux*)((uintptr_t)ver + ver->vn_aux);
+            if(!strcmp(filename, libname))
+                return ver->vn_cnt;
+            ver = ver->vn_next?((Elf32_Verneed*)((uintptr_t)ver + ver->vn_next)):NULL;
+        }
+    }
+    return 0;
+}
+
+const char* GetNeededVersionString(elfheader_t* h, const char* libname, int idx)
+{
+    if(!libname)
+        return 0;
+    if(h->VerNeed) {
+        Elf32_Verneed *ver = (Elf32_Verneed*)((uintptr_t)h->VerNeed + h->delta);
+        while(ver) {
+            char *filename = h->DynStr + ver->vn_file;
+            Elf32_Vernaux *aux = (Elf32_Vernaux*)((uintptr_t)ver + ver->vn_aux);
+            if(!strcmp(filename, libname)) {
+                for(int j=0; j<ver->vn_cnt; ++j) {
+                    if(j==idx) 
+                        return h->DynStr+aux->vna_name;
+                    aux = (Elf32_Vernaux*)((uintptr_t)aux + aux->vna_next);
+                }
+                return NULL;    // idx out of bound, return NULL...
+           }
+            ver = ver->vn_next?((Elf32_Verneed*)((uintptr_t)ver + ver->vn_next)):NULL;
+        }
+    }
+    return NULL;
+}
+
+int GetNeededVersionForLib(elfheader_t* h, const char* libname, const char* ver)
+{
+    if(!libname || !ver)
+        return 0;
+    int n = GetNeededVersionCnt(h, libname);
+    if(!n)
+        return 0;
+    for(int i=0; i<n; ++i) {
+        const char* vername = GetNeededVersionString(h, libname, i);
+        if(vername && !strcmp(ver, vername))
+            return 1;
+    }
+    return 0;
+}
@@ -1,4 +1,27 @@
+#include <string.h>
+
+#define F8      *(uint8_t*)(addr++)
+#define F8S     *(int8_t*)(addr++)
+#define F16     *(uint16_t*)(addr+=2, addr-2)
+#define F16S    *(int16_t*)(addr+=2, addr-2)
+#define F32     *(uint32_t*)(addr+=4, addr-4)
+#define F32S    *(int32_t*)(addr+=4, addr-4)
+#define F32S64  (uint64_t)(int64_t)F32S
+#define F64     *(uint64_t*)(addr+=8, addr-8)
+#define F64S    *(int64_t*)(addr+=8, addr-8)
+#define PK(a)   *(uint8_t*)(addr+a)
+#ifdef DYNAREC
+#define STEP if(step) return 0;
+#define STEP2 if(step) {R_EIP = addr; return 0;}
+#define STEP3 if(*step) (*step)++;
+#else
+#define STEP
+#define STEP2
+#define STEP3
+#endif
+
 // ModRM utilities macros
+#define MODREG ((nextop&0xC0)==0xC0)
 #define getecommon(A, T) \
    if(!(nextop&0xC0)) { \
        if((nextop&7)==4) { \
@@ -85,7 +108,7 @@
            case 7: base =      R_BX; break;    \
        }                                       \
        switch((m>>6)&3) {                      \
-            case 0: if(m==6) base = F16; break; \
+            case 0: if((m&7)==6) base = F16S; break; \
            case 1: base += F8S; break;         \
            case 2: base += F16S; break;        \
        }                                       \
@@ -94,47 +117,119 @@


 #define geteb(A) \
-    if((nextop&0xC0)==0xC0) { \
+    if(MODREG) { \
        A = (reg32_t*)&emu->regs[(nextop&3)].byte[((nextop&0x4)>>2)]; \
    } else getecommon(A, reg32_t)
+#define testeb(A, O) \
+    if(MODREG) { \
+        A = (reg32_t*)&emu->regs[(nextop&3)].byte[((nextop&0x4)>>2)]; \
+    } else { \
+        reg32_t* ret; \
+        getecommono(ret, reg32_t, O);\
+        test->memsize = 1; \
+        test->memaddr = (uintptr_t)ret;\
+        test->mem[0] = ret->byte[0];\
+        A = (reg32_t*)test->mem;\
+    }
 #define getebo(A, O)          \
-    if((nextop&0xC0)==0xC0) { \
+    if(MODREG) { \
        A = (reg32_t*)&emu->regs[(nextop&3)].byte[((nextop&0x4)>>2)]; \
    } else getecommono(A, reg32_t, O)
 #define geted(A) \
-    if((nextop&0xC0)==0xC0) { \
+    if(MODREG) { \
        A = &emu->regs[(nextop&7)]; \
    } else getecommon(A, reg32_t)
 #define getedo(A, O) \
-    if((nextop&0xC0)==0xC0) { \
+    if(MODREG) { \
        A = &emu->regs[(nextop&7)]; \
    } else getecommono(A, reg32_t, O)
+#define tested(A, SZ, O) \
+    if(MODREG) { \
+        A = &emu->regs[(nextop&7)]; \
+    } else {\
+        reg32_t* ret; \
+        getecommono(ret, reg32_t, O);\
+        test->memsize = SZ; \
+        test->memaddr = (uintptr_t)ret;\
+        memcpy(test->mem, ret->dword, SZ);\
+        A = (reg32_t*)test->mem;\
+    }
 #define getem(A) \
-    if((nextop&0xC0)==0xC0) { \
+    if(MODREG) { \
        A = &emu->mmx[(nextop&7)]; \
    } else getecommon(A, mmx87_regs_t)
+#define testem(A) \
+    if(MODREG) { \
+        A = &emu->mmx[(nextop&7)]; \
+    } else {\
+        mmx87_regs_t* ret; \
+        getecommon(ret, mmx87_regs_t) \
+        test->memsize = 8; \
+        test->memaddr = (uintptr_t)ret;\
+        memcpy(test->mem, ret->ub, 8);\
+        A = (mmx87_regs_t*)test->mem;\
+    }
 #define getex(A) \
-    if((nextop&0xC0)==0xC0) { \
+    if(MODREG) { \
        A = &emu->xmm[(nextop&7)]; \
    } else getecommon(A, sse_regs_t)
+#define testex(A) \
+    if(MODREG) { \
+        A = &emu->xmm[(nextop&7)]; \
+    } else {\
+        sse_regs_t* ret; \
+        getecommon(ret, sse_regs_t);\
+        test->memsize = 16; \
+        test->memaddr = (uintptr_t)ret;\
+        memcpy(test->mem, ret->ub, 16);\
+        A = (sse_regs_t*)test->mem;\
+    }
 #define getew16(A)  \
-    if((nextop&0xC0)==0xC0) { \
+    if(MODREG) { \
        A = &emu->regs[(nextop&7)]; \
    } else getecommon16(A, reg32_t)
 #define getew16o(A, O)              \
-    if((nextop&0xC0)==0xC0) {       \
+    if(MODREG) {       \
        A = &emu->regs[(nextop&7)]; \
    } else getecommon16o(A, reg32_t, O)
+#define testtew16(A, O)                 \
+    if(MODREG) {           \
+        A = &emu->regs[(nextop&7)];     \
+    } else {                            \
+        reg32_t* ret;                   \
+        getecommon16o(A, reg32_t, O);   \
+        test->memsize = 2;              \
+        test->memaddr = (uintptr_t)ret; \
+        *(uint16_t*)test->mem = ret->word[0];\
+        A = (reg32_t*)test->mem;        \
+    }

