Bumped version to v0.2.4

Update CHANGELOG.md
[DYNAREC] Small cosmetic change
2026-11-08 02:08:33 -10:00 · 2021-11-19 21:57:50 +01:00 · 2021-11-19 21:41:31 +01:00 · 2021-11-19 21:41:15 +01:00 · 2021-11-19 15:39:12 +01:00 · 2021-11-19 15:34:57 +01:00
868 changed files with 72365 additions and 18536 deletions
@@ -1,6 +1,6 @@
 # These are supported funding model platforms

-github: # Replace with up to 4 GitHub Sponsors-enabled usernames e.g., [user1, user2]
+github: ptitSeb
 patreon: # Replace with a single Patreon username
 open_collective: # Replace with a single Open Collective username
 ko_fi: # Replace with a single Ko-fi username
@@ -0,0 +1,88 @@
+### Auto build Box86 and release its binary with Github Action
+name: Build and Release Box86
+
+on: 
+  workflow_dispatch:
+  release:
+  push:
+    paths:
+      - "**/*.c"
+      - "**/*.h"
+      - "**/*.S"
+      - "**/*.py"
+      - "CMakeList.txt"
+      - "**/*.yml"
+  pull_request: 
+    types: [assigned, opened, synchronize, reopened]
+    paths:
+      - "**/*.c"
+      - "**/*.h"
+      - "**/*.S"
+      - "**/*.py"
+      - "CMakeList.txt"
+      - "**/*.yml"
+
+jobs:
+  build:
+    strategy:
+      fail-fast: false
+      matrix:
+        platform: [I386, OTHER_ARM, PANDORA, PYRA, GAMESHELL, RPI4, RPI4ARM64, ODROID, RK3399, RK3288, A64, SD845, PHYTIUM]
+        type: [Release, Trace]
+
+    runs-on: ubuntu-latest
+    steps:
+      - name: "Checkout Box86 Repository"
+        uses: actions/checkout@v2
+
+      - name: "Environment preparation"
+        run: |
+          sudo apt-get update
+
+          if [[ ${{ matrix.platform }} != 'I386' && ${{ matrix.platform }} != 'OTHER_ARM' ]]; then
+            echo "BOX86_PLATFORM_MARCRO=-D${{ matrix.platform }}=1" >> $GITHUB_ENV
+            echo "BOX86_COMPILER=arm-linux-gnueabihf-gcc" >> $GITHUB_ENV
+            sudo apt-get -y install git gcc-arm-linux-gnueabihf cmake make python3
+          else
+            if [[ ${{ matrix.platform }} == 'I386' ]]; then
+              echo "BOX86_PLATFORM_MARCRO=-DLD80BITS=1 -DNOALIGN=1" >> $GITHUB_ENV
+              echo "BOX86_COMPILER=/bin/i686-linux-gnu-gcc-10" >> $GITHUB_ENV
+              sudo apt-get -y install git libc6-dev-i386-amd64-cross gcc-10-i686-linux-gnu cmake make python3
+              ls /bin/*gcc*
+            else
+              echo BOX86_PLATFORM_MARCRO="-DARM_DYNAREC=ON -DCMAKE_C_FLAGS=-marm" >> $GITHUB_ENV
+              echo "BOX86_COMPILER=arm-linux-gnueabihf-gcc" >> $GITHUB_ENV
+              sudo apt-get -y install git gcc-arm-linux-gnueabihf cmake make python3
+            fi
+          fi
+
+          if [[ ${{ matrix.type }} == 'Release' ]]; then
+            echo BOX86_BUILD_TYPE=Release >> $GITHUB_ENV
+            echo BOX86_HAVE_TRACE=0 >> $GITHUB_ENV
+          else
+            echo BOX86_BUILD_TYPE=RelWithDebInfo >> $GITHUB_ENV
+            echo BOX86_HAVE_TRACE=1 >> $GITHUB_ENV
+          fi
+            
+      - name: "Display Build info"
+        run: |
+          echo "CMake Platform Macro: ${{ env.BOX86_PLATFORM_MARCRO }}"
+          echo "CMake C Compiler: ${{ env.BOX86_COMPILER }}"
+          echo "Build type: ${{ env.BOX86_BUILD_TYPE }}"
+          echo "Trace Enabled: ${{ env.BOX86_HAVE_TRACE }}"
+          
+      - name: "Build Box86"
+        run: |
+          mkdir build
+          cd build
+          cmake .. -DCMAKE_C_COMPILER=${{ env.BOX86_COMPILER }} ${{ env.BOX86_PLATFORM_MARCRO }} -DCMAKE_BUILD_TYPE=${{ env.BOX86_BUILD_TYPE }} -DHAVE_TRACE=${{ env.BOX86_HAVE_TRACE }} -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON
+          make -j$(nproc) VERBOSE=1
+      
+      - name: "Upload Artifact"
+        uses: actions/upload-artifact@v2
+        with:
+          name: box86-${{ matrix.platform }}-${{ matrix.type }}
+          path: build/box86
+
+          
+
@@ -3,3 +3,6 @@ build*/
 .vscode
 .gdb_history
 src/git_head.h
+backup/
+box86_update.sh
+restore_backup.sh
@@ -0,0 +1,26 @@
+language: c
+arch:
+  - arm64-graviton2
+  - amd64
+os: linux
+dist: focal
+
+addons:
+  apt:
+    packages:
+      - python3
+
+compiler:
+    - gcc
+
+      #Build steps
+before_script:
+    - if [ "$TRAVIS_CPU_ARCH" = "arm64" ]; then sudo dpkg --add-architecture armhf ; else sudo dpkg --add-architecture i386 ; fi
+    - if [ "$TRAVIS_CPU_ARCH" = "arm64" ]; then sudo apt-get -y install gcc-arm-linux-gnueabihf ; else sudo apt-get -y install gcc-multilib libc6-dev:i386 libgcc1:i386 ; fi
+    - mkdir build
+    - cd build
+    - if [ "$TRAVIS_CPU_ARCH" = "arm64" ]; then cmake .. -DRK3399=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo ; else CFLAGS="-m32" cmake .. -DLD80BITS=1 -DNOALIGN=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo ; fi
+
+script:
+    - make
+    - if [ "$TRAVIS_CPU_ARCH" != "arm64" ]; then ctest --output-on-failure ; fi
@@ -1,58 +0,0 @@
-v0.1.2
-======
-* The Dynarec now handle JIT code
-* Added support for Unity games (not perfect yet)
-* Added support for Wine (not perfect yet). You need an x86 build of Wine to use it
-* Added support for Steam (not perfect yet). Note that steam have limited functionnalities on 32bits (only mini-mode is available)
-* More wrapped libs
-* Added support for "PltResolver", so order of library is less important and many symbol are resolved at runtime
-* Added an option to build box86 as a lib (to wrapped dynamic library)
-* Better Signal handling (not perfect yet)
-* More opcodes added, more opcode fixes.
-* More Dynarec opcodes.
-* Added support for FS:, and creating custom selector (needed by Wine)
-* There is now 1 x86emu_t structure per thread (simplifying/optimising many callback handling)
-* Box86 now have a logo!
-* Added options handling (only version and help for now). Now it's much usefull exept for version printing.
-
-
-v0.1.0
-=======
-* Dynarec!!! Only for ARM (note that dynarec doesn't support JITed code for now)
-* Added real support for getcontext/set/context/makecontext/swapcontext
-* Preliminary signal handling
-* Fixes to SDL(1/2) Image and SDL1 Mixer, and to SDL1.2 RWops usage
-* Fixed numerous issues in opcodes (both interpretor and dynarec). FTL works fine now, among many others
-* Added wrapped gtk support (still ongoing, many libs involved)
-* Make loading of libs more configurable
-* If a wrapped native libs is not found, try to use emulated one
-* Add en env. var. to force use of emulated lib for certain libs
-* Add an env. var. to define wich libGL to use
-* Added Install / Uninstall target (using systemd binfmt)
-* Added more hardware target (RPis, GameShell...)
-* Wrapped more libs (including FreeType, smpeg, ncurses, sndfile...)
-
-v0.0.4
-=======
-* Improved the Init of dependant Libs. More things work now.
-* Added a lot of wrapped functions.
-* Added a few wrapped libs (like libz or some other x11 related libs)
-* For trace enabled build, Trace can be enabled only after a certain amount of opcodes (still, a debugger would be better)
-* Some fixes in a few opcodes, and implemented x87 rounding (SuperMeatBoy behaves better now)
-* FTL 1.6.9 still have corrupted music, but older 1.5.13 seems fine (different set of libs)
-
-v0.0.2
-=======
-* A full commercial games runs fine on the Pandora: Airline Tycoon Deluxe
-* Implemented all planed subpart of Box86 except JIT
-* CPU Emulation is at 75%, roughly, including x87 and SSE/SSE2. MMX is barely implemented, but barely used anyway
-* x87 emulation is simplified, no real x87 State handling (but should not be mandatory, as native libm is used)
-* No Signal handling yet
-* ELF Loader is crude and probably full of bug. Also, the Init of libs is defered after all symbols are resolved for now.
-* Wrapped lib include libc, libm, rt, pthread, libdl, dllinux, libasound, GL, GLU, SDL1/mixer/image, SDL2/mixer/image/smpeg, OpenAL/ALUT, libz, libpng16, vorbisfile, x11/xrandr/xxf86vm.
-* Most wrapped libs are still partially implemented (SDL1 & GL should be complete)
-* Implemented specific mecanism for SDL(1/2) RWops, to be able to used them both in Native and x86 world
-* WorldOfGoo works, but painfully slow on th Pandora (too much double math, and lack of JIT)
-* FTL works, but sound is broken (issue with thread? asound? or CPU core?)
-* Limbo launch but crash before main menu
-
@@ -1,25 +1,39 @@
-cmake_minimum_required(VERSION 2.6)
-
-project(box86)
-
-enable_testing()
-enable_language(C)
+cmake_minimum_required(VERSION 3.0)

 option(PANDORA "Set to ON if targeting an OpenPandora device" ${PANDORA})
 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(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(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(POWER9 "Set to ON if targeting a POWER9 processor" ${POWER9})
+option(SD845 "Set to ON if targeting a Snapragon 845 based device" ${SD845})
+option(A64 "Set to ON if targeting an Allwinner A64 based device" ${A64})
+option(ANDROID "Set to ON if targeting an Android device" ${ANDROID})
 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 RPI4ARM64 OR PHYTIUM OR SD845 OR A64)
+    set(LD80BITS OFF CACHE BOOL "")
+    set(NOALIGN OFF CACHE BOOL "")
+    set(ARM_DYNAREC ON CACHE BOOL "")
+endif()
+if(ANDROID)
+    set(NOLOADADDR ON CACHE BOOL "")
+endif()
 option(LD80BITS "Set to ON if host device have 80bits long double (i.e. i386)" ${LD80BITS})
 option(NOALIGN "Set to ON if host device doesn't need re-align (i.e. i386)" ${NOALIGN})
-option(HAVE_TRACE "Set to ON to have Trace ability (needs ZydisInfo library)" ${HAVE_TRACE})
-option(USE_FLOAT "Set to ON to use only float, no double, in all x87 Emulation" ${USE_FLOAT})
-option(NOLOADADDR "Set to ON to avoid fixing the load address of Box86" ${NO_LOADAADR})
-option(ARM_DYNAREC "Set to ON to use ARM Dynamic Recompilation (WIP, don't use yet)" ${ARM_DYNAREC})
-option(NOGETCLOCK "Set to ON to avoid using clock_gettime with CLOCK_MONOTONIC_COARSE for RDTSC opcode (use gltimeofday instead)" ${NOGETCLOCK})
+option(ARM_DYNAREC "Set to ON to use ARM Dynamic Recompilation" ${ARM_DYNAREC})
+option(POWERPCLE "Set to ON if targeting a little-endian PowerPC device" ${POWERPCLE})
+option(SYSV_VARARG "Set to ON if host device uses SysVR4 vararg calling convention" ${SYSV_VARARG})

 if(${CMAKE_VERSION} VERSION_LESS "3.12.2")
    find_package(PythonInterp 3)
@@ -37,6 +51,16 @@ else()
    set(PYTHON_EXECUTABLE ${Python3_EXECUTABLE} CACHE INTERNAL "The Python3 executable" FORCE)
 endif()

+if(RK3399 OR RPI4ARM64 OR PHYTIUM OR SD845 OR A64)
+    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)
+endif()
+
+project(box86 C ASM)
+
+enable_testing()
+
 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.")
@@ -44,69 +68,88 @@ if(NOT CMAKE_BUILD_TYPE AND NOT CMAKE_CONFIGURATION_TYPES)
  set_property(CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS "Debug" "Release" "MinSizeRel" "RelWithDebInfo")
 endif()

-
-# Pandora
 if(PANDORA)
    add_definitions(-DPANDORA)
    #add_definitions(-mcpu=cortex-a8 -mfpu=neon -mfloat-abi=softfp -ftree-vectorize -fsingle-precision-constant -ffast-math)
-    set(ARM_DYNAREC ON)
-    set(LD80BITS OFF)
-    set(NOALIGN OFF)
-endif()
-
-# Gameshell
-if(GAMESHELL)
+elseif(GAMESHELL)
+    add_definitions(-DGAMESHELL)
    add_definitions(-marm -mcpu=cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard)
    set(CMAKE_ASM_FLAGS  "-marm -mcpu=cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard")
-    set(ARM_DYNAREC ON)
-    set(LD80BITS OFF)
-    set(NOALIGN OFF)
-endif()
-
-if(PYRA)
+elseif(PYRA)
+    add_definitions(-DPYRA)
    add_definitions(-mcpu=cortex-a15 -mfpu=neon -mfloat-abi=hard -marm)
    set(CMAKE_ASM_FLAGS  "-marm -mcpu=cortex-a15 -mfpu=neon -mfloat-abi=hard")
-    set(ARM_DYNAREC ON)
-    set(LD80BITS OFF)
-    set(NOALIGN OFF)
-endif()
-
-if(RPI2)
+elseif(RPI2)
    add_definitions(-DRPI)
    add_definitions(-DRPI2)
    add_definitions(-marm -mcpu=cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard)
    set(CMAKE_ASM_FLAGS  "-marm -mcpu=cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard")
-    set(ARM_DYNAREC ON)
-    set(LD80BITS OFF)
-    set(NOALIGN OFF)
-endif()
-if(RPI3)
+elseif(RPI3)
    add_definitions(-DRPI)
    add_definitions(-DRPI3)
    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")
-    set(ARM_DYNAREC ON)
-    set(LD80BITS OFF)
-    set(NOALIGN OFF)
-endif()
-if(RPI4)
+elseif(RPI4)
    add_definitions(-DRPI)
-    add_definitions(-DRPI3)
+    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")
-    set(ARM_DYNAREC ON)
-    set(LD80BITS OFF)
-    set(NOALIGN OFF)
-endif()
-if(ODROID)
+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")
-    set(ARM_DYNAREC ON)
-    set(LD80BITS OFF)
-    set(NOALIGN OFF)
+elseif(RK3288)
+    add_definitions(-DRK3288)
+    add_definitions(-mcpu=cortex-a17 -mfpu=neon -mfloat-abi=hard -marm)
+    set(CMAKE_ASM_FLAGS  "-marm -mcpu=cortex-a17 -mfpu=neon -mfloat-abi=hard")
+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(A64)
+    add_definitions(-DA64)
+    add_definitions(-marm -march=armv8-a+crc+simd+crypto -mcpu=cortex-a53+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-marm -march=armv8-a+crc+simd+crypto -mcpu=cortex-a53+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(GOA_CLONE)
+    add_definitions(-DGOA_CLONE)
+    add_definitions(-marm -mcpu=cortex-a35 -mfpu=neon-vfpv3 -march=armv8-a+crc+simd+crypto -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-marm -mcpu=cortex-a35 -mfpu=neon-vfpv3 -march=armv8-a+crc+simd+crypto -mfloat-abi=hard")
+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(PHYTIUM)
+    add_definitions(-DPHYTIUM)
+    add_definitions(-pipe -march=armv8.1-a+simd+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8.1-a+simd+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(ARM_DYNAREC)
+    #if DYNAREC is selected alone, without any arch
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv7-a+simd -mfpu=neon")
 endif()

-set(BOX86_ELF_ADDRESS "0x02800000")
+if(POWER9)
+  add_definitions(-mcpu=power9 -mtune=power9)
+endif()
+
+if(NOGIT)
+    add_definitions(-DNOGIT)
+endif()
+
+if(BOX86LIB)
+    set(BOX86 box86)
+elseif(HAVE_TRACE)
+    set(BOX86 box86)
+else()
+    set(BOX86 box86)
+endif()
+
+
+set(BOX86_ELF_ADDRESS "0x62800000")

 if(LD80BITS)
    add_definitions(-DHAVE_LD80BITS)
@@ -120,10 +163,6 @@ if(HAVE_TRACE)
    add_definitions(-DHAVE_TRACE)
 endif()

-if(USE_FLOAT)
-    add_definitions(-DUSE_FLOAT)
-endif()
-
 if(ARM_DYNAREC)
    add_definitions(-DDYNAREC)
    add_definitions(-DARM)
@@ -131,6 +170,14 @@ if(ARM_DYNAREC)
    enable_language(ASM)
 endif()

+if(POWERPCLE)
+  add_definitions(-DPOWERPCLE)
+endif()
+
+if(SYSV_VARARG)
+  add_definitions(-DSYSV_VARARG)
+endif()
+
 if(NOGETCLOCK)
    add_definitions(-DNOGETCLOCK)
 endif()
@@ -145,11 +192,11 @@ if(BOX86LIB)
        message(FATAL_ERROR "BOX86_ROOT not set when build as lib")
    endif()
 else()
-    set(BOX86_ROOT ${CMAKE_HOME_DIRECTORY})
+    set(BOX86_ROOT ${CMAKE_SOURCE_DIR})
 endif()


-add_definitions(-std=gnu99 -funwind-tables -fvisibility=hidden)
+add_definitions(-std=gnu11 -funwind-tables -fvisibility=hidden)

 if(USE_CCACHE)
    find_program(CCACHE_FOUND ccache)
@@ -165,18 +212,17 @@ include_directories(
    "${BOX86_ROOT}/src/wrapped/generated"
 )

-add_custom_command(
-    OUTPUT "${BOX86_ROOT}/src/git_head.h"
-    COMMAND bash -c "echo \\\#define GITREV \\\"$(git rev-parse --short HEAD)\\\"\>${BOX86_ROOT}/src/git_head.h"
-    DEPEND "${BOX86_ROOT}/src/build_info.c"
-    VERBATIM
-)
+# git_head.h is a generated file
+set_source_files_properties(
+    "${BOX86_ROOT}/src/git_head.h"
+    PROPERTIES GENERATED TRUE
+    HEADER_FILE_ONLY TRUE)

 set(ELFLOADER_SRC
-    "${BOX86_ROOT}/src/git_head.h"
    "${BOX86_ROOT}/src/main.c"
    "${BOX86_ROOT}/src/box86context.c"
    "${BOX86_ROOT}/src/build_info.c"
+    "${BOX86_ROOT}/src/custommem.c"
    "${BOX86_ROOT}/src/tools/bridge.c"
    "${BOX86_ROOT}/src/tools/pathcoll.c"
    "${BOX86_ROOT}/src/tools/fileutils.c"
@@ -190,9 +236,13 @@ set(ELFLOADER_SRC
    "${BOX86_ROOT}/src/elfs/elfload_dump.c"
    "${BOX86_ROOT}/src/librarian/library.c"
    "${BOX86_ROOT}/src/librarian/librarian.c"
+    "${BOX86_ROOT}/src/librarian/dictionnary.c"
+    "${BOX86_ROOT}/src/librarian/symbols.c"
+    "${BOX86_ROOT}/src/librarian/globalsymbols.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"
    "${BOX86_ROOT}/src/emu/x86primop.c"
@@ -205,14 +255,17 @@ set(ELFLOADER_SRC
    "${BOX86_ROOT}/src/libtools/myalign.c"
    "${BOX86_ROOT}/src/libtools/myalign64.c"
    "${BOX86_ROOT}/src/libtools/myfts.c"
-    "${BOX86_ROOT}/src/libtools/obstack.c"
    "${BOX86_ROOT}/src/libtools/sdl1rwops.c"
    "${BOX86_ROOT}/src/libtools/sdl2rwops.c"
    "${BOX86_ROOT}/src/libtools/signals.c"
    "${BOX86_ROOT}/src/libtools/threads.c"
+    "${BOX86_ROOT}/src/libtools/vkalign.c"
    "${BOX86_ROOT}/src/dynarec/dynarec.c"
    "${BOX86_ROOT}/src/wrapped/generated/wrapper.c"
 )
+if(NOT ANDROID)
+    list(APPEND ELFLOADER_SRC "${BOX86_ROOT}/src/libtools/obstack.c")
+endif()

 set(WRAPPEDS
    "${BOX86_ROOT}/src/wrapped/wrappedlibc.c"
@@ -226,11 +279,15 @@ set(WRAPPEDS
    "${BOX86_ROOT}/src/wrapped/wrappedlibm.c"
    "${BOX86_ROOT}/src/wrapped/wrappedsdl1.c"
    "${BOX86_ROOT}/src/wrapped/wrappedsdl1mixer.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedsdl1net.c"
    "${BOX86_ROOT}/src/wrapped/wrappedsdl1image.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedsdl1sound.c"
    "${BOX86_ROOT}/src/wrapped/wrappedsdl1ttf.c"
    "${BOX86_ROOT}/src/wrapped/wrappedsdl2.c"
    "${BOX86_ROOT}/src/wrapped/wrappedsdl2mixer.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedsdl2net.c"
    "${BOX86_ROOT}/src/wrapped/wrappedsdl2image.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedsdl2ttf.c"
    "${BOX86_ROOT}/src/wrapped/wrappedsmpeg.c"
    "${BOX86_ROOT}/src/wrapped/wrappedsmpeg2.c"
    "${BOX86_ROOT}/src/wrapped/wrappedvorbisfile.c"
@@ -245,6 +302,7 @@ set(WRAPPEDS
    "${BOX86_ROOT}/src/wrapped/wrappedlibxfixes.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibxcursor.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibxrender.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibxft.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibxi.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibxss.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibxpm.c"
@@ -252,6 +310,16 @@ set(WRAPPEDS
    "${BOX86_ROOT}/src/wrapped/wrappedlibxdmcp.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibx11xcb.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibxcb.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibxcbxfixes.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibxcbshape.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibxcbshm.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibxcbrandr.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibxcbimage.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibxcbkeysyms.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibxcbxtest.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibxcbdri2.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibxcbdri3.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibxcbpresent.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibxtst.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibxt.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibxcomposite.c"
@@ -259,8 +327,8 @@ set(WRAPPEDS
    "${BOX86_ROOT}/src/wrapped/wrappedlibxmu.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibz.c"
    "${BOX86_ROOT}/src/wrapped/wrappedopenal.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedalure.c"
    "${BOX86_ROOT}/src/wrapped/wrappedalut.c"
-    "${BOX86_ROOT}/src/wrapped/wrappedlibjpeg.c"
    "${BOX86_ROOT}/src/wrapped/wrappedcurl.c"
    "${BOX86_ROOT}/src/wrapped/wrappedudev0.c"
    "${BOX86_ROOT}/src/wrapped/wrappedudev1.c"
@@ -269,6 +337,7 @@ set(WRAPPEDS
    "${BOX86_ROOT}/src/wrapped/wrappedlibdrm.c"
    "${BOX86_ROOT}/src/wrapped/wrappedcrypto.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibssl.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibcrypt.c"
    "${BOX86_ROOT}/src/wrapped/wrappedutil.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibuuid.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibresolv.c"
@@ -278,6 +347,7 @@ set(WRAPPEDS
    "${BOX86_ROOT}/src/wrapped/wrappedfontconfig.c"
    "${BOX86_ROOT}/src/wrapped/wrappedfreetype.c"
    "${BOX86_ROOT}/src/wrapped/wrappedbz2.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedexpat.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibsm.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibice.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibusb1.c"
@@ -285,20 +355,26 @@ set(WRAPPEDS
    "${BOX86_ROOT}/src/wrapped/wrappedlibformw.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibncurses.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibtinfo.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibpanel.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibncurses6.c"
    "${BOX86_ROOT}/src/wrapped/wrappedlibtinfo6.c"
    "${BOX86_ROOT}/src/wrapped/wrappedgtkx112.c"
    "${BOX86_ROOT}/src/wrapped/wrappedgdkx112.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedgtk3.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedgdk3.c"
    "${BOX86_ROOT}/src/wrapped/wrappedgdkpixbuf2.c"
    "${BOX86_ROOT}/src/wrapped/wrappedgio2.c"
    "${BOX86_ROOT}/src/wrapped/wrappedgmodule2.c"
    "${BOX86_ROOT}/src/wrapped/wrappedgobject2.c"
    "${BOX86_ROOT}/src/wrapped/wrappedglib2.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedgconf2.c"
    "${BOX86_ROOT}/src/wrapped/wrappedgstreamer010.c"
    "${BOX86_ROOT}/src/wrapped/wrappedgstinterfaces010.c"
    "${BOX86_ROOT}/src/wrapped/wrappeddbusmenugtk.c"
    "${BOX86_ROOT}/src/wrapped/wrappeddbusmenuglib.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedappindicator.c"
    "${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"
@@ -310,12 +386,50 @@ set(WRAPPEDS
    "${BOX86_ROOT}/src/wrapped/wrappedtcmallocminimal.c"
    "${BOX86_ROOT}/src/wrapped/wrappedmpg123.c"
    "${BOX86_ROOT}/src/wrapped/wrappedgnutls.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibcups.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedd3dadapter9.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedvulkan.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedwaylandclient.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedxml2.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedxslt.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedldapr.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlber.c"
+    "${BOX86_ROOT}/src/wrapped/wrappednsl.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlcms2.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedkrb5.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedgssapikrb5.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibtiff.c"
    "${BOX86_ROOT}/src/wrapped/wrappedldlinux.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedcrashhandler.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibv4l2.c"
+    "${BOX86_ROOT}/src/wrapped/wrappednspr4.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedsmime3.c"
+    "${BOX86_ROOT}/src/wrapped/wrappednssutil3.c"
+    "${BOX86_ROOT}/src/wrapped/wrappednss3.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedplds4.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedplc4.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedssl3.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedsecret1.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedtbbmalloc.c"
 )
+if(ANDROID)
+    list(APPEND WRAPPEDS
+        "${BOX86_ROOT}/src/wrapped/wrappedandroidshmem.c"
+    )
+else()
+    list(APPEND WRAPPEDS
+        "${BOX86_ROOT}/src/wrapped/wrappedlibjpeg.c"
+        "${BOX86_ROOT}/src/wrapped/wrappedlibjpeg62.c"
+        "${BOX86_ROOT}/src/wrapped/wrappedturbojpeg.c"
+    )
+endif()

-set(WRAPPEDS_HEAD "")
+# If BOX86_ROOT contains a ".c", the build breaks...
+string(REPLACE ".c" "_private.h" MODROOT ${BOX86_ROOT})
+set(WRAPPEDS_HEAD "${BOX86_ROOT}/src/wrapped/wrappedd3dadapter9_genvate.h")
 foreach(A ${WRAPPEDS})
-    string(REPLACE ".c" "_private.h" B ${A})
+    string(REPLACE ".c" "_private.h" C ${A})
+    string(REPLACE "${MODROOT}" "${BOX86_ROOT}" B ${C})
    set(WRAPPEDS_HEAD ${WRAPPEDS_HEAD} ${B})
    set_source_files_properties(${A} PROPERTIES OBJECT_DEPENDS ${B})
 endforeach()
@@ -324,7 +438,10 @@ set(WRAPPER "${BOX86_ROOT}/src/wrapped/generated/wrapper.c" "${BOX86_ROOT}/src/w

 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" "USE_FLOAT" "--" ${WRAPPEDS_HEAD}
+    COMMAND "${PYTHON_EXECUTABLE}" "${BOX86_ROOT}/rebuild_wrappers.py" 
+    "${BOX86_ROOT}" 
+    "PANDORA" "HAVE_LD80BITS" "NOALIGN" "HAVE_TRACE" "POWERPCLE" "ANDROID" "--" 
+    ${WRAPPEDS_HEAD}
    MAIN_DEPENDENCY "${BOX86_ROOT}/rebuild_wrappers.py"
    DEPENDS ${WRAPPEDS} ${WRAPPEDS_HEAD}
    BYPRODUCTS ${WRAPPER}
@@ -341,9 +458,6 @@ add_custom_command(
 add_custom_target(WRAPPERS DEPENDS "${BOX86_ROOT}/src/wrapped/generated/functions_list.txt")
 add_custom_target(PRINTER DEPENDS "${BOX86_ROOT}/src/dynarec/last_run.txt")

-#this is to force build_info.c to be rebuilt all the time
-set_source_files_properties("${BOX86_ROOT}/src/build_info.c" PROPERTIES OBJECT_OUTPUTS "dummy.c")
-
 if(ARM_DYNAREC)
    set(DYNAREC_SRC
        "${BOX86_ROOT}/src/dynarec/dynablock.c"
@@ -354,8 +468,8 @@ if(ARM_DYNAREC)

        "${BOX86_ROOT}/src/dynarec/arm_prolog.S"
        "${BOX86_ROOT}/src/dynarec/arm_epilog.S"
-        "${BOX86_ROOT}/src/dynarec/arm_linker.S"
-        "${BOX86_ROOT}/src/dynarec/arm_table.S"
+        "${BOX86_ROOT}/src/dynarec/arm_next.S"
+        "${BOX86_ROOT}/src/dynarec/arm_lock_helper.S"
    )

    set(DYNAREC_PASS
@@ -409,40 +523,59 @@ if(ARM_DYNAREC)
        $<TARGET_OBJECTS:arm_pass2>
        $<TARGET_OBJECTS:arm_pass3>
    )
+
+endif()
+
+# creates git_head.h
+if(ARM_DYNAREC)
+    add_custom_command(
+        OUTPUT "${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} ${WRAPPEDS}
+        VERBATIM)
+else()
+    add_custom_command(
+        OUTPUT "${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}
+        VERBATIM)
 endif()

 if(BOX86LIB)
 #build a static or dynamic lib...
 if(BOX86DYNAMIC)
-add_library(box86 SHARED ${ELFLOADER_SRC} ${WRAPPEDS})
+    add_library(box86 SHARED ${ELFLOADER_SRC} ${WRAPPEDS} "${BOX86_ROOT}/src/git_head.h")
 else()
-    add_library(box86 OBJECT ${ELFLOADER_SRC} ${WRAPPEDS})
+    add_library(${BOX86} OBJECT ${ELFLOADER_SRC} ${WRAPPEDS} "${BOX86_ROOT}/src/git_head.h")
 endif()
 if(ARM_DYNAREC)
-    target_link_libraries(box86 dynarec m dl rt pthread)
+    if(ANDROID)
+        target_link_libraries(${BOX86} dynarec m dl)
+    else()
+        target_link_libraries(${BOX86} dynarec m dl rt pthread)
+    endif()
 endif()
 if(${CMAKE_VERSION} VERSION_LESS "3.13")
-    set_target_properties(box86 PROPERTIES LINK_FLAGS -rdynamic)
+    set_target_properties(${BOX86} PROPERTIES LINK_FLAGS -rdynamic)
 else()
-    target_link_options(box86 PUBLIC -rdynamic)
+    target_link_options(${BOX86} PUBLIC -rdynamic)
 endif()
-add_dependencies(box86 WRAPPERS)
-add_dependencies(box86 PRINTER)
+add_dependencies(${BOX86} WRAPPERS)
+add_dependencies(${BOX86} PRINTER)
+
+add_custom_target(buildinfo ${BOX86}
+    DEPENDS dynarec ${BOX86} )

 else(BOX86LIB)

-if(USE_FLOAT)
-    set(BOX86 box86f)
-elseif(HAVE_TRACE)
-    set(BOX86 box86t)
-else()
-    set(BOX86 box86)
-endif()
-
-add_executable(${BOX86} ${ELFLOADER_SRC} ${WRAPPEDS})
+add_executable(${BOX86} ${ELFLOADER_SRC} ${WRAPPEDS} "${BOX86_ROOT}/src/git_head.h")
 add_dependencies(${BOX86} WRAPPERS)
 add_dependencies(${BOX86} PRINTER)
-target_link_libraries(${BOX86} m dl rt pthread)
+if(ANDROID)
+    target_link_libraries(${BOX86} m dl)
+else()
+    target_link_libraries(${BOX86} m dl rt pthread)
+endif()
 if(ARM_DYNAREC)
    target_link_libraries(${BOX86} dynarec)
 endif()
@@ -462,11 +595,14 @@ endif()

 string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "i686"  _x86)
 string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "x86_64"  _x86_64)
+string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "aarch64"  _aarch64)
+string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "armhf"  _armhf)

 if(NOT _x86 AND NOT _x86_64)
  install(TARGETS ${BOX86}
    RUNTIME DESTINATION bin)
-  install(FILES ${CMAKE_SOURCE_DIR}/system/box86.conf DESTINATION /etc/binfmt.d/)
+  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/)
@@ -482,76 +618,131 @@ if(NOT TARGET uninstall)
        COMMAND ${CMAKE_COMMAND} -P ${CMAKE_CURRENT_BINARY_DIR}/cmake_uninstall.cmake)
 endif()

-add_test(test01 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+set(CPACK_GENERATOR "DEB")
+set(CPACK_DEBIAN_PACKAGE_MAINTAINER "ptitSeb")
+set(CPACK_PACKAGE_CONTACT "[email protected]")
+set(CPACK_PACKAGE_DESCRIPTION "Box86 - Linux Userspace x86 Emulator with a twist")
+set(CPACK_DEBIAN_PACKAGE_DEPENDS "libc6")
+set(CPACK_PACKAGE_HOMEPAGE_URL, "https://box86.org")
+file(STRINGS "${BOX86_ROOT}/src/box86version.h" TEMP_STRING REGEX "BOX86_MAJOR\\s*")
+string(REGEX REPLACE "BOX86_MAJOR" "" TEMP_STRING ${TEMP_STRING})
+string(REGEX MATCH "[0-9]" BOX86_MAJOR ${TEMP_STRING})
+file(STRINGS "${BOX86_ROOT}/src/box86version.h" TEMP_STRING REGEX "BOX86_MINOR\\s*")
+string(REGEX REPLACE "BOX86_MINOR" "" TEMP_STRING ${TEMP_STRING})
+string(REGEX MATCH "[0-9]" BOX86_MINOR ${TEMP_STRING})
+file(STRINGS "${BOX86_ROOT}/src/box86version.h" TEMP_STRING REGEX "BOX86_REVISION\\s*")
+string(REGEX REPLACE "BOX86_REVISION" "" TEMP_STRING ${TEMP_STRING})
+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 ${BOX86_REVISION})
+set(CPACK_DEBIAN_PACKAGE_CONTROL_EXTRA "${CMAKE_CURRENT_SOURCE_DIR}/postinst")
+if(_aarch64 OR _armhf)
+    set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "armhf")
+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")
+INCLUDE(CPack)
+
+add_test(NAME bootSyscall COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test01 -D TEST_OUTPUT=tmpfile.txt 
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref01.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-add_test(test02 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+add_test(NAME bootSyscallC COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test02 -D TEST_OUTPUT=tmpfile.txt 
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref02.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-add_test(test03 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+add_test(NAME printf COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test03 -D TEST_OUTPUT=tmpfile.txt 
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref03.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-add_test(test04 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+add_test(NAME args COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test04 -D TEST_ARGS2=yeah -D TEST_OUTPUT=tmpfile.txt 
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref04.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-add_test(test05 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+add_test(NAME maths1 COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test05 -D TEST_ARGS2=7 -D TEST_OUTPUT=tmpfile.txt 
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref05.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-add_test(test06 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+add_test(NAME threadsStart COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test06 -D TEST_OUTPUT=tmpfile.txt 
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref06.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-add_test(test07 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+add_test(NAME trig COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test07 -D TEST_OUTPUT=tmpfile.txt 
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref07.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-add_test(test08 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+add_test(NAME pi COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test08 -D TEST_OUTPUT=tmpfile.txt 
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref08.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-add_test(test09 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+add_test(NAME fork COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test09 -D TEST_OUTPUT=tmpfile.txt 
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref09.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-add_test(test10 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+add_test(NAME cppThreads COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test10 -D TEST_OUTPUT=tmpfile.txt 
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref10.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-add_test(test11 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+add_test(NAME tlsData COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test11 -D TEST_OUTPUT=tmpfile.txt 
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref11.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-add_test(test12 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+add_test(NAME fpu COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test12 -D TEST_OUTPUT=tmpfile.txt 
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref12.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-add_test(test13 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+add_test(NAME contexts COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test13 -D TEST_OUTPUT=tmpfile.txt 
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref13.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

-add_test(test14 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+add_test(NAME conditionalThreads COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test14 -D TEST_OUTPUT=tmpfile.txt 
    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref14.txt
    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )

+add_test(NAME linkingIndirectNoversion COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test15 -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref15.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(NAME linkingIndirectVersion COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test16 -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref16.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(NAME ucomiss COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test17 -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref17.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+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=tmpfile.txt 
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref18.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+    add_test(NAME x87 COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test19 -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref19.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)
@@ -1,64 +0,0 @@
-Compiling
----
-#### for Pandora
-
- `mkdir build; cd build; cmake .. -DPANDORA=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make`
-
-#### for Pyra
-
- `mkdir build; cd build; cmake .. -DPYRA=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make`
-
-#### for Gameshell
-
-`mkdir build; cd build; cmake .. -DGAMESHELL=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make`
-
-#### for RaspberryPI
-
-  _a build for model 2, 3 and 4 can be done. Model 1 and 0 cannot (at least not with Dynarec, as they lack NEON support)_
- 
-`mkdir build; cd build; cmake .. -DRPI4=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make`
- 
-  _For Pi4. Change to RPI2 or RPI3 for other models._
-
-#### for ODROID
-
-`mkdir build; cd build; cmake .. -DODROID=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make`
-
-#### for Other ARM Linux
-
- `mkdir build; cd build; cmake .. -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo; make`
-
-#### for x86 Linux
-
- `mkdir build; cd build; cmake .. -DLD80BITS=1 -DNOALIGN=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make`
-
-#### use ccmake
-
-Alternatively, you can use the curses-bases ccmake (or any other gui frontend for cmake) to select wich platform to use interactively.
-
-#### Customize your build
-
-*use ccache if you have it* 
-
-Add `-DUSE_CCACHE=1` if you have ccache (it's better if you plan to touch the sources)
-
-*have some debug info* 
-
-The `-DCMAKE_BUILD_TYPE=RelWithDebInfo` argument make a build that is both optimized for speed, and have debug information embedded. That way, if you have a crash or try to analyse performances, you'll have some symbols.
-
-*to have a Trace Enabled build* 
-
-To have a trace enabled build (warning, it will be slower), add `-DHAVE_TRACE=1` but you will need, at runtime, to have the [Zydis library](https://github.com/zyantific/zydis) library in your `LD_LIBRARY_PATH` or in the system lib folders.
-
-*to have ARM Dynarec*
-
-The Dynarec is only avaiable on ARM Cpu. Notes also that VFPv3 and NEON are required for the Dynarec. Activate it by using `-DARM_DYNAREC=1`. Also, be sure to use `-marm` in compilation flags (because many compileur use Thumb as default, and the dynarec will not work in this mode).
-
----
-
-Testing
----
-A few tests are included.
-They can be launched with `ctest`
-They are very basic and don't test much for now.
-
@@ -1,4 +1,4 @@
-Copyright (c) 2018-2019 Sebastien Chevalier ("ptitSeb")
+Copyright (c) 2018-2021 Sebastien Chevalier ("ptitSeb")

 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal
@@ -1,90 +0,0 @@
-# box86
-
-![Official logo](Box86Logo.png "Official Logo")
-Linux Userspace x86 Emulator with a twist
-
-Box86 lets you run x86 Linux programs (such as games) on non-x86 Linux, like ARM (host system needs to be 32bit little-endian).
-
-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 support 64-bit will not compile Box86, and you'll get errors (typically on aarch64, you get "-marm" not recognized).
-
-Because Box86 uses the native versions of some "system" libraries, like libc, libm, SDL, and OpenGL, it's easy to integrate and use, and performance can be surprisingly high in some cases.
-
-Most x86 Games need OpenGL, so on ARM platforms a solution like [gl4es](https://github.com/ptitSeb/gl4es) is usually necessary.
-
-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.
-
-Many games already work, for example: WorldOfGoo, Airline Tycoon Deluxe, and FTL. Many of the GameMaker Linux games also run fine (there a long list, among them are UNDERTALE, A Risk of Rain, and Cook Server Delicious)
-
-If you are serious about developing Box86, you should install ccache and activate its support in the cmake project (use ccmake for example)
-To enable TRACE (i.e. dumping to stdout all individual x86 instructions executed, with dump of registers), you'll also need [Zydis library](https://github.com/zyantific/zydis) available on your system.
-
-Some x86 internal opcodes use parts of "Realmode X86 Emulator Library", see [x86primop.c](src/x86primop.c) for copyright details
-
-Here are 6 videos, the first 2 of "Airline Tycoon Deluxe" and "Heretic 2" running on a gigahertz OpenPandora (the second one use the dynarec), and the next 2 of "Bit.Trip.Runner" and "Neverwinter Night" running on an ODroid XU4 (without dynarec), and the last 2 on a Pi4: Shovel Knight (video from @ITotalJustice) and Freedom Planet (video from @djazz), also without 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)
-
-You can find many more Box86 video on [PI Lab Channel](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g) [![PI Lab Channel](https://yt3.ggpht.com/a/AATXAJyMeWrgCjs78gr6To6yX4KtDPUCS7hsbX1rRA=s100-c-k-c0xffffffff-no-rj-mo)](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g).
-
-Compatibility list is there: https://github.com/ptitSeb/box86-compatibility-list/issues
-
-<img src="Box86Icon.png" width="96" height="96">
-
-Logo and Icon made by @grayduck, thanks!
-
----
-
-Compiling
----
-Compilation instructions can be found [here](COMPILE.md)
-
----
-
-Usage
----
-
-There are a few environment variables avaiable to control the behaviour of Box86.
-
-See [here](USAGE.md) for all variables and what they do.
-
-Note that now the Dynarec of Box86 uses a mecanism with Memory Protection and a SegFault signal handler to handle JIT code. That means if you want to use gdb to debug running a program that use JIT'd code (like mono/Unity3D), you will have many "normal" segfaults triggering. I suggest you use something like `handle SIGSEGV nostop` in gdb to not stop at each segfault, and maybe put a breakpoint inside `my_memprotectionhandler` in `signals.c` if you want to trap SegFaults.
-
----
-
-Version history
----
-
-The change log is available [here](CHANGELOG.md)
-
----
-
-A note about 64-bit platforms
----
-
-Because Box86 works by directly translating function calls from x86 to host system, the host system (the one Box86 is running on) needs to have 32-bit libraries. Box86 doesn't include any 32-bit <-> 64-bit translation. So basically, to run Box86 on, for example, an ARM64 platform, you will need to build Box86 for ARM 32-bit, and also need to have a chroot with 32-bit libraries.
-
-Also note that, even if, on day, there is a Box86_64, this one will only be able to run x86_64 binaries on 64-bit platforms. You will still need Box86 (and see 32-bit chroot) to run x86 binaries (in fact, the same is the case on actual x86_64 Linux).
-
----
-
-A note about Unity game emulation
----
-
-Running Unity games is a hit or miss for now. Unity uses Mono (which uses signals that are not well emulated enough), and a runtime embedded in the main binary. A solution would be to use a native version of the libmono library used by Unity (it can be found here: https://github.com/Unity-Technologies/mono and needs to be built from source). But the wrapping of this lib is tricky, and not done for now, so the only solution is to emulate everything. The tricky part is to emulate the "JIT" code emitted by Mono, however with he new "protected memory" mechanism implemented it is running with correct performance now.
-You should also note that some Unity3D games require OpenGL 3+ which can be tricky to provide on ARM SBC (single-board computers) for now.
-
-TL;DR: Not all Mono games work and can require a high OpenGL profile, but the speed, for the ones running, should be correct now.
-
----
-
-A note about Wine
----
-
-Wine is now partly supported. Wine integreted program all runs, annd some windows program and games now run fine. Don't forget most Windows games use Direct3D, that may require a complete OpenGL driver and as high profile as possible (and gl4es with ES2 backend have issue with Wine for now). Also, vulkan is not wrapped on box86, so vk3d is not usable yet, even if supported by the hardware.
-
----
-
-Final words
----
-
-(If you use Box86 in your project, please don't forget to mention Box86)
@@ -1,132 +0,0 @@
-Usage
----
-
-There are many environment variable to control Box86 behaviour. 
-
-##### BOX86_LOG
-Controls the Verbose level of the log
- * 0
- * NONE : No message (except some fatal error)
- * 1
- * INFO : Current default. Show some log
- * 2
- * DEBUG : Verbose a lot of stuffs (like relocations or function called)
- * 3
- * DUMP : All DEBUG plus DUMP of all ELF Info
-
-#### BOX86_LD_LIBRARY_PATH
-Controls the folder scanned to look for x86 libs. Default is current folder and `lib` in current folder.
-Also `/usr/lib/i386-linux-gnu` and `/lib/i386-linux-gnu` are added if they exists
-
-#### BOX86_PATH
-Controls the folder scanned to look for x86 executable. Default is current folder and `bin` in current folder.
-
-#### BOX86_DLSYM_ERROR
-* 0 : default. Don't log `dlsym` error
-* 1 : Log dlsym error
-
-#### BOX86_TRACE_FILE
-Send all log and trace to a file instead of `stdout`
-Also, if name contains `%pid` then this is replaced by actual PID of box86 instance
-
-#### BOX86_TRACE
-Only on build with trace enabled. Trace allow the logging of all instruction execute, along with register dump
-* 0 : No trace
-* 1 : Trace enabled. The trace start after the init of all depending libs 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
-Use BOX86_TRACE_INIT instead of BOX_TRACE to start trace before init of Libs and main program
-* 0 : No trace
-* 1 : Trace enabled. The trace start with the init of all depending libs is done
-
-#### BOX86_TRACE_START
-Only on build with trace enabled.
-* NNNNNNN : Start trace only after NNNNNNNN opcode execute (number is an `uint64_t`)
-
-#### BOX86_TRACE_XMM
-Only on build with trace enabled.
-* 0 : Default, the XMM (i.e. SSE/SSE2) register will not be logged with the general and x87 registers
-* 1 : Dump the XMM registers
-
-#### BOX86_LOAD_ADDR
-Try to load at 0xXXXXXX main binaray (if binary is a PIE)
-* 0xXXXXXXXX the load address (only active on PIE programs)
-
-#### BOX86_NOSIGSEGV
-To disable handling of SigSEGV (to ease debugging mainly)
-* 0 : default, let x86 program set sighandler for SEGV
-* 1 : disable handling of SigSEGV
-
-#### BOX86_NOSIGILL
-To disable handling of SigILL (to ease debugging mainly)
-* 0 : default, let x86 program set sighandler for Illegal Instruction
-* 1 : disable handling of SigILL
-
-#### BOX86_X11COLOR16
-PANDORA only: to try convert X11 color from 32 bits to 16 bits (to avoid light green on light cyan windows
-* 0 : default, don't touch X11 colors
-* 1 : Change colors arguments in XSetForeground, XSetBackground and XCreateGC
-
-#### BOX86_X11THREADS
-To call XInitThreads when loading X11 (for old Loki games with Loki_Compat lib)
-* 0 : default, don't force call XInitThreads
-* 1 : Call XInitThreads as soon as libX11 is loaded
-
-#### BOX86_X11GLX
-To Force libX11 GLX extension to be present
-* 0 : Disable the force.
-* 1 : default, GLX will always be present when using XQueryExtension
-
-#### BOX86_DYNAREC_DUMP
- * 0 : Disable Dynarec blocks dump (default)
- * 1 : Enable Dynarec blocks dump
- * 2 : Enable Dynarec blocks dump with some colors
-
-#### BOX86_DYNAREC_LOG
-Set level of DynaRec log
- * 0 :
- * NONE : No Log for DynaRec
- * 1 :
- * INFO : Minimum Dynarec Log (only unimplemented OpCode)
- * 2 :
- * DEBUG : Debug Log for Dynarec (with detail on block created / executed)
- * 3 :
- * VERBOSE : All above plus more
-
-#### BOX86_DYNAREC
- * 0 : Disable Dynarec
- * 1 : Enable Dynarec (default)
-
-#### BOX86_DYNAREC_LINKER
- * 0 : Disable Dynarec Linker (use that on debug, with dynarec log >= 2, to have detail on wich block get executed)
- * 1 : Enable Dynarec Linker (default)
-
-#### BOX86_DYNAREC_TRACE
- * 0 : Disable trace for generated code (default)
- * 1 : Enable trace for generated code (like regular Trace, this will slow down a lot and generate huge logs)
-
-#### BOX86_LIBGL
- * libXXXX set the name for libGL (default to libGL.so.1)
- * /PATH/TO/libGLXXX set the name and path for libGL
- You can also use SDL_VIDEO_GL_DRIVER
-
-#### 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
- 
-#### BOX86_EMULATED_LIBS
- * XXXX[:YYYYY] force lib XXXX (and YYYY...) to be emulated (and not wrapped)
-Some games use old version of some lib, 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_ALLOWMISSINGLIBS
-Allow box86 to continue even if a lib is missing
- * 0 : default, stop if a lib cannot be loaded
- * 1 : continue even if a needed lib cannot be loaded. Unadvised, this will, in most cases, crash later on.
-
-#### BOX86_FIX_64BIT_INODES
- * 0 : Don't fix 64bit inodes (default)
- * 1 : Fix 64bit inodes. Helps when running on filesystems with 64bit inodes, the program uses API functions which don't support it and the program doesn't use inodes information.
-
@@ -0,0 +1 @@
+theme: jekyll-theme-hacker
@@ -0,0 +1,153 @@
+v0.2.4
+======
+* Some work on Dynarec to limit the number of mutex use, and also allow smaller block to be built (for JIT'd programs)
+* Introduce BOX86_DYNAREC_STRONGMEM, like with box64 (not found any program that needs that yet)
+* More vulkan wrapped function (so dkvk works, at least with RX550)
+* Reworked Elfloader part, with better handling of versionned symbols
+* Improve speed for x87 code (by avoiding many float<->double conversion)
+* Improve speed of the custom allocator used by dynarec and Hash (speedup loading)
+* Added a detection of box64 in uname call, to expose an x86_64 system instead of i386 one
+* Added a fix for linking box86 on system using glibc 2.34+
+* A few more wrapped libs
+* More ARM hardware option in CMake (Phytium, SD845)
+* Changed a bit struture of Dynarec, switching some operations from Pass1 to Pass0
+* Small fixes to elfloader
+* Some fixes and improvments on wrapped libs (mainly gtk2, gtk3 and d3dadpter)
+* Better VSyscall implementation
+* Fixed the "Alternate" handling (address shadowing on x86 side)
+* A few opcodes border case fixes
+* Improved wrapper helpers (from rajdakin)
+* Added a mecanism to cancel a Dynarec block construction if it triggers a segfault (instead of a crash)
+* Improvment to the handling of longjmp inside signal handlers
+
+v0.2.2
+======
+* Reworked some elfloader relocation mecanism
+* Removed old "smart linker" and replace with a simple "jump table"
+* Also removed CALL/RET optomisation from Dynarec, simple Jump Table is faster
+* Reworked all memory protection tracking
+* A few more wrapped libs and wrapped functions
+* A few more opcodes
+* A few more dynarec opcodes
+* Some optimisation in Dynarec around ZF and CF flags handlings, and LL/GE/LE/G conditionnal opcodes
+* Some changes to the "cancel threads" handling
+* Some changes to the "thread once" handling
+
+v0.2.0
+======
+* Improvements on x86 Flags handling.
+* A few more opcode has been added.
+* A few potential BusError are now fixed.
+* Added the BOX86_NOVULKAN en. var. . This is used to disable the wrapping of vulkan libraries.
+* Improvments on libc O_XXXX flags handling.
+* Box86 now uses a custom allocator for Dynarec and Hash tables.
+* Improved the wrapping of pulse audio.
+* Optimisation to a few Dynarec opcode (like SHRD/SHLD).
+* Improved the tracking of memory protection.
+* A few more wrapping of libraries (like libgssapi_krb5).
+* More function wrapping on libturbojpeg
+* Preliminary support for POWER9 (ppcle) build
+* Many contributions to remove typos and rephrase the README, COMPILE and USAGE documents
+
+v0.1.8
+======
+* Fixes on some float to int x86 convertions opcodes.
+* Reworked all callback mecanisms.
+* Added libturbojpeg wrapping (and a hack for zoom to force using native one).
+* Added the BOX86_SAFEMMAP env. var.
+* Reworked Dynarec's memory manager.
+* Added a few opcodes.
+* Improve elfloader to not force PltResolver all the time (fixing SuperHexagon and maybe other)
+* Reworked Exet and Cancel Thread mecanism.
+* Added wrapped libldap_r and liblber library (used by wine)
+* Reworked Dynarec block handling, and remove the "AddMark" mecanism that wasn't efficient enough
+* Added TokiTori 2+ detection and runtime patch on the Raspberry Pi platform
+* Fixes on elfloader with TLS object.
+
+v0.1.6
+======
+* Changes in Dynarec to make flags optimizations before CALL and RET opcode less aggressive.
+* Added a Vulkan wrapper.
+* Improved wrapping of SDL_mixer/ SDL2_mixer libraries.
+* Improved wrapping of some GTK structures.
+* Added a quick and dirty wrapping of GTK3 (based on current GTK2).
+* Improved the signal handling, should be more stable now (the signal handler from syscall still need some works).
+* Added the RK3399 profile, and some hints to build 32bits box86 on 64bits OS.
+* Fixed some wrapped printf formating not handled correctly.
+* Fixed some buserror with the new LOCK mecanism handling.
+* A few more Dynarec opcodes added.
+* If winedbg is tried to be launched, exit without launching it (it doesn't work anyways).
+
+v0.1.4
+======
+* Change in Dynarec's memory handling, to simplify it (and hopefully optimized it).
+* Even more opcodes added.
+* And some more Dynarec opcodes.
+* Fixed some issue were PltResolver was injected but should not be.
+* Fixed many Dynarec and non dynarec opcodes.
+* Improved Signal handling.
+* Added a few more wrapped libraries (like curl).
+* Gallium9 is now wrapped (thx @icecream95).
+* Fixed and simplied many wrapped functions (especialy the one where a callback is involved).
+* Fixed Dynarec "freeing" a Dynablock sometimes causing a corruption of the heap (generally happens in case of JIT code).
+* Optimized the way LOCK prefix work when using Dynarec on ARM.
+* improvements to the ARM opcode Printer (for dumping Dynarec blocks) improvement (thx @rajdakin).
+
+v0.1.2
+======
+* The Dynarec now handle JIT code
+* Added support for Unity games (not perfect yet).
+* Added support for Wine (not perfect yet). You need an x86 build of Wine to use it.
+* Added support for Steam (not perfect yet). Note that Steam have limited functionalities on 32bits (only mini-mode is available).
+* More wrapped libs
+* Added support for the "PltResolver". This makes the order of libraries less important and many symbols are now resolved at runtime.
+* Added an option to build Box86 as a library (to wrap dynamic library).
+* Better Signal handling (not perfect yet).
+* More opcodes added, more opcode fixes.
+* More Dynarec opcodes.
+* Added support for FS:, and creating custom selector (needed by Wine).
+* There is now 1 x86emu_t structure per thread (simplifying/optimising many callback handling).
+* Box86 now has a logo!
+* Added options handling (only version and help for now). Now it's much usefull exept for version printing.
+
+
+v0.1.0
+=======
+* Dynarec!!! Only for ARM (note that Dynarec doesn't support JITed code for now).
+* Added real support for getcontext/set/context/makecontext/swapcontext.
+* Preliminary signal handling.
+* Fixes to SDL(1/2) Image and SDL1 Mixer, and to SDL1.2 RWops usage.
+* Fixed numerous issues in opcodes (both interpretor and dynarec). FTL works fine now, among many others.
+* Added wrapped GTK support (still ongoing, many libraries involved).
+* Make loading of libraries more configurable.
+* If a wrapped native library is not found, try to use an emulated one.
+* Added an env. var. to force the use of emulated libraries for certain libraries.
+* Added an env. var. to precise which libGL to use.
+* Added Install / Uninstall target (using systemd binfmt).
+* Added more hardware targets (RPis, GameShell...).
+* Wrapped more libraries (including FreeType, smpeg, ncurses, sndfile...).
+
+v0.0.4
+=======
+* Improved the initialisation of dependent libraries. More things work now.
+* Added a lot of wrapped functions.
+* Added a few wrapped libraries (like libz or some other x11 related library).
+* For trace enabled build, Trace can be enabled only after a certain amount of opcodes (still, a debugger would be better).
+* Some fixes in a few opcodes, and implemented x87 rounding (SuperMeatBoy behaves better now).
+* FTL 1.6.9 still has corrupted music, but older 1.5.13 seems fine (different set of libraries).
+
+v0.0.2
+=======
+* A full commercial games runs fine on the Pandora platform: Airline Tycoon Deluxe.
+* Implemented all planned subparts of Box86 except JIT support.
+* CPU Emulation is at roughly 75%. This includes x87 and SSE/SSE2. MMX is barely implemented (but barely used anyway).
+* x87 emulation is simplified, no real x87 State handling (but should not be mandatory, as the native libm library is used)
+* No Signal handling yet
+* ELF Loader is crude and probably full of bugs. Also, the initialisation of libraries are defered after all symbols are resolved for now.
+* Wrapped libraries include libc, libm, rt, pthread, libdl, dllinux, libasound, GL, GLU, SDL1/mixer/image, SDL2/mixer/image/smpeg, OpenAL/ALUT, libz, libpng16, vorbisfile, x11/xrandr/xxf86vm.
+* Most wrapped libraries are still partially implemented ( the SDL1 & GL  libraries should be complete).
+* Implemented specific mecanism for SDL(1/2) RWops, to be able to used them both in Native and x86 world.
+* WorldOfGoo works, but is painfully slow on the Pandora platform (there is too much double math, and lack of JIT).
+* FTL works, but sound is broken (issue with thread? asound? or CPU core?)
+* Limbo launches but crashes before main menu.
+
@@ -0,0 +1,120 @@
+Compiling
+----
+#### for Pandora
+
+ `mkdir build; cd build; cmake .. -DPANDORA=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make`
+
+#### for Pyra
+
+ `mkdir build; cd build; cmake .. -DPYRA=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make`
+
+#### for Gameshell
+
+`mkdir build; cd build; cmake .. -DGAMESHELL=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make`
+
+#### for Raspberry Pi
+
+  _a build for model 2, 3 and 4 can be done. Model 1 and 0 cannot (at least not with Dynarec, as they lack NEON support)_
+ 
+```
+git clone https://github.com/ptitSeb/box86
+cd box86
+mkdir build; cd build; cmake .. -DRPI4=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
+make
+sudo make install
+sudo systemctl restart systemd-binfmt
+```
+  Note that you can use `make -j2` or more to speed up 1st build, but beware of the memory requirement if you go to high.
+ 
+#### for Raspberry Pi on 64bits OS
+
+  _For Pi4. Change to RPI2 or RPI3 for other models.  Change `-DRPI4=1` to `-DRPI4ARM64=1` for compiling on arm64. (armhf multiarch or chroot required alongside armhf gcc. Install it with 'sudo apt install gcc-arm-linux-gnueabihf'.)_
+
+```
+git clone https://github.com/ptitSeb/box86
+cd box86
+mkdir build; cd build; cmake .. -DRPI4ARM64=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
+make -j2
+sudo make install
+sudo systemctl restart systemd-binfmt
+```
+
+#### for ODROID
+
+`mkdir build; cd build; cmake .. -DODROID=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+
+#### for RK3399
+
+`mkdir build; cd build; cmake .. -DRK3399=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+
+As most RK3399 device run 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 Tinker Board (1/1S) or RK3288
+
+`mkdir build; cd build; cmake .. -DRK3288=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+
+#### for Allwinner A64
+
+`mkdir build; cd build; cmake .. -DA64=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+
+As most Allwinner A64 device run 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 Snapdragon 845
+
+`mkdir build; cd build; cmake .. -DSD845=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j4`
+
+As most Snapdragon 845 device run AARCH64 OS, in most cases 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
+
+`mkdir build; cd build; cmake .. -DPHYTIUM=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j4`
+
+As most Phytium (D2000 or FT2000/4) device run AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On debian/ubuntu, install it with `sudo apt install gcc-arm-linux-gnueabihf`.
+
+#### for Other ARM Linux platforms
+
+ `mkdir build; cd build; cmake .. -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+
+#### for x86 Linux
+
+ `mkdir build; cd build; cmake .. -DCMAKE_C_FLAGS="-m32" -DLD80BITS=1 -DNOALIGN=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+
+#### use ccmake
+
+Alternatively, you can use the curses-bases ccmake (or any other gui frontend for cmake) to select wich platform to use interactively.
+
+#### Customize your build
+
+*use ccache if you have it* 
+
+Add `-DUSE_CCACHE=1` if you have ccache (it's better if you plan to touch the sources)
+
+*have some debug info* 
+
+The `-DCMAKE_BUILD_TYPE=RelWithDebInfo` argument makes a build that is both optimized for speed, and has debug information embedded. That way, if you have a crash or try to analyse performance, you'll have some symbols.
+
+*to have a Trace Enabled build* 
+
+To have a trace enabled build ( ***it will be slower***), add `-DHAVE_TRACE=1` but you will need, at runtime, to have the [Zydis library](https://github.com/zyantific/zydis) library in your `LD_LIBRARY_PATH` or in the system library folders.
+
+*to have ARM Dynarec*
+
+The Dynarec is only available on the ARM architecture(Right now, anyways.). Notes also that VFPv3 and NEON are required for the Dynarec. Activate it by using `-DARM_DYNAREC=1`. Also, be sure to use `-marm` in compilation flags (because many compileur use Thumb as default, and the dynarec will not work in this mode). Note that compiling with `ARM_DYNAREC` but without hardware profile is not advise. If you get error building that "target CPU does not support ARM mode", then try to pick hardware profile (like ODROID for armv7 or PI4 for armv8). Also, if you are using a 64bits with armhf multiarch, it's much easier to pick one of the hardware profile like `RPI4ARM64`, `RK3399`, `PHYTIUM` or `SD845`.
+
+*not building from a git clone*
+
+If you are not building from a git clone (for example, downloading a release source zipped from github), you need to activate `-DNOGIT=1` from cmake to be able to build (normal process include git sha1 of HEAD in the version that box86 print).
+
+----
+
+Testing
+----
+A few tests are included.
+They can be launched with `ctest`
+They are very basic and don't test much for now.
+
+----
+
+Note about devices with Tegra X1 and newer.
+
+Nvidia don't provide armhf libraries for their GPU drivers at this moment so there is no special variable to compile it for them as it would be missleading for many people. If you still want to use it wihout GPU acceleration, building it with RPI4 configuration should work just fine. As instalation of Mesa can break Nvidia driver, safest option is usage of chroot.
@@ -0,0 +1,126 @@
+# box86
+
+![Official logo](img/Box86Logo.png "Official Logo")
+Linux Userspace x86 Emulator with a twist
+
+[![Build Status](https://app.travis-ci.com/ptitSeb/box86.svg?branch=master)](https://app.travis-ci.com/ptitSeb/box86)
+
+----
+
+Compiling/Installation
+----
+> Compilation instructions can be found [here](COMPILE.md)  
+> Instructions for installing Wine for Box86 can be found [here](X86WINE.md)  
+
+Box86 lets you run x86 Linux programs (such as games) on non-x86 Linux systems, like ARM (host system needs to be 32bit little-endian).
+
+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 environnement).
+
+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 thoses bench analysis for an example [here](https://box86.org/index.php/2021/06/game-performances/).
+
+Most x86 Games need OpenGL, so on ARM platforms a solution like [gl4es](https://github.com/ptitSeb/gl4es) is usually necessary. (Most 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/).
+
+Many games already work without much tweaking, for example: WorldOfGoo, Airline Tycoon Deluxe, and FTL. Many of the GameMaker Linux games also run fine. (there's a long list, among them are UNDERTALE, A Risk of Rain, and Cook Serve Delicious)
+
+If you are serious about developing Box86, you should install ccache and build Box86 with it. (Use ccmake for example.)
+To enable TRACE (i.e. dumping to stdout all individual x86 instructions executed, with dump of registers), you'll also need [Zydis library](https://github.com/zyantific/zydis) available on your system.
+
+Some x86 internal opcodes use parts of "Realmode X86 Emulator Library", see [x86primop.c](../src/emu/x86primop.c) for copyright details
+
+Here's  6 videos, the first 2 videos are videos of "Airline Tycoon Deluxe" and "Heretic 2" running on a GigaHertz OpenPandora (the second one is using  the dynarec), and the next 2 videos are videos of of "Bit.Trip.Runner" and "Neverwinter Night" running on an ODroid XU4 (without dynarec), and the last 2  videos are on on a Pi4: Shovel Knight (video from @ITotalJustice) and Freedom Planet (video from @djazz), also without 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)
+
+You can find many more Box86 video on Youtube in the [MicroLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ), [Pi Labs](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g) or [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA) Channels
+
+Compatibility list is there: https://github.com/ptitSeb/box86-compatibility-list/issues
+
+<img src="img/Box86Icon.png" width="96" height="96">
+
+Logo and Icon made by @grayduck, thanks!
+
+Note that this project is not to be mistaken with [86box](https://github.com/86Box/86Box), a nice "Full system" emulator specialized in early (to fairly recent) PC hardware.
+
+----
+
+Usage
+----
+
+There are a few environment variables to control the behaviour of Box86.
+
+See [here](USAGE.md) for all environment variables and what they do.
+
+Note: Box86's Dynarec uses a mechanism with Memory Protection and a SegFault signal handler to handle JIT code. In simpler terms, if you want to use GDB to debug a running program that use JIT'd code (like mono/Unity3D), you will still have many "normal" segfaults triggering. It is suggested to use something like `handle SIGSEGV nostop` in GDB to not stop at each segfault, and maybe put a breakpoint inside `my_memprotectionhandler` in `signals.c` if you want to trap SegFaults.
+
+----
+
+Version history
+----
+
+The change log is available [here](CHANGELOG.md)
+
+----
+
+Notes about 64-bit platforms
+----
+
+Because Box86 works by directly translating function calls from x86 to host system, the host system (the one Box86 is running on) needs to have 32-bit libraries. Box86 doesn't include any 32-bit <-> 64-bit translation. So basically, to run Box86 on, for example, an ARM64 platform, you will need to build Box86 for ARM 32-bit, and also need to have a chroot with 32-bit libraries.
+
+If you look at a 64bits version of box86, look at [Box64](https://github.com/ptitSeb/box64): this one is able to run x86_64 binaries on 64-bit platforms. But note that you still need Box86 (and a 32-bit chroot) to run x86 binaries (as it also happens on actual x86_64 Linux that need x86 libs and binary on multiarch to run).
+
+----
+
+Notes about Unity game emulation
+----
+
+Running Unity games is a hit or miss for now. Unity uses Mono (which uses signals that are not well emulated enough), and a runtime embedded in the main binary. A solution would be to use a native version of the libmono library used by Unity (it can be found here: https://github.com/Unity-Technologies/mono and needs to be built from source). But the wrapping of this lib is tricky, and not done for now, so the only solution is to emulate everything. The tricky part is to emulate the "JIT" code emitted by Mono, however with the new "protected memory" mechanism implemented it should be running with correct performance now.
+You should also note that some Unity3D games require OpenGL 3+ which can be tricky to provide on ARM SBC (single-board computers) for now.
+
+TL;DR: Not all Unity games work and can require a high OpenGL profile, but the speed, for the ones running, should be correct now.
+
+----
+
+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 intalled. To install it on Debian, run `sudo apt install libnm0`.
+- 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. 
+- 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 aformentionned feature, it is safer to add it to your system if you intend to play those game. To install it on Debian, run `sudo apt install libtcmalloc-minimal4`.
+
+Steam for Windows installs fine but doesn't work yet.
+
+----
+
+Notes about Wine
+----
+
+Wine is now supported. Wine integrated program all runs, and many windows programs and games also runs fine. Don't forget most Windows games use Direct3D, this may require a complete OpenGL driver and a as high profile as possible (and gl4es with ES2 backend have issue with Wine for now).
+Note: if you plan to use box86 with Wine on Raspberry Pi 3 or earlier, those models use a default OS that have a kernel with a 2/2 Split (meaning 2G of space for user program, and 2G of space for the Kernel). This is not compatible with Wine programs that needs to access memory > 2Gb address. So you'll need to reconfigure your kernel for a 3G/1G split.
+
+----
+
+Notes about Vulkan
+----
+
+Box86 already wrap vulikan. If your system has a 32bits vulkan driver, box86 will use it when needed. Note that vulkan wrapping has not been tested much, due to the limited vulkan support on the hardware I currently own. Profile 1.0 and 1.1, whith some extension, should be OK. 1.2 is not really wrapped. I know some demos work on Pi4 (Sascha Willems demos build for x86 work the same as if build on armhf directly). Note that the Vulkan driver of the Pi4 DOES NOT support dxvk for now (wine DirectX->Vulkan wrapper). It's not a box86 issue, it's missing extensions (hardxware support) and a few other things that make dxvk not working on pi4. On Panfrost side, PanVK is a bit young and I haven't tested dxvk with it yet.
+
+----
+Final word
+----
+
+I want to thank everyone who has contributed to box86 developpement.
+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
+ * 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 continous 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.
+
+(If you use Box86 in your project, please don't forget to mention it!)
@@ -0,0 +1,178 @@
+
+Usage
+----
+
+There are many environment variables to control Box86 behaviour. 
+
+#### 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
+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_NOBANNER
+Disables Box86 printing its version and build
+ * 0 : Enable printing its banner. (Default.)
+ * 1 : Disable printing its banner. 
+
+#### 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.
+
+#### BOX86_PATH
+Path to look for x86 executable. Default is current folder and `bin` in current folder.
+
+#### BOX86_DLSYM_ERROR
+Enables/Disables the logging of `dlsym` errors.
+ * 0 : Don't log `dlsym` errors. (Default.)
+ * 1 : Log dlsym errors.
+
+#### 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
+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
+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
+Only on builds with trace enabled.
+ * NNNNNNN : Start trace only after NNNNNNNN opcode execute (number is an `uint64_t`).
+
+#### 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
+Try to load at 0xXXXXXX main binary (if binary is a PIE)
+ * 0xXXXXXXXX : The load address . (Only active on PIE programs.)
+
+#### 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
+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
+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
+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
+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
+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
+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
+Enables/Disables Box86's Dynarec.
+ * 0 : Disables Dynarec.
+ * 1 : Enable Dynarec. (Default.)
+
+#### 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 
+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
+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.)
+
+#### 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
+
+#### 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
+ * 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
+ 
+#### 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_ALLOWMISSINGLIBS
+Allow Box86 to continue even if a library is missing.
+ * 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_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
+Disables the loading of wrapped GTK libraries.
+ * 0 : Load wrapped GTK libraries if found. (Default.)
+ * 1 : Disables loading wrapped GTK libraries (can be used with Steam, along with STEAM_RUNTIME=1 to use the i386 version of GTK).
+
+#### 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_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_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.
+ When in gdb, you need to find the correct thread yourself (the one with `my_box86signalhandler` in is stack)
+ then probably need to `finish` 1 or 2 functions (inside `usleep(..)`) and then you'll be in `my_box86signalhandler`, 
+ just before the printf of the Segfault message. Then simply 
+ `set waiting=0` to exit the infinite loop.
+ * 2 : Launch `gdbserver` when a segfault, bus error or illegal instruction signal is trapped, attached to the offending process, and go in an endless loop, waiting.
+ Use `gdb /PATH/TO/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)
@@ -0,0 +1,137 @@
+# Installing Wine (and winetricks)
+_TwisterOS users: Wine, winetricks, and Box86 are already installed in TwisterOS. You do not have to install anything._
+
+_Raspberry Pi users: Wine requires a 3G/1G split memory kernel. Raspberry Pi OS for the Pi 4 already has a 3G/1G split kernel and works with Wine, but **the Pi 3B+ and earlier models have a 2G/2G kernel and require a custom-compiled 3G/1G kernel to get Wine working.**_
+
+Using Wine with Box86 allows (x86) Windows programs to run on ARM Linux computers (x64 is not yet implemented).  
+See installation steps below (in the [Examples](#examples) section).
+
+Box86 needs `wine-i386` to be installed **manually** on ARM devices.  Even though `wine-armhf` is available in many repo's on ARM devices (ie using _apt-get_ will attempt to install `wine-armhf` by default), `wine-armhf` will not work with Box86.  Note that manual installation is required since using `multiarch` will result in your ARM device thinking it needs to install lots of i386 dependencies to make `wine-i386` work. The "twist" in Box86 is that Box86 "wraps" many of Wine's core Linux i386 libraries (.so files) so that their calls are interpretable by your ARM devices other armhf Linux system libraries.  Also note that wrapping libraries is an ongoing process throughout Box86 development and that some programs may not run properly until all of their i386 library dependencies are wrapped.
+
+Installation files for Wine can be found in the [WineHQ repository](https://dl.winehq.org/wine-builds/debian/dists/buster/main/binary-i386/), the [TwisterOS FAQ](https://twisteros.com/faq.html) page, or the [PlayOnLinux website repository](https://www.playonlinux.com/wine/). Box86 requires the "i386" (x86) version of Wine even though we are installing it on an ARM processor.
+
+## Overview
+The general procedure for installing Wine for Box86 is to...
+ - Download all the install files (.deb, .zip, or even .pol files) for the version of Wine you wish to install
+ - Unzip or dpkg the install files into one folder
+ - Move that folder to the directory that you wish Wine to run from (in TwisterOS, this is `~/wine/` by default)
+ - Go to `/usr/local/bin` and make symlinks or scripts that will point to your main wine binaries (`wine`, `winecfg`, and `wineserver`).
+ - Boot wine to create a new wineprefix (`wine wineboot`).
+ - Download winetricks (which is simply a very complicated bash script), make it executable, then copy it to `/usr/local/bin`.
+
+## Examples
+### Installing Wine for Box86 on Raspberry Pi OS from Twister OS FAQ .tgz files
+_Links from the [TwisterOS FAQ](https://twisteros.com/faq.html)_
+
+This installs Wine v5.13 (which is the version of Wine that comes pre-installed in TwisterOS).  When submitting bug reports, other Box86 users may assume you are using this version of Wine unless you specify otherwise.
+```
+# Backup any old wine installations
+sudo mv ~/wine ~/wine-old
+sudo mv ~/.wine ~/.wine-old
+sudo mv /usr/local/bin/wine /usr/local/bin/wine-old
+sudo mv /usr/local/bin/wineboot /usr/local/bin/wineboot-old
+sudo mv /usr/local/bin/winecfg /usr/local/bin/winecfg-old
+sudo mv /usr/local/bin/wineserver /usr/local/bin/wineserver-old
+
+# Download, extract wine, and install wine (last I checked, the Twister OS FAQ page had Wine 5.13-devel)
+wget https://twisteros.com/wine.tgz -O ~/wine.tgz
+tar -xzvf ~/wine.tgz
+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/wineboot /usr/local/bin/wineboot
+sudo ln -s ~/wine/bin/winecfg /usr/local/bin/winecfg
+sudo ln -s ~/wine/bin/wineserver /usr/local/bin/wineserver
+sudo chmod +x /usr/local/bin/wine /usr/local/bin/wineboot /usr/local/bin/winecfg /usr/local/bin/wineserver
+
+# Boot wine (make fresh wineprefix in ~/.wine )
+wine wineboot
+```
+
+### Installing Wine for Box86 on Raspberry Pi OS Buster from WineHQ .deb files
+_Links from the [WineHQ repo](https://dl.winehq.org/wine-builds/debian/dists/buster/main/binary-i386/)_
+
+This install method allows you to install different versions of Wine.  You may install any version and any branch of Wine that you wish (wine-devel, wine-stable, or wine-staging).
+```
+# Backup any old wine installations
+sudo mv ~/wine ~/wine-old
+sudo mv ~/.wine ~/.wine-old
+sudo mv /usr/local/bin/wine /usr/local/bin/wine-old
+sudo mv /usr/local/bin/wineboot /usr/local/bin/wineboot-old
+sudo mv /usr/local/bin/winecfg /usr/local/bin/winecfg-old
+sudo mv /usr/local/bin/wineserver /usr/local/bin/wineserver-old
+
+# Download, extract wine, and install wine
+# (Replace the links/versions below with links/versions from the WineHQ site for the version of wine you wish to install. Note that we need the i386 version for Box86 even though we're installing it on our ARM processor.)
+# (Pick an i386 version of wine-devel, wine-staging, or wine-stable)
+cd ~/Downloads
+wget https://dl.winehq.org/wine-builds/debian/dists/buster/main/binary-i386/wine-devel-i386_5.21~buster_i386.deb # NOTE: Replace this link with the version you want
+wget https://dl.winehq.org/wine-builds/debian/dists/buster/main/binary-i386/wine-devel_5.21~buster_i386.deb  # NOTE: Also replace this link with the version you want
+dpkg-deb -xv wine-devel-i386_5.21~buster_i386.deb wine-installer # NOTE: Make sure these dpkg command matches the filename of the deb package you just downloaded
+dpkg-deb -xv wine-devel_5.21~buster_i386.deb wine-installer
+mv ~/Downloads/wine-installer/opt/wine* ~/wine
+rm wine*.deb # clean up
+rm -rf wine-installer # clean up
+
+# Install shortcuts (make 32bit 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/wineboot /usr/local/bin/wineboot
+sudo ln -s ~/wine/bin/winecfg /usr/local/bin/winecfg
+sudo ln -s ~/wine/bin/wineserver /usr/local/bin/wineserver
+sudo chmod +x /usr/local/bin/wine /usr/local/bin/wineboot /usr/local/bin/winecfg /usr/local/bin/wineserver
+
+# These packages are needed for running wine-staging on RPi 4 (Credits: chills340)
+sudo apt install libstb0 -y
+cd ~/Downloads
+wget -r -l1 -np -nd -A "libfaudio0_*~bpo10+1_i386.deb" http://ftp.us.debian.org/debian/pool/main/f/faudio/ # Download libfaudio i386 no matter its version number
+dpkg-deb -xv libfaudio0_*~bpo10+1_i386.deb libfaudio
+sudo cp -TRv libfaudio/usr/ /usr/
+rm libfaudio0_*~bpo10+1_i386.deb # clean up
+rm -rf libfaudio # clean up
+
+# Boot wine (make fresh wineprefix in ~/.wine )
+wine wineboot
+```
+
+## Installing winetricks
+Winetricks is a bash script which makes it easier to install & configure any desired Windows core system software packages which may be dependencies for certain Windows programs.  You can either install it with `apt`, or manually using the steps below.
+
+```
+sudo apt-get install cabextract -y                                                                   # winetricks needs this installed
+sudo mv /usr/local/bin/winetricks /usr/local/bin/winetricks-old                                      # Backup old winetricks
+cd ~/Downloads && wget https://raw.githubusercontent.com/Winetricks/winetricks/master/src/winetricks # Download
+sudo chmod +x winetricks && sudo mv winetricks /usr/local/bin/                                       # Install
+```
+
+Whenever we run winetricks, we must tell Box86 to supress its info banner, otherwise winetricks will crash.  Box86's info banner can be suppressed using the `BOX86_NOBANNER=1` environment variable before invoking winetricks (example: `BOX86_NOBANNER=1 winetricks`).
+
+If `apt` installed a desktop menu shortcut for winetricks (or if you make your own desktop shortcut for winetricks) then you may need to edit that shortcut to include Box86's BOX86_NOBANNER=1 env var.  With your favorite text editor, edit `/usr/share/applications/winetricks.desktop` and change `Exec=winetricks --gui` to `Exec=env BOX86_NOBANNER=1 winetricks --gui`.
+
+## Example commands
+Here is an example of how we should run a winetricks command with box86:  
+`BOX86_NOBANNER=1 winetricks -q corefonts vcrun2010 dotnet20sp1`  
+_This command will silently install three packages in series: Windows core fonts, VC++ 2010 Runtimes, and .NET 2.0 SP1.  `-q` is the "install silent/quiet" command._
+
+Whenever we run winetricks, we must suppress Box86's banner by typing `BOX86_NOBANNER=1` to prevent winetricks from crashing.  Invoking Box86's logging features (with `BOX86_LOG=1` or similar) will also cause winetricks to crash.  (If Box86 logging is necessary though, we can patch winetricks to avoid these crashes - see the *Troubleshooting* section) # future work.
+
+For a list of all the different Windows packages & libraries that winetricks can help you install, run `winetricks list-all`  
+
+## Other notes
+
+### Wineprefixes (and Wine initialization)
+When you first run or boot Wine (`wine wineboot`), Wine will create a new user environment within which to install Windows software.  This user environment is called a "wineprefix" (or "wine bottle") and is located (by default) in `~/.wine` (note that Linux folders with a `.` in front of them are "hidden" folders).  Think of a wineprefix as Wine's virtual 'harddrive' where it installs software and saves settings.  Wineprefixes are portable and deletable.  For more Wine documentation, see [WineHQ](https://www.winehq.org/documentation).
+
+If you at any point corrupt something inside your default wineprefix, you can start "fresh" by deleting your `~/.wine` directory (with the `rm -rf ~/.wine` command) and boot wine again to create a new default wineprefix.
+
+### Transplanting wineprefixes (side-loading)
+If software isn't installing in Wine with Box86, but is installing for you in Wine on a regular x86 Linux computer, you can copy a wineprefix from your x86 Linux computer to the device you're running Box86 on. This is most easily done by tarring the `~/.wine` folder on your x86 Linux computer (`tar -cvf winebottle.tar ~/.wine`), transferring the tar file to your device, then un-tarring the tar file (`tar -xf winebottle.tar`) on your device running Box86 & Wine. Tarring the wineprefix preserves any symlinks that are inside it.
+
+### Swapping out different versions of Wine
+Some versions of Wine work better with certain software. It is best to install a version of Wine that is known to work with the software you would like to run. There are three main development branches of Wine you can pick from, referred to as wine-stable, wine-devel, and wine-staging. _Note that the wine-staging branch requires extra installation steps on Raspberry Pi._
+
+Your entire Wine installation can reside within a single folder on your Linux computer. TwisterOS assumes that your Wine installation is located inside the `~/wine/` directory. The actual directory where you put your `wine` folder doesn't matter as long as you have symlinks within the `/usr/local/bin/` directory which point to the `wine` folder so that Linux can find Wine when you type `wine` into the terminal).
+
+You can change which version of Wine you are running simply by renaming your old `wine` and `.wine` folders to something else, then putting a new `wine` folder (containing your new version of Wine) in its place. Running `wine wineboot` again will migrate a wineprefix from an older version of Wine to be usable by the newer version of Wine you just installed.  You can check which version of Wine you're running with the `wine --version` command.
@@ -0,0 +1 @@
+theme: jekyll-theme-hacker
@@ -0,0 +1,44 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-arm-git
+pkgver=3494.bedacef
+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%-arm-git}-git" "${pkgname%-arm-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-arm-git}-git" "${pkgname%-arm-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	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}"
+	if [[ ! -d ./build ]]; then
+		mkdir build && cd build
+		cmake .. -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo
+	fi
+}
+
+build() {
+	cd "$srcdir/${pkgname%-arm-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "$srcdir/${pkgname%-arm-git}/build"
+	make DESTDIR="${pkgdir}/" install
+	# /usr/local/bin isn't in PATH by the default,
+	# we should move it to /usr/bin
+	cd ${pkgdir}
+	mv usr/local/bin/ usr/bin/
+	# cleanup when dir is empty
+	rmdir usr/local || exit 0
+}
@@ -0,0 +1,44 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-gameshell-git
+pkgver=3494.bedacef
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-gameshell-git}-git" "${pkgname%-gameshell-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-gameshell-git}-git" "${pkgname%-gameshell-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "$srcdir/${pkgname%-gameshell-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-gameshell-git}"
+	if [[ ! -d ./build ]]; then
+		mkdir build && cd build
+		cmake .. -DGAMESHELL=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
+	fi
+}
+
+build() {
+	cd "$srcdir/${pkgname%-gameshell-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "$srcdir/${pkgname%-gameshell-git}/build"
+	make DESTDIR="${pkgdir}/" install
+	# /usr/local/bin isn't in PATH by the default,
+	# we should move it to /usr/bin
+	cd ${pkgdir}
+	mv usr/local/bin/ usr/bin/
+	# cleanup when dir is empty
+	rmdir usr/local || exit 0
+}
@@ -0,0 +1,44 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-odroid-git
+pkgver=3494.bedacef
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-odroid-git}-git" "${pkgname%-odroid-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-odroid-git}-git" "${pkgname%-odroid-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "$srcdir/${pkgname%-odroid-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-odroid-git}"
+	if [[ ! -d ./build ]]; then
+		mkdir build && cd build
+		cmake .. -DODROID=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
+	fi
+}
+
+build() {
+	cd "$srcdir/${pkgname%-odroid-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "$srcdir/${pkgname%-odroid-git}/build"
+	make DESTDIR="${pkgdir}/" install
+	# /usr/local/bin isn't in PATH by the default,
+	# we should move it to /usr/bin
+	cd ${pkgdir}
+	mv usr/local/bin/ usr/bin/
+	# cleanup when dir is empty
+	rmdir usr/local || exit 0
+}
@@ -0,0 +1,44 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-pandora-git
+pkgver=3494.bedacef
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-pandora-git}"-git "${pkgname%-pandora-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-pandora-git}"-git "${pkgname%-pandora-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "$srcdir/${pkgname%-pandora-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-pandora-git}"
+	if [[ ! -d ./build ]]; then
+		mkdir build && cd build
+		cmake .. -DPANDORA=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
+	fi
+}
+
+build() {
+	cd "$srcdir/${pkgname%-pandora-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "$srcdir/${pkgname%-git}/build"
+	make DESTDIR="${pkgdir}/" install
+	# /usr/local/bin isn't in PATH by the default,
+	# we should move it to /usr/bin
+	cd ${pkgdir}
+	mv usr/local/bin/ usr/bin/
+	# cleanup when dir is empty
+	rmdir usr/local || exit 0
+}
@@ -0,0 +1,44 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-pyra-git
+pkgver=3494.bedacef
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-pyra-git}-git" "${pkgname%-pyra-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-pyra-git}-git" "${pkgname%-pyra-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "$srcdir/${pkgname%-pyra-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-pyra-git}"
+	if [[ ! -d ./build ]]; then
+		mkdir build && cd build
+		cmake .. -DPYRA=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
+	fi
+}
+
+build() {
+	cd "$srcdir/${pkgname%-pyra-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "$srcdir/${pkgname%-pyra-git}/build"
+	make DESTDIR="${pkgdir}/" install
+	# /usr/local/bin isn't in PATH by the default,
+	# we should move it to /usr/bin
+	cd ${pkgdir}
+	mv usr/local/bin/ usr/bin/
+	# cleanup when dir is empty
+	rmdir usr/local || exit 0
+}
@@ -0,0 +1,59 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-rpi-git
+pkgver=3494.bedacef
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-rpi-git}-git" "${pkgname%-rpi-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-rpi-git}-git" "${pkgname%-rpi-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+_pimodel=`tr -d '\0' < /sys/firmware/devicetree/base/model`
+
+pkgver() {
+	cd "$srcdir/${pkgname%-rpi-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-rpi-git}"
+	# Check for Pi version number
+	if [[ ${_pimodel} =~ "Raspberry Pi 4" ]]; then
+		_piversion=4
+	elif [[ ${_pimodel} =~ "Raspberry Pi 3" ]]; then
+		_piversion=3
+	elif [[ ${_pimodel} =~ "Raspberry Pi 2" ]]; then
+		_piversion=2
+	else
+		# This RPI is not supported or this device is not a pi.
+		# We should cancel then the build.
+		echo -e "\033[01m\033[01;31m==> ERROR:\033[00;01m Your device is not supported! This script is for RPi model 4/3/2 (1/0 isn't supported yet).\033[00m"
+		exit 1
+	fi
+	echo -e "\033[01m\033[01;34m==> INFO:\033[00;01m Your detected Pi verion number is: \033[01;31m${_piversion}\033[01m.\033[00m"
+	if [[ ! -d ./build ]]; then
+		mkdir build && cd build
+		cmake .. -DRPI${_piversion}=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
+	fi
+}
+
+build() {
+	cd "$srcdir/${pkgname%-rpi-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "$srcdir/${pkgname%-rpi-git}/build"
+	make DESTDIR="${pkgdir}/" install
+	# /usr/local/bin isn't in PATH by the default,
+	# we should move it to /usr/bin
+	cd ${pkgdir}
+	mv usr/local/bin/ usr/bin/
+	# cleanup when dir is empty
+	rmdir usr/local || exit 0
+}
@@ -0,0 +1,44 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-rk3399-git
+pkgver=3494.bedacef
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-rk3399-git}-git" "${pkgname%-rk3399-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-rk3399-git}-git" "${pkgname%-rk3399-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "$srcdir/${pkgname%-rk3399-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-rk3399-git}"
+	if [[ ! -d ./build ]]; then
+		mkdir build && cd build
+		cmake .. -DRK3399=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
+	fi
+}
+
+build() {
+	cd "$srcdir/${pkgname%-rk3399-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "$srcdir/${pkgname%-arm-git}/build"
+	make DESTDIR="${pkgdir}/" install
+	# /usr/local/bin isn't in PATH by the default,
+	# we should move it to /usr/bin
+	cd ${pkgdir}
+	mv usr/local/bin/ usr/bin/
+	# cleanup when dir is empty
+	rmdir usr/local || exit 0
+}
@@ -0,0 +1,43 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-x86-git
+pkgver=3494.bedacef
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('i386')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-x86-git}-git" "${pkgname%-x86-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-x86-git}-git" "${pkgname%-x86-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "$srcdir/${pkgname%-x86-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-x86-git}"
+	if [[ ! -d ./build ]]; then
+		mkdir build && cd build
+		cmake .. -DLD80BITS=1 -DNOALIGN=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
+	fi
+}
+
+build() {
+	cd "$srcdir/${pkgname%-x86-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "$srcdir/${pkgname%-x86-git}/build"
+	make DESTDIR="${pkgdir}/" install
+	# /usr/local/bin isn't in PATH by the default,
+	# we should move it to /usr/bin
+	cd ${pkgdir}
+	mv usr/local/bin/ usr/bin/
+	# cleanup when dir is empty
+	rmdir usr/local || exit 0
+}
@@ -0,0 +1,24 @@
+# The directory containing all PKGBUILDs for BOX86
+## FAQ
+### 1. How to use it?
+Just download the `PKGBUILD-<your-platform>` (no need to download the whole repository), go to directory where you downloaded and type:
+```sh
+mv ./PKGBUILD-<your-platform> ./PKGBUILD
+cd ..
+makepkg -si
+```
+This will copy the PKGBUILD for your platform (where you type your platform name instead of `<your-platform>`), build it and install the PKGBUILD on 32-bit Archlinux ARM.
+
+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](../COMPILE.md).
+
+### 3. Any advantages with that sollution?
+This might give you some advantages like those:
+- automatic updates (with some AUR frontends like [`pamac-aur`](https://aur.archlinux.org/packages/pamac-aur/) or [`yay`](https://aur.archlinux.org/packages/yay));
+- easily uninstall with `pacman -R` when no needed (no need other scripts nor Makefile to do this);
+- your `box86` installation is recognized as a package (where it wasn't when installing with `make install`).
+
+
+### Note: You can use [makedeb](https://github.com/makedeb/makedeb) (or) [pkgbuild2spec](https://github.com/prozum/pkgbuild2spec) to compile box86 to .deb or .rpm installer files.
@@ -0,0 +1,2 @@
+#!/bin/sh
+systemctl restart systemd-binfmt
@@ -5,6 +5,13 @@ import re
 import sys
 from math import ceil

+"""
+Generates src/dynarec/arm_printer.c
+===
+
+See src/dynarec/arm_instructions.txt (the input file) for the syntax documentation.
+"""
+
 # Helper class to avoid displaying '\x1b[' on errors
 class string(str):
 	def __repr__(self):
@@ -172,13 +179,6 @@ def main(root, ver, __debug_forceAllDebugging=False):
 			- specifics is an array of arrays the same length as positions that contains a tuple (mask, correctBit)
 			  that is at position positions[current_pos]
 			"""
-			if len(positions) != len(specifics):
-				fail(
-					AssertionError,
-					"generate_bin_tests requires the same length for positions ({}) and specifics ({})!".format(
-						len(positions), len(specifics)
-					)
-				)
 			
 			if specifics == []:
 				append("if ((opcode & " + arr2hex(mask) + ") == " + arr2hex(correctBits) + ") {\n")
@@ -194,8 +194,9 @@ def main(root, ver, __debug_forceAllDebugging=False):
 				if any(map(lambda s: len(s) != l, specifics)):
 					fail(
 						AssertionError,
-						"generate_bin_tests requires the same length for positions ({}) and specifics ({})!".format(
-							len(positions), len(specifics)
+						"generate_bin_tests requires the same length for positions ({}) and each element "
+						"of the specifics array ({})!".format(
+							len(positions), [len(s) for s in specifics]
 						)
 					)
 				
@@ -290,7 +291,7 @@ def main(root, ver, __debug_forceAllDebugging=False):
 							# Custom ifs
 							# Also, requires only a single '='
 							if eq != "=":
-								fail(ValueError, "Too many '=' switches (!@ modifier)")
+								fail(ValueError, "Too many '=' switches (@@ modifier)")
 							
 							# Extract the statements
 							statements = []
@@ -575,11 +576,11 @@ def main(root, ver, __debug_forceAllDebugging=False):
 					append("int rot = (opcode >> " + str(variables["r"]) + ") & 3;\nchar tmprot[8] = {0};\n")
 					append("if (rot) {\nsprintf(tmprot, \" ror %d\", rot * 8);\n}\n")
 				if variables["sb"][0] != -1:
-					append("int lsb = (opcode >> " + str(variables["sb"][0]) + ") & 0xF;\n")
+					append("int lsb = (opcode >> " + str(variables["sb"][0]) + ") & 0x1F;\n")
 				if variables["sb"][1] != -1:
-					append("int msb = (opcode >> " + str(variables["sb"][1]) + ") & 0xF;\n")
+					append("int msb = (opcode >> " + str(variables["sb"][1]) + ") & 0x1F;\n")
 				if variables["sb"][2] != -1:
-					append("int widthm1 = (opcode >> " + str(variables["sb"][2]) + ") & 0xF;\n")
+					append("int widthm1 = (opcode >> " + str(variables["sb"][2]) + ") & 0x1F;\n")
 				if variables["reglist16"] != -1:
 					append("int reglist = (opcode >> " + str(variables["reglist16"]) + ") & 0xFFFF;\n")
 				if imms != []:
@@ -620,7 +621,7 @@ def main(root, ver, __debug_forceAllDebugging=False):
 					skiprbr = False
 					if idx % 2:
 						if text == "c":
-							if variables["cond"] != -1:
+							if variables["cond"] >= 0:
 								append("%s")
 								printf_args = printf_args + ", cond"
 						elif text == "Rd":
@@ -1105,7 +1106,7 @@ def main(root, ver, __debug_forceAllDebugging=False):
 						
 						l = len(imms) == 2
 						if l:
-							append("uint8_t l = (opcode >> " + str(imms[1][1]) + ") & 0x1;\n")
+							append("uint8_t l = (opcode >> " + str(imms[1][0]) + ") & 0x1;\n")
 						
 						immssz = 6
 						append("uint8_t imm6 = (opcode >> " + str(imms[0][0]) + ") & 0x3F;\n")
@@ -1653,7 +1654,7 @@ def main(root, ver, __debug_forceAllDebugging=False):
 					
 					# No C variable since we're invalidating!
 					
-					# Now print the invalidation, numerote if debugging
+					# Now print the invalidation, number if debugging
 					if numberInvalids:
 						append("strcpy(ret, \"??? #" + str(invalidationCount) + "\");\n} else ")
 						invalidationCount = invalidationCount + 1
@@ -1674,301 +1675,309 @@ def main(root, ver, __debug_forceAllDebugging=False):
 	
 	# Now the files rebuilding part
 	# File header and guard
-	header = """/*****************************************************************
- * File automatically generated by rebuild_printer.py (v{version})
- *****************************************************************/
-"""
 	files_header = {
-		"arm_printer.c": """#include <stdint.h>
-#include <stddef.h>
-#include <string.h>
-#include <stdio.h>
-
-#include "arm_printer.h"
-
-// conditions
-#define cEQ (0b0000<<28)
-#define cNE (0b0001<<28)
-#define cCS (0b0010<<28)
-#define cCC (0b0011<<28)
-#define cMI (0b0100<<28)
-#define cPL (0b0101<<28)
-#define cVS (0b0110<<28)
-#define cVC (0b0111<<28)
-#define cHI (0b1000<<28)
-#define cLS (0b1001<<28)
-#define cGE (0b1010<<28)
-#define cLT (0b1011<<28)
-#define cGT (0b1100<<28)
-#define cLE (0b1101<<28)
-#define c__ (0b1110<<28)	// means all
-
-static const char* conds[16] = {
-	"EQ", "NE", "CS", "CC",
-	"MI", "PL", "VS", "VC",
-	"HI", "LS", "GE", "LT",
-	"GT", "LE",   "", "##"
-};
-
-static const char* regname[16] = {
-	"r0", "r1", "r2", "r3", "r4",
-	"r5", "r6", "r7", "r8", "r9",
-	"r10", "r11", "r12", "SP", "LR", "PC"
-};
-
-// Single precision are V_:_
-// Double and quad precision are _:V_
-// So we always use _:V_ and invert the corresponding single-precision (cleaner? code)
-static const char* vecname[96] = {
-	 "S0",  "S2",  "S4",  "S6",  "S8", "S10", "S12", "S14",
-	"S16", "S18", "S20", "S22", "S24", "S26", "S28", "S30",
-	 "S1",  "S3",  "S5",  "S7",  "S9", "S11", "S13", "S15",
-	"S17", "S19", "S21", "S23", "S25", "S27", "S29", "S31",
-	"D0", "D1", "D2", "D3", "D4", "D5", "D6", "D7",
-	"D8", "D9", "D10", "D11", "D12", "D13", "D14", "D15",
-	"D16", "D17", "D18", "D19", "D20", "D21", "D22", "D23",
-	"D24", "D25", "D26", "D27", "D28", "D29", "D30", "D31",
-	"Q0", "!!!", "Q1", "!!!", "Q2", "!!!", "Q3", "!!!",
-	"Q4", "!!!", "Q5", "!!!", "Q6", "!!!", "Q7", "!!!",
-	"Q8", "!!!", "Q9", "!!!", "Q10", "!!!", "Q11", "!!!",
-	"Q12", "!!!", "Q13", "!!!", "Q14", "!!!", "Q15", "!!!"
-};
-
-static const char* dts[16] = {
-	"S8", "S16", "S32", "S64",
-	"U8", "U16", "U32", "U64",
-	"I8", "I16", "I32", "I64",
-	"F8", "F16", "F32", "F64"
-};
-
-static const char* shift_type[4] = {
-	"lsl ", "lsr ", "asr ", "ror "
-};
-
-const char* print_shift(int shift, int comma) {
-	static __thread char ret[20];
-	ret[0] = '\\0';
-	
-	int sh_op = (shift >> 1) & 3;
-	if(shift & 1) {
-		int rs = shift >> 4;
-		sprintf(ret, "%s%s%s", (comma ? ", " : ""), shift_type[sh_op], regname[rs]);
-	} else {
-		uint8_t amount = shift >> 3;
-		if (!amount) {
-			switch (sh_op) {
-			case 0b00:
-				break;
-			case 0b01:
-			case 0b10:
-				sprintf(ret, "%s%s#32", (comma ? ", " : ""), shift_type[sh_op]);
-				break;
-			case 0b11:
-				sprintf(ret, "%srrx", (comma ? ", " : ""));
-				break;
-			}
-		} else {
-			sprintf(ret, "%s%s#%u", (comma ? ", " : ""), shift_type[sh_op], amount);
-		}
-	}
-	return ret;
-}
-
-#define print_modified_imm_ARM(imm12) (((imm12 & 0xFF) >> (imm12 >> 8)) | ((imm12 & 0xFF) << (32 - (imm12 >> 8))))
-
-const char* print_register_list(int list, int size) {
-	int last = -2;
-	int cnt = 0;
-	static __thread char ret[68];
-	
-	ret[0] = '{';
-	ret[1] = '\\0';
-	for (int i = 0; i < size; ++i) {
-		if (list & (1 << i)) {
-			if (last >= 0) {
-				++cnt;
+		"arm_printer.c": """
+		#include <stdint.h>
+		#include <stddef.h>
+		#include <string.h>
+		#include <stdio.h>
+		
+		#include "arm_printer.h"
+		
+		// conditions
+		#define cEQ (0b0000<<28)
+		#define cNE (0b0001<<28)
+		#define cCS (0b0010<<28)
+		#define cCC (0b0011<<28)
+		#define cMI (0b0100<<28)
+		#define cPL (0b0101<<28)
+		#define cVS (0b0110<<28)
+		#define cVC (0b0111<<28)
+		#define cHI (0b1000<<28)
+		#define cLS (0b1001<<28)
+		#define cGE (0b1010<<28)
+		#define cLT (0b1011<<28)
+		#define cGT (0b1100<<28)
+		#define cLE (0b1101<<28)
+		#define c__ (0b1110<<28)	// means all
+		
+		static const char* conds[16] = {
+			"EQ", "NE", "CS", "CC",
+			"MI", "PL", "VS", "VC",
+			"HI", "LS", "GE", "LT",
+			"GT", "LE",   "", "##"
+		};
+		
+		static const char* regname[16] = {
+			"r0", "r1", "r2", "r3", "r4",
+			"r5", "r6", "r7", "r8", "r9",
+			"r10", "r11", "r12", "SP", "LR", "PC"
+		};
+		
+		// Single precision are V_:_
+		// Double and quad precision are _:V_
+		// So we always use _:V_ and invert the corresponding single-precision (cleaner? code)
+		static const char* vecname[96] = {
+			"S0",  "S2",  "S4",  "S6",  "S8", "S10", "S12", "S14",
+			"S16", "S18", "S20", "S22", "S24", "S26", "S28", "S30",
+			"S1",  "S3",  "S5",  "S7",  "S9", "S11", "S13", "S15",
+			"S17", "S19", "S21", "S23", "S25", "S27", "S29", "S31",
+			"D0", "D1", "D2", "D3", "D4", "D5", "D6", "D7",
+			"D8", "D9", "D10", "D11", "D12", "D13", "D14", "D15",
+			"D16", "D17", "D18", "D19", "D20", "D21", "D22", "D23",
+			"D24", "D25", "D26", "D27", "D28", "D29", "D30", "D31",
+			"Q0", "!!!", "Q1", "!!!", "Q2", "!!!", "Q3", "!!!",
+			"Q4", "!!!", "Q5", "!!!", "Q6", "!!!", "Q7", "!!!",
+			"Q8", "!!!", "Q9", "!!!", "Q10", "!!!", "Q11", "!!!",
+			"Q12", "!!!", "Q13", "!!!", "Q14", "!!!", "Q15", "!!!"
+		};
+		
+		static const char* dts[16] = {
+			"S8", "S16", "S32", "S64",
+			"U8", "U16", "U32", "U64",
+			"I8", "I16", "I32", "I64",
+			"F8", "F16", "F32", "F64"
+		};
+		
+		static const char* shift_type[4] = {
+			"lsl ", "lsr ", "asr ", "ror "
+		};
+		
+		const char* print_shift(int shift, int comma) {
+			static __thread char ret[20];
+			ret[0] = '\\0';
+			
+			int sh_op = (shift >> 1) & 3;
+			if(shift & 1) {
+				int rs = shift >> 4;
+				sprintf(ret, "%s%s%s", (comma ? ", " : ""), shift_type[sh_op], regname[rs]);
 			} else {
-				if (last == -1)
-					strcat(ret, ", ");
-				strcat(ret, regname[i]);
-				last = i;
+				uint8_t amount = shift >> 3;
+				if (!amount) {
+					switch (sh_op) {
+					case 0b00:
+						break;
+					case 0b01:
+					case 0b10:
+						sprintf(ret, "%s%s#32", (comma ? ", " : ""), shift_type[sh_op]);
+						break;
+					case 0b11:
+						sprintf(ret, "%srrx", (comma ? ", " : ""));
+						break;
+					}
+				} else {
+					sprintf(ret, "%s%s#%u", (comma ? ", " : ""), shift_type[sh_op], amount);
+				}
+			}
+			return ret;
+		}
+		
+		#define print_modified_imm_ARM(imm12) (((imm12 & 0xFF) >> (2*(imm12 >> 8))) | ((imm12 & 0xFF) << (32 - 2*(imm12 >> 8))))
+		
+		const char* print_register_list(int list, int size) {
+			int last = -2;
+			int cnt = 0;
+			static __thread char ret[68];
+			
+			ret[0] = '{';
+			ret[1] = '\\0';
+			for (int i = 0; i < size; ++i) {
+				if (list & (1 << i)) {
+					if (last >= 0) {
+						++cnt;
+					} else {
+						if (last == -1)
+							strcat(ret, ", ");
+						strcat(ret, regname[i]);
+						last = i;
+					}
+				} else {
+					if (cnt) {
+						strcat(ret, "-");
+						strcat(ret, regname[i - 1]);
+					}
+					if (last != -2) last = -1;
+					cnt = 0;
+				}
 			}
-		} else {
 			if (cnt) {
 				strcat(ret, "-");
-				strcat(ret, regname[i - 1]);
+				strcat(ret, regname[size - 1]);
 			}
-			if (last != -2) last = -1;
-			cnt = 0;
+			strcat(ret, "}");
+			
+			return ret;
 		}
-	}
-	if (cnt) {
-		strcat(ret, "-");
-		strcat(ret, regname[size - 1]);
-	}
-	strcat(ret, "}");
-	
-	return ret;
-}
-
-const char* print_register_list_fpu(uint8_t start, uint8_t size, uint8_t double_prec) {
-	static __thread char ret[68];
-	
-	// Assume VFPSmallRegisterBank(), so no D16 and no S32 (and above)
-	if (start + (double_prec + 1) * size > 32) {
-		return "!!!";
-	}
-	if (size == 0) {
-		return "!!!";
-	}
-	if (double_prec && (size % 2 == 1)) {
-		strcpy(ret, "!! {");
-	} else {
-		ret[0] = '{';
-		ret[1] = '\\0';
-	}
-	
-	char regChr;
-	if (double_prec) {
-		regChr = 'D';
-		size /= 2;
-	} else {
-		regChr = 'S';
-		start = (start >> 4) + ((start & 0xF) << 1);
-	}
-	
-	char tmp[6];
-	for (int cur = start; cur < start + size - 1; ++cur) {
-		sprintf(tmp, "%c%d, ", regChr, cur);
-		strcat(ret, tmp);
-	}
-	sprintf(tmp, "%c%d}", regChr, start + size - 1);
-	strcat(ret, tmp);
-	
-	return ret;
-}
-
-static const char* _print_modified_imm_I64_bytes[2] = {"00", "FF"};
-
-const char* print_modified_imm_NEON(int op_cmode, int imm) {
-	static __thread char ret[34] = "???\\0-"; // [dt: 4][imm: 30]
-	
-	strcpy(&ret[4], "UNPREDICTABLE");
-	
-	switch (op_cmode) {
-	case 0b00000: // VMOV.I32
-	case 0b00001: // VORR.I32
-	case 0b10000: // VMVN.I32
-	case 0b10001: // VBIC.I32
-		strcpy(ret, "I32");
-		sprintf(&ret[4], "0x%02X", imm);
-		break;
-	case 0b00010: // VMOV.I32
-	case 0b00011: // VORR.I32
-	case 0b10010: // VMVN.I32
-	case 0b10011: // VBIC.I32
-		if (imm) {
-			strcpy(ret, "I32");
-			sprintf(&ret[4], "0x%02X00", imm);
+		
+		const char* print_register_list_fpu(uint8_t start, uint8_t size, uint8_t double_prec) {
+			static __thread char ret[68];
+			
+			// Assume VFPSmallRegisterBank(), so no D16 and no S32 (and above)
+			if (start + (double_prec + 1) * size > 32) {
+				return "!!!";
+			}
+			if (size == 0) {
+				return "!!!";
+			}
+			if (double_prec && (size % 2 == 1)) {
+				strcpy(ret, "!! {");
+			} else {
+				ret[0] = '{';
+				ret[1] = '\\0';
+			}
+			
+			char regChr;
+			if (double_prec) {
+				regChr = 'D';
+				size /= 2;
+			} else {
+				regChr = 'S';
+				start = (start >> 4) + ((start & 0xF) << 1);
+			}
+			
+			char tmp[6];
+			for (int cur = start; cur < start + size - 1; ++cur) {
+				sprintf(tmp, "%c%d, ", regChr, cur);
+				strcat(ret, tmp);
+			}
+			sprintf(tmp, "%c%d}", regChr, start + size - 1);
+			strcat(ret, tmp);
+			
+			return ret;
 		}
-		break;
-	case 0b00100: // VMOV.I32
-	case 0b00101: // VORR.I32
-	case 0b10100: // VMVN.I32
-	case 0b10101: // VBIC.I32
-		if (imm) {
-			strcpy(ret, "I32");
-			sprintf(&ret[4], "0x%02X0000", imm);
+		
+		static const char* _print_modified_imm_I64_bytes[2] = {"00", "FF"};
+		
+		const char* print_modified_imm_NEON(int op_cmode, int imm) {
+			static __thread char ret[34] = "???\\0-"; // [dt: 4][imm: 30]
+			
+			strcpy(&ret[4], "UNPREDICTABLE");
+			
+			switch (op_cmode) {
+			case 0b00000: // VMOV.I32
+			case 0b00001: // VORR.I32
+			case 0b10000: // VMVN.I32
+			case 0b10001: // VBIC.I32
+				strcpy(ret, "I32");
+				sprintf(&ret[4], "0x%02X", imm);
+				break;
+			case 0b00010: // VMOV.I32
+			case 0b00011: // VORR.I32
+			case 0b10010: // VMVN.I32
+			case 0b10011: // VBIC.I32
+				if (imm) {
+					strcpy(ret, "I32");
+					sprintf(&ret[4], "0x%02X00", imm);
+				}
+				break;
+			case 0b00100: // VMOV.I32
+			case 0b00101: // VORR.I32
+			case 0b10100: // VMVN.I32
+			case 0b10101: // VBIC.I32
+				if (imm) {
+					strcpy(ret, "I32");
+					sprintf(&ret[4], "0x%02X0000", imm);
+				}
+				break;
+			case 0b00110: // VMOV.I32
+			case 0b00111: // VORR.I32
+			case 0b10110: // VMVN.I32
+			case 0b10111: // VBIC.I32
+				if (imm) {
+					strcpy(ret, "I32");
+					sprintf(&ret[4], "0x%02X000000", imm);
+				}
+				break;
+			case 0b01000: // VMOV.I16
+			case 0b01001: // VORR.I16
+			case 0b11000: // VMVN.I16
+			case 0b11001: // VBIC.I16
+				strcpy(ret, "I16");
+				sprintf(&ret[4], "0x%02X", imm);
+				break;
+			case 0b01010: // VMOV.I16
+			case 0b01011: // VORR.I16
+			case 0b11010: // VMVN.I16
+			case 0b11011: // VBIC.I16
+				if (imm) {
+					strcpy(ret, "I16");
+					sprintf(&ret[4], "0x%02X00", imm);
+				}
+				break;
+			case 0b01100: // VMOV.I32
+			case 0b11100: // VMVN.I32
+				if (imm) {
+					strcpy(ret, "I32");
+					sprintf(&ret[4], "0x%02XFF", imm);
+				}
+				break;
+			case 0b01101: // VMOV.I32
+			case 0b11101: // VMVN.I32
+				if (imm) {
+					strcpy(ret, "I32");
+					sprintf(&ret[4], "0x%02XFFFF", imm);
+				}
+				break;
+			case 0b01110: // VMOV.I8
+				strcpy(ret, "I8");
+				sprintf(&ret[4], "0x%02X", imm);
+				break;
+			case 0b01111: // VMOV.F32
+				strcpy(ret, "F32");
+				sprintf(&ret[4], "%e", (imm & 0x80 ? -1 : 1) * (1 << (((imm >> 4) & 0x7) ^ 0b100)) * (float) (16 + (imm & 0xF)) / 128.);
+				break;
+			case 0b11110: // VMOV.I64
+				{
+					strcpy(ret, "I64");
+					sprintf(&ret[4], "#0x%s%s%s%s%s%s%s%s",
+						_print_modified_imm_I64_bytes[(imm >> 7) & 1], _print_modified_imm_I64_bytes[(imm >> 6) & 1],
+						_print_modified_imm_I64_bytes[(imm >> 5) & 1], _print_modified_imm_I64_bytes[(imm >> 4) & 1],
+						_print_modified_imm_I64_bytes[(imm >> 3) & 1], _print_modified_imm_I64_bytes[(imm >> 2) & 1],
+						_print_modified_imm_I64_bytes[(imm >> 1) & 1], _print_modified_imm_I64_bytes[(imm >> 0) & 1]);
+				}
+				break;
+			case 0b11111: // V???.-
+			default:
+				strcpy(ret, "-");
+				strcpy(&ret[4], "UNDEFINED");
+				break;
+			}
+			
+			return ret;
 		}
-		break;
-	case 0b00110: // VMOV.I32
-	case 0b00111: // VORR.I32
-	case 0b10110: // VMVN.I32
-	case 0b10111: // VBIC.I32
-		if (imm) {
-			strcpy(ret, "I32");
-			sprintf(&ret[4], "0x%02X000000", imm);
-		}
-		break;
-	case 0b01000: // VMOV.I16
-	case 0b01001: // VORR.I16
-	case 0b11000: // VMVN.I16
-	case 0b11001: // VBIC.I16
-		strcpy(ret, "I16");
-		sprintf(&ret[4], "0x%02X", imm);
-		break;
-	case 0b01010: // VMOV.I16
-	case 0b01011: // VORR.I16
-	case 0b11010: // VMVN.I16
-	case 0b11011: // VBIC.I16
-		if (imm) {
-			strcpy(ret, "I16");
-			sprintf(&ret[4], "0x%02X00", imm);
-		}
-		break;
-	case 0b01100: // VMOV.I32
-	case 0b11100: // VMVN.I32
-		if (imm) {
-			strcpy(ret, "I32");
-			sprintf(&ret[4], "0x%02XFF", imm);
-		}
-		break;
-	case 0b01101: // VMOV.I32
-	case 0b11101: // VMVN.I32
-		if (imm) {
-			strcpy(ret, "I32");
-			sprintf(&ret[4], "0x%02XFFFF", imm);
-		}
-		break;
-	case 0b01110: // VMOV.I8
-		strcpy(ret, "I8");
-		sprintf(&ret[4], "0x%02X", imm);
-		break;
-	case 0b01111: // VMOV.F32
-		strcpy(ret, "F32");
-		sprintf(&ret[4], "%e", (imm & 0x80 ? -1 : 1) * (1 << (((imm >> 4) & 0x7) ^ 0b100)) * (float) (16 + (imm & 0xF)) / 128.);
-		break;
-	case 0b11110: // VMOV.I64
+		
+		const char* arm_print(uint32_t opcode) {
+			static __thread char ret[100];
+			memset(ret, 0, sizeof(ret));
+			
+			"""
+	}
+	files_guard = {
+		"arm_printer.c": """
 		{
-			strcpy(ret, "I64");
-			sprintf(&ret[4], "#0x%s%s%s%s%s%s%s%s",
-				_print_modified_imm_I64_bytes[(imm >> 7) & 1], _print_modified_imm_I64_bytes[(imm >> 6) & 1],
-				_print_modified_imm_I64_bytes[(imm >> 5) & 1], _print_modified_imm_I64_bytes[(imm >> 4) & 1],
-				_print_modified_imm_I64_bytes[(imm >> 3) & 1], _print_modified_imm_I64_bytes[(imm >> 2) & 1],
-				_print_modified_imm_I64_bytes[(imm >> 1) & 1], _print_modified_imm_I64_bytes[(imm >> 0) & 1]);
+				strcpy(ret, "???");
+			}
+			
+			return ret;
 		}
-		break;
-	case 0b11111: // V???.-
-	default:
-		strcpy(ret, "-");
-		strcpy(&ret[4], "UNDEFINED");
-		break;
+		"""
 	}
-	
-	return ret;
-}
-
-const char* arm_print(uint32_t opcode) {
-	static __thread char ret[100];
-	memset(ret, 0, sizeof(ret));
-	
-	"""
-	}
-	files_guard = {"arm_printer.c": """{
-		strcpy(ret, "???");
-	}
-	
-	return ret;
-}
-"""}
+	banner = "/*******************************************************" + ('*'*len(ver)) + "***\n" \
+	         " * File automatically generated by rebuild_printer.py (v" + ver + ") *\n" \
+	         " *******************************************************" + ('*'*len(ver)) + "***/\n"
+	trim = lambda string: '\n'.join(line[2:] for line in string.splitlines())[1:]
+	# Yes, the for loops are inverted. This is because both dicts should have the same keys.
+	for fhdr in files_guard:
+		files_header[fhdr] = banner + trim(files_header[fhdr])
+	for fhdr in files_header:
+		files_guard[fhdr] = trim(files_guard[fhdr])
 	
 	for f in files_header:
-	# Save the string for the next iteration, writing was successful
 		with open(os.path.join(root, "src", "dynarec", f), 'w') as file:
-			file.write(header.format(version = ver))
 			file.write(files_header[f])
 			file.write(output)
 			file.write(files_guard[f])
+	# Save the string for the next iteration, writing was successful
 	with open(os.path.join(root, "src", "dynarec", "last_run.txt"), 'w') as file:
 		file.write('\n'.join(insts))
 	
@@ -1979,6 +1988,6 @@ if __name__ == '__main__':
 	for i, v in enumerate(sys.argv):
 		if v == "--":
 			limit.append(i)
-	if main(sys.argv[1], "1.1.0.04") != 0:
+	if main(sys.argv[1], "1.2.0.05") != 0:
 		exit(2)
 	exit(0)
@@ -13,6 +13,7 @@ if( NOT TEST_REFERENCE )
 endif( NOT TEST_REFERENCE )

 set(ENV{BOX86_LOG} 0)
+set(ENV{BOX86_NOBANNER} 1)
 if( EXISTS ${CMAKE_SOURCE_DIR}/x86lib )
  # we are inside box86 folder
  set(ENV{LD_LIBRARY_PATH} ${CMAKE_SOURCE_DIR}/x86lib)
@@ -31,17 +32,23 @@ execute_process(

 # if the return value is !=0 bail out
 if( TEST_RESULT )
+  get_filename_component(TESTNAME "${TEST_ARGS}" NAME)
+  file(RENAME "${TEST_OUTPUT}" "${CMAKE_BINARY_DIR}/${TESTNAME}.out")
+  file(WRITE  "${CMAKE_BINARY_DIR}/${TESTNAME}.err" ${TEST_ERROR})
  message( FATAL_ERROR "Failed: Test program ${TEST_PROGRAM} exited != 0.\n${TEST_ERROR}" )
 endif( TEST_RESULT )

 # now compare the output with the reference
 execute_process(
-  COMMAND ${CMAKE_COMMAND} -E compare_files ${TEST_OUTPUT} ${TEST_REFERENCE}
+  COMMAND "${CMAKE_COMMAND}" -E compare_files "${TEST_OUTPUT}" "${TEST_REFERENCE}"
  RESULT_VARIABLE TEST_RESULT
  )

 # again, if return value is !=0 scream and shout
 if( TEST_RESULT )
+  get_filename_component(TESTNAME "${TEST_ARGS}" NAME)
+  file(RENAME "${TEST_OUTPUT}" "${CMAKE_BINARY_DIR}/${TESTNAME}.out")
+  file(WRITE  "${CMAKE_BINARY_DIR}/${TESTNAME}.err" ${TEST_ERROR})
  message( FATAL_ERROR "Failed: The output of ${TEST_PROGRAM} did not match ${TEST_REFERENCE}")
 endif( TEST_RESULT )

@@ -19,201 +19,16 @@
 #include "threads.h"
 #include "x86trace.h"
 #include "signals.h"
-#ifdef DYNAREC
 #include <sys/mman.h>
+#include "custommem.h"
+#include "dictionnary.h"
+#ifdef DYNAREC
 #include "dynablock.h"
-
-#define MMAPSIZE (4*1024*1024)      // allocate 4Mo sized blocks
-#define MMAPBLOCK   256             // minimum size of a block
-
-typedef struct mmaplist_s {
-    void*         block;
-    uint8_t       map[MMAPSIZE/(8*MMAPBLOCK)];  // map of allocated sub-block
-} mmaplist_t;
-
-// get first subblock free in map, stating at start. return -1 if no block, else first subblock free, filling size (in subblock unit)
-static int getFirstBlock(mmaplist_t *map, int start, int maxsize, int* size)
-{
-    #define ISFREE(A) (((map->map[(A)>>3]>>((A)&7))&1)?0:1)
-    if(start<0) start = 0;
-    while(start<MMAPSIZE/(8*MMAPBLOCK)) {   // still a chance...
-        if(ISFREE(start)) {
-            // free, now get size...
-            int end = start+1;
-            while(end<MMAPSIZE/(8*MMAPBLOCK)) {
-                if(!ISFREE(end) || (end-start==maxsize)) {
-                    if(size) *size = end-start;
-                    return start;
-                }
-                ++end;
-            }
-            if(size) *size = end-start;
-            return start;
-        }
-        ++start;
-    }
-    return -1;
-}
-
-static void allocBlock(mmaplist_t *map, int start, int size)
-{
-    for(int i=0; i<size; ++i) {
-        map->map[start/8]|=(1<<(start&7));
-        ++start;
-    }
-}
-static void freeBlock(mmaplist_t *map, int start, int size)
-{
-    for(int i=0; i<size; ++i) {
-        map->map[start/8]&=~(1<<(start&7));
-        ++start;
-    }
-}
-
-uintptr_t AllocDynarecMap(int size, int nolinker)
-{
-    if(size>MMAPSIZE) {
-        void* p = mmap(NULL, size, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
-        if(p==MAP_FAILED) {
-            dynarec_log(LOG_DEBUG, "Cannot create dynamic map of %d bytes\n", size);
-            return 0;
-        }
-        return (uintptr_t)p;
-    }
-    pthread_mutex_lock(&my_context->mutex_mmap);
-    int bsize = (size+MMAPBLOCK-1)/MMAPBLOCK;
-    // look for free space
-    for(int i=0; i<my_context->mmapsize; ++i) {
-        int rsize = 0;
-        int start = 0;
-        do {
-            start = getFirstBlock(my_context->mmaplist+i, start, bsize, &rsize);
-            if(start!=-1 && rsize>=bsize) {
-                allocBlock(my_context->mmaplist+i, start, bsize);
-                uintptr_t ret = (uintptr_t)my_context->mmaplist[i].block + start*MMAPBLOCK;
-                pthread_mutex_unlock(&my_context->mutex_mmap);
-                return ret;
-            }
-            if(start!=-1)
-                start += rsize;
-        } while (start!=-1);
-    }
-    // no luck, add a new one !
-    int i = my_context->mmapsize++;    // yeah, useful post incrementation
-    dynarec_log(LOG_DEBUG, "Ask for DynaRec Block Alloc #%d\n", my_context->mmapsize);
-    my_context->mmaplist = (mmaplist_t*)realloc(my_context->mmaplist, my_context->mmapsize*sizeof(mmaplist_t));
-    void* p = mmap(NULL, MMAPSIZE, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
-    if(p==MAP_FAILED) {
-        dynarec_log(LOG_DEBUG, "Cannot create memory map of %d byte for dynarec block #%d\n", MMAPSIZE, i);
-        --my_context->mmapsize;
-        pthread_mutex_unlock(&my_context->mutex_mmap);
-        return 0;
-    }
-    my_context->mmaplist[i].block = p;
-    memset(my_context->mmaplist[i].map, 0, sizeof(my_context->mmaplist[i].map));
-    allocBlock(my_context->mmaplist+i, 0, bsize);
-    pthread_mutex_unlock(&my_context->mutex_mmap);
-    return (uintptr_t)p;
-}
-
-void FreeDynarecMap(uintptr_t addr, uint32_t size)
-{
-    if(size>MMAPSIZE) {
-        munmap((void*)addr, size);
-        return;
-    }
-    pthread_mutex_lock(&my_context->mutex_mmap);
-    int bsize = (size+MMAPBLOCK-1)/MMAPBLOCK;
-    // look for free space
-    for(int i=0; i<my_context->mmapsize; ++i) {
-        if(addr>=(uintptr_t)my_context->mmaplist[i].block && ((uintptr_t)my_context->mmaplist[i].block+MMAPSIZE)>=addr+size) {
-            int start = (addr - (uintptr_t)my_context->mmaplist[i].block) / MMAPBLOCK;
-            freeBlock(my_context->mmaplist+i, start, bsize);
-            pthread_mutex_unlock(&my_context->mutex_mmap);
-            return;
-        }
-    }
-    pthread_mutex_unlock(&my_context->mutex_mmap);
-}
-
-// each dynmap is 64k of size
-dynablocklist_t* getDBFromAddress(uintptr_t addr)
-{
-    int idx = (addr>>16);
-    if(!my_context->dynmap[idx]) {
-        return NULL;
-    }
-    return my_context->dynmap[idx]->dynablocks;
-}
-
-void addDBFromAddressRange(uintptr_t addr, uintptr_t size)
-{
-    dynarec_log(LOG_DEBUG, "addDBFromAddressRange %p -> %p\n", (void*)addr, (void*)(addr+size-1));
-    uintptr_t idx = (addr>>16);
-    uintptr_t end = ((addr+size-1)>>16);
-    for (uintptr_t i=idx; i<=end; ++i) {
-        if(!my_context->dynmap[i]) {
-            my_context->dynmap[i] = (dynmap_t*)calloc(1, sizeof(dynmap_t));
-            my_context->dynmap[i]->dynablocks = NewDynablockList(i<<16, i<<16, 1<<16, 1, 0);
-        }
-    }
-    protectDB(addr, size);
-}
-
-void cleanDBFromAddressRange(uintptr_t addr, uintptr_t size, int destroy)
-{
-    dynarec_log(LOG_DEBUG, "cleanDBFromAddressRange %p -> %p %s\n", (void*)addr, (void*)(addr+size-1), destroy?"destroy":"mark");
-    uintptr_t idx = (addr>box86_dynarec_largest && !destroy)?((addr-box86_dynarec_largest)>>16):(addr>>16);
-    uintptr_t end = ((addr+size-1)>>16);
-    for (uintptr_t i=idx; i<=end; ++i) {
-        dynmap_t* dynmap = my_context->dynmap[i];
-        if(dynmap) {
-            uintptr_t startdb = StartDynablockList(dynmap->dynablocks);
-            uintptr_t enddb = EndDynablockList(dynmap->dynablocks);
-            uintptr_t startaddr = addr;
-            if(startaddr<startdb) startaddr = startdb;
-            uintptr_t endaddr = addr + size - 1;
-            if(endaddr>enddb) endaddr = enddb;
-            if(startaddr==startdb && endaddr==enddb) {
-                if(destroy) {
-                    my_context->dynmap[i] = NULL;
-                    FreeDynablockList(&dynmap->dynablocks);
-                    free(dynmap);
-                } else
-                    MarkDynablockList(&dynmap->dynablocks);
-            } else
-                if(destroy)
-                    FreeRangeDynablock(dynmap->dynablocks, startaddr, endaddr-startaddr+1);
-                else
-                    MarkRangeDynablock(dynmap->dynablocks, startaddr, endaddr-startaddr+1);
-
-        }
-    }
-}
-
-// Remove the Write flag from an adress range, so DB can be executed
-// no log, as it can be executed inside a signal handler
-void protectDB(uintptr_t addr, uintptr_t size)
-{
-    uintptr_t start = (addr)&~(box86_pagesize-1);
-    uintptr_t end = (addr+size+(box86_pagesize-1))&~(box86_pagesize-1);
-    // should get "end" according to last block inside the window
-    mprotect((void*)start, end-start, PROT_READ|PROT_EXEC);
-}
-
-// Add the Write flag from an adress range, and mark all block as dirty
-// no log, as it can be executed inside a signal handler
-void unprotectDB(uintptr_t addr, uintptr_t size)
-{
-    uintptr_t start = (addr)&~(box86_pagesize-1);
-    uintptr_t end = (addr+size+(box86_pagesize-1))&~(box86_pagesize-1);
-    // should get "end" according to last block inside the window
-    mprotect((void*)start, end-start, PROT_READ|PROT_WRITE|PROT_EXEC);
-    cleanDBFromAddressRange(start, end-start, 0);
-}
-
+#include "dynarec/arm_lock_helper.h"
 #endif

+
+
 EXPORTDYN
 void initAllHelpers(box86context_t* context)
 {
@@ -222,18 +37,21 @@ void initAllHelpers(box86context_t* context)
        return;
    my_context = context;
    init_pthread_helper();
-    init_signal_helper();
+    init_bridge_helper();
+    init_signal_helper(context);
    inited = 1;
 }

 EXPORTDYN
-void finiAllHelpers()
+void finiAllHelpers(box86context_t* context)
 {
    static int finied = 0;
    if(finied)
        return;
-    fini_pthread_helper();
+    fini_pthread_helper(context);
    fini_signal_helper();
+    fini_bridge_helper();
+    fini_custommem_helper(context);
    finied = 1;
 }

@@ -244,11 +62,11 @@ int getrand(int maxval)
 {
    if(maxval<1024) {
        return ((random()&0x7fff)*maxval)/0x7fff;
-    } else {
+    } 
        uint64_t r = random();
        r = (r*maxval) / RAND_MAX;
        return r;
-    }
+
 }

 void free_tlsdatasize(void* p)
@@ -260,6 +78,79 @@ void free_tlsdatasize(void* p)
    free(p);
 }

+int unlockMutex()
+{
+    int ret = unlockCustommemMutex();
+    int i;
+    #define GO(A, B)                    \
+        i = checkUnlockMutex(&A);       \
+        if(i) {                         \
+            ret|=(1<<B);                \
+        }
+
+    GO(my_context->mutex_once, 5)
+    GO(my_context->mutex_once2, 6)
+    GO(my_context->mutex_trace, 7)
+    #ifdef DYNAREC
+    GO(my_context->mutex_dyndump, 8)
+    #else
+    GO(my_context->mutex_lock, 8)
+    #endif
+    GO(my_context->mutex_tls, 9)
+    GO(my_context->mutex_thread, 10)
+    GO(my_context->mutex_bridge, 11)
+    #undef GO
+
+    return ret;
+}
+
+void relockMutex(int locks)
+{
+    #define GO(A, B)                    \
+        if(locks&(1<<B))                \
+            pthread_mutex_lock(&A);     \
+
+    GO(my_context->mutex_once, 5)
+    GO(my_context->mutex_once2, 6)
+    GO(my_context->mutex_trace, 7)
+    #ifdef DYNAREC
+    GO(my_context->mutex_dyndump, 8)
+    #else
+    GO(my_context->mutex_lock, 8)
+    #endif
+    GO(my_context->mutex_tls, 9)
+    GO(my_context->mutex_thread, 10)
+    GO(my_context->mutex_bridge, 11)
+    #undef GO
+    relockCustommemMutex(locks);
+}
+
+static void init_mutexes(box86context_t* context)
+{
+    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_mutexattr_destroy(&attr);
+}
+
+static void atfork_child_box64context(void)
+{
+    // (re)init mutex if it was lock before the fork
+    init_mutexes(my_context);
+}
+
 EXPORTDYN
 box86context_t *NewBox86Context(int argc)
 {
@@ -270,18 +161,24 @@ box86context_t *NewBox86Context(int argc)
    }
 #endif
    // init and put default values
-    box86context_t *context = (box86context_t*)calloc(1, sizeof(box86context_t));
+    box86context_t *context = my_context = (box86context_t*)calloc(1, sizeof(box86context_t));

 #ifdef BUILD_LIB
    context->deferedInit = 0;
 #else
    context->deferedInit = 1;
 #endif
+    context->sel_serial = 1;
+
+    init_custommem_helper(context);
+
    context->maplib = NewLibrarian(context, 1);
    context->local_maplib = NewLibrarian(context, 1);
+    context->versym = NewDictionnary();
    context->system = NewBridge();
    // create vsyscall
-    context->vsyscall = AddBridge(context->system, vFv, x86Syscall, 0);
+    context->vsyscall = AddBridge(context->system, iFEv, x86Syscall, 0, NULL);
+    addAlternate((void*)0xffffe400, (void*)context->vsyscall);
 #ifdef BUILD_LIB
    context->box86lib = RTLD_DEFAULT;   // not ideal
 #else
@@ -289,28 +186,18 @@ box86context_t *NewBox86Context(int argc)
 #endif
    context->dlprivate = NewDLPrivate();

-    context->callbacks = NewCallbackList();
-
    context->argc = argc;
    context->argv = (char**)calloc(context->argc+1, sizeof(char*));

-    pthread_mutex_init(&context->mutex_once, NULL);
-    pthread_mutex_init(&context->mutex_once2, NULL);
-    pthread_mutex_init(&context->mutex_trace, NULL);
-    pthread_mutex_init(&context->mutex_lock, NULL);
-    pthread_mutex_init(&context->mutex_thread, NULL);
+    init_mutexes(context);
+    pthread_atfork(NULL, NULL, atfork_child_box64context);

    pthread_key_create(&context->tlskey, free_tlsdatasize);

-#ifdef DYNAREC
-    pthread_mutex_init(&context->mutex_blocks, NULL);
-    pthread_mutex_init(&context->mutex_mmap, NULL);
-    context->dynablocks = NewDynablockList(0, 0, 0, 0, 0);
-#endif
    InitFTSMap(context);

    for (int i=0; i<4; ++i) context->canary[i] = 1 +  getrand(255);
-    context->canary[getrand(4)] = 0;
+    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);

    initAllHelpers(context);
@@ -333,6 +220,12 @@ void FreeBox86Context(box86context_t** context)

    box86context_t* ctx = *context;   // local copy to do the cleanning

+    if(ctx->local_maplib)
+        FreeLibrarian(&ctx->local_maplib, NULL);
+    if(ctx->maplib)
+        FreeLibrarian(&ctx->maplib, NULL);
+    FreeDictionnary(&ctx->versym);
+
    for(int i=0; i<ctx->elfsize; ++i) {
        FreeElfHeader(&ctx->elfs[i]);
    }
@@ -340,29 +233,6 @@ void FreeBox86Context(box86context_t** context)

    FreeFTSMap(ctx);

-    if(ctx->maplib)
-        FreeLibrarian(&ctx->maplib);
-    if(ctx->local_maplib)
-        FreeLibrarian(&ctx->local_maplib);
-
-#ifdef DYNAREC
-    dynarec_log(LOG_DEBUG, "Free global Dynarecblocks\n");
-    if(ctx->dynablocks)
-        FreeDynablockList(&ctx->dynablocks);
-    for (int i=0; i<ctx->mmapsize; ++i)
-        if(ctx->mmaplist[i].block)
-            munmap(ctx->mmaplist[i].block, MMAPSIZE);
-    free(ctx->mmaplist);
-    pthread_mutex_destroy(&ctx->mutex_blocks);
-    pthread_mutex_destroy(&ctx->mutex_mmap);
-    dynarec_log(LOG_DEBUG, "Free dynamic Dynarecblocks\n");
-    cleanDBFromAddressRange(0, 0xffffffff, 1);
-#endif
-    
-    *context = NULL;                // bye bye my_context
-
-    CleanStackSize(ctx);
-
    FreeCollection(&ctx->box86_path);
    FreeCollection(&ctx->box86_ld_lib);
    FreeCollection(&ctx->box86_emulated_libs);
@@ -375,6 +245,27 @@ void FreeBox86Context(box86context_t** context)
    if(ctx->deferedInitList)
        free(ctx->deferedInitList);