 // Macros for ModR/M gets
+#ifdef TEST_INTERPRETER
+#define GET_EB      testeb(oped, 0)
+#define GET_ED      tested(oped, 4, 0)
+#define GET_ED8     tested(oped, 8, 0)
+#define GET_EDT     tested(oped, 10, 0)
+#define GET_ED_OFFS(o) tested(oped, 4, o)
+#define GET_EB_OFFS(o) testeb(oped, o)
+#define GET_EM      testem(opem)
+#define GET_EX      testex(opex)
+#define GET_EW16    testew16(oped, 0)
+#define GET_EW16_OFFS(o)    testtew16(oped, o)
+#else
 #define GET_EB      geteb(oped)
 #define GET_ED      geted(oped)
+#define GET_ED8     geted(oped)
+#define GET_EDT     geted(oped)
 #define GET_ED_OFFS(o) getedo(oped, o)
 #define GET_EB_OFFS(o) getebo(oped, o)
 #define GET_EM      getem(opem)
 #define GET_EX      getex(opex)
 #define GET_EW16    getew16(oped)
 #define GET_EW16_OFFS(o)    getew16o(oped, o)
+#endif
+#define GET_ED_     geted(oped)
+#define GET_EW16_   getew16(oped)
 #define EB          oped
 #define ED          oped
 #define EM          opem
@@ -1,555 +0,0 @@
-    opcode = F8;
-    goto *opcodes66[opcode];
-
-    #define GO(B, OP)                       \
-    _66_##B##_0: \
-        nextop = F8;               \
-        GET_EB;             \
-        EB->byte[0] = OP##8(emu, EB->byte[0], GB);  \
-        NEXT;                              \
-    _66_##B##_1:                               \
-        nextop = F8;                \
-        GET_EW;                   \
-        EW->word[0] = OP##16(emu, EW->word[0], GW.word[0]); \
-        NEXT;                              \
-    _66_##B##_2: \
-        nextop = F8;               \
-        GET_EB;                   \
-        GB = OP##8(emu, GB, EB->byte[0]); \
-        NEXT;                              \
-    _66_##B##_3:                               \
-        nextop = F8;                \
-        GET_EW;                   \
-        GW.word[0] = OP##16(emu, GW.word[0], EW->word[0]); \
-        NEXT;                              \
-    _66_##B##_4: \
-        R_AL = OP##8(emu, R_AL, F8); \
-        NEXT;                              \
-    _66_##B##_5:                               \
-        R_AX = OP##16(emu, R_AX, F16); \
-        NEXT;
-
-    GO(0x00, add)                   /* ADD 0x01 ~> 0x05 */
-    GO(0x08, or)                    /*  OR 0x09 ~> 0x0D */
-    GO(0x10, adc)                   /* ADC 0x11 ~> 0x15 */
-    GO(0x18, sbb)                   /* SBB 0x19 ~> 0x1D */
-    GO(0x20, and)                   /* AND 0x21 ~> 0x25 */
-    GO(0x28, sub)                   /* SUB 0x29 ~> 0x2D */
-    GO(0x30, xor)                   /* XOR 0x31 ~> 0x35 */
-    //GO(0x38, cmp)                   /* CMP 0x39 ~> 0x3D */
-    #undef GO
-    _66_0x06:                      /* PUSH ES */
-        Push16(emu, emu->segs[_ES]);
-        NEXT;
-    _66_0x07:                      /* POP ES */
-        emu->segs[_ES] = Pop16(emu);    // no check, no use....
-        emu->segs_serial[_ES] = 0;
-        NEXT;
-
-    _66_0x0F:                      /* 66 0f prefix */
-        emu->old_ip = R_EIP;
-        R_EIP = ip-2;   // don't count 66 0F yet
-        Run660F(emu); // implemented in Run660f.c
-        ip = R_EIP;
-        if(emu->quit) goto fini;
-        STEP
-        NEXT;
-
-    _66_0x1E:                      /* PUSH DS */
-        Push16(emu, emu->segs[_DS]);
-        NEXT;
-    _66_0x1F:                      /* POP DS */
-        emu->segs[_DS] = Pop16(emu);    // no check, no use....
-        emu->segs_serial[_DS] = 0;
-        NEXT;
-        
-    _66_0x26:                      /* ES: */
-        // ignored
-        opcode = F8;
-        goto *opcodes66[opcode];
-    _66_0x2E:                      /* CS: */
-        // ignored
-        opcode = F8;
-        goto *opcodes66[opcode];
-    _66_0x36:                      /* SS: */
-        // ignored
-        opcode = F8;
-        goto *opcodes66[opcode];
-
-    _66_0x39:
-        nextop = F8;
-        GET_EW;
-        cmp16(emu, EW->word[0], GW.word[0]);
-        NEXT;
-    _66_0x3B:
-        nextop = F8;
-        GET_EW;
-        cmp16(emu, GW.word[0], EW->word[0]);
-        NEXT;
-    _66_0x3D:
-        cmp16(emu, R_AX, F16);
-        NEXT;
-    _66_0x3E:                      /* DS: */
-        // ignored
-        opcode = F8;
-        goto *opcodes66[opcode];
-    
-    _66_0x40:
-    _66_0x41:
-    _66_0x42:
-    _66_0x43:
-    _66_0x44:
-    _66_0x45:
-    _66_0x46:
-    _66_0x47:                              /* INC Reg */
-        tmp8u = opcode&7;
-        emu->regs[tmp8u].word[0] = inc16(emu, emu->regs[tmp8u].word[0]);
-        NEXT;
-    _66_0x48:
-    _66_0x49:
-    _66_0x4A:
-    _66_0x4B:
-    _66_0x4C:
-    _66_0x4D:
-    _66_0x4E:
-    _66_0x4F:                              /* DEC Reg */
-        tmp8u = opcode&7;
-        emu->regs[tmp8u].word[0] = dec16(emu, emu->regs[tmp8u].word[0]);
-        NEXT;
-    _66_0x50:
-    _66_0x51:
-    _66_0x52:
-    _66_0x53:
-    _66_0x54:
-    _66_0x55:
-    _66_0x56:
-    _66_0x57:                              /* PUSH Reg */
-        tmp8u = opcode&7;
-        Push16(emu, emu->regs[tmp8u].word[0]);
-        NEXT;
-    _66_0x58:
-    _66_0x59:
-    _66_0x5A:
-    _66_0x5B:
-    _66_0x5C:                      
-    _66_0x5D:
-    _66_0x5E:
-    _66_0x5F:                              /* POP Reg */
-        tmp8u = opcode&7;
-        emu->regs[tmp8u].word[0] = Pop16(emu);
-        NEXT;
-    _66_0x60:                              /* PUSHA */
-        tmp16u = R_SP;
-        Push16(emu, R_AX);
-        Push16(emu, R_CX);
-        Push16(emu, R_DX);
-        Push16(emu, R_BX);
-        Push16(emu, tmp16u);
-        Push16(emu, R_BP);
-        Push16(emu, R_SI);
-        Push16(emu, R_DI);
-        NEXT;
-    _66_0x61:                              /* POPA */
-        R_DI = Pop16(emu);
-        R_SI = Pop16(emu);
-        R_BP = Pop16(emu);
-        R_ESP+=2;   // POP ESP
-        R_BX = Pop16(emu);
-        R_DX = Pop16(emu);
-        R_CX = Pop16(emu);
-        R_AX = Pop16(emu);
-        NEXT;
-
-    _66_0x64:
-        R_EIP = ip-2;
-        RunFS66(emu, GetFSBaseEmu(emu));
-        ip = R_EIP;
-        STEP;
-        NEXT;
-
-    _66_0x66:
-        goto _0x66; // 0x66 0x66 => can remove one 0x66
-
-    _66_0x68:                       /* PUSH u16 */
-        tmp16u = F16;
-        Push16(emu, tmp16u);
-        NEXT;
-    _66_0x69:                      /* IMUL Gw,Ew,Iw */
-        nextop = F8;
-        GET_EW;
-        tmp16u = F16;
-        GW.word[0] = imul16(emu, EW->word[0], tmp16u);
-        NEXT;
-    _66_0x6A:                      /* PUSH Ib (as signed word) */
-        tmp16s = F8S;
-        Push16(emu, (uint16_t)tmp16s);
-        NEXT;
-    _66_0x6B:                      /* IMUL Gw,Ew,Ib */
-        nextop = F8;
-        GET_EW;
-        tmp16s = F8S;
-        GW.word[0] = imul16(emu, EW->word[0], (uint16_t)tmp16s);
-        NEXT;
-
-    _66_0x81:                              /* GRP3 Ew,Iw */
-    _66_0x83:                              /* GRP3 Ew,Ib */
-        nextop = F8;
-        GET_EW;
-        if(opcode==0x81) 
-            tmp16u = F16;
-        else {
-            tmp16s = F8S;
-            tmp16u = (uint16_t)tmp16s;
-        }
-        switch((nextop>>3)&7) {
-            case 0: EW->word[0] = add16(emu, EW->word[0], tmp16u); break;
-            case 1: EW->word[0] =  or16(emu, EW->word[0], tmp16u); break;
-            case 2: EW->word[0] = adc16(emu, EW->word[0], tmp16u); break;
-            case 3: EW->word[0] = sbb16(emu, EW->word[0], tmp16u); break;
-            case 4: EW->word[0] = and16(emu, EW->word[0], tmp16u); break;
-            case 5: EW->word[0] = sub16(emu, EW->word[0], tmp16u); break;
-            case 6: EW->word[0] = xor16(emu, EW->word[0], tmp16u); break;
-            case 7:               cmp16(emu, EW->word[0], tmp16u); break;
-        }
-        NEXT;
-
-    _66_0x85:                              /* TEST Ew,Gw */
-        nextop = F8;
-        GET_EW;
-        test16(emu, EW->word[0], GW.word[0]);
-        NEXT;
-
-    _66_0x87:                              /* XCHG Ew,Gw */
-        nextop = F8;
-        GET_EW;
-        tmp16u = GW.word[0];
-        GW.word[0] = EW->word[0];
-        EW->word[0] = tmp16u;
-        NEXT;
-
-    _66_0x89:                              /* MOV Ew,Gw */
-        nextop = F8;
-        GET_EW;
-        EW->word[0] = GW.word[0];
-        NEXT;
-
-    _66_0x8B:                              /* MOV Gw,Ew */
-        nextop = F8;
-        GET_EW;
-        GW.word[0] = EW->word[0];
-        NEXT;
-    _66_0x8C:                              /* MOV Ew,Seg */
-        nextop = F8;
-        GET_EW;
-        EW->word[0] = emu->segs[(nextop&0x38)>>3];
-        NEXT;
-    
-    _66_0x8E:                               /* MOV Seg,Ew */
-        nextop = F8;
-        GET_EW;
-        emu->segs[((nextop&0x38)>>3)] = EW->word[0];
-        emu->segs_serial[((nextop&0x38)>>3)] = 0;
-        NEXT;
-    _66_0x8F:                              /* POP Ew */
-        nextop = F8;
-        GET_EW;
-        EW->word[0] = Pop16(emu);
-        NEXT;
-    _66_0x90:                              /* NOP */
-        NEXT;
-
-    _66_0x91:
-    _66_0x92:
-    _66_0x93:
-    _66_0x94:
-    _66_0x95:
-    _66_0x96:
-    _66_0x97:                      /* XCHG reg,EAX */
-        tmp16u = R_AX;
-        R_AX = emu->regs[opcode&7].word[0];
-        emu->regs[opcode&7].word[0] = tmp16u;
-        NEXT;
-
-    _66_0x98:                               /* CBW */
-        emu->regs[_AX].sword[0] = emu->regs[_AX].sbyte[0];
-        NEXT;
-    _66_0x99:                              /* CWD */
-        R_DX=((R_AX & 0x8000)?0xFFFF:0x0000);
-        NEXT;
-
-    _66_0x9C:                              /* PUSHFW */
-        CHECK_FLAGS(emu);
-        Push16(emu, (uint16_t)emu->eflags.x32);
-        NEXT;
-    _66_0x9D:                              /* POPFW */
-        CHECK_FLAGS(emu);
-        emu->eflags.x32 &=0xffff0000;
-        emu->eflags.x32 |= (Pop16(emu) & 0x3F7FD7) | 0x2;
-        NEXT;
-
-    _66_0xA1:                              /* MOV AX,Ow */
-        R_AX = *(uint16_t*)F32;
-        NEXT;
-    _66_0xA3:                              /* MOV Ow,AX */
-        *(uint16_t*)F32 = R_AX;
-        NEXT;
-    _66_0xA5:                              /* MOVSW */
-        tmp8s = ACCESS_FLAG(F_DF)?-2:+2;
-        *(uint16_t*)R_EDI = *(uint16_t*)R_ESI;
-        R_EDI += tmp8s;
-        R_ESI += tmp8s;
-        NEXT;
-
-    _66_0xA7:                              /* CMPSW */
-        tmp8s = ACCESS_FLAG(F_DF)?-2:+2;
-        tmp16u  = *(uint16_t*)R_EDI;
-        tmp16u2 = *(uint16_t*)R_ESI;
-        R_EDI += tmp8s;
-        R_ESI += tmp8s;
-        cmp16(emu, tmp16u2, tmp16u);
-        NEXT;
-
-    _66_0xA9:                             /* TEST AX,Iw */
-        test16(emu, R_AX, F16);
-        NEXT;
-
-    _66_0xAB:                              /* STOSW */
-        tmp8s = ACCESS_FLAG(F_DF)?-2:+2;
-        *(uint16_t*)R_EDI = R_AX;
-        R_EDI += tmp8s;
-        NEXT;
-    _66_0xAD:                              /* LODSW */
-        tmp8s = ACCESS_FLAG(F_DF)?-2:+2;
-        R_AX = *(uint16_t*)R_ESI;
-        R_ESI += tmp8s;
-        NEXT;
-    _66_0xAF:                              /* SCASW */
-        tmp8s = ACCESS_FLAG(F_DF)?-2:+2;
-        cmp16(emu, R_AX, *(uint16_t*)R_EDI);
-        R_EDI += tmp8s;
-        NEXT;
-
-    _66_0xB8:                              /* MOV AX,Iw */
-    _66_0xB9:                              /* MOV CX,Iw */
-    _66_0xBA:                              /* MOV DX,Iw */
-    _66_0xBB:                              /* MOV BX,Iw */
-    _66_0xBC:                              /*    ...     */
-    _66_0xBD:
-    _66_0xBE:
-    _66_0xBF:
-        emu->regs[opcode&7].word[0] = F16;
-        NEXT;
-
-    _66_0xC1:                              /* GRP2 Ew,Ib */
-        nextop = F8;
-        GET_EW;
-        tmp8u = F8 /*& 0x1f*/;
-        switch((nextop>>3)&7) {
-            case 0: EW->word[0] = rol16(emu, EW->word[0], tmp8u); break;
-            case 1: EW->word[0] = ror16(emu, EW->word[0], tmp8u); break;
-            case 2: EW->word[0] = rcl16(emu, EW->word[0], tmp8u); break;