+    free(ctx->argv);
+    
+    for (int i=0; i<ctx->envc; ++i)
+        free(ctx->envv[i]);
+    free(ctx->envv);
+
+    if(ctx->atfork_sz) {
+        free(ctx->atforks);
+        ctx->atforks = NULL;
+        ctx->atfork_sz = ctx->atfork_cap = 0;
+    }
+
+    for(int i=0; i<MAX_SIGNAL; ++i)
+        if(ctx->signals[i]!=0 && ctx->signals[i]!=1) {
+            signal(i, SIG_DFL);
+        }
+
+    *context = NULL;                // bye bye my_context
+
+    CleanStackSize(ctx);
+
 #ifndef BUILD_LIB
    if(ctx->box86lib)
        dlclose(ctx->box86lib);
@@ -382,13 +273,7 @@ void FreeBox86Context(box86context_t** context)

    FreeDLPrivate(&ctx->dlprivate);

-    free(ctx->argv);
-    
-    for (int i=0; i<ctx->envc; ++i)
-        free(ctx->envv[i]);
-    free(ctx->envv);
-
-    free(ctx->stack);
+//    free(ctx->stack); // don't free the stack, it's owned by main emu!

    free(ctx->fullpath);
    free(ctx->box86path);
@@ -398,8 +283,6 @@ void FreeBox86Context(box86context_t** context)
    freeGLProcWrapper(ctx);
    freeALProcWrapper(ctx);

-    FreeCallbackList(&ctx->callbacks);
-
    void* ptr;
    if ((ptr = pthread_getspecific(ctx->tlskey)) != NULL) {
        free_tlsdatasize(ptr);
@@ -417,29 +300,24 @@ void FreeBox86Context(box86context_t** context)
        }
    }

-    pthread_mutex_destroy(&ctx->mutex_once);
-    pthread_mutex_destroy(&ctx->mutex_once2);
-    pthread_mutex_destroy(&ctx->mutex_trace);
-    pthread_mutex_destroy(&ctx->mutex_lock);
-    pthread_mutex_destroy(&ctx->mutex_thread);
-
    free_neededlib(&ctx->neededlibs);

-    if(ctx->atfork_sz) {
-        free(ctx->atforks);
-        ctx->atforks = NULL;
-        ctx->atfork_sz = ctx->atfork_cap = 0;
-    }
-
-    for(int i=0; i<MAX_SIGNAL; ++i)
-        if(ctx->signals[i]!=0 && ctx->signals[i]!=1) {
-            signal(i, SIG_DFL);
-        }
-
    if(ctx->emu_sig)
        FreeX86Emu(&ctx->emu_sig);

-    finiAllHelpers();
+    finiAllHelpers(ctx);
+
+    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);

    free(ctx);
 }
@@ -463,27 +341,12 @@ int AddTLSPartition(box86context_t* context, int tlssize) {
    context->tlsdata = realloc(context->tlsdata, context->tlssize);
    memmove(context->tlsdata+tlssize, context->tlsdata, oldsize);   // move to the top, using memmove as regions will probably overlap
    memset(context->tlsdata, 0, tlssize);           // fill new space with 0 (not mandatory)
+    // clean GS segment for current emu
+    if(my_context) {
+        ResetSegmentsCache(thread_get_emu());
+        if(!(++context->sel_serial))
+            ++context->sel_serial;
+    }
+
    return -context->tlssize;   // negative offset
 }
-
-void add_neededlib(needed_libs_t* needed, library_t* lib)
-{
-    if(!needed)
-        return;
-    if(needed->size == needed->cap) {
-        needed->cap += 8;
-        needed->libs = (library_t**)realloc(needed->libs, needed->cap*sizeof(library_t*));
-    }
-    needed->libs[needed->size++] = lib;
-}
-
-void free_neededlib(needed_libs_t* needed)
-{
-    if(!needed)
-        return;
-    needed->cap = 0;
-    needed->size = 0;
-    if(needed->libs)
-        free(needed->libs);
-    needed->libs = NULL;
-}
@@ -2,7 +2,7 @@
 #define __BOX86_VERSION_H_

 #define BOX86_MAJOR 0
-#define BOX86_MINOR 1
-#define BOX86_REVISION 2
+#define BOX86_MINOR 2
+#define BOX86_REVISION 4

 #endif //__BOX86_VERSION_H_
@@ -1,11 +1,15 @@
 #include <stdio.h>
 #include "debug.h"
 #include "box86version.h"
+#ifdef NOGIT
+#define GITREV "nogit"
+#else
 #include "git_head.h"
+#endif

 void PrintBox86Version()
 {
-    printf("Box86%s%s%s v%d.%d.%d %s built on %s %s\n", 
+    printf("Box86%s%s v%d.%d.%d %s built on %s %s\n", 
    #ifdef HAVE_TRACE
        " with trace",
    #else
@@ -15,11 +19,6 @@ void PrintBox86Version()
        " with Dynarec",
    #else
        "",
-    #endif
-    #ifdef USE_FLOAT
-        " (float only)",
-    #else
-        "",
    #endif
        BOX86_MAJOR, BOX86_MINOR, BOX86_REVISION,
        GITREV,
@@ -0,0 +1,824 @@
+#define _GNU_SOURCE         /* See feature_test_macros(7) */
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <dlfcn.h>
+#include <signal.h>
+
+#include "box86context.h"
+#include "elfloader.h"
+#include "debug.h"
+#include "x86trace.h"
+#include "x86emu.h"
+#include "librarian.h"
+#include "bridge.h"
+#include "library.h"
+#include "callback.h"
+#include "wrapper.h"
+#include "myfts.h"
+#include "threads.h"
+#include "x86trace.h"
+#include "signals.h"
+#include <sys/mman.h>
+#include "custommem.h"
+#include "threads.h"
+#ifdef DYNAREC
+#include "dynablock.h"
+#include "dynarec/arm_lock_helper.h"
+#include "khash.h"
+
+#define USE_MMAP
+
+// init inside dynablocks.c
+KHASH_MAP_INIT_INT(dynablocks, dynablock_t*)
+static dynablocklist_t*    dynmap[DYNAMAP_SIZE];     // 4G of memory mapped by 4K block
+static mmaplist_t          *mmaplist = NULL;
+static int                 mmapsize = 0;
+static int                 mmapcap = 0;
+static kh_dynablocks_t     *dblist_oversized;      // store the list of oversized dynablocks (normal sized are inside mmaplist)
+static uintptr_t           *box86_jumptable[JMPTABL_SIZE];
+static uintptr_t           box86_jmptbl_default[1<<JMPTABL_SHIFT];
+#endif
+#define MEMPROT_SHIFT 12
+#define MEMPROT_SIZE (1<<(32-MEMPROT_SHIFT))
+static uint8_t             memprot[MEMPROT_SIZE] = {0};    // protection flags by 4K block
+static int inited = 0;
+
+typedef struct blocklist_s {
+    void*               block;
+    int                 maxfree;
+    size_t              size;
+} blocklist_t;
+
+#define MMAPSIZE (256*1024)      // allocate 256kb sized blocks
+
+static pthread_mutex_t     mutex_blocks;
+static int                 n_blocks = 0;       // number of blocks for custom malloc
+static blocklist_t*        p_blocks = NULL;    // actual blocks for custom malloc
+
+typedef union mark_s {
+    struct {
+        unsigned int    size:31;
+        unsigned int    fill:1;
+    };
+    uint32_t            x32;
+} mark_t;
+typedef struct blockmark_s {
+    mark_t  prev;
+    mark_t  next;
+} blockmark_t;
+
+#define NEXT_BLOCK(b) (blockmark_t*)((uintptr_t)(b) + (b)->next.size + sizeof(blockmark_t))
+#define PREV_BLOCK(b) (blockmark_t*)(((uintptr_t)(b) - (b)->prev.size) - sizeof(blockmark_t))
+#define LAST_BLOCK(b, s) (blockmark_t*)(((uintptr_t)(b)+(s))-sizeof(blockmark_t))
+
+static void printBlock(blockmark_t* b, void* start)
+{
+    printf_log(LOG_INFO, "========== Block is:\n");
+    do {
+        printf_log(LOG_INFO, "%c%p, fill=%d, size=0x%x (prev=%d/0x%x)\n", b==start?'*':' ', b, b->next.fill, b->next.size, b->prev.fill, b->prev.size);
+        b = NEXT_BLOCK(b);
+    } while(b->next.x32);
+    printf_log(LOG_INFO, "===================\n");
+}
+
+// get first subblock free in block. Return NULL if no block, else first subblock free (mark included), filling size
+static void* getFirstBlock(void* block, size_t maxsize, size_t* size, void* start)
+{
+    // get start of block
+    blockmark_t *m = (blockmark_t*)((start)?start:block);
+    while(m->next.x32) {    // while there is a subblock
+        if(!m->next.fill && m->next.size>=maxsize) {
+            *size = m->next.size;
+            return m;
+        }
+        m = NEXT_BLOCK(m);
+    }
+
+    return NULL;
+}
+
+static size_t getMaxFreeBlock(void* block, size_t block_size, void* start)
+{
+    // get start of block
+    if(start) {
+        blockmark_t *m = (blockmark_t*)start;
+        int maxsize = 0;
+        while(m->next.x32) {    // while there is a subblock
+            if(!m->next.fill && m->next.size>maxsize) {
+                maxsize = m->next.size;
+            }
+            m = NEXT_BLOCK(m);
+        }
+        return (maxsize>=sizeof(blockmark_t))?maxsize:0;
+    } else {
+        blockmark_t *m = LAST_BLOCK(block, block_size); // start with the end
+        int maxsize = 0;
+        while(m->prev.x32) {    // while there is a subblock
+            if(!m->prev.fill && m->prev.size>maxsize) {
+                maxsize = m->prev.size;
+                if((uintptr_t)block+maxsize>(uintptr_t)m)
+                    return (maxsize>=sizeof(blockmark_t))?maxsize:0; // no block large enough left...
+            }
+            m = PREV_BLOCK(m);
+        }
+        return (maxsize>=sizeof(blockmark_t))?maxsize:0;
+    }
+}
+
+static void* allocBlock(void* block, void *sub, size_t size, void** pstart)
+{
+    (void)block;
+
+    blockmark_t *s = (blockmark_t*)sub;
+    blockmark_t *n = NEXT_BLOCK(s);
+
+    s->next.fill = 1;
+    // check if a new mark is worth it
+    if(s->next.size>size+2*sizeof(blockmark_t))
+        s->next.size = size;
+    blockmark_t *m = NEXT_BLOCK(s);   // this is new n
+    m->prev.fill = 1;
+    if(n!=m) {
+        // new mark
+        m->prev.size = s->next.size;
+        m->next.fill = 0;
+        m->next.size = ((uintptr_t)n - (uintptr_t)m) - sizeof(blockmark_t);
+        n->prev.fill = 0;
+        n->prev.size = m->next.size;
+    }
+
+    if(pstart && sub==*pstart) {
+        // get the next free block
+        m = (blockmark_t*)*pstart;
+        while(m->next.fill)
+            m = NEXT_BLOCK(m);
+        *pstart = (void*)m;
+    }
+    return (void*)((uintptr_t)sub + sizeof(blockmark_t));
+}
+static size_t freeBlock(void *block, void* sub, void** pstart)
+{
+    blockmark_t *m = (blockmark_t*)block;
+    blockmark_t *s = (blockmark_t*)sub;
+    blockmark_t *n = NEXT_BLOCK(s);
+    if(block!=sub)
+        m = PREV_BLOCK(s);
+    s->next.fill = 0;
+    n->prev.fill = 0;
+    // check if merge with previous
+    if (s->prev.x32 && !s->prev.fill) {
+        // remove s...
+        m->next.size += s->next.size + sizeof(blockmark_t);
+        n->prev.size = m->next.size;
+        s = m;
+    }
+    // check if merge with next
+    if(n->next.x32 && !n->next.fill) {
+        blockmark_t *n2 = NEXT_BLOCK(n);
+        //remove n
+        s->next.size += n->next.size + sizeof(blockmark_t);
+        n2->prev.size = s->next.size;
+    }
+    if(pstart && (uintptr_t)*pstart>(uintptr_t)sub) {
+        *pstart = (void*)s;
+    }
+    // return free size at current block (might be bigger)
+    return s->next.size;
+}
+// return 1 if block has been expanded to new size, 0 if not
+static int expandBlock(void* block, void* sub, size_t newsize)
+{
+    (void)block;
+
+    newsize = (newsize+3)&~3;
+    blockmark_t *s = (blockmark_t*)sub;
+    blockmark_t *n = NEXT_BLOCK(s);
+    if(s->next.size>=newsize)
+        // big enough, no shrinking...
+        return 1;
+    if(s->next.fill)
+        return 0;   // next block is filled
+    // unsigned bitfield of this length gets "promoted" to *signed* int...
+    if((size_t)(s->next.size + n->next.size + sizeof(blockmark_t)) < newsize)
+        return 0;   // free space too short
+    // ok, doing the alloc!
+    if((s->next.size+n->next.size+sizeof(blockmark_t))-newsize<2*sizeof(blockmark_t))
+        s->next.size += n->next.size+sizeof(blockmark_t);
+    else
+        s->next.size = newsize+sizeof(blockmark_t);
+    blockmark_t *m = NEXT_BLOCK(s);   // this is new n
+    m->prev.fill = 1;
+    m->prev.size = s->next.size;
+    if(n!=m) {
+        // new mark
+        m->prev.fill = 1;
+        m->prev.size = s->next.size;
+        m->next.fill = 0;
+        m->next.size = (uintptr_t)n - (uintptr_t)m;
+        n->prev.fill = 0;
+        n->prev.size = m->next.size;
+    }
+    return 1;
+}
+// return size of block
+static size_t sizeBlock(void* sub)
+{
+    blockmark_t *s = (blockmark_t*)sub;
+    return s->next.size;
+}
+
+void* customMalloc(size_t size)
+{
+    // look for free space
+    void* sub = NULL;
+    pthread_mutex_lock(&mutex_blocks);
+    for(int i=0; i<n_blocks; ++i) {
+        if(p_blocks[i].maxfree>=size) {
+            size_t rsize = 0;
+            sub = getFirstBlock(p_blocks[i].block, size, &rsize, NULL);
+            if(sub) {
+                void* ret = allocBlock(p_blocks[i].block, sub, size, NULL);
+                if(rsize==p_blocks[i].maxfree)
+                    p_blocks[i].maxfree = getMaxFreeBlock(p_blocks[i].block, p_blocks[i].size, NULL);
+                pthread_mutex_unlock(&mutex_blocks);
+                return ret;
+            }
+        }
+    }
+    // add a new block
+    int i = n_blocks++;
+    p_blocks = (blocklist_t*)realloc(p_blocks, n_blocks*sizeof(blocklist_t));
+    size_t allocsize = MMAPSIZE;
+    if(size+2*sizeof(blockmark_t)>allocsize)
+        allocsize = size+2*sizeof(blockmark_t);
+    #ifdef USE_MMAP
+    void* p = mmap(NULL, allocsize, PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
+    memset(p, 0, allocsize);
+    #else
+    void* p = calloc(1, allocsize);
+    #endif
+    p_blocks[i].block = p;
+    p_blocks[i].size = allocsize;
+    // setup marks
+    blockmark_t* m = (blockmark_t*)p;
+    m->prev.x32 = 0;
+    m->next.fill = 0;
+    m->next.size = allocsize-2*sizeof(blockmark_t);
+    blockmark_t* n = NEXT_BLOCK(m);
+    n->next.x32 = 0;
+    n->prev.fill = 0;
+    n->prev.size = m->next.size;
+    // alloc 1st block
+    void* ret  = allocBlock(p_blocks[i].block, p, size, NULL);
+    p_blocks[i].maxfree = getMaxFreeBlock(p_blocks[i].block, p_blocks[i].size, NULL);
+    pthread_mutex_unlock(&mutex_blocks);
+    return ret;
+}
+void* customCalloc(size_t n, size_t size)
+{
+    size_t newsize = n*size;
+    void* ret = customMalloc(newsize);
+    memset(ret, 0, newsize);
+    return ret;
+}
+void* customRealloc(void* p, size_t size)
+{
+    if(!p)
+        return customMalloc(size);
+    uintptr_t addr = (uintptr_t)p;
+    pthread_mutex_lock(&mutex_blocks);
+    for(int i=0; i<n_blocks; ++i) {
+        if ((addr>(uintptr_t)p_blocks[i].block) 
+         && (addr<((uintptr_t)p_blocks[i].block+p_blocks[i].size))) {
+            void* sub = (void*)(addr-sizeof(blockmark_t));
+            if(expandBlock(p_blocks[i].block, sub, size)) {
+                p_blocks[i].maxfree = getMaxFreeBlock(p_blocks[i].block, p_blocks[i].size, NULL);
+                pthread_mutex_unlock(&mutex_blocks);
+                return p;
+            }
+            pthread_mutex_unlock(&mutex_blocks);
+            void* newp = customMalloc(size);
+            memcpy(newp, p, sizeBlock(sub));
+            customFree(p);
+            return newp;
+        }
+    }
+    pthread_mutex_unlock(&mutex_blocks);
+    if(n_blocks)
+        dynarec_log(LOG_NONE, "Warning, block %p not found in p_blocks for realloc, malloc'ing again without free\n", (void*)addr);
+    return customMalloc(size);
+}
+void customFree(void* p)
+{
+    if(!p)
+        return;
+    uintptr_t addr = (uintptr_t)p;
+    pthread_mutex_lock(&mutex_blocks);
+    for(int i=0; i<n_blocks; ++i) {
+        if ((addr>(uintptr_t)p_blocks[i].block) 
+         && (addr<((uintptr_t)p_blocks[i].block+p_blocks[i].size))) {
+            void* sub = (void*)(addr-sizeof(blockmark_t));
+            size_t newfree = freeBlock(p_blocks[i].block, sub, NULL);
+            if(p_blocks[i].maxfree < newfree) p_blocks[i].maxfree = newfree;
+            pthread_mutex_unlock(&mutex_blocks);
+            return;
+        }
+    }
+    pthread_mutex_unlock(&mutex_blocks);
+    if(n_blocks)
+        dynarec_log(LOG_NONE, "Warning, block %p not found in p_blocks for Free\n", (void*)addr);
+}
+
+#ifdef DYNAREC
+typedef struct mmaplist_s {
+    void*               block;
+    int                 maxfree;
+    size_t              size;
+    kh_dynablocks_t*    dblist;
+    uint8_t*            helper;
+    void*               first;  // first free block, to speed up things
+    int                 locked; // don't try to add stuff on locked block
+} mmaplist_t;
+
+uintptr_t FindFreeDynarecMap(dynablock_t* db, int size)
+{
+    // look for free space
+    void* sub = NULL;
+    for(int i=0; i<mmapsize; ++i) {
+        if(mmaplist[i].maxfree>=size+sizeof(blockmark_t) && !mmaplist[i].locked) {
+            mmaplist[i].locked = 1;
+            int rsize = 0;
+            sub = getFirstBlock(mmaplist[i].block, size, &rsize, mmaplist[i].first);
+            if(sub) {
+                uintptr_t ret = (uintptr_t)allocBlock(mmaplist[i].block, sub, size, &mmaplist[i].first);
+                if(rsize==mmaplist[i].maxfree) {
+                    mmaplist[i].maxfree = getMaxFreeBlock(mmaplist[i].block, mmaplist[i].size, mmaplist[i].first);
+                }
+                kh_dynablocks_t *blocks = mmaplist[i].dblist;
+                if(!blocks) {
+                    blocks = mmaplist[i].dblist = kh_init(dynablocks);
+                    kh_resize(dynablocks, blocks, 64);
+                }
+                khint_t k;
+                int r;
+                k = kh_put(dynablocks, blocks, (uintptr_t)ret, &r);
+                kh_value(blocks, k) = db;
+                int size255=(size<256)?size:255;
+                for(size_t j=0; j<size255; ++j)
+                    mmaplist[i].helper[(uintptr_t)ret-(uintptr_t)mmaplist[i].block+j] = j;
+                if(size!=size255)
+                    memset(&mmaplist[i].helper[(uintptr_t)ret-(uintptr_t)mmaplist[i].block+256], -1, size-255);
+                mmaplist[i].locked = 0;
+                return ret;
+            } else {
+                printf_log(LOG_INFO, "BOX86: Warning, sub not found, corrupted mmaplist[%d] info?\n", i);
+                if(box86_log >= LOG_DEBUG)
+                    printBlock(mmaplist[i].block, mmaplist[i].first);
+            }
+        }
+    }
+    return 0;
+}
+
+uintptr_t AddNewDynarecMap(dynablock_t* db, int size)
+{
+    int i = mmapsize++;    // yeah, useful post incrementation
+    dynarec_log(LOG_DEBUG, "Ask for DynaRec Block Alloc #%zu/%zu\n", mmapsize, mmapcap);
+    if(mmapsize>mmapcap) {
+        mmapcap += 32;
+        mmaplist = (mmaplist_t*)realloc(mmaplist, mmapcap*sizeof(mmaplist_t));
+    }
+    #ifndef USE_MMAP
+    void *p = NULL;
+    if(posix_memalign(&p, box86_pagesize, MMAPSIZE)) {
+        dynarec_log(LOG_INFO, "Cannot create memory map of %d byte for dynarec block #%d\n", MMAPSIZE, i);
+        --mmapsize;
+        return 0;
+    }
+    mprotect(p, MMAPSIZE, PROT_READ | PROT_WRITE | PROT_EXEC);
+    #else
+    void* p = mmap(NULL, MMAPSIZE, PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
+    if(p==(void*)-1) {
+        dynarec_log(LOG_INFO, "Cannot create memory map of %d byte for dynarec block #%d\n", MMAPSIZE, i);
+        --mmapsize;
+        return 0;
+    }
+    #endif
+    setProtection((uintptr_t)p, MMAPSIZE, PROT_READ | PROT_WRITE | PROT_EXEC);
+
+    mmaplist[i].locked = 1;
+    mmaplist[i].block = p;
+    mmaplist[i].size = MMAPSIZE;
+    mmaplist[i].helper = (uint8_t*)calloc(1, MMAPSIZE);
+    mmaplist[i].first = p;
+    // setup marks
+    blockmark_t* m = (blockmark_t*)p;
+    m->prev.x32 = 0;
+    m->next.fill = 0;
+    m->next.size = MMAPSIZE-2*sizeof(blockmark_t);
+    blockmark_t* n = NEXT_BLOCK(m);
+    n->next.x32 = 0;
+    n->prev.fill = 0;
+    n->prev.size = m->next.size;
+    // alloc 1st block
+    uintptr_t sub  = (uintptr_t)allocBlock(mmaplist[i].block, p, size, &mmaplist[i].first);
+    mmaplist[i].maxfree = getMaxFreeBlock(mmaplist[i].block, mmaplist[i].size, mmaplist[i].first);
+    kh_dynablocks_t *blocks = mmaplist[i].dblist = kh_init(dynablocks);
+    kh_resize(dynablocks, blocks, 64);
+    khint_t k;
+    int ret;
+    k = kh_put(dynablocks, blocks, (uintptr_t)sub, &ret);
+    kh_value(blocks, k) = db;
+    for(int j=0; j<size; ++j)
+        mmaplist[i].helper[(uintptr_t)sub-(uintptr_t)mmaplist[i].block + j] = (j<256)?j:255;
+    mmaplist[i].locked = 0;
+    return sub;
+}
+
+void ActuallyFreeDynarecMap(dynablock_t* db, uintptr_t addr, int size)
+{
+    if(!addr || !size)
+        return;
+    for(int i=0; i<mmapsize; ++i) {
+        if ((addr>(uintptr_t)mmaplist[i].block) 
+         && (addr<((uintptr_t)mmaplist[i].block+mmaplist[i].size))) {
+            void* sub = (void*)(addr-sizeof(blockmark_t));
+            size_t newfree = freeBlock(mmaplist[i].block, sub, &mmaplist[i].first);
+            if(mmaplist[i].maxfree < newfree) mmaplist[i].maxfree = newfree;
+            kh_dynablocks_t *blocks = mmaplist[i].dblist;
+            if(blocks) {
+                khint_t k = kh_get(dynablocks, blocks, (uintptr_t)sub);
+                if(k!=kh_end(blocks))
+                    kh_del(dynablocks, blocks, k);
+                memset(&mmaplist[i].helper[(uintptr_t)sub-(uintptr_t)mmaplist[i].block], 0, size);
+            }
+            if(mmaplist[i].locked) {
+                printf_log(LOG_INFO, "BOX86: Warning, Free a chunk in a locked mmaplist[%d]\n", i);
+                ++mmaplist[i].locked;
+            }
+            return;
+        }
+    }
+    if(mmapsize)
+        dynarec_log(LOG_NONE, "Warning, block %p (size %d) not found in mmaplist for Free\n", (void*)addr, size);
+}
+
+dynablock_t* FindDynablockFromNativeAddress(void* addr)
+{
+    // look in actual list
+    for(int i=0; i<mmapsize; ++i) {
+        if ((uintptr_t)addr>=(uintptr_t)mmaplist[i].block 
+        && ((uintptr_t)addr<(uintptr_t)mmaplist[i].block+mmaplist[i].size)) {
+            if(!mmaplist[i].helper)
+                return FindDynablockDynablocklist(addr, mmaplist[i].dblist);
+            else {
+                uintptr_t p = (uintptr_t)addr - (uintptr_t)mmaplist[i].block;
+                while(mmaplist[i].helper[p]) p -= mmaplist[i].helper[p];
+                khint_t k = kh_get(dynablocks, mmaplist[i].dblist, (uintptr_t)mmaplist[i].block + p);
+                if(k!=kh_end(mmaplist[i].dblist))
+                    return kh_value(mmaplist[i].dblist, k);
+                return NULL;
+            }
+        }
+    }
+    // look in oversized
+    return FindDynablockDynablocklist(addr, dblist_oversized);
+}
+
+uintptr_t AllocDynarecMap(dynablock_t* db, int size)
+{
+    if(!size)
+        return 0;
+    if(size>MMAPSIZE-2*sizeof(blockmark_t)) {
+        #ifndef USE_MMAP
+        void *p = NULL;
+        if(posix_memalign(&p, box86_pagesize, size)) {
+            dynarec_log(LOG_INFO, "Cannot create dynamic map of %d bytes\n", size);
+            return 0;
+        }
+        mprotect(p, size, PROT_READ | PROT_WRITE | PROT_EXEC);
+        #else
+        void* p = mmap(NULL, size, PROT_READ|PROT_WRITE|PROT_EXEC, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
+        if(p==(void*)-1) {
+            dynarec_log(LOG_INFO, "Cannot create dynamic map of %d bytes\n", size);
+            return 0;
+        }
+        #endif
+        setProtection((uintptr_t)p, size, PROT_READ | PROT_WRITE | PROT_EXEC);
+        kh_dynablocks_t *blocks = dblist_oversized;
+        if(!blocks) {
+            blocks = dblist_oversized = kh_init(dynablocks);
+            kh_resize(dynablocks, blocks, 64);
+        }
+        khint_t k;
+        int ret;
+        k = kh_put(dynablocks, blocks, (uintptr_t)p, &ret);
+        kh_value(blocks, k) = db;
+        return (uintptr_t)p;
+    }
+    
+    uintptr_t ret = FindFreeDynarecMap(db, size);
+    if(!ret)
+        ret = AddNewDynarecMap(db, size);
+
+    return ret;
+}
+
+void FreeDynarecMap(dynablock_t* db, uintptr_t addr, uint32_t size)
+{
+    if(!addr || !size)
+        return;
+    if(size>MMAPSIZE-2*sizeof(blockmark_t)) {
+        #ifndef USE_MMAP
+        free((void*)addr);
+        #else
+        munmap((void*)addr, size);
+        #endif
+        kh_dynablocks_t *blocks = dblist_oversized;
+        if(blocks) {
+            khint_t k = kh_get(dynablocks, blocks, addr);
+            if(k!=kh_end(blocks))
+                kh_del(dynablocks, blocks, k);
+        }
+        return;
+    }
+    ActuallyFreeDynarecMap(db, addr, size);
+}
+
+dynablocklist_t* getDB(uintptr_t idx)
+{
+    return dynmap[idx];
+}
+
+// each dynmap is 64k of size
+
+void addDBFromAddressRange(uintptr_t addr, uintptr_t size)
+{
+    dynarec_log(LOG_DEBUG, "addDBFromAddressRange %p -> %p\n", (void*)addr, (void*)(addr+size-1));
+    uintptr_t idx = (addr>>DYNAMAP_SHIFT);
+    uintptr_t end = ((addr+size-1)>>DYNAMAP_SHIFT);
+    for (uintptr_t i=idx; i<=end; ++i) {
+        if(!dynmap[i]) {
+            dynablocklist_t* p = NewDynablockList(i<<DYNAMAP_SHIFT, 1<<DYNAMAP_SHIFT, 0);
+            if(arm_lock_storeifnull(&dynmap[i], p)!=p)
+                FreeDynablockList(&p);
+        }
+    }
+}
+
+void cleanDBFromAddressRange(uintptr_t addr, uintptr_t size, int destroy)
+{
+    dynarec_log(LOG_DEBUG, "cleanDBFromAddressRange %p -> %p %s\n", (void*)addr, (void*)(addr+size-1), destroy?"destroy":"mark");
+    uintptr_t idx = (addr>>DYNAMAP_SHIFT);
+    uintptr_t end = ((addr+size-1)>>DYNAMAP_SHIFT);
+    for (uintptr_t i=idx; i<=end; ++i) {
+        dynablocklist_t* dblist = dynmap[i];
+        if(dblist) {
+            if(destroy)
+                FreeRangeDynablock(dblist, addr, size);
+            else
+                MarkRangeDynablock(dblist, addr, size);
+        }
+    }
+}
+
+#ifdef ARM
+void arm_next(void);
+#endif
+
+void addJumpTableIfDefault(void* addr, void* jmp)
+{
+    const uintptr_t idx = ((uintptr_t)addr>>JMPTABL_SHIFT);
+    if(box86_jumptable[idx] == box86_jmptbl_default) {
+        uintptr_t* tbl = (uintptr_t*)malloc((1<<JMPTABL_SHIFT)*sizeof(uintptr_t));
+        for(int i=0; i<(1<<JMPTABL_SHIFT); ++i)
+            tbl[i] = (uintptr_t)arm_next;
+        if(arm_lock_storeifref(&box86_jumptable[idx], tbl, box86_jmptbl_default)!=tbl)
+            free(tbl);
+    }
+    const uintptr_t off = (uintptr_t)addr&((1<<JMPTABL_SHIFT)-1);
+    arm_lock_storeifref(&box86_jumptable[idx][off], jmp, arm_next);
+}
+void setJumpTableDefault(void* addr)
+{
+    const uintptr_t idx = ((uintptr_t)addr>>JMPTABL_SHIFT);
+    if(box86_jumptable[idx] == box86_jmptbl_default) {
+        return;
+    }
+    const uintptr_t off = (uintptr_t)addr&((1<<JMPTABL_SHIFT)-1);
+    box86_jumptable[idx][off] = (uintptr_t)arm_next;
+}
+int isJumpTableDefault(void* addr)
+{
+    const uintptr_t idx = ((uintptr_t)addr>>JMPTABL_SHIFT);
+    if(box86_jumptable[idx] == box86_jmptbl_default) {
+        return 1;
+    }
+    const uintptr_t off = (uintptr_t)addr&((1<<JMPTABL_SHIFT)-1);
+    return (box86_jumptable[idx][off]==(uintptr_t)arm_next)?1:0;
+}
+uintptr_t getJumpTable()
+{
+    return (uintptr_t)box86_jumptable;
+}
+
+uintptr_t getJumpTableAddress(uintptr_t addr)
+{
+    const uintptr_t idx = ((uintptr_t)addr>>JMPTABL_SHIFT);
+    if(box86_jumptable[idx] == box86_jmptbl_default) {
+        uintptr_t* tbl = (uintptr_t*)malloc((1<<JMPTABL_SHIFT)*sizeof(uintptr_t));
+        for(int i=0; i<(1<<JMPTABL_SHIFT); ++i)
+            tbl[i] = (uintptr_t)arm_next;
+        if(arm_lock_storeifref(&box86_jumptable[idx], tbl, box86_jmptbl_default)!=tbl)
+            free(tbl);
+    }
+    const uintptr_t off = (uintptr_t)addr&((1<<JMPTABL_SHIFT)-1);
+    return (uintptr_t)&box86_jumptable[idx][off];
+}
+
+// Remove the Write flag from an adress range, so DB can be executed
+// no log, as it can be executed inside a signal handler
+void protectDB(uintptr_t addr, uintptr_t size)
+{
+    dynarec_log(LOG_DEBUG, "protectDB %p -> %p\n", (void*)addr, (void*)(addr+size-1));
+    uintptr_t idx = (addr>>MEMPROT_SHIFT);
+    uintptr_t end = ((addr+size-1)>>MEMPROT_SHIFT);
+    for (uintptr_t i=idx; i<=end; ++i) {
+        uint32_t prot = memprot[i];
+        if(!(prot&PROT_DYNAREC)) {
+            if(!prot)
+                prot = PROT_READ | PROT_WRITE;    // comes from malloc & co, so should not be able to execute
+            memprot[i] = prot|PROT_DYNAREC;
+            mprotect((void*)(i<<MEMPROT_SHIFT), 1<<MEMPROT_SHIFT, prot&~PROT_WRITE);
+        }
+    }
+}
+
+// Add the Write flag from an adress range, and mark all block as dirty
+// no log, as it can be executed inside a signal handler
+void unprotectDB(uintptr_t addr, uintptr_t size)
+{
+    dynarec_log(LOG_DEBUG, "unprotectDB %p -> %p\n", (void*)addr, (void*)(addr+size-1));
+    uintptr_t idx = (addr>>MEMPROT_SHIFT);
+    uintptr_t end = ((addr+size-1)>>MEMPROT_SHIFT);
+    for (uintptr_t i=idx; i<=end; ++i) {
+        uint32_t prot = memprot[i];
+        if(prot&PROT_DYNAREC) {
+            memprot[i] = prot&~PROT_DYNAREC;
+            mprotect((void*)(i<<MEMPROT_SHIFT), 1<<MEMPROT_SHIFT, prot&~PROT_DYNAREC);
+            cleanDBFromAddressRange((i<<MEMPROT_SHIFT), 1<<MEMPROT_SHIFT, 0);
+        }
+    }
+}
+
+#endif
+
+void updateProtection(uintptr_t addr, uintptr_t size, uint32_t prot)
+{
+    const uintptr_t idx = (addr>>MEMPROT_SHIFT);
+    const uintptr_t end = ((addr+size-1)>>MEMPROT_SHIFT);
+    for (uintptr_t i=idx; i<=end; ++i) {
+        uint32_t dyn=(memprot[i]&PROT_DYNAREC);
+        if(dyn && (prot&PROT_WRITE))    // need to remove the write protection from this block
+            mprotect((void*)(i<<MEMPROT_SHIFT), 1<<MEMPROT_SHIFT, prot&~PROT_WRITE);
+        memprot[i] = prot|dyn;
+    }
+}
+
+void forceProtection(uintptr_t addr, uintptr_t size, uint32_t prot)
+{
+    const uintptr_t idx = (addr>>MEMPROT_SHIFT);
+    const uintptr_t end = ((addr+size-1)>>MEMPROT_SHIFT);
+    for (uintptr_t i=idx; i<=end; ++i) {
+        mprotect((void*)(i<<MEMPROT_SHIFT), 1<<MEMPROT_SHIFT, prot&~PROT_DYNAREC);
+        memprot[i] = prot;
+    }
+}
+
+void setProtection(uintptr_t addr, uintptr_t size, uint32_t prot)
+{
+    const uintptr_t idx = (addr>>MEMPROT_SHIFT);
+    const uintptr_t end = ((addr+size-1)>>MEMPROT_SHIFT);
+    for (uintptr_t i=idx; i<=end; ++i) {
+        memprot[i] = prot;
+    }
+}
+
+uint32_t getProtection(uintptr_t addr)
+{
+    const uintptr_t idx = (addr>>MEMPROT_SHIFT);
+    uint32_t ret = memprot[idx];
+    return ret;
+}
+
+int unlockCustommemMutex()
+{
+    int ret = 0;
+    int i = 0;
+    #define GO(A, B)                    \
+        i = checkUnlockMutex(&A);       \
+        if(i) {                         \
+            ret|=(1<<B);                \
+        }
+    GO(mutex_blocks, 0)
+    #undef GO
+    return ret;
+}
+
+void relockCustommemMutex(int locks)
+{
+    #define GO(A, B)                    \
+        if(locks&(1<<B))                \
+            pthread_mutex_lock(&A);     \
+
+    GO(mutex_blocks, 0)
+    #undef GO
+}
+
+static void init_mutexes(void)
+{
+    pthread_mutexattr_t attr;
+    pthread_mutexattr_init(&attr);
+    pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_ERRORCHECK);
+    pthread_mutex_init(&mutex_blocks, &attr);
+
+    pthread_mutexattr_destroy(&attr);
+}
+
+static void atfork_child_custommem(void)
+{
+    // (re))init mutex if it was lock before the fork
+    init_mutexes();
+}
+
+void init_custommem_helper(box86context_t* ctx)
+{
+    if(inited) // already initialized
+        return;
+    inited = 1;
+    init_mutexes();
+#ifdef DYNAREC
+#ifdef ARM
+    for(int i=0; i<(1<<JMPTABL_SHIFT); ++i)
+        box86_jmptbl_default[i] = (uintptr_t)arm_next;
+    for(int i=0; i<JMPTABL_SIZE; ++i)
+        box86_jumptable[i] = box86_jmptbl_default;
+#else
+#error Unsupported architecture!
+#endif
+#endif
+    pthread_atfork(NULL, NULL, atfork_child_custommem);
+}
+
+void fini_custommem_helper(box86context_t *ctx)
+{
+    if(!inited)
+        return;
+    inited = 0;
+#ifdef DYNAREC
+    if(box86_dynarec) {
+        dynarec_log(LOG_DEBUG, "Free global Dynarecblocks\n");
+        for (int i=0; i<mmapsize; ++i) {
+            if(mmaplist[i].block)
+                #ifdef USE_MMAP
+                munmap(mmaplist[i].block, mmaplist[i].size);
+                #else
+                free(mmaplist[i].block);
+                #endif
+            if(mmaplist[i].dblist) {
+                kh_destroy(dynablocks, mmaplist[i].dblist);
+                mmaplist[i].dblist = NULL;
+            }
+            if(mmaplist[i].helper) {
+                free(mmaplist[i].helper);
+                mmaplist[i].helper = NULL;
+            }
+        }
+        if(dblist_oversized) {
+            kh_destroy(dynablocks, dblist_oversized);
+            dblist_oversized = NULL;
+        }
+        mmapsize = 0;
+        mmapcap = 0;
+        dynarec_log(LOG_DEBUG, "Free dynamic Dynarecblocks\n");
+        uintptr_t idx = 0;
+        uintptr_t end = ((0xFFFFFFFF)>>DYNAMAP_SHIFT);
+        for (uintptr_t i=idx; i<=end; ++i)
+            if(dynmap[i])
+                FreeDynablockList(&dynmap[i]);
+        free(mmaplist);
+        for (int i=0; i<DYNAMAP_SIZE; ++i)
+            if(box86_jumptable[i]!=box86_jmptbl_default)
+                free(box86_jumptable[i]);
+    }
+#endif
+    for(int i=0; i<n_blocks; ++i)
+        #ifdef USE_MMAP
+        munmap(p_blocks[i].block, p_blocks[i].size);
+        #else
+        free(p_blocks[i].block);
+        #endif
+    free(p_blocks);
+    pthread_mutex_destroy(&mutex_blocks);
+}
@@ -21,12 +21,13 @@ Op is 20-27
 #define xEBP    9
 #define xESI    10
 #define xEDI    11
-#define xEIP    12
+#define xFlags  12
+#define xEIP    14
 // scratch registers
 #define x1      1
 #define x2      2
 #define x3      3
-#define x12     12
+#define x14     14
 // emu is r0
 #define xEmu    0
 // ARM SP is r13
@@ -77,13 +78,17 @@ Op is 20-27
 // movt dst, #imm16
 #define MOVT(dst, imm16) EMIT(0xe3400000 | ((dst) << 12) | (((imm16) & 0xf000) << 4) | brIMM((imm16) & 0x0fff) )
 // pseudo insruction: mov reg, #imm with imm a 32bits value
-#define MOV32(dst, imm32)                   \
-    MOVW(dst, ((uint32_t)imm32)&0xffff);    \
-    if (((uint32_t)(imm32))>>16) {            \
-        MOVT(dst, (((uint32_t)imm32)>>16)); }
+#define MOV32(dst, imm32)                       \
+    if((~(uint32_t)(imm32))<0x100) {            \
+        MVN_IMM8(dst, ~(uint32_t)(imm32), 0);   \
+    } else {                                    \
+        MOVW(dst, ((uint32_t)imm32)&0xffff);    \
+        if (((uint32_t)(imm32))>>16) {          \
+        MOVT(dst, (((uint32_t)imm32)>>16));}    \
+    }
 // pseudo insruction: mov reg, #imm with imm a 32bits value, fixed size
-#define MOV32_(dst, imm32)                   \
-    MOVW(dst, ((uint32_t)(imm32))&0xffff);    \
+#define MOV32_(dst, imm32)                  \
+    MOVW(dst, ((uint32_t)(imm32))&0xffff);  \
    MOVT(dst, (((uint32_t)(imm32))>>16))
 // movw.cond dst, #imm16
 #define MOVW_COND(cond, dst, imm16) EMIT(cond | 0x03000000 | ((dst) << 12) | (((imm16) & 0xf000) << 4) | brIMM((imm16) & 0x0fff) )
@@ -96,10 +101,14 @@ Op is 20-27
 #define MOV_REG_LSL_IMM5(dst, src, imm5) EMIT(0xe1a00000 | ((dst) << 12) | (src) | (0<<4) | (0<<5) | ((imm5)<<7))
 // mov dst, src lsr imm5
 #define MOV_REG_LSR_IMM5(dst, src, imm5) EMIT(0xe1a00000 | ((dst) << 12) | (src) | (0<<4) | (1<<5) | ((imm5)<<7))
+// mov.cond dst, src lsr imm5
+#define MOV_REG_LSR_IMM5_COND(cond, dst, src, imm5) EMIT(cond | 0x01a00000 | ((dst) << 12) | (src) | (0<<4) | (1<<5) | ((imm5)<<7))
 // mov dst, src asr imm5
 #define MOV_REG_ASR_IMM5(dst, src, imm5) EMIT(0xe1a00000 | ((dst) << 12) | (src) | (0<<4) | (2<<5) | ((imm5)<<7))
 // mov dst, src ror imm5
 #define MOV_REG_ROR_IMM5(dst, src, imm5) EMIT(0xe1a00000 | ((dst) << 12) | (src) | (0<<4) | (3<<5) | ((imm5)<<7))
+// mov dst, src rrx (1)
+#define MOV_REG_RRX(dst, src) EMIT(0xe1a00000 | ((dst) << 12) | (src) | (0<<4) | (3<<5) | ((0)<<7))
 // mov.s dst, src lsl imm5
 #define MOVS_REG_LSL_IMM5(dst, src, imm5) EMIT(0xe1b00000 | ((dst) << 12) | (src) | (0<<4) | (0<<5) | ((imm5)<<7))
 // mov.s dst, src lsr imm5
@@ -108,9 +117,17 @@ Op is 20-27
 #define MOVS_REG_ASR_IMM5(dst, src, imm5) EMIT(0xe1b00000 | ((dst) << 12) | (src) | (0<<4) | (2<<5) | ((imm5)<<7))
 // mov.s dst, src ror imm5
 #define MOVS_REG_ROR_IMM5(dst, src, imm5) EMIT(0xe1b00000 | ((dst) << 12) | (src) | (0<<4) | (3<<5) | ((imm5)<<7))
+// mov.s dst, src rrx (1)
+#define MOVS_REG_RRX(dst, src) EMIT(0xe1b00000 | ((dst) << 12) | (src) | (0<<4) | (3<<5) | ((0)<<7))

+// mov.s dst, src lsl rs
+#define MOVS_REG_LSL_REG(dst, src, rs) EMIT(0xe1b00000 | ((dst) << 12) | (src) | (1<<4) | (0<<5) | ((rs)<<8))
+// mov.s dst, src lsr rs
+#define MOVS_REG_LSR_REG(dst, src, rs) EMIT(0xe1b00000 | ((dst) << 12) | (src) | (1<<4) | (1<<5) | ((rs)<<8))
 // mov dst, src lsl rs
 #define MOV_REG_LSL_REG(dst, src, rs) EMIT(0xe1a00000 | ((dst) << 12) | (src) | (1<<4) | (0<<5) | ((rs)<<8))
+// mov.cond dst, src lsl rs
+#define MOV_REG_LSL_REG_COND(cond, dst, src, rs) EMIT(cond | 0x01a00000 | ((dst) << 12) | (src) | (1<<4) | (0<<5) | ((rs)<<8))
 // mov dst, src lsr rs
 #define MOV_REG_LSR_REG(dst, src, rs) EMIT(0xe1a00000 | ((dst) << 12) | (src) | (1<<4) | (1<<5) | ((rs)<<8))
 // mov dst, src asr rs
@@ -122,7 +139,7 @@ Op is 20-27
 #define SUB_IMM8(dst, src, imm8) \
    EMIT(0xe2400000 | ((dst) << 12) | ((src) << 16) | brIMM(imm8) )
 // sub cond dst, src, #(imm8)
-#define SUB_COND_IMM8(dst, src, imm8) \
+#define SUB_COND_IMM8(cond, dst, src, imm8) \
    EMIT((cond) | 0x02400000 | ((dst) << 12) | ((src) << 16) | brIMM(imm8) )
 // sub.s dst, src, #(imm8)
 #define SUBS_IMM8(dst, src, imm8) \
@@ -156,6 +173,9 @@ Op is 20-27
 // rsb cond dst, src, #(imm8)
 #define RSB_COND_IMM8(cond, dst, src, imm8) \
    EMIT((cond) | 0x02600000 | ((dst) << 12) | ((src) << 16) | brIMM(imm8) )
+// rsc dst, src, #(imm8)
+#define RSC_IMM8(dst, src, imm8) \
+    EMIT(0xe2E00000 | ((dst) << 12) | ((src) << 16) | brIMM(imm8) )

 // and dst, src1, src2, lsl #imm
 #define AND_REG_LSL_IMM5(dst, src1, src2, imm5) \
@@ -169,15 +189,21 @@ Op is 20-27
 // and dst, src, #(imm8)
 #define AND_IMM8(dst, src, imm8) \
    EMIT(0xe2000000 | ((dst) << 12) | ((src) << 16) | brIMM(imm8) )
+// 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.s dst, src, #(imm8)
 #define ANDS_IMM8(dst, src, imm8) \
    EMIT(0xe2100000 | ((dst) << 12) | ((src) << 16) | brIMM(imm8) )
 // and.s dst, src1, #imm ror rot*2
-#define ANDS_IMM8_ROR(src, imm8, rot) \
+#define ANDS_IMM8_ROR(dst, src, imm8, rot) \
    EMIT(0xe2100000 | ((dst) << 12) | ((src) << 16) | ((rot)<<8) | brIMM(imm8) )
 // add dst, src, #(imm8)
 #define ADD_IMM8(dst, src, imm8) \
    EMIT(0xe2800000 | ((dst) << 12) | ((src) << 16) | brIMM(imm8) )
+// add dst, src, #imm8 ror rot*2
+#define ADD_IMM8_ROR(dst, src, imm8, rot) \
+    EMIT(0xe2800000 | ((dst) << 12) | ((src) << 16) | ((rot)<<8) | brIMM((imm8)) )
 // add.s dst, src, #(imm8)
 #define ADDS_IMM8(dst, src, imm8) \
    EMIT(0xe2900000 | ((dst) << 12) | ((src) << 16) | brIMM(imm8) )
@@ -207,9 +233,18 @@ Op is 20-27
 // cmp.s dst, src1, src2, lsl #imm
 #define CMPS_REG_LSL_IMM5(src1, src2, imm5) \
    EMIT(0xe1500000 | ((0) << 12) | ((src1) << 16) | brLSL(imm5, src2) )
+// cmp.cond.s dst, src1, src2, lsl #imm
+#define CMPS_REG_LSL_IMM5_COND(cond, src1, src2, imm5) \
+    EMIT((cond) | 0x01500000 | ((0) << 12) | ((src1) << 16) | brLSL(imm5, src2) )
+// cmp.s dst, src, #imm
+#define CMPS_IMM8_COND(cond, src, imm8) \
+    EMIT((cond) | 0x03500000 | ((0) << 12) | ((src) << 16) | brIMM(imm8) )
 // cmp.s dst, src, #imm
 #define CMPS_IMM8(src, imm8) \
    EMIT(0xe3500000 | ((0) << 12) | ((src) << 16) | brIMM(imm8) )
+// cmn.s dst, src, #imm
+#define CMNS_IMM8_COND(cond, src, imm8) \
+    EMIT((cond) | 0x03700000 | ((0) << 12) | ((src) << 16) | brIMM(imm8) )
 // tst.s dst, src1, src2, lsl #imm
 #define TSTS_REG_LSL_IMM5(src1, src2, imm5) \
    EMIT(0xe1100000 | ((0) << 12) | ((src1) << 16) | brLSL(imm5, src2) )
@@ -225,13 +260,25 @@ Op is 20-27
 // orr.s dst, src1, src2, lsl #imm
 #define ORRS_REG_LSL_IMM5(dst, src1, src2, imm5) \
    EMIT(0xe1900000 | ((dst) << 12) | ((src1) << 16) | brLSL(imm5, src2) )
+// orr dst, src1, src2, lsr #imm
+#define ORR_REG_LSR_IMM5(dst, src1, src2, imm5) \
+    EMIT(0xe1800000 | ((dst) << 12) | ((src1) << 16) | brLSR(imm5, src2) )
+// orr.s dst, src1, src2, lsr #imm
+#define ORRS_REG_LSR_IMM5(dst, src1, src2, imm5) \
+    EMIT(0xe1900000 | ((dst) << 12) | ((src1) << 16) | brLSR(imm5, src2) )
+// orr.s dst, src1, src2, lsl rs
+#define ORRS_REG_LSL_REG(dst, src1, src2, rs) \
+    EMIT(0xe1900000 | ((dst) << 12) | ((src1) << 16) | brRLSL(rs, src2) )
 // orr dst, src1, #imm8
 #define ORR_IMM8(dst, src, imm8, rot) \
    EMIT(0xe3800000 | ((dst) << 12) | ((src) << 16) | ((rot)<<8) | imm8 )
 // orr.s dst, src1, #imm8
 #define ORRS_IMM8(dst, src, imm8, rot) \
    EMIT(0xe3900000 | ((dst) << 12) | ((src) << 16) | ((rot)<<8) | imm8 )
-// orr dst, src1, src2, lsl #rs
+// orr.cond dst, src1, #imm8 ror rot*2
+#define ORR_IMM8_COND(cond, dst, src, imm8, rot) \
+    EMIT((cond) | 0x03800000 | ((dst) << 12) | ((src) << 16) | ((rot)<<8) | imm8 )
+// orr dst, src1, src2, lsl rs
 #define ORR_REG_LSL_REG(dst, src1, src2, rs) \
    EMIT(0xe1800000 | ((dst) << 12) | ((src1) << 16) | brRLSL(rs, src2) )
 // xor dst, src1, src2, lsl #imm
@@ -247,6 +294,9 @@ Op is 20-27
    EMIT(0xe2200000 | ((dst) << 12) | ((src) << 16) | brIMM(imm8) )
 #define XORS_IMM8(dst, src, imm8) \
    EMIT(0xe2300000 | ((dst) << 12) | ((src) << 16) | brIMM(imm8) )
+// xor dst, src, #(imm8 ror rot*2)
+#define XOR_IMM8_ROR(dst, src, imm8, rot) \
+    EMIT(0xe2200000 | ((dst) << 12) | ((src) << 16) | ((rot)<<8) | brIMM(imm8) )
 // xor.cond dst, src, #(imm8)
 #define XOR_IMM8_COND(cond, dst, src, imm8) \
    EMIT(cond | 0x02200000 | ((dst) << 12) | ((src) << 16) | brIMM(imm8) )
@@ -256,12 +306,21 @@ Op is 20-27
 // xor dst, src1, src2, lsl #imm
 #define XOR_REG_LSR_IMM8(dst, src1, src2, imm8) \
    EMIT(0xe0200000 | ((dst) << 12) | ((src1) << 16) | brLSR(imm8, src2) )
+// xor.cond dst, src1, src2, lsl #imm
+#define XOR_REG_LSR_IMM8_COND(cond, dst, src1, src2, imm8) \
+    EMIT(cond | 0x00200000 | ((dst) << 12) | ((src1) << 16) | brLSR(imm8, src2) )
 // bic dst, src1, src2, lsl #imm
 #define BIC_REG_LSL_IMM5(dst, src1, src2, imm5) \
    EMIT(0xe1c00000 | ((dst) << 12) | ((src1) << 16) | brLSL(imm5, src2) )
-// bic dst, src, IMM8
+// bic dst, src, IMM8 ror rot*2
 #define BIC_IMM8(dst, src, imm8, rot) \
    EMIT(0xe3c00000 | ((dst) << 12) | ((src) << 16) | ((rot)<<8) | imm8 )
+// bic.cond dst, src, IMM8
+#define BIC_IMM8_COND(cond, dst, src, imm8, rot) \
+    EMIT((cond) | 0x03c00000 | ((dst) << 12) | ((src) << 16) | ((rot)<<8) | imm8 )
+// bic dst, src1, #imm ror rot*2
+#define BIC_IMM8_ROR(dst, src, imm8, rot) \
+    EMIT(0xe3c00000 | ((dst) << 12) | ((src) << 16) | ((rot)<<8) | brIMM(imm8) )
 // bic.s dst, src1, #imm ror rot*2
 #define BICS_IMM8_ROR(dst, src, imm8, rot) \
    EMIT(0xe3d00000 | ((dst) << 12) | ((src) << 16) | ((rot)<<8) | brIMM(imm8) )
@@ -272,8 +331,8 @@ Op is 20-27
 #define MVN_REG_LSL_IMM5(dst, rm, imm5) \
    EMIT(0xe1e00000 | ((dst) << 12) | (0 << 16) | brLSL(imm5, rm) )
 // mvn dst, src, IMM8
-#define MVN_IMM8(dst, src, imm8, rot) \
-    EMIT(0xe3e00000 | ((dst) << 12) | ((src) << 16) | ((rot)<<8) | imm8 )
+#define MVN_IMM8(dst, imm8, rot) \
+    EMIT(0xe3e00000 | ((dst) << 12) | (rot)<<8 | ((imm8)&0xff) )
 // mvn dst, src1, src2, lsr #imm
 #define MVN_REG_LSR_IMM8(dst, rm, imm8) \
    EMIT(0xe1e00000 | ((dst) << 12) | (0 << 16) | brLSR(imm8, rm) )
@@ -284,9 +343,6 @@ Op is 20-27
 #define MVN_COND_REG_LSL_IMM5(cond, dst, rm, imm5) \
    EMIT(cond | 0x01e00000 | ((dst) << 12) | (0 << 16) | brLSL(imm5, rm) )

-// Single data transfert construction
-#define SDT_REG(Cond, P, U, B, W, L, Rn, Rd, ShiftRm) (Cond | (0b00<<26) | (1<<25) | (P<<24) | (U<<23) | (B<<22) | (U<<23) | (W<<21) | (L<<20) | (Rn<<16) | (Rd<<12) | ShiftRm)
-#define SDT_OFF(Cond, P, U, B, W, L, Rn, Rd, Imm12)   (Cond | (0b00<<26) | (0<<25) | (P<<24) | (U<<23) | (B<<22) | (U<<23) | (W<<21) | (L<<20) | (Rn<<16) | (Rd<<12) | Imm12)
 // ldr reg, [addr, #+/-imm9]
 #define LDR_IMM9(reg, addr, imm9) EMIT(0xe5100000 | (((imm9)<0)?0:1)<<23 | ((reg) << 12) | ((addr) << 16) | brIMM(imm9) )
 // ldrxx reg, [addr, #+/-imm9]
@@ -316,6 +372,10 @@ Op is 20-27

 // str reg, [addr, #+/-imm9]
 #define STR_IMM9(reg, addr, imm9) EMIT(0xe5000000 | (((imm9)<0)?0:1)<<23 | ((reg) << 12) | ((addr) << 16) | brIMM(imm9) )
+// str with cond reg, [addr, #+/-imm9]
+#define STR_IMM9_COND(cond, reg, addr, imm9) EMIT(cond | 0x05000000 | (((imm9)<0)?0:1)<<23 | ((reg) << 12) | ((addr) << 16) | brIMM(imm9) )
+// str reg, [addr, #+/-imm9]!
+#define STR_IMM9_W(reg, addr, imm9) EMIT(0xe5200000 | (((imm9)<0)?0:1)<<23 | ((reg) << 12) | ((addr) << 16) | brIMM(imm9) )
 // strb reg, [addr, #+/-imm9]
 #define STRB_IMM9(reg, addr, imm9) EMIT(0xe5400000 | (((imm9)<0)?0:1)<<23 | ((reg) << 12) | ((addr) << 16) | brIMM(imm9) )
 // str reg, [addr], #+/-imm9
@@ -353,10 +413,16 @@ Op is 20-27
 //                           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))

+// push reg to xESP
+#define PUSH1(reg)   STR_IMM9_W(reg, xESP, -4)
+
 // pop reg!, {list}
 //                           all |    const    |postindex|  add    | no PSR  |writeback|  load   |   base    |reg list
 #define POP(reg, list)  EMIT(c__ | (0b100<<25) | (0<<24) | (1<<23) | (0<<22) | (1<<21) | (1<<20) | ((reg)<<16) | (list))

+// pop reg from xESP
+#define POP1(reg)   LDRAI_IMM9_W(reg, xESP, 4)
+
 // STMDB reg, {list}
 //                            all |    const    |pre-index| subs    | no PSR  |  no wb  | store   |   base    |reg list
 #define STMDB(reg, list) EMIT(c__ | (0b100<<25) | (1<<24) | (0<<23) | (0<<22) | (0<<21) | (0<<20) | ((reg)<<16) | (list))
@@ -382,8 +448,14 @@ Op is 20-27
 #define LDRH_REG(reg, addr, rm) EMIT(HWS_REG(c__, 1, 1, 0, 1, addr, reg, 0, 1, rm))
 #define STRH_REG(reg, addr, rm) EMIT(HWS_REG(c__, 1, 1, 0, 0, addr, reg, 0, 1, rm))

-#define LDRHAI_REG_LSL_IMM5(reg, addr, rm) EMIT(HWS_REG(c__, 0, 1, 0, 1, addr, reg, 0, 1, rm))
-#define STRHAI_REG_LSL_IMM5(reg, addr, rm) EMIT(HWS_REG(c__, 0, 1, 0, 0, addr, reg, 0, 1, rm))
+#define LDRHAI_REG_LSL_IMM5(reg, addr, rm)  EMIT(HWS_REG(c__, 0, 1, 0, 1, addr, reg, 0, 1, rm))
+#define STRHAI_REG_LSL_IMM5(reg, addr, rm)  EMIT(HWS_REG(c__, 0, 1, 0, 0, addr, reg, 0, 1, rm))
+#define LDRHB_IMM8(reg, addr, imm8)         EMIT(HWS_OFF(c__, 1, (((imm8)<0)?0:1), 1, 1, addr, reg, 0, 1, abs(imm8)))
+#define LDRHA_IMM8(reg, addr, imm8)         EMIT(HWS_OFF(c__, 0, (((imm8)<0)?0:1), 0, 1, addr, reg, 0, 1, abs(imm8)))
+#define STRHB_IMM8(reg, addr, imm8)         EMIT(HWS_OFF(c__, 1, (((imm8)<0)?0:1), 1, 0, addr, reg, 0, 1, abs(imm8)))
+#define STRHA_IMM8(reg, addr, imm8)         EMIT(HWS_OFF(c__, 0, (((imm8)<0)?0:1), 0, 0, addr, reg, 0, 1, abs(imm8)))
+#define PUSH16(reg, addr)                   EMIT(HWS_OFF(c__, 1, 0, 1, 0, addr, reg, 0, 1, 2))
+#define POP16(reg, addr)                    EMIT(HWS_OFF(c__, 0, 1, 0, 1, addr, reg, 0, 1, 2))

 // Mul Long construction
 #define MULLONG(Cond, U, A, S, RdHi, RdLo, Rs, Rm)     (Cond | 0b00001<<23 | (U)<<22 | (A)<<21 | (S)<<20 | (RdHi)<<16 | (RdLo)<<12 | (Rs)<<8 | 0b1001<<4 | (Rm))
@@ -395,6 +467,10 @@ Op is 20-27
 #define MULMULA(Cond, A, S, Rd, Rn, Rs, Rm)     (Cond | 0b000000<<22 | (A)<<21 | (S)<<20 | (Rd)<<16 | (Rn)<<12 | (Rs)<<8 | 0b1001<<4 | (Rm))
 #define MUL(Rd, Rm, Rn)     EMIT(MULMULA(c__, 0, 0, (Rd), 0, (Rm), (Rn)))

+#define MLS_gen(Cond, Rd, Ra, Rm, Rn)     (Cond | 0b00000110<<20 | (Rd)<<16 | (Ra)<<12 | (Rm)<<8 | 0b1001<<4 | (Rn))
+//MLS : Rd = Ra - Rm*Rn
+#define MLS(Rd, Rm, Rn, Ra)         EMIT(MLS_gen(c__, Rd, Ra, Rm, Rn))
+
 #define SMUL_16_gen(cond, Rd, Rm, M, N, Rn) (cond | 0b00010110<<20 | (Rd)<<16 | (Rm)<<8 | 1<<7 | (M)<<6 | (N)<<5 | (Rn))
 // Signed Mul between Rn[0..15] * Rm[0..15] => Rd
 #define SMULBB(Rd, Rn, Rm)  EMIT(SMUL_16_gen(c__, Rd, Rm, 0, 0, Rn))
@@ -410,18 +486,30 @@ Op is 20-27

 // UBFX: Unsigned Bit Field Extract: extract any number of bits from Rn, zero extend and put in Rd
 #define UBFX(rd, rn, lsb, width)    EMIT(c__ | (0b0111111<<21) | (((width)-1)<<16) | ((rd)<<12) | ((lsb)<<7) | (0b101<<4) | (rn))
+// SBFX: Signed Bit Field Extract: extract any number of bits from Rn, zero extend and put in Rd
+#define SBFX(rd, rn, lsb, width)    EMIT(c__ | (0b0111101<<21) | (((width)-1)<<16) | ((rd)<<12) | ((lsb)<<7) | (0b101<<4) | (rn))

 // BFI: Bit Field Insert: copy any number of low order bit from Rn to any position of Rd
 #define BFI(rd, rn, lsb, width) EMIT(c__ | (0b0111110<<21) | (((lsb)+(width)-1)<<16) | ((rd)<<12) | ((lsb)<<7) | (0b001<<4) | (rn))
 // BFI_COND: Bit Field Insert with condition: copy any number of low order bit from Rn to any position of Rd
 #define BFI_COND(cond, rd, rn, lsb, width) EMIT(cond | (0b0111110<<21) | (((lsb)+(width)-1)<<16) | ((rd)<<12) | ((lsb)<<7) | (0b001<<4) | (rn))

+// BFC: Bit Field Clear: clear any number of adjacent bit from Rd
+#define BFC(rd, lsb, width) EMIT(c__ | (0b0111110<<21) | (((lsb)+(width)-1)<<16) | ((rd)<<12) | ((lsb)<<7) | (0b001<<4) | 0b1111)
+// BFC_COND: Bit Field Clear with condition: clear any number of adjacent bit from Rd
+#define BFC_COND(cond, rd, lsb, width) EMIT(cond | (0b0111110<<21) | (((lsb)+(width)-1)<<16) | ((rd)<<12) | ((lsb)<<7) | (0b001<<4) | 0b1111)
+
 // REV: Reverse byte of a 32bits word
-#define REV(rd, rm) EMIT(c__ | (0b01101<<23) | (0<<22) | (0b11<<20) | (0b1111<<16) | ((rd)<<12) | (0b1111<<8) | (0b0011<<4) | (rm))
+#define REV(rd, rm)     EMIT(c__ | (0b01101<<23) | (0<<22) | (0b11<<20) | (0b1111<<16) | ((rd)<<12) | (0b1111<<8) | (0b0011<<4) | (rm))
+// REV16: Reverse byte of a 16bits word
+#define REV16(rd, rm)   EMIT(c__ | (0b01101<<23) | (0<<22) | (0b11<<20) | (0b1111<<16) | ((rd)<<12) | (0b1111<<8) | (0b1011<<4) | (rm))

 #define SSAT_gen(cond, sat_imm, Rd, imm5, sh, Rn) (cond | 0b0110101<<21 | (sat_imm)<<16 | (Rd)<<12 | (imm5)<<7 | (sh)<<6 | 0b01<<4 | (Rn))
 // Signed Staturate Rn to 2^(staturate_to-1) into Rd. Optionnaly shift left Rn before saturate
 #define SSAT_REG_LSL_IMM5(Rd, saturate_to, Rn, shift)   EMIT(SSAT_gen(c__, (saturate_to)-1, Rd, shift, 0, Rn))
+// Signed Staturate Rn to 2^(staturate_to-1) into Rd. Optionnaly shift left Rn before saturate
+#define SSAT_REG_LSL_IMM5_COND(cond, Rd, saturate_to, Rn, shift)   EMIT(SSAT_gen((cond), (saturate_to)-1, Rd, shift, 0, Rn))
+
 #define USAT_gen(cond, sat_imm, Rd, imm5, sh, Rn) (cond | 0b0110111<<21 | (sat_imm)<<16 | (Rd)<<12 | (imm5)<<7 | (sh)<<6 | 0b01<<4 | (Rn))
 // Unsigned Staturate Rn to 2^(staturate_to-1) into Rd. Optionnaly shift left Rn before saturate
 #define USAT_REG_LSL_IMM5(Rd, saturate_to, Rn, shift)   EMIT(SSAT_gen(c__, (saturate_to), Rd, shift, 0, Rn))
@@ -438,11 +526,33 @@ Op is 20-27

 #define LDREXD_gen(cond, Rn, Rt) (cond | 0b000<<25 | 0b11011<<20 | (Rn)<<16 | (Rt)<<12 | 0b1111<<8 | 0b1001<<4 | 0b1111)
 // Load Exclusive Rt/Rt+1 from Rn (tagging the memory)
-#define LDREXD(Rn, Rt)  EMIT(LDREXD_gen(c__, Rn, Rt))
+#define LDREXD(Rt, Rn)  EMIT(LDREXD_gen(c__, Rn, Rt))
+#define LDREXD_COND(cond, Rt, Rn)  EMIT(LDREXD_gen(cond, Rn, Rt))

 #define STREXD_gen(cond, Rd, Rn, Rt)   (cond | 0b000<<25 | 0b11010<<20 | (Rn)<<16 | (Rd)<<12 | 0b1111<<8 | 0b1001<<4 | (Rt))
 // Store Exclusive Rt/Rt+1 to Rn, with result in Rd if tag is ok (Rd!=Rn && Rd!=Rt && Rd!=Rt+1), Rd==1 if store failed
-#define STREXD(Rd, Rn, Rt)  EMIT(STREXD_gen(c__, Rd, Rn, Rt))
+#define STREXD(Rd, Rt, Rn)  EMIT(STREXD_gen(c__, Rd, Rn, Rt))
+#define STREXD_COND(cond, Rd, Rt, Rn)  EMIT(STREXD_gen(cond, Rd, Rn, Rt))
+
+#define LDREX_gen(cond, Rn, Rt)     (cond | 0b0001100<<21 | 1<<20 | (Rn)<<16 | (Rt)<<12 | 0b1111<<8 | 0b1001<<4 | 0b1111)
+// Load Exclusive Rt from Rn (tagging the memory)
+#define LDREX(Rt, Rn)       EMIT(LDREX_gen(c__, Rn, Rt))
+#define LDREX_COND(cond, Rt, Rn)       EMIT(LDREX_gen(cond, Rn, Rt))
+
+#define STREX_gen(cond, Rd, Rn, Rt) (cond | 0b0001100<<21 | 0<<20 | (Rn)<<16 | (Rd)<<12 | 0b1111<<8 | 0b1001<<4 | (Rt))
+// Store Exclusive Rt to Rn, with result in Rd=0 if tag is ok, Rd==1 if store failed (Rd!=Rn && Rd!=Rt)
+#define STREX(Rd, Rt, Rn)   EMIT(STREX_gen(c__, Rd, Rn, Rt))
+#define STREX_COND(cond, Rd, Rt, Rn)   EMIT(STREX_gen(cond, Rd, Rn, Rt))
+
+#define LDREXB_gen(cond, Rn, Rt)        (cond | 0b0001110<<21 | 1<<20 | (Rn)<<16 | (Rt)<<12 | 0b1111<<8 | 0b1001<<4 | 0b1111)
+// Load Exclusive Byte Rt from Rn (tagging the memory)
+#define LDREXB(Rt, Rn)      EMIT(LDREXB_gen(c__, Rn, Rt))
+#define LDREXB_COND(cond, Rt, Rn)      EMIT(LDREXB_gen(cond, Rn, Rt))
+
+#define STREXB_gen(cond, Rd, Rn, Rt)    (cond | 0b0001110<<21 | 0<<20 | (Rn)<<16 | (Rd)<<12 | 0b1111<<8 | 0b1001<<4 | (Rt))
+// Store Exclusive byte Rt to Rn, with result in Rd=0 if tag is ok, Rd==1 if store failed (Rd!=Rn && Rd!=Rt)
+#define STREXB(Rd, Rt, Rn)  EMIT(STREXB_gen(c__, Rd, Rn, Rt))
+#define STREXB_COND(cond, Rd, Rt, Rn)  EMIT(STREXB_gen(cond, Rd, Rn, Rt))