-            case 3: EW->word[0] = rcr16(emu, EW->word[0], tmp8u); break;
-            case 4:
-            case 6: EW->word[0] = shl16(emu, EW->word[0], tmp8u); break;
-            case 5: EW->word[0] = shr16(emu, EW->word[0], tmp8u); break;
-            case 7: EW->word[0] = sar16(emu, EW->word[0], tmp8u); break;
-        }
-        NEXT;
-
-    _66_0xC7:                              /* MOV Ew,Iw */
-        nextop = F8;
-        GET_EW;
-        EW->word[0] = F16;
-        NEXT;
-
-    _66_0xCB:                               /* FAR RET */
-        ip = Pop(emu);
-        emu->segs[_CS] = Pop(emu);    // no check, no use....
-        emu->segs_serial[_CS] = 0;
-        // need to check status of CS register!
-        STEP
-        NEXT;
-    _66_0xCC:                              /* INT3 */
-        emu->old_ip = R_EIP;
-        R_EIP = ip-1;
-        if(my_context->signals[SIGTRAP])
-            raise(SIGTRAP);
-        ip = R_EIP;
-        if(emu->quit) goto fini;
-        STEP
-        NEXT;
-    _66_0xD1:                              /* GRP2 Ew,1  */
-    _66_0xD3:                              /* GRP2 Ew,CL */
-        nextop = F8;
-        GET_EW;
-        tmp8u=(opcode==0xD3)?R_CL:1;
-        switch((nextop>>3)&7) {
-            case 0: EW->word[0] = rol16(emu, EW->word[0], tmp8u); break;
-            case 1: EW->word[0] = ror16(emu, EW->word[0], tmp8u); break;
-            case 2: EW->word[0] = rcl16(emu, EW->word[0], tmp8u); break;
-            case 3: EW->word[0] = rcr16(emu, EW->word[0], tmp8u); break;
-            case 4: 
-            case 6: EW->word[0] = shl16(emu, EW->word[0], tmp8u); break;
-            case 5: EW->word[0] = shr16(emu, EW->word[0], tmp8u); break;
-            case 7: EW->word[0] = sar16(emu, EW->word[0], tmp8u); break;
-        }
-        NEXT;
-    
-    _66_0xD9:
-        emu->old_ip = R_EIP;
-        R_EIP = ip;
-        Run66D9(emu);
-        ip = R_EIP;
-        if(emu->quit) goto fini;
-        STEP
-        NEXT;
-
-    _66_0xDD:
-        emu->old_ip = R_EIP;
-        R_EIP = ip;
-        Run66DD(emu);
-        ip = R_EIP;
-        if(emu->quit) goto fini;
-        STEP
-        NEXT;
-
-    _66_0xF0:                      /* LOCK prefix */
-        emu->old_ip = R_EIP;
-        R_EIP = ip - 2;
-        RunLock66(emu);
-        ip = R_EIP;
-        if(emu->quit) goto fini;
-        STEP
-        NEXT;
-
-    _66_0xF2:                      /* REPNZ prefix */
-    _66_0xF3:                      /* REPZ prefix */
-        nextop = F8;
-        tmp8s = ACCESS_FLAG(F_DF)?-2:+2;
-        tmp32u = R_ECX;
-        switch(nextop) {
-            case 0xA4:              /* REP MOVSB */
-                while(tmp32u) {
-                    --tmp32u;
-                    *(uint8_t*)R_EDI = *(uint8_t*)R_ESI;
-                    R_EDI += tmp8s;
-                    R_ESI += tmp8s;
-                }
-                break;
-            case 0xA5:              /* REP MOVSW */
-                while(tmp32u) {
-                    --tmp32u;
-                    *(uint16_t*)R_EDI = *(uint16_t*)R_ESI;
-                    R_EDI += tmp8s;
-                    R_ESI += tmp8s;
-                }
-                break;
-            case 0xA7:              /* REP(N)Z CMPSW */
-                tmp16u = 0;
-                tmp16u2 = 0;
-                while(tmp32u) {
-                    --tmp32u;
-                    tmp16u  = *(uint16_t*)R_EDI;
-                    tmp16u2 = *(uint16_t*)R_ESI;
-                    R_EDI += tmp8s;
-                    R_ESI += tmp8s;
-                    if((tmp16u==tmp16u2)==(opcode==0xF2))
-                        break;
-                }
-                if(R_ECX) cmp16(emu, tmp16u2, tmp16u);
-                break;
-            case 0xAB:              /* REP STOSW */
-                while(tmp32u) {
-                    --tmp32u;
-                    *(uint16_t*)R_EDI = R_AX;
-                    R_EDI += tmp8s;
-                }
-                break;
-            case 0xAD:              /* REP LODSW */
-                while(tmp32u) {
-                    --tmp32u;
-                    R_AX = *(uint16_t*)R_ESI;
-                    R_ESI += tmp8s;
-                }
-                break;
-            case 0xAF:              /* REP(N)Z SCASW */
-                tmp16u = 0;
-                while(tmp32u) {
-                    --tmp32u;
-                    tmp16u = *(uint16_t*)R_EDI;
-                    R_EDI += tmp8s;
-                    if((R_AX==tmp16u)==(opcode==0xF2))
-                        break;
-                }
-                if(R_ECX) cmp16(emu, R_AX, tmp16u);
-                break;
-            default:
-                goto _default;
-        }
-        R_ECX = tmp32u;
-        NEXT;
-
-    _66_0xF5:                      /* CMC */
-        CHECK_FLAGS(emu);
-        CONDITIONAL_SET_FLAG(!ACCESS_FLAG(F_CF), F_CF);
-        NEXT;
-
-    _66_0xF7:                      /* GRP3 Ew(,Iw) */
-        nextop = F8;
-        GET_EW;
-        switch((nextop>>3)&7) {
-            case 0: 
-            case 1:                 /* TEST Ew,Iw */
-                test16(emu, EW->word[0], F16);
-                break;
-            case 2:                 /* NOT Ew */
-                EW->word[0] = not16(emu, EW->word[0]);
-                break;
-            case 3:                 /* NEG Ew */
-                EW->word[0] = neg16(emu, EW->word[0]);
-                break;
-            case 4:                 /* MUL AX,Ew */
-                mul16(emu, EW->word[0]);
-                break;
-            case 5:                 /* IMUL AX,Ew */
-                imul16_eax(emu, EW->word[0]);
-                break;
-            case 6:                 /* DIV Ew */
-                div16(emu, EW->word[0]);
-                break;
-            case 7:                 /* IDIV Ew */
-                idiv16(emu, EW->word[0]);
-                break;
-        }
-        NEXT;
-    _66_0xF8:                       /* CLC */
-        CHECK_FLAGS(emu);
-        CLEAR_FLAG(F_CF);
-        NEXT;
-    _66_0xF9:                       /* STC */
-        CHECK_FLAGS(emu);
-        SET_FLAG(F_CF);
-        NEXT;
-
-    _66_0xFF:                      /* GRP 5 Ew */
-        nextop = F8;
-        GET_EW;
-        switch((nextop>>3)&7) {
-            case 0:                 /* INC Ed */
-                EW->word[0] = inc16(emu, EW->word[0]);
-                break;
-            case 1:                 /* DEC Ed */
-                EW->word[0] = dec16(emu, EW->word[0]);
-                break;
-            case 6:
-                Push16(emu, EW->word[0]);
-                break;
-            default:
-                emu->old_ip = R_EIP;
-                R_EIP = ip;
-                printf_log(LOG_NONE, "Illegal Opcode 66 %02X %02X\n", opcode, nextop);
-                emu->quit=1;
-                emu->error |= ERR_ILLEGAL;
-                goto fini;
-        }
-        NEXT;
@@ -0,0 +1,127 @@
+#include <stdint.h>
+
+#include "box86stack.h"
+#include "x86emu.h"
+#include "x86run_private.h"
+#include "x86emu_private.h"
+#include "x86compstrings.h"
+
+static int overrideIfDataInvalid(sse_regs_t* mem, int lmem, sse_regs_t* reg, int lreg, int j, int i, int imm8)
+{
+    int valid1 = (i<lreg);
+    int valid2 = (j<lmem);
+    if(!valid1 && !valid2)
+        switch((imm8>>2)&3) {
+            case 0b00:
+            case 0b01:  return 0;
+            case 0b10:
+            case 0b11:  return 1;
+        }
+    if(!valid1 && valid2)
+        switch((imm8>>2)&3) {
+            case 0b00:
+            case 0b01:
+            case 0b10:  return 0;
+            case 0b11:  return 1;
+        }
+    if(valid1 && !valid2)
+        return 0;
+    switch((imm8>>2)&3) {
+        case 0b01:  // range
+            switch (imm8&3) {
+                case 0b00:  // ub
+                    return (i&1)?((reg->ub[i]>=mem->ub[j])):((reg->ub[i]<=mem->ub[j]));
+                case 0b01:  // uw
+                    return (i&1)?((reg->uw[i]>=mem->uw[j])):((reg->uw[i]<=mem->uw[j]));
+                case 0b10:  // sb
+                    return (i&1)?((reg->sb[i]>=mem->sb[j])):((reg->sb[i]<=mem->sb[j]));
+                case 0b11:  // sw
+                    return (i&1)?((reg->sw[i]>=mem->sw[j])):((reg->sw[i]<=mem->sw[j]));
+            }
+            break;
+        default:    // the others
+            switch (imm8&1) {
+                case 0: // byte
+                    return (reg->ub[i] == mem->ub[j]);
+                case 1: // word
+                    return (reg->uw[i] == mem->uw[j]);
+            }
+    }
+}
+
+uint32_t sse42_compare_string_explicit_len(x86emu_t* emu, sse_regs_t* mem, int lmem, sse_regs_t* reg, int lreg, uint8_t imm8)
+{
+    // get number of packed byte/word
+    int n_packed = (imm8&1)?8:16;
+    if(lreg<0) lreg = -lreg;
+    if(lmem<0) lmem = -lmem;
+    if(lreg>n_packed) lreg = n_packed;
+    if(lmem>n_packed) lmem = n_packed;
+    // aggregate to intres1
+    uint32_t intres1 = 0;
+    switch((imm8>>2)&3) {
+        case 0b00:  //Equal any
+            for(int j=0; j<n_packed; ++j)
+                for(int i=0; i<n_packed; ++i) {
+                    intres1 |= overrideIfDataInvalid(mem, lmem, reg, lreg, j, i, imm8)<<j;
+                }
+            break;
+        case 0b01:  // Range
+            for(int j=0; j<n_packed; ++j)
+                for(int i=0; i<n_packed; i+=2) {
+                    intres1 |= (overrideIfDataInvalid(mem, lmem, reg, lreg, j, i, imm8) & overrideIfDataInvalid(mem, lmem, reg, lreg, j, i+1, imm8))<<j;
+                }
+            break;
+        case 0b10:  // Equal each
+            for(int i=0; i<n_packed; ++i) {
+                intres1 |= overrideIfDataInvalid(mem, lmem, reg, lreg, i, i, imm8)<<i;
+            }
+            break;
+        case 0b11:  // Equal ordered
+            intres1 = (1<<n_packed)-1;
+            for(int j=0; j<n_packed; ++j)
+                for(int i=0; i<n_packed-j; ++i) {
+                    #if 1
+                    if(!overrideIfDataInvalid(mem, lmem, reg, lreg, i+j, i, imm8)) {
+                        intres1 &= ~(1<<j);
+                        break;
+                    }
+                    #else
+                    int k = i+j;
+                    intres1 &= (((1<<n_packed)-1)^(1<<j)) | overrideIfDataInvalid(mem, lmem, reg, lreg, k, i, imm8)<<j;
+                    #endif
+                }
+            break;
+    }
+    // build intres2
+    uint32_t intres2 = intres1;
+    switch((imm8>>4)&3) {
+        case 0b01: intres2 ^= ((1<<n_packed)-1); break;
+        case 0b11: intres2 ^= ((1<<lmem)-1); break;
+    }
+    // and now set the flags
+    RESET_FLAGS(emu);
+    CONDITIONAL_SET_FLAG(intres2, F_CF);
+    CONDITIONAL_SET_FLAG(lmem<n_packed, F_ZF);
+    CONDITIONAL_SET_FLAG(lreg<n_packed, F_SF);
+    CONDITIONAL_SET_FLAG(intres2&1, F_OF);
+    CONDITIONAL_SET_FLAG(((!intres2) && (lmem==n_packed)), F_AF);
+    CLEAR_FLAG(F_PF);
+    
+    return intres2;
+}
+
+uint32_t sse42_compare_string_implicit_len(x86emu_t* emu, sse_regs_t* mem, sse_regs_t* reg, uint8_t imm8)
+{
+    int lmem = 0;
+    int lreg = 0;
+    // get lmem and lreg
+    if(imm8&1) {
+        while(lmem<8 && mem->uw[lmem]) ++lmem;
+        while(lreg<8 && reg->uw[lreg]) ++lreg;
+    } else {
+        while(lmem<16 && mem->ub[lmem]) ++lmem;
+        while(lreg<16 && reg->ub[lreg]) ++lreg;
+    }
+    return sse42_compare_string_explicit_len(emu, mem, lmem, reg, lreg, imm8);
+}
@@ -0,0 +1,13 @@
+#ifndef __X86_CMPSTRINGS_H__
+#define __X86_CMPSTRINGS_H__
+
+#include <stdint.h>
+
+#include "regs.h"
+
+typedef struct x86emu_s x86emu_t;
+
+uint32_t sse42_compare_string_explicit_len(x86emu_t* emu, sse_regs_t* a, int la, sse_regs_t* b, int lb, uint8_t imm8);
+uint32_t sse42_compare_string_implicit_len(x86emu_t* emu, sse_regs_t* a, sse_regs_t* b, uint8_t imm8);
+
+#endif //__X86_CMPSTRINGS_H__
@@ -23,6 +23,13 @@
 #ifdef DYNAREC
 #include "custommem.h"
 #endif
+// for the applyFlushTo0
+#ifdef __i386__
+#include <immintrin.h>
+#elif defined(__arm__)
+#else
+#warning Architecture cannot follow SSE Flush to 0 flag
+#endif