 // 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))
@@ -455,6 +565,10 @@ Op is 20-27
 // Preload Cache Rn-Rm
 #define PLDn(Rn, Rm) EMIT(PLD_gen(0, 1, Rn, 0, 0, Rm))

+#define DMB_gen(opt)    (0b1111<<28 | 0b01010111<<20 | 0b1111<<16 | 0b1111<<12 | 0b0000<<8 | 0b0101<<4 | (opt))
+// Data memory barrier Inner Sharable
+#define DMB_ISH()   EMIT(DMB_gen(0b1011))
+
 #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
 #define SWP(Rt, Rt2, Rn)    EMIT(SWP_gen(c__, 0, Rn, Rt, Rt2))
@@ -463,14 +577,17 @@ Op is 20-27

 #define SDIV_gen(cond, Rd, Rm, Rn)  (cond | 0b0111<<24 | 0b0001<<20 | (Rd)<<16 | 0b1111<<12 | (Rm)<<8 | 0b0001<<4 | (Rn))
 // Signed Div Rd <- Rn/Rm
-#define SDIV(Rd, Rm, Rn)    EMIT(SDIV_gen(c__, Rd, Rm, Rn))
+#define SDIV(Rd, Rn, Rm)    EMIT(SDIV_gen(c__, Rd, Rm, Rn))

 #define UDIV_gen(cond, Rd, Rm, Rn)  (cond | 0b0111<<24 | 0b0011<<20 | (Rd)<<16 | 0b1111<<12 | (Rm)<<8 | 0b0001<<4 | (Rn))
 // Unsigned Div Rd <- Rn/Rm
-#define UDIV(Rd, Rm, Rn)    EMIT(UDIV_gen(c__, Rd, Rm, Rn))
+#define UDIV(Rd, Rn, Rm)    EMIT(UDIV_gen(c__, Rd, Rm, Rn))

-// VFPU
-#define TRANSFERT64(C, op) ((0b1100<<24) | (0b010<<21) | (0b101<<9) | ((C)<<8) | ((op)<<4))
+// Yield
+#define YIELD(cond) EMIT(cond | 0b00110010<<20 | 0b1111<<12 | 1)
+
+// UDF
+#define UDF(imm16)    EMIT(0b1110<<28 | 0b011<<25 | 0b11111<<20 | (((imm16)>>4)&0xfff) | 0b1111<<4 | ((imm16)&0xf))

 // Move from FPSCR to Arm register
 #define VMRS(Rt)    EMIT(c__ | (0b1110<<24) | (0b1111<<20) | (0b0001<<16) | ((Rt)<<12) | (0b1010<<8) | (0b0001<<4) | (0b0000))
@@ -504,6 +621,8 @@ Op is 20-27
 #define VMOVcond_32(cond, Sd, Sm)     EMIT(cond | (0b11101<<23) | (((Sd)&1)<<22) | (0b11<<20) | ((((Sd)>>1)&15)<<12) | (0b101<<9) | (0<<8) | (0b01<<6) | (((Sm)&1)<<5) | (((Sm)>>1)&15))
 // Move between Sd and Imm
 #define VMOV_i_32(Sd, Imm8)     EMIT(c__ | (0b11101<<23) | (((Sd)&1)<<22) | (0b11<<20) | ((((Imm8)>>4)&0xf)<<16) | ((((Sd)>>1)&15)<<12) | (0b101<<9) | (0<<8) | (0b00<<6) | ((Imm8)&15))
+// Move between Dd and Imm
+#define VMOV_i_64(Dd, Imm8)     EMIT(c__ | (0b11101<<23) | ((((Dd)>>4)&1)<<22) | (0b11<<20) | ((((Imm8)>>4)&0xf)<<16) | (((Dd)&15)<<12) | (0b101<<9) | (1<<8) | (0b00<<6) | ((Imm8)&15))

 #define VMOVtoDx_gen(Vd, D, Rt, opc1, opc2)   (0b1110<<24 | 0<<23 | (opc1)<<21 | (Vd)<<16 | (Rt)<<12 | 0b1011<<8 | (D)<<7 | (opc2)<<5 | 1<<4)
 // Move between Rt and Dm[x]
@@ -542,10 +661,14 @@ Op is 20-27

 // Convert from double Dm to int32 Sd, with Round toward Zero mode
 #define VCVT_S32_F64(Sd, Dm)    EMIT(c__ | (0b1110<<24) | (1<<23) | (((Sd)&1)<<22) | (0b111<<19) | (0b101<<16) | ((((Sd)>>1)&15)<<12) | (0b101<<9) | (1<<8) | (1<<7) | (1<<6) | ((((Dm)>>4)&1)<<5) | ((Dm)&15) )
+// Convert from double Dm to uint32 Sd, with Round toward Zero mode
+#define VCVT_U32_F64(Sd, Dm)    EMIT(c__ | (0b1110<<24) | (1<<23) | (((Sd)&1)<<22) | (0b111<<19) | (0b100<<16) | ((((Sd)>>1)&15)<<12) | (0b101<<9) | (1<<8) | (1<<7) | (1<<6) | ((((Dm)>>4)&1)<<5) | ((Dm)&15) )
 // Convert from double Dm to int32 Sd, with Round selection from FPSCR
 #define VCVTR_S32_F64(Sd, Dm)   EMIT(c__ | (0b1110<<24) | (1<<23) | (((Sd)&1)<<22) | (0b111<<19) | (0b101<<16) | ((((Sd)>>1)&15)<<12) | (0b101<<9) | (1<<8) | (0<<7) | (1<<6) | ((((Dm)>>4)&1)<<5) | ((Dm)&15) )
 // Convert from int32 Sm to double Dd
 #define VCVT_F64_S32(Dd, Sm)    EMIT(c__ | (0b1110<<24) | (1<<23) | ((((Dd)>>4)&1)<<22) | (0b111<<19) | (0b000<<16) | (((Dd)&15)<<12) | (0b101<<9) | (1<<8) | (1<<7) | (1<<6) | (((Sm)&1)<<5) | (((Sm)>>1)&15) )
+// Convert from uint32 Sm to double Dd
+#define VCVT_F64_U32(Dd, Sm)    EMIT(c__ | (0b1110<<24) | (1<<23) | ((((Dd)>>4)&1)<<22) | (0b111<<19) | (0b000<<16) | (((Dd)&15)<<12) | (0b101<<9) | (1<<8) | (0<<7) | (1<<6) | (((Sm)&1)<<5) | (((Sm)>>1)&15) )
 // Convert from single Sm to int32 Sd, with Round toward Zero mode
 #define VCVT_S32_F32(Sd, Sm)    EMIT(c__ | (0b1110<<24) | (1<<23) | (((Sd)&1)<<22) | (0b111<<19) | (0b101<<16) | ((((Sd)>>1)&15)<<12) | (0b101<<9) | (0<<8) | (1<<7) | (1<<6) | (((Sm)&1)<<5) | (((Sm)>>1)&15) )
 // Convert from single Sm to int32 Sd, with Round selection from FPSCR
@@ -553,9 +676,9 @@ Op is 20-27
 // Convert from int32 Sm to single Dd
 #define VCVT_F32_S32(Sd, Sm)    EMIT(c__ | (0b1110<<24) | (1<<23) | (((Sd)&1)<<22) | (0b111<<19) | (0b000<<16) | ((((Sd)>>1)&15)<<12) | (0b101<<9) | (0<<8) | (1<<7) | (1<<6) | (((Sm)&1)<<5) | (((Sm)>>1)&15) )

-#define VCVT_16_gen(cond, D, op, U, Vd, sf, sx, i, imm4)   (cond | 0b1110<<24 | 1<<23 | (D)<<22 | 0b11<<20 | 1<<19 | (op)<<18 | 1<<17 | (U)<<16 | (Vd)<<12 | 0b101<<9 | (sf)<<8 | (sx)<<7 | 1<<6 | (i)<<5 | (imm4))
+#define VCVT_FP_gen(cond, D, op, U, Vd, sf, sx, i, imm4)   (cond | 0b1110<<24 | 1<<23 | (D)<<22 | 0b11<<20 | 1<<19 | (op)<<18 | 1<<17 | (U)<<16 | (Vd)<<12 | 0b101<<9 | (sf)<<8 | (sx)<<7 | 1<<6 | (i)<<5 | (imm4))
 // Inplace convert from F64 to S16. Rounding is not selectable
-#define VCVT_S16_F64(Dd)    EMIT(VCVT_16_gen(c__, ((Dd)>>4)&1, 1, 0, (Dd)&15, 1, 0, 16&1, (16>>1)&15))
+#define VCVT_S16_F64(Dd)    EMIT(VCVT_FP_gen(c__, ((Dd)>>4)&1, 1, 0, (Dd)&15, 1, 0, 16&1, (16>>1)&15))

 // Mutiply F64 Dd = Dn*Dm
 #define VMUL_F64(Dd, Dn, Dm)    EMIT(c__ | (0b1110<<24) | (0<<23) | ((((Dd)>>4)&1)<<22) | (0b10<<20) | (((Dn)&15)<<16) | (((Dd)&15)<<12) | (0b101<<9) | (1<<8) | ((((Dn)>>4)&1)<<7) | ((((Dm)>>4)&1)<<5) | ((Dm)&15) )
@@ -592,6 +715,10 @@ Op is 20-27

 // Neg F64 Dd = - Dm
 #define VNEG_F64(Dd, Dm)    EMIT(c__ | (0b1110<<24) | (1<<23) | ((((Dd)>>4)&1)<<22) | (0b11<<20) | (0b0001<<16) | (((Dd)&15)<<12) | (0b101<<9) | (1<<8) | (0b01<<6) | ((((Dm)>>4)&1)<<5) | ((Dm)&15) )
+// Neg F64 Dd = - Dm
+#define VNEG_F64_cond(cond, Dd, Dm)    EMIT((cond) | (0b1110<<24) | (1<<23) | ((((Dd)>>4)&1)<<22) | (0b11<<20) | (0b0001<<16) | (((Dd)&15)<<12) | (0b101<<9) | (1<<8) | (0b01<<6) | ((((Dm)>>4)&1)<<5) | ((Dm)&15) )
+// Neg F32 Dd = - Dm
+#define VNEG_F32(Sd, Sm)    EMIT(c__ | (0b1110<<24) | (1<<23) | (((Sd)&1)<<22) | (0b11<<20) | (0b0001<<16) | ((((Sd)>>1)&15)<<12) | (0b101<<9) | (0<<8) | (0b01<<6) | (((Sm)&1)<<5) | (((Sm)>>1)&15) )

 // Sqrt F64 Dd = - Dm
 #define VSQRT_F64(Dd, Dm)    EMIT(c__ | (0b1110<<24) | (1<<23) | ((((Dd)>>4)&1)<<22) | (0b11<<20) | (0b0001<<16) | (((Dd)&15)<<12) | (0b101<<9) | (1<<8) | (0b11<<6) | ((((Dm)>>4)&1)<<5) | ((Dm)&15) )
@@ -600,6 +727,13 @@ Op is 20-27

 // Abs Dd = |Dm|
 #define VABS_F64(Dd, Dm)     EMIT(c__ | (0b11101<<23) | ((((Dd)>>4)&1)<<22) | (0b11<<20) | (((Dd)&15)<<12) | (0b101<<9) | (1<<8) | (0b11<<6) | ((((Dm)>>4)&1)<<5) | ((Dm)&15))
+// Abs Dd = |Dm|
+#define VABS_F32(Sd, Sm)     EMIT(c__ | (0b11101<<23) | (((Sd)&1)<<22) | (0b11<<20) | ((((Sd)>>1)&15)<<12) | (0b101<<9) | (0<<8) | (0b11<<6) | (((Sm)&1)<<5) | (((Sm)>>1)&15))
+
+// MLS Dd = Dd - Dn*Dm
+#define VMLS_F64(Dd, Dn, Dm)    EMIT(c__ | (0b1110<<24) | (0<<23) | ((((Dd)>>4)&1)<<22) | (0b00<<20) | (((Dn)&15)<<16) | (((Dd)&15)<<12) | (0b101<<9) | (1<<8) | ((((Dn)>>4)&1)<<7) | 1<<6 | ((((Dm)>>4)&1)<<5) | ((Dm)&15) )
+// MLA Dd = Dd + Dn*Dm
+#define VMLA_F64(Dd, Dn, Dm)    EMIT(c__ | (0b1110<<24) | (0<<23) | ((((Dd)>>4)&1)<<22) | (0b00<<20) | (((Dn)&15)<<16) | (((Dd)&15)<<12) | (0b101<<9) | (1<<8) | ((((Dn)>>4)&1)<<7) | 0<<6 | ((((Dm)>>4)&1)<<5) | ((Dm)&15) )

 // NEON

@@ -658,16 +792,26 @@ Op is 20-27
 #define VEORQ(Dd, Dn, Dm)   EMIT(VEOR_gen(((Dd)>>4)&1, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 1, ((Dm)>>4)&1, (Dm)&15))

 #define VMOVL_gen(U, D, imm3, Vd, M, Vm) (0b1111<<28 | 0b001<<25 | (U)<<24 | 1<<23 | (D)<<22 | (imm3)<<19 | (Vd)<<12 | 0b1010<<8 | (M)<<5 | 1<<4 | (Vm))
+#define VMOVL_S8(Dd, Dm)    EMIT(VMOVL_gen(0, ((Dd)>>4)&1, 0b001, (Dd)&15, ((Dm)>>4)&1, (Dm)&15))
+#define VMOVL_U8(Dd, Dm)    EMIT(VMOVL_gen(1, ((Dd)>>4)&1, 0b001, (Dd)&15, ((Dm)>>4)&1, (Dm)&15))
 #define VMOVL_S32(Dd, Dm)   EMIT(VMOVL_gen(0, ((Dd)>>4)&1, 0b100, (Dd)&15, ((Dm)>>4)&1, (Dm)&15))
 #define VMOVL_U32(Dd, Dm)   EMIT(VMOVL_gen(1, ((Dd)>>4)&1, 0b100, (Dd)&15, ((Dm)>>4)&1, (Dm)&15))

 #define VMOV_gen(D,Vd, M, Vm, Q) (0b1111<<28 | 0b0010<<24 | 0<<23 | (D)<<22 | 0b10<<20 | (Vm)<<16 | (Vd)<<12 | 0b0001<<8 | (M)<<7 | (Q)<<6 | (M)<<5 | 1<<4 | (Vm))
-#define VMOVD(Dd, Dm) EMIT(VMOV_gen(((Dd)>>4)&1, (Dd)&15, ((Dm)>>4)&1, (Dm)&15, 0))
-#define VMOVQ(Dd, Dm) EMIT(VMOV_gen(((Dd)>>4)&1, (Dd)&15, ((Dm)>>4)&1, (Dm)&15, 1))
+#define VMOVD(Dd, Dm) if((Dd)!=(Dm)) {EMIT(VMOV_gen(((Dd)>>4)&1, (Dd)&15, ((Dm)>>4)&1, (Dm)&15, 0));}
+#define VMOVQ(Dd, Dm) if((Dd)!=(Dm)) {EMIT(VMOV_gen(((Dd)>>4)&1, (Dd)&15, ((Dm)>>4)&1, (Dm)&15, 1));}

 #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 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
+#define VMOV_H16(Dd, imm8)  EMIT(VMOV_igen(((imm8)>>7)&1, ((Dd)>>4)&1, ((imm8)>>4)&7, (Dd)&15, 0b1010, 0, 0, (imm8)&15))
+// Qd <= imm8 in high bits
+#define VMOVQ_H32(Dd, imm8) EMIT(VMOV_igen(((imm8)>>7)&1, ((Dd)>>4)&1, ((imm8)>>4)&7, (Dd)&15, 0b0110, 1, 0, (imm8)&15))
+// Qd <= imm8 in high bits
+#define VMOVQ_H16(Dd, imm8) EMIT(VMOV_igen(((imm8)>>7)&1, ((Dd)>>4)&1, ((imm8)>>4)&7, (Dd)&15, 0b1010, 1, 0, (imm8)&15))

 #define VLD1LANE_gen(D, Rn, Vd, size, index_align, Rm) (0b1111<<28 | 0b0100<<24 | 1<<23 | (D)<<22 | 0b10<<20 | (Rn)<<16 | (Vd)<<12 | (size)<<10 | (index_align)<<4 | (Rm))
 #define VLD1LANE_8(Dd, Rn, index)    EMIT(VLD1LANE_gen(((Dd)>>4)&1, Rn, (Dd)&15, 0, (index)<<1, 15))
@@ -732,6 +876,14 @@ Op is 20-27
 #define VCEQ_F32(Dd, Dn, Dm)    EMIT(VCEQ_F_gen(0, ((Dd)>>4)&1, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 0, ((Dm)>>4)&1, (Dm)&15))
 #define VCEQQ_F32(Dd, Dn, Dm)   EMIT(VCEQ_F_gen(0, ((Dd)>>4)&1, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 1, ((Dm)>>4)&1, (Dm)&15))

+#define VCEQ_0_gen(D, size, Vd, F, Q, M, Vm)    (0b1111<<28 | 0b0011<<24 | 1<<23 | (D)<<22 | 0b11<<20 | (size)<<18 | 0b01<<16 | (Vd)<<12 | (F)<<10 | 0b010<<7 | (Q)<<6 | (M)<<5 | (Vm))
+#define VCEQ_0_8(Dd, Dm)        EMIT(VCEQ_0_gen(((Dd)>>4)&1, 0, (Dd)&15, 0, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VCEQ_0_16(Dd, Dm)       EMIT(VCEQ_0_gen(((Dd)>>4)&1, 1, (Dd)&15, 0, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VCEQ_0_32(Dd, Dm)       EMIT(VCEQ_0_gen(((Dd)>>4)&1, 2, (Dd)&15, 0, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VCEQQ_0_8(Dd, Dm)       EMIT(VCEQ_0_gen(((Dd)>>4)&1, 0, (Dd)&15, 0, 1, ((Dm)>>4)&1, (Dm)&15))
+#define VCEQQ_0_16(Dd, Dm)      EMIT(VCEQ_0_gen(((Dd)>>4)&1, 1, (Dd)&15, 0, 1, ((Dm)>>4)&1, (Dm)&15))
+#define VCEQQ_0_32(Dd, Dm)      EMIT(VCEQ_0_gen(((Dd)>>4)&1, 2, (Dd)&15, 0, 1, ((Dm)>>4)&1, (Dm)&15))
+
 #define VCGT_I_gen(U, D, size, Vn, Vd, N, Q, M, Vm) (0b1111<<28 | 0b001<<25 | (U)<<24 | 0<<23 | (D)<<22 | (size)<<20 | (Vn)<<16 | (Vd)<<12 | 0b0011<<8 | (N)<<7 | (Q)<<6 | (M)<<5 | (Vm))
 #define VCGT_U8(Dd, Dn, Dm)     EMIT(VCGT_I_gen(1, ((Dd)>>4)&1, 0, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 0, ((Dm)>>4)&1, (Dm)&15))
 #define VCGTQ_U8(Dd, Dn, Dm)    EMIT(VCGT_I_gen(1, ((Dd)>>4)&1, 0, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 1, ((Dm)>>4)&1, (Dm)&15))
@@ -757,6 +909,15 @@ Op is 20-27
 #define VCGE_F32(Dd, Dn, Dm)    EMIT(VCGE_F_gen(((Dd)>>4)&1, 0, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 0, ((Dm)>>4)&1, (Dm)&15))
 #define VCGEQ_F32(Dd, Dn, Dm)   EMIT(VCGE_F_gen(((Dd)>>4)&1, 0, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 1, ((Dm)>>4)&1, (Dm)&15))

+#define VCLT_0_gen(D, size, Vd, F, Q, M, Vm)    (0b1111<<28 | 0b0011<<24 | 1<<23 | (D)<<22 | 0b11<<20 | (size)<<18 | 0b01<<16 | (Vd)<<12 | (F)<<10 | 0b100<<7 | (Q)<<6 | (M)<<5 | (Vm))
+#define VCLT_0_8(Dd, Dm)        EMIT(VCLT_0_gen(((Dd)>>4)&1, 0, (Dd)&15, 0, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VCLT_0_16(Dd, Dm)       EMIT(VCLT_0_gen(((Dd)>>4)&1, 1, (Dd)&15, 0, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VCLT_0_32(Dd, Dm)       EMIT(VCLT_0_gen(((Dd)>>4)&1, 2, (Dd)&15, 0, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VCLTQ_0_8(Dd, Dm)       EMIT(VCLT_0_gen(((Dd)>>4)&1, 0, (Dd)&15, 0, 1, ((Dm)>>4)&1, (Dm)&15))
+#define VCLTQ_0_16(Dd, Dm)      EMIT(VCLT_0_gen(((Dd)>>4)&1, 1, (Dd)&15, 0, 1, ((Dm)>>4)&1, (Dm)&15))
+#define VCLTQ_0_32(Dd, Dm)      EMIT(VCLT_0_gen(((Dd)>>4)&1, 2, (Dd)&15, 0, 1, ((Dm)>>4)&1, (Dm)&15))
+
+
 #define VSHR_gen(U, D, imm6, Vd, L, Q, M, Vm) (0b1111<<28 | 0b001<<25 | (U)<<24 | 1<<23 | (D)<<22 | (imm6)<<16 | (Vd)<<12 | (L)<<7 | (Q)<<6 | (M)<<5 | 1<<4 | (Vm))
 #define VSHR_U8(Dd, Dm, imm3)    EMIT(VSHR_gen(1, ((Dd)>>4)&1, 0b001<<3 | (8-(imm3)), (Dd)&15, 0, 0, ((Dm)>>4)&1, (Dm)&15))
 #define VSHR_S8(Dd, Dm, imm3)    EMIT(VSHR_gen(0, ((Dd)>>4)&1, 0b001<<3 | (8-(imm3)), (Dd)&15, 0, 0, ((Dm)>>4)&1, (Dm)&15))
@@ -803,6 +964,11 @@ Op is 20-27
 #define VSHLQ_S32(Dd, Dm, Dn)   EMIT(VSHLR_gen(0, ((Dd)>>4)&1, 0b10, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 1, ((Dm)>>4)&1, (Dm)&15))
 #define VSHLQ_S64(Dd, Dm, Dn)   EMIT(VSHLR_gen(0, ((Dd)>>4)&1, 0b11, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 1, ((Dm)>>4)&1, (Dm)&15))

+#define VSHRN_gen(D, imm6, Vd, M, Vm)   (0b1111<<28 | 0b0010<<24 | 1<<23 | (D)<<22 | (imm6)<<16 | (Vd)<<12 | 1<<11 | (M)<<5 | 1<<4 | (Vm))
+#define VSHRN_16(Dd, Dm, imm3)          EMIT(VSHRN_gen(((Dd)>>4)&1, 0b001<<3 | (8-(imm3)), (Dd)&15, ((Dm)>>4)&1, (Dm)&15))
+#define VSHRN_32(Dd, Dm, imm4)          EMIT(VSHRN_gen(((Dd)>>4)&1, 0b01<<4  | (16-(imm4)), (Dd)&15, ((Dm)>>4)&1, (Dm)&15))
+#define VSHRN_64(Dd, Dm, imm5)          EMIT(VSHRN_gen(((Dd)>>4)&1, 0b1<<5   | (32-(imm5)), (Dd)&15, ((Dm)>>4)&1, (Dm)&15))
+
 #define VTRN_gen(D, size, Vd, Q, M, Vm) (0b1111<<28 | 0b0011<<24 | 1<<23 | (D)<<22 | 0b11<<20 | (size)<<18 | 0b10<<16 | (Vd)<<12 | 0b0000<<8 | 1<<7 | (Q)<<6 | (M)<<5 | (Vm))
 #define VTRN_32(Dd, Dm)     EMIT(VTRN_gen(((Dd)>>4)&1, 2, (Dd)&15, 0, ((Dm)>>4)&1, (Dm)&15))

@@ -855,8 +1021,8 @@ Op is 20-27
 #define VMUL_16(Dd, Dn, Dm)      EMIT(VMUL_NEON_gen(0, ((Dd)>>4)&1, 0b01, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 0, ((Dm)>>4)&1, (Dm)&15))

 #define VEXT_gen(D, Vn, Vd, imm4, N, Q, M, Vm)  (0b1111<<28 | 0b0010<<24 | 1<<23 | (D)<<22 | 0b11<<20 | (Vn)<<16 | (Vd)<<12 | (imm4)<<8 | (N)<<7 | (Q)<<6 | (M)<<5 | (Vm))
-#define VEXT_8(Dd, Dn, Dm, imm4)    EMIT(VEXT_gen(((Dd)>>4)&1, (Dn)&15, (Dd)&15, (imm4), ((Dn)>>4)&1, 0, ((Dm)>>4)&1, (Dm)&15))
-#define VEXTQ_8(Dd, Dn, Dm, imm4)   EMIT(VEXT_gen(((Dd)>>4)&1, (Dn)&15, (Dd)&15, (imm4), ((Dn)>>4)&1, 1, ((Dm)>>4)&1, (Dm)&15))
+#define VEXT_8(Dd, Dn, Dm, imm4)    EMIT(VEXT_gen(((Dd)>>4)&1, (Dn)&15, (Dd)&15, (imm4)&0xf, ((Dn)>>4)&1, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VEXTQ_8(Dd, Dn, Dm, imm4)   EMIT(VEXT_gen(((Dd)>>4)&1, (Dn)&15, (Dd)&15, (imm4)&0xf, ((Dn)>>4)&1, 1, ((Dm)>>4)&1, (Dm)&15))

 #define VMVN_gen(D, size, Vd, Q, M, Vm) (0b1111<<28 | 0b0011<<24 | 1<<23 | (D)<<22 | 0b11<<20 | (size)<<18 | (Vd)<<12 | 0b1011<<7 | (Q)<<6 | (M)<<5 | 0<<4 | (Vm))
 #define VMVNQ(Dd, Dm)       EMIT(VMVN_gen(((Dd)>>4)&1, 0, (Dd)&15, 1, ((Dm)>>4)&1, (Dm)&15))
@@ -1062,7 +1228,7 @@ Op is 20-27

 #define VPADD_Fgen(D, size, Vn, Vd, N, Q, M, Vm)   (0b1111<<28 | 0b0011<<24 | (D)<<22 | (size)<<20 | (Vn)<<16 | (Vd)<<12 | 0b1101<<8 | (N)<<7 | (Q)<<6 | (M)<<5 | 0<<4 | (Vm))
 // Add pair of F32, store result in F32
-#define VPADD_F32(Dd, Dn, Dm)   EMIT(VPADD_gen(((Dd)>>4)&1, 0, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VPADD_F32(Dd, Dn, Dm)   EMIT(VPADD_Fgen(((Dd)>>4)&1, 0, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 0, ((Dm)>>4)&1, (Dm)&15))

 #define VMINMAX_gen(U, D, size, Vn, Vd, N, Q, M, op, Vm)    (0b1111<<28 | 0b001<<25 | (U)<<24 | (D)<<22 | (size)<<20 | (Vn)<<16 | (Vd)<<12 | 0b110<<8 | (N)<<7 | (Q)<<6 | (M)<<5 | (op)<<4 | (Vm))
 #define VMIN_U8(Dd, Dn, Dm)     EMIT(VMINMAX_gen(1, ((Dd)>>4)&1, 0b00, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 0, ((Dm)>>4)&1, 1, (Dm)&15))
@@ -1113,4 +1279,25 @@ Op is 20-27
 #define VRHADDQ_S16(Dd, Dn, Dm) EMIT(VRHADD_gen(0, ((Dd)>>4)&1, 0b01, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 1, ((Dm)>>4)&1, (Dm)&15))
 #define VRHADDQ_S32(Dd, Dn, Dm) EMIT(VRHADD_gen(0, ((Dd)>>4)&1, 0b10, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 1, ((Dm)>>4)&1, (Dm)&15))

+#define VQRDMULH_gen(D, size, Vn, Vd, N, Q, M, Vm)  (0b1111<<28 | 0b0011 <<24 | (D)<<22 | (size)<<20 | (Vn)<<16 | (Vd)<<12 | 0b1011<<8 | (N)<<7 | (Q)<<6 | (M)<<5 | (Vm))
+// Dd <= (2*Dn*Dm + 1<<(size-1))>>size
+#define VQRDMULH_S16(Dd, Dn, Dm)    EMIT(VQRDMULH_gen(((Dd)>>4)&1, 0b01, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VQRDMULH_S32(Dd, Dn, Dm)    EMIT(VQRDMULH_gen(((Dd)>>4)&1, 0b10, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VQRDMULHQ_S16(Dd, Dn, Dm)   EMIT(VQRDMULH_gen(((Dd)>>4)&1, 0b01, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 1, ((Dm)>>4)&1, (Dm)&15))
+#define VQRDMULHQ_S32(Dd, Dn, Dm)   EMIT(VQRDMULH_gen(((Dd)>>4)&1, 0b10, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 1, ((Dm)>>4)&1, (Dm)&15))
+
+#define VDUP_sgen(D, imm4, Vd, Q, M, Vm)    (0b1111<<28 | 0b0011<<24 | 1<<23 | (D)<<22 | 0b11<<20 | (imm4)<<16 | (Vd)<<12 | 0b1100<<8 | (Q)<<6 | (M)<<5 | (Vm))
+// Dd <= Dup(Dm[x])
+#define VDUP_32(Dd, Dm, x)      EMIT(VDUP_sgen(((Dd)>>4)&1, (((x)<<3)|0b100), (Dd)&15, 0, ((Dm)>>4)&1, (Dm)&15))
+
+#define VABS_vgen(D, size, Vd, F, Q, M, Vm) (0b1111<<28 | 0b0011<<24 | 1<<23 | (D)<<22 | 0b11<<20 | (size)<<18 | 0b01<<16 | (Vd)<<12 | (F)<<10 | 0b110<<7 | (Q)<<6 | (M)<<5 | (Vm))
+#define VABS_S32(Dd, Dm)        EMIT(VABS_vgen(((Dd)>>4)&1, 2, (Dd)&15, 0, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VABS_S16(Dd, Dm)        EMIT(VABS_vgen(((Dd)>>4)&1, 1, (Dd)&15, 0, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VABS_S8(Dd, Dm)         EMIT(VABS_vgen(((Dd)>>4)&1, 0, (Dd)&15, 0, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VABS_F(Dd, Dm)          EMIT(VABS_vgen(((Dd)>>4)&1, 2, (Dd)&15, 1, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VABSQ_S32(Dd, Dm)       EMIT(VABS_vgen(((Dd)>>4)&1, 2, (Dd)&15, 0, 1, ((Dm)>>4)&1, (Dm)&15))
+#define VABSQ_S16(Dd, Dm)       EMIT(VABS_vgen(((Dd)>>4)&1, 1, (Dd)&15, 0, 1, ((Dm)>>4)&1, (Dm)&15))
+#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))
+
 #endif  //__ARM_EMITTER_H__
@@ -10,10 +10,10 @@
 arm_epilog:
    //update register -> emu
    //pop     {r0}
-    stm     r0, {r4-r12}   // put back reg value in emu, including EIP (so r12 must be EIP now)
+    stm     r0, {r4-r12,r14}   // put back reg value in emu, including EIP (so r14 must be Flags now)
    //restore all used register
-    //vpop     {d8-d15}
-    pop     {r4-r11, pc}
+    vpop     {d8-d15}
+    pop     {r4-r12, pc}
    //end, return...
    //bx      lr

@@ -21,7 +21,7 @@ arm_epilog:
 .global arm_epilog_fast
 arm_epilog_fast:
    //restore all used register
-    //vpop     {d8-d15}
-    pop     {r4-r11, pc}
+    vpop     {d8-d15}
+    pop     {r4-r12, pc}
    //end, return...
    //bx      lr
@@ -73,12 +73,19 @@
 ! VMOV is VORR constrained -> (removed VORR)
 ! VEOR, VBSL, VBIT, VBIF differ only depending on 2 bits: 0/1 0/1 ... VEOR -> @<2> ... @ ... @ V\op\ (field, switchable var, assembly)
 ! Three registers of different lengths (when having L and W variants): op (field) -> Q (field)
+! MRS: always APSR since program is always in userspace
+! DBG: option (field and assembly) -> imm (field and assembly)

 ; ========== BEGINNING OF REGISTERED INSTRUCTIONS =========

 # This instruction changes the instruction set to Thumb! Disabled (but valid)
 #ARM_ 1 1 1 1 1 0 1 @<1> imm24 @ set imm24 = (imm24 << 2) + (param << 1)@ @ BLX <label>

+# CPS marked as invalid (in userspace, treated as NOP)
+;INVALIDATE 1 1 1 1 0 0 0 1 0 0 0 0 <4> <8> <2> 0 <1> <4>
+ARM_ 1 1 1 1 0 0 0 1 0 0 0 0 (0) (0) (0) 1 (0) (0) (0) (0) (0) (0) @<1> (0) 0 0 0 0 (0) (0) (0) (0) SETEND \%cparam ? 'B' : 'L'\E
+;INVALIDATE 1 1 1 1 0 0 0 1 0 0 1 0 <4> <8> 0 1 1 1 <4>
+
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 0 op 0 N Q M 0 Vm VH\(op ? "SUB" : "ADD")\<c>.<dt> <Qd>, <Qn>, <Qm>
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 0 0 0 N Q M 1 Vm VQADD<c>.<dt> <Qd>, <Qn>, <Qm>
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 0 0 1 N Q M 0 Vm VRHADD<c>.<dt> <Qd>, <Qn>, <Qm>
@@ -89,10 +96,10 @@ NEON 1 1 1 1 0 0 1 0 0 D 1 1 Vn Vd 0 0 0 1 N Q M 1 Vm VORN<c> <Qd>, <Qn>, <Qm>
 NEON 1 1 1 1 0 0 1 1 0 D @<2> Vn Vd 0 0 0 1 N Q M 1 Vm @ param=0b00=0b01=0b10 const char* op op,"EOR","BSL","BIT","BIF"@ @ V\op\<c> <Qd>, <Qn>, <Qm>
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 0 1 0 N Q M 1 Vm VQSUB<c>.<dt> <Qd>, <Qn>, <Qm>
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 0 1 1 N Q M op Vm @ op=0 const char* strict strict,"T","E"@ @ VCG\strict\<c>.<dt> <Qd>, <Qn>, <Qm>
-NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 0 0 N Q M 0 Vm VSHL<c>.<dt> <Qd>, <Qn>, <Qm>
-NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 0 0 N Q M 1 Vm VQSHL<c>.<dt> <Qd>, <Qn>, <Qm>
-NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 0 1 N Q M 0 Vm VRSHL<c>.<dt> <Qd>, <Qn>, <Qm>
-NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 0 1 N Q M 1 Vm VQRSHL<c>.<dt> <Qd>, <Qn>, <Qm>
+NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 0 0 N Q M 0 Vm VSHL<c>.<dt> <Qd>, <Qm>, <Qn>
+NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 0 0 N Q M 1 Vm VQSHL<c>.<dt> <Qd>, <Qm>, <Qn>
+NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 0 1 N Q M 0 Vm VRSHL<c>.<dt> <Qd>, <Qm>, <Qn>
+NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 0 1 N Q M 1 Vm VQRSHL<c>.<dt> <Qd>, <Qm>, <Qn>
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 1 0 N Q M op Vm @ op=0 const char* opname opname,"MAX","MIN"@ @ V\opname\<c>.<dt> <Qd>, <Qn>, <Qm>
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 1 1 N Q M 0 Vm VABD<c>.<dt> <Qd>, <Qn>, <Qm>
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 1 1 N Q M 1 Vm VABA<c>.<dt> <Qd>, <Qn>, <Qm>
@@ -135,12 +142,12 @@ NEON 1 1 1 1 0 0 1 i 1 D 0 0 0 imm3 Vd cmode/=1110 0 Q op/=1 1 imm4 VMOV<c>.<dt>
 INVALIDATE 1 1 1 1 0 0 1 <1> 1 <1> 0 0 0 <11> 0 <2> 1 <4>

 # Special case for VMOVL (harder to do it on the documented place)
-NEON 1 1 1 1 0 0 1 0 1 D 0 0 1 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.S8 <Qd>, <Qm>
-NEON 1 1 1 1 0 0 1 0 1 D 0 1 0 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.S16 <Qd>, <Qm>
-NEON 1 1 1 1 0 0 1 0 1 D 1 0 0 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.S32 <Qd>, <Qm>
-NEON 1 1 1 1 0 0 1 1 1 D 0 0 1 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.U8 <Qd>, <Qm>
-NEON 1 1 1 1 0 0 1 1 1 D 0 1 0 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.U16 <Qd>, <Qm>
-NEON 1 1 1 1 0 0 1 1 1 D 1 0 0 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.U32 <Qd>, <Qm>
+NEON 1 1 1 1 0 0 1 0 1 D 0 0 1 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.S8 <Qd>, <Dm>
+NEON 1 1 1 1 0 0 1 0 1 D 0 1 0 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.S16 <Qd>, <Dm>
+NEON 1 1 1 1 0 0 1 0 1 D 1 0 0 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.S32 <Qd>, <Dm>
+NEON 1 1 1 1 0 0 1 1 1 D 0 0 1 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.U8 <Qd>, <Dm>
+NEON 1 1 1 1 0 0 1 1 1 D 0 1 0 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.U16 <Qd>, <Dm>
+NEON 1 1 1 1 0 0 1 1 1 D 1 0 0 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.U32 <Qd>, <Dm>

 NEON 1 1 1 1 0 0 1 U 1 D imm6 Vd 0 0 0 0 L Q M 1 Vm VSHR<c>.<dt> <Qd>, <Qm>, #<imm>
 NEON 1 1 1 1 0 0 1 U 1 D imm6 Vd 0 0 0 1 L Q M 1 Vm VSRA<c>.<dt> <Qd>, <Qm>, #<imm>
@@ -150,7 +157,7 @@ NEON 1 1 1 1 0 0 1 1 1 D imm6 Vd 0 1 0 0 L Q M 1 Vm VSRI<c>.<size> <Qd>, <Qm>, #
 NEON 1 1 1 1 0 0 1 0 1 D Imm6 Vd 0 1 0 1 L Q M 1 Vm VSHL<c>.I<size> <Qd>, <Qm>, #<imm>
 NEON 1 1 1 1 0 0 1 1 1 D Imm6 Vd 0 1 0 1 L Q M 1 Vm VSLI<c>.<size> <Qd>, <Qm>, #<imm>
 NEON 1 1 1 1 0 0 1 U 1 D Imm6 Vd 0 1 1 op L Q M 1 Vm VQSHL\((u & op) ? "U" : "")\<c>.<dt> <Qd>, <Qm>, #<imm>
-NEON 1 1 1 1 0 0 1 0 1 D imm6 Vd 1 0 0 0 0 0 M 1 Vm VSHRN<c>.I<size> <Dd>, <Qm>, #<imm>
+NEON 1 1 1 1 0 0 1 0 1 D imm6 Vd 1 0 0 0 0 0 M 1 Vm @ set ++size@ @ VSHRN<c>.I<size> <Dd>, <Qm>, #<imm>
 NEON 1 1 1 1 0 0 1 0 1 D imm6 Vd 1 0 0 0 0 1 M 1 Vm VRSHRN<c>.I<size> <Dd>, <Qm>, #<imm>
 NEON 1 1 1 1 0 0 1 U 1 D imm6 Vd 1 0 0 op 0 0 M 1 Vm VQSHR\((u & op) ? "U" : "")\N<c>.<dt> <Dd>, <Qm>, #<imm>
 NEON 1 1 1 1 0 0 1 U 1 D imm6 Vd 1 0 0 op 0 1 M 1 Vm VQRSHR\((u & op) ? "U" : "")\N<c>.<dt> <Dd>, <Qm>, #<imm>
@@ -194,7 +201,7 @@ NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 1 0 Vd 0 0 0 1 0 Q M 0 Vm VUZP<c>.\%d8 &l&l si
 NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 1 0 Vd 0 0 0 1 1 Q M 0 Vm VZIP<c>.\%d8 &l&l size\ <Qd>, <Qm>
 NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 1 0 Vd 0 0 1 0 0 0 M 0 Vm VMOVN<c>.<dt> <Dd>, <Qm>
 NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 1 0 Vd 0 0 1 0 0 1 M 0 Vm VQMOVUN<c>.S\%d16 &l&l size\ <Dd>, <Qm>
-NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 1 0 Vd 0 0 1 0 1 U M 0 Vm @ set u = 1 - u@ set ++size@ @ VQMOVN<c>.<dt> <Dd>, <Qm>
+NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 1 0 Vd 0 0 1 0 1 U M 0 Vm @ set ++size@ @ VQMOVN<c>.<dt> <Dd>, <Qm>
 NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 1 0 Vd 0 0 1 1 0 0 M 0 Vm VSHLL<c>.I\%d8 &l&l size\ <Qd>, <Dm>, #\%d8 &l&l size\
 NEON 1 1 1 1 0 0 1 1 1 D 1 1 size/=01 1 0 Vd 0 1 1 0 0 0 M 0 Vm VCVT<c>.F16.F32 <Dd>, <Qm>
 NEON 1 1 1 1 0 0 1 1 1 D 1 1 size/=01 1 0 Vd 0 1 1 1 0 0 M 0 Vm VCVT<c>.F32.F16 <Qd>, <Dm>
@@ -237,107 +244,141 @@ NEON 1 1 1 1 0 0 1 Q 1 D size Vn Vd 1 1 0 0 N 1 M 0 Vm @ size=0b01=0b10 unsigned
 NEON 1 1 1 1 0 0 1 Q 1 D size Vn Vd 1 1 0 1 N 1 M 0 Vm @ size=0b01=0b10 unsigned int decodedImm !@ decodedImm = m >> %; m %@ 3@ &= 0x7@ 4@ &= 0xF@ 5@ = 0x21@ @ VQRDMULH<c>.<dt> <Qd>, <Qn>, <Dm[x]>
 INVALIDATE 1 1 1 1 0 0 1 <1> 1 <16> 0 <1> 0 <4>

-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 0 size @<2> Rm @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 1 size @<2> Rm @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\], <Rm>
-;INVALIDATE 1 1 1 1 0 1 0 0 0 <1> 0 0 <22>
-INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 0 0 <8> 1 1 <10>
-NEON 1 1 1 1 0 1 0 0 1 D 0 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)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]
-NEON 1 1 1 1 0 1 0 0 1 D 0 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)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]!
-NEON 1 1 1 1 0 1 0 0 1 D 0 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)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 0 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@ @ VST2<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 0 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@ @ VST2<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 0 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@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]!
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> Rm @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 0 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=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]
-NEON 1 1 1 1 0 1 0 0 1 D 0 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=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]!
-NEON 1 1 1 1 0 1 0 0 1 D 0 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=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
-;INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 0 0 <20>
-;INVALIDATE 1 1 1 1 0 1 0 0 <2> 0 0 <20>
-
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 0 size @<2> Rm @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 1 size @<2> Rm @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\], <Rm>
-;INVALIDATE 1 1 1 1 0 1 0 0 0 <1> 1 0 <22>
-INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <8> 1 1 <10>
-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\]
+INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 0 0 <8> 1 1 <10>
+NEON 1 1 1 1 0 1 0 0 1 D 0 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)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 0 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)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 0 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)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 0 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@ @ VST2<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 0 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@ @ VST2<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 0 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@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 0 @<4> 1 1 0 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]!
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 0 @<4> 1 1 1 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 0 @<4> Rm @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
+INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <8> 1 1 <2> 1 1 <6>
+INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <8> 1 1 <2> 0 0 <1> 1 <4>
+INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <8> 1 1 1 0 <3> 1 <4>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 0 @<1> 1 1 0 1 @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 0 @<1> 1 1 1 1 @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 0 @<1> Rm @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 1 @<1> 1 1 0 1 @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 1 @<1> 1 1 1 1 @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + d]\[]}, [<Rn>\align\]
+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] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + d]\[]}, [<Rn>\align\], <Rm>
 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 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 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 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>
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]!
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> Rm @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>], <Rm>
-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=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*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 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*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=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
-;INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <20>
-;INVALIDATE 1 1 1 1 0 1 0 0 <2> 1 0 <20>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 0 size @<1> @<1> 1 1 0 1 VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[]}, [<Rn>]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 0 size @<1> @<1> 1 1 1 1 VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[]}, [<Rn>]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 0 size @<1> @<1> Rm VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[]}, [<Rn>], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 0 @<4> 1 1 0 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 0 @<4> 1 1 1 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 0 @<4> Rm @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 1 size @<1> @<1> 1 1 0 1 @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 32 << size)@ align[0] = '\0'@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[], \vecname[0x23 + param1_1+param1_1+param1_1 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 1 size @<1> @<1> 1 1 1 1 @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 32 << size)@ align[0] = '\0'@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[], \vecname[0x23 + param1_1+param1_1+param1_1 + d]\[]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 1 size @<1> @<1> Rm @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 32 << size)@ align[0] = '\0'@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[], \vecname[0x23 + param1_1+param1_1+param1_1 + d]\[]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
 INVALIDATE 1 1 1 1 0 1 0 0 <3> 0 <20>

+ARM_ 1 1 1 1 0 1 0 0 @<1> 0 0 1 Rn @<8> @<4> @<4> 'NOP'
+ARM_ 1 1 1 1 0 1 0 0 U 1 0 1 Rn (1) (1) (1) (1) imm12 PLI [<Rn>, #+/-<imm12>]
+;INVALIDATE 1 1 1 1 0 1 0 0 <2> 1 1 <4> <8> <4> <4>
+ARM_ 1 1 1 1 0 1 0 1 U @<1> 0 1 Rn (1) (1) (1) (1) imm12 PLD\param ? "W" : ""\ [<Rn>, #+/-<imm12>]
+ARM_ 1 1 1 1 0 1 0 1 U (1) 0 1 1 1 1 1 (1) (1) (1) (1) imm12 PLD [PC, #+/-<imm>]
+;INVALIDATE 1 1 1 1 0 1 0 1 0 0 1 1 <4> <8> <4> <4>
+;INVALIDATE 1 1 1 1 0 1 0 1 0 1 1 1 <4> <8> 0 0 0 0 <4>
+ARM_ 1 1 1 1 0 1 0 1 0 1 1 1 (1) (1) (1) (1) (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 1 (1) (1) (1) (1) CLREX
+;INVALIDATE 1 1 1 1 0 1 0 1 0 1 1 1 <4> <8> 0 0 1 <1> <4>
+ARM_ 1 1 1 1 0 1 0 1 0 1 1 1 (1) (1) (1) (1) (1) (1) (1) (1) (0) (0) (0) (0) 0 1 0 0 @<4> @ param=0b1111=0b1110=0b1011=0b1010=0b0111=0b0110=0b0011=0b0010 const char* option option,"SY","ST","ISH","ISHST","NSH","NSHST","OSH","OSHST","!!!"@ @ DSB \option\
+ARM_ 1 1 1 1 0 1 0 1 0 1 1 1 (1) (1) (1) (1) (1) (1) (1) (1) (0) (0) (0) (0) 0 1 0 1 @<4> @ param=0b1111=0b1110=0b1011=0b1010=0b0111=0b0110=0b0011=0b0010 const char* option option,"SY","ST","ISH","ISHST","NSH","NSHST","OSH","OSHST","!!!"@ @ DMB \option\
+ARM_ 1 1 1 1 0 1 0 1 0 1 1 1 (1) (1) (1) (1) (1) (1) (1) (1) (0) (0) (0) (0) 0 1 1 0 @<4> @ param=0b1111 const char* option option,"SY","!!!"@ @ ISB \option\
+;INVALIDATE 1 1 1 1 0 1 0 1 0 1 1 1 <4> <8> 0 1 1 1 <4>
+;INVALIDATE 1 1 1 1 0 1 0 1 0 1 1 1 <4> <8> 1 <3> <4>
+;INVALIDATE 1 1 1 1 0 1 0 1 0 1 1 1 <4> <8> <4> <4>
+;INVALIDATE 1 1 1 1 0 1 0 1 1 <1> 1 1 <4> <8> <4> <4>
+ARM_ 1 1 1 1 0 1 1 0 @<1> 0 0 1 Rn @<8> @<3> 0 @<4> 'NOP'
+ARMs 1 1 1 1 0 1 1 0 U 1 0 1 Rn (1) (1) (1) (1) imm5 type 0 Rm PLI [<Rn>, +/-<Rm>{, <shift>}]
+ARMs 1 1 1 1 0 1 1 1 U @<1> 0 1 Rn (1) (1) (1) (1) imm5 type 0 Rm PLD\param ? "W" : ""\ [<Rn>, +/-<Rm>{, <shift>}]
+;INVALIDATE 1 1 1 1 0 1 1 <3> 1 1 <4> <8> <3> 0 <4>
+# Permanently undefined
+INVALIDATE 1 1 1 1 0 1 1 1 1 1 1 1 <4> <8> 1 1 1 1 <4>
+;INVALIDATE 1 1 1 1 0 <7> <4> <8> <4> <4>
+
 ; ===== No more 1 1 1 1 ..., so invalidate all of them ====
 INVALIDATE 1 1 1 1 <28>

@@ -390,12 +431,15 @@ ARMS cond 0 0 0 1 0 0 0 1 Rn (0) (0) (0) (0) imm5 type 0 Rm TST<c> <Rn>, <Rm>{,
 ARM_ cond 0 0 0 1 U 0 W 1 Rn Rt (0) (0) (0) (0) 1 0 1 1 Rm LDRH<c> <Rt>, [<Rn>, +/-<Rm>]{!}
 ARM_ cond 0 0 0 1 U 0 W 1 Rn Rt (0) (0) (0) (0) 1 1 0 1 Rm LDRSB<c> <Rt>, [<Rn>, +/-<Rm>]{!}
 ARM_ cond 0 0 0 1 U 0 W 1 Rn Rt (0) (0) (0) (0) 1 1 1 1 Rm LDRSH<c> <Rt>, [<Rn>, +/-<Rm>]{!}
+ARM_ cond 0 0 0 1 0 0 0 0 (1) (1) (1) (1) Rd (0) (0) 0 (0) 0 0 0 0 (0) (0) (0) (0) MRS<c> <Rd>, APSR
+ARM_ cond 0 0 0 1 0 0 1 0 @<2> 0 0 (1) (1) (1) (1) (0) (0) 0 (0) 0 0 0 0 Rn @ param=0b01=0b10 const char* spec_reg spec_reg,"APSR_g","APSR_nzcvq","APSR_nzcvqg"@ @ MSR<c> \spec_reg\, <Rn>
 ARM_ cond 0 0 0 1 0 0 1 0 (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) 0 0 0 1 Rm BX<c> <Rm>
 ARM_ cond 0 0 0 1 0 0 1 0 (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) 0 0 1 0 Rm BXJ<c> <Rm>
 ARM_ cond 0 0 0 1 0 0 1 0 (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) 0 0 1 1 Rm BLX<c> <Rm>
 ARM_ cond/=1110 0 0 0 1 0 0 1 0 imm12 0 1 1 1 imm4 BKPT #<imm16>
 ARMS cond 0 0 0 1 0 0 1 1 Rn (0) (0) (0) (0) imm5 type 0 Rm TEQ<c> <Rn>, <Rm>{, <shift>}
 ARM_ cond 0 0 0 1 0 1 0 0 Rn Rt (0) (0) (0) (0) 1 0 0 1 Rm SWPB<c> <Rt>, <Rm>, [<Rn>]
+ARM_ cond 0 0 0 1 0 1 1 0 (1) (1) (1) (1) Rd (1) (1) (1) (1) 0 0 0 1 Rm CLZ<c> <Rd>, <Rm>
 ARM_ cond 0 0 0 1 U 1 W 0 Rn Rt imm4H 1 0 1 1 imm4L STRH<c> <Rt>, [<Rn>, #+/-<imm8>]{!}
 ARM_ cond 0 0 0 1 U 1 W 0 Rn Rt imm4H 1 1 0 1 imm4L LDRD<c> <Rt>, <Rt2>, [<Rn>, #+/-<imm8>]{!}
 ARM_ cond 0 0 0 1 U 1 W 0 Rn Rt imm4H 1 1 1 1 imm4L STRD<c> <Rt>, <Rt2>, [<Rn>, #+/-<imm8>]{!}
@@ -436,6 +480,18 @@ ARM_ cond 0 0 1 0 1 1 1 S Rn Rd imm12 RSC{S}<c> <Rd>, <Rn>, #<const>
 ARM_ cond 0 0 1 1 0 0 0 0 imm4 Rd imm12 MOVW<c> <Rd>, #<imm16>
 ARM_ cond 0 0 1 1 0 0 0 1 Rn (0) (0) (0) (0) imm12 TST<c> <Rn>, #<const>
 ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 0 0 0 0 0 NOP<c>
+ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 0 0 0 0 1 YIELD<c>
+ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 0 0 0 1 0 WFE<c>
+ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 0 0 0 1 1 WFI<c>
+ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 0 0 1 0 0 SEV<c>
+ARM_ cond/=1110 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 1 0 1 0 0 CSDB<c>
+ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 1 1 1 1 imm4 DBG<c> #<imm>
+INVALIDATE <4> 0 0 1 1 0 0 1 0 0 0 0 0 <8> <8>
+ARM_ cond 0 0 1 1 0 0 1 0 @<2> 0 0 (1) (1) (1) (1) imm12 @ param=0b01=0b10 const char* spec_reg spec_reg,"APSR_g","APSR_nzcvq","APSR_nzcvqg"@ @ MSR<c> \spec_reg\, #<imm>
+# MSR in system level is invalid/useless in application level
+INVALIDATE <4> 0 0 1 1 0 0 1 0 <2> 0 1 <8> <8>
+INVALIDATE <4> 0 0 1 1 0 0 1 0 <2> 1 <1> <8> <8>
+INVALIDATE <4> 0 0 1 1 0 1 1 0 <4> <8> <8>
 ARM_ cond 0 0 1 1 0 0 1 1 Rn (0) (0) (0) (0) imm12 TEQ<c> <Rn>, #<const>
 ARM_ cond 0 0 1 1 0 1 0 0 imm4 Rd imm12 MOVT<c> <Rd>, #<imm16>
 ARM_ cond 0 0 1 1 0 1 0 1 Rn (0) (0) (0) (0) imm12 CMP<c> <Rn>, #<const>
@@ -482,6 +538,7 @@ ARMs cond 0 1 1 0 1 1 1 sat_imm Rd imm5 sh 0 1 Rn USAT<c> <Rd>, #<imm>, <Rn>{, <
 ARM_ cond 0 1 1 0 1 1 1 0 sat_imm Rd (1) (1) (1) (1) 0 0 1 1 Rn USAT16<c> <Rd>, #<imm>, <Rn>
 ARM_ cond 0 1 1 0 1 1 1 0 1 1 1 1 Rd rotate (0) (0) 0 1 1 1 Rm UXTB<c> <Rd>, <Rm>{, <rotation>}
 ARM_ cond 0 1 1 0 1 1 1 0 Rn Rd rotate (0) (0) 0 1 1 1 Rm UXTAB<c> <Rd>, <Rn>, <Rm>{, <rotation>}
+ARM_ cond 0 1 1 0 1 1 1 1 (1) (1) (1) (1) Rd (1) (1) (1) (1) 0 0 1 1 Rm RBIT<c> <Rd>, <Rm>
 ARM_ cond 0 1 1 0 1 1 1 1 1 1 1 1 Rd rotate (0) (0) 0 1 1 1 Rm UXTH<c> <Rd>, <Rm>{, <rotation>}
 ARM_ cond 0 1 1 0 1 1 1 1 (1) (1) (1) (1) Rd (1) (1) (1) (1) 1 0 1 1 Rm REVSH<c> <Rd>, <Rm>
 ARM_ cond 0 1 1 0 1 1 1 1 Rn Rd rotate (0) (0) 0 1 1 1 Rm UXTAH<c> <Rd>, <Rn>, <Rm>{, <rotation>}
@@ -493,6 +550,8 @@ ARMs cond 0 1 1 0 U 1 0 0 Rn Rt imm5 type 0 Rm STRB<c> <Rt>, [<Rn>], +/-<Rm>{, <
 ARMs cond 0 1 1 0 U 1 0 1 Rn Rt imm5 type 0 Rm LDRB<c> <Rt>, [<Rn>], +/-<Rm>{, <shift>}
 ARMs cond 0 1 1 0 U 1 1 0 Rn Rt imm5 type 0 Rm STRBT<c> <Rt>, [<Rn>], +/-<Rm>{, <shift>}
 ARMs cond 0 1 1 0 U 1 1 1 Rn Rt imm5 type 0 Rm LDRBT<c> <Rt>, [<Rn>], +/-<Rm>{, <shift>}
+ARM_ cond 0 1 1 1 0 0 0 1 Rd (1) (1) (1) (1) Rm 0 0 0 1 Rn SDIV<c> <Rd>, <Rn>, <Rm>
+ARM_ cond 0 1 1 1 0 0 1 1 Rd (1) (1) (1) (1) Rm 0 0 0 1 Rn UDIV<c> <Rd>, <Rn>, <Rm>
 ARMs cond 0 1 1 1 U 0 W 0 Rn Rt imm5 type 0 Rm STR<c> <Rt>, [<Rn>, +/-<Rm>{, <shift>}]{!}
 ARMs cond 0 1 1 1 U 0 W 1 Rn Rt imm5 type 0 Rm LDR<c> <Rt>, [<Rn>, +/-<Rm>{, <shift>}]{!}
 ARMs cond 0 1 1 1 U 1 W 0 Rn Rt imm5 type 0 Rm STRB<c> <Rt>, [<Rn>, +/-<Rm>{, <shift>}]{!}
@@ -502,6 +561,7 @@ ARM_ cond 0 1 1 1 1 0 0 0 Rd Ra Rm 0 0 0 1 Rn USADA8<c> <Rd>, <Rn>, <Rm>, <Ra>
 ARM_ cond 0 1 1 1 1 1 0 msb Rd lsb 0 0 1 1 1 1 1 BFC<c> <Rd>, #<lsb>, #<width>
 ARM_ cond 0 1 1 1 1 1 0 msb Rd lsb 0 0 1 Rn BFI<c> <Rd>, <Rn>, #<lsb>, #<width>
 ARM_ cond 0 1 1 1 1 1 1 widthm1 Rd lsb 1 0 1 Rn UBFX<c> <Rd>, <Rn>, #<lsb>, #<width>
+ARM_ cond/=1110 0 1 1 1 1 1 1 1 imm12 1 1 1 1 imm4 UDF<c> #<imm16>
 ARM_ cond 1 0 0 0 0 0 W 0 Rn/=1101 register_list STMED<c> <Rn>{!}, <registers>
 ARM_ cond 1 0 0 0 0 0 W 0 Rn register_list STMDA<c> <Rn>{!}, <registers>
 ARM_ cond 1 0 0 0 0 0 W 1 Rn/=1101 register_list LDMED<c> <Rn>{!}, <registers>
@@ -535,14 +595,14 @@ INVALIDATE <4> 1 1 0 1 1 <1> 1 <9> 1 0 1 <9>
 INVALIDATE <4> 1 1 0 1 1 <1> 1 <9> 1 0 1 <9>
 VFPU cond 1 1 0 1 0 D 1 0 1 1 0 1 Vd 1 0 1 1 imm8/01 @ set --imm8@ @ FSTMX<c> <LIst>
 VFPU cond 1 1 0 1 0 D 1 0 1 1 0 1 Vd 1 0 1 sz imm8 VPUSH<c> <list>
-VFPU cond 1 1 0 1 U D 0 0 Rn Vd 1 0 1 sz imm8 VSTR<c> <Dd>, [<Rn>{, #+/-<imm>}]
+VFPU cond 1 1 0 1 U D 0 0 Rn Vd 1 0 1 sz imm8 @ set imm8*=4@ @ VSTR<c> <Dd>, [<Rn>{, #+/-<imm>}]
 VFPU cond 1 1 0 @<2> D W 0 Rn Vd 1 0 1 1 imm8/01 @ set --imm8@ @ FSTM\(param == 0b01 ? "IA" : "DB")\X<c> <Rn>{!}, <LIst>
 VFPU cond 1 1 0 @<2> D W 0 Rn Vd 1 0 1 sz imm8 VSTM\(param == 0b01 ? "IA" : "DB")\<c> <Rn>{!}, <list>
 VFPU cond 1 1 0 0 1 D 1 1 1 1 0 1 Vd 1 0 1 1 imm8/01 @ set --imm8@ @ FLDMX<c> <LIst>
 VFPU cond 1 1 0 0 1 D 1 1 1 1 0 1 Vd 1 0 1 sz imm8 VPOP<c> <list>
 VFPU cond 1 1 0 1 0 D 0 1 1 1 1 1 Vd 1 0 1 sz 0 0 0 0 0 0 0 0 VLDR<c> <Dd>, [PC, #-0]
 VFPU cond 1 1 0 1 U D 0 1 1 1 1 1 Vd 1 0 1 sz imm8 VLDR<c> <Dd>, <label>
-VFPU cond 1 1 0 1 U D 0 1 Rn Vd 1 0 1 sz imm8 VLDR<c> <Dd>, [<Rn>{, #+/-<imm>}]
+VFPU cond 1 1 0 1 U D 0 1 Rn Vd 1 0 1 sz imm8 @ set imm8*=4@ @ VLDR<c> <Dd>, [<Rn>{, #+/-<imm>}]
 VFPU cond 1 1 0 @<2> D W 1 Rn Vd 1 0 1 1 imm8/01 @ set --imm8@ @ FLDM\(param == 0b01 ? "IA" : "DB")\X<c> <Rn>{!}, <LIst>
 VFPU cond 1 1 0 @<2> D W 1 Rn Vd 1 0 1 sz imm8 VLDM\(param == 0b01 ? "IA" : "DB")\<c> <Rn>{!}, <list>
 INVALIDATE <4> 1 1 0 <25>
@@ -556,8 +616,9 @@ VFPU cond 1 1 1 0 1 0 Q 0 Vd Rt 1 0 1 1 D 0 0 1 (0) (0) (0) (0) VDUP<c>.32 <Qd>,
 VFPU cond 1 1 1 0 1 0 Q 0 Vd Rt 1 0 1 1 D 0 1 1 (0) (0) (0) (0) VDUP<c>.16 <Qd>, <Rt>
 VFPU cond 1 1 1 0 1 1 Q 0 Vd Rt 1 0 1 1 D 0 0 1 (0) (0) (0) (0) VDUP<c>.8 <Qd>, <Rt>
 VFPU cond 1 1 1 0 0 0 0 1 Vn Rt 1 0 1 0 N (0) (0) 1 (0) (0) (0) (0) VMOV<c> <Rt>, <Sn>
+VFPU cond 1 1 1 0 1 1 1 1 0 0 0 1 Rt/=1111 1 0 1 0 (0) (0) (0) 1 (0) (0) (0) (0) VMRS<c> APSR_nzcv, FPSCR
 VFPU cond 1 1 1 0 1 1 1 1 0 0 0 1 Rt 1 0 1 0 (0) (0) (0) 1 (0) (0) (0) (0) VMRS<c> <Rt>, FPSCR
-VFPU cond 1 1 1 0 U @<2> 1 Vn Rt 1 0 1 1 N @<2> 1 (0) (0) (0) (0) @ set uint8_t opc = (u << 3) + (param1_2 << 2) + param2_2@ = uint8_t shift, size @@ (opc & 0b01000) == 0b01000@ @@ (opc & 0b01001) == 0b00001@ @@ (opc & 0b11011) == 0b00000@ shift,0,1,2,0xFF;size,0,1,2,4@ set uint8_t decodedImm = (opc & 0x7) >> shift@ @ VMOV<c>.<dt> <Rt>, <Dn[x]>
+VFPU cond 1 1 1 0 U @<2> 1 Vn Rt 1 0 1 1 N @<2> 1 (0) (0) (0) (0) @ set uint8_t opc = (u << 4) + (param1_2 << 2) + param2_2@ = uint8_t shift, size @@ (opc & 0b01000) == 0b01000@ @@ (opc & 0b01001) == 0b00001@ @@ (opc & 0b11011) == 0b00000@ shift,0,1,2,0xFF;size,0,1,2,4@ set uint8_t decodedImm = (opc & 0x7) >> shift@ @ VMOV<c>.<dt> <Rt>, <Dn[x]>
 INVALIDATE <4> 1 1 1 0 <12> 1 0 1 <4> 1 <4>

 VFPU cond 1 1 1 0 0 D 0 0 Vn Vd 1 0 1 sz N op M 0 Vm VML\(op ? "S" : "A")\<c>.F\(size ? "64" : "32")\ <Dd>, <Dn>, <Dm>
@@ -585,7 +646,7 @@ VFPU cond 1 1 1 0 1 D 1 1 0 1 1 1 Vd 1 0 1 0 1 1 M 0 Vm VCVT<c>.F64.F32 <Dd>, <S
 VFPU cond 1 1 1 0 1 D 1 1 0 1 1 1 Vd 1 0 1 1 1 1 M 0 Vm VCVT<c>.F32.F64 <Sd>, <Dm>
 VFPU cond 1 1 1 0 1 D 1 1 1 0 0 0 Vd 1 0 1 sz op 1 M 0 Vm VCVT<c>.F\(size ? "64" : "32")\.\(op ? "S" : "U")\32 <Dd>, <Sm>
 VFPU cond 1 1 1 0 1 D 1 1 1 0 1 U Vd 1 0 1 sz op 1 i 0 imm4 @ set uint8_t td = (u << 2) + op + 1@ td=0b001=0b010=0b101 const char* Td Td,"S16","U16","S32","U32"@ set imm5 = (16 << op) - (imm5 >> 4) - ((imm5 & 0xF) << 1)@ @ VCVT<c>.F\(size ? "64" : "32")\.\Td\ <Dd>, <Dd>, #<imm>
-VFPU cond 1 1 1 0 1 D 1 1 1 1 0 U Vd 1 0 1 sz op 1 M 0 Vm VCVT\(op ? "" : "R")\<c>.\(u ? "U" : "S")\32.F\(size ? "64" : "32")\ <Sd>, <Dm>
+VFPU cond 1 1 1 0 1 D 1 1 1 1 U 1 Vd 1 0 1 sz op 1 M 0 Vm VCVT\(op ? "" : "R")\<c>.\(u ? "U" : "S")\32.F\(size ? "64" : "32")\ <Sd>, <Dm>
 VFPU cond 1 1 1 0 1 D 1 1 1 1 1 U Vd 1 0 1 sz op 1 i 0 imm4 @ set uint8_t td = (u << 2) + op + 1@ td=0b001=0b010=0b101 const char* Td Td,"S16","U16","S32","U32"@ set imm5 = (16 << op) - (imm5 >> 4) - ((imm5 & 0xF) << 1)@ @ VCVT<c>.\Td\.F\(size ? "64" : "32")\ <Dd>, <Dd>, #<imm>
 ;INVALIDATE <4> 1 1 1 0 1 <1> 1 1 <8> 1 0 1 <4> 0 <4>
 ;INVALIDATE <4> 1 1 1 0 <12> 1 0 1 <4> 0 <4>
@@ -1,25 +0,0 @@
-//arm update linker table for dynarec
-//called with pointer to emu as 1st parameter
-//and address of table to as 2nd parameter
-//ip is at r12
-
-.text
-.align 4
-
-.extern UpdateLinkTable
-
-.global arm_linker
-arm_linker:
-    // emu is r0
-    stm     r0, {r4-r12}   // put back reg value in emu, including EIP (easier debugging, and safer)
-    // table offset is r1
-    // IP address is r12
-    push    {r0, r1, r12}
-    mov     r2, r12
-    // call the function
-    bl      UpdateLinkTable
-    // return offset is jump address
-    mov     r3, r0
-    pop     {r0, r1, r12}
-    bx      r3
-
@@ -0,0 +1,115 @@
+// arm lock helper
+// there is 2 part: read and write
+// write return 0 on success, 1 on fail (value has been changed)
+
+.text
+.align 4
+
+.global arm_lock_read_b
+.global arm_lock_write_b
+.global arm_lock_read_h
+.global arm_lock_write_h
+.global arm_lock_read_d
+.global arm_lock_write_d
+.global arm_lock_read_dd
+.global arm_lock_write_dd
+.global arm_lock_xchg
+.global arm_lock_storeifnull
+.global arm_lock_storeifref
+
+
+arm_lock_read_b:
+    // address is r0, return is r0
+    dmb     ish
+    ldrexb  r0, [r0]
+    bx      lr
+
+arm_lock_write_b:
+    // address is r0, value is r1, return is r0
+    mov     r2, r0
+    strexb  r0, r1, [r2]
+    bx      lr
+
+arm_lock_read_h:
+    // address is r0, return is r0
+    dmb     ish
+    ldrexh  r0, [r0]
+    bx      lr
+
+arm_lock_write_h:
+    // address is r0, value is r1, return is r0
+    mov     r2, r0
+    strexh  r0, r1, [r2]
+    bx      lr
+
+arm_lock_read_d:
+    // address is r0, return is r0
+    // r0 needs to be aligned
+    dmb     ish
+    ldrex   r0, [r0]
+    bx      lr
+
+arm_lock_write_d:
+    // address is r0, value is r1, return is r0
+    // r0 needs to be aligned
+    mov     r2, r0
+    strex   r0, r1, [r2]
+    bx      lr
+
+arm_lock_read_dd:
+    // address is r2, return is r0, r1
+    dmb     ish
+    ldrexd  r2, r3, [r2]
+    str     r2, [r0]
+    str     r3, [r1]
+    bx      lr
+
+arm_lock_write_dd:
+    // address is r2, value is r0, r1, return is r0
+    // r0 needs to be aligned
+    strexd  r3, r0, r1, [r2]
+    mov     r0, r3
+    bx      lr
+
+arm_lock_xchg:
+    // address is r0, value is r1, return old value in r0
+    // r0 needs to be aligned
+    dmb     ish
+    ldrex   r2, [r0]
+    strex   r3, r1, [r0]
+    cmp     r3, #1
+    beq     arm_lock_xchg
+    mov     r0, r2
+    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)
+    // 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]
+    cmp     r3, #1
+    beq     arm_lock_storeifnull
+arm_lock_storeifnull_exit:
+    mov     r0, r2
+    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
+    bne     arm_lock_storeifref_exit
+    mov     r3, r1
+    strex   r4, r3, [r0]
+    cmp     r4, #1
+    beq     arm_lock_storeifref_loop
+arm_lock_storeifref_exit:
+    mov     r0, r3
+    pop     {r4}
+    bx      lr
@@ -0,0 +1,34 @@
+#ifndef __ARM_LOCK_HELPER__H__
+#define __ARM_LOCK_HELPER__H__
+#include <stdint.h>
+
+// LDREXB of ADDR
+extern uint8_t arm_lock_read_b(void* addr);
+// STREXB of ADDR, return 0 if ok, 1 if not
+extern int arm_lock_write_b(void* addr, uint8_t val);
+
+// LDREXH of ADDR
+extern uint16_t arm_lock_read_h(void* addr);
+// STREXH of ADDR, return 0 if ok, 1 if not
+extern int arm_lock_write_h(void* addr, uint16_t val);
+
+// LDREX of ADDR
+extern uint32_t arm_lock_read_d(void* addr);
+// STREX of ADDR, return 0 if ok, 1 if not
+extern int arm_lock_write_d(void* addr, uint32_t val);
+
+// LDREXD of ADDR
+extern void arm_lock_read_dd(uint32_t* a, uint32_t* b, void* addr);
+// STREX of ADDR, return 0 if ok, 1 if not
+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)
+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);
+
+#endif  //__ARM_LOCK_HELPER__H__
@@ -0,0 +1,23 @@
+//arm update linker table for dynarec
+//called with pointer to emu as 1st parameter
+//and address of table to as 2nd parameter
+//ip is at r12
+
+.text
+.align 4
+
+.extern LinkNext
+
+.global arm_next
+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
+    // IP address is r1
+    push    {r0, r1, r12, r14}
+    // call the function
+    bl      LinkNext
+    // return offset is jump address
+    mov     r3, r0
+    pop     {r0, r1, r12, r14}
+    bx      r3
+
@@ -9,11 +9,11 @@
 .global arm_prolog
 arm_prolog:
    //save all used register
-    push     {r4-r11, lr}
-    //vpush     {d8-d15}
+    push     {r4-r12, lr}
+    vpush     {d8-d15}
    //setup emu -> register
    //push     {r0}
-    ldm      r0, {r4-r11}    // all 8 register in direct access, no EIP (so r12 can be used as scratch)
+    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
@@ -1,19 +0,0 @@
-//arm try to update linker table for dynarec
-//and address of table to as 1st parameter
-//jump addres as second and ip as third
-// don't retry if failed to write...
-
-.text
-.align 4
-
-.global arm_tableupdate
-arm_tableupdate:
-    // jump address is r0 and IP address is r1
-    // table offset is r2
-    push    {r4, r5}
-    ldrexd  r4, r5, [r2]
-    strexd  r3, r0, r1, [r2]
-    pop     {r4, r5}
-    mov     r0, r3
-    bx      lr
-
@@ -2,6 +2,8 @@
 #include <stdlib.h>
 #include <pthread.h>
 #include <errno.h>
+#include <sys/mman.h>
+#include <setjmp.h>

 #include "debug.h"
 #include "box86context.h"
@@ -20,16 +22,15 @@
 #include "elfloader.h"
 #ifdef ARM
 #include "dynarec_arm.h"
+#include "arm_lock_helper.h"
 #else
 #error Unsupported architecture!
 #endif
-
+#include "custommem.h"
 #include "khash.h"

 KHASH_MAP_INIT_INT(dynablocks, dynablock_t*)

-KHASH_SET_INIT_INT(mark)
-
 uint32_t X31_hash_code(void* addr, int len)
 {
    if(!len) return 0;
@@ -39,27 +40,33 @@ uint32_t X31_hash_code(void* addr, int len)
 	return (uint32_t)h;
 }

-dynablocklist_t* NewDynablockList(uintptr_t base, uintptr_t text, int textsz, int nolinker, int direct)
+dynablocklist_t* NewDynablockList(uintptr_t text, int textsz, int direct)
 {
+    if(!textsz) {
+        printf_log(LOG_NONE, "Error, creating a NULL sized Dynablock\n");
+        return NULL;
+    }
    dynablocklist_t* ret = (dynablocklist_t*)calloc(1, sizeof(dynablocklist_t));
-    ret->blocks = (nolinker && direct)?NULL:kh_init(dynablocks);
-    ret->base = base;
    ret->text = text;
    ret->textsz = textsz;
-    ret->nolinker = nolinker;
-    pthread_rwlock_init(&ret->rwlock_blocks, NULL);
    if(direct && textsz) {
        ret->direct = (dynablock_t**)calloc(textsz, sizeof(dynablock_t*));
        if(!ret->direct) {printf_log(LOG_NONE, "Warning, fail to create direct block for dynablock @%p\n", (void*)text);}
    }
+    dynarec_log(LOG_DEBUG, "New Dynablocklist %p, from %p->%p\n", ret, (void*)text, (void*)(text+textsz));
    return ret;
 }

-void FreeDynablock(dynablock_t* db)
+void FreeDynablock(dynablock_t* db, int need_lock)
 {
    if(db) {
-        dynarec_log(LOG_DEBUG, "FreeDynablock(%p), db->block=%p x86=%p:%p father=%p, tablesz=%d, %swith %d son(s)\n", db, db->block, db->x86_addr, db->x86_addr+db->x86_size, db->father, db->tablesz, db->marks?"with marks, ":"", db->sons_size);
+        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);
+        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;
@@ -68,53 +75,23 @@ void FreeDynablock(dynablock_t* db)
            if(addr>=startdb && addr<enddb)
                db->parent->direct[addr-startdb] = NULL;
        }
-       // remove from hash if there
-	if (db->parent->blocks) {
-            khint_t kdb;
-            kdb = kh_get(dynablocks, db->parent->blocks, (uintptr_t) db->x86_addr - db->parent->base);
-            if(kdb!=kh_end(db->parent->blocks))
-                kh_del(dynablocks, db->parent->blocks, kdb);
-	}
-
-        if(db->marks) {
-            // Follow mark and set arm_linker instead
-            khint_t k;
-            kh_foreach_key(db->marks, k,
-                void** p = (void**)(uintptr_t)k;
-                dynarec_log(LOG_DEBUG, " -- resettable(%p)\n", p);
-                resettable(p);
-            );
-            // free mark
-            kh_destroy(mark, db->marks);
-            db->marks = NULL;
-        }
-        for(int i=0; i<db->tablesz; i+=4) {
-            dynablock_t* p = (dynablock_t*)db->table[i+3];
-            dynarec_log(LOG_DEBUG, "  -- table[%d+3] = %p ", i, p);
-            if(p && p!=db && p->marks) {
-                khint_t kd = kh_get(mark, p->marks, (uintptr_t)(&db->table[i+0]));
-                if(kd!=kh_end(p->marks)) {
-                    kh_del(mark, p->marks, kd);
-                    dynarec_log(LOG_DEBUG, "cleaned");
-                }
-                else {dynarec_log(LOG_DEBUG, "not found");}
-            }
-            dynarec_log(LOG_DEBUG, "\n");
-        }
+        // remove jumptable
+        setJumpTableDefault(db->x86_addr);
        // remove and free the sons
        for (int i=0; i<db->sons_size; ++i) {
-            dynablock_t *son = db->sons[i];
-            db->sons[i] = NULL;
-            FreeDynablock(son);
+            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((uintptr_t)db->block, db->size);
+            FreeDynarecMap(db, (uintptr_t)db->block, db->size);
+            free(db->sons);
+            free(db->instsize);
        }
-        free(db->sons);
-        free(db->table);
        free(db);
+        if(need_lock)
+            pthread_mutex_unlock(&my_context->mutex_dyndump);
    }
 }

@@ -124,31 +101,15 @@ void FreeDynablockList(dynablocklist_t** dynablocks)
        return;
    if(!*dynablocks)
        return;
-    dynarec_log(LOG_INFO, "Free %d Blocks from Dynablocklist (with %d buckets, nolinker=%d) %s\n", (*dynablocks)->blocks?kh_size((*dynablocks)->blocks):0, (*dynablocks)->blocks?kh_n_buckets((*dynablocks)->blocks):0, (*dynablocks)->nolinker, ((*dynablocks)->direct)?" With Direct mapping enabled":"");
-    dynablock_t* db;
-    if((*dynablocks)->blocks) {
-        // create a list of the parent db to delete (because there are sons in the middle that will be invalid if father is removed)
-        dynablock_t** list = (dynablock_t**)calloc(kh_size((*dynablocks)->blocks), sizeof(dynablock_t*));
-        int n = 0;
-        kh_foreach_value((*dynablocks)->blocks, db, 
-            if(!db->father) list[n++] = db;
-        );
-        for (int i=0; i<n; ++i)
-            FreeDynablock(list[i]);
-        kh_destroy(dynablocks, (*dynablocks)->blocks);
-        (*dynablocks)->blocks = NULL;
-        free(list);
-    }
+    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]);
+                FreeDynablock((*dynablocks)->direct[i], my_context?1:0);
        }
        free((*dynablocks)->direct);
    }
-    (*dynablocks)->direct = 0;
-
-    pthread_rwlock_destroy(&(*dynablocks)->rwlock_blocks);
+    (*dynablocks)->direct = NULL;

    free(*dynablocks);
    *dynablocks = NULL;
@@ -159,103 +120,12 @@ void MarkDynablock(dynablock_t* db)
    if(db) {
        if(db->father)
            db = db->father;    // mark only father
-        if(db->marks && !db->need_test) {
-            // Follow mark and set arm_linker instead
-            khint_t k;
-            kh_foreach_key(db->marks, k,
-                void** p = (void**)(uintptr_t)k;
-                resettable(p);
-            );
-            // free mark
-            kh_clear(mark, db->marks);
-        }
+        if(db->need_test)
+            return; // already done
        db->need_test = 1;
-    }
-}
-
-void ProtectDynablock(dynablock_t* db)
-{
-    if(db) {
-        if(db->father)
-            return;    // protect only father, child(ren) will be automatically
-        protectDB((uintptr_t)db->x86_addr, db->x86_size);
-    }
-}
-
-void RemoveMark(void** table)
-{
-    // remove old value if any
-    dynablock_t *old = (dynablock_t*)table[3];
-    if(old && old->father)
-        old = old->father;
-    if(old && old->marks) {
-        khint_t kd = kh_get(mark, old->marks, (uintptr_t)table);
-        if(kd!=kh_end(old->marks))
-            kh_del(mark, old->marks, kd);
-    }
-    table[3] = NULL;
-}
-
-// source is linked to dest (i.e. source->table[x] = dest->block), so add a "mark" in dest, add a "linked" info to source
-void AddMark(dynablock_t* source, dynablock_t* dest, void** table)
-{
-    int ret;
-    // remove old value if any
-    dynablock_t *old = (dynablock_t*)table[3];
-    if(old && old->father)
-        old = old->father;
-    if(old && old->marks) {
-        khint_t kd = kh_get(mark, old->marks, (uintptr_t)table);
-        if(kd!=kh_end(old->marks))
-            kh_del(mark, old->marks, kd);
-    }
-    // add mark if needed
-    if(dest->father)
-        dest = dest->father;
-    if(dest->marks) {
-        kh_put(mark, dest->marks, (uintptr_t)table, &ret);
-        table[3] = (void*)dest;
-    } else
-        table[3] = NULL;
-}
-
-void MarkDynablockList(dynablocklist_t** dynablocks)
-{
-    if(!dynablocks)
-        return;
-    if(!*dynablocks)
-        return;
-    dynarec_log(LOG_DEBUG, "Marked %d Blocks from Dynablocklist (with %d buckets, nolinker=%d) %p:0x%x %s\n", (*dynablocks)->blocks?(kh_size((*dynablocks)->blocks)):0, (*dynablocks)->blocks?(kh_n_buckets((*dynablocks)->blocks)):0, (*dynablocks)->nolinker, (void*)(*dynablocks)->text, (*dynablocks)->textsz, ((*dynablocks)->direct)?" With Direct mapping enabled":"");
-    dynablock_t* db;
-    if((*dynablocks)->blocks) {
-        kh_foreach_value((*dynablocks)->blocks, db, 
-            MarkDynablock(db);
-        );
-    }
-    if((*dynablocks)->direct) {
-        for (int i=0; i<(*dynablocks)->textsz; ++i) {
-            MarkDynablock((*dynablocks)->direct[i]);
-        }
-    }
-}
-
-void ProtectkDynablockList(dynablocklist_t** dynablocks)
-{
-    if(!dynablocks)
-        return;
-    if(!*dynablocks)
-        return;
-    dynarec_log(LOG_DEBUG, "Protect %d Blocks from Dynablocklist (with %d buckets, nolinker=%d) %p:0x%x %s\n", (*dynablocks)->blocks?(kh_size((*dynablocks)->blocks)):0, (*dynablocks)->blocks?(kh_n_buckets((*dynablocks)->blocks)):0, (*dynablocks)->nolinker, (void*)(*dynablocks)->text, (*dynablocks)->textsz, ((*dynablocks)->direct)?" With Direct mapping enabled":"");
-    dynablock_t* db;
-    if((*dynablocks)->blocks) {
-        kh_foreach_value((*dynablocks)->blocks, db, 
-            ProtectDynablock(db);
-        );
-    }
-    if((*dynablocks)->direct) {
-        for (int i=0; i<(*dynablocks)->textsz; ++i) {
-            ProtectDynablock((*dynablocks)->direct[i]);
-        }
+        setJumpTableDefault(db->x86_addr);
+        for(int i=0; i<db->sons_size; ++i)
+            setJumpTableDefault(db->sons[i]->x86_addr);
    }
 }

@@ -268,320 +138,311 @@ uintptr_t StartDynablockList(dynablocklist_t* db)
 uintptr_t EndDynablockList(dynablocklist_t* db)
 {
    if(db)
-        return db->text+db->textsz;
+        return db->text+db->textsz-1;
    return 0;
 }
-void FreeDirectDynablock(dynablocklist_t* dynablocks, uintptr_t addr, uintptr_t size)
+
+int IntervalIntersects(uintptr_t start1, uintptr_t end1, uintptr_t start2, uintptr_t end2)
 {
-    if(!dynablocks)
-        return;
-    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)
-            if(dynablocks->direct[i-startdb]) {
-                FreeDynablock(dynablocks->direct[i-startdb]);
-            }
+    if(start1 > end2 || start2 > end1)
+        return 0;
+    return 1;
 }
+
 void MarkDirectDynablock(dynablocklist_t* dynablocks, uintptr_t addr, uintptr_t size)
 {
    if(!dynablocks)
        return;
-    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)
-            if(dynablocks->direct[i-startdb]) {
-                MarkDynablock(dynablocks->direct[i-startdb]);
-            }
-}
-
-void ProtectDirectDynablock(dynablocklist_t* dynablocks, uintptr_t addr, uintptr_t size)
-{
-    if(!dynablocks)
+    if(!dynablocks->direct)
        return;
    uintptr_t startdb = dynablocks->text;
-    uintptr_t enddb = startdb + dynablocks->textsz;
+    uintptr_t enddb = startdb + dynablocks->textsz -1;
    uintptr_t start = addr;
-    uintptr_t end = addr+size;
+    uintptr_t end = addr+size-1;
    if(start<startdb)
        start = startdb;
    if(end>enddb)
        end = enddb;
+    dynablock_t *db;
    if(end>startdb && start<enddb)
-        protectDB(start, end-start); //no +1; as end/enddb is exclusive and not inclusive
+        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)
-        FreeDirectDynablock(dynablocks, addr, size);
-    if(dynablocks->blocks) {
+
+    if(dynablocks->direct) {
        dynablock_t* db;
-        uintptr_t s, e;
-        kh_foreach_value(dynablocks->blocks, db, 
-            s = (uintptr_t)db->x86_addr;
-            e = (uintptr_t)db->x86_addr+db->x86_size-1;
-            if((s>=addr && s<(addr+size)) || (e>=addr && e<(addr+size)))
-                ProtectDynablock(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, 1);
        );
+        kh_destroy(dynablocks, blocks);
    }
 }
 void MarkRangeDynablock(dynablocklist_t* dynablocks, uintptr_t addr, uintptr_t size)
 {
    if(!dynablocks)
        return;
-    if(dynablocks->direct)
-        MarkDirectDynablock(dynablocks, addr, size);
-    if(dynablocks->blocks) {
-        dynablock_t* db;
-        uintptr_t s, e;
-        kh_foreach_value(dynablocks->blocks, db, 
-            s = (uintptr_t)db->x86_addr;
-            e = (uintptr_t)db->x86_addr+db->x86_size-1;
-            if((s>=addr && s<(addr+size)) || (e>=addr && e<(addr+size)))
-                MarkDynablock(db);
-        );
+    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);
    }
 }

-dynablock_t* FindDynablockDynablocklist(void* addr, dynablocklist_t* dynablocks)
+dynablock_t* FindDynablockDynablocklist(void* addr, kh_dynablocks_t* dynablocks)
 {
    if(!dynablocks)
        return NULL;
-    if(dynablocks->direct)
-        for(int i=0; i<dynablocks->textsz; ++i) {
-            if(dynablocks->direct[i]) {
-                dynablock_t* db = dynablocks->direct[i];
-                uintptr_t s = (uintptr_t)db->block;
-                uintptr_t e = (uintptr_t)db->block+db->size;
-                if(s>=(uintptr_t)addr && e<(uintptr_t)addr)
-                    return db->father?db->father:db;
-            }
-        }
-    if(dynablocks->blocks) {
-        dynablock_t* db;
-        uintptr_t s, e;
-        kh_foreach_value(dynablocks->blocks, db, 
-            s = (uintptr_t)db->block;
-            e = (uintptr_t)db->block+db->size;
-                if((uintptr_t)addr>=s && (uintptr_t)addr<e)
-                    return db->father?db->father:db;
-        );
-    }
+    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;
 }

-dynablock_t* FindDynablockFromNativeAddress(void* addr)
+static dynablocklist_t* getDBFromAddress(uintptr_t addr)
 {
-    // unoptimized search through all dynablockslist for the dynablock that contains native addr (NULL if not found)
-    // search "volatile" dynarec first
-    dynablock_t *ret = NULL;
-    for(int idx=0; idx<65536 && !ret; ++idx)
-        if(my_context->dynmap[idx])
-            ret = FindDynablockDynablocklist(addr, my_context->dynmap[idx]->dynablocks);
-    if(ret)
-        return ret;
-    // search elfs
-    for(int idx=0; idx<my_context->elfsize && !ret; ++idx)
-        ret = FindDynablockDynablocklist(addr, GetDynablocksFromElf(my_context->elfs[idx]));
-    // last, search context dynablocks
-    if(my_context->dynablocks)
-        ret = FindDynablockDynablocklist(addr, my_context->dynablocks);
-    return ret;
+    const int idx = (addr>>DYNAMAP_SHIFT);
+    return getDB(idx);
 }

-void ConvertHash2Direct(dynablocklist_t* dynablocks)
-{
-    if(dynablocks->textsz==0 || dynablocks->text==0 || !dynablocks->blocks)
-        return; // nothing to do
-    // create the new set
-    dynablock_t **direct = (dynablock_t**)calloc(dynablocks->textsz, sizeof(dynablock_t*));
-    kh_dynablocks_t *blocks = kh_init(dynablocks);
-    // transfert
-    int ret;
-    khint_t k;
-    dynablock_t* db;
-    uintptr_t key;
-    uintptr_t start = dynablocks->text;
-    uintptr_t end = start + dynablocks->textsz;
-    kh_foreach(dynablocks->blocks, key, db,
-        key += dynablocks->base;
-        if(key>=start && key<end)
-            direct[key-start] = db;
-        else {
-            k = kh_put(dynablocks, blocks, key-dynablocks->base, &ret);
-            if(ret) {   // don't try to insert if already done...
-                kh_value(blocks, k) = db;
-            }
-        }
-    );
-    // destroy old and do the swap (should swap before destroy, but hash access is always behind mutex)
-    kh_destroy(dynablocks, dynablocks->blocks);
-    dynablocks->blocks = blocks;
-    dynablocks->direct = direct;
-}
-
-#define MAGIC_SIZE 256
-
-dynablock_t *AddNewDynablock(dynablocklist_t* dynablocks, uintptr_t addr, int with_marks, int* created)
+dynablock_t *AddNewDynablock(dynablocklist_t* dynablocks, uintptr_t addr, int* created)
 {
    if(!dynablocks) {
-        dynarec_log(LOG_INFO, "Warning: Ask to create a dynblock with a NULL dynablocklist (addr=%p)\n", (void*)addr);
+        dynarec_log(LOG_INFO, "Warning: Ask to create a dynablock with a NULL dynablocklist (addr=%p)\n", (void*)addr);
        *created = 0;
        return NULL;
    }
-    khint_t k;
-    int ret;
+    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 && (addr>=dynablocks->text) && (addr<(dynablocks->text+dynablocks->textsz))) {
+    if(dynablocks->direct) {
        block = dynablocks->direct[addr-dynablocks->text];
        if(block) {
-            dynarec_log(LOG_DUMP, "Block already exist in Direct Map\n");
+            dynarec_log(LOG_VERBOSE, "Block already exist in Direct Map\n");
            *created = 0;
            return block;
        }
    }
    
-    if(dynablocks->blocks) {
-        pthread_rwlock_rdlock(&dynablocks->rwlock_blocks);
-        // check if the block exist
-        khint_t k;
-        k = kh_get(dynablocks, dynablocks->blocks, addr-dynablocks->base);
-        if(k!=kh_end(dynablocks->blocks)) {
-            block = kh_value(dynablocks->blocks, k);
-            pthread_rwlock_unlock(&dynablocks->rwlock_blocks);
-            dynarec_log(LOG_DUMP, "Block already exist in Hash Map\n");
-            *created = 0;
-            return block;
-        }
-        pthread_rwlock_unlock(&dynablocks->rwlock_blocks);
-    }
    if (!*created)
        return block;
    
-    // Lock as write now!
-    pthread_rwlock_wrlock(&dynablocks->rwlock_blocks);
-    // create and add new block
-    dynarec_log(LOG_DUMP, "Ask for DynaRec Block creation @%p\n", (void*)addr);
-    if(dynablocks->direct && (addr>=dynablocks->text) && (addr<(dynablocks->text+dynablocks->textsz))) {
-        block = dynablocks->direct[addr-dynablocks->text] = (dynablock_t*)calloc(1, sizeof(dynablock_t));
-    } else {
-        if(dynablocks->nolinker && ((addr<dynablocks->text) || (addr>=(dynablocks->text+dynablocks->textsz)))) {
-            pthread_rwlock_unlock(&dynablocks->rwlock_blocks);
-            //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, with_marks, created);
-        }
-        k = kh_put(dynablocks, dynablocks->blocks, addr-dynablocks->base, &ret);
-        if(!ret) {
-            // Ooops, block is already there? retreive it and bail out
-            dynarec_log(LOG_DUMP, "Block already exist in Hash Map\n");
-            block = kh_value(dynablocks->blocks, k);    
-            pthread_rwlock_unlock(&dynablocks->rwlock_blocks);
-            *created = 0;
-            return block;
-        }
-        block = kh_value(dynablocks->blocks, k) = (dynablock_t*)calloc(1, sizeof(dynablock_t));
-        // check if size of hash map == magic size, if yes, convert to direct before unlocking
-        if(!dynablocks->direct && kh_size(dynablocks->blocks)==MAGIC_SIZE)
-            ConvertHash2Direct(dynablocks);
+    pthread_mutex_lock(&my_context->mutex_dyndump);
+    if(!dynablocks->direct) {
+        dynablock_t** p = (dynablock_t**)calloc(dynablocks->textsz, sizeof(dynablock_t*));
+        if(arm_lock_storeifnull(&dynablocks->direct, p)!=p)
+            free(p);    // someone already create the direct array, too late...
    }
-    block->parent = dynablocks;
-    // add mark if needed
-    if(with_marks)
-        block->marks = kh_init(mark);

-    // create an empty block first, so if other thread want to execute the same block, they can, but using interpretor path
-    pthread_rwlock_unlock(&dynablocks->rwlock_blocks);
+    // create and add new block
+    dynarec_log(LOG_VERBOSE, "Ask for DynaRec Block creation @%p\n", (void*)addr);

+    block = (dynablock_t*)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!
+        free(block);
+        *created = 0;
+        return tmp;
+    }
    *created = 1;
    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;
+
+void cancelFillBlock()
+{
+    longjmp(&dynarec_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, int create, dynablock_t* current)
+static dynablock_t* internalDBGetBlock(x86emu_t* emu, uintptr_t addr, uintptr_t filladdr, int create, dynablock_t* current, int need_lock)
 {
    // try the quickest way first: get parent of current and check if ok!
    dynablocklist_t *dynablocks = NULL;
    dynablock_t* block = NULL;
    if(current) {
-        dynablocks = current->parent;    
-        if(!(addr>=dynablocks->text && addr<(dynablocks->text+dynablocks->textsz)))
+        dynablocks = current->parent;
+        if(dynablocks && !(addr>=dynablocks->text && addr<(dynablocks->text+dynablocks->textsz)))
            dynablocks = NULL;
-        else if(dynablocks->direct && (addr>=dynablocks->text) && (addr<(dynablocks->text+dynablocks->textsz))) {
-            block = dynablocks->direct[addr-dynablocks->text];
-            if(block)
-                return block;
-        }
    }
    // nope, lets do the long way
-    if(!dynablocks)
-        dynablocks = GetDynablocksFromAddress(emu->context, addr);
-    if(!dynablocks)
-        return NULL;
+    if(!dynablocks) {
+        dynablocks = getDBFromAddress(addr);
+        if(!dynablocks) {
+            dynablocks = GetDynablocksFromAddress(emu->context, addr);
+            if(!dynablocks)
+                return NULL;
+        }
+    }
    // check direct first, without lock
-    if(dynablocks->direct && (addr>=dynablocks->text) && (addr<(dynablocks->text+dynablocks->textsz)))
-        block = dynablocks->direct[addr-dynablocks->text];
-    if(block)
-        return block;
+    if(dynablocks->direct/* && (addr>=dynablocks->text) && (addr<(dynablocks->text+dynablocks->textsz))*/)
+        if((block = dynablocks->direct[addr-dynablocks->text]))
+            return block;

    int created = create;
-    block = AddNewDynablock(dynablocks, addr, (dynablocks->nolinker)&&(dynablocks->direct), &created);
-    if(!created || !create)
+    block = AddNewDynablock(dynablocks, addr, &created);
+    if(!created)
        return block;   // existing block...

-    if(box86_dynarec_dump)
-        pthread_mutex_lock(&my_context->mutex_lock);
    // fill the block
    block->x86_addr = (void*)addr;
-    FillBlock(block);
-    if(box86_dynarec_dump)
-        pthread_mutex_unlock(&my_context->mutex_lock);
+    if(need_lock)
+        pthread_mutex_lock(&my_context->mutex_dyndump);
+    if(sigsetjmp(&dynarec_jmpbuf, 1)) {
+        printf_log(LOG_INFO, "FillBlock at %p triggered a segfault, cancelling\n", (void*)addr);
+        if(need_lock)
+            pthread_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
+        }
+        free(block);
+        block = NULL;
+    }
+    // check size
+    if(block && block->x86_size) {
+        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;
+        // 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);
+    }

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

    return block;
 }

 dynablock_t* DBGetBlock(x86emu_t* emu, uintptr_t addr, int create, dynablock_t** current)
 {
-    dynablock_t *db = internalDBGetBlock(emu, addr, create, *current);
-    if(db && (db->need_test || (db->father && db->father->need_test))) {
+    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 = X31_hash_code(father->x86_addr, father->x86_size);
+        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)%s with %d son(s)\n", father, father->x86_addr, father->x86_addr+father->x86_size, hash, father->hash, father->marks?" with Mark,":"", father->sons_size);
+            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);
+            FreeDynablock(father, 0);
            // start again... (will create a new block)
-            db = internalDBGetBlock(emu, addr, create, *current);
+            db = internalDBGetBlock(emu, addr, addr, create, *current, 0);
        } else {
            father->need_test = 0;
            protectDB((uintptr_t)father->x86_addr, father->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);
        }
+        pthread_mutex_unlock(&my_context->mutex_dyndump);
+    } 
+    return db;
+}
+
+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);
+            // start again... (will create a new block)
+            db = internalDBGetBlock(emu, addr, filladdr, create, NULL, 0);
+        } else {
+            father->need_test = 0;
+            protectDB((uintptr_t)father->x86_addr, father->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);
+        }
+        pthread_mutex_unlock(&my_context->mutex_dyndump);
    } 
    return db;
 }
@@ -2,35 +2,35 @@
 #define __DYNABLOCK_PRIVATE_H_

 typedef struct dynablocklist_s  dynablocklist_t;
-typedef struct kh_dynablocks_s  kh_dynablocks_t;
-typedef struct kh_mark_s        kh_mark_t;
+
+typedef struct instsize_s {
+    unsigned int x86:4;
+    unsigned int nat:4;
+} instsize_t;

 typedef struct dynablock_s {
    dynablocklist_t* parent;
-    kh_mark_t*      marks; // List of blocks that marked this block
    void*           block;
    int             size;
    void*           x86_addr;
-    int             x86_size;
+    uintptr_t       x86_size;
    uint32_t        hash;
-    int             need_test;
-    uintptr_t*      table;
-    int             tablesz;
-    int             done;
+    uint8_t         need_test;
+    uint8_t         done;
+    uint8_t         gone;
+    uint8_t         dummy;
    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;
 } dynablock_t;

 typedef struct dynablocklist_s {
-    kh_dynablocks_t     *blocks;
-    pthread_rwlock_t    rwlock_blocks;
-    uintptr_t           base;
    uintptr_t           text;
    int                 textsz;
-    int                 nolinker;    // in case this dynablock can disapear (also, block memory are allocated with a temporary scheme)
-    dynablock_t         **direct;    // direct mapping (waste of space, so not always there)
+    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_
@@ -2,6 +2,7 @@
 #include <stdlib.h>
 #include <pthread.h>
 #include <errno.h>
+#include <setjmp.h>

 #include "debug.h"
 #include "box86context.h"
@@ -14,84 +15,87 @@
 #include "callback.h"
 #include "emu/x86run_private.h"
 #include "x86trace.h"
+#include "threads.h"
 #ifdef DYNAREC
 #include "dynablock.h"
 #include "dynablock_private.h"
+#include "bridge.h"
 #endif

 #ifdef ARM
 void arm_prolog(x86emu_t* emu, void* addr) EXPORTDYN;
 void arm_epilog() EXPORTDYN;
 void arm_epilog_fast() EXPORTDYN;
-void arm_linker() EXPORTDYN;
-int arm_tableupdate(void* jump, uintptr_t addr, void** table) EXPORTDYN;
 #endif

-void tableupdate(void* jumpto, uintptr_t ref, void** table)
-{
-    #ifdef ARM
-    if(arm_tableupdate(jumpto, ref, table))
-    #endif
-    {
-        table[0] = jumpto;
-        table[1] = (void*)ref;
-    }
-}
-
-void resettable(void** table)
-{
-    void* p = table[1];
-    #ifdef ARM
-    tableupdate(arm_linker, (uintptr_t)p, table);
-    #endif
-    //table[2] = own_dynablock // unchanged
-    table[3] = NULL;    // removed "linked" information
-}
-
 #ifdef DYNAREC
-void* UpdateLinkTable(x86emu_t* emu, void** table, uintptr_t addr)
+#ifdef HAVE_TRACE
+uintptr_t getX86Address(dynablock_t* db, uintptr_t arm_addr);
+#endif
+void* LinkNext(x86emu_t* emu, uintptr_t addr, void* x2)
 {
-    dynablock_t* current = (dynablock_t*)table[2];
-    if(current->father)
-        current = current->father;
+    #ifdef HAVE_TRACE
+    if(!addr) {
+        x2-=8;  // actual PC is 2 instructions ahead
+        dynablock_t* db = FindDynablockFromNativeAddress(x2);
+        printf_log(LOG_NONE, "Warning, jumping to NULL address from %p (db=%p, x86addr=%p)\n", x2, db, db?(void*)getX86Address(db, (uintptr_t)x2):NULL);
+    }
+    #endif
+    dynablock_t* current = NULL;
+    void * jblock;
    dynablock_t* block = DBGetBlock(emu, addr, 1, &current);
-    if(!current)  {  // current has been invalidated, stop running it...
-        //dynarec_log(LOG_DEBUG, "--- Current invalidated while linking.\n");
-        return arm_epilog_fast;
-    }
-    if(table[3]) {
-        //dynarec_log(LOG_DEBUG, "--- Remove mark from %p (%p)\n", current, table[3]);
-        RemoveMark(table);
-    }
    if(!block) {
-        // no block, don't try again, ever
-        tableupdate(arm_epilog, addr, table);
-        return arm_epilog_fast;
+        // 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;
+        }
    }
    if(!block->done) {
        // not finished yet... leave linker
        //tableupdate(arm_linker, addr, table);
-        return arm_epilog_fast;
+        return arm_epilog;
    }
-    if(!block->block) {
+    if(!(jblock=block->block)) {
        // null block, but done: go to epilog, no linker here
-        tableupdate(arm_epilog, addr, table);
-        return arm_epilog_fast;
+        return arm_epilog;
    }
-    dynablock_t *father = block->father?block->father:block;
-    if(!block->parent->nolinker || (current && father->marks)) {
-        //dynarec_log(LOG_DEBUG, "--- Linking %p/%p to %p (table=%p[%p/%p/%p/%p])\n", block, block->block, current, table, table[0], table[1], table[2], table[3]);
-        // only update block if linker is allowed or if marks is possibe
-        if(current && father->marks && current!=father)
-            AddMark(current, father, table);
-        tableupdate(block->block, addr, table);
-    }
-    return block->block;
+    //dynablock_t *father = block->father?block->father:block;
+    return jblock;
 }
 #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
+            }
+        }
+    }
 #ifdef DYNAREC
    if(!box86_dynarec)
 #endif
@@ -111,14 +115,14 @@ void DynaCall(x86emu_t* emu, uintptr_t addr)
        dynablock_t* current = NULL;
        while(!emu->quit) {
            block = DBGetBlock(emu, R_EIP, 1, &current);
-            current = (block && !block->parent->nolinker)?block:NULL;
+            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, "Calling Interpretor @%p, emu=%p\n", (void*)R_EIP, emu);
+                dynarec_log(LOG_DEBUG, "%04d|Calling Interpretor @%p, emu=%p\n", GetTID(), (void*)R_EIP, emu);
                Run(emu, 1);
            } else {
-                dynarec_log(LOG_DEBUG, "Calling DynaRec Block @%p (%p) of %d x86 instructions (nolinker=%d, father=%p) emu=%p\n", (void*)R_EIP, block->block, block->isize ,block->parent->nolinker, block->father, emu);
+                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
@@ -130,6 +134,16 @@ void DynaCall(x86emu_t* emu, uintptr_t addr)
                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...
@@ -146,10 +160,35 @@ void DynaCall(x86emu_t* emu, uintptr_t addr)
        }
    }
 #endif
+    // clear the setjmp
+    if(ejb && jmpbuf_reset)
+        ejb->jmpbuf_ok = 0;
 }

 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
+
+            }
+        }
+    }
 #ifdef DYNAREC
    if(!box86_dynarec)
 #endif
@@ -160,14 +199,14 @@ int DynaRun(x86emu_t* emu)
        dynablock_t* current = NULL;
        while(!emu->quit) {
            block = DBGetBlock(emu, R_EIP, 1, &current);
-            current = (block && !block->parent->nolinker)?block:NULL;
+            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, "Running Interpretor @%p, emu=%p\n", (void*)R_EIP, emu);
+                dynarec_log(LOG_DEBUG, "%04d|Running Interpretor @%p, emu=%p\n", GetTID(), (void*)R_EIP, emu);
                Run(emu, 1);
            } else {
-                dynarec_log(LOG_DEBUG, "Running DynaRec Block @%p (%p) of %d x86 insts (nolinker=%d, father=%p) emu=%p\n", (void*)R_EIP, block->block, block->isize, block->parent->nolinker, block->father, emu);
+                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);
                // block is here, let's run it!
                #ifdef ARM
                arm_prolog(emu, block->block);
@@ -178,9 +217,20 @@ int DynaRun(x86emu_t* emu)
                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]);
+                }
            }
        }
    }
+    // clear the setjmp
+    if(ejb && jmpbuf_reset)
+        ejb->jmpbuf_ok = 0;
    return 0;
 #endif
 }
@@ -6,6 +6,7 @@

 #include "debug.h"
 #include "box86context.h"
+#include "custommem.h"
 #include "dynarec.h"
 #include "emu/x86emu_private.h"
 #include "tools/bridge_private.h"
@@ -19,6 +20,7 @@
 #include "dynablock_private.h"
 #include "dynarec_arm.h"
 #include "dynarec_arm_private.h"
+#include "dynarec_arm_functions.h"
 #include "elfloader.h"

 void printf_x86_instruction(zydis_dec_t* dec, instruction_x86_t* inst, const char* name) {
@@ -56,13 +58,15 @@ void printf_x86_instruction(zydis_dec_t* dec, instruction_x86_t* inst, const cha
 }

 void add_next(dynarec_arm_t *dyn, uintptr_t addr) {
+    if(!box86_dynarec_bigblock)
+        return;
+    for(int i=0; i<dyn->next_sz; ++i)
+        if(dyn->next[i]==addr)
+            return;
    if(dyn->next_sz == dyn->next_cap) {
        dyn->next_cap += 16;
        dyn->next = (uintptr_t*)realloc(dyn->next, dyn->next_cap*sizeof(uintptr_t));
    }
-    for(int i=0; i<dyn->next_sz; ++i)
-        if(dyn->next[i]==addr)
-            return;
    dyn->next[dyn->next_sz++] = addr;
 }
 uintptr_t get_closest_next(dynarec_arm_t *dyn, uintptr_t addr) {
@@ -81,9 +85,9 @@ uintptr_t get_closest_next(dynarec_arm_t *dyn, uintptr_t addr) {
    }
    return best;
 }
+#define PK(A) (*((uint8_t*)(addr+(A))))
 int is_nops(dynarec_arm_t *dyn, uintptr_t addr, int n)
 {
-    #define PK(A) (*((uint8_t*)(addr+(A))))
    if(!n)
        return 1;
    if(PK(0)==0x90)
@@ -109,84 +113,232 @@ int is_nops(dynarec_arm_t *dyn, uintptr_t addr, int n)
    if(n>7 && PK(0)==0x0f && PK(1)==0x1f && PK(2)==0x84 && PK(3)==0x00 && PK(4)==0x00 && PK(5)==0x00 && PK(6)==0x00 && PK(7)==0x00)
        return is_nops(dyn, addr+8, n-8);
    return 0;
-    #undef PK
 }

-uint32_t needed_flags(dynarec_arm_t *dyn, int ninst, uint32_t setf, int recurse)
+// return size of next instuciton, -1 is unknown
+// not all instrction are setup
+int next_instruction(dynarec_arm_t *dyn, uintptr_t addr)
 {
-    if(ninst == dyn->size || (recurse==5))
-        return X_ALL; // no more instructions, or too many jmp loop, stop
-    uint32_t needed = dyn->insts[ninst].x86.use_flags;
-    if(needed) {
-        setf &= ~needed;
-        if(!setf)   // all flags already used, no need to continue
-            return needed;
+    uint8_t opcode = PK(0);
+    uint8_t nextop;
+    switch (opcode) {
+        case 0x66:
+            opcode = PK(1);
+            switch(opcode) {
+                case 0x90:
+                    return 2;
+            }
+            break;
+        case 0x81:
+            nextop = PK(1);
+            return fakeed(dyn, addr+2, 0, nextop)-addr + 4;
+        case 0x83:
+            nextop = PK(1);
+            return fakeed(dyn, addr+2, 0, nextop)-addr + 1;
+        case 0x84:
+        case 0x85:
+        case 0x88:
+        case 0x89:
+        case 0x8A:
+        case 0x8B:
+        case 0x8C:
+        case 0x8D:
+        case 0x8E:
+        case 0x8F:
+            nextop = PK(1);
+            return fakeed(dyn, addr+2, 0, nextop)-addr;
+        case 0x50:
+        case 0x51:
+        case 0x52:
+        case 0x53:
+        case 0x54:
+        case 0x55:
+        case 0x56:
+        case 0x57:
+        case 0x58:
+        case 0x59:
+        case 0x5A:
+        case 0x5B:
+        case 0x5C:
+        case 0x5D:
+        case 0x5E:
+        case 0x5F:
+        case 0x90:
+        case 0x91:
+        case 0x92:
+        case 0x93:
+        case 0x94:
+        case 0x95:
+        case 0x96:
+        case 0x97:
+        case 0x98:
+        case 0x99:
+        case 0x9B:
+        case 0x9C:
+        case 0x9D:
+        case 0x9E:
+        case 0x9F:
+            return 1;
+        case 0xA0:
+        case 0xA1:
+        case 0xA2:
+        case 0xA3:
+            return 5;
+        case 0xB0:
+        case 0xB1:
+        case 0xB2:
+        case 0xB3:
+        case 0xB4:
+        case 0xB5:
+        case 0xB6:
+        case 0xB7:
+            return 2;
+        case 0xB8:
+        case 0xB9:
+        case 0xBA:
+        case 0xBB:
+        case 0xBC:
+        case 0xBD:
+        case 0xBE:
+        case 0xBF:
+            return 5;
+        case 0xFF:
+            nextop = PK(1);
+            switch((nextop>>3)&7) {
+                case 0: // INC Ed
+                case 1: //DEC Ed
+                case 2: // CALL Ed
+                case 4: // JMP Ed
+                case 6: // Push Ed
+                    return fakeed(dyn, addr+2, 0, nextop)-addr;
+            }
+            break;
+        default:
+            break;
    }
+    return -1;
+}
+#undef PK

-    if(!needed && !dyn->insts[ninst].x86.set_flags && !dyn->insts[ninst].x86.jmp_insts) {
-        int start = ninst;
-        int end = ninst;
-        while(end<dyn->size && !dyn->insts[end].x86.use_flags && !dyn->insts[end].x86.set_flags && !dyn->insts[end].x86.jmp_insts)
-            ++end;
-        needed = needed_flags(dyn, end, setf, recurse);
-        for(int i=start; i<end; ++i)
-            dyn->insts[i].x86.need_flags = needed;
-        return needed;
+int is_instructions(dynarec_arm_t *dyn, uintptr_t addr, int n)
+{
+    int i = 0;
+    while(i<n) {
+        int j=next_instruction(dyn, addr+i);
+        if(j<=0) return 0;
+        i+=j;
    }
-
-    if(dyn->insts[ninst].x86.set_flags && (dyn->insts[ninst].x86.state_flags!=SF_MAYSET)) {
-        if((setf & ~dyn->insts[ninst].x86.set_flags) == 0)
-            return needed;    // all done, gives all the flags needed
-        setf |= dyn->insts[ninst].x86.set_flags;    // add new flags to continue
-    }
-
-    int jinst = dyn->insts[ninst].x86.jmp_insts;
-    if(dyn->insts[ninst].x86.jmp) {
-        dyn->insts[ninst].x86.need_flags = (jinst==-1)?X_PEND:needed_flags(dyn, jinst, setf, recurse+1);
-        if(dyn->insts[ninst].x86.use_flags)  // conditionnal jump
-             dyn->insts[ninst].x86.need_flags |= needed_flags(dyn, ninst+1, setf, recurse);
-    } else
-        dyn->insts[ninst].x86.need_flags = needed_flags(dyn, ninst+1, setf, recurse);
-    if(dyn->insts[ninst].x86.state_flags==SF_MAYSET)
-        needed |= dyn->insts[ninst].x86.need_flags;
-    else
-        needed |= (dyn->insts[ninst].x86.need_flags & ~dyn->insts[ninst].x86.set_flags);
-    if(needed == (X_PEND|X_ALL))
-        needed = X_ALL;
-    return needed;
+    return (i==n)?1:0;
 }

-void arm_pass0(dynarec_arm_t* dyn, uintptr_t addr);
-void arm_pass1(dynarec_arm_t* dyn, uintptr_t addr);
-void arm_pass2(dynarec_arm_t* dyn, uintptr_t addr);
-void arm_pass3(dynarec_arm_t* dyn, uintptr_t addr);
+instsize_t* addInst(instsize_t* insts, size_t* size, size_t* cap, int x86_size, int arm_size)
+{
+    // x86 instruction is <16 bytes
+    int toadd;
+    if(x86_size>arm_size)
+        toadd = 1 + x86_size/15;
+    else
+        toadd = 1 + arm_size/15;
+    if((*size)+toadd>(*cap)) {
+        *cap = (*size)+toadd;
+        insts = (instsize_t*)realloc(insts, (*cap)*sizeof(instsize_t));
+    }
+    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)
+            insts[*size].nat = 15;
+        else
+            insts[*size].nat = arm_size;
+        arm_size -= insts[*size].nat;
+        ++(*size);
+        --toadd;
+    }
+    return insts;
+}

-void* FillBlock(dynablock_t* block) {
+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);
+
+__thread void* current_helper = NULL;
+
+void CancelBlock()
+{
+    dynarec_arm_t* helper = (dynarec_arm_t*)current_helper;
+    current_helper = NULL;
+    if(!helper)
+        return;
+    free(helper->next);
+    free(helper->insts);
+    free(helper->sons_x86);
+    free(helper->sons_arm);
+    if(helper->dynablock && helper->dynablock->block)
+        FreeDynarecMap(helper->dynablock, (uintptr_t)helper->dynablock->block, helper->dynablock->size);
+}
+
+void* FillBlock(dynablock_t* block, uintptr_t addr) {
+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 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;
+    }
    // init the helper
-    uintptr_t addr = (uintptr_t)block->x86_addr;
    dynarec_arm_t helper = {0};
-    helper.nolinker = box86_dynarec_linker?(block->parent->nolinker):1;
+    current_helper = &helper;
+    helper.dynablock = block;
    helper.start = addr;
-    arm_pass0(&helper, addr);
+    uintptr_t start = addr;
+    helper.cap = 64; // needs epilog handling
+    helper.insts = (instruction_arm_t*)calloc(helper.cap, sizeof(instruction_arm_t));
+    // pass 0, addresses, x86 jump addresses, overall size of the block
+    uintptr_t end = arm_pass0(&helper, addr);
+    // no need for next anymore
+    free(helper.next);
+    helper.next_sz = helper.next_cap = 0;
+    helper.next = NULL;
+    // basic checks
    if(!helper.size) {
        dynarec_log(LOG_DEBUG, "Warning, null-sized dynarec block (%p)\n", (void*)addr);
-        block->done = 1;
-        free(helper.next);
+        CancelBlock();
        return (void*)block;
    }
-    helper.cap = helper.size+3; // needs epilog handling
-    helper.insts = (instruction_arm_t*)calloc(helper.cap, sizeof(instruction_arm_t));
-    // pass 1, addresses, x86 jump addresses, flags
-    arm_pass1(&helper, addr);
+    // already protect the block and compute hash signature
+    protectDB(addr, end-addr);  //end is 1byte after actual end
+    uint32_t hash = X31_hash_code((void*)addr, end-addr);
+    // Compute flag_need, without taking into account any barriers
+    uint32_t last_need = X_PEND;
+    for(int i = helper.size; i-- > 0;) {
+        last_need |= helper.insts[i].x86.use_flags;
+        if((helper.insts[i].x86.state_flags == SF_SUBSET) && (last_need&X_PEND))
+            last_need = (X_ALL & (~helper.insts[i].x86.set_flags) ) | helper.insts[i].x86.use_flags;
+        if (last_need == (X_PEND | X_ALL)) {
+            last_need = X_ALL;
+        }
+        helper.insts[i].x86.need_flags = last_need;
+        if ((helper.insts[i].x86.set_flags) && !(helper.insts[i].x86.state_flags & SF_MAYSET)) {
+            if (last_need & X_PEND) {
+                last_need = (~helper.insts[i].x86.set_flags) & X_ALL;
+            } else {
+                last_need &= ~helper.insts[i].x86.set_flags;
+            }
+        }
+    }
    // calculate barriers
-    uintptr_t start = helper.insts[0].x86.addr;
-    uintptr_t end = helper.insts[helper.size].x86.addr+helper.insts[helper.size].x86.size;
    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)
+            if(j<start || j>=end) {
                helper.insts[i].x86.jmp_insts = -1;
-            else {
+            } else {
                // find jump address instruction
                int k=-1;
                for(int i2=0; i2<helper.size && k==-1; ++i2) {
@@ -198,58 +350,140 @@ void* FillBlock(dynablock_t* block) {
                helper.insts[i].x86.jmp_insts = k;
            }
        }
-    for(int i=0; i<helper.size; ++i)
-        if(helper.insts[i].x86.set_flags && !helper.insts[i].x86.need_flags) {
-            helper.insts[i].x86.need_flags = needed_flags(&helper, i+1, helper.insts[i].x86.set_flags, 0);
-            if((helper.insts[i].x86.need_flags&X_PEND) && (helper.insts[i].x86.state_flags==SF_MAYSET))
-                helper.insts[i].x86.need_flags = X_ALL;
+    // check for the optionnal barriers now
+    for(int i=helper.size-1; i>=0; --i) {
+        if(helper.insts[i].x86.barrier==3)
+            if(helper.insts[i].x86.jmp_insts == -1) {
+                if(i==helper.size-1 || helper.insts[i+1].x86.barrier)
+                    helper.insts[i].x86.barrier=2;  // nope, end of block or barrier just after
+                else
+                    helper.insts[i].x86.barrier=0;  // ok, no need for a barrier here after all
+            } else 
+                helper.insts[i].x86.barrier=2;
+    }
+    // check to remove useless barrier, in case of jump when destination doesn't needs flags
+    for(int i=helper.size-1; i>=0; --i) {
+        if(helper.insts[i].x86.jmp
+        && helper.insts[i].x86.jmp_insts>=0
+        && helper.insts[helper.insts[i].x86.jmp_insts].x86.barrier==1) {
+            int k = helper.insts[i].x86.jmp_insts;
+            //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 = 4; // remove barrier (keep FPU barrier, and still reset state flag)
+             }
        }
+    }
+    // reset need_flags and compute again, now taking barrier into account (because barrier change use_flags)
+    for(int i = helper.size; i-- > 0;) {
+        if(helper.insts[i].x86.barrier==1)
+            // immediate barrier
+            helper.insts[i].x86.use_flags |= X_PEND;
+        else if(helper.insts[i].x86.jmp 
+        && helper.insts[i].x86.jmp_insts>=0
+        ) {
+            if(helper.insts[helper.insts[i].x86.jmp_insts].x86.barrier==1)
+                // jumpto barrier
+                helper.insts[i].x86.use_flags |= X_PEND;
+            else
+                helper.insts[i].x86.use_flags |= helper.insts[helper.insts[i].x86.jmp_insts].x86.need_flags;
+        }
+    }
+    for(int i = helper.size; i-- > 0;)
+        helper.insts[i].x86.need_flags = 0;
+    last_need = X_PEND;
+    for(int i = helper.size; i-- > 0;) {
+        helper.insts[i].x86.need_flags = last_need;
+        if((helper.insts[i].x86.state_flags == SF_SUBSET) && (last_need&X_PEND))
+            last_need = (X_ALL & (~helper.insts[i].x86.set_flags) ) | helper.insts[i].x86.use_flags;
+        if ((helper.insts[i].x86.set_flags) && !(helper.insts[i].x86.state_flags & SF_MAYSET)) {
+            if (last_need & X_PEND) {
+                last_need = (~helper.insts[i].x86.set_flags) & X_ALL;
+            } else {
+                last_need &= ~helper.insts[i].x86.set_flags;
+            }
+        }
+        last_need |= helper.insts[i].x86.use_flags;
+        if (last_need == (X_PEND | X_ALL)) {
+            last_need = X_ALL;
+        }
+    }
+
+    // pass 1, float optimisations
+    arm_pass1(&helper, addr);
    
    // pass 2, instruction size
    arm_pass2(&helper, addr);
    // ok, now allocate mapped memory, with executable flag on
    int sz = helper.arm_size;
-    void* p = (void*)AllocDynarecMap(sz, block->parent->nolinker);
+    void* p = (void*)AllocDynarecMap(block, sz);
    if(p==NULL) {
-        free(helper.insts);
-        free(helper.next);
-        return (void*)block;
+        dynarec_log(LOG_DEBUG, "AllocDynarecMap(%p, %d) failed, cancelling block\n", block, sz);
+        CancelBlock();
+        return NULL;
    }
    helper.block = p;
    helper.arm_start = (uintptr_t)p;
-    helper.tablesz = helper.tablei;
-    if(helper.tablesz)
-        helper.table = (uintptr_t*)calloc(helper.tablesz, sizeof(uintptr_t));
    if(helper.sons_size) {
        helper.sons_x86 = (uintptr_t*)calloc(helper.sons_size, sizeof(uintptr_t));
        helper.sons_arm = (void**)calloc(helper.sons_size, sizeof(void*));
    }
    // pass 3, emit (log emit arm opcode)
    if(box86_dynarec_dump) {
-        dynarec_log(LOG_NONE, "%sEmitting %d bytes for %d x86 bytes", (box86_dynarec_dump>1)?"\e[01;36m":"", helper.arm_size, helper.isize); 
+        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":"");
    }
    helper.arm_size = 0;
    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(sz!=helper.arm_size) {
+        printf_log(LOG_NONE, "BOX86: Warning, size difference in block between pass2 (%d) & pass3 (%d)!\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) {
+            printf_log(LOG_NONE, "%p:", dump);
+            for(; dump<(uint8_t*)helper.insts[i+1].x86.addr; ++dump)
+                printf_log(LOG_NONE, " %02X", *dump);
+            printf_log(LOG_NONE, "\t%d -> %d\n", helper.insts[i].size2, helper.insts[i].size);
+        }
+        printf_log(LOG_NONE, " ------------\n");
+    }
    // all done...
-    __builtin___clear_cache(p, p+sz);   // need to clear the cache before execution...
+    __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*)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
    free(helper.insts);
+    helper.insts = NULL;
    free(helper.next);
-    block->table = helper.table;
-    block->tablesz = helper.tablesz;
-    for (int i=0; i<helper.tablesz/4; ++i)
-        block->table[i*4+2] = (uintptr_t)block;
+    helper.next = NULL;
    block->size = sz;
    block->isize = helper.size;
    block->block = p;
    block->need_test = 0;
-    block->x86_addr = (void*)start;
+    //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 = (helper.nolinker)?X31_hash_code(block->x86_addr, block->x86_size):0;
+    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();
+        return NULL;
+    }
    // fill sons if any
    dynablock_t** sons = NULL;
    int sons_size = 0;
@@ -257,8 +491,8 @@ void* FillBlock(dynablock_t* block) {
        sons = (dynablock_t**)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], 0, &created);
-            if(created) {    // avoid breaking a working block! also, block could be outside this parent...
+            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];
@@ -266,6 +500,8 @@ void* FillBlock(dynablock_t* block) {
                son->father = block;
                son->done = 1;
                sons[sons_size++] = son;
+                if(!son->parent)
+                    son->parent = block->parent;
            }
        }
        if(sons_size) {
@@ -274,8 +510,11 @@ void* FillBlock(dynablock_t* block) {
        } else
            free(sons);
    }
-    if(block->parent->nolinker)
-        protectDB((uintptr_t)block->x86_addr, block->x86_size);
+    free(helper.sons_x86);
+    helper.sons_x86 = NULL;
+    free(helper.sons_arm);
+    helper.sons_arm = NULL;
+    current_helper = NULL;
    block->done = 1;
    return (void*)block;
 }
@@ -25,71 +25,207 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
 {
    uint8_t opcode = F8;
    uint8_t nextop;
-    int32_t i32;
+    int32_t i32, j32;
    uint32_t u32;
    uint8_t gd, ed;
    uint8_t wback;
    int fixedaddress;
+
+    MAYUSE(j32);
+
    switch(opcode) {
-        case 0x33:
-            INST_NAME("XOR Gd, FS:Ed");
-            SETFLAGS(X_ALL, SF_SET);
-            grab_fsdata(dyn, addr, ninst, x12);
+
+        case 0x03:
+            INST_NAME("ADD Gd, FS:Ed");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_fsdata(dyn, addr, ninst, x14);
            nextop = F8;
            GETGD;
-            GETEDO(x12);
-            emit_xor32(dyn, ninst, gd, ed, x3, x12);
+            GETEDO(x14);
+            emit_add32(dyn, ninst, gd, ed, x3, x14);
+            break;
+
+        case 0x2B:
+            INST_NAME("SUB Gd, FS:Ed");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_fsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGD;
+            GETEDO(x14);
+            emit_sub32(dyn, ninst, gd, ed, x3, x14);
+            break;
+
+        case 0x33:
+            INST_NAME("XOR Gd, FS:Ed");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_fsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGD;
+            GETEDO(x14);
+            emit_xor32(dyn, ninst, gd, ed, x3, x14);
+            break;
+
+        case 0x39:
+            INST_NAME("CMP FS:Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_fsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGD;
+            GETEDO(x14);
+            emit_cmp32(dyn, ninst, ed, gd, x3, x14);
            break;

        case 0x3B:
            INST_NAME("CMP Gd, FS:Ed");
-            SETFLAGS(X_ALL, SF_SET);
-            grab_fsdata(dyn, addr, ninst, x12);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_fsdata(dyn, addr, ninst, x14);
            nextop = F8;
            GETGD;
-            GETEDO(x12);
-            emit_cmp32(dyn, ninst, gd, ed, x3, x12);
+            GETEDO(x14);
+            emit_cmp32(dyn, ninst, gd, ed, x3, x14);
+            break;
+
+        case 0x67:
+            // reduced EA size...
+            opcode = F8;
+            switch(opcode) {
+                case 0x89:
+                    INST_NAME("MOV FS:Ew16, Gw");
+                    nextop = F8;
+                    grab_fsdata(dyn, addr, ninst, x14);
+                    GETGD;  // don't need GETGW here
+                    if((nextop&0xC0)==0xC0) {
+                        ed = xEAX+(nextop&7);
+                        if(ed!=gd) {
+                            BFI(ed, gd, 0, 16);
+                        }
+                    } else {
+                        addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
+                        STRH_REG(gd, x14, ed);
+                    }
+                    break;
+
+                case 0x8B:
+                    INST_NAME("MOV Gd, FS:Ew16");
+                    nextop=F8;
+                    grab_fsdata(dyn, addr, ninst, x14);
+                    GETGD;
+                    if((nextop&0xC0)==0xC0) {   // reg <= reg
+                        MOV_REG(gd, xEAX+(nextop&7));
+                    } else {                    // mem <= reg
+                        addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
+                        LDR_REG_LSL_IMM5(gd, ed, x14, 0);
+                    }
+                    break;
+
+                case 0x8F:
+                    INST_NAME("POP FS:Ew16");
+                    nextop=F8;
+                    grab_fsdata(dyn, addr, ninst, x14);
+                    if((nextop&0xC0)==0xC0) {
+                        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);
+                    }
+                    break;
+
+                case 0xA1:
+                    INST_NAME("MOV EAX, FS:Ow");
+                    grab_fsdata(dyn, addr, ninst, x1);
+                    u32 = F16;
+                    if(u32) {
+                        MOV32(x2, u32);
+                        ADD_REG_LSL_IMM5(x1, x1, x2, 0);
+                    }
+                    LDR_IMM9(xEAX, x1, 0);
+                    break;
+
+                case 0xA3:
+                    INST_NAME("MOV FS:Ow, EAX");
+                    grab_fsdata(dyn, addr, ninst, x1);
+                    u32 = F16;
+                    if(u32) {
+                        MOV32(x2, u32);
+                        ADD_REG_LSL_IMM5(x1, x1, x2, 0);
+                    }
+                    STR_IMM9(xEAX, x1, 0);
+                    break;
+
+                case 0xFF:
+                    nextop = F8;
+                    grab_fsdata(dyn, addr, ninst, x14);
+                    switch((nextop>>3)&7) {
+                        case 6: // Push Ed
+                            INST_NAME("PUSH FS:Ew");
+                            if((nextop&0xC0)==0xC0) {   // reg
+                                DEFAULT;
+                            } else {                    // mem <= i32
+                                addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
+                                LDR_REG_LSL_IMM5(x3, ed, x14, 0);
+                                PUSH1(x3);
+                            }
+                            break;
+                        default:
+                            DEFAULT;
+                    }
+                    break;
+                default:
+                    DEFAULT;
+            }
            break;

        case 0x89:
            INST_NAME("MOV FS:Ed, Gd");
-            grab_fsdata(dyn, addr, ninst, x12);
+            grab_fsdata(dyn, addr, ninst, x14);
            nextop=F8;
            GETGD;
            if((nextop&0xC0)==0xC0) {   // reg <= reg
                MOV_REG(xEAX+(nextop&7), gd);
            } else {                    // mem <= reg
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
-                STR_REG_LSL_IMM5(gd, ed, x12, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0);
+                STR_REG_LSL_IMM5(gd, ed, x14, 0);
            }
            break;

        case 0x8B:
            INST_NAME("MOV Gd, FS:Ed");
-            grab_fsdata(dyn, addr, ninst, x12);
+            grab_fsdata(dyn, addr, ninst, x14);
            nextop=F8;
            GETGD;
            if((nextop&0xC0)==0xC0) {   // reg <= reg
                MOV_REG(gd, xEAX+(nextop&7));
            } else {                    // mem <= reg
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
-                LDR_REG_LSL_IMM5(gd, ed, x12, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0);
+                LDR_REG_LSL_IMM5(gd, ed, x14, 0);
            }
            break;

        case 0x8F:
            INST_NAME("POP FS:Ed");
-            grab_fsdata(dyn, addr, ninst, x12);
+            grab_fsdata(dyn, addr, ninst, x14);
            nextop = F8;
            if((nextop&0xC0)==0xC0) {
-                POP(xESP, (1<<(xEAX+(nextop&7))));
+                POP1(xEAX+(nextop&7));
            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
-                POP(xESP, (1<<x2));
-                STR_REG_LSL_IMM5(x2, ed, x12, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
+                POP1(x2);
+                STR_REG_LSL_IMM5(x2, ed, x14, 0);
            }
            break;

+        case 0xA0:
+            INST_NAME("MOV AL, FS:Id");
+            grab_fsdata(dyn, addr, ninst, x1);
+            i32 = F32S;
+            if(i32>-4096 && i32<4096) {
+                LDRB_IMM9(xEAX, x1, i32);
+            } else {
+                MOV32(x2, i32);
+                LDRB_REG_LSL_IMM5(xEAX, x1, x2, 0);
+            }
+            break;
        case 0xA1:
            INST_NAME("MOV EAX, FS:Id");
            grab_fsdata(dyn, addr, ninst, x1);
@@ -98,48 +234,90 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                LDR_IMM9(xEAX, x1, i32);
            } else {
                MOV32(x2, i32);
-                ADD_REG_LSL_IMM5(x1, x1, x2, 0);
-                LDR_IMM9(xEAX, x1, 0);
+                LDR_REG_LSL_IMM5(xEAX, x1, x2, 0);
+            }
+            break;
+        case 0xA2:
+            INST_NAME("MOV FS:Od, AL");
+            grab_fsdata(dyn, addr, ninst, x1);
+            i32 = F32S;
+            if(i32>-4096 && i32<4096) {
+                STRB_IMM9(xEAX, x1, i32);
+            } else {
+                MOV32(x2, i32);
+                STRB_REG_LSL_IMM5(xEAX, x1, x2, 0);
            }
            break;
-
        case 0xA3:
            INST_NAME("MOV FS:Od, EAX");
            grab_fsdata(dyn, addr, ninst, x1);
-            u32 = F32;
-            MOV32(x2, u32);
-            ADD_REG_LSL_IMM5(x2, x1, x2, 0);
-            STR_IMM9(xEAX, x2, 0);
+            i32 = F32S;
+            if(i32>-4096 && i32<4096) {
+                STR_IMM9(xEAX, x1, i32);
+            } else {
+                MOV32(x2, i32);
+                STR_REG_LSL_IMM5(xEAX, x2, x1, 0);
+            }
            break;

        case 0xC7:
            INST_NAME("MOV FS:Ed, Id");
-            grab_fsdata(dyn, addr, ninst, x12);
+            grab_fsdata(dyn, addr, ninst, x14);
            nextop=F8;
            if((nextop&0xC0)==0xC0) {   // reg <= i32
                i32 = F32S;
                ed = xEAX+(nextop&7);
                MOV32(ed, i32);
            } else {                    // mem <= i32
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0);
                i32 = F32S;
                MOV32(x3, i32);
-                STR_REG_LSL_IMM5(x3, ed, x12, 0);
+                STR_REG_LSL_IMM5(x3, ed, x14, 0);
            }
            break;
            