 typedef struct cleanup_s {
    void*       f;
@@ -42,22 +49,6 @@ static uint32_t x86emu_parity_tab[8] =
 	0x69969669,
 };

-uint32_t* GetParityTab()
-{
-    return x86emu_parity_tab;
-}
-
-void PushExit(x86emu_t* emu)
-{
-    uintptr_t endMarker = AddCheckBridge(my_context->system, NULL, NULL, 0, "(ExitEmulation)");
-    Push(emu, endMarker);
-}
-
-void* GetExit()
-{
-    return (void*)AddCheckBridge(my_context->system, NULL, NULL, 0, "(ExitEmulation)");
-}
-
 static void internalX86Setup(x86emu_t* emu, box86context_t *context, uintptr_t start, uintptr_t stack, int stacksize, int ownstack)
 {
    emu->context = context;
@@ -81,13 +72,13 @@ static void internalX86Setup(x86emu_t* emu, box86context_t *context, uintptr_t s
    emu->segs[_GS] = 0x33;
    // setup fpu regs
    reset_fpu(emu);
-    emu->mxcsr = 0x1f80;  
+    emu->mxcsr.x32 = 0x1f80;  
 }

 EXPORTDYN
 x86emu_t *NewX86Emu(box86context_t *context, uintptr_t start, uintptr_t stack, int stacksize, int ownstack)
 {
-    printf_log(LOG_DEBUG, "Allocate a new X86 Emu, with EIP=%p and Stack=%p/0x%X\n", (void*)start, (void*)stack, stacksize);
+    printf_log(LOG_DEBUG, "%04d|Allocate a new X86 Emu, with EIP=%p and Stack=%p/0x%X\n", GetTID(), (void*)start, (void*)stack, stacksize);

    x86emu_t *emu = (x86emu_t*)box_calloc(1, sizeof(x86emu_t));

@@ -98,7 +89,7 @@ x86emu_t *NewX86Emu(box86context_t *context, uintptr_t start, uintptr_t stack, i

 x86emu_t *NewX86EmuFromStack(x86emu_t* emu, box86context_t *context, uintptr_t start, uintptr_t stack, int stacksize, int ownstack)
 {
-    printf_log(LOG_DEBUG, "New X86 Emu from stack, with EIP=%p and Stack=%p/0x%X\n", (void*)start, (void*)stack, stacksize);
+    printf_log(LOG_DEBUG, "%04d|New X86 Emu from stack, with EIP=%p and Stack=%p/0x%X\n", GetTID(), (void*)start, (void*)stack, stacksize);

    internalX86Setup(emu, context, start, stack, stacksize, ownstack);
    
@@ -108,6 +99,7 @@ x86emu_t *NewX86EmuFromStack(x86emu_t* emu, box86context_t *context, uintptr_t s
 EXPORTDYN
 void SetupX86Emu(x86emu_t *emu)
 {
+    (void)emu;
    printf_log(LOG_DEBUG, "Setup X86 Emu\n");
 }

@@ -128,6 +120,7 @@ void SetTraceEmu(uintptr_t start, uintptr_t end)

 void AddCleanup(x86emu_t *emu, void *p)
 {
+    (void)emu;
    if(my_context->clean_sz == my_context->clean_cap) {
        my_context->clean_cap += 4;
        my_context->cleanups = (cleanup_t*)box_realloc(my_context->cleanups, sizeof(cleanup_t)*my_context->clean_cap);
@@ -139,6 +132,7 @@ void AddCleanup(x86emu_t *emu, void *p)

 void AddCleanup1Arg(x86emu_t *emu, void *p, void* a)
 {
+    (void)emu;
    if(my_context->clean_sz == my_context->clean_cap) {
        my_context->clean_cap += 4;
        my_context->cleanups = (cleanup_t*)box_realloc(my_context->cleanups, sizeof(cleanup_t)*my_context->clean_cap);
@@ -188,6 +182,11 @@ void FreeX86Emu(x86emu_t **emu)
        return;
    printf_log(LOG_DEBUG, "%04d|Free a X86 Emu (%p)\n", GetTID(), *emu);

+    if((*emu)->test.emu) {
+        internalFreeX86((*emu)->test.emu);
+        box_free((*emu)->test.emu);
+        (*emu)->test.emu = NULL;
+    }
    internalFreeX86(*emu);

    box_free(*emu);
@@ -215,19 +214,47 @@ void CloneEmu(x86emu_t *newemu, const x86emu_t* emu)
 	memcpy(newemu->mmx, emu->mmx, sizeof(emu->mmx));
    memcpy(newemu->fpu_ld, emu->fpu_ld, sizeof(emu->fpu_ld));
    memcpy(newemu->fpu_ll, emu->fpu_ll, sizeof(emu->fpu_ll));
-	memcpy(newemu->p_regs, emu->p_regs, sizeof(emu->p_regs));
+	newemu->fpu_tags = emu->fpu_tags;
 	newemu->cw = emu->cw;
 	newemu->sw = emu->sw;
 	newemu->top = emu->top;
    newemu->fpu_stack = emu->fpu_stack;
    memcpy(newemu->xmm, emu->xmm, sizeof(emu->xmm));
+    newemu->df = emu->df;
+    newemu->op1 = emu->op1;
+    newemu->op2 = emu->op2;
+    newemu->res = emu->res;
+    newemu->mxcsr = emu->mxcsr;
+    newemu->quit = emu->quit;
+    newemu->error = emu->error;
+}
+
+void CopyEmu(x86emu_t *newemu, const x86emu_t* emu)
+{
+	memcpy(newemu->regs, emu->regs, sizeof(emu->regs));
+    memcpy(&newemu->ip, &emu->ip, sizeof(emu->ip));
+	memcpy(&newemu->eflags, &emu->eflags, sizeof(emu->eflags));
+    newemu->old_ip = emu->old_ip;
+    memcpy(newemu->segs, emu->segs, sizeof(emu->segs));
+    memcpy(newemu->segs_serial, emu->segs_serial, sizeof(emu->segs_serial));
+    memcpy(newemu->segs_offs, emu->segs_offs, sizeof(emu->segs_offs));
+	memcpy(newemu->x87, emu->x87, sizeof(emu->x87));
+	memcpy(newemu->mmx, emu->mmx, sizeof(emu->mmx));
+    memcpy(newemu->xmm, emu->xmm, sizeof(emu->xmm));
+    memcpy(newemu->fpu_ld, emu->fpu_ld, sizeof(emu->fpu_ld));
+    memcpy(newemu->fpu_ll, emu->fpu_ll, sizeof(emu->fpu_ll));
+	newemu->fpu_tags = emu->fpu_tags;
+	newemu->cw = emu->cw;
+	newemu->sw = emu->sw;
+	newemu->top = emu->top;
+    newemu->fpu_stack = emu->fpu_stack;
+    newemu->df = emu->df;
+    newemu->op1 = emu->op1;
+    newemu->op2 = emu->op2;
+    newemu->res = emu->res;
    newemu->mxcsr = emu->mxcsr;
    newemu->quit = emu->quit;
    newemu->error = emu->error;
-    // adapt R_ESP to new stack frame
-    uintptr_t oldst = (uintptr_t)((emu->init_stack)?emu->init_stack:emu->context->stack);
-    uintptr_t newst = (uintptr_t)((newemu->init_stack)?newemu->init_stack:newemu->context->stack);
-    newemu->regs[_SP].dword[0] = emu->regs[_SP].dword[0] + (intptr_t)(newst - oldst);
 }

 box86context_t* GetEmuContext(x86emu_t* emu)
@@ -301,7 +328,8 @@ void ResetFlags(x86emu_t *emu)
 const char* DumpCPURegs(x86emu_t* emu, uintptr_t ip)
 {
    static char buff[800];
-    char* regname[] = {"EAX", "ECX", "EDX", "EBX", "ESP", "EBP", "ESI", "EDI"};
+    static const char* regname[] = {"EAX", "ECX", "EDX", "EBX", "ESP", "EBP", "ESI", "EDI"};
+    static const char* segname[] = {"ES", "CS", "SS", "DS", "FS", "GS"};
    char tmp[80];
    buff[0] = '\0';
    if(trace_emm) {
@@ -322,21 +350,27 @@ const char* DumpCPURegs(x86emu_t* emu, uintptr_t ip)
    }
    // start with FPU regs...
    if(emu->fpu_stack) {
-        if(emu->fpu_stack<9)
-            for (int i=0; i<emu->fpu_stack; i++) {
-                //if(i==4) strcat(buff, "\n");
-                sprintf(tmp, "ST%d=%f", i, emu->x87[(emu->top+i)&7].d);
-                strcat(buff, tmp);
-                int c = 10-strlen(tmp);
-                if(c<1) c=1;
-                while(c--) strcat(buff, " ");
-            }
-        else {
-            sprintf(tmp, "Warning, incoherent fpu_stack=%d", emu->fpu_stack);
+        int stack = emu->fpu_stack;
+        if(stack>8) stack = 8;
+        for (int i=0; i<stack; i++) {
+            sprintf(tmp, "ST%d=%f", i, ST(i).d);
            strcat(buff, tmp);
+            int c = 10-strlen(tmp);
+            if(c<1) c=1;
+            while(c--) strcat(buff, " ");
+            if(i==3) strcat(buff, "\n");
        }
+        sprintf(tmp, " C3210 = %d%d%d%d", emu->sw.f.F87_C3, emu->sw.f.F87_C2, emu->sw.f.F87_C1, emu->sw.f.F87_C0);
+        strcat(buff, tmp);
        strcat(buff, "\n");
    }
+    for (int i=0; i<6; ++i) {
+            sprintf(tmp, "%s=0x%04x", segname[i], emu->segs[i]);
+            strcat(buff, tmp);
+            if(i!=_GS)
+                strcat(buff, " ");
+    }
+    strcat(buff, "\n");
    for (int i=_AX; i<=_DI; ++i) {
        sprintf(tmp, "%s=%08x ", regname[i], emu->regs[i].dword[0]);
        strcat(buff, tmp);
@@ -386,8 +420,6 @@ void UnimpOpcode(x86emu_t* emu)
        tid, (void*)emu->old_ip, Peek(emu, -1),
        Peek(emu, 0), Peek(emu, 1), Peek(emu, 2), Peek(emu, 3),
        Peek(emu, 4), Peek(emu, 5), Peek(emu, 6), Peek(emu, 7));
-    emu->quit=1;
-    emu->error |= ERR_UNIMPL;
 }

 void EmuCall(x86emu_t* emu, uintptr_t addr)
@@ -404,9 +436,9 @@ void EmuCall(x86emu_t* emu, uintptr_t addr)
    Run(emu, 0);
    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
+    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 {
        R_EBX = old_ebx;
        R_EDI = old_edi;
@@ -458,14 +490,15 @@ static inline void init_perfcounters (int32_t do_reset, int32_t enable_divider)

 uint64_t ReadTSC(x86emu_t* emu)
 {
+    (void)emu;
    // Read the TimeStamp Counter as 64bits.
    // this is supposed to be the number of instructions executed since last reset
 #if defined(__i386__)
    uint64_t ret;
    __asm__ volatile("rdtsc" : "=A"(ret));
    return ret;
-#if 0
 #elif defined(__ARM_ARCH)
+#if 0
 #if (__ARM_ARCH >= 6)
    static int init = 1;
    if(init) {
@@ -497,3 +530,17 @@ void ResetSegmentsCache(x86emu_t *emu)
        return;
    memset(emu->segs_serial, 0, sizeof(emu->segs_serial));
 }
+
+void applyFlushTo0(x86emu_t* emu)
+{
+#ifdef __i386__
+    _mm_setcsr(_mm_getcsr() | (emu->mxcsr.x32&0x8040));
+#elif defined(__arm__)
+    uint64_t fpscr = __builtin_arm_get_fpscr();
+    fpscr &= ~(1<<24);                     // clear bit FZ (24)
+    fpscr |= (emu->mxcsr.f.MXCSR_FZ)<<24;  // set FZ as mxcsr FZ
+    __builtin_arm_set_fpscr(fpscr);
+#else
+    (void)emu;
+#endif
+}
@@ -23,11 +23,38 @@ typedef struct forkpty_s {
    void*   f;  // forkpty function
 } forkpty_t;

+typedef struct x86emu_s x86emu_t;
+
+typedef struct x86test_s {
+    x86emu_t*   emu;
+    uintptr_t   memaddr;
+    int         memsize;
+    int         test;
+    int         clean;
+    uint8_t     mem[16];
+} x86test_t;
+
+typedef struct emu_flags_s {
+    uint32_t    need_jmpbuf:1;    // need a new jmpbuff for signal handling
+    uint32_t    quitonlongjmp:2;  // quit if longjmp is called
+    uint32_t    quitonexit:2;     // quit if exit/_exit is called
+    uint32_t    longjmp:1;        // if quit because of longjmp
+    uint32_t    jmpbuf_ready:1;   // the jmpbuf in the emu is ok and don't need refresh
+} emu_flags_t;
+
+#ifdef ANDROID
+#include <setjmp.h>
+#define JUMPBUFF sigjmp_buf
+#else
+#define JUMPBUFF struct __jmp_buf_tag
+#endif
+
 typedef struct x86emu_s {
    // cpu
 	reg32_t     regs[8];
 	x86flags_t  eflags;
    reg32_t     ip;
+    uintptr_t   xSPSave;
    // fpu / mmx
 	x87control_t cw;
 	x87flags_t  sw;
@@ -35,12 +62,12 @@ typedef struct x86emu_s {
 	mmx87_regs_t mmx[8];
 	uint32_t    top;        // top is part of sw, but it's faster to have it separatly
    int         fpu_stack;
+	uint32_t    fpu_tags; // tags for the x87 regs, stacked, only on a 16bits anyway
    fpu_ld_t    fpu_ld[8]; // for long double emulation / 80bits fld fst
    fpu_ll_t    fpu_ll[8]; // for 64bits fild / fist sequence
-	fpu_p_reg_t p_regs[8];
    // sse
    sse_regs_t  xmm[8];
-    uint32_t    mxcsr;
+    mmxcontrol_t mxcsr;
    uintptr_t   old_ip;
    // defered flags
    defered_flags_t df;
@@ -52,17 +79,18 @@ typedef struct x86emu_s {
    uintptr_t   prev2_ip, prev_ip;
    #endif
    // segments
-    uint32_t    segs[6];        // only 32bits value?
+    uint16_t    segs[6];
+    uint16_t    dummy_seg6, dummy_seg7; // to stay aligned
    uintptr_t   segs_offs[6];   // computed offset associate with segment
    uint32_t    segs_serial[6];  // are seg offset clean (not 0) or does they need to be re-computed (0)? For GS, serial need to be the same as context->sel_serial
    // emu control
    int         quit;
    int         error;
    int         fork;   // quit because need to fork
-    forkpty_t*  forkpty_info;
    int         exit;
-    int         quitonlongjmp;  // quit if longjmp is called
-    int         longjmp;        // if quit because of longjmp
+    forkpty_t*  forkpty_info;
+    emu_flags_t flags;
+    x86test_t   test;       // used for dynarec testing
    // parent context
    box86context_t *context;
    // cpu helpers
@@ -74,6 +102,8 @@ typedef struct x86emu_s {
    void*       stack2free; // this is the stack to free (can be NULL)
    void*       init_stack; // initial stack (owned or not)
    uint32_t    size_stack; // stack size (owned or not)
+    JUMPBUFF*   jmpbuf;
+    uintptr_t   old_savedsp;

    i386_ucontext_t *uc_link; // to handle setcontext

@@ -88,4 +118,6 @@ typedef struct x86emu_s {
 //#define INTR_RAISE_DIV0(emu) {emu->error |= ERR_DIVBY0; emu->quit=1;}
 #define INTR_RAISE_DIV0(emu) {emu->error |= ERR_DIVBY0;} // should rise a SIGFPE and not quit

+void applyFlushTo0(x86emu_t* emu);
+
 #endif //__X86EMU_PRIVATE_H_
@@ -10,6 +10,7 @@
 #include <sys/types.h>
 #include <pthread.h>
 #include <signal.h>
+#include <wait.h>