+        case 0xE9:
+        case 0xEB:
+            INST_NAME("(ignored) FS:");
+            --addr;
+            break;
+
        case 0xFF:
            nextop = F8;
-            grab_fsdata(dyn, addr, ninst, x12);
+            grab_fsdata(dyn, addr, ninst, x14);
            switch((nextop>>3)&7) {
+                case 2: // CALL Ed
+                    INST_NAME("CALL FS:Ed");
+                    PASS2IF(ninst && dyn->insts[ninst-1].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
+                    }
+                    if((nextop&0xC0)==0xC0) {   // reg
+                        MOV_REG(xEIP, xEAX+(nextop&7));
+                    } else {                    // mem
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0);
+                        LDR_REG_LSL_IMM5(xEIP, ed, x14, 0);
+                    }
+                    BARRIER(1);
+                    BARRIER_NEXT(1);
+                    if(!dyn->insts || ninst!=dyn->size-1) {
+                    } else {
+                        *need_epilog = 0;
+                        *ok = 0;
+                    }
+                    MOV32(x2, addr);
+                    PUSH1(x2);
+                    jump_to_next(dyn, 0, xEIP, ninst);
+                    break;
                case 6: // Push Ed
-                    INST_NAME("PUSH Ed");
+                    INST_NAME("PUSH FS:Ed");
                    if((nextop&0xC0)==0xC0) {   // reg
                        DEFAULT;
                    } else {                    // mem <= i32
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
-                        LDR_REG_LSL_IMM5(x3, ed, x12, 0);
-                        PUSH(xESP, 1<<x3);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0);
+                        LDR_REG_LSL_IMM5(x3, ed, x14, 0);
+                        PUSH1(x3);
                    }
                    break;
                default:
@@ -25,65 +25,356 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
 {
    uint8_t opcode = F8;
    uint8_t nextop;
-    int32_t i32;
+    int32_t i32, j32;
    uint32_t u32;
    uint8_t gd, ed;
-    uint8_t wback;
+    uint8_t wback, wb1, wb2, wb;
+    uint8_t u8;
    int fixedaddress;
+
+    MAYUSE(j32);
+    MAYUSE(wback);
+    MAYUSE(wb2);
+    
    switch(opcode) {
+
        case 0x03:
            INST_NAME("ADD Gd, GS:Ed");
-            SETFLAGS(X_ALL, SF_SET);
-            grab_tlsdata(dyn, addr, ninst, x12);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_tlsdata(dyn, addr, ninst, x14);
            nextop = F8;
            GETGD;
-            GETEDO(x12);
-            emit_add32(dyn, ninst, gd, ed, x3, x12);
+            GETEDO(x14);
+            emit_add32(dyn, ninst, gd, ed, x3, x14);
+            break;
+
+        case 0x0F:  // more opcodes
+            opcode = F8;
+            switch(opcode) {
+                
+                case 0xAF:
+                    INST_NAME("IMUL Gd, Ed");
+                    SETFLAGS(X_ALL, SF_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x14);
+                    nextop = F8;
+                    GETGD;
+                    GETEDO(x14);
+                    UFLAG_IF {
+                        SMULL(x3, gd, gd, ed);
+                        UFLAG_OP1(x3);
+                        UFLAG_RES(gd);
+                        UFLAG_DF(x3, d_imul32);
+                    } else {
+                        MUL(gd, gd, ed);
+                    }
+                    break;
+
+                default:
+                    DEFAULT;
+            }
+            break;
+
+        case 0x21:
+            INST_NAME("AND Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_tlsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGD;
+            GETEDO2(x14);
+            emit_and32(dyn, ninst, ed, gd, x3, x14);
+            WBACK;
+            break;
+
+        case 0x23:
+            INST_NAME("AND Gd, GS:Ed");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_tlsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGD;
+            GETEDO(x14);
+            emit_and32(dyn, ninst, gd, ed, x3, x14);
+            break;
+
+        case 0x29:
+            INST_NAME("SUB GS:Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_tlsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGD;
+            GETEDO2(x14);
+            emit_sub32(dyn, ninst, ed, gd, x3, x14);
+            WBACK;
            break;

        case 0x2B:
            INST_NAME("SUB Gd, GS:Ed");
-            SETFLAGS(X_ALL, SF_SET);
-            grab_tlsdata(dyn, addr, ninst, x12);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_tlsdata(dyn, addr, ninst, x14);
            nextop = F8;
            GETGD;
-            GETEDO(x12);
-            emit_sub32(dyn, ninst, gd, ed, x3, x12);
+            GETEDO(x14);
+            emit_sub32(dyn, ninst, gd, ed, x3, x14);
+            break;
+
+        case 0x31:
+            INST_NAME("XOR GS:Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_tlsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGD;
+            GETEDO2(x14);
+            emit_xor32(dyn, ninst, gd, ed, x3, x14);
+            WBACK;
            break;

        case 0x33:
            INST_NAME("XOR Gd, GS:Ed");
-            SETFLAGS(X_ALL, SF_SET);
-            grab_tlsdata(dyn, addr, ninst, x12);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_tlsdata(dyn, addr, ninst, x14);
            nextop = F8;
            GETGD;
-            GETEDO(x12);
-            emit_xor32(dyn, ninst, gd, ed, x3, x12);
+            GETEDO(x14);
+            emit_xor32(dyn, ninst, gd, ed, x3, x14);
+            break;
+
+        case 0x3B:
+            INST_NAME("CMP Gd, GS:Ed");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_tlsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGD;
+            GETEDO(x14);
+            emit_cmp32(dyn, ninst, gd, ed, x3, x14);
+            break;
+
+        case 0x69:
+            INST_NAME("IMUL Gd, GS:Ed, Id");
+            SETFLAGS(X_ALL, SF_PENDING);
+            grab_tlsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGD;
+            GETEDO(x14);
+            i32 = F32S;
+            MOV32(x14, i32);
+            UFLAG_IF {
+                SMULL(x3, gd, x14, ed);
+                UFLAG_OP1(x3);
+                UFLAG_RES(gd);
+                UFLAG_DF(x3, d_imul32);
+            } else {
+                MUL(gd, ed, x14);
+            }
+            break;
+
+        case 0x80:
+        case 0x82:
+            nextop = F8;
+            switch((nextop>>3)&7) {
+                case 0: //ADD
+                    INST_NAME("ADD GS:Eb, Ib");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x1);
+                    GETEBO(x1);
+                    u8 = F8;
+                    emit_add8c(dyn, ninst, x1, u8, x2, x14);
+                    EBBACK;
+                    break;
+                case 1: //OR
+                    INST_NAME("OR GS:Eb, Ib");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x1);
+                    GETEBO(x1);
+                    u8 = F8;
+                    emit_or8c(dyn, ninst, x1, u8, x2, x14);
+                    EBBACK;
+                    break;
+                case 2: //ADC
+                    INST_NAME("ADC GS:Eb, Ib");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x1);
+                    GETEBO(x1);
+                    u8 = F8;
+                    emit_adc8c(dyn, ninst, x1, u8, x2, x14);
+                    EBBACK;
+                    break;
+                case 3: //SBB
+                    INST_NAME("SBB GS:Eb, Ib");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x1);
+                    GETEBO(x1);
+                    u8 = F8;
+                    emit_sbb8c(dyn, ninst, x1, u8, x2, x14);
+                    EBBACK;
+                    break;
+                case 4: //AND
+                    INST_NAME("AND GS:Eb, Ib");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x1);
+                    GETEBO(x1);
+                    u8 = F8;
+                    emit_and8c(dyn, ninst, x1, u8, x2, x14);
+                    EBBACK;
+                    break;
+                case 5: //SUB
+                    INST_NAME("SUB GS:Eb, Ib");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x1);
+                    GETEBO(x1);
+                    u8 = F8;
+                    emit_sub8c(dyn, ninst, x1, u8, x2, x14);
+                    EBBACK;
+                    break;
+                case 6: //XOR
+                    INST_NAME("XOR GS:Eb, Ib");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x1);
+                    GETEBO(x1);
+                    u8 = F8;
+                    emit_xor8c(dyn, ninst, x1, u8, x2, x14);
+                    EBBACK;
+                    break;
+                case 7: //CMP
+                    INST_NAME("CMP GS:Eb, Ib");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x1);
+                    GETEBO(x1);
+                    u8 = F8;
+                    if(u8) {
+                        MOVW(x2, u8);
+                        emit_cmp8(dyn, ninst, x1, x2, x3, x14);
+                    } else {
+                        emit_cmp8_0(dyn, ninst, x1, x3, x14);
+                    }
+                    break;
+            }
+            break;
+        case 0x81:
+        case 0x83:
+            nextop = F8;
+            switch((nextop>>3)&7) {
+                case 0: //ADD
+                    if(opcode==0x81) {
+                        INST_NAME("ADD Ed, Id");
+                    } else {
+                        INST_NAME("ADD Ed, Ib");
+                    }
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x14);
+                    GETEDO2(x14);
+                    if(opcode==0x81) i32 = F32S; else i32 = F8S;
+                    emit_add32c(dyn, ninst, ed, i32, x3, x14);
+                    WBACK;
+                    break;
+                case 1: //OR
+                    if(opcode==0x81) {INST_NAME("OR Ed, Id");} else {INST_NAME("OR Ed, Ib");}
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x14);
+                    GETEDO2(x14);
+                    if(opcode==0x81) i32 = F32S; else i32 = F8S;
+                    emit_or32c(dyn, ninst, ed, i32, x3, x14);
+                    WBACK;
+                    break;
+                case 2: //ADC
+                    if(opcode==0x81) {INST_NAME("ADC Ed, Id");} else {INST_NAME("ADC Ed, Ib");}
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x14);
+                    GETEDO2(x14);
+                    if(opcode==0x81) i32 = F32S; else i32 = F8S;
+                    emit_adc32c(dyn, ninst, ed, i32, x3, x14);
+                    WBACK;
+                    break;
+                case 3: //SBB
+                    if(opcode==0x81) {INST_NAME("SBB Ed, Id");} else {INST_NAME("SBB Ed, Ib");}
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x14);
+                    GETEDO2(x14);
+                    if(opcode==0x81) i32 = F32S; else i32 = F8S;
+                    emit_sbb32c(dyn, ninst, ed, i32, x3, x14);
+                    WBACK;
+                    break;
+                case 4: //AND
+                    if(opcode==0x81) {INST_NAME("AND Ed, Id");} else {INST_NAME("AND Ed, Ib");}
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x14);
+                    GETEDO2(x14);
+                    if(opcode==0x81) i32 = F32S; else i32 = F8S;
+                    emit_and32c(dyn, ninst, ed, i32, x3, x14);
+                    WBACK;
+                    break;
+                case 5: //SUB
+                    if(opcode==0x81) {INST_NAME("SUB Ed, Id");} else {INST_NAME("SUB Ed, Ib");}
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x14);
+                    GETEDO2(x14);
+                    if(opcode==0x81) i32 = F32S; else i32 = F8S;
+                    emit_sub32c(dyn, ninst, ed, i32, x3, x14);
+                    WBACK;
+                    break;
+                case 6: //XOR
+                    if(opcode==0x81) {INST_NAME("XOR Ed, Id");} else {INST_NAME("XOR Ed, Ib");}
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x14);
+                    GETEDO2(x14);
+                    if(opcode==0x81) i32 = F32S; else i32 = F8S;
+                    emit_xor32c(dyn, ninst, ed, i32, x3, x14);
+                    WBACK;
+                    break;
+                case 7: //CMP
+                    if(opcode==0x81) {INST_NAME("CMP Ed, Id");} else {INST_NAME("CMP Ed, Ib");}
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x14);
+                    GETEDO2(x14);
+                    if(opcode==0x81) i32 = F32S; else i32 = F8S;
+                    if(i32) {
+                        MOV32(x2, i32);
+                        emit_cmp32(dyn, ninst, ed, x2, x3, x14);
+                    } else {
+                        emit_cmp32_0(dyn, ninst, ed, x3, x14);
+                    }
+                    break;
+            }
            break;

        case 0x89:
            INST_NAME("MOV GS:Ed, Gd");
-            grab_tlsdata(dyn, addr, ninst, x12);
+            grab_tlsdata(dyn, addr, ninst, x14);
            nextop=F8;
            GETGD;
            if((nextop&0xC0)==0xC0) {   // reg <= reg
                MOV_REG(xEAX+(nextop&7), gd);
            } else {                    // mem <= reg
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
-                STR_REG_LSL_IMM5(gd, ed, x12, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0);
+                STR_REG_LSL_IMM5(gd, ed, x14, 0);
            }
            break;

        case 0x8B:
            INST_NAME("MOV Gd, GS:Ed");
-            grab_tlsdata(dyn, addr, ninst, x12);
+            grab_tlsdata(dyn, addr, ninst, x14);
            nextop=F8;
            GETGD;
            if((nextop&0xC0)==0xC0) {   // reg <= reg
                MOV_REG(gd, xEAX+(nextop&7));
            } else {                    // mem <= reg
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
-                LDR_REG_LSL_IMM5(gd, ed, x12, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0);
+                LDR_REG_LSL_IMM5(gd, ed, x14, 0);
+            }
+            break;
+
+        case 0x8F:
+            INST_NAME("POP GS:Ed");
+            grab_tlsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            if((nextop&0xC0)==0xC0) {
+                POP1((xEAX+(nextop&7)));
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
+                POP1(x2);
+                STR_REG_LSL_IMM5(x2, ed, x14, 0);
            }
            break;

@@ -111,17 +402,208 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,

        case 0xC7:
            INST_NAME("MOV GS:Ed, Id");
-            grab_tlsdata(dyn, addr, ninst, x12);
+            grab_tlsdata(dyn, addr, ninst, x14);
            nextop=F8;
            if((nextop&0xC0)==0xC0) {   // reg <= i32
                i32 = F32S;
                ed = xEAX+(nextop&7);
                MOV32(ed, i32);
            } else {                    // mem <= i32
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0);
                i32 = F32S;
                MOV32(x3, i32);
-                STR_REG_LSL_IMM5(x3, ed, x12, 0);
+                STR_REG_LSL_IMM5(x3, ed, x14, 0);
+            }
+            break;
+
+        case 0xD1:
+            nextop = F8;
+            grab_tlsdata(dyn, addr, ninst, x14);
+            switch((nextop>>3)&7) {
+                case 0:
+                    INST_NAME("ROL Ed, 1");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    GETEDO2(x14);
+                    emit_rol32c(dyn, ninst, ed, 1, x3, x14);
+                    WBACK;
+                    break;
+                case 1:
+                    INST_NAME("ROR Ed, 1");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    GETEDO2(x14);
+                    emit_ror32c(dyn, ninst, ed, 1, x3, x14);
+                    WBACK;
+                    break;
+                case 2:
+                    INST_NAME("RCL Ed, 1");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    MOVW(x2, 1);
+                    GETEDO2(x14);
+                    wb = ed;
+                    if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
+                    CALL_(rcl32, ed, (1<<x14));
+                    SBACK(wb);
+                    break;
+                case 3:
+                    INST_NAME("RCR Ed, 1");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    MOVW(x2, 1);
+                    wb = ed;
+                    if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
+                    CALL_(rcr32, ed, (1<<x14));
+                    SBACK(wb);
+                    break;
+                case 4:
+                case 6:
+                    INST_NAME("SHL Ed, 1");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                    GETEDO2(x14);
+                    emit_shl32c(dyn, ninst, ed, 1, x3, x14);
+                    WBACK;
+                    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;
+                    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;
+                    break;
+            }
+            break;
+        case 0xD3:
+            nextop = F8;
+            grab_tlsdata(dyn, addr, ninst, x14);
+            switch((nextop>>3)&7) {
+                case 0:
+                    INST_NAME("ROL Ed, CL");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    AND_IMM8(x3, xECX, 0x1f);
+                    TSTS_REG_LSL_IMM5(x3, x3, 0);
+                    B_MARK2(cEQ);
+                    RSB_IMM8(x3, x3, 0x20);
+                    GETEDO2(x14);
+                    MOV_REG_ROR_REG(ed, ed, x3);
+                    WBACK;
+                    UFLAG_IF {  // calculate flags directly
+                        CMPS_IMM8(x3, 31);
+                        B_MARK(cNE);
+                            MOV_REG_LSR_IMM5(x1, ed, 31);
+                            ADD_REG_LSL_IMM5(x1, x1, ed, 0);
+                            BFI(xFlags, x1, F_OF, 1);
+                        MARK;
+                        BFI(xFlags, ed, F_CF, 1);
+                        UFLAG_DF(x2, d_none);
+                    }
+                    MARK2;
+                    break;
+                case 1:
+                    INST_NAME("ROR Ed, CL");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    AND_IMM8(x3, xECX, 0x1f);
+                    TSTS_REG_LSL_IMM5(x3, x3, 0);
+                    B_MARK2(cEQ);
+                    GETEDO2(x14);
+                    MOV_REG_ROR_REG(ed, ed, x3);
+                    WBACK;
+                    UFLAG_IF {  // calculate flags directly
+                        CMPS_IMM8(x3, 1);
+                        B_MARK(cNE);
+                            MOV_REG_LSR_IMM5(x2, ed, 30); // x2 = d>>30
+                            XOR_REG_LSR_IMM8(x2, x2, x2, 1); // x2 = ((d>>30) ^ ((d>>30)>>1))
+                            BFI(xFlags, x2, F_OF, 1);
+                        MARK;
+                        MOV_REG_LSR_IMM5(x2, ed, 31);
+                        BFI(xFlags, x2, F_CF, 1);
+                        UFLAG_DF(x2, d_none);
+                    }
+                    MARK2;
+                    break;
+                case 2:
+                    INST_NAME("RCL Ed, CL");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    AND_IMM8(x2, xECX, 0x1f);
+                    wb = ed;
+                    if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
+                    CALL_(rcl32, ed, (1<<x14));
+                    SBACK(wb);
+                    break;
+                case 3:
+                    INST_NAME("RCR Ed, CL");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    AND_IMM8(x2, xECX, 0x1f);
+                    wb = ed;
+                    if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
+                    CALL_(rcr32, ed, (1<<x14));
+                    SBACK(wb);
+                    break;
+                case 4:
+                case 6:
+                    INST_NAME("SHL Ed, CL");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                    AND_IMM8(x3, xECX, 0x1f);
+                    GETEDO2(x14);
+                    emit_shl32(dyn, ninst, ed, x3, x3, x14);
+                    WBACK;
+                    break;
+                case 5:
+                    INST_NAME("SHR Ed, CL");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                    AND_IMM8(x3, xECX, 0x1f);
+                    GETEDO2(x14);
+                    emit_shr32(dyn, ninst, ed, x3, x3, x14);
+                    WBACK;
+                    break;
+                case 7:
+                    INST_NAME("SAR Ed, CL");
+                    SETFLAGS(X_ALL, SF_PENDING);
+                    AND_IMM8(x3, xECX, 0x1f);
+                    GETEDO2(x14);
+                    UFLAG_OP12(ed, x3);
+                    MOV_REG_ASR_REG(ed, ed, x3);
+                    WBACK;
+                    UFLAG_RES(ed);
+                    UFLAG_DF(x3, d_sar32);
+                    break;
+            }
+            break;
+
+        case 0xE9:
+        case 0xEB:
+            INST_NAME("(ignored) GS:");
+            --addr;
+            break;
+
+        case 0xFF:
+            nextop = F8;
+            grab_tlsdata(dyn, addr, ninst, x14);
+            switch((nextop>>3)&7) {
+                case 6: // Push Ed
+                    INST_NAME("PUSH GS:Ed");
+                    if((nextop&0xC0)==0xC0) {   // reg
+                        DEFAULT;
+                    } else {                    // mem <= i32
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0);
+                        LDR_REG_LSL_IMM5(x3, ed, x14, 0);
+                        PUSH1(x3);
+                    }
+                    break;
+                default:
+                    DEFAULT;
            }
            break;

@@ -20,6 +20,7 @@
 #include "arm_printer.h"

 #include "dynarec_arm_helper.h"
+#include "dynarec_arm_functions.h"


 uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog)
@@ -27,7 +28,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
    uint8_t opcode = F8;
    uint8_t nextop, u8;
    uint32_t u32;
-    int32_t i32;
+    int32_t i32, j32;
    int16_t i16;
    uint16_t u16;
    uint8_t gd, ed;
@@ -35,34 +36,35 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
    int fixedaddress;
    MAYUSE(u16);
    MAYUSE(u8);
+    MAYUSE(j32);
    while(opcode==0x66) opcode = F8;    // "unlimited" 0x66 as prefix for variable sized NOP
    if(opcode==0x2E) opcode = F8;       // cs: is ignored
    switch(opcode) {
        
        case 0x01:
            INST_NAME("ADD Ew, Gw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGW(x2);
            GETEW(x1);
-            emit_add16(dyn, ninst, x1, x2, x12, x3, (wb1 && (wback==x3))?1:0);
+            emit_add16(dyn, ninst, x1, x2, x14, x3, (wb1 && (wback==x3))?1:0);
            EWBACK;
            break;
        case 0x03:
            INST_NAME("ADD Gw, Ew");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGW(x1);
            GETEW(x2);
-            emit_add16(dyn, ninst, x1, x2, x3, x12, 0);
+            emit_add16(dyn, ninst, x1, x2, x3, x14, 0);
            GWBACK;
            break;
        case 0x05:
            INST_NAME("ADD AX, Iw");
-            SETFLAGS(X_ALL, SF_PENDING);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            i32 = F16;
            UXTH(x1, xEAX, 0);
-            emit_add16c(dyn, ninst, x1, i32, x3, x12);
+            emit_add16c(dyn, ninst, x1, i32, x3, x14);
            BFI(xEAX, x1, 0, 16);
            break;
        case 0x06:
@@ -81,33 +83,33 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            STRH_IMM8(x2, x1, 0);
            ADD_IMM8(xESP, xESP, 2);
            MOVW(x1, 0);
-            STR_IMM9(x1, xEmu, offsetof(x86emu_t, segs_clean[_ES]));
+            STR_IMM9(x1, xEmu, offsetof(x86emu_t, segs_serial[_ES]));
            break;

        case 0x09:
            INST_NAME("OR Ew, Gw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGW(x2);
            GETEW(x1);
-            emit_or16(dyn, ninst, x1, x2, x12, x3, (wb1 && (wback==x3))?1:0);
+            emit_or16(dyn, ninst, x1, x2, x14, x2);
            EWBACK;
            break;
        case 0x0B:
            INST_NAME("OR Gw, Ew");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGW(x1);
            GETEW(x2);
-            emit_or16(dyn, ninst, x1, x2, x12, x3, 0);
+            emit_or16(dyn, ninst, x1, x2, x14, x3);
            GWBACK;
            break;
        case 0x0D:
            INST_NAME("OR AX, Iw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            i32 = F16;
            UXTH(x1, xEAX, 0);
-            emit_or16c(dyn, ninst, x1, i32, x3, x12);
+            emit_or16c(dyn, ninst, x1, i32, x3, x14);
            BFI(xEAX, x1, 0, 16);
            break;
                
@@ -117,88 +119,87 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0x11:
            INST_NAME("ADC Ew, Gw");
            READFLAGS(X_CF);
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGW(x2);
            GETEW(x1);
-            emit_adc16(dyn, ninst, x1, x2, x12, x3, (wb1 && (wback==x3))?1:0);
+            emit_adc16(dyn, ninst, x1, x2, x14, x3, (wb1 && (wback==x3))?1:0);
            EWBACK;
            break;
        case 0x13:
            INST_NAME("ADC Gw, Ew");
            READFLAGS(X_CF);
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGW(x1);
            GETEW(x2);
-            emit_adc16(dyn, ninst, x1, x2, x12, x3, 0);
+            emit_adc16(dyn, ninst, x1, x2, x14, x3, 0);
            GWBACK;
            break;
        case 0x15:
            INST_NAME("ADC AX, Iw");
            READFLAGS(X_CF);
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            i32 = F16;
            UXTH(x1, xEAX, 0);
-            emit_adc16c(dyn, ninst, x1, i32, x3, x12);
+            emit_adc16c(dyn, ninst, x1, i32, x3, x14);
            BFI(xEAX, x1, 0, 16);
            break;

        case 0x19:
            INST_NAME("SBB Ew, Gw");
            READFLAGS(X_CF);
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGW(x2);
            GETEW(x1);
-            emit_sbb16(dyn, ninst, x1, x2, x12, x3, (wb1 && (wback==x3))?1:0);
+            emit_sbb16(dyn, ninst, x1, x2, x14, x3, (wb1 && (wback==x3))?1:0);
            EWBACK;
            break;
        case 0x1B:
            INST_NAME("SBB Gw, Ew");
            READFLAGS(X_CF);
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGW(x1);
            GETEW(x2);
-            emit_sbb16(dyn, ninst, x1, x2, x12, x3, 0);
+            emit_sbb16(dyn, ninst, x1, x2, x14, x3, 0);
            GWBACK;
            break;
        case 0x1D:
            INST_NAME("SBB AX, Iw");
            READFLAGS(X_CF);
-            SETFLAGS(X_ALL, SF_SET);
-            i32 = F16;
-            MOV32(x2, i32);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            i16 = F16;
            UXTH(x1, xEAX, 0);
-            emit_sbb16c(dyn, ninst, x1, i32, x3, x12);
+            emit_sbb16c(dyn, ninst, x1, i16, x3, x14);
            BFI(xEAX, x1, 0, 16);
            break;

        case 0x21:
            INST_NAME("AND Ew, Gw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGW(x2);
            GETEW(x1);
-            emit_and16(dyn, ninst, x1, x2, x12, x3, (wb1 && (wback==x3))?1:0);
+            emit_and16(dyn, ninst, x1, x2, x14, x2);
            EWBACK;
            break;
        case 0x23:
            INST_NAME("AND Gw, Ew");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGW(x1);
            GETEW(x2);
-            emit_and16(dyn, ninst, x1, x2, x3, x12, 0);
+            emit_and16(dyn, ninst, x1, x2, x3, x14);
            GWBACK;
            break;
        case 0x25:
            INST_NAME("AND AX, Iw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            i32 = F16;
            UXTH(x1, xEAX, 0);
-            emit_and16c(dyn, ninst, x1, i32, x3, x12);
+            emit_and16c(dyn, ninst, x1, i32, x3, x14);
            BFI(xEAX, x1, 0, 16);
            break;
        case 0x26:
@@ -208,81 +209,81 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,

        case 0x29:
            INST_NAME("SUB Ew, Gw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGW(x2);
            GETEW(x1);
-            emit_sub16(dyn, ninst, x1, x2, x12, x3, (wb1 && (wback==x3))?1:0);
+            emit_sub16(dyn, ninst, x1, x2, x14, x3, (wb1 && (wback==x3))?1:0);
            EWBACK;
            break;
        case 0x2B:
            INST_NAME("SUB Gw, Ew");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGW(x1);
            GETEW(x2);
-            emit_sub16(dyn, ninst, x1, x2, x3, x12, 0);
+            emit_sub16(dyn, ninst, x1, x2, x3, x14, 0);
            GWBACK;
            break;
        case 0x2D:
            INST_NAME("SUB AX, Iw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            i32 = F16;
            UXTH(x1, xEAX, 0);
-            emit_sub16c(dyn, ninst, x1, i32, x3, x12);
+            emit_sub16c(dyn, ninst, x1, i32, x3, x14);
            BFI(xEAX, x1, 0, 16);
            break;

        case 0x31:
            INST_NAME("XOR Ew, Gw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGW(x2);
            GETEW(x1);
-            emit_xor16(dyn, ninst, x1, x2, x12, x3, (wb1 && (wback==x3))?1:0);
+            emit_xor16(dyn, ninst, x1, x2, x14, x2);
            EWBACK;
            break;
        case 0x33:
            INST_NAME("XOR Gw, Ew");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGW(x1);
            GETEW(x2);
-            emit_xor16(dyn, ninst, x1, x2, x3, x12, 0);
+            emit_xor16(dyn, ninst, x1, x2, x3, x14);
            GWBACK;
            break;
        case 0x35:
            INST_NAME("XOR AX, Iw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            i32 = F16;
            UXTH(x1, xEAX, 0);
-            emit_xor16c(dyn, ninst, x1, i32, x3, x12);
+            emit_xor16c(dyn, ninst, x1, i32, x3, x14);
            BFI(xEAX, x1, 0, 16);
            break;

        case 0x39:
            INST_NAME("CMP Ew, Gw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGW(x2);
            GETEW(x1);
-            emit_cmp16(dyn, ninst, x1, x2, x3, x12);
+            emit_cmp16(dyn, ninst, x1, x2, x3, x14);
            break;
        case 0x3B:
            INST_NAME("CMP Gw, Ew");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETGW(x1);
            GETEW(x2);
-            emit_cmp16(dyn, ninst, x1, x2, x3, x12);
+            emit_cmp16(dyn, ninst, x1, x2, x3, x14);
            break;
        case 0x3D:
            INST_NAME("CMP AX, Iw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            i32 = F16;
            MOVW(x2, i32);
            UXTH(x1, xEAX, 0);
-            emit_cmp16(dyn, ninst, x1, x2, x3, x12);
+            emit_cmp16(dyn, ninst, x1, x2, x3, x14);
            break;

        case 0x40:
@@ -297,7 +298,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
            gd = xEAX+(opcode&7);
            UXTH(x1, gd, 0);
-            emit_inc16(dyn, ninst, x1, x3, x12);
+            emit_inc16(dyn, ninst, x1, x3, x14);
            BFI(gd, x1, 0, 16);
            break;
        case 0x48:
@@ -312,7 +313,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
            gd = xEAX+(opcode&7);
            UXTH(x1, gd, 0);
-            emit_dec16(dyn, ninst, x1, x3, x12);
+            emit_dec16(dyn, ninst, x1, x3, x14);
            BFI(gd, x1, 0, 16);
            break;
        case 0x50:
@@ -324,8 +325,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0x56:
        case 0x57:
            INST_NAME("PUSH Reg16");
-            SUB_IMM8(xESP, xESP, 2);
-            STRH_IMM8(xEAX+(opcode&7), xESP, 0);    //TODO: use a single STRH pre-decrement instruction
+            PUSH16(xEAX+(opcode&7), xESP);
            break;
        case 0x58:
        case 0x59:
@@ -336,17 +336,45 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0x5E:
        case 0x5F:
            INST_NAME("POP Reg16");
-            LDRH_IMM8(x1, xESP, 0);     //TODO: use a single LDRH post-increment instruction
+            POP16(x1, xESP);
            BFI(xEAX+(opcode&7), x1, 0, 16);
-            ADD_IMM8(xESP, xESP, 2);
+            break;
+        case 0x60:
+            INST_NAME("PUSHA");
+            MOV_REG(x1, xESP);
+            PUSH16(xEAX, xESP);
+            PUSH16(xECX, xESP);
+            PUSH16(xEDX, xESP);
+            PUSH16(xEBX, xESP);
+            PUSH16(x1, xESP);
+            PUSH16(xEBP, xESP);
+            PUSH16(xESI, xESP);
+            PUSH16(xEDI, xESP);
+            break;
+        case 0x61:
+            INST_NAME("POPA");
+            POP16(x1, xESP);
+            BFI(xEDI, x1, 0, 16);
+            POP16(x1, xESP);
+            BFI(xESI, x1, 0, 16);
+            POP16(x1, xESP);
+            BFI(xEBP, x1, 0, 16);
+            ADD_IMM8(xESP, xESP, 2);    //POP16(xESP, xESP);
+            POP16(x1, xESP);
+            BFI(xEBX, x1, 0, 16);
+            POP16(x1, xESP);
+            BFI(xEDX, x1, 0, 16);
+            POP16(x1, xESP);
+            BFI(xECX, x1, 0, 16);
+            POP16(x1, xESP);
+            BFI(xEAX, x1, 0, 16);
            break;

        case 0x68:
            INST_NAME("PUSH Iw");
            u16 = F16;
            MOVW(x2, u16);
-            SUB_IMM8(xESP, xESP, 2);
-            STRH_IMM8(x2, xESP, 0);
+            PUSH16(x2, xESP);
            break;
        case 0x69:
            INST_NAME("IMUL Gw,Ew,Iw");
@@ -364,8 +392,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            INST_NAME("PUSH Ib");
            i16 = F8S;
            MOVW(x2, i16);
-            SUB_IMM8(xESP, xESP, 2);
-            STRH_IMM8(x2, xESP, 0);
+            PUSH16(x2, xESP);
            break;
        case 0x6B:
            INST_NAME("IMUL Gw,Ew,Ib");
@@ -390,83 +417,114 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    } else {
                        INST_NAME("ADD Ew, Ib");
                    }
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    GETEW(x1);
                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    emit_add16c(dyn, ninst, ed, i16, x2, x12);
+                    emit_add16c(dyn, ninst, ed, i16, x2, x14);
                    EWBACK;
                    break;
                case 1: //OR
                    if(opcode==0x81) {INST_NAME("OR Ew, Iw");} else {INST_NAME("OR Ew, Ib");}
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    GETEW(x1);
                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    emit_or16c(dyn, ninst, x1, i16, x2, x12);
+                    emit_or16c(dyn, ninst, x1, i16, x2, x14);
                    EWBACK;
                    break;
                case 2: //ADC
                    if(opcode==0x81) {INST_NAME("ADC Ew, Iw");} else {INST_NAME("ADC Ew, Ib");}
                    READFLAGS(X_CF);
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    GETEW(x1);
                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    emit_adc16c(dyn, ninst, x1, i16, x2, x12);
+                    emit_adc16c(dyn, ninst, x1, i16, x2, x14);
                    EWBACK;
                    break;
                case 3: //SBB
                    if(opcode==0x81) {INST_NAME("SBB Ew, Iw");} else {INST_NAME("SBB Ew, Ib");}
                    READFLAGS(X_CF);
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    GETEW(x1);
                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    emit_sbb16c(dyn, ninst, x1, i16, x2, x12);
+                    emit_sbb16c(dyn, ninst, x1, i16, x2, x14);
                    EWBACK;
                    break;
                case 4: //AND
                    if(opcode==0x81) {INST_NAME("AND Ew, Iw");} else {INST_NAME("AND Ew, Ib");}
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    GETEW(x1);
                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    emit_and16c(dyn, ninst, x1, i16, x2, x12);
+                    emit_and16c(dyn, ninst, x1, i16, x2, x14);
                    EWBACK;
                    break;
                case 5: //SUB
                    if(opcode==0x81) {INST_NAME("SUB Ew, Iw");} else {INST_NAME("SUB Ew, Ib");}
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    GETEW(x1);
                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    emit_sub16c(dyn, ninst, x1, i16, x2, x12);
+                    emit_sub16c(dyn, ninst, x1, i16, x2, x14);
                    EWBACK;
                    break;
                case 6: //XOR
                    if(opcode==0x81) {INST_NAME("XOR Ew, Iw");} else {INST_NAME("XOR Ew, Ib");}
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    GETEW(x1);
                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
-                    emit_xor16c(dyn, ninst, x1, i16, x2, x12);
+                    emit_xor16c(dyn, ninst, x1, i16, x2, x14);
                    EWBACK;
                    break;
                case 7: //CMP
                    if(opcode==0x81) {INST_NAME("CMP Ew, Iw");} else {INST_NAME("CMP Ew, Ib");}
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    GETEW(x1);
                    if(opcode==0x81) i16 = F16S; else i16 = F8S;
                    if(i16) {
                        MOVW(x2, i16);
-                        emit_cmp16(dyn, ninst, x1, x2, x3, x12);
+                        emit_cmp16(dyn, ninst, x1, x2, x3, x14);
                    } else
-                        emit_cmp16_0(dyn, ninst, x1, x3, x12);
+                        emit_cmp16_0(dyn, ninst, x1, x3, x14);
                    break;
            }
            break;
                
        case 0x85:
            INST_NAME("TEST Ew, Gw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            nextop = F8;
            GETEW(x1);
            GETGW(x2);
-            emit_test16(dyn, ninst, x1, x2, x3, x12);
+            emit_test16(dyn, ninst, x1, x2, x3, x14);
+            break;
+
+        case 0x87:
+            INST_NAME("XCHG Ew, Gw");
+            nextop = F8;
+            if((nextop&0xC0)==0xC0) {
+                GETGD;
+                UXTH(x14, gd, 0);
+                ed = xEAX + (nextop&7);
+                UXTH(x1, ed, 0);
+                // do the swap 14 -> ed, 1 -> gd
+                BFI(gd, x1, 0, 16);
+                BFI(ed, x14, 0, 16);
+            } else {
+                GETGD;
+                UXTH(x14, gd, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0);
+                #if 0
+                MARKLOCK;
+                // do the swap with exclusive locking
+                LDREXH(x1, ed);
+                // do the swap 14 -> strb(ed), 1 -> gd
+                STREXH(x3, x14, ed);
+                CMPS_IMM8(x3, 0);
+                B_MARKLOCK(cNE);
+                #else
+                LDRH_IMM8(x1, ed, fixedaddress);
+                STRH_IMM8(x14, ed, fixedaddress);
+                #endif
+                BFI(gd, x1, 0, 16);
+            }
            break;

        case 0x89:
@@ -479,7 +537,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    BFI(ed, gd, 0, 16);
                }
            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0);
                STRH_IMM8(gd, ed, fixedaddress);
            }
            break;
@@ -493,7 +551,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    BFI(gd, ed, 0, 16);
                }
            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0);
                LDRH_IMM8(x1, ed, fixedaddress);
                BFI(gd, x1, 0, 16);
            }
@@ -501,13 +559,13 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0x8C:
            INST_NAME("MOV Ew,Seg");
            nextop = F8;
-            MOV32(x2, offsetof(x86emu_t, segs[(nextop&38)>>3]));
+            MOV32(x2, offsetof(x86emu_t, segs[(nextop&0x38)>>3]));
            LDRH_REG(x1, xEmu, x2);
            if((nextop&0xC0)==0xC0) {
                ed = xEAX+(nextop&7);
                BFI(ed, x1, 0, 16);
            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0);
                STRH_IMM8(x1, ed, fixedaddress);
            }
            break;
@@ -517,16 +575,16 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            nextop = F8;
            if((nextop&0xC0)==0xC0) {
                ed = xEAX+(nextop&7);
-                MOV32(x2, offsetof(x86emu_t, segs[(nextop&38)>>3]));
+                MOV32(x2, offsetof(x86emu_t, segs[(nextop&0x38)>>3]));
                STRH_REG(ed, xEmu, x2);
            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0);
                LDRH_IMM8(x1, ed, fixedaddress);
-                MOV32(x2, offsetof(x86emu_t, segs[(nextop&38)>>3]));
+                MOV32(x2, offsetof(x86emu_t, segs[(nextop&0x38)>>3]));
                STRH_REG(x1, xEmu, x2);
            }
            MOVW(x1, 0);
-            STR_IMM9(x1, xEmu, offsetof(x86emu_t, segs_clean[(nextop&38)>>3]));
+            STR_IMM9(x1, xEmu, offsetof(x86emu_t, segs_serial[(nextop&0x38)>>3]));
            break;

        case 0x90:
@@ -545,12 +603,33 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            BFI(xEAX, gd, 0, 16);
            BFI(gd, x2, 0, 16);
            break;
-
        case 0x98:
            INST_NAME("CBW");
            SXTB(x1, xEAX, 0);
            BFI(xEAX, x1, 0, 16);
            break;
+        case 0x99:
+            INST_NAME("CWD");
+            SXTH(x1, xEAX, 0);
+            MOV_REG_LSR_IMM5(x1, x1, 16);
+            BFI(xEDX, x1, 0, 16);
+            break;
+
+        case 0x9C:
+            INST_NAME("PUSHF (16b)");
+            READFLAGS(X_ALL);
+            STRHB_IMM8(xFlags, xESP, -2);
+            break;
+        case 0x9D:
+            INST_NAME("POPF (16b)");
+            SETFLAGS(X_ALL, SF_SET);    // lower 16bits is all flags handled in dynarec
+            LDRHA_IMM8(x2, xESP, 2);
+            MOV32(x1, 0x7FD7);
+            AND_REG_LSL_IMM5(x2, x2, x1, 0);
+            ORR_IMM8(x2, x2, 2, 0);
+            BFI(xFlags, x2, 0, 16);
+            SET_DFNONE(x1);
+            break;

        case 0xA1:
            INST_NAME("MOV, AX, Od");
@@ -573,20 +652,20 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            break;
        case 0xA7:
            INST_NAME("CMPSW");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            GETDIR(x3, 2);
            LDRHAI_REG_LSL_IMM5(x1, xESI, x3);
            LDRHAI_REG_LSL_IMM5(x2, xEDI, x3);
-            CALL(cmp16, -1, 0);
+            emit_cmp16(dyn, ninst, x1, x2, x3, x14);
            break;

        case 0xA9:
            INST_NAME("TEST AX,Iw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
            u16 = F16;
            MOVW(x2, u16);
            UBFX(x1, xEAX, 0, 16);
-            emit_test16(dyn, ninst, x1, x2, x3, x12);
+            emit_test16(dyn, ninst, x1, x2, x3, x14);
            break;

        case 0xAB:
@@ -605,7 +684,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            GETDIR(x3, 2);
            UXTH(x1, xEAX, 0);
            LDRHAI_REG_LSL_IMM5(x2, xEDI, x3);
-            CALL(cmp16, -1, 0);
+            emit_cmp16(dyn, ninst, x1, x2, x3, x14);
            break;

        case 0xB8:
@@ -712,7 +791,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                MOVW(x1, u16);
                BFI(ed, x1, 0, 16);
            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0);
                u16 = F16;
                MOVW(x1, u16);
                STRH_IMM8(x1, ed, fixedaddress);
@@ -771,15 +850,15 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                case 6:
                    if(opcode==0xD1) {
                        INST_NAME("SHL Ew, 1");
-                        MOVW(x12, 1);
+                        MOVW(x14, 1);
                    } else {
                        INST_NAME("SHL Ew, CL");
-                        AND_IMM8(x12, xECX, 0x1f);
+                        AND_IMM8(x14, xECX, 0x1f);
                    }
                    SETFLAGS(X_ALL, SF_PENDING);
                    GETEW(x1);
-                    UFLAG_OP12(ed, x12)
-                    MOV_REG_LSL_REG(ed, ed, x12);
+                    UFLAG_OP12(ed, x14)
+                    MOV_REG_LSL_REG(ed, ed, x14);
                    EWBACK;
                    UFLAG_RES(ed);
                    UFLAG_DF(x3, d_shl16);
@@ -787,15 +866,15 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                case 5:
                    if(opcode==0xD1) {
                        INST_NAME("SHR Ew, 1");
-                        MOVW(x12, 1);
+                        MOVW(x14, 1);
                    } else {
                        INST_NAME("SHR Ew, CL");
-                        AND_IMM8(x12, xECX, 0x1f);
+                        AND_IMM8(x14, xECX, 0x1f);
                    }
                    SETFLAGS(X_ALL, SF_PENDING);
                    GETEW(x1);
-                    UFLAG_OP12(ed, x12)
-                    MOV_REG_LSR_REG(ed, ed, x12);
+                    UFLAG_OP12(ed, x14)
+                    MOV_REG_LSR_REG(ed, ed, x14);
                    EWBACK;
                    UFLAG_RES(ed);
                    UFLAG_DF(x3, d_shr16);
@@ -803,15 +882,15 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                case 7:
                    if(opcode==0xD1) {
                        INST_NAME("SAR Ew, 1");
-                        MOVW(x12, 1);
+                        MOVW(x14, 1);
                    } else {
                        INST_NAME("SAR Ew, CL");
-                        AND_IMM8(x12, xECX, 0x1f);
+                        AND_IMM8(x14, xECX, 0x1f);
                    }
                    SETFLAGS(X_ALL, SF_PENDING);
                    GETSEW(x1);
-                    UFLAG_OP12(ed, x12)
-                    MOV_REG_ASR_REG(ed, ed, x12);
+                    UFLAG_OP12(ed, x14)
+                    MOV_REG_ASR_REG(ed, ed, x14);
                    EWBACK;
                    UFLAG_RES(ed);
                    UFLAG_DF(x3, d_sar16);
@@ -869,7 +948,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    MARK2;
                    SUB_IMM8(xECX, xECX, 1);
                    MARK3;
-                    emit_cmp16(dyn, ninst, x1, x2, x3, x12);
+                    emit_cmp16(dyn, ninst, x1, x2, x3, x14);
                    break;

                case 0xAB:
@@ -905,7 +984,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    MARK2;
                    SUB_IMM8(xECX, xECX, 1);
                    MARK3;
-                    emit_cmp16(dyn, ninst, x1, x2, x3, x12);
+                    emit_cmp16(dyn, ninst, x1, x2, x3, x14);
                    break;
                default:
                    DEFAULT;
@@ -918,11 +997,11 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                case 0:
                case 1:
                    INST_NAME("TEST Ew, Iw");
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    GETEW(x1);
                    u16 = F16;
                    MOVW(x2, u16);
-                    emit_test16(dyn, ninst, x1, x2, x3, x12);
+                    emit_test16(dyn, ninst, x1, x2, x3, x14);
                    break;
                case 2:
                    INST_NAME("NOT Ew");
@@ -932,13 +1011,14 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 3:
                    INST_NAME("NEG Ew");
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                    GETEW(x1);
-                    emit_neg16(dyn, ninst, ed, x2, x12);
+                    emit_neg16(dyn, ninst, ed, x2, x14);
                    EWBACK;
                    break;
                case 4:
                    INST_NAME("MUL AX, Ew");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
                    SETFLAGS(X_ALL, SF_PENDING);
                    GETEW(x1);
                    STM(xEmu, (1<<xEAX) | (1<<xECX) | (1<<xEDX));
@@ -958,14 +1038,27 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 6:
                    INST_NAME("DIV Ew");
-                    SETFLAGS(X_ALL, SF_SET);
-                    GETEW(x1);
-                    STM(xEmu, (1<<xEAX) | (1<<xECX) | (1<<xEDX));
-                    CALL(div16, -1, 0);
-                    LDM(xEmu, (1<<xEAX) | (1<<xECX) | (1<<xEDX));
+                    if(arm_div) {
+                        GETEW(x1);
+                        SETFLAGS(X_ALL, SF_SET);
+                        UXTH(x2, xEAX, 0);
+                        ORR_REG_LSL_IMM5(x2, x2, xEDX, 16);
+                        UDIV(x3, x2, ed);
+                        MLS(x14, x3, ed, x2);  // x14 = x2 mod ed (i.e. x2 - x3*ed)
+                        BFI(xEAX, x3, 1, 16);
+                        BFI(xEDX, x14, 1, 16);
+                    } else {
+                        MESSAGE(LOG_DUMP, "Need Optimization\n");
+                        SETFLAGS(X_ALL, SF_SET);
+                        GETEW(x1);
+                        STM(xEmu, (1<<xEAX) | (1<<xECX) | (1<<xEDX));
+                        CALL(div16, -1, 0);
+                        LDM(xEmu, (1<<xEAX) | (1<<xECX) | (1<<xEDX));
+                    }
                    break;
                case 7:
                    INST_NAME("IDIV Ew");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
                    SETFLAGS(X_ALL, SF_SET);
                    GETEW(x1);
                    STM(xEmu, (1<<xEAX) | (1<<xECX) | (1<<xEDX));
@@ -974,6 +1067,20 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
            }
            break;
+        case 0xF8:
+            INST_NAME("CLC");
+            SETFLAGS(X_CF, SF_SUBSET);
+            SET_DFNONE(x1);
+            BFC(xFlags, F_CF, 1);
+            break;
+        case 0xF9:
+            INST_NAME("STC");
+            SETFLAGS(X_CF, SF_SUBSET);
+            SET_DFNONE(x1);
+            MOVW(x1, 1);
+            BFI(xFlags, x1, F_CF, 1);
+            break;
+
        case 0xFF:
            nextop = F8;
            switch((nextop>>3)&7) {
@@ -981,14 +1088,14 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    INST_NAME("INC Ew");
                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
                    GETEW(x1);
-                    emit_inc16(dyn, ninst, x1, x2, x12);
+                    emit_inc16(dyn, ninst, x1, x2, x14);
                    EWBACK;
                    break;
                case 1:
                    INST_NAME("DEC Ew");
-                    SETFLAGS(X_ALL&~X_CF, SF_SET);
+                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
                    GETEW(x1);
-                    emit_dec16(dyn, ninst, x1, x2, x12);
+                    emit_dec16(dyn, ninst, x1, x2, x14);
                    EWBACK;
                    break;
                default:
@@ -21,26 +21,126 @@

 #include "dynarec_arm_helper.h"

-
 uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog)
 {
-    uint8_t nextop;
+    uint8_t nextop, opcode;
    int8_t  i8;
-    int32_t i32;
+    int32_t i32, j32;
+    uint32_t u32;
+    uint8_t gd, ed;
+    int fixedaddress;
+
    MAYUSE(i32);
+    MAYUSE(j32);
+    
    nextop = F8;
+
    switch(nextop) {
       

-        #define GO(NO, YES)   \
-            BARRIER(2); \
-            JUMP(addr+i8);\
+        case 0x64:
+            // reduced EA size...
+            opcode = F8;
+            switch(opcode) {
+                case 0x89:
+                    INST_NAME("MOV FS:Ew16, Gw");
+                    nextop = F8;
+                    grab_fsdata(dyn, addr, ninst, x14);
+                    GETGD;  // don't need GETGW here
+                    if((nextop&0xC0)==0xC0) {
+                        ed = xEAX+(nextop&7);
+                        if(ed!=gd) {
+                            BFI(ed, gd, 0, 16);
+                        }
+                    } else {
+                        addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
+                        STRH_REG(gd, x14, ed);
+                    }
+                    break;
+
+                case 0x8B:
+                    INST_NAME("MOV Gd, FS:Ew16");
+                    nextop=F8;
+                    grab_fsdata(dyn, addr, ninst, x14);
+                    GETGD;
+                    if((nextop&0xC0)==0xC0) {   // reg <= reg
+                        ed = xEAX+(nextop&7);
+                        if(ed!=gd) {
+                            BFI(gd, ed, 0, 16);
+                        }
+                    } else {                    // mem <= reg
+                        addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
+                        LDR_REG_LSL_IMM5(gd, x14, ed, 0);
+                    }
+                    break;
+
+                case 0x8F:
+                    INST_NAME("POP FS:Ew16");
+                    nextop=F8;
+                    grab_fsdata(dyn, addr, ninst, x14);
+                    if((nextop&0xC0)==0xC0) {
+                        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);
+                    }
+                    break;
+
+                case 0xA1:
+                    INST_NAME("MOV EAX, FS:Ow");
+                    grab_fsdata(dyn, addr, ninst, x1);
+                    u32 = F16;
+                    if(u32) {
+                        MOV32(x2, u32);
+                        ADD_REG_LSL_IMM5(x1, x1, x2, 0);
+                    }
+                    LDR_IMM9(xEAX, x1, 0);
+                    break;
+
+                case 0xA3:
+                    INST_NAME("MOV FS:Ow, EAX");
+                    grab_fsdata(dyn, addr, ninst, x1);
+                    u32 = F16;
+                    if(u32) {
+                        MOV32(x2, u32);
+                        ADD_REG_LSL_IMM5(x1, x1, x2, 0);
+                    }
+                    STR_IMM9(xEAX, x1, 0);
+                    break;
+
+                case 0xFF:
+                    nextop = F8;
+                    grab_fsdata(dyn, addr, ninst, x14);
+                    switch((nextop>>3)&7) {
+                        case 6: // Push Ed
+                            INST_NAME("PUSH FS:Ew");
+                            if((nextop&0xC0)==0xC0) {   // reg
+                                DEFAULT;
+                            } else {                    // mem <= i32
+                                addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
+                                LDR_REG_LSL_IMM5(x3, ed, x14, 0);
+                                PUSH1(x3);
+                            }
+                            break;
+                        default:
+                            DEFAULT;
+                    }
+                    break;
+                default:
+                    DEFAULT;
+            }
+            break;
+            
+        #define GO(NO, YES)     \
+            BARRIER(2);         \
+            JUMP(addr+i8, 1);   \
            if(dyn->insts) {    \
                if(dyn->insts[ninst].x86.jmp_insts==-1) {   \
                    /* out of the block */                  \
                    i32 = dyn->insts[ninst+1].address-(dyn->arm_size+8); \
                    Bcond(NO, i32);     \
-                    jump_to_linker(dyn, addr+i8, 0, ninst); \
+                    jump_to_next(dyn, addr+i8, 0, ninst); \
                } else {    \
                    /* inside the block */  \
                    i32 = dyn->insts[dyn->insts[ninst].x86.jmp_insts].address-(dyn->arm_size+8);    \
@@ -55,8 +155,7 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            SUBS_IMM8(x1, x1, 1);
            BFI(xECX, x1, 0, 16);
            B_NEXT(cEQ);    // CX is 0, no LOOP
-            LDR_IMM9(x1, xEmu, offsetof(x86emu_t, flags[F_ZF]));
-            TSTS_REG_LSL_IMM5(x1, x1, 0);
+            TSTS_IMM8(xFlags, 1<<F_ZF);
            GO(cNE, cEQ);
            break;
        case 0xE1:
@@ -67,8 +166,7 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            SUBS_IMM8(x1, x1, 1);
            BFI(xECX, x1, 0, 16);
            B_NEXT(cEQ);    // CX is 0, no LOOP
-            LDR_IMM9(x1, xEmu, offsetof(x86emu_t, flags[F_ZF]));
-            TSTS_REG_LSL_IMM5(x1, x1, 0);
+            TSTS_IMM8(xFlags, 1<<F_ZF);
            GO(cEQ, cNE);
            break;
        case 0xE2:
@@ -50,9 +50,13 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xC6:
        case 0xC7:
            INST_NAME("FADD ST0, STx");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VADD_F64(v1, v1, v2);
+            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)) {
+                VADD_F32(v1, v1, v2);
+            } else {
+                VADD_F64(v1, v1, v2);
+            }
            break;
        case 0xC8:
        case 0xC9:
@@ -63,9 +67,13 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xCE:
        case 0xCF:
            INST_NAME("FMUL ST0, STx");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VMUL_F64(v1, v1, v2);
+            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)) {
+                VMUL_F32(v1, v1, v2);
+            } else {
+                VMUL_F64(v1, v1, v2);
+            }
            break;
        case 0xD0:
        case 0xD1:
@@ -76,9 +84,13 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xD6:
        case 0xD7:
            INST_NAME("FCOM ST0, STx");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VCMP_F64(v1, v2);
+            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)) {
+                VCMP_F32(v1, v2);
+            } else {
+                VCMP_F64(v1, v2);
+            }
            FCOM(x1, x2);
            break;
        case 0xD8:
@@ -90,11 +102,15 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xDE:
        case 0xDF:
            INST_NAME("FCOMP ST0, STx");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VCMP_F64(v1, v2);
+            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)) {
+                VCMP_F32(v1, v2);
+            } else {
+                VCMP_F64(v1, v2);
+            }
            FCOM(x1, x2);
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
            break;
        case 0xE0:
        case 0xE1:
@@ -105,9 +121,13 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xE6:
        case 0xE7:
            INST_NAME("FSUB ST0, STx");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VSUB_F64(v1, v1, v2);
+            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)) {
+                VSUB_F32(v1, v1, v2);
+            } else {
+                VSUB_F64(v1, v1, v2);
+            }
            break;
        case 0xE8:
        case 0xE9:
@@ -118,9 +138,13 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xEE:
        case 0xEF:
            INST_NAME("FSUBR ST0, STx");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VSUB_F64(v1, v2, v1);
+            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)) {
+                VSUB_F32(v1, v2, v1);
+            } else {
+                VSUB_F64(v1, v2, v1);
+            }
            break;
        case 0xF0:
        case 0xF1:
@@ -131,9 +155,13 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xF6:
        case 0xF7:
            INST_NAME("FDIV ST0, STx");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VDIV_F64(v1, v1, v2);
+            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)) {
+                VDIV_F32(v1, v1, v2);
+            } else {
+                VDIV_F64(v1, v1, v2);
+            }
            break;
        case 0xF8:
        case 0xF9:
@@ -144,143 +172,179 @@ uintptr_t dynarecD8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xFE:
        case 0xFF:
            INST_NAME("FDIVR ST0, STx");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VDIV_F64(v1, v2, v1);
+            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)) {
+                VDIV_F32(v1, v2, v1);
+            } else {
+                VDIV_F64(v1, v2, v1);
+            }
            break;
      
        default:
            switch((nextop>>3)&7) {
                case 0:
                    INST_NAME("FADD ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
                    s0 = fpu_get_scratch_single(dyn);
                    d1 = fpu_get_scratch_double(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0, 0);
                        VLDR_32(s0, ed, fixedaddress);
                    } else {
                        GETED;
                        VMOVtoV(s0, ed);
                    }
-                    VCVT_F64_F32(d1, s0);
-                    VADD_F64(v1, v1, d1);
+                    if(ST_IS_F(0)) {
+                        VADD_F32(v1, v1, s0);
+                    } else {
+                        VCVT_F64_F32(d1, s0);
+                        VADD_F64(v1, v1, d1);
+                    }
                    break;
                case 1:
                    INST_NAME("FMUL ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
                    s0 = fpu_get_scratch_single(dyn);
                    d1 = fpu_get_scratch_double(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0, 0);
                        VLDR_32(s0, ed, fixedaddress);
                    } else {
                        GETED;
                        VMOVtoV(s0, ed);
                    }
-                    VCVT_F64_F32(d1, s0);
-                    VMUL_F64(v1, v1, d1);
+                    if(ST_IS_F(0)) {
+                        VMUL_F32(v1, v1, s0);
+                    } else {
+                        VCVT_F64_F32(d1, s0);
+                        VMUL_F64(v1, v1, d1);
+                    }
                    break;
                case 2:
                    INST_NAME("FCOM ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
                    s0 = fpu_get_scratch_single(dyn);
                    d1 = fpu_get_scratch_double(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0, 0);
                        VLDR_32(s0, ed, fixedaddress);
                    } else {
                        GETED;
                        VMOVtoV(s0, ed);
                    }
-                    VCVT_F64_F32(d1, s0);
-                    VCMP_F64(v1, d1);
+                    if(ST_IS_F(0)) {
+                        VCMP_F32(v1, s0);
+                    } else {
+                        VCVT_F64_F32(d1, s0);
+                        VCMP_F64(v1, d1);
+                    }
                    FCOM(x1, x2);
                    break;
                case 3:
                    INST_NAME("FCOMP ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
                    s0 = fpu_get_scratch_single(dyn);
                    d1 = fpu_get_scratch_double(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0, 0);
                        VLDR_32(s0, ed, fixedaddress);
                    } else {
                        GETED;
                        VMOVtoV(s0, ed);
                    }
-                    VCVT_F64_F32(d1, s0);
-                    VCMP_F64(v1, d1);
+                    if(ST_IS_F(0)) {
+                        VCMP_F32(v1, s0);
+                    } else {
+                        VCVT_F64_F32(d1, s0);
+                        VCMP_F64(v1, d1);
+                    }
                    FCOM(x1, x2);
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                    break;
                case 4:
                    INST_NAME("FSUB ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
                    s0 = fpu_get_scratch_single(dyn);
                    d1 = fpu_get_scratch_double(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0, 0);
                        VLDR_32(s0, ed, fixedaddress);
                    } else {
                        GETED;
                        VMOVtoV(s0, ed);
                    }
-                    VCVT_F64_F32(d1, s0);
-                    VSUB_F64(v1, v1, d1);
+                    if(ST_IS_F(0)) {
+                        VSUB_F32(v1, v1, s0);
+                    } else {
+                        VCVT_F64_F32(d1, s0);
+                        VSUB_F64(v1, v1, d1);
+                    }
                    break;
                case 5:
                    INST_NAME("FSUBR ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
                    s0 = fpu_get_scratch_single(dyn);
                    d1 = fpu_get_scratch_double(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0, 0);
                        VLDR_32(s0, ed, fixedaddress);
                    } else {
                        GETED;
                        VMOVtoV(s0, ed);
                    }
-                    VCVT_F64_F32(d1, s0);
-                    VSUB_F64(v1, d1, v1);
+                    if(ST_IS_F(0)) {
+                        VSUB_F32(v1, s0, v1);
+                    } else {
+                        VCVT_F64_F32(d1, s0);
+                        VSUB_F64(v1, d1, v1);
+                    }
                    break;
                case 6:
                    INST_NAME("FDIV ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
                    s0 = fpu_get_scratch_single(dyn);
                    d1 = fpu_get_scratch_double(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0, 0);
                        VLDR_32(s0, ed, fixedaddress);
                    } else {
                        GETED;
                        VMOVtoV(s0, ed);
                    }
-                    VCVT_F64_F32(d1, s0);
-                    VDIV_F64(v1, v1, d1);
+                    if(ST_IS_F(0)) {
+                        VDIV_F32(v1, v1, s0);
+                    } else {
+                        VCVT_F64_F32(d1, s0);
+                        VDIV_F64(v1, v1, d1);
+                    }
                    break;
                case 7:
                    INST_NAME("FDIVR ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
                    s0 = fpu_get_scratch_single(dyn);
                    d1 = fpu_get_scratch_double(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0, 0);
                        VLDR_32(s0, ed, fixedaddress);
                    } else {
                        GETED;
                        VMOVtoV(s0, ed);
                    }
-                    VCVT_F64_F32(d1, s0);
-                    VDIV_F64(v1, d1, v1);
+                    if(ST_IS_F(0)) {
+                        VDIV_F32(v1, s0, v1);
+                    } else {
+                        VCVT_F64_F32(d1, s0);
+                        VDIV_F64(v1, d1, v1);
+                    }
                    break;
                default:
                    DEFAULT;
@@ -23,13 +23,16 @@
 #include "dynarec_arm_functions.h"
 #include "dynarec_arm_helper.h"

-static double d_1   = 1.0;
-static double d_l2t = L2T;
-static double d_l2e = L2E;
-static double d_pi  = PI;
-static double d_lg2 = LG2;
-static double d_ln2 = LN2;
-static double d_0   = 0.0;
+static const double d_l2t = L2T;
+static const double d_l2e = L2E;
+static const double d_pi  = PI;
+static const double d_lg2 = LG2;
+static const double d_ln2 = LN2;
+static const float  f_l2t = L2T;
+static const float  f_l2e = L2E;
+static const float  f_pi  = PI;
+static const float  f_lg2 = LG2;
+static const float  f_ln2 = LN2;

 uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog)
 {
@@ -40,11 +43,13 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
    int v1, v2;
    int s0;
    int i1, i2, i3;
+    int j32;
    int parity;