 #include "debug.h"
 #include "box86stack.h"
@@ -42,7 +43,9 @@ x86emu_t* x86emu_fork(x86emu_t* emu, int forktype)
        iFpppp_t forkpty = (iFpppp_t)emu->forkpty_info->f;
        v = forkpty(emu->forkpty_info->amaster, emu->forkpty_info->name, emu->forkpty_info->termp, emu->forkpty_info->winp);
        emu->forkpty_info = NULL;
-    } else
+    } else if(forktype==3)
+        v = vfork();
+    else
        v = fork();
    if(type == EMUTYPE_MAIN)
        thread_set_emu(emu);
@@ -50,20 +53,19 @@ x86emu_t* x86emu_fork(x86emu_t* emu, int forktype)
        // error...
    } else if(v!=0) {  
        // execute atforks parent functions
-        for (int i=0; i<my_context->atfork_sz; --i)
+        for (int i=0; i<my_context->atfork_sz; ++i)
            if(my_context->atforks[i].parent)
                EmuCall(emu, my_context->atforks[i].parent);
-
    } else if(v==0) {
        // execute atforks child functions
-        for (int i=0; i<my_context->atfork_sz; --i)
+        for (int i=0; i<my_context->atfork_sz; ++i)
            if(my_context->atforks[i].child)
                EmuCall(emu, my_context->atforks[i].child);
    }
    R_EAX = v;
    return emu;
 }
-
+void emit_signal(x86emu_t* emu, int sig, void* addr, int code); // in signals.c
 extern int errno;
 void x86Int3(x86emu_t* emu)
 {
@@ -87,6 +89,8 @@ void x86Int3(x86emu_t* emu)
            if(h && strstr(ElfName(h), "libMiles")) have_trace = 1;*/
            if(box86_log>=LOG_DEBUG  || cycle_log) {
                int tid = GetTID();
+                const char *s = NULL;
+                s = GetNativeName((void*)addr);
                char t_buff[256] = "\0";
                char buff2[64] = "\0";
                char buff3[64] = "\0";
@@ -99,210 +103,223 @@ void x86Int3(x86emu_t* emu)
                int post = 0;
                int perr = 0;
                uint32_t *pu32 = NULL;
-                const char *s = NULL;
-                s = GetNativeName((void*)addr);
                if(addr==(uintptr_t)PltResolver) {
                    snprintf(buff, 256, "%s", cycle_log?"PltResolver ":" ... ");
                } else
                if(strstr(s, "SDL_RWFromFile")==s || strstr(s, "SDL_RWFromFile")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%s, %s)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%s, %s)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
                } else  if(strstr(s, "glColor4f")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%f, %f, %f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8), *(float*)(R_ESP+12), *(float*)(R_ESP+16));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%f, %f, %f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8), *(float*)(R_ESP+12), *(float*)(R_ESP+16));
                } else  if(strstr(s, "glTexCoord2f")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8));
                } else  if(strstr(s, "glVertex2f")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8));
                } else  if(strstr(s, "glVertex3f")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%f, %f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8), *(float*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%f, %f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8), *(float*)(R_ESP+12));
                } else  if(strstr(s, "__open64")==s || strcmp(s, "open64")==0) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", %d, %d)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(int*)(R_ESP+8), *(int*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", %d, %d)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(int*)(R_ESP+8), *(int*)(R_ESP+12));
                    perr = 1;
                } else  if(!strcmp(s, "opendir")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
                    perr = 1;
                } else  if(strstr(s, "__open")==s || !strcmp(s, "open") || !strcmp(s, "my_open64")) {
                    tmp = *(char**)(R_ESP+4);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", %d (,%d))", tid, *(void**)(R_ESP), s, (tmp)?tmp:"(nil)", *(int*)(R_ESP+8), *(int*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", %d (,%d))", tid, *(void**)(R_ESP), s, (tmp)?tmp:"(nil)", *(int*)(R_ESP+8), *(int*)(R_ESP+12));
                    perr = 1;
                } else  if(!strcmp(s, "shm_open")) {
                    tmp = *(char**)(R_ESP+4);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", %d, %d)", tid, *(void**)(R_ESP), s, (tmp)?tmp:"(nil)", *(int*)(R_ESP+8), *(int*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", %d, %d)", tid, *(void**)(R_ESP), s, (tmp)?tmp:"(nil)", *(int*)(R_ESP+8), *(int*)(R_ESP+12));
                    perr = 1;
                } else  if(strcmp(s, "mkdir")==0) {
                    tmp = *(char**)(R_ESP+4);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", %d)", tid, *(void**)(R_ESP), s, (tmp)?tmp:"(nil)", *(int*)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", %d)", tid, *(void**)(R_ESP), s, (tmp)?tmp:"(nil)", *(int*)(R_ESP+8));
                    perr = 1;
                } else  if(!strcmp(s, "fopen")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
                    perr = 2;
                } else  if(!strcmp(s, "freopen")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", \"%s\", %p)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", \"%s\", %p)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
                    perr = 2;
                } else  if(!strcmp(s, "fopen64")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
                    perr = 2;
                } else  if(!strcmp(s, "chdir")) {
                    pu32=*(uint32_t**)(R_ESP+4);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, pu32?((pu32==(uint32_t*)1)?"/1/":(char*)pu32):"/0/");
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, pu32?((pu32==(uint32_t*)1)?"/1/":(char*)pu32):"/0/");
                } else  if(strstr(s, "getenv")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
                    post = 2;
                } else  if(strstr(s, "putenv")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
                } else  if(strstr(s, "pread")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p, %u, %d)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(void**)(R_ESP+8), *(uint32_t*)(R_ESP+12), *(int32_t*)(R_ESP+16));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p, %u, %d)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(void**)(R_ESP+8), *(uint32_t*)(R_ESP+12), *(int32_t*)(R_ESP+16));
                    perr = 1;
                } else  if(!strcmp(s, "read")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p, %u)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(void**)(R_ESP+8), *(uint32_t*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p, %u)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(void**)(R_ESP+8), *(uint32_t*)(R_ESP+12));
                    perr = 1;
                } else  if(strstr(s, "ioctl")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, 0x%x, %p)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(int32_t*)(R_ESP+8), *(void**)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, 0x%x, %p)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(int32_t*)(R_ESP+8), *(void**)(R_ESP+12));
                    perr = 1;
                } else  if(strstr(s, "statvfs64")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p(\"%s\"), %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+4), *(void**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p(\"%s\"), %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+4), *(void**)(R_ESP+8));
                } else  if(strstr(s, "index")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p(\"%s\"), %i(%c))", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+4), *(int32_t*)(R_ESP+8), *(int32_t*)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p(\"%s\"), %i(%c))", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+4), *(int32_t*)(R_ESP+8), *(int32_t*)(R_ESP+8));
                } else  if(strstr(s, "rindex")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p(\"%s\"), %i(%c))", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+4), *(int32_t*)(R_ESP+8), *(int32_t*)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p(\"%s\"), %i(%c))", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+4), *(int32_t*)(R_ESP+8), *(int32_t*)(R_ESP+8));
                } else  if(strstr(s, "__xstat64")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p(\"%s\"), %p)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(char**)(R_ESP+8), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p(\"%s\"), %p)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(char**)(R_ESP+8), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
                    perr = 1;
                } else  if(strcmp(s, "__xstat")==0) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p(\"%s\"), %p)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(char**)(R_ESP+8), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p(\"%s\"), %p)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(char**)(R_ESP+8), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
                    perr = 1;
                } else  if(strstr(s, "__lxstat64")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p(\"%s\"), %p)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(char**)(R_ESP+8), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p(\"%s\"), %p)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(char**)(R_ESP+8), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
                    perr = 1;
                } else  if(strstr(s, "sem_timedwait")==s) {
                    pu32 = *(uint32_t**)(R_ESP+8);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, %p[%d sec %d ns])", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(void**)(R_ESP+8), pu32?pu32[0]:-1, pu32?pu32[1]:-1);
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, %p[%d sec %d ns])", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(void**)(R_ESP+8), pu32?(int32_t)pu32[0]:-1, pu32?(int32_t)pu32[1]:-1);
                    perr = 1;
                } else  if(strstr(s, "waitpid")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p, 0x%x)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(void**)(R_ESP+8), *(uint32_t*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p, 0x%x)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(void**)(R_ESP+8), *(uint32_t*)(R_ESP+12));
                    perr = 1;
                } else  if(strstr(s, "clock_gettime")==s || strstr(s, "__clock_gettime")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(void**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(void**)(R_ESP+8));
                    post = 1;
                    pu32 = *(uint32_t**)(R_ESP+8);
                } else  if(strstr(s, "semop")==s) {
                    int16_t* p16 = *(int16_t**)(R_ESP+8);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p[%u/%d/0x%x], %d)", tid, *(void**)(R_ESP), s, *(int*)(R_ESP+4), p16, p16[0], p16[1], p16[2], *(int*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p[%u/%d/0x%x], %d)", tid, *(void**)(R_ESP), s, *(int*)(R_ESP+4), p16, p16[0], p16[1], p16[2], *(int*)(R_ESP+12));
                    perr = 1;
                } else  if(!strcmp(s, "mmap64")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, 0x%x, %d, 0x%x, %d, %lld)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(int*)(R_ESP+20), *(int64_t*)(R_ESP+24));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, 0x%x, %d, 0x%x, %d, %lld)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(int*)(R_ESP+20), *(int64_t*)(R_ESP+24));
                    perr = 3;
                } else  if(!strcmp(s, "mmap")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, 0x%x, %d, 0x%x, %d, %d)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(int*)(R_ESP+20), *(int*)(R_ESP+24));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, 0x%x, %d, 0x%x, %d, %d)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(int*)(R_ESP+20), *(int*)(R_ESP+24));
                    perr = 3;
                } else  if(strstr(s, "strcasecmp")==s || strstr(s, "__strcasecmp")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                } else  if(strstr(s, "wcsncasecmp")==s) {
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%S\", \"%S\", %d)", tid, *(void**)(R_ESP), s, *(wchar_t**)(R_ESP+4), *(wchar_t**)(R_ESP+8), *(int*)(R_ESP+12));
                } else  if(strstr(s, "gtk_signal_connect_full")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, \"%s\", %p, %p, %p, %p, %d, %d)", tid, *(void**)(R_ESP), "gtk_signal_connect_full", *(void**)(R_ESP+4), *(char**)(R_ESP+8), *(void**)(R_ESP+12), *(void**)(R_ESP+16), *(void**)(R_ESP+20), *(void**)(R_ESP+24), *(int32_t*)(R_ESP+28), *(int32_t*)(R_ESP+32));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, \"%s\", %p, %p, %p, %p, %d, %d)", tid, *(void**)(R_ESP), "gtk_signal_connect_full", *(void**)(R_ESP+4), *(char**)(R_ESP+8), *(void**)(R_ESP+12), *(void**)(R_ESP+16), *(void**)(R_ESP+20), *(void**)(R_ESP+24), *(int32_t*)(R_ESP+28), *(int32_t*)(R_ESP+32));
                } else  if(strstr(s, "strcmp")==s || strstr(s, "__strcmp")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
                } else  if(strstr(s, "strstr")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%.127s\", \"%.127s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%.127s\", \"%.127s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
                } else  if(strstr(s, "strlen")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p(\"%s\"))", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), ((R_ESP+4))?(*(char**)(R_ESP+4)):"nil");
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p(\"%s\"))", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), ((R_ESP+4))?(*(char**)(R_ESP+4)):"nil");
                } else  if(strstr(s, "vsnprintf")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%08X, %u, %08X...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(uint32_t*)(R_ESP+8), *(uint32_t*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%08X, %u, %p(\"%s\")...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(uint32_t*)(R_ESP+8), *(void**)(R_ESP+12), *(char**)(R_ESP+12));
                    pu32 = *(uint32_t**)(R_ESP+4);
                    post = 3;
                } else  if(strstr(s, "vsprintf")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, \"%s\", %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, \"%s\", %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8), *(void**)(R_ESP+12));
                    pu32 = *(uint32_t**)(R_ESP+4);
                    post = 3;
                } else  if(strstr(s, "__vsprintf_chk")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, %d, %zu, \"%s\", %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(int*)(R_ESP+8), *(size_t*)(R_ESP+12), *(char**)(R_ESP+16), *(void**)(R_ESP+20));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, %d, %zu, \"%s\", %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(int*)(R_ESP+8), *(size_t*)(R_ESP+12), *(char**)(R_ESP+16), *(void**)(R_ESP+20));
                    pu32 = *(uint32_t**)(R_ESP+4);
                    post = 3;
                } else  if(strstr(s, "__snprintf_chk")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, %zu, %d, %d, \"%s\", %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(char**)(R_ESP+20), *(void**)(R_ESP+24));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, %zu, %d, %d, \"%s\", %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(char**)(R_ESP+20), *(void**)(R_ESP+24));
                    pu32 = *(uint32_t**)(R_ESP+4);
                    post = 3;
                } else  if(strstr(s, "snprintf")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, %zu, \"%s\", ...)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(char**)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, %zu, \"%s\", ...)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(char**)(R_ESP+12));
                    pu32 = *(uint32_t**)(R_ESP+4);
                    post = 3;
                } else  if(strstr(s, "sprintf")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%08X, %08X...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(uint32_t*)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%08X, %08X...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(uint32_t*)(R_ESP+8));
                    pu32 = *(uint32_t**)(R_ESP+4);
                    post = 3;
                } else  if(strstr(s, "printf")==s) {
                    pu32 = *(uint32_t**)(R_ESP+4);
                    if(((uintptr_t)pu32)<0x5) // probably a _chk function
                        pu32 = *(uint32_t**)(R_ESP+8);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\"...)", tid, *(void**)(R_ESP), s, pu32?((char*)(pu32)):"nil");
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\"...)", tid, *(void**)(R_ESP), s, pu32?((char*)(pu32)):"nil");
                } else  if(strstr(s, "wprintf")==s) {
                    pu32 = *(uint32_t**)(R_ESP+4);
                    if(((uintptr_t)pu32)<0x5) // probably a _chk function
                        pu32 = *(uint32_t**)(R_ESP+8);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%S\"...)", tid, *(void**)(R_ESP), s, pu32?((wchar_t*)(pu32)):L"nil");
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%S\"...)", tid, *(void**)(R_ESP), s, pu32?((wchar_t*)(pu32)):L"nil");
                } else  if(strstr(s, "__vswprintf")==s) {
                    if(*(size_t*)(R_ESP+12)<2) {
-                        snprintf(buff, 255, "%04d|%p: Calling %s(%p, %zu, %p, %p, %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(void**)(R_ESP+12), *(void**)(R_ESP+16), *(void**)(R_ESP+20));
+                        snprintf(buff, 256, "%04d|%p: Calling %s(%p, %zu, %p, %p, %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(void**)(R_ESP+12), *(void**)(R_ESP+16), *(void**)(R_ESP+20));
                    } else {
-                        snprintf(buff, 255, "%04d|%p: Calling %s(%p, %zu, \"%S\", %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(wchar_t**)(R_ESP+12), *(void**)(R_ESP+16));
+                        snprintf(buff, 256, "%04d|%p: Calling %s(%p, %zu, \"%S\", %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(size_t*)(R_ESP+8), *(wchar_t**)(R_ESP+12), *(void**)(R_ESP+16));
                        pu32 = *(uint32_t**)(R_ESP+4);
                        post = 6;
                    }
                } else  if(strstr(s, "puts")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\"...)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\"...)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
                } else  if(strstr(s, "fputs")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", %p...)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(void**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", %p...)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(void**)(R_ESP+8));
                } else  if(strstr(s, "fprintf")==s) {
                    pu32 = *(uint32_t**)(R_ESP+8);
                    if(((uintptr_t)pu32)<0x5) // probably a __fprint_chk
                        pu32 = *(uint32_t**)(R_ESP+12);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%08X, \"%s\", ...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), pu32?((char*)(pu32)):"nil");
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%08X, \"%s\", ...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), pu32?((char*)(pu32)):"nil");
                } else  if(strstr(s, "vfprintf")==s) {
                    pu32 = *(uint32_t**)(R_ESP+8);
                    if(((uintptr_t)pu32)<0x5) // probably a _chk function
                        pu32 = *(uint32_t**)(R_ESP+12);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%08X, \"%s\", ...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), pu32?((char*)(pu32)):"nil");
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%08X, \"%s\", ...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), pu32?((char*)(pu32)):"nil");
                } else  if(strstr(s, "vkGetInstanceProcAddr")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, \"%s\")", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, \"%s\")", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8));
                } else  if(strstr(s, "vkGetDeviceProcAddr")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, \"%s\")", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, \"%s\")", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8));
                } else  if(strstr(s, "glXGetProcAddress")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\")", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4));
+                } else if (!strcmp(s, "glTexImage2D")) {
+                    snprintf(buff, 256, "%04d|%p: Calling %s(0x%x, %d, 0x%x, %d, %d, %d, 0x%x, 0x%x, %p)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(int*)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(int*)(R_ESP+20), *(int*)(R_ESP+24), *(uint32_t*)(R_ESP+28), *(uint32_t*)(R_ESP+32), *(void**)(R_ESP+36));
                } else  if(strstr(s, "sscanf")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", \"%s\", ...)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", \"%s\", ...)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
                } else  if(!strcmp(s, "vsscanf")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", \"%s\", ...)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(\"%s\", \"%s\", ...)", tid, *(void**)(R_ESP), s, *(char**)(R_ESP+4), *(char**)(R_ESP+8));
                } else if(strstr(s, "XCreateWindow")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, %p, %d, %d, %u, %u, %u, %d, %u, %p, %u, %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(void**)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(uint32_t*)(R_ESP+20), *(uint32_t*)(R_ESP+24), *(uint32_t*)(R_ESP+28), *(int32_t*)(R_ESP+32), *(uint32_t*)(R_ESP+36), *(void**)(R_ESP+40), *(uint32_t*)(R_ESP+44), *(void**)(R_ESP+48));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, %p, %d, %d, %u, %u, %u, %d, %u, %p, %u, %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(void**)(R_ESP+8), *(int*)(R_ESP+12), *(int*)(R_ESP+16), *(uint32_t*)(R_ESP+20), *(uint32_t*)(R_ESP+24), *(uint32_t*)(R_ESP+28), *(int32_t*)(R_ESP+32), *(uint32_t*)(R_ESP+36), *(void**)(R_ESP+40), *(uint32_t*)(R_ESP+44), *(void**)(R_ESP+48));
                } else if(strstr(s, "XLoadQueryFont")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, \"%s\")", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, \"%s\")", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8));
                } else if(strstr(s, "pthread_mutex_lock")==s) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4));
                } else if(!strcmp(s, "fmodf")) {
                    post = 4;
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%f, %f)", tid, *(void**)(R_ESP), s, *(float*)(R_ESP+4), *(float*)(R_ESP+8));
                } else if(!strcmp(s, "fmod")) {
                    post = 4;
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%f, %f)", tid, *(void**)(R_ESP), s, *(double*)(R_ESP+4), *(double*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%f, %f)", tid, *(void**)(R_ESP), s, *(double*)(R_ESP+4), *(double*)(R_ESP+12));
                } else if(strstr(s, "SDL_GetWindowSurface")==s) {
                    post = 5;
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4));
                } else if(strstr(s, "udev_monitor_new_from_netlink")==s) {
                    post = 5;
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%p, \"%s\")", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, \"%s\")", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(char**)(R_ESP+8));
                } else  if(!strcmp(s, "syscall")) {
-                    snprintf(buff, 255, "%04d|%p: Calling %s(%d, %p, %p, %p...)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(void**)(R_ESP+8), *(void**)(R_ESP+12), *(void**)(R_ESP+16));
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%d, %p, %p, %p...)", tid, *(void**)(R_ESP), s, *(int32_t*)(R_ESP+4), *(void**)(R_ESP+8), *(void**)(R_ESP+12), *(void**)(R_ESP+16));
                    perr = 1;
+                } else  if(strstr(s, "___tls_get_addr")) {
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p[%d, %d])", tid, *(void**)(R_ESP), s, (void*)(R_EAX), ((int*)(R_EAX))[0], ((int*)(R_EAX))[1]);
+                } else  if(strstr(s, "__tls_get_addr")) {
+                    pu32 = *(uint32_t**)(R_ESP+4);
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p[%d, %d])", tid, *(void**)(R_ESP), s, pu32, pu32[0], pu32[1]);
+                } else if (!strcmp(s, "FT_Outline_Get_CBox")) {
+                    pu32 = *(uint32_t**)(R_ESP+8);
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, %p)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(void**)(R_ESP+8));
+                    post = 7;
+                } else if (!strcmp(s, "XDrawImageString")) {
+                    snprintf(buff, 256, "%04d|%p: Calling %s(%p, %p, %p, %i, %i, %p\"%s\", %d)", tid, *(void**)(R_ESP), s, *(void**)(R_ESP+4), *(void**)(R_ESP+8), *(void**)(R_ESP+12), *(int*)(R_ESP+16), *(int*)(R_ESP+20), *(char**)(R_ESP+24), *(char**)(R_ESP+24), *(int*)(R_ESP+28));
                } else {
-                    snprintf(buff, 255, "%04d|%p: Calling %s (%08X, %08X, %08X...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(uint32_t*)(R_ESP+8), *(uint32_t*)(R_ESP+12));
+                    snprintf(buff, 256, "%04d|%p: Calling %s (%08X, %08X, %08X...)", tid, *(void**)(R_ESP), s, *(uint32_t*)(R_ESP+4), *(uint32_t*)(R_ESP+8), *(uint32_t*)(R_ESP+12));
                }
                if(!cycle_log) {
-                    pthread_mutex_lock(&emu->context->mutex_trace);
+                    mutex_lock(&emu->context->mutex_trace);
                    printf_log(LOG_NONE, "%s =>", buff);
-                    pthread_mutex_unlock(&emu->context->mutex_trace);
+                    mutex_unlock(&emu->context->mutex_trace);
                }
                w(emu, addr);   // some function never come back, so unlock the mutex first!
                if(post)
                    switch(post) {
-                    case 1: snprintf(buff2, 63, " [%d sec %d nsec]", pu32?pu32[0]:-1, pu32?pu32[1]:-1);
+                    case 1: snprintf(buff2, 63, " [%d sec %d nsec]", pu32?(int32_t)pu32[0]:-1, pu32?(int32_t)pu32[1]:-1);
                            break;
                    case 2: snprintf(buff2, 63, "(%s)", R_EAX?((char*)R_EAX):"nil");
                            break;
@@ -320,6 +337,8 @@ void x86Int3(x86emu_t* emu)
                            break;
                    case 6: snprintf(buff2, 63, "(%S)", pu32?((wchar_t*)pu32):L"nil");
                            break;
+                    case 7: snprintf(buff2, 63, " [%d / %d / %d /%d]", pu32[0], pu32[1], pu32[2], pu32[3]);
+                            break;
                }
                if(perr==1 && ((int)R_EAX)<0)
                    snprintf(buff3, 63, " (errno=%d:\"%s\")", errno, strerror(errno));
@@ -330,17 +349,17 @@ void x86Int3(x86emu_t* emu)
                if(cycle_log)
                    snprintf(buffret, 127, "0x%X%s%s", R_EAX, buff2, buff3);
                else {
-                    pthread_mutex_lock(&emu->context->mutex_trace);
+                    mutex_lock(&emu->context->mutex_trace);
                    printf_log(LOG_NONE, " return 0x%08X%s%s\n", R_EAX, buff2, buff3);
-                    pthread_mutex_unlock(&emu->context->mutex_trace);
+                    mutex_unlock(&emu->context->mutex_trace);
                }
            } else
                w(emu, addr);
        }
        return;
    }
-    if(0 && my_context->signals[SIGTRAP])
-        raise(SIGTRAP);
+    if(1 && my_context->signals[SIGTRAP])
+        emit_signal(emu, SIGTRAP, (void*)R_EIP, 128);
    else
        printf_log(LOG_INFO, "%04d|Warning, ignoring unsupported Int 3 call @%p\n", GetTID(), (void*)R_EIP);
    //emu->quit = 1;
@@ -362,4 +381,4 @@ void print_cycle_log(int loglevel) {
            }
        }
    }
-}
+}
@@ -484,10 +484,8 @@ uint8_t rcl8(x86emu_t *emu, uint8_t d, uint8_t s)
 		CONDITIONAL_SET_FLAG(cf, F_CF);
        /* OVERFLOW is set *IFF* cnt==1, then it is the 
           xor of CF and the most significant bit.  Blecck. */
-        /* parenthesized this expression since it appears to
-           be causing OF to be misset */
-        CONDITIONAL_SET_FLAG(cnt == 1 && XOR2(cf + ((res >> 6) & 0x2)),
-							 F_OF);
+		if(cnt == 1)
+        	CONDITIONAL_SET_FLAG((cf ^ (res >> 7)) & 0x1, F_OF);

    }
 	return (uint8_t)res;
@@ -513,8 +511,8 @@ uint16_t rcl16(x86emu_t *emu, uint16_t d, uint8_t s)
 			res |= 1 << (cnt - 1);
 		}
 		CONDITIONAL_SET_FLAG(cf, F_CF);
-		CONDITIONAL_SET_FLAG(cnt == 1 && XOR2(cf + ((res >> 14) & 0x2)),
-							 F_OF);
+		if(cnt == 1)
+        	CONDITIONAL_SET_FLAG((cf ^ (res >> 15)) & 0x1, F_OF);
 	}
 	return (uint16_t)res;
 }
@@ -539,8 +537,8 @@ uint32_t rcl32(x86emu_t *emu, uint32_t d, uint8_t s)
 			res |= 1 << (cnt - 1);
 		}
 		CONDITIONAL_SET_FLAG(cf, F_CF);
-		CONDITIONAL_SET_FLAG(cnt == 1 && XOR2(cf + ((res >> 30) & 0x2)),
-							 F_OF);
+		if(cnt == 1)
+        	CONDITIONAL_SET_FLAG((cf ^ (res >> 31)) & 0x1, F_OF);
 	}
 	return res;
 }
@@ -592,6 +590,11 @@ uint8_t rcr8(x86emu_t *emu, uint8_t d, uint8_t s)
               (i.e. packed bit array or unpacked.)
             */
 			ocf = ACCESS_FLAG(F_CF) != 0;
+			/* OVERFLOW is set *IFF* cnt==1, then it is the 
+			xor of CF and the most significant bit.  Blecck. */
+			/* parenthesized... */
+			CONDITIONAL_SET_FLAG((ocf ^ (d >> 7)) & 0x1,
+								 F_OF);
        } else
            cf = (d >> (cnt - 1)) & 0x1;