    MAYUSE(s0);
    MAYUSE(v2);
    MAYUSE(v1);
+    MAYUSE(j32);

    switch(nextop) {

@@ -57,9 +62,13 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xC6:
        case 0xC7:
            INST_NAME("FLD STx");
-            v1 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            v2 = x87_do_push(dyn, ninst);
-            VMOV_64(v2, v1);
+            v2 = x87_do_push(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);
+            } else {
+                VMOV_64(v2, v1);
+            }
            break;

        case 0xC8:
@@ -72,11 +81,17 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xCF:
            INST_NAME("FXCH STx");
            // swap the cache value, not the double value itself :p
-            i1 = x87_get_cache(dyn, ninst, x1, x2, nextop&7);
-            i2 = x87_get_cache(dyn, ninst, x1, x2, 0);
+            i1 = x87_get_cache(dyn, ninst, 1, x1, x2, nextop&7, X87_COMBINE(0, nextop&7));
+            i2 = x87_get_cache(dyn, ninst, 1, x1, x2, 0, X87_COMBINE(0, nextop&7));
            i3 = dyn->x87cache[i1];
            dyn->x87cache[i1] = dyn->x87cache[i2];
            dyn->x87cache[i2] = i3;
+            // swap those too
+            i1 = x87_get_neoncache(dyn, ninst, x1, x2, nextop&7);
+            i2 = x87_get_neoncache(dyn, ninst, x1, x2, 0);
+            i3 = dyn->n.neoncache[i1].v;
+            dyn->n.neoncache[i1].v = dyn->n.neoncache[i2].v;
+            dyn->n.neoncache[i2].v = i3;
            break;

        case 0xD0:
@@ -85,78 +100,184 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,

        case 0xE0:
            INST_NAME("FCHS");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            VNEG_F64(v1, v1);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+            if(ST_IS_F(0)) {
+                VNEG_F32(v1, v1);
+            } else {
+                VNEG_F64(v1, v1);
+            }
            break;
        case 0xE1:
            INST_NAME("FABS");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            VABS_F64(v1, v1);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+            if(ST_IS_F(0)) {
+                VABS_F32(v1, v1);
+            } else {
+                VABS_F64(v1, v1);
+            }
            break;

        case 0xE4:
            INST_NAME("FTST");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            VCMP_F64_0(v1);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+            if(ST_IS_F(0)) {
+                VCMP_F32_0(v1);
+            } else {
+                VCMP_F64_0(v1);
+            }
            FCOM(x1, x2);   // same flags...
            break;
        case 0xE5:
            INST_NAME("FXAM");
+            #if 1
+            i1 = x87_get_current_cache(dyn, ninst, 0, NEON_CACHE_ST_D);
+            // value put in x14
+            if(i1==-1) {
+                // 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->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, 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->x87reg[i1];
+                VMOVfrV_D(x2, x3, v1);
+            }
+            // get exponant in r1
+            MOV_REG_LSR_IMM5(x1, x3, 20);
+            MOVW(x14, 0x7ff);
+            ANDS_REG_LSL_IMM5(x1, x1, x14, 0);
+            B_MARK(cNE); // not zero or denormal
+            BIC_IMM8_ROR(x1, x3, 0b10, 1); // remove sign bit
+            ORRS_REG_LSL_IMM5(x1, x1, x2, 0);
+            MOVW_COND(cEQ, x14, 0b100000000000000); // Zero: C3,C2,C0 = 100
+            MOVW_COND(cNE, x14, 0b100010000000000); // Denormal: C3,C2,C0 = 110
+            B_MARK3(c__);
+            MARK;
+            CMPS_REG_LSL_IMM5(x1, x14, 0);   // infinite/NaN?
+            MOVW_COND(cNE, x14, 0b000010000000000); // normal: C3,C2,C0 = 010
+            B_MARK3(cNE);
+            ORR_IMM8(x1, x1, 0x08, 12);  //prepare mask, 0x7ff | 0x800 => 0xfff
+            BIC_REG_LSL_IMM5(x1, x3, x1, 20);
+            ORRS_REG_LSL_IMM5(x1, x1, x2, 0);
+            MOVW_COND(cEQ, x14, 0b000010100000000); // infinity: C3,C2,C0 = 011
+            MOVW_COND(cNE, x14, 0b000000100000000); // NaN: C3,C2,C0 = 001
+            MARK3;
+            // Extract signa & Update SW
+            MOV_REG_LSR_IMM5(x1, x3, 31);
+            BFI(x14, x1, 9, 1); //C1
+            LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
+            BIC_IMM8(x1, x1, 0b01000111, 12);
+            ORR_REG_LSL_IMM5(x14, x14, x1, 0);
+            STRH_IMM8(x14, xEmu, offsetof(x86emu_t, sw));
+            #else
+            MESSAGE(LOG_DUMP, "Need Optimization\n");
            x87_refresh(dyn, ninst, x1, x2, 0);
            CALL(fpu_fxam, -1, 0);  // should be possible inline, but is it worth it?
+            #endif
            break;

        case 0xE8:
            INST_NAME("FLD1");
-            v1 = x87_do_push(dyn, ninst);
-            MOV32(x2, (&d_1));
-            VLDR_64(v1, x2, 0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                VMOV_i_32(v1, 0b01110000);
+            } else {
+                VMOV_i_64(v1, 0b01110000);
+            }
            break;
        case 0xE9:
            INST_NAME("FLDL2T");
-            v1 = x87_do_push(dyn, ninst);
-            MOV32(x2, (&d_l2t));
-            VLDR_64(v1, x2, 0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                MOV32(x2, (&f_l2t));
+                VLDR_32(v1, x2, 0);
+            } else {
+                MOV32(x2, (&d_l2t));
+                VLDR_64(v1, x2, 0);
+            }
            break;
        case 0xEA:     
            INST_NAME("FLDL2E");
-            v1 = x87_do_push(dyn, ninst);
-            MOV32(x2, (&d_l2e));
-            VLDR_64(v1, x2, 0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                MOV32(x2, (&f_l2e));
+                VLDR_32(v1, x2, 0);
+            } else {
+                MOV32(x2, (&d_l2e));
+                VLDR_64(v1, x2, 0);
+            }
            break;
        case 0xEB:
            INST_NAME("FLDPI");
-            v1 = x87_do_push(dyn, ninst);
-            MOV32(x2, (&d_pi));
-            VLDR_64(v1, x2, 0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                MOV32(x2, (&f_pi));
+                VLDR_32(v1, x2, 0);
+            } else {
+                MOV32(x2, (&d_pi));
+                VLDR_64(v1, x2, 0);
+            }
            break;
        case 0xEC:
            INST_NAME("FLDLG2");
-            v1 = x87_do_push(dyn, ninst);
-            MOV32(x2, (&d_lg2));
-            VLDR_64(v1, x2, 0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                MOV32(x2, (&f_lg2));
+                VLDR_32(v1, x2, 0);
+            } else {
+                MOV32(x2, (&d_lg2));
+                VLDR_64(v1, x2, 0);
+            }
            break;
        case 0xED:
            INST_NAME("FLDLN2");
-            v1 = x87_do_push(dyn, ninst);
-            MOV32(x2, (&d_ln2));
-            VLDR_64(v1, x2, 0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                MOV32(x2, (&f_ln2));
+                VLDR_32(v1, x2, 0);
+            } else {
+                MOV32(x2, (&d_ln2));
+                VLDR_64(v1, x2, 0);
+            }
            break;
        case 0xEE:
            INST_NAME("FLDZ");
-            v1 = x87_do_push(dyn, ninst);
-            MOV32(x2, (&d_0));
-            VLDR_64(v1, x2, 0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                VMOV_8(v1/2, 0);  // float is *2...
+            } else {
+                VMOV_8(v1, 0);  // cannot use VMOV_i_F64
+            }
            break;

        case 0xFA:
            INST_NAME("FSQRT");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            VSQRT_F64(v1, v1);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+            if(ST_IS_F(0)) {
+                VSQRT_F32(v1, v1);
+            } else {
+                VSQRT_F64(v1, v1);
+            }
            break;

        case 0xFC:
            INST_NAME("FRNDINT");
+            MESSAGE(LOG_DUMP, "Need Optimization\n");
            // use C helper for now, nothing staightforward is available
            x87_forget(dyn, ninst, x1, x2, 0);
            CALL(arm_frndint, -1, 0);
@@ -174,71 +295,142 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            break;
        case 0xF0:
            INST_NAME("F2XM1");
+            MESSAGE(LOG_DUMP, "Need Optimization\n");
            x87_forget(dyn, ninst, x1, x2, 0);
            CALL(arm_f2xm1, -1, 0);
            break;
        case 0xF1:
            INST_NAME("FYL2X");
-            x87_forget(dyn, ninst, x1, x2, 0);
-            x87_forget(dyn, ninst, x1, x2, 1);
-            CALL(arm_fyl2x, -1, 0);
-            x87_do_pop(dyn, 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);
+            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);
            break;
        case 0xF2:
            INST_NAME("FTAN");
-            x87_forget(dyn, ninst, x1, x2, 0);
-            CALL(arm_ftan, -1, 0);
-            v1 = x87_do_push(dyn, ninst);
-            MOV32(x2, (&d_1));
-            VLDR_64(v1, x2, 0);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            VMOV_64(0, v1);    // prepare call to tan
+            CALL_1D(tan, 0);
+            VMOV_64(v1, 0);
+            //emu->sw.f.F87_C2 = 0;
+            LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
+            BFC(x1, 10, 1); //C2 = 0
+            STRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
+            v2 = x87_do_push(dyn, ninst, x3, NEON_CACHE_ST_D);
+            // 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
+            VMOV_i_64(v2, 0b01110000);
            break;
        case 0xF3:
            INST_NAME("FPATAN");
-            x87_forget(dyn, ninst, x1, x2, 0);
-            x87_forget(dyn, ninst, x1, x2, 1);
-            CALL(arm_fpatan, -1, 0);
-            x87_do_pop(dyn, 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);
+            VMOV_64(0, v2);    // prepare call to atan2
+            VMOV_64(1, v1);
+            CALL_2D(atan2, 0);
+            VMOV_64(v2, 0);    //ST(1).d = atan2(ST1.d, ST0.d);
+            x87_do_pop(dyn, ninst, x3);
            break;
        case 0xF4:
            INST_NAME("FXTRACT");
+            MESSAGE(LOG_DUMP, "Need Optimization\n");
            x87_do_push_empty(dyn, ninst, 0);
            x87_forget(dyn, ninst, x1, x2, 1);
            CALL(arm_fxtract, -1, 0);
            break;
+        case 0xF5:
+            INST_NAME("FPREM1");
+            MESSAGE(LOG_DUMP, "Need Optimization\n");
+            x87_forget(dyn, ninst, x1, x2, 0);
+            x87_forget(dyn, ninst, x1, x2, 1);
+            CALL(arm_fprem1, -1, 0);
+            break;
+        case 0xF6:
+            INST_NAME("FDECSTP");
+            fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+            LDR_IMM9(x2, xEmu, offsetof(x86emu_t, top));
+            SUB_IMM8(x2, x2, 1);
+            AND_IMM8(x2, x2, 7);
+            STR_IMM9(x2, xEmu, offsetof(x86emu_t, top));
+            break;
+        case 0xF7:
+            INST_NAME("FINCSTP");
+            fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+            LDR_IMM9(x2, xEmu, offsetof(x86emu_t, top));
+            ADD_IMM8(x2, x2, 1);
+            AND_IMM8(x2, x2, 7);
+            STR_IMM9(x2, xEmu, offsetof(x86emu_t, top));
+            break;
        case 0xF8:
            INST_NAME("FPREM");
+            MESSAGE(LOG_DUMP, "Need Optimization\n");
            x87_forget(dyn, ninst, x1, x2, 0);
            x87_forget(dyn, ninst, x1, x2, 1);
            CALL(arm_fprem, -1, 0);
            break;
        case 0xF9:
            INST_NAME("FYL2XP1");
-            x87_forget(dyn, ninst, x1, x2, 0);
-            x87_forget(dyn, ninst, x1, x2, 1);
-            CALL(arm_fyl2xp1, -1, 0);
-            x87_do_pop(dyn, 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);
+            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);
            break;
        case 0xFB:
            INST_NAME("FSINCOS");
-            x87_do_push_empty(dyn, ninst, 0);
-            x87_forget(dyn, ninst, x1, x2, 1);
-            CALL(arm_fsincos, -1, 0);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            v2 = x87_do_push(dyn, ninst, x3, NEON_CACHE_ST_D);
+            VMOV_64(0, v1);
+            CALL_1D(sin, 0);
+            VSWP(v1, 0);
+            CALL_1D(cos, 0);
+            VMOV_64(v2, 0);
+            //emu->sw.f.F87_C2 = 0;
+            LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
+            BFC(x1, 10, 1); //C2 = 0
+            STRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
            break;
        case 0xFD:
            INST_NAME("FSCALE");
-            x87_forget(dyn, ninst, x1, x2, 0);
-            x87_forget(dyn, ninst, x1, x2, 1);
-            CALL(arm_fscale, -1, 0);
+            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);
+            VMOV_64(0, v2);
+            CALL_1DD(trunc, exp2, 0);
+            VMUL_F64(v1, v1, 0);
            break;
        case 0xFE:
            INST_NAME("FSIN");
-            x87_forget(dyn, ninst, x1, x2, 0);
-            CALL(arm_fsin, -1, 0);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            VMOV_64(0, v1);    // prepare call to sin
+            CALL_1D(sin, 0);
+            VMOV_64(v1, 0);
+            //emu->sw.f.F87_C2 = 0;
+            LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
+            BFC(x1, 10, 1); //C2 = 0
+            STRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
            break;
        case 0xFF:
            INST_NAME("FCOS");
-            x87_forget(dyn, ninst, x1, x2, 0);
-            CALL(arm_fcos, -1, 0);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            VMOV_64(0, v1);    // prepare call to cos
+            CALL_1D(cos, 0);
+            VMOV_64(v1, 0);
+            //emu->sw.f.F87_C2 = 0;
+            LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
+            BFC(x1, 10, 1); //C2 = 0
+            STRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
            break;


@@ -260,9 +452,6 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xE6:
        case 0xE7:
        case 0xEF:
-        case 0xF5:
-        case 0xF6:
-        case 0xF7:
            DEFAULT;
            break;
             
@@ -270,54 +459,68 @@ 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);
-                    s0 = fpu_get_scratch_single(dyn);
+                    v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+                    if(ST_IS_F(0))
+                        s0 = v1;
+                    else
+                        s0 = fpu_get_scratch_single(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0, 0);
                        VLDR_32(s0, ed, fixedaddress);
                    } else {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0, 0);
                        LDR_IMM9(x2, ed, fixedaddress);
                        VMOVtoV(s0, x2);
                    }
-                    VCVT_F64_F32(v1, s0);
+                    if(!ST_IS_F(0)) {
+                        VCVT_F64_F32(v1, s0);
+                    }
                    break;
                case 2:
                    INST_NAME("FST float[ED], ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-                    s0 = fpu_get_scratch_single(dyn);
-                    VCVT_F32_F64(s0, v1);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                    if(ST_IS_F(0))
+                        s0 = v1;
+                    else {
+                        s0 = fpu_get_scratch_single(dyn);
+                        VCVT_F32_F64(s0, v1);
+                    }
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0, 0);
                        VSTR_32(s0, ed, fixedaddress);
                    } else {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0, 0);
                        VMOVfrV(x2, s0);
                        STR_IMM9(x2, ed, fixedaddress);
                    }
                    break;
                case 3:
                    INST_NAME("FSTP float[ED], ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-                    s0 = fpu_get_scratch_single(dyn);
-                    VCVT_F32_F64(s0, v1);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                    if(ST_IS_F(0))
+                        s0 = v1;
+                    else {
+                        s0 = fpu_get_scratch_single(dyn);
+                        VCVT_F32_F64(s0, v1);
+                    }
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0, 0);
                        VSTR_32(s0, ed, fixedaddress);
                    } else {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0, 0);
                        VMOVfrV(x2, s0);
                        STR_IMM9(x2, ed, fixedaddress);
                    }
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                    break;
                case 4:
                    INST_NAME("FLDENV Ed");
-                    fpu_purgecache(dyn, ninst, x1, x2, x3); // maybe only x87, not SSE?
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    fpu_purgecache(dyn, ninst, 0, x1, x2, x3); // maybe only x87, not SSE?
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
                    if(ed!=x1) {
                        MOV_REG(x1, ed);
                    }
@@ -331,9 +534,20 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    UBFX(x1, x1, 10, 2);    // extract round
                    STR_IMM9(x1, xEmu, offsetof(x86emu_t, round));
                    break;
+                case 6:
+                    INST_NAME("FNSTENV Ed");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    fpu_purgecache(dyn, ninst, 0, x1, x2, x3); // maybe only x87, not SSE?
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
+                    if(ed!=x1) {
+                        MOV_REG(x1, ed);
+                    }
+                    MOVW(x2, 0);
+                    CALL(fpu_savenv, -1, 0);
+                    break;
                case 7:
                    INST_NAME("FNSTCW Ew");
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0);
                    ed = x1;
                    wb1 = 1;
                    LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, cw));
@@ -26,7 +26,7 @@
 uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog)
 {
    uint8_t nextop = F8;
-    int32_t i32;
+    int32_t j32;
    uint8_t ed;
    uint8_t wback;
    int v1, v2;
@@ -39,7 +39,7 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
    MAYUSE(v2);
    MAYUSE(v1);
    MAYUSE(ed);
-    MAYUSE(i32);
+    MAYUSE(j32);

    switch(nextop) {
        case 0xC0:
@@ -52,11 +52,14 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xC7:
            INST_NAME("FCMOVB ST0, STx");
            READFLAGS(X_CF);
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            LDR_IMM9(x1, xEmu, offsetof(x86emu_t, flags[F_CF]));
-            TSTS_REG_LSL_IMM5(x1, x1, 0);
-            VMOVcond_64(cNE, v1, v2);   // F_CF==1
+            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));
+            TSTS_IMM8(xFlags, 1<<F_CF);
+            if(ST_IS_F(0)) {
+                VMOVcond_32(cNE, v1, v2);   // F_CF==1
+            } else {
+                VMOVcond_64(cNE, v1, v2);   // F_CF==1
+            }
            break;
        case 0xC8:
        case 0xC9:
@@ -68,11 +71,14 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xCF:
            INST_NAME("FCMOVE ST0, STx");
            READFLAGS(X_ZF);
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            LDR_IMM9(x1, xEmu, offsetof(x86emu_t, flags[F_ZF]));
-            TSTS_REG_LSL_IMM5(x1, x1, 0);
-            VMOVcond_64(cNE, v1, v2);   // F_ZF==1
+            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));
+            TSTS_IMM8(xFlags, 1<<F_ZF);
+            if(ST_IS_F(0)) {
+                VMOVcond_32(cNE, v1, v2);   // F_ZF==1
+            } else {
+                VMOVcond_64(cNE, v1, v2);   // F_ZF==1
+            }
            break;
        case 0xD0:
        case 0xD1:
@@ -84,12 +90,14 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xD7:
            INST_NAME("FCMOVBE ST0, STx");
            READFLAGS(X_CF|X_ZF);
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            LDR_IMM9(x1, xEmu, offsetof(x86emu_t, flags[F_ZF]));
-            LDR_IMM9(x2, xEmu, offsetof(x86emu_t, flags[F_CF]));
-            ORRS_REG_LSL_IMM5(x1, x1, x2, 0);
-            VMOVcond_64(cNE, v1, v2);   // F_CF==1 | F_ZF==1
+            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));
+            TSTS_IMM8(xFlags, (1<<F_CF)|(1<<F_ZF));
+            if(ST_IS_F(0)) {
+                VMOVcond_32(cNE, v1, v2);   // F_CF==1 | F_ZF==1
+            } else {
+                VMOVcond_64(cNE, v1, v2);   // F_CF==1 | F_ZF==1
+            }
            break;
        case 0xD8:
        case 0xD9:
@@ -101,20 +109,27 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xDF:
            INST_NAME("FCMOVU ST0, STx");
            READFLAGS(X_PF);
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            LDR_IMM9(x1, xEmu, offsetof(x86emu_t, flags[F_PF]));
-            TSTS_REG_LSL_IMM5(x1, x1, 0);
-            VMOVcond_64(cNE, v1, v2);   // F_PF==1
+            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));
+            TSTS_IMM8(xFlags, 1<<F_PF);
+            if(ST_IS_F(0)) {
+                VMOVcond_32(cNE, v1, v2);   // F_PF==1
+            } else {
+                VMOVcond_64(cNE, v1, v2);   // F_PF==1
+            }
            break;       
        case 0xE9:
            INST_NAME("FUCOMPP ST0, ST1");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, 1);
-            VCMP_F64(v1, v2);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
+            v2 = x87_get_st(dyn, ninst, x1, x2, 1, X87_COMBINE(0, nextop&7));
+            if(ST_IS_F(0)) {
+                VCMP_F32(v1, v2);
+            } else {
+                VCMP_F64(v1, v2);
+            }
            FCOM(x1, x2);
-            x87_do_pop(dyn, ninst);
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
+            x87_do_pop(dyn, ninst, x3);
            break;

        case 0xE4:
@@ -134,7 +149,7 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
            switch((nextop>>3)&7) {
                case 0:
                    INST_NAME("FIADD ST0, Ed");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    GETED;
                    d0 = fpu_get_scratch_double(dyn);
                    s0 = fpu_get_scratch_single(dyn);
@@ -144,7 +159,7 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 1:
                    INST_NAME("FIMUL ST0, Ed");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    GETED;
                    d0 = fpu_get_scratch_double(dyn);
                    s0 = fpu_get_scratch_single(dyn);
@@ -154,7 +169,7 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 2:
                    INST_NAME("FICOM ST0, Ed");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    GETED;
                    d0 = fpu_get_scratch_double(dyn);
                    s0 = fpu_get_scratch_single(dyn);
@@ -165,7 +180,7 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 3:
                    INST_NAME("FICOMP ST0, Ed");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    GETED;
                    d0 = fpu_get_scratch_double(dyn);
                    s0 = fpu_get_scratch_single(dyn);
@@ -173,11 +188,11 @@ 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);
+                    x87_do_pop(dyn, ninst, x3);
                    break;
                case 4:
                    INST_NAME("FISUB ST0, Ed");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    GETED;
                    d0 = fpu_get_scratch_double(dyn);
                    s0 = fpu_get_scratch_single(dyn);
@@ -187,7 +202,7 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 5:
                    INST_NAME("FISUBR ST0, Ed");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    GETED;
                    d0 = fpu_get_scratch_double(dyn);
                    s0 = fpu_get_scratch_single(dyn);
@@ -197,7 +212,7 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 6:
                    INST_NAME("FIDIV ST0, Ed");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    GETED;
                    d0 = fpu_get_scratch_double(dyn);
                    s0 = fpu_get_scratch_single(dyn);
@@ -207,7 +222,7 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 7:
                    INST_NAME("FIDIVR ST0, Ed");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    GETED;
                    d0 = fpu_get_scratch_double(dyn);
                    s0 = fpu_get_scratch_single(dyn);
@@ -26,7 +26,7 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
 {
    uint8_t nextop = F8;
    uint8_t u8;
-    int32_t i32;
+    int32_t j32;
    uint8_t ed;
    uint8_t wback;
    int v1, v2;
@@ -37,7 +37,7 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
    MAYUSE(s0);
    MAYUSE(v1);
    MAYUSE(v2);
-    MAYUSE(i32);
+    MAYUSE(j32);

    switch(nextop) {
        case 0xC0:
@@ -50,11 +50,14 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xC7:
            INST_NAME("FCMOVNB ST0, STx");
            READFLAGS(X_CF);
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            LDR_IMM9(x1, xEmu, offsetof(x86emu_t, flags[F_CF]));
-            TSTS_REG_LSL_IMM5(x1, x1, 0);
-            VMOVcond_64(cEQ, v1, v2);   // F_CF==0
+            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));
+            TSTS_IMM8(xFlags, 1<<F_CF);
+            if(ST_IS_F(0)) {
+                VMOVcond_32(cEQ, v1, v2);   // F_CF==0
+            } else {
+                VMOVcond_64(cEQ, v1, v2);   // F_CF==0
+            }
            break;
        case 0xC8:
        case 0xC9:
@@ -66,11 +69,14 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xCF:
            INST_NAME("FCMOVNE ST0, STx");
            READFLAGS(X_ZF);
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            LDR_IMM9(x1, xEmu, offsetof(x86emu_t, flags[F_ZF]));
-            TSTS_REG_LSL_IMM5(x1, x1, 0);
-            VMOVcond_64(cEQ, v1, v2);   // F_ZF==0
+            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));
+            TSTS_IMM8(xFlags, 1<<F_ZF);
+            if(ST_IS_F(0)) {
+                VMOVcond_32(cEQ, v1, v2);   // F_ZF==0
+            } else {
+                VMOVcond_64(cEQ, v1, v2);   // F_ZF==0
+            }
            break;
        case 0xD0:
        case 0xD1:
@@ -82,12 +88,14 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xD7:
            INST_NAME("FCMOVNBE ST0, STx");
            READFLAGS(X_CF|X_ZF);
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            LDR_IMM9(x1, xEmu, offsetof(x86emu_t, flags[F_ZF]));
-            LDR_IMM9(x2, xEmu, offsetof(x86emu_t, flags[F_CF]));
-            ORRS_REG_LSL_IMM5(x1, x1, x2, 0);
-            VMOVcond_64(cEQ, v1, v2);   // F_CF==0 | F_ZF==0
+            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));
+            TSTS_IMM8(xFlags, (1<<F_CF)|(1<<F_ZF));
+            if(ST_IS_F(0)) {
+                VMOVcond_32(cEQ, v1, v2);   // F_CF==0 & F_ZF==0
+            } else {
+                VMOVcond_64(cEQ, v1, v2);   // F_CF==0 & F_ZF==0
+            }
            break;
        case 0xD8:
        case 0xD9:
@@ -99,23 +107,30 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xDF:
            INST_NAME("FCMOVNU ST0, STx");
            READFLAGS(X_PF);
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            LDR_IMM9(x1, xEmu, offsetof(x86emu_t, flags[F_PF]));
-            TSTS_REG_LSL_IMM5(x1, x1, 0);
-            VMOVcond_64(cEQ, v1, v2);   // F_PF==0
+            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));
+            TSTS_IMM8(xFlags, 1<<F_PF);
+            if(ST_IS_F(0)) {
+                VMOVcond_32(cEQ, v1, v2);   // F_PF==0
+            } else {
+                VMOVcond_64(cEQ, v1, v2);   // F_PF==0
+            }
+            break;
+        case 0xE1:
+            INST_NAME("FDISI8087_NOP"); // so.. NOP?
            break;
        case 0xE2:
            INST_NAME("FNCLEX");
-            MOVW(x1, 0);
            LDRH_IMM8(x2, xEmu, offsetof(x86emu_t, sw));
+            MOVW(x1, 0);
            BFI(x2, x1, 0, 8);  // IE .. PE, SF, ES
            BFI(x2, x1, 15, 1); // B
            STRH_IMM8(x2, xEmu, offsetof(x86emu_t, sw));
            break;
        case 0xE3:
            INST_NAME("FNINIT");
-            fpu_purgecache(dyn, ninst, x1, x2, x3); // maybe only x87, not SSE?
+            MESSAGE(LOG_DUMP, "Need Optimization\n");
+            fpu_purgecache(dyn, ninst, 0, x1, x2, x3); // maybe only x87, not SSE?
            CALL(reset_fpu, -1, 0);
            break;
        case 0xE8:
@@ -128,9 +143,13 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xEF:
            INST_NAME("FUCOMI ST0, STx");
            SETFLAGS(X_ALL, SF_SET);
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VCMP_F64(v1, v2);
+            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)) {
+                VCMP_F32(v1, v2);
+            } else {
+                VCMP_F64(v1, v2);
+            }
            FCOMI(x1, x2);
            break;
        case 0xF0:  
@@ -143,14 +162,17 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xF7:
            INST_NAME("FCOMI ST0, STx");
            SETFLAGS(X_ALL, SF_SET);
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VCMP_F64(v1, v2);
+            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)) {
+                VCMP_F32(v1, v2);
+            } else {
+                VCMP_F64(v1, v2);
+            }
            FCOMI(x1, x2);
            break;

        case 0xE0:
-        case 0xE1:
        case 0xE4:
        case 0xE5:
        case 0xE6:
@@ -162,11 +184,11 @@ 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);
+                    v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
                    s0 = fpu_get_scratch_single(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 1023, 0, 0);
                        VLDR_32(s0, ed, fixedaddress);
                    } else {
                        GETED;
@@ -176,18 +198,18 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 1:
                    INST_NAME("FISTTP Ed, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    if((nextop&0xC0)==0xC0) {
                        ed = xEAX+(nextop&7);
                        wback = 0;
                    } else {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0, 0);
                        ed = x1;
                    }
                    s0 = fpu_get_scratch_single(dyn);
                    MSR_nzcvq_0();
-                    VMRS(x12);   // Get FPCSR reg to clear exceptions flags
-                    ORR_IMM8(x3, x12, 0b001, 6); // enable exceptions
+                    VMRS(x14);   // Get FPCSR reg to clear exceptions flags
+                    ORR_IMM8(x3, x14, 0b001, 6); // enable exceptions
                    BIC_IMM8(x3, x3, 0b10011111, 0);
                    VMSR(x3);
                    VCVT_S32_F64(s0, v1);
@@ -196,18 +218,18 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    TSTS_IMM8_ROR(x3, 0b00000001, 0);
                    MOV_IMM_COND(cNE, ed, 0b10, 1);   // 0x80000000
                    WBACK;
-                    VMSR(x12);  // put back values
-                    x87_do_pop(dyn, ninst);
+                    VMSR(x14);  // put back values
+                    x87_do_pop(dyn, ninst, x3);
                    break;
                case 2:
                    INST_NAME("FIST Ed, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-                    u8 = x87_setround(dyn, ninst, x1, x2, x12); // x1 have the modified RPSCR reg
+                    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
                    if((nextop&0xC0)==0xC0) {
                        ed = xEAX+(nextop&7);
                        wback = 0;
                    } else {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0, 0);
                        ed = x1;
                    }
                    s0 = fpu_get_scratch_single(dyn);
@@ -226,13 +248,13 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 3:
                    INST_NAME("FISTP Ed, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-                    u8 = x87_setround(dyn, ninst, x1, x2, x12); // x1 have the modified RPSCR reg
+                    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
                    if((nextop&0xC0)==0xC0) {
                        ed = xEAX+(nextop&7);
                        wback = 0;
                    } else {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0, 0);
                        ed = x1;
                    }
                    s0 = fpu_get_scratch_single(dyn);
@@ -247,27 +269,44 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    TSTS_IMM8_ROR(x3, 0b00000001, 0);
                    MOV_IMM_COND(cNE, ed, 0b10, 1);   // 0x80000000
                    WBACK;
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                    x87_restoreround(dyn, ninst, u8);
                    break;
                case 5:
                    INST_NAME("FLD tbyte");
-                    x87_do_push_empty(dyn, ninst, x1);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
-                    if(ed!=x1) {
-                        MOV_REG(x1, ed);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
+                    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);
+                        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);
+                        STR_IMM9(x2, ed, 0);
+                        STR_IMM9(x3, ed, 4);
+                        STRH_IMM8(x14, ed, 8);
+                    } else {
+                        MESSAGE(LOG_DUMP, "Need Optimization\n");
+                        if(ed!=x1) {
+                            MOV_REG(x1, ed);
+                        }
+                        x87_do_push_empty(dyn, ninst, x3);
+                        CALL(arm_fld, -1, 0);
                    }
-                    CALL(arm_fld, -1, 0);
                    break;
                case 7:
                    INST_NAME("FSTP tbyte");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
                    x87_forget(dyn, ninst, x1, x3, 0);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
                    if(ed!=x1) {
                        MOV_REG(x1, ed);
                    }
                    CALL(arm_fstp, -1, 0);
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                    break;
                default:
                    DEFAULT;
@@ -46,9 +46,13 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xC6:
        case 0xC7:
            INST_NAME("FADD STx, ST0");
-            v2 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v1 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VADD_F64(v1, v1, v2);
+            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(ST_IS_F(0)) {
+                VADD_F32(v1, v1, v2);
+            } else {
+                VADD_F64(v1, v1, v2);
+            }
            break;
        case 0xC8:
        case 0xC9:
@@ -59,9 +63,13 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xCE:
        case 0xCF:
            INST_NAME("FMUL STx, ST0");
-            v2 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v1 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VMUL_F64(v1, v1, v2);
+            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(ST_IS_F(0)) {
+                VMUL_F32(v1, v1, v2);
+            } else {
+                VMUL_F64(v1, v1, v2);
+            }
            break;
        case 0xD0:
        case 0xD1:
@@ -72,9 +80,13 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xD6:
        case 0xD7:
            INST_NAME("FCOM ST0, STx"); //yep
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VCMP_F64(v1, v2);
+            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)) {
+                VCMP_F32(v1, v2);
+            } else {
+                VCMP_F64(v1, v2);
+            }
            FCOM(x1, x2);
            break;
        case 0xD8:
@@ -86,11 +98,15 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xDE:
        case 0xDF:
            INST_NAME("FCOMP ST0, STx");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VCMP_F64(v1, v2);
+            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)) {
+                VCMP_F32(v1, v2);
+            } else {
+                VCMP_F64(v1, v2);
+            }
            FCOM(x1, x2);
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
            break;
        case 0xE0:
        case 0xE1:
@@ -101,9 +117,13 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xE6:
        case 0xE7:
            INST_NAME("FSUBR STx, ST0");
-            v2 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v1 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VSUB_F64(v1, v2, v1);
+            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(ST_IS_F(0)) {
+                VSUB_F32(v1, v2, v1);
+            } else {
+                VSUB_F64(v1, v2, v1);
+            }
            break;
        case 0xE8:
        case 0xE9:
@@ -114,9 +134,13 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xEE:
        case 0xEF:
            INST_NAME("FSUB STx, ST0");
-            v2 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v1 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VSUB_F64(v1, v1, v2);
+            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(ST_IS_F(0)) {
+                VSUB_F32(v1, v1, v2);
+            } else {
+                VSUB_F64(v1, v1, v2);
+            }
            break;
        case 0xF0:
        case 0xF1:
@@ -127,9 +151,13 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xF6:
        case 0xF7:
            INST_NAME("FDIVR STx, ST0");
-            v2 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v1 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VDIV_F64(v1, v2, v1);
+            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(ST_IS_F(0)) {
+                VDIV_F32(v1, v2, v1);
+            } else {
+                VDIV_F64(v1, v2, v1);
+            }
            break;       
        case 0xF8:
        case 0xF9:
@@ -140,22 +168,26 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xFE:
        case 0xFF:
            INST_NAME("FDIV STx, ST0");
-            v2 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v1 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VDIV_F64(v1, v1, v2);
+            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(ST_IS_F(0)) {
+                VDIV_F32(v1, v1, v2);
+            } else {
+                VDIV_F64(v1, v1, v2);
+            }
            break;
        default:
            switch((nextop>>3)&7) {
                case 0:
                    INST_NAME("FADD ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    d1 = fpu_get_scratch_double(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 1023, 3);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 1023, 3, 0);
                        VLDR_64(d1, wback, fixedaddress);
                    } else {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 4095-4, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 4095-4, 0, 0);
                        LDR_IMM9(x2, wback, fixedaddress);
                        LDR_IMM9(x3, wback, fixedaddress+4);
                        VMOVtoV_D(d1, x2, x3);
@@ -164,14 +196,14 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 1:
                    INST_NAME("FMUL ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    d1 = fpu_get_scratch_double(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 1023, 3);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 1023, 3, 0);
                        VLDR_64(d1, wback, fixedaddress);
                    } else {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 4095-4, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 4095-4, 0, 0);
                        LDR_IMM9(x2, wback, fixedaddress);
                        LDR_IMM9(x3, wback, fixedaddress+4);
                        VMOVtoV_D(d1, x2, x3);
@@ -180,14 +212,14 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 2:
                    INST_NAME("FCOM ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    d1 = fpu_get_scratch_double(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 1023, 3);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 1023, 3, 0);
                        VLDR_64(d1, wback, fixedaddress);
                    } else {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 4095-4, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 4095-4, 0, 0);
                        LDR_IMM9(x2, wback, fixedaddress);
                        LDR_IMM9(x3, wback, fixedaddress+4);
                        VMOVtoV_D(d1, x2, x3);
@@ -197,32 +229,32 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 3:
                    INST_NAME("FCOMP ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    d1 = fpu_get_scratch_double(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 1023, 3);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 1023, 3, 0);
                        VLDR_64(d1, wback, fixedaddress);
                    } else {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 4095-4, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 4095-4, 0, 0);
                        LDR_IMM9(x2, wback, fixedaddress);
                        LDR_IMM9(x3, wback, fixedaddress+4);
                        VMOVtoV_D(d1, x2, x3);
                    }
                    VCMP_F64(v1, d1);
                    FCOM(x1, x2);
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                    break;
                case 4:
                    INST_NAME("FSUB ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    d1 = fpu_get_scratch_double(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 1023, 3);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 1023, 3, 0);
                        VLDR_64(d1, wback, fixedaddress);
                    } else {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 4095-4, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 4095-4, 0, 0);
                        LDR_IMM9(x2, wback, fixedaddress);
                        LDR_IMM9(x3, wback, fixedaddress+4);
                        VMOVtoV_D(d1, x2, x3);
@@ -231,14 +263,14 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 5:
                    INST_NAME("FSUBR ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    d1 = fpu_get_scratch_double(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 1023, 3);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 1023, 3, 0);
                        VLDR_64(d1, wback, fixedaddress);
                    } else {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 4095-4, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 4095-4, 0, 0);
                        LDR_IMM9(x2, wback, fixedaddress);
                        LDR_IMM9(x3, wback, fixedaddress+4);
                        VMOVtoV_D(d1, x2, x3);
@@ -247,14 +279,14 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 6:
                    INST_NAME("FDIV ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    d1 = fpu_get_scratch_double(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 1023, 3);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 1023, 3, 0);
                        VLDR_64(d1, wback, fixedaddress);
                    } else {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 4095-4, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 4095-4, 0, 0);
                        LDR_IMM9(x2, wback, fixedaddress);
                        LDR_IMM9(x3, wback, fixedaddress+4);
                        VMOVtoV_D(d1, x2, x3);
@@ -263,14 +295,14 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 7:
                    INST_NAME("FDIVR ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                    d1 = fpu_get_scratch_double(dyn);
                    parity = getedparity(dyn, ninst, addr, nextop, 2);
                    if(parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 1023, 3);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 1023, 3, 0);
                        VLDR_64(d1, wback, fixedaddress);
                    } else {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 4095-4, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 4095-4, 0, 0);
                        LDR_IMM9(x2, wback, fixedaddress);
                        LDR_IMM9(x3, wback, fixedaddress+4);
                        VMOVtoV_D(d1, x2, x3);
@@ -31,6 +31,7 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
    int parity;
    uint8_t ed;
    int v1, v2;
+    int i1, i2, i3;

    MAYUSE(v2);
    MAYUSE(v1);
@@ -45,7 +46,10 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xC6:
        case 0xC7:
            INST_NAME("FFREE STx");
-            break;  // not handling Tag...
+            x87_purgecache(dyn, ninst, 0, x1, x2, x3);
+            MOVW(x1, nextop&7);
+            CALL(fpu_do_free, -1, 0);
+            break;
        case 0xD0:
        case 0xD1:
        case 0xD2:
@@ -55,13 +59,17 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xD6:
        case 0xD7:
            INST_NAME("FST ST0, STx");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VMOV_64(v2, v1);
+            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)) {
+                VMOV_32(v2, v1);
+            } else {
+                VMOV_64(v2, v1);
+            }
            break;
        case 0xD8:
            INST_NAME("FSTP ST0, ST0");
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
            break;
        case 0xD9:
        case 0xDA:
@@ -71,10 +79,19 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xDE:
        case 0xDF:
            INST_NAME("FSTP ST0, STx");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VMOV_64(v2, v1);
-            x87_do_pop(dyn, ninst);
+            // copy the cache value for st0 to stx
+            i1 = x87_get_cache(dyn, ninst, 0, x1, x2, nextop&7, X87_COMBINE(0, nextop&7));
+            i2 = x87_get_cache(dyn, ninst, 1, x1, x2, 0, X87_COMBINE(0, nextop&7));
+            i3 = dyn->x87cache[i1];
+            dyn->x87cache[i1] = dyn->x87cache[i2];
+            dyn->x87cache[i2] = i3;
+            // swap those too
+            i1 = x87_get_neoncache(dyn, ninst, x1, x2, nextop&7);
+            i2 = x87_get_neoncache(dyn, ninst, x1, x2, 0);
+            i3 = dyn->n.neoncache[i1].v;
+            dyn->n.neoncache[i1].v = dyn->n.neoncache[i2].v;
+            dyn->n.neoncache[i2].v = i3;
+            x87_do_pop(dyn, ninst, x3);
            break;

        case 0xE0:
@@ -86,9 +103,13 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xE6:
        case 0xE7:
            INST_NAME("FUCOM ST0, STx");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VCMP_F64(v1, v2);
+            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)) {
+                VCMP_F32(v1, v2);
+            } else {
+                VCMP_F64(v1, v2);
+            }
            FCOM(x1, x2);
            break;
        case 0xE8:
@@ -100,11 +121,15 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xEE:
        case 0xEF:
            INST_NAME("FUCOMP ST0, STx");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VCMP_F64(v1, v2);
+            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)) {
+                VCMP_F32(v1, v2);
+            } else {
+                VCMP_F64(v1, v2);
+            }
            FCOM(x1, x2);
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
            break;

        case 0xC8:
@@ -138,41 +163,35 @@ 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);
-                    #if 0
-                    // can bus error...
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3);
-                    VLDR_64(v1, ed, fixedaddress);
-                    #else
-		    parity = getedparity(dyn, ninst, addr, nextop, 3);
-		    if (parity) {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3);
+                    v1 = x87_do_push(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);
                        VLDR_64(v1, ed, fixedaddress);
-		    } else {
-			addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0);
-			LDR_IMM9(x2, ed, fixedaddress);
-			LDR_IMM9(x3, ed, fixedaddress + 4);
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0);
+                        LDR_IMM9(x2, ed, fixedaddress);
+                        LDR_IMM9(x3, ed, fixedaddress + 4);
                        VMOVtoV_D(v1, x2, x3);
-		    }
-                    #endif
+        		    }
                    break;
                case 1:
                    INST_NAME("FISTTP i64, ST0");
                    x87_forget(dyn, ninst, x1, x2, 0);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
                    if(ed!=x1) {MOV_REG(x1, ed);}
-                    CALL(arm_fistp64, -1, 0);
-                    x87_do_pop(dyn, ninst);
+                    CALL(arm_fistt64, -1, 0);
+                    x87_do_pop(dyn, ninst, x3);
                    break;
                case 2:
                    INST_NAME("FST double");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    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);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0);
                        VSTR_64(v1, ed, fixedaddress);
                    } else {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0);
                        VMOVfrV_D(x2, x3, v1);
                        STR_IMM9(x2, ed, fixedaddress);
                        STR_IMM9(x3, ed, fixedaddress+4);
@@ -180,33 +199,50 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                    break;
                case 3:
                    INST_NAME("FSTP double");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    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);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0);
                        VSTR_64(v1, ed, fixedaddress);
                    } else {
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0);
                        VMOVfrV_D(x2, x3, v1);
                        STR_IMM9(x2, ed, fixedaddress);
                        STR_IMM9(x3, ed, fixedaddress+4);
                    }
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                    break;
                case 4: 
                    INST_NAME("FRSTOR m108byte");
-                    fpu_purgecache(dyn, ninst, x1, x2, x3);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                    fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
                    if(ed!=x1) {MOV_REG(x1, ed);}
                    CALL(arm_frstor, -1, 0);
                    break;
                case 6: 
                    INST_NAME("FSAVE m108byte");
-                    fpu_purgecache(dyn, ninst, x1, x2, x3);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                    fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
                    if(ed!=x1) {MOV_REG(x1, ed);}
                    CALL(arm_fsave, -1, 0);
                    break;
+                case 7:
+                    INST_NAME("FNSTSW m2byte");
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x14, &fixedaddress, 0, 0, 0);
+                    LDR_IMM9(x2, xEmu, offsetof(x86emu_t, top));
+                    LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
+                    if(dyn->x87stack) {
+                        // update top
+                        if(dyn->x87stack>0) {
+                            SUB_IMM8(x2, x2, dyn->x87stack);
+                        } else {
+                            ADD_IMM8(x2, x2, -dyn->x87stack);
+                        }
+                        AND_IMM8(x2, x2, 7);
+                    }
+                    BFI(x1, x2, 11, 3); // inject top
+                    STRH_IMM8(x1, ed, 0);   // store whole sw flags
+                    break;
                default:
                    DEFAULT;
            }
@@ -17,6 +17,7 @@
 #include "x86trace.h"
 #include "dynarec_arm.h"
 #include "dynarec_arm_private.h"
+#include "dynarec_arm_functions.h"
 #include "arm_printer.h"

 #include "dynarec_arm_helper.h"
@@ -40,10 +41,14 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xC6:
        case 0xC7:
            INST_NAME("FADDP STx, ST0");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VADD_F64(v2, v2, v1);
-            x87_do_pop(dyn, ninst);
+            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)) {
+                VADD_F32(v2, v2, v1);
+            } else {
+                VADD_F64(v2, v2, v1);
+            }
+            x87_do_pop(dyn, ninst, x3);
            break;
        case 0xC8:
        case 0xC9:
@@ -54,19 +59,47 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xCE:
        case 0xCF:
            INST_NAME("FMULP STx, ST0");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VMUL_F64(v2, v2, v1);
-            x87_do_pop(dyn, ninst);
+            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)) {
+                VMUL_F32(v2, v2, v1);
+            } else {
+                VMUL_F64(v2, v2, v1);
+            }
+            x87_do_pop(dyn, ninst, x3);
            break;
-        case 0xD9:  /* FCOMPP */
-            INST_NAME("FCOMPP ST0, ST1");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, 1);
-            VCMP_F64(v1, v2);
+        case 0xD0:
+        case 0xD1:
+        case 0xD2:
+        case 0xD3:
+        case 0xD4:
+        case 0xD5:
+        case 0xD6:
+        case 0xD7:
+            INST_NAME("FCOMP 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(ST_IS_F(0)) {
+                VCMP_F32(v1, v2);
+            } else {
+                VCMP_F64(v1, v2);
+            }
            FCOM(x1, x2);
-            x87_do_pop(dyn, ninst);
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
+            break;
+
+        case 0xD9: 
+            INST_NAME("FCOMPP ST0, ST1");
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, 1));
+            v2 = x87_get_st(dyn, ninst, x1, x2, 1, X87_COMBINE(0, 1));
+            if(ST_IS_F(0)) {
+                VCMP_F32(v1, v2);
+            } else {
+                VCMP_F64(v1, v2);
+            }
+            FCOM(x1, x2);
+            x87_do_pop(dyn, ninst, x3);
+            x87_do_pop(dyn, ninst, x3);
            break;
        case 0xE0:
        case 0xE1:
@@ -77,10 +110,14 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xE6:
        case 0xE7:
            INST_NAME("FSUBRP STx, ST0");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VSUB_F64(v2, v1, v2);
-            x87_do_pop(dyn, ninst);
+            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)) {
+                VSUB_F32(v2, v1, v2);
+            } else {
+                VSUB_F64(v2, v1, v2);
+            }
+            x87_do_pop(dyn, ninst, x3);
            break;
        case 0xE8:
        case 0xE9:
@@ -91,10 +128,14 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xEE:
        case 0xEF:
            INST_NAME("FSUBP STx, ST0");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VSUB_F64(v2, v2, v1);
-            x87_do_pop(dyn, ninst);
+            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)) {
+                VSUB_F32(v2, v2, v1);
+            } else {
+                VSUB_F64(v2, v2, v1);
+            }
+            x87_do_pop(dyn, ninst, x3);
            break;
        case 0xF0:
        case 0xF1:
@@ -105,10 +146,14 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xF6:
        case 0xF7:
            INST_NAME("FDIVRP STx, ST0");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VDIV_F64(v2, v1, v2);
-            x87_do_pop(dyn, ninst);
+            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)) {
+                VDIV_F32(v2, v1, v2);
+            } else {
+                VDIV_F64(v2, v1, v2);
+            }
+            x87_do_pop(dyn, ninst, x3);
            break;
        case 0xF8:
        case 0xF9:
@@ -119,20 +164,16 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xFE:
        case 0xFF:
            INST_NAME("FDIVP STx, ST0");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VDIV_F64(v2, v2, v1);
-            x87_do_pop(dyn, ninst);
+            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)) {
+                VDIV_F32(v2, v2, v1);
+            } else {
+                VDIV_F64(v2, v2, v1);
+            }
+            x87_do_pop(dyn, ninst, x3);
            break;

-        case 0xD0:
-        case 0xD1:
-        case 0xD2:
-        case 0xD3:
-        case 0xD4:
-        case 0xD5:
-        case 0xD6:
-        case 0xD7:
        case 0xD8:
        case 0xDA:
        case 0xDB:
@@ -28,24 +28,47 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
 {
    uint8_t nextop = F8;
    uint8_t u8;
-    int32_t i32;
+    int32_t j32;
    uint8_t wback;
    uint8_t ed;
-    int v1, v2;
+    int v0, v1, v2;
    int s0;
    int fixedaddress;
+    int parity;

    MAYUSE(s0);
    MAYUSE(v2);
    MAYUSE(v1);
-    MAYUSE(i32);
+    MAYUSE(v0);
+    MAYUSE(j32);

    switch(nextop) {
+        case 0xC0:
+        case 0xC1:
+        case 0xC2:
+        case 0xC3:
+        case 0xC4:
+        case 0xC5:
+        case 0xC6:
+        case 0xC7:
+            INST_NAME("FFREEP STx");
+            // not handling Tag...
+            x87_do_pop(dyn, ninst, x3);
+            break;
+
        case 0xE0:
            INST_NAME("FNSTSW AX");
            LDR_IMM9(x2, xEmu, offsetof(x86emu_t, top));
            LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
-            AND_IMM8(x2, x2, 7);
+            if(dyn->x87stack) {
+                // update top
+                if(dyn->x87stack>0) {
+                    SUB_IMM8(x2, x2, dyn->x87stack);
+                } else {
+                    ADD_IMM8(x2, x2, -dyn->x87stack);
+                }
+                AND_IMM8(x2, x2, 7);
+            }
            BFI(x1, x2, 11, 3); // inject top
            BFI(xEAX, x1, 0, 16);
            break;
@@ -59,11 +82,16 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xEF:
            INST_NAME("FUCOMIP ST0, STx");
            SETFLAGS(X_ALL, SF_SET);
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VCMP_F64(v1, v2);
+            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))
+            {
+                VCMP_F32(v1, v2);
+            } else {
+                VCMP_F64(v1, v2);
+            }
            FCOMI(x1, x2);
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
            break;
        case 0xF0:
        case 0xF1:
@@ -75,21 +103,18 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
        case 0xF7:
            INST_NAME("FCOMIP ST0, STx");
            SETFLAGS(X_ALL, SF_SET);
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            VCMP_F64(v1, v2);
+            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))
+            {
+                VCMP_F32(v1, v2);
+            } else {
+                VCMP_F64(v1, v2);
+            }
            FCOMI(x1, x2);
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
            break;

-        case 0xC0:
-        case 0xC1:
-        case 0xC2:
-        case 0xC3:
-        case 0xC4:
-        case 0xC5:
-        case 0xC6:
-        case 0xC7:
        case 0xC8:
        case 0xC9:
        case 0xCA:
@@ -136,101 +161,260 @@ 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);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0);
+                    v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0);
                    LDRSH_IMM8(x1, wback, fixedaddress);
-                    s0 = fpu_get_scratch_single(dyn);
-                    VMOVtoV(s0, x1);
-                    VCVT_F64_S32(v1, s0);
+                    if(ST_IS_F(0)) {
+                        VMOVtoV(v1, x1);
+                        VCVT_F32_S32(v1, v1);
+                    } else {
+                        s0 = fpu_get_scratch_single(dyn);
+                        VMOVtoV(s0, x1);
+                        VCVT_F64_S32(v1, s0);
+                    }
                    break;
                case 1:
                    INST_NAME("FISTTP Ew, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-                    u8 = x87_setround(dyn, ninst, x1, x2, x12);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0, 0);
                    ed = x1;
                    s0 = fpu_get_scratch_single(dyn);
-                    VCVT_S32_F64(s0, v1);
+                    MSR_nzcvq_0();
+                    VMRS(x14);               // get fpscr
+                    // x1 already have FPCSR reg to clear exceptions flags
+                    ORR_IMM8(x3, x14, 0b001, 6); // enable exceptions
+                    BIC_IMM8(x3, x3, 0b10011111, 0);
+                    VMSR(x3);
+                    if(ST_IS_F(0)) {
+                        VCVTR_S32_F32(s0, v1);
+                    } else {
+                        VCVTR_S32_F64(s0, v1);
+                    }
+                    VMRS(x3);   // get the FPCSR reg and test FPU execption (invalid operation only)
                    VMOVfrV(ed, s0);
-                    MSR_nzcvq_0();    // reset Q flag (and all other flags too)
-                    SSAT_REG_LSL_IMM5(ed, 16, ed, 0);
-                    MRS_aspr(x3);   // check Q flag
-                    TSTS_IMM8_ROR(x3, 0b10, 3);
-                    B_MARK(cEQ);    // not saturated, continue
-                    MOV32(ed, 0xffff8000);
-                    MARK;
-                    STRH_IMM8(ed, wback, fixedaddress);
-                    x87_do_pop(dyn, ninst);
-                    x87_restoreround(dyn, ninst, u8);
+                    TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                    SSAT_REG_LSL_IMM5_COND(cEQ, x3, 16, ed, 0);
+                    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);
+                    VMSR(x14);
                    break;
                case 2:
                    INST_NAME("FIST Ew, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-                    u8 = x87_setround(dyn, ninst, x1, x2, x12);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                    u8 = x87_setround(dyn, ninst, x1, x2, x14);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0, 0);
                    ed = x1;
                    s0 = fpu_get_scratch_single(dyn);
-                    VCVTR_S32_F64(s0, v1);
+                    MSR_nzcvq_0();
+                    // x1 already have FPCSR reg to clear exceptions flags
+                    ORR_IMM8(x3, x1, 0b001, 6); // enable exceptions
+                    BIC_IMM8(x3, x3, 0b10011111, 0);
+                    VMSR(x3);
+                    if(ST_IS_F(0)) {
+                        VCVTR_S32_F32(s0, v1);
+                    } else {
+                        VCVTR_S32_F64(s0, v1);
+                    }
+                    VMRS(x3);   // get the FPCSR reg and test FPU execption (invalid operation only)
                    VMOVfrV(ed, s0);
-                    MSR_nzcvq_0();    // reset Q flag (and all other flags too)
-                    SSAT_REG_LSL_IMM5(ed, 16, ed, 0);
-                    MRS_aspr(x3);   // check Q flag
-                    TSTS_IMM8_ROR(x3, 0b10, 3);
-                    B_MARK(cEQ);    // not saturated, continue
-                    MOV32(ed, 0xffff8000);
-                    MARK;
-                    STRH_IMM8(ed, wback, fixedaddress);
+                    TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                    SSAT_REG_LSL_IMM5_COND(cEQ, x3, 16, ed, 0);
+                    CMPS_REG_LSL_IMM5_COND(cEQ, ed, x3, 0);
+                    MOVW_COND(cNE, x3, 0x8000); // saturated
+                    STRH_IMM8(x3, wback, fixedaddress);
                    x87_restoreround(dyn, ninst, u8);
                    break;
                case 3:
                    INST_NAME("FISTP Ew, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-                    u8 = x87_setround(dyn, ninst, x1, x2, x12);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                    u8 = x87_setround(dyn, ninst, x1, x2, x14);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0, 0);
                    ed = x1;
                    s0 = fpu_get_scratch_single(dyn);
-                    VCVTR_S32_F64(s0, v1);
+                    MSR_nzcvq_0();
+                    // x1 already have FPCSR reg to clear exceptions flags
+                    ORR_IMM8(x3, x1, 0b001, 6); // enable exceptions
+                    BIC_IMM8(x3, x3, 0b10011111, 0);
+                    VMSR(x3);
+                    if(ST_IS_F(0)) {
+                        VCVTR_S32_F32(s0, v1);
+                    } else {
+                        VCVTR_S32_F64(s0, v1);
+                    }
+                    VMRS(x3);   // get the FPCSR reg and test FPU execption (invalid operation only)
                    VMOVfrV(ed, s0);
-                    MSR_nzcvq_0();    // reset Q flag (and all other flags too)
-                    SSAT_REG_LSL_IMM5(ed, 16, ed, 0);
-                    MRS_aspr(x3);   // check Q flag
-                    TSTS_IMM8_ROR(x3, 0b10, 3);
-                    B_MARK(cEQ);    // not saturated, continue
-                    MOV32(ed, 0xffff8000);
-                    MARK;
-                    STRH_IMM8(ed, wback, fixedaddress);
-                    x87_do_pop(dyn, ninst);
+                    TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                    SSAT_REG_LSL_IMM5_COND(cEQ, x3, 16, ed, 0);
+                    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_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);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
                    if(ed!=x1) {MOV_REG(x1, ed);}
                    CALL(fpu_fbld, -1, 0);
                    break;
-                case 5: // could be inlined for most thing, but is it usefull?
+                case 5:
                    INST_NAME("FILD ST0, i64");
-                    x87_do_push_empty(dyn, ninst, x1);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
-                    if(ed!=x1) {MOV_REG(x1, ed);}
-                    CALL(arm_fild64, -1, 0);
+                    v1 = x87_do_push(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);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, 0, 0, 0);
+                    if(parity) {
+                        LDRD_IMM8(x2, wback, 0);    // x2/x3 is 64bits
+                    } else {
+                        LDR_IMM9(x2, wback, 0);
+                        LDR_IMM9(x3, wback, 4);
+                    }
+                    if(ninst<dyn->size
+                      && 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)==7)
+                    {
+                        MESSAGE(LOG_DUMP, "Hack for FILD/FISTP i64");
+                    } else {
+                        // Save xFlags to use it as scratch...
+                        STR_IMM9(xFlags, xEmu, offsetof(x86emu_t, eflags));
+                        // set STll(0).ll=i64 and ref=ST(0).q later (emu->fpu_ll[emu->top].ref == emu->mmx87[emu->top])
+                        //  get TOP
+                        LDR_IMM9(xFlags, xEmu, offsetof(x86emu_t, top));
+                        int a = 0 - dyn->x87stack;
+                        if(a<0) {
+                            SUB_IMM8(xFlags, xFlags, -a);
+                            AND_IMM8(xFlags, xFlags, 7);
+                        } else if(a>0) {
+                            ADD_IMM8(xFlags, xFlags, a);
+                            AND_IMM8(xFlags, xFlags, 7);
+                        }
+                        ADD_REG_LSL_IMM5(xFlags, xEmu, xFlags, 4);  // each fpu_ll is 2 int64: ref than ll
+                        ADD_IMM8_ROR(xFlags, xFlags, offsetof(x86emu_t, fpu_ll)>>2, 15);
+                        STRD_IMM8(x2, xFlags, 8);  // save ll
+                        // continue with conversion
+                        MOVS_REG_LSR_IMM5(x14, x3, 31);    // x14 is sign bit
+                        B_MARK(cEQ);    // no NEG is no sign bit
+                        RSBS_IMM8(x2, x2, 0);
+                        RSC_IMM8(x3, x3, 0);
+                        MARK;
+                        VMOVtoV(s0, x3);
+                        VCVT_F64_U32(v1, s0);
+                        VEOR(v2, v2, v2);
+                        MOVW(x1, 0x41F0);
+                        VMOVtoDx_16(v2, 3, x1);
+                        VMUL_F64(v1, v1, v2); // v1 = double high part of i64
+                        VMOVtoV(s0, x2);
+                        VCVT_F64_U32(v2, s0);
+                        VADD_F64(v1, v1, v2);
+                        TSTS_IMM8(x14, 1);
+                        VNEG_F64_cond(cNE, v1, v1);
+                        VST1_64(v1, xFlags);    // Store STll(0).ref
+                        LDR_IMM9(xFlags, xEmu, offsetof(x86emu_t, eflags));
+                    }
                    break;
                case 6:
                    INST_NAME("FBSTP tbytes, ST0");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
                    x87_forget(dyn, ninst, x1, x2, 0);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
                    if(ed!=x1) {MOV_REG(x1, ed);}
                    CALL(fpu_fbst, -1, 0);
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                    break;
                case 7: // could be inlined for most thing, but is it usefull?
                    INST_NAME("FISTP i64, ST0");
-                    x87_forget(dyn, ninst, x1, x2, 0);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
-                    if(ed!=x1) {MOV_REG(x1, ed);}
-                    CALL(arm_fistp64, -1, 0);
-                    x87_do_pop(dyn, ninst);
+                    parity = getedparity(dyn, ninst, addr, nextop, 3);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
+                    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 {
+                        #if 0
+                        v1 = x87_get_st(dyn, ninst, x2, x3, 0);
+                        //addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
+                        v2 = fpu_get_scratch_double(dyn);
+                        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->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, 0b001, 6); // 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);
+                        #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);
                    break;
                default:
                    DEFAULT;
@@ -23,6 +23,63 @@
 #include "dynarec_arm_functions.h"
 #include "dynarec_arm_helper.h"

+// emit SHL32 instruction, from s1 , shift s2, store result in s1 using s3 and s4 as scratch. s3 can be same as s2
+void emit_shl32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
+{
+    int32_t j32;
+    MAYUSE(j32);
+
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_shl32);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    IFX(F_OF) {
+        CMPS_IMM8(s2, 0);
+        IFX(F_OF) {
+            BIC_IMM8_COND(cEQ, xFlags, xFlags, 0b10, 0x0b);
+        }
+        IFX(X_PEND) {
+            STR_IMM9_COND(cEQ, s1, xEmu, offsetof(x86emu_t, res));
+        }
+        B_NEXT(cEQ);
+    }
+    IFX(X_CF | X_OF) {
+        RSB_IMM8(s4, s2, 32);
+        MOV_REG_LSR_REG(s4, s1, s4);
+        BFI(xFlags, s4, F_CF, 1);
+    }
+    IFX(X_ZF) {
+        MOVS_REG_LSL_REG(s1, s1, s2);
+    } else {
+        MOV_REG_LSL_REG(s1, s1, s2);
+    }
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_ZF) {
+        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
+        BIC_IMM8_COND(cNE, xFlags, xFlags, 1<<F_ZF, 0);
+    }
+    IFX(X_SF) {
+        MOV_REG_LSR_IMM5(s4, s1, 31);
+        BFI(xFlags, s4, F_SF, 1);
+    }
+    IFX(X_OF) {
+        CMPS_IMM8(s2, 1);   // if s3==1
+            IFX(X_SF) {} else {MOV_REG_LSR_IMM5_COND(cEQ, s4, s1, 31);}
+            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);
+        // else
+            BFC_COND(cNE, xFlags, F_OF, 1);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
 // emit SHL32 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
 void emit_shl32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
 {
@@ -30,17 +87,13 @@ void emit_shl32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        MOVW(s3, c);
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
-        MOVW(s4, d_shl32);
+        SET_DF(s4, d_shl32);
    } else IFX(X_ALL) {
-        MOVW(s4, d_none);
-    }
-    IFX(X_PEND|X_ALL) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, df));
+        SET_DFNONE(s4);
    }
    if(c==0) {
        IFX(F_OF) {
-            MOVW(s4, 0);
-            STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_OF]));
+            BIC_IMM8(xFlags, xFlags, 0b10, 0x0b);
        }
        IFX(X_PEND) {
            STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
@@ -48,8 +101,8 @@ void emit_shl32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        return;
    }
    IFX(X_CF) {
-        UBFX(s3, s1, c, 1);
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, flags[F_CF]));
+        MOV_REG_LSR_IMM5(s3, s1, 32-c);
+        BFI(xFlags, s3, F_CF, 1);
    }
    IFX(X_ZF|X_OF) {
        MOVS_REG_LSL_IMM5(s1, s1, c);
@@ -60,33 +113,76 @@ void emit_shl32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
    }
    IFX(X_ZF) {
-        MOVW_COND(cNE, s3, 0);
-        MOVW_COND(cEQ, s3, 1);
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, flags[F_ZF]));
+        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
+        BIC_IMM8_COND(cNE, xFlags, xFlags, 1<<F_ZF, 0);
    }
    IFX(X_SF) {
-        UBFX(s3, s1, 31, 1);
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, flags[F_SF]));
+        MOV_REG_LSR_IMM5(s3, s1, 31);
+        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_OF) {
        if(c==1) {
-            IFX(X_SF) {} else {UBFX(s3, s1, 31, 1);}
+            IFX(X_SF) {} else {MOV_REG_LSR_IMM5(s3, s1, 31);}
            XOR_IMM8_COND(cCS, s3, s3, 1);
-            STR_IMM9(s3, xEmu, offsetof(x86emu_t, flags[F_OF]));
+            BFI(xFlags, s3, F_OF, 1);
        } else {
-            MOVW(s4, 0);
-            STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_OF]));
+            BFC(xFlags, F_OF, 1);
        }
    }
    IFX(X_PF) {
-        // PF: (((emu->x86emu_parity_tab[(res) / 32] >> ((res) % 32)) & 1) == 0)
-        AND_IMM8(s3, s1, 0xE0); // lsr 5 masking pre-applied
-        LDR_IMM9(s4, xEmu, offsetof(x86emu_t, x86emu_parity_tab));
-        LDR_REG_LSR_IMM5(s4, s4, s3, 5-2);   // x/32 and then *4 because array is integer
-        AND_IMM8(s3, s1, 31);
-        MVN_REG_LSR_REG(s4, s4, s3);
-        AND_IMM8(s4, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_PF]));
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit SHR32 instruction, from s1 , s2, store result in s1 using s3 and s4 as scratch, s2 can be same as s3
+void emit_shr32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
+{
+    int32_t j32;
+    MAYUSE(j32);
+
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_shr32);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    IFX(X_ALL) {
+        CMPS_IMM8(s2, 0); //if(!c)
+            IFX(X_PEND) {
+                STR_IMM9_COND(cEQ, s1, xEmu, offsetof(x86emu_t, res));
+            }
+            B_NEXT(cEQ);
+    }
+    IFX(X_ZF|X_CF) {
+        MOVS_REG_LSR_REG(s1, s1, s2);
+    } else {
+        MOV_REG_LSR_REG(s1, s1, s2);
+    }
+    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) {
+        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);
    }
 }

@@ -97,12 +193,9 @@ void emit_shr32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        MOVW(s3, c);
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
-        MOVW(s4, d_shr32);
+        SET_DF(s4, d_shr32);
    } else IFX(X_ALL) {
-        MOVW(s4, d_none);
-    }
-    IFX(X_PEND|X_ALL) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, df));
+        SET_DFNONE(s4);
    }
    if(!c) {
        IFX(X_PEND) {
@@ -118,37 +211,28 @@ 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) {
-        MOVW_COND(cNE, s3, 0);
-        MOVW_COND(cEQ, s3, 1);
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, flags[F_ZF]));
+        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
    }
    IFX(X_CF) {
-        MOVW_COND(cCC, s3, 0);
-        MOVW_COND(cCS, s3, 1);
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, flags[F_CF]));
+        ORR_IMM8_COND(cCS, xFlags, xFlags, 1<<F_CF, 0);
    }
    IFX(X_SF) {
-        UBFX(s3, s1, 31, 1);
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, flags[F_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);
-            AND_IMM8(s4, s4, 1);
-            STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_OF]));
+            BFI(xFlags, s4, F_OF, 1);
        }
    }
    IFX(X_PF) {
-        // PF: (((emu->x86emu_parity_tab[(res) / 32] >> ((res) % 32)) & 1) == 0)
-        AND_IMM8(s3, s1, 0xE0); // lsr 5 masking pre-applied
-        LDR_IMM9(s4, xEmu, offsetof(x86emu_t, x86emu_parity_tab));
-        LDR_REG_LSR_IMM5(s4, s4, s3, 5-2);   // x/32 and then *4 because array is integer
-        AND_IMM8(s3, s1, 31);
-        MVN_REG_LSR_REG(s4, s4, s3);
-        AND_IMM8(s4, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_PF]));
+        emit_pf(dyn, ninst, s1, s3, s4);
    }
 }

@@ -159,12 +243,9 @@ void emit_sar32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        MOVW(s3, c);
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
-        MOVW(s4, d_sar32);
+        SET_DF(s4, d_sar32);
    } else IFX(X_ALL) {
-        MOVW(s4, d_none);
-    }
-    IFX(X_PEND|X_ALL) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, df));
+        SET_DFNONE(s4);
    }
    if(!c) {
        IFX(X_PEND) {
@@ -180,44 +261,35 @@ void emit_sar32c(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) {
-        MOVW_COND(cNE, s3, 0);
-        MOVW_COND(cEQ, s3, 1);
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, flags[F_ZF]));
+        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
    }
    IFX(X_CF) {
-        MOVW_COND(cCC, s3, 0);
-        MOVW_COND(cCS, s3, 1);
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, flags[F_CF]));
+        ORR_IMM8_COND(cCS, xFlags, xFlags, 1<<F_CF, 0);
    }
    IFX(X_SF) {
-        UBFX(s3, s1, 31, 1);
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, flags[F_SF]));
+        MOV_REG_LSR_IMM5(s3, s1, 31);
+        BFI(xFlags, s3, F_SF, 1);
    }
    IFX(X_PF) {
-        // PF: (((emu->x86emu_parity_tab[(res) / 32] >> ((res) % 32)) & 1) == 0)
-        AND_IMM8(s3, s1, 0xE0); // lsr 5 masking pre-applied
-        LDR_IMM9(s4, xEmu, offsetof(x86emu_t, x86emu_parity_tab));
-        LDR_REG_LSR_IMM5(s4, s4, s3, 5-2);   // x/32 and then *4 because array is integer
-        AND_IMM8(s3, s1, 31);
-        MVN_REG_LSR_REG(s4, s4, s3);
-        AND_IMM8(s4, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_PF]));
+        emit_pf(dyn, ninst, s1, s3, s4);
    }
 }