@@ -620,13 +623,6 @@ uint8_t rcr8(x86emu_t *emu, uint8_t d, uint8_t s)
        }
        /* set the new carry flag, based on the variable "cf" */
 		CONDITIONAL_SET_FLAG(cf, F_CF);
-        /* OVERFLOW is set *IFF* cnt==1, then it is the 
-           xor of CF and the most significant bit.  Blecck. */
-        /* parenthesized... */
-		if (cnt == 1) {
-			CONDITIONAL_SET_FLAG(XOR2(ocf + ((d >> 6) & 0x2)),
-								 F_OF);
-		}
 	}
 	return (uint8_t)res;
 }
@@ -648,6 +644,8 @@ uint16_t rcr16(x86emu_t *emu, uint16_t d, uint8_t s)
 		if (cnt == 1) {
 			cf = d & 0x1;
 			ocf = ACCESS_FLAG(F_CF) != 0;
+			CONDITIONAL_SET_FLAG((ocf ^ (d >> 15)) & 0x1,
+								 F_OF);
 		} else
 			cf = (d >> (cnt - 1)) & 0x1;
 		mask = (1 << (16 - cnt)) - 1;
@@ -657,10 +655,6 @@ uint16_t rcr16(x86emu_t *emu, uint16_t d, uint8_t s)
 			res |= 1 << (16 - cnt);
 		}
 		CONDITIONAL_SET_FLAG(cf, F_CF);
-		if (cnt == 1) {
-			CONDITIONAL_SET_FLAG(XOR2(ocf + ((d >> 14) & 0x2)),
-								 F_OF);
-		}
 	}
 	return (uint16_t)res;
 }
@@ -682,6 +676,8 @@ uint32_t rcr32(x86emu_t *emu, uint32_t d, uint8_t s)
 		if (cnt == 1) {
 			cf = d & 0x1;
 			ocf = ACCESS_FLAG(F_CF) != 0;
+			CONDITIONAL_SET_FLAG((ocf ^ (d >> 31)) & 0x1,
+								 F_OF);
 		} else
 			cf = (d >> (cnt - 1)) & 0x1;
 		mask = (1 << (32 - cnt)) - 1;
@@ -692,10 +688,6 @@ uint32_t rcr32(x86emu_t *emu, uint32_t d, uint8_t s)
 			res |= 1 << (32 - cnt);
 		}
 		CONDITIONAL_SET_FLAG(cf, F_CF);
-		if (cnt == 1) {
-			CONDITIONAL_SET_FLAG(XOR2(ocf + ((d >> 30) & 0x2)),
-								 F_OF);
-		}
 	}
 	return res;
 }
@@ -885,8 +877,7 @@ uint16_t shld16 (x86emu_t *emu, uint16_t d, uint16_t fill, uint8_t s)
 			res = d;
 		}
 		if (cnt == 1) {
-			CONDITIONAL_SET_FLAG((((res & 0x8000) == 0x8000) ^
-								  (ACCESS_FLAG(F_CF) != 0)), F_OF);
+			CONDITIONAL_SET_FLAG(((res ^ d) >> 15)&1, F_OF);
 		} else {
 			CLEAR_FLAG(F_OF);
 		}
@@ -927,8 +918,7 @@ uint32_t shld32 (x86emu_t *emu, uint32_t d, uint32_t fill, uint8_t s)
 		res = d;
 	}
 	if (cnt == 1) {
-		CONDITIONAL_SET_FLAG((((res & 0x80000000) == 0x80000000) ^
-								(ACCESS_FLAG(F_CF) != 0)), F_OF);
+		CONDITIONAL_SET_FLAG(((res ^ d) >> 31)&1, F_OF);
 	} else {
 		CLEAR_FLAG(F_OF);
 	}
@@ -959,7 +949,7 @@ uint16_t shrd16 (x86emu_t *emu, uint16_t d, uint16_t fill, uint8_t s)
 		}

 		if (cnt == 1) {
-			CONDITIONAL_SET_FLAG(XOR2(res >> 14), F_OF);
+			CONDITIONAL_SET_FLAG(((res ^ d) >> 15)&1, F_OF);
        } else {
 			CLEAR_FLAG(F_OF);
        }
@@ -1005,7 +995,7 @@ uint32_t shrd32 (x86emu_t *emu, uint32_t d, uint32_t fill, uint8_t s)
 		res = d;
 	}
 	if (cnt == 1) {
-		CONDITIONAL_SET_FLAG(XOR2(res >> 30), F_OF);
+		CONDITIONAL_SET_FLAG(((res ^ d) >> 31)&1, F_OF);
 	} else {
 		CLEAR_FLAG(F_OF);
 	}
@@ -1044,14 +1034,14 @@ Implements the SBB instruction and side effects.
 ****************************************************************************/
 uint16_t sbb16(x86emu_t *emu, uint16_t d, uint16_t s)
 {
-    uint32_t res;   /* all operands in native machine order */
-    uint32_t bc;
+	uint32_t res;   /* all operands in native machine order */
+	uint32_t bc;
 	CHECK_FLAGS(emu);

 	if (ACCESS_FLAG(F_CF))
-        res = d - s - 1;
-    else
-        res = d - s;
+		res = d - s - 1;
+	else
+		res = d - s;
 	CONDITIONAL_SET_FLAG(res & 0x8000, F_SF);
 	CONDITIONAL_SET_FLAG((res & 0xffff) == 0, F_ZF);
 	CONDITIONAL_SET_FLAG(PARITY(res & 0xff), F_PF);
@@ -1075,9 +1065,9 @@ uint32_t sbb32(x86emu_t *emu, uint32_t d, uint32_t s)
 	CHECK_FLAGS(emu);

 	if (ACCESS_FLAG(F_CF))
-        res = d - s - 1;
-    else
-        res = d - s;
+		res = d - s - 1;
+	else
+		res = d - s;
 	CONDITIONAL_SET_FLAG(res & 0x80000000, F_SF);
 	CONDITIONAL_SET_FLAG(!res, F_ZF);
 	CONDITIONAL_SET_FLAG(PARITY(res & 0xff), F_PF);
@@ -247,6 +247,7 @@ Implements the NOT instruction
 ****************************************************************************/
 static inline uint8_t not8(x86emu_t *emu, uint8_t s)
 {
+    (void)emu;
 	return ~s;
 }

@@ -256,6 +257,7 @@ Implements the NOT instruction
 ****************************************************************************/
 static inline uint16_t not16(x86emu_t *emu, uint16_t s)
 {
+    (void)emu;
 	return ~s;
 }

@@ -265,6 +267,7 @@ Implements the NOT instruction
 ****************************************************************************/
 static inline uint32_t not32(x86emu_t *emu, uint32_t s)
 {
+    (void)emu;
 	return ~s;
 }