 // emit ROL32 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
 void emit_rol32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
 {
+    if(!c)
+        return;
    IFX(X_PEND) {
        MOVW(s3, c);
        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
-        MOVW(s4, d_rol32);
+        SET_DF(s4, d_rol32);
    } else IFX(X_ALL) {
-        MOVW(s4, d_none);
-    }
-    IFX(X_PEND|X_ALL) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, df));
+        SET_DFNONE(s4);
    }
    if(!c) {
        IFX(X_PEND) {
@@ -230,14 +302,12 @@ void emit_rol32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
    }
    IFX(X_CF) {
-        AND_IMM8(s3, s1, 1);
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, flags[F_CF]));
+        BFI(xFlags, s1, F_CF, 1);
    }
    IFX(X_OF) {
        if(c==1) {
            ADD_REG_LSR_IMM5(s3, s1, s1, 31);
-            AND_IMM8(s3, s3, 1);
-            STR_IMM9(s3, xEmu, offsetof(x86emu_t, flags[F_OF]));
+            BFI(xFlags, s3, F_OF, 1);
        }
    }
 }
@@ -245,15 +315,14 @@ void emit_rol32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
 // emit ROR32 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
 void emit_ror32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
 {
+    if(!c)
+        return;
    IFX(X_PEND) {
        MOVW(s3, c);
        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
-        MOVW(s4, d_ror32);
+        SET_DF(s4, d_ror32);
    } else IFX(X_ALL) {
-        MOVW(s4, d_none);
-    }
-    IFX(X_PEND|X_ALL) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, df));
+        SET_DFNONE(s4);
    }
    if(!c) {
        IFX(X_PEND) {
@@ -266,15 +335,260 @@ void emit_ror32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
    }
    IFX(X_CF) {
-        UBFX(s3, s1, 31, 1);
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, flags[F_CF]));
+        MOV_REG_LSR_IMM5(s3, s1, 31);
+        BFI(xFlags, s3, F_CF, 1);
    }
    IFX(X_OF) {
        if(c==1) {
            MOV_REG_LSR_IMM5(s4, s1, 30);
            XOR_REG_LSR_IMM8(s4, s4, s4, 1);
-            AND_IMM8(s4, s4, 1);
-            STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_OF]));
+            BFI(xFlags, s4, F_OF, 1);
+        }
+    }
+}
+
+// emit SHRD32 instruction, from s1, fill s2 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_shrd32c(dynarec_arm_t* dyn, int ninst, int s1, int s2, int32_t c, int s3, int s4)
+{
+    c&=0x1f;
+    IFX(X_PEND) {
+        MOVW(s3, c);
+        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);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    if(!c) {
+        IFX(X_PEND) {
+            STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+        }
+        return;
+    }
+    IFX(X_CF) {
+        MOVS_REG_LSR_IMM5(s1, s1, c);
+    } else {
+        MOV_REG_LSR_IMM5(s1, s1, c);
+    }
+    IFX(X_ZF|X_CF) {
+        BIC_IMM8(xFlags, xFlags, (1<<F_ZF)|(1<<F_CF), 0);
+    }
+    IFX(X_CF) {
+        ORR_IMM8_COND(cCS, xFlags, xFlags, 1<<F_CF, 0);
+    }
+    IFX(X_ZF) {
+        ORRS_REG_LSL_IMM5(s1, s1, s2, 32-c);
+    } else {
+        ORR_REG_LSL_IMM5(s1, s1, s2, 32-c);
+    }
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_ZF) {
+        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 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);
+    }
+}
+
+void emit_shld32c(dynarec_arm_t* dyn, int ninst, int s1, int s2, int32_t c, int s3, int s4)
+{
+    c&=0x1f;
+    IFX(X_PEND) {
+        MOVW(s3, c);
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
+        // same flags computation as with shl32
+        SET_DF(s4, d_shl32);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    if(c==0) {
+        IFX(F_OF) {
+            BFC(xFlags, F_OF, 1);
+        }
+        IFX(X_PEND) {
+            STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+        }
+        return;
+    }
+    IFX(X_CF) {
+        MOV_REG_LSR_IMM5(s3, s1, 32-c);
+        BFI(xFlags, s3, F_CF, 1);
+    }
+    IFX(X_OF) {
+        MOVS_REG_LSL_IMM5(s1, s1, c);
+    } else {
+        MOV_REG_LSL_IMM5(s1, s1, c);
+    }
+    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);
+        } else {
+            BFC(xFlags, F_OF, 1);
+        }
+    }
+    IFX(X_ZF) {
+        ORRS_REG_LSR_IMM5(s1, s1, s2, 32-c);
+    } else {
+        ORR_REG_LSR_IMM5(s1, s1, s2, 32-c);
+    }
+
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_ZF) {
+        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
+        BIC_IMM8_COND(cNE, xFlags, xFlags, 1<<F_ZF, 0);
+    }
+    IFX(X_SF) {
+        MOV_REG_LSR_IMM5(s3, s1, 31);
+        BFI(xFlags, s3, F_SF, 1);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit SHRD32 instruction, from s1, fill s2 , shift s3, store result in s1 using s4 and s3 as scratch
+void emit_shrd32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
+{
+    int j32;
+    IFX(X_PEND) {
+        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);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    CMPS_IMM8(s3, 0);
+        IFX(X_PEND) {
+            STR_IMM9_COND(cEQ, s1, xEmu, offsetof(x86emu_t, res));
+        }
+        B_NEXT(cEQ);
+
+    IFX(X_CF) {
+        MOVS_REG_LSR_REG(s1, s1, s3);
+    } else {
+        MOV_REG_LSR_REG(s1, s1, s3);
+    }
+    IFX(X_ZF|X_CF) {
+        BIC_IMM8(xFlags, xFlags, (1<<F_ZF)|(1<<F_CF), 0);
+    }
+    IFX(X_CF) {
+        ORR_IMM8_COND(cCS, xFlags, xFlags, 1<<F_CF, 0);
+    }
+    RSB_IMM8(s3, s3, 32);
+    IFX(X_ZF) {
+        ORRS_REG_LSL_REG(s1, s1, s2, s3);
+    } else {
+        ORR_REG_LSL_REG(s1, s1, s2, s3);
+    }
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_ZF) {
+        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
+    }
+    IFX(X_SF) {
+        MOV_REG_LSR_IMM5(s4, s1, 31);
+        BFI(xFlags, s4, F_SF, 1);
+    }
+    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);
+            BFI_COND(cEQ, xFlags, s4, F_OF, 1);
+    }
+    IFX(X_PF) {
+        emit_pf(dyn, ninst, s1, s3, s4);
+    }
+}
+
+// emit ROL8 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_rol8c(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_rol8);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    if(!c) {
+        IFX(X_PEND) {
+            STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+        }
+        return;
+    }
+    if(c&7) {
+        BFI(s1, s1, 24, 8);  // duplicate the value for rotate left
+        MOV_REG_ROR_IMM5(s1, s1, 32-(c&7));
+    }
+    //AND_IMM8(s1, s1, 0xff);
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_CF) {
+        BFI(xFlags, s1, F_CF, 1);
+    }
+    IFX(X_OF) {
+        if(c==1) {
+            ADD_REG_LSR_IMM5(s3, s1, s1, 7);
+            BFI(xFlags, s3, F_OF, 1);
+        }
+    }
+}
+
+// emit ROR8 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
+void emit_ror8c(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_ror8);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    if(!c) {
+        IFX(X_PEND) {
+            STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+        }
+        return;
+    }
+    if(c&7) {
+        BFI(s1, s1, 8, 8);  // duplicate the value for Rotate right
+        MOV_REG_ROR_IMM5(s1, s1, c&7);
+    }
+    IFX(X_PEND) {
+        AND_IMM8(s1, s1, 0xff);
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+    }
+    IFX(X_CF) {
+        MOV_REG_LSR_IMM5(s3, s1, 7);
+        BFI(xFlags, s3, F_CF, 1);
+    }
+    IFX(X_OF) {
+        if(c==1) {
+            MOV_REG_LSR_IMM5(s4, s1, 6);
+            XOR_REG_LSR_IMM8(s4, s4, s4, 1);
+            BFI(xFlags, s4, F_OF, 1);
        }
    }
 }
@@ -26,43 +26,34 @@
 // emit CMP32 instruction, from cmp s1 , s2, using s3 and s4 as scratch
 void emit_cmp32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 {
-    IFX(X_PEND) {
+    IFX_PENDOR0 {
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
-        MOVW(s4, d_cmp32);
+        SET_DF(s4, d_cmp32);
    } else {
-        MOVW(s4, 0);
+        SET_DFNONE(s4);
    }
-    STR_IMM9(s4, xEmu, offsetof(x86emu_t, df)); // reset flags
    SUBS_REG_LSL_IMM5(s3, s1, s2, 0);   // res = s1 - s2
-    IFX(X_PEND) {
+    IFX_PENDOR0 {
        STR_IMM9(s3, xEmu, offsetof(x86emu_t, res));
    }
+    IFX(X_ZF|X_CF) {
+        BIC_IMM8(xFlags, xFlags, (1<<F_ZF)|(1<<F_CF), 0);
+    }
    IFX(X_ZF) {
-        IFX(X_PEND) {
-            MOVW(s4, 0);
-        }
-        // first the easy flag, also found on ARM
-        MOVW_COND(cEQ, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_ZF]));
+        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
    }
    IFX(X_OF) {
-        IFX(X_ZF|X_PEND) {
-            MOVW(s4, 0);
-        }
-        MOVW_COND(cVS, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_OF]));
+        ORR_IMM8_COND(cVS, xFlags, xFlags, 0b10, 0x0b);
+        BIC_IMM8_COND(cVC, xFlags, xFlags, 0b10, 0x0b);
    }
    IFX(X_CF) {
-        IFX(X_ZF|X_OF|X_PEND) {
-            MOVW(s4, 0);
-        }
-        MOVW_COND(cCC, s4, 1);      // Carry flags on ARM is reversed compared to x86 one
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_CF]));
+        // reversed carry
+        ORR_IMM8_COND(cCC, xFlags, xFlags, 1<<F_CF, 0);
    }
    IFX(X_SF) {
-        UBFX(s4, s3, 31, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_SF]));
+        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_AF) {
@@ -73,8 +64,8 @@ void emit_cmp32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BIC_REG_LSL_IMM5(s3, s2, s1, 0);    // loosing res... s3 = s & ~d
        ORR_REG_LSL_IMM5(s3, s4, s3, 0);    // s3 = (res & (~d | s)) | (s & ~d)
        IFX(X_AF) {
-            UBFX(s4, s3, 3, 1);
-            STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_AF]));    // AF: bc & 0x08
+            MOV_REG_LSR_IMM5(s4, s3, 3);
+            BFI(xFlags, s4, F_AF, 1);    // AF: bc & 0x08
        }
    }
    IFX(X_PF) {
@@ -83,90 +74,71 @@ void emit_cmp32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
            SUB_REG_LSL_IMM5(s3, s1, s2, 0);
        }
        AND_IMM8(s3, s3, 0xE0); // lsr 5 masking pre-applied
-        LDR_IMM9(s4, xEmu, offsetof(x86emu_t, x86emu_parity_tab));
+        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);
-        AND_IMM8(s4, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_PF]));
+        BFI(xFlags, s4, F_PF, 1);
    }
 }

 // emit CMP32 instruction, from cmp s1 , 0, using s3 and s4 as scratch
 void emit_cmp32_0(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
 {
-    MOVW(s4, 0);
-    IFX(X_PEND) {
+    IFX_PENDOR0 {
+        MOVW(s4, 0);
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
        STR_IMM9(s4, xEmu, offsetof(x86emu_t, op2));
-        MOVW(s4, d_cmp32);
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+        SET_DF(s4, d_cmp32);
+    } else {
+        SET_DFNONE(s4);
    }
-    STR_IMM9(s4, xEmu, offsetof(x86emu_t, df)); // reset flags
    SUBS_IMM8(s3, s1, 0);   // res = s1 - 0
-    IFX(X_PEND) {
-        IFX(X_ZF|X_OF|X_CF|X_AF) {
-            MOVW(s4, 0);
-        }
-    }
    // and now the tricky ones (and mostly unused), PF and AF
    // bc = (res & (~d | s)) | (~d & s) => is 0 here...
-    IFX(X_CF) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_CF]));    // CF : bc & 0x80000000
-    }
-    IFX(X_AF) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_AF]));    // AF: bc & 0x08
+    IFX(X_CF | X_AF | X_ZF) {
+        BIC_IMM8(xFlags, xFlags, (1<<F_CF)|(1<<F_AF)|(1<<F_ZF), 0);
    }
    IFX(X_OF) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_OF]));
+        BFC(xFlags, F_OF, 1);
    }
    IFX(X_ZF) {
-        // first the easy flags, also found on ARM
-        MOVW_COND(cEQ, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_ZF]));
+        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
    }
    IFX(X_SF) {
-        UBFX(s4, s3, 31, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_SF]));
+        MOV_REG_LSR_IMM5(s4, s3, 31);
+        BFI(xFlags, s4, F_SF, 1);
    }
    IFX(X_PF) {
-        // PF: (((emu->x86emu_parity_tab[(res) / 32] >> ((res) % 32)) & 1) == 0)
-        AND_IMM8(s3, s1, 0xE0); // lsr 5 masking pre-applied
-        LDR_IMM9(s4, xEmu, offsetof(x86emu_t, x86emu_parity_tab));
-        LDR_REG_LSR_IMM5(s4, s4, s3, 5-2);   // x/32 and then *4 because array is integer
-        AND_IMM8(s3, s1, 31);
-        MVN_REG_LSR_REG(s4, s4, s3);
-        AND_IMM8(s4, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_PF]));
+        emit_pf(dyn, ninst, s1, s3, s4);
    }
 }

 // emit CMP16 instruction, from cmp s1 , s2, using s3 and s4 as scratch
 void emit_cmp16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 {
-    IFX(X_PEND) {
+    IFX_PENDOR0 {
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
-        MOVW(s3, d_cmp16);
+        SET_DF(s3, d_cmp16);
    } else {
-        MOVW(s3, 0);
+        SET_DFNONE(s3);
    }
-    STR_IMM9(s3, xEmu, offsetof(x86emu_t, df)); // reset flags
    SUB_REG_LSL_IMM5(s3, s1, s2, 0);   // res = s1 - s2
-    IFX(X_PEND) {
+    IFX_PENDOR0 {
        STR_IMM9(s3, xEmu, offsetof(x86emu_t, res));
    }
    IFX(X_ZF) {
        MOVW(s4, 0xffff);
        TSTS_REG_LSL_IMM5(s3, s4, 0);
-        MOVW_COND(cNE, s4, 0);
-        MOVW_COND(cEQ, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_ZF]));
+        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
+        BIC_IMM8_COND(cNE, xFlags, xFlags, 1<<F_ZF, 0);
    }
    IFX(X_SF) {
-        UBFX(s4, s3, 15, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_SF]));
+        MOV_REG_LSR_IMM5(s4, s3, 15);
+        BFI(xFlags, s4, F_SF, 1);
    }
    // bc = (res & (~d | s)) | (~d & s)
    IFX(X_CF|X_AF|X_OF) {
@@ -176,108 +148,87 @@ void emit_cmp16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BIC_REG_LSL_IMM5(s3, s2, s1, 0);    // loosing res... s3 = s & ~d
        ORR_REG_LSL_IMM5(s3, s4, s3, 0);    // s3 = (res & (~d | s)) | (s & ~d)
        IFX(X_CF) {
-            UBFX(s4, s3, 15, 1);
-            STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_CF]));    // CF : bc & 0x8000
+            MOV_REG_LSR_IMM5(s4, s3, 15);
+            BFI(xFlags, s4, F_CF, 1);    // CF : bc & 0x8000
        }
        IFX(X_AF) {
-            UBFX(s4, s3, 3, 1);
-            STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_AF]));    // AF: bc & 0x08
+            MOV_REG_LSR_IMM5(s4, s3, 3);
+            BFI(xFlags, s4, F_AF, 1);    // AF: bc & 0x08
        }
        IFX(X_OF) {
            MOV_REG_LSR_IMM5(s4, s3, 14);
            XOR_REG_LSR_IMM8(s4, s4, s4, 1);
-            AND_IMM8(s4, s4, 1);
-            STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_OF]));    // OF: ((bc >> 14) ^ ((bc>>14)>>1)) & 1
+            BFI(xFlags, s4, F_OF, 1);    // OF: ((bc >> 14) ^ ((bc>>14)>>1)) & 1
        }
    }
-    // PF: (((emu->x86emu_parity_tab[(res) / 32] >> ((res) % 32)) & 1) == 0)
    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
-        LDR_IMM9(s4, xEmu, offsetof(x86emu_t, x86emu_parity_tab));
+        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);
-        AND_IMM8(s4, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_PF]));
+        BFI(xFlags, s4, F_PF, 1);
    }
 }

 // emit CMP16 instruction, from cmp s1 , #0, using s3 and s4 as scratch
 void emit_cmp16_0(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
 {
-    MOVW(s3, 0);
-    IFX(X_PEND) {
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
+    IFX_PENDOR0 {
+        MOVW(s3, 0);
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
-        MOVW(s3, d_cmp16);
-    }
-    STR_IMM9(s3, xEmu, offsetof(x86emu_t, df)); // reset flags
-    IFX(X_OF|X_CF|X_AF) {
-        MOVW(s4, 0);
+        SET_DF(s3, d_cmp16);
+    } else {
+        SET_DFNONE(s3);
    }
    // bc = (res & (~d | s)) | (~d & s) = 0
-    IFX(X_CF) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_CF]));    // CF : bc & 0x8000
-    }
-    IFX(X_AF) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_AF]));    // AF: bc & 0x08
+    IFX(X_CF | X_AF | X_ZF) {
+        BIC_IMM8(xFlags, xFlags, (1<<F_CF)|(1<<F_AF)|(1<<F_ZF), 0);
    }
    IFX(X_OF) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_OF]));    // OF: ((bc >> 14) ^ ((bc>>14)>>1)) & 1
+        BFC(xFlags, F_OF, 1);
    }
    IFX(X_ZF) {
        MOVW(s4, 0xffff);
        TSTS_REG_LSL_IMM5(s1, s4, 0);
-        MOVW(s4, 0);
-        MOVW_COND(cEQ, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_ZF]));
+        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
    }
    IFX(X_SF) {
-        UBFX(s4, s1, 15, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_SF]));
+        MOV_REG_LSR_IMM5(s4, s1, 15);
+        BFI(xFlags, s4, F_SF, 1);
    }
-    // PF: (((emu->x86emu_parity_tab[(res) / 32] >> ((res) % 32)) & 1) == 0)
    IFX(X_PF) {
-        AND_IMM8(s3, s1, 0xE0); // lsr 5 masking pre-applied
-        LDR_IMM9(s4, xEmu, offsetof(x86emu_t, x86emu_parity_tab));
-        LDR_REG_LSR_IMM5(s4, s4, s3, 5-2);   // x/32 and then *4 because array is integer
-        AND_IMM8(s3, s1, 31);
-        MVN_REG_LSR_REG(s4, s4, s3);
-        AND_IMM8(s4, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_PF]));
+        emit_pf(dyn, ninst, s1, s3, s4);
    }
 }
 // emit CMP8 instruction, from cmp s1 , s2, using s3 and s4 as scratch
 void emit_cmp8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 {
-    IFX(X_PEND) {
-        MOVW(s4, d_cmp8);
+    IFX_PENDOR0 {
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
+        SET_DF(s4, d_cmp8);
    } else {
-        MOVW(s4, 0);
+        SET_DFNONE(s4);
    }
-    STR_IMM9(s4, xEmu, offsetof(x86emu_t, df)); // reset flags
    SUB_REG_LSL_IMM5(s3, s1, s2, 0);   // res = s1 - s2
-    IFX(X_PEND) {
+    IFX_PENDOR0 {
        STR_IMM9(s3, xEmu, offsetof(x86emu_t, res));
    }
    IFX(X_ZF) {
-        IFX(X_PEND) {
-            MOVW(s4, 0);
-        }
        TSTS_IMM8(s3, 0xff);
-        MOVW_COND(cEQ, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_ZF]));
+        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
+        BIC_IMM8_COND(cNE, xFlags, xFlags, 1<<F_ZF, 0);
    }
    IFX(X_SF) {
-        UBFX(s4, s3, 7, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_SF]));
+        MOV_REG_LSR_IMM5(s4, s3, 7);
+        BFI(xFlags, s4, F_SF, 1);
    }
    // bc = (res & (~d | s)) | (~d & s)
    IFX(X_CF|X_AF|X_OF) {
@@ -287,198 +238,175 @@ void emit_cmp8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
        BIC_REG_LSL_IMM5(s3, s2, s1, 0);    // loosing res... s3 = s & ~d
        ORR_REG_LSL_IMM5(s3, s4, s3, 0);    // s3 = (res & (~d | s)) | (s & ~d)
        IFX(X_CF) {
-            UBFX(s4, s3, 7, 1);
-            STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_CF]));    // CF : bc & 0x80
+            MOV_REG_LSR_IMM5(s4, s3, 7);
+            BFI(xFlags, s4, F_CF, 1);    // CF : bc & 0x80
        }
        IFX(X_AF) {
-            UBFX(s4, s3, 3, 1);
-            STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_AF]));    // AF: bc & 0x08
+            MOV_REG_LSR_IMM5(s4, s3, 3);
+            BFI(xFlags, s4, F_AF, 1);    // AF: bc & 0x08
        }
        IFX(X_OF) {
            MOV_REG_LSR_IMM5(s4, s3, 6);
            XOR_REG_LSR_IMM8(s4, s4, s4, 1);
-            AND_IMM8(s4, s4, 1);
-            STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_OF]));    // OF: ((bc >> 6) ^ ((bc>>6)>>1)) & 1
+            BFI(xFlags, s4, F_OF, 1);    // OF: ((bc >> 6) ^ ((bc>>6)>>1)) & 1
        }
    }
-    // PF: (((emu->x86emu_parity_tab[(res) / 32] >> ((res) % 32)) & 1) == 0)
    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
-        LDR_IMM9(s4, xEmu, offsetof(x86emu_t, x86emu_parity_tab));
+        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);
-        AND_IMM8(s4, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_PF]));
+        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)
 {
-    MOVW(s4, 0);
-    IFX(X_PEND) {
-        MOVW(s3, d_cmp8);
+    IFX_PENDOR0 {
        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+        MOVW(s4, 0);
        STR_IMM9(s4, xEmu, offsetof(x86emu_t, op2));
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, df));
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, res));
+        SET_DF(s3, d_cmp8);
    } else {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, df)); // reset flags
+        SET_DFNONE(s4);
    }
    // bc = (res & (~d | s)) | (~d & s) = 0
-    IFX(X_CF) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_CF]));    // CF : bc & 0x80
-    }
-    IFX(X_AF) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_AF]));    // AF: bc & 0x08
+    IFX(X_CF | X_AF | X_ZF) {
+        BIC_IMM8(xFlags, xFlags, (1<<F_CF)|(1<<F_AF)|(1<<F_ZF), 0);
    }
    IFX(X_OF) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_OF]));    // OF: ((bc >> 6) ^ ((bc>>6)>>1)) & 1
+        BFC(xFlags, F_OF, 1);
    }
    IFX(X_ZF) {
        TSTS_IMM8(s1, 0xff);
-        MOVW_COND(cEQ, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_ZF]));
+        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
    }
    IFX(X_SF) {
-        UBFX(s4, s1, 7, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_SF]));
+        MOV_REG_LSR_IMM5(s4, s1, 7);
+        BFI(xFlags, s4, F_SF, 1);
    }
-    // PF: (((emu->x86emu_parity_tab[(res) / 32] >> ((res) % 32)) & 1) == 0)
    IFX(X_PF) {
-        AND_IMM8(s3, s1, 0xE0); // lsr 5 masking pre-applied
-        LDR_IMM9(s4, xEmu, offsetof(x86emu_t, x86emu_parity_tab));
-        LDR_REG_LSR_IMM5(s4, s4, s3, 5-2);   // x/32 and then *4 because array is integer
-        AND_IMM8(s3, s1, 31);
-        MVN_REG_LSR_REG(s4, s4, s3);
-        AND_IMM8(s4, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_PF]));
+        emit_pf(dyn, ninst, s1, s3, s4);
    }
 }

 // emit TEST32 instruction, from test s1 , s2, using s3 and s4 as scratch
 void emit_test32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 {
-    MOVW(s4, 0);
-    IFX(X_PEND) {
-        MOVW(s3, d_tst32);
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, df));
+    IFX_PENDOR0 {
+        SET_DF(s3, d_tst32);
    } else {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, df)); // reset flags
+        SET_DFNONE(s4);
    }
    IFX(X_OF) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_OF]));
+        BFC(xFlags, F_OF, 1);
    }
-    IFX(X_CF) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_CF]));
+    IFX(X_ZF|X_CF) {
+        BIC_IMM8(xFlags, xFlags, (1<<F_ZF)|(1<<F_CF), 0);
    }
    ANDS_REG_LSL_IMM5(s3, s1, s2, 0);   // res = s1 & s2
-    IFX(X_PEND) {
+    IFX_PENDOR0 {
        STR_IMM9(s3, xEmu, offsetof(x86emu_t, res));
    }
    IFX(X_ZF) {
-        MOVW_COND(cEQ, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_ZF]));
+        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
    }
    IFX(X_SF) {
-        UBFX(s4, s3, 31, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_SF]));
+        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
-        LDR_IMM9(s4, xEmu, offsetof(x86emu_t, x86emu_parity_tab));
+        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);
+        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);
-        AND_IMM8(s4, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_PF]));
+        BFI(xFlags, s4, F_PF, 1);
    }
 }

 // emit TEST16 instruction, from test s1 , s2, using s3 and s4 as scratch
 void emit_test16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 {
-    MOVW(s4, 0);
-    IFX(X_PEND) {
-        MOVW(s3, d_tst16);
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, df));
+    IFX_PENDOR0 {
+        SET_DF(s3, d_tst16);
    } else {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, df)); // reset flags
+        SET_DFNONE(s4);
    }
    IFX(X_OF) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_OF]));
+        BFC(xFlags, F_OF, 1);
    }
-    IFX(X_CF) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_CF]));
+    IFX(X_ZF|X_CF) {
+        BIC_IMM8(xFlags, xFlags, (1<<F_ZF)|(1<<F_CF), 0);
    }
    ANDS_REG_LSL_IMM5(s3, s1, s2, 0);   // res = s1 & s2
-    IFX(X_PEND) {
+    IFX_PENDOR0 {
        STR_IMM9(s3, xEmu, offsetof(x86emu_t, res));
    }
    IFX(X_ZF) {
-        MOVW_COND(cEQ, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_ZF]));
+        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
    }
    IFX(X_SF) {
-        UBFX(s4, s3, 15, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_SF]));
+        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
-        LDR_IMM9(s4, xEmu, offsetof(x86emu_t, x86emu_parity_tab));
+        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);
-        AND_IMM8(s4, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_PF]));
+        BFI(xFlags, s4, F_PF, 1);
    }
 }

 // emit TEST8 instruction, from test s1 , s2, using s3 and s4 as scratch
 void emit_test8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 {
-    MOVW(s4, 0);
-    IFX(X_PEND) {
-        MOVW(s3, d_tst8);
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, df));
+    IFX_PENDOR0 {
+        SET_DF(s3, d_tst8);
    } else {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, df)); // reset flags
+        SET_DFNONE(s4);
    }
    IFX(X_OF) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_OF]));
+        BFC(xFlags, F_OF, 1);
    }
-    IFX(X_CF) {
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_CF]));
+    IFX(X_ZF|X_CF) {
+        BIC_IMM8(xFlags, xFlags, (1<<F_ZF)|(1<<F_CF), 0);
    }
    ANDS_REG_LSL_IMM5(s3, s1, s2, 0);   // res = s1 & s2
-    IFX(X_PEND) {
+    IFX_PENDOR0 {
        STR_IMM9(s3, xEmu, offsetof(x86emu_t, res));
    }
    IFX(X_ZF) {
-        MOVW_COND(cEQ, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_ZF]));
+        ORR_IMM8_COND(cEQ, xFlags, xFlags, 1<<F_ZF, 0);
    }
    IFX(X_SF) {
-        UBFX(s4, s3, 7, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_SF]));
+        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(s3, s3, 0xE0); // lsr 5 masking pre-applied
-        LDR_IMM9(s4, xEmu, offsetof(x86emu_t, x86emu_parity_tab));
+        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);
-        AND_IMM8(s4, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_PF]));
+        BFI(xFlags, s4, F_PF, 1);
    }
 }
@@ -23,17 +23,17 @@
 #include "dynarec_arm_helper.h"

 // Get Ex as a double, not a quad (warning, x2 and x3 may get used)
-#define GETEX(a) \
-    if((nextop&0xC0)==0xC0) { \
-        a = sse_get_reg(dyn, ninst, x1, nextop&7); \
+#define GETEX(a, w)                                 \
+    if((nextop&0xC0)==0xC0) {                       \
+        a = sse_get_reg(dyn, ninst, x1, nextop&7,w);\
    } else {    \
        parity = getedparity(dyn, ninst, addr, nextop, 3);  \
        a = fpu_get_scratch_double(dyn);            \
        if(parity) {                                \
-            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3); \
+            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0); \
            VLDR_64(a, ed, fixedaddress);           \
        } else {                                    \
-            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0);\
+            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0);\
            LDR_IMM9(x2, ed, fixedaddress+0);       \
            LDR_IMM9(x3, ed, fixedaddress+4);       \
            VMOVtoV_D(a, x2, x3);                   \
@@ -58,17 +58,18 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
    MAYUSE(d0);

    switch(nextop) {
-        
+
        case 0x10:
            INST_NAME("MOVSD Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
            if((nextop&0xC0)==0xC0) {
-                d0 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+                d0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                VMOVD(v0, d0);
            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0);
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0);
                LDR_IMM9(x2, ed, fixedaddress+0);
                LDR_IMM9(x3, ed, fixedaddress+4);
                VMOVtoV_D(v0, x2, x3);
@@ -79,13 +80,13 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("MOVSD Ex, Gx");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
            if((nextop&0xC0)==0xC0) {
-                d0 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                d0 = sse_get_reg(dyn, ninst, x1, nextop&7, 1);
                VMOVD(d0, v0);
            } else {
                VMOVfrV_D(x2, x3, v0);
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0);
                STR_IMM9(x2, ed, fixedaddress+0);
                STR_IMM9(x3, ed, fixedaddress+4);
            }
@@ -94,25 +95,25 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("MOVDDUP Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
            if((nextop&0xC0)==0xC0) {
-                d0 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                d0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                VMOVD(v0, d0);
-                VMOVD(v0+1, d0);
            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0);
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0);
                LDR_IMM9(x2, ed, fixedaddress+0);
                LDR_IMM9(x3, ed, fixedaddress+4);
                VMOVtoV_D(v0, x2, x3);
-                VMOVD(v0+1, v0);
            }
+            VMOVD(v0+1, v0);
            break;

        case 0x2A:
            INST_NAME("CVTSI2SD Gx, Ed");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
            GETED;
            s0 = fpu_get_scratch_single(dyn);
            VMOVtoV(s0, ed);
@@ -125,9 +126,9 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            GETGD;
            s0 = fpu_get_scratch_single(dyn);
            if((nextop&0xC0)==0xC0) {
-                d0 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                d0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0);
                d0 = fpu_get_scratch_double(dyn);
                VLDR_64(d0, ed, fixedaddress);
            }
@@ -136,14 +137,14 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            break;
        case 0x2D:
            INST_NAME("CVTSD2SI Gd, Ex");
-            u8 = x87_setround(dyn, ninst, x1, x2, x12);
+            u8 = sse_setround(dyn, ninst, x1, x2, x14);
            nextop = F8;
            GETGD;
            s0 = fpu_get_scratch_single(dyn);
            if((nextop&0xC0)==0xC0) {
-                d0 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                d0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0);
                d0 = fpu_get_scratch_double(dyn);
                VLDR_64(d0, ed, fixedaddress);
            }
@@ -156,8 +157,8 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("SQRTSD Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
-            GETEX(d0);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+            GETEX(d0, 0);
            VSQRT_F64(v0, d0);
            break;

@@ -165,80 +166,82 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("ADDSD Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
-            GETEX(d0);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+            GETEX(d0, 0);
            VADD_F64(v0, v0, d0);
            break;
        case 0x59:
            INST_NAME("MULSD Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
-            GETEX(d0);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+            GETEX(d0, 0);
            VMUL_F64(v0, v0, d0);
            break;
        case 0x5A:
            INST_NAME("CVTSD2SS Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
-            GETEX(d0);
-            d1 = fpu_get_scratch_double(dyn);
-            s0 = fpu_get_scratch_single(dyn);
-            VCVT_F32_F64(s0, d0);
-            VMOV_64(d1, v0);
-            VMOV_32(d1*2, s0);
-            VMOV_64(v0, d1);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+            GETEX(d0, 0);
+            if(v0<16) {
+                VCVT_F32_F64(v0*2, d0);
+            } else {
+                d1 = fpu_get_scratch_double(dyn);
+                VMOV_64(d1, v0);
+                VCVT_F32_F64(d1*2, d0);
+                VMOV_64(v0, d1);
+            }
            break;

        case 0x5C:
            INST_NAME("SUBSD Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
-            GETEX(d0);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+            GETEX(d0, 0);
            VSUB_F64(v0, v0, d0);
            break;
        case 0x5D:
            INST_NAME("MINSD Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
-            GETEX(d0);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+            GETEX(d0, 0);
            // MINSD: if any input is NaN, or Ex[0]<Gx[0], copy Ex[0] -> Gx[0]
            VCMP_F64(v0, d0);
            VMRS_APSR();
-            VMOVcond_64(cPL, v0, d0);
+            VMOVcond_64(cCS, v0, d0); // move v0 <- d0 if v0 or d0 unordered or v0 >= d0
            break;
        case 0x5E:
            INST_NAME("DIVSD Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
-            GETEX(d0);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+            GETEX(d0, 0);
            VDIV_F64(v0, v0, d0);
            break;
        case 0x5F:
            INST_NAME("MAXSD Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
-            GETEX(d0);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+            GETEX(d0, 0);
            // MAXSD: if any input is NaN, or Ex[0]>Gx[0], copy Ex[0] -> Gx[0]
            VCMP_F64(d0, v0);
            VMRS_APSR();
-            VMOVcond_64(cPL, v0, d0);
+            VMOVcond_64(cCS, v0, d0); // move v0 <- d0 if v0 or d0 is unordered or d0 >= v0
            break;

        case 0x70:
            INST_NAME("PSHUFLW Gx, Ex, Ib");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
            if((nextop&0xC0)==0xC0) {
-                v1 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
                v1 = fpu_get_scratch_quad(dyn);
                VLD1Q_64(v1, ed);
            }
@@ -269,11 +272,11 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("HADDPS Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
            if((nextop&0xC0)==0xC0) {
-                v1 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
                v1 = fpu_get_scratch_quad(dyn);
                VLD1Q_64(v1, ed);
            }
@@ -289,25 +292,45 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("CMPSD Gx, Ex");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
-            GETEX(d0);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+            GETEX(d0, 0);
+            u8 = F8;
            VCMP_F64(v0, d0);
            VMRS_APSR();
            MOVW(x2, 0);
-            u8 = F8;
            switch(u8&7) {
                case 0: MVN_COND_REG_LSL_IMM5(cEQ, x2, x2, 0); break;   // Equal
                case 1: MVN_COND_REG_LSL_IMM5(cCC, x2, x2, 0); break;   // Less than
                case 2: MVN_COND_REG_LSL_IMM5(cLS, x2, x2, 0); break;   // Less or equal
                case 3: MVN_COND_REG_LSL_IMM5(cVS, x2, x2, 0); break;   // NaN
-                case 4: MVN_COND_REG_LSL_IMM5(cNE, x2, x2, 0); break;   // Not Equal (or unordered on ARM, not on X86...)
-                case 5: MVN_COND_REG_LSL_IMM5(cPL, x2, x2, 0); break;   // Greater or equal or unordered
-                case 6: MVN_COND_REG_LSL_IMM5(cHI, x2, x2, 0); break;   // Greater or unordered
+                case 4: MVN_COND_REG_LSL_IMM5(cNE, x2, x2, 0); break;   // Not Equal or unordered
+                case 5: MVN_COND_REG_LSL_IMM5(cCS, x2, x2, 0); break;   // Greater or equal or unordered
+                case 6: MVN_COND_REG_LSL_IMM5(cHI, x2, x2, 0); break;   // Greater or unordered, test inverted, N!=V so unordereded or less than (inverted)
                case 7: MVN_COND_REG_LSL_IMM5(cVC, x2, x2, 0); break;   // not NaN
            }
            VMOVtoV_D(v0, x2, x2);
            break;

+        case 0xF0:
+            INST_NAME("LDDQ Gx,Ex");
+            nextop = F8;
+            gd = (nextop&0x38)>>3;
+            if((nextop&0xC0)==0xC0) {
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                VMOVQ(v0, v1);
+            } else {
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-12, 0, 0);
+                LDR_IMM9(x2, ed, fixedaddress+0);
+                LDR_IMM9(x3, ed, fixedaddress+4);
+                VMOVtoV_D(v0, x2, x3);
+                LDR_IMM9(x2, ed, fixedaddress+8);
+                LDR_IMM9(x3, ed, fixedaddress+12);
+                VMOVtoV_D(v0+1, x2, x3);
+            }
+            break;
+
        default:
            DEFAULT;
    }
@@ -23,9 +23,9 @@
 #include "dynarec_arm_helper.h"

 // Get Ex as a single, not a quad, with b a intermediary double
-#define GETEX(a, b) \
-    if((nextop&0xC0)==0xC0) { \
-        b = sse_get_reg(dyn, ninst, x1, nextop&7);  \
+#define GETEX(a, b)                                 \
+    if((nextop&0xC0)==0xC0) {                       \
+        b = sse_get_reg(dyn, ninst, x1, nextop&7,0);\
        if(b<16)                                    \
            a = b;                                  \
        else {                                      \
@@ -38,7 +38,7 @@
        a = b*2;                                    \
        parity = getedparity(dyn, ninst, addr, nextop, 2);  \
        if(parity) {                                \
-            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 0); \
+            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 0, 0); \
            VLDR_32(a, ed, fixedaddress);           \
        } else {                                    \
            GETED;                                  \
@@ -46,16 +46,16 @@
        }                                           \
    }

-#define GETGX(a) \
-    gd = (nextop&0x38)>>3;                          \
-    a = sse_get_reg(dyn, ninst, x1, gd)
+#define GETGX(a, w)                         \
+    gd = (nextop&0x38)>>3;                  \
+    a = sse_get_reg(dyn, ninst, x1, gd, w)


 uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog)
 {
    uint8_t opcode = F8;
    uint8_t nextop, u8;
-    int32_t i32;
+    int32_t j32;
    uint32_t u32;
    uint8_t gd, ed;
    uint8_t wback;
@@ -63,12 +63,13 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
    int v0, v1;
    int s0, s1, s2;
    int d0, d1;
-    int q0;
+    int q0, q1;
    int parity;

+    MAYUSE(q1);
    MAYUSE(s1);
    MAYUSE(s2);
-    MAYUSE(i32);
+    MAYUSE(j32);

    switch(opcode) {

@@ -77,8 +78,8 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            nextop = F8;
            gd = (nextop&0x38)>>3;
            if((nextop&0xC0)==0xC0) {
-                v0 = sse_get_reg(dyn, ninst, x1, gd);
-                q0 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+                q0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                if(v0<16)
                    d0 = v0;
                else {
@@ -97,7 +98,7 @@ 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);
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0, 0);
                LDR_IMM9(x2, ed, fixedaddress);   // to avoid bus errors
                VEORQ(v0, v0, v0);
                VMOVtoDx_32(v0, 0, x2);
@@ -107,9 +108,9 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("MOVSS Ex, Gx");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
            if((nextop&0xC0)==0xC0) {
-                q0 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                q0 = sse_get_reg(dyn, ninst, x1, nextop&7, 1);
                if(v0<16)
                    d0 = v0;
                else {
@@ -128,10 +129,42 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                }
            } else {
                VMOVfrDx_32(x2, v0, 0);
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0, 0);
                STR_IMM9(x2, ed, fixedaddress);
            }
            break;
+        case 0x12:
+            INST_NAME("MOVSLDUP Gx, Ex");
+            nextop = F8;
+            gd = (nextop&0x38)>>3;
+            if((nextop&0xC0)==0xC0) {
+                q1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
+                q1 = fpu_get_scratch_quad(dyn);
+                VLD1Q_64(q1, ed);
+            }
+            q0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+            VDUP_32(q0+0, q1+0, 0);
+            VDUP_32(q0+1, q1+1, 0);
+            break;
+        
+        case 0x16:
+            INST_NAME("MOVSHDUP Gx, Ex");
+            nextop = F8;
+            gd = (nextop&0x38)>>3;
+            if((nextop&0xC0)==0xC0) {
+                q1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
+                q1 = fpu_get_scratch_quad(dyn);
+                VLD1Q_64(q1, ed);
+            }
+            q0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+            VDUP_32(q0+0, q1+0, 1);
+            VDUP_32(q0+1, q1+1, 1);
+            break;
+
        
        case 0x1E:
            INST_NAME("NOP / ENDBR32 / ENDBR64");
@@ -148,7 +181,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("CVTSI2SS Gx, Ed");
            nextop = F8;
            gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
            GETED;
            if(v0<16)
                d0 = v0;
@@ -170,7 +203,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            GETGD;
            s0 = fpu_get_scratch_single(dyn);
            if((nextop&0xC0)==0xC0) {
-                v0 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                v0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                if(v0<16)
                    d0 = v0;
                else {
@@ -179,20 +212,27 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                }
                VMOV_32(s0, d0*2);
            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3);
-                VLDR_32(s0, ed, fixedaddress);
+                parity = getedparity(dyn, ninst, addr, nextop, 3);
+                if(parity) {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0);
+                    VLDR_32(s0, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0, 0);
+                    LDR_IMM9(x2, ed, fixedaddress);
+                    VMOVtoV(s0, x2);
+                }
            }
            VCVT_S32_F32(s0, s0);
            VMOVfrV(gd, s0);
            break;
        case 0x2D:
            INST_NAME("CVTSS2SI Gd, Ex");
-            u8 = x87_setround(dyn, ninst, x1, x2, x12);
+            u8 = sse_setround(dyn, ninst, x1, x2, x14);
            nextop = F8;
            GETGD;
            s0 = fpu_get_scratch_single(dyn);
            if((nextop&0xC0)==0xC0) {
-                v0 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                v0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                if(v0<16)
                    d0 = v0;
                else {
@@ -201,8 +241,15 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                }
                VMOV_32(s0, d0*2);
            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3);
-                VLDR_32(s0, ed, fixedaddress);
+                parity = getedparity(dyn, ninst, addr, nextop, 3);
+                if(parity) {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0);
+                    VLDR_32(s0, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0, 0);
+                    LDR_IMM9(x2, ed, fixedaddress);
+                    VMOVtoV(s0, x2);
+                }
            }
            VCVTR_S32_F32(s0, s0);
            VMOVfrV(gd, s0);
@@ -212,7 +259,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
        case 0x51:
            INST_NAME("SQRTSS Gx, Ex");
            nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
            GETEX(s1, v1);
            if(v0==v1)
                d0 = s1/2;
@@ -232,7 +279,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
        case 0x52:
            INST_NAME("RSQRTSS Gx, Ex");
            nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
            GETEX(s1, v1);
            if(v0==v1)
                d0 = s1/2;
@@ -260,7 +307,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
        case 0x53:
            INST_NAME("RCPSS Gx, Ex");
            nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
            GETEX(s1, v1);
            if(v0==v1)
                d0 = s1/2;
@@ -283,7 +330,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
        case 0x58:
            INST_NAME("ADDSS Gx, Ex");
            nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
            GETEX(s1, v1);
            if(v0==v1)
                d0 = s1/2;
@@ -303,7 +350,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
        case 0x59:
            INST_NAME("MULSS Gx, Ex");
            nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
            GETEX(s1, v1);
            if(v0==v1)
                d0 = s1/2;
@@ -323,7 +370,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
        case 0x5A:
            INST_NAME("CVTSS2SD Gx, Ex");
            nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
            GETEX(s0, d0);
            VCVT_F64_F32(v0, s0);
            break;
@@ -331,9 +378,9 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("CVTTPS2DQ Gx, Ex");
            nextop = F8;
            if((nextop&0xC0)==0xC0) {
-                v1 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
                v1 = fpu_get_scratch_quad(dyn);
                VLD1Q_64(v1, ed);
            }
@@ -344,7 +391,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
        case 0x5C:
            INST_NAME("SUBSS Gx, Ex");
            nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
            GETEX(s1, v1);
            if(v0==v1)
                d0 = s1/2;
@@ -364,7 +411,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
        case 0x5D:
            INST_NAME("MINSS Gx, Ex");
            nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
            GETEX(s1, v1);
            if(v0==v1)
                d0 = s1/2;
@@ -379,7 +426,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            // MINSS: if any input is NaN, or Ex[0]<Gx[0], copy Ex[0] -> Gx[0]
            VCMP_F32(s0, s1);
            VMRS_APSR();
-            VMOVcond_32(cPL, s0, s1);
+            VMOVcond_32(cCS, s0, s1); // move s0 <- s1 if s0 or s1 unordered or s0 >= s1
            if(v0!=d0) {
                VMOV_64(v0, d0);
            }
@@ -387,7 +434,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
        case 0x5E:
            INST_NAME("DIVSS Gx, Ex");
            nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
            GETEX(s1, v1);
            if(v0==v1)
                d0 = s1/2;
@@ -407,7 +454,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
        case 0x5F:
            INST_NAME("MAXSS Gx, Ex");
            nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
            GETEX(s1, v1);
            if(v0==v1)
                d0 = s1/2;
@@ -422,7 +469,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            // MAXSS: if any input is NaN, or Ex[0]>Gx[0], copy Ex[0] -> Gx[0]
            VCMP_F32(s1, s0);
            VMRS_APSR();
-            VMOVcond_32(cPL, s0, s1);
+            VMOVcond_32(cCS, s0, s1); // move s0 <- s1 if s0 or s1 is unordered or s1 >= s0
            if(v0!=d0) {
                VMOV_64(v0, d0);
            }
@@ -433,12 +480,12 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            nextop = F8;
            gd = (nextop&0x38)>>3;
            if((nextop&0xC0)==0xC0) {
-                v1 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                VMOVQ(v0, v1);
            } else {
                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-12, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-12, 0, 0);
                LDR_IMM9(x2, ed, fixedaddress+0);
                LDR_IMM9(x3, ed, fixedaddress+4);
                VMOVtoV_D(v0, x2, x3);
@@ -451,16 +498,16 @@ 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);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
            if((nextop&0xC0)==0xC0) {
-                v1 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
            } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
                v1 = fpu_get_scratch_quad(dyn);
                VLD1Q_64(v1, ed);
            }
            u8 = F8;
-            // only high part need to be suffled. VTBL only handle 8bits value, so the 16bits suffles need to be changed in 8bits
+            // only high part need to be shuffled. VTBL only handle 8bits value, so the 16bits shuffles need to be changed in 8bits
            u32 = 0;
            for (int i=0; i<2; ++i) {
                u32 |= (((u8>>(i*2))&3)*2+0)<<(i*16+0);
@@ -487,12 +534,12 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            nextop = F8;
            gd = (nextop&0x38)>>3;
            if((nextop&0xC0)==0xC0) {
-                v1 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                VMOVD(v0, v1);
            } else {
                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0);
                VLD1_64(v0, ed);
            }
            VEOR(v0+1, v0+1, v0+1);
@@ -501,13 +548,12 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
            INST_NAME("MOVDQU Ex,Gx");
            nextop = F8;
            gd = (nextop&0x38)>>3;
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
            if((nextop&0xC0)==0xC0) {
                v1 = sse_get_reg_empty(dyn, ninst, x1, nextop&7);
-                v0 = sse_get_reg(dyn, ninst, x1, gd);
                VMOVQ(v1, v0);
            } else {
-                v0 = sse_get_reg(dyn, ninst, x1, gd);
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-12, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-12, 0, 0);
                VMOVfrV_D(x2, x3, v0);
                STR_IMM9(x2, ed, fixedaddress+0);
                STR_IMM9(x3, ed, fixedaddress+4);
@@ -519,26 +565,27 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins

        case 0xBC:
            INST_NAME("TZCNT Gd, Ed");
-            SETFLAGS(X_CF|X_ZF, SF_SET);
+            SETFLAGS(X_CF|X_ZF, SF_SUBSET);
+            SET_DFNONE(x3);
            nextop = F8;
            GETED;
            GETGD;
            TSTS_REG_LSL_IMM5(ed, ed, 0);
-            RBIT(x1, ed);   // reverse
-            CLZ(gd, x1);    // x2 gets leading 0 == TZCNT
-            MOVW_COND(cEQ, x1, 1);
-            MOVW_COND(cNE, x1, 0);
-            STR_IMM9(x1, xEmu, offsetof(x86emu_t, flags[F_CF]));
-            RSB_IMM8(x1, x1, 1);
-            STR_IMM9(x1, xEmu, offsetof(x86emu_t, flags[F_ZF]));
-            MOVW(x1, d_none);
-            STR_IMM9(x1, xEmu, offsetof(x86emu_t, df));
+            MOVW_COND(cEQ, x3, 1);
+            MOVW_COND(cNE, x3, 0);
+            BFI(xFlags, x3, F_CF, 1);   // CF = is source 0?
+            RBIT(x3, ed);   // reverse
+            CLZ(gd, x3);    // x2 gets leading 0 == TZCNT
+            TSTS_REG_LSL_IMM5(gd, gd, 0);
+            MOVW_COND(cEQ, x3, 1);
+            MOVW_COND(cNE, x3, 0);
+            BFI(xFlags, x3, F_ZF, 1);   // ZF = is dest 0?
            break;

        case 0xC2:
            INST_NAME("CMPSS Gx, Ex");
            nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
            GETEX(s1, v1);
            if(v0==v1)
                d0 = s1/2;
@@ -550,22 +597,53 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                    VMOV_64(d0, v0);
                }
            s0 = d0*2;
+            u8 = F8;
            VCMP_F32(s0, s1);
            VMRS_APSR();
            MOVW(x2, 0);
-            u8 = F8;
            switch(u8&7) {
                case 0: MVN_COND_REG_LSL_IMM5(cEQ, x2, x2, 0); break;   // Equal
                case 1: MVN_COND_REG_LSL_IMM5(cCC, x2, x2, 0); break;   // Less than
                case 2: MVN_COND_REG_LSL_IMM5(cLS, x2, x2, 0); break;   // Less or equal
                case 3: MVN_COND_REG_LSL_IMM5(cVS, x2, x2, 0); break;   // NaN
-                case 4: MVN_COND_REG_LSL_IMM5(cNE, x2, x2, 0); break;   // Not Equal (or unordered on ARM, not on X86...)
-                case 5: MVN_COND_REG_LSL_IMM5(cPL, x2, x2, 0); break;   // Greater or equal or unordered
+                case 4: MVN_COND_REG_LSL_IMM5(cNE, x2, x2, 0); break;   // Not Equal or unordered
+                case 5: MVN_COND_REG_LSL_IMM5(cCS, x2, x2, 0); break;   // Greater or equal or unordered
                case 6: MVN_COND_REG_LSL_IMM5(cHI, x2, x2, 0); break;   // Greater or unordered
                case 7: MVN_COND_REG_LSL_IMM5(cVC, x2, x2, 0); break;   // not NaN
            }
            VMOVtoDx_32(v0, 0, x2);
            break;
+        
+        case 0xE6:
+            INST_NAME("CVTDQ2PD Gx, Ex");
+            nextop = F8;
+            if((nextop&0xC0)==0xC0) {
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
+                if(v1<16)
+                    v0 = v1;
+                else {
+                    v0 = fpu_get_scratch_double(dyn);
+                    VMOV_64(v0, v1);    // vfpu opcode here
+                }
+            } else {
+                v0 = fpu_get_scratch_double(dyn);
+                parity = getedparity(dyn, ninst, addr, nextop, 3);
+                if(parity) {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0);
+                    VLDR_64(v0, ed, fixedaddress);    // vfpu opcode here
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-8, 0, 0);
+                    LDR_IMM9(x2, ed, fixedaddress);
+                    LDR_IMM9(x3, ed, fixedaddress+4);
+                    VMOVtoV_D(v0, x2, x3);
+                }
+            }
+            gd = (nextop&0x38)>>3;
+            q0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+            //v0 is a low reg now
+            VCVT_F64_S32(q0+1, v0*2+1);
+            VCVT_F64_S32(q0+0, v0*2+0);
+            break;

        default:
            DEFAULT;
@@ -21,20 +21,15 @@
 #include "emu/x86run_private.h"
 #include "emu/x87emu_private.h"
 #include "x86trace.h"
+#include "signals.h"
 #include "dynarec_arm.h"
 #include "dynarec_arm_private.h"
 #include "dynarec_arm_functions.h"
-
-void arm_popf(x86emu_t* emu, uint32_t f)
-{
-    emu->packed_eflags.x32 = (f & 0x3F7FD7) | 0x2; // mask off res2 and res3 and on res1
-    UnpackFlags(emu);
-    RESET_FLAGS(emu);
-}
+#include "bridge.h"

 void arm_fstp(x86emu_t* emu, void* p)
 {
-    if(ST0.ll!=STld(0).ref)
+    if(ST0.q!=STld(0).uref)
        D2LD(&ST0.d, p);
    else
        memcpy(p, &STld(0).ld, 10);
@@ -56,6 +51,7 @@ void arm_fyl2x(x86emu_t* emu)
 void arm_ftan(x86emu_t* emu)
 {
    ST0.d = tan(ST0.d);
+    emu->sw.f.F87_C2 = 0;
 }
 void arm_fpatan(x86emu_t* emu)
 {
@@ -63,7 +59,7 @@ void arm_fpatan(x86emu_t* emu)
 }
 void arm_fxtract(x86emu_t* emu)
 {
-    int32_t tmp32s = (ST1.ll&0x7ff0000000000000LL)>>52;
+    int32_t tmp32s = (ST1.q&0x7ff0000000000000LL)>>52;
    tmp32s -= 1023;
    ST1.d /= exp2(tmp32s);
    ST0.d = tmp32s;
@@ -77,31 +73,15 @@ void arm_fprem(x86emu_t* emu)
    emu->sw.f.F87_C3 = ((tmp32s>>1)&1);
    emu->sw.f.F87_C1 = ((tmp32s>>2)&1);
 }
-void arm_fyl2xp1(x86emu_t* emu)
-{
-    ST(1).d = log2(ST0.d + 1.0)*ST(1).d;
-}
-void arm_fsincos(x86emu_t* emu)
-{
-    sincos(ST1.d, &ST1.d, &ST0.d);
-}
 void arm_frndint(x86emu_t* emu)
 {
    ST0.d = fpu_round(emu, ST0.d);
 }
 void arm_fscale(x86emu_t* emu)
 {
-    ST0.d *= exp2(trunc(ST1.d));
+    if(ST0.d!=0.0)
+        ST0.d *= exp2(trunc(ST1.d));
 }
-void arm_fsin(x86emu_t* emu)
-{
-    ST0.d = sin(ST0.d);
-}
-void arm_fcos(x86emu_t* emu)
-{
-    ST0.d = cos(ST0.d);
-}
-
 void arm_fbld(x86emu_t* emu, uint8_t* ed)
 {
    fpu_fbld(emu, ed);
@@ -109,9 +89,11 @@ void arm_fbld(x86emu_t* emu, uint8_t* ed)

 void arm_fild64(x86emu_t* emu, int64_t* ed)
 {
-    ST0.d = *ed;
-    STll(0).ll = *ed;
-    STll(0).ref = ST0.ll;
+    int64_t tmp;
+    memcpy(&tmp, ed, sizeof(tmp));
+    ST0.d = tmp;
+    STll(0).sq = tmp;
+    STll(0).sref = ST0.sq;
 }

 void arm_fbstp(x86emu_t* emu, uint8_t* ed)
@@ -121,51 +103,36 @@ void arm_fbstp(x86emu_t* emu, uint8_t* ed)

 void arm_fistp64(x86emu_t* emu, int64_t* ed)
 {
-    if(STll(0).ref==ST(0).ll) {
-        *ed = STll(0).ll;
+    // used of memcpy to avoid aligments issues
+    if(STll(0).sref==ST(0).sq) {
+        memcpy(ed, &STll(0).sq, sizeof(int64_t));
    } else {
-        if(isgreater(ST0.d, (double)(int64_t)0x7fffffffffffffffLL) || isless(ST0.d, (double)(int64_t)0x8000000000000000LL))
-            *ed = 0x8000000000000000LL;
+        int64_t tmp;
+        if(isgreater(ST0.d, (double)(int64_t)0x7fffffffffffffffLL) || isless(ST0.d, (double)(int64_t)0x8000000000000000LL) || !isfinite(ST0.d))
+            tmp = 0x8000000000000000LL;
        else
-            *ed = (int64_t)ST0.d;
+            tmp = fpu_round(emu, ST0.d);
+        memcpy(ed, &tmp, sizeof(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);
    LD2D(&STld(0), &ST(0).d);
-    STld(0).ref = ST0.ll;
+    STld(0).uref = ST0.q;
 }

 void arm_ud(x86emu_t* emu)
 {
-    kill(getpid(), SIGILL);
-}
-
-void arm_fxsave(x86emu_t* emu, uint8_t* ed)
-{
-    // should save flags & all
-    // copy MMX regs...
-    for(int i=0; i<8; ++i)
-        memcpy(ed+32+i*16, &emu->mmx[0], sizeof(emu->mmx[0]));
-    // copy SSE regs
-    memcpy(ed+160, &emu->xmm[0], sizeof(emu->xmm));
-    // put also FPU regs in a reserved area...
-    for(int i=0; i<8; ++i)
-        memcpy(ed+416+i*8, &emu->fpu[0], sizeof(emu->fpu[0]));
-}
-
-void arm_fxrstor(x86emu_t* emu, uint8_t* ed)
-{
-    // should restore flags & all
-    // copy back MMX regs...
-    for(int i=0; i<8; ++i)
-        memcpy(&emu->mmx[i], ed+32+i*16, sizeof(emu->mmx[0]));
-    // copy SSE regs
-    memcpy(&emu->xmm[0], ed+160/4, sizeof(emu->xmm));
-    for(int i=0; i<8; ++i)
-        memcpy(&emu->fpu[0], ed+416+i*8, sizeof(emu->fpu[0]));
+    emit_signal(emu, SIGILL, (void*)R_EIP, 0);
 }

 void arm_fsave(x86emu_t* emu, uint8_t* ed)
@@ -175,12 +142,7 @@ void arm_fsave(x86emu_t* emu, uint8_t* ed)
    uint8_t* p = ed;
    p += 28;
    for (int i=0; i<8; ++i) {
-        #ifdef USE_FLOAT
-        d = ST(i).f;
-        LD2D(p, &d);
-        #else
        LD2D(p, &ST(i).d);
-        #endif
        p+=10;
    }
 }
@@ -191,17 +153,23 @@ void arm_frstor(x86emu_t* emu, uint8_t* ed)
    uint8_t* p = ed;
    p += 28;
    for (int i=0; i<8; ++i) {
-        #ifdef USE_FLOAT
-        D2LD(&d, p);
-        ST(i).f = d;
-        #else
        D2LD(&ST(i).d, p);
-        #endif
        p+=10;
    }

 }

+void arm_fprem1(x86emu_t* emu)
+{
+    // simplified version
+    int32_t tmp32s = round(ST0.d / ST1.d);
+    ST0.d -= ST1.d*tmp32s;
+    emu->sw.f.F87_C2 = 0;
+    emu->sw.f.F87_C0 = (tmp32s&1);
+    emu->sw.f.F87_C3 = ((tmp32s>>1)&1);
+    emu->sw.f.F87_C1 = ((tmp32s>>2)&1);
+}
+

 // Get a FPU single scratch reg
 int fpu_get_scratch_single(dynarec_arm_t* dyn)
@@ -218,6 +186,13 @@ int fpu_get_scratch_double(dynarec_arm_t* dyn)
 // Get a FPU quad scratch reg
 int fpu_get_scratch_quad(dynarec_arm_t* dyn)
 {
+    if(dyn->fpu_scratch>4) {
+        if(dyn->fpu_extra_qscratch) {
+            dynarec_log(LOG_NONE, "Warning, Extra QScratch slot taken and need another one!\n");
+        } else
+            dyn->fpu_extra_qscratch = fpu_get_reg_quad(dyn, NEON_CACHE_SCR, 0);
+        return dyn->fpu_extra_qscratch;
+    }
    int i = (dyn->fpu_scratch+3)&(~3);
    dyn->fpu_scratch = i+4;
    return i/2; // Return a Dx, not a Qx
@@ -226,15 +201,20 @@ int fpu_get_scratch_quad(dynarec_arm_t* dyn)
 void fpu_reset_scratch(dynarec_arm_t* dyn)
 {
    dyn->fpu_scratch = 0;
+    if(dyn->fpu_extra_qscratch) {
+        fpu_free_reg_quad(dyn, dyn->fpu_extra_qscratch);
+        dyn->fpu_extra_qscratch = 0;
+    }
 }
-#define FPUFIRST    8
 // Get a FPU double reg
-int fpu_get_reg_double(dynarec_arm_t* dyn)
+int fpu_get_reg_double(dynarec_arm_t* dyn, unsigned int t, unsigned int n)
 {
    // TODO: check upper limit?
    int i=0;
    while (dyn->fpuused[i]) ++i;
    dyn->fpuused[i] = 1;
+    dyn->n.neoncache[i].n = n;
+    dyn->n.neoncache[i].t = t;
    return i+FPUFIRST; // return a Dx
 }
 // Free a FPU double reg
@@ -243,13 +223,19 @@ void fpu_free_reg_double(dynarec_arm_t* dyn, int reg)
    // TODO: check upper limit?
    int i=reg-FPUFIRST;
    dyn->fpuused[i] = 0;
+    if(dyn->n.neoncache[i].t!=NEON_CACHE_ST_F && dyn->n.neoncache[i].t!=NEON_CACHE_ST_D)
+        dyn->n.neoncache[i].v = 0;
 }
 // Get a FPU quad reg
-int fpu_get_reg_quad(dynarec_arm_t* dyn)
+int fpu_get_reg_quad(dynarec_arm_t* dyn, unsigned int t, unsigned int n)
 {
    int i=0;
    while (dyn->fpuused[i] || dyn->fpuused[i+1]) i+=2;
    dyn->fpuused[i] = dyn->fpuused[i+1] = 1;
+    dyn->n.neoncache[i].t = t;
+    dyn->n.neoncache[i].n = n;
+    dyn->n.neoncache[i+1].t = t;
+    dyn->n.neoncache[i+1].n = n;
    return i+FPUFIRST; // Return a Dx, not a Qx
 }
 // Free a FPU quad reg
@@ -257,20 +243,171 @@ void fpu_free_reg_quad(dynarec_arm_t* dyn, int reg)
 {
    int i=reg-FPUFIRST;
    dyn->fpuused[i] = dyn->fpuused[i+1] = 0;
+    dyn->n.neoncache[i].v = 0;
+    dyn->n.neoncache[i+1].v = 0;
 }
 // Reset fpu regs counter
 void fpu_reset_reg(dynarec_arm_t* dyn)
 {
    dyn->fpu_reg = 0;
-    for (int i=0; i<24; ++i)
+    for (int i=0; i<24; ++i) {
        dyn->fpuused[i]=0;
+        dyn->n.neoncache[i].v = 0;
+    }
 }

+int neoncache_get_st(dynarec_arm_t* dyn, int ninst, int a)
+{
+    if(!dyn->insts)
+        return NEON_CACHE_ST_D;
+    for(int i=0; i<24; ++i)
+        if((dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_F
+         || dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_D)
+         && dyn->insts[ninst].n.neoncache[i].n==a)
+            return dyn->insts[ninst].n.neoncache[i].t;
+    // not in the cache yet, so will be fetched...
+    return NEON_CACHE_ST_D;
+}
+
+int neoncache_get_current_st(dynarec_arm_t* dyn, int ninst, int a)
+{
+    (void)ninst;
+    if(!dyn->insts)
+        return NEON_CACHE_ST_D;
+    for(int i=0; i<24; ++i)
+        if((dyn->n.neoncache[i].t==NEON_CACHE_ST_F
+         || dyn->n.neoncache[i].t==NEON_CACHE_ST_D)
+         && dyn->n.neoncache[i].n==a)
+            return dyn->n.neoncache[i].t;
+    // not in the cache yet, so will be fetched...
+    return NEON_CACHE_ST_D;
+}
+
+int neoncache_get_st_f(dynarec_arm_t* dyn, int ninst, int a, int st)
+{
+    if(a+dyn->insts[ninst].n.stack_next-st<0)
+        // The STx has been pushed at the end of instructon, so stop going back
+        return -1;
+    for(int i=0; i<24; ++i)
+        if(dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_F
+         && dyn->insts[ninst].n.neoncache[i].n==a+dyn->insts[ninst].n.stack-st)
+            return i;
+    return -1;
+} 
+int neoncache_get_st_f_noback(dynarec_arm_t* dyn, int ninst, int a, int st)
+{
+    for(int i=0; i<24; ++i)
+        if(dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_F
+         && dyn->insts[ninst].n.neoncache[i].n==a+dyn->insts[ninst].n.stack-st)
+            return i;
+    return -1;
+} 
+int neoncache_get_current_st_f(dynarec_arm_t* dyn, int a, int st)
+{
+    for(int i=0; i<24; ++i)
+        if(dyn->n.neoncache[i].t==NEON_CACHE_ST_F
+         && dyn->n.neoncache[i].n==a+dyn->n.stack-st)
+            return i;
+    return -1;
+} 
+static void neoncache_promote_double_forward(dynarec_arm_t* dyn, int ninst, int maxinst, int a, int st);
+static void neoncache_promote_double_internal(dynarec_arm_t* dyn, int ninst, int maxinst, int a, int st);
+static void neoncache_promote_double_combined(dynarec_arm_t* dyn, int ninst, int maxinst, int a, int st)
+{
+    if(a == dyn->insts[ninst].n.combined1) {
+        int i = neoncache_get_st_f_noback(dyn, ninst, dyn->insts[ninst].n.combined2, st);
+        //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double_internal, ninst=%d combined2 %d i=%d\n", ninst, dyn->insts[ninst].n.combined2 ,i);
+        if(i>=0) {
+            dyn->insts[ninst].n.neoncache[i].t = NEON_CACHE_ST_D;
+            neoncache_promote_double_internal(dyn, ninst-1, maxinst, dyn->insts[ninst].n.combined2, st);
+            // go forward is combined is not pop'd
+            if(dyn->insts[ninst].n.stack_next-dyn->insts[ninst].n.stack<=dyn->insts[ninst].n.combined2)
+                neoncache_promote_double_forward(dyn, ninst+1, maxinst, dyn->insts[ninst].n.combined2, st);
+        }
+    } else if(a == dyn->insts[ninst].n.combined2) {
+        int i = neoncache_get_st_f_noback(dyn, ninst, dyn->insts[ninst].n.combined1, st);
+        //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double_internal, ninst=%d combined1 %d i=%d\n", ninst, dyn->insts[ninst].n.combined1 ,i);
+        if(i>=0) {
+            dyn->insts[ninst].n.neoncache[i].t = NEON_CACHE_ST_D;
+            neoncache_promote_double_internal(dyn, ninst-1, maxinst, dyn->insts[ninst].n.combined1, st);
+            // go forward is combined is not pop'd
+            if(dyn->insts[ninst].n.stack_next-dyn->insts[ninst].n.stack<=dyn->insts[ninst].n.combined1)
+                neoncache_promote_double_forward(dyn, ninst+1, maxinst, dyn->insts[ninst].n.combined1, st);
+        }
+    }
+}
+static void neoncache_promote_double_internal(dynarec_arm_t* dyn, int ninst, int maxinst, int a, int st)
+{
+    while(ninst>=0) {
+        int i = neoncache_get_st_f(dyn, ninst, a, st);
+        //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double_internal, ninst=%d, a=%d st=%d:%d/%d, i=%d\n", ninst, a, st, dyn->insts[ninst].n.stack, dyn->insts[ninst].n.stack_next, i);
+        if(i<0) return;
+        dyn->insts[ninst].n.neoncache[i].t = NEON_CACHE_ST_D;
+        a+=dyn->insts[ninst].n.stack-st;    // adjust STx
+        st = dyn->insts[ninst].n.stack;     // new stack depth
+        // check combined propagation too
+        if(dyn->insts[ninst].n.combined1 || dyn->insts[ninst].n.combined2) {
+            //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double_internal, ninst=%d combined %d/%d vs %d with st %d/%d\n", ninst, dyn->insts[ninst].n.combined1 ,dyn->insts[ninst].n.combined2, a, st, dyn->insts[ninst].n.stack);
+            neoncache_promote_double_combined(dyn, ninst, maxinst, a, st);
+        }
+        --ninst;
+    }
+}
+
+static void neoncache_promote_double_forward(dynarec_arm_t* dyn, int ninst, int maxinst, int a, int st)
+{
+    while(ninst<maxinst) {
+        int i = neoncache_get_st_f_noback(dyn, ninst, a, st);
+        //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double_forward, ninst=%d, a=%d st=%d:%d/%d, i=%d\n", ninst, a, st, dyn->insts[ninst].n.stack, dyn->insts[ninst].n.stack_next, i);
+        if(i<0) return;
+        dyn->insts[ninst].n.neoncache[i].t = NEON_CACHE_ST_D;
+        a+=dyn->insts[ninst].n.stack-st;    // adjust STx
+        st = dyn->insts[ninst].n.stack;     // new stack depth
+        // check combined propagation too
+        if(dyn->insts[ninst].n.combined1 || dyn->insts[ninst].n.combined2) {
+            //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double_forward, ninst=%d combined %d/%d vs %d with st %d/%d\n", ninst, dyn->insts[ninst].n.combined1 ,dyn->insts[ninst].n.combined2, a, st, dyn->insts[ninst].n.stack);
+            neoncache_promote_double_combined(dyn, ninst, maxinst, a, st);
+        }
+        ++ninst;
+    }
+    neoncache_promote_double(dyn, ninst, a, st);
+}
+
+void neoncache_promote_double(dynarec_arm_t* dyn, int ninst, int a, int st)
+{
+    int i = neoncache_get_current_st_f(dyn, a, st);
+    //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double, ninst=%d a=%d st=%d/%d i=%d\n", ninst, a, st, dyn->n.stack, i);
+    if(i<0) return;
+    dyn->n.neoncache[i].t = NEON_CACHE_ST_D;
+    a+=dyn->n.stack-st;
+    st = dyn->n.stack;
+    // check combined propagation too
+    if(dyn->n.combined1 || dyn->n.combined2) {
+        //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double, ninst=%d combined! %d/%d vs %d\n", ninst, dyn->n.combined1 ,dyn->n.combined2, a);
+        if(dyn->n.combined1 == a)
+            neoncache_promote_double(dyn, ninst, dyn->n.combined2, st);
+        else if(dyn->n.combined2 == a)
+            neoncache_promote_double(dyn, ninst, dyn->n.combined1, st);
+    }
+    if(!ninst) return;
+    neoncache_promote_double_internal(dyn, ninst-1, ninst, a, st);
+}
+
+int neoncache_combine_st(dynarec_arm_t* dyn, int ninst, int a, int b)
+{
+    dyn->n.combined1=a;
+    dyn->n.combined2=b;
+    if( neoncache_get_current_st(dyn, ninst, a)==NEON_CACHE_ST_F
+     && neoncache_get_current_st(dyn, ninst, b)==NEON_CACHE_ST_F )
+        return NEON_CACHE_ST_F;
+    return NEON_CACHE_ST_D;
+}
+
+#define F8      *(uint8_t*)(addr++)
+#define F32     *(uint32_t*)(addr+=4, addr-4)
 // Get if ED will have the correct parity. Not emiting anything. Parity is 2 for DWORD or 3 for QWORD
 int getedparity(dynarec_arm_t* dyn, int ninst, uintptr_t addr, uint8_t nextop, int parity)
 {
-#define F8      *(uint8_t*)(addr++)
-#define F32     *(uint32_t*)(addr+=4, addr-4)

    uint32_t tested = (1<<parity)-1;
    if((nextop&0xC0)==0xC0)
@@ -289,7 +426,7 @@ int getedparity(dynarec_arm_t* dyn, int ninst, uintptr_t addr, uint8_t nextop, i
                    return (tmp&tested)?0:1;
                }
            } else {
-                if(sib_reg==4)
+                if(sib_reg==4 && parity<3)
                    return 0;   // simple [reg]
                // don't try [reg1 + reg2<<N], unless reg1 is ESP
                return ((sib&0x7)==4 && (sib>>6)>=parity)?1:0;
@@ -303,6 +440,59 @@ int getedparity(dynarec_arm_t* dyn, int ninst, uintptr_t addr, uint8_t nextop, i
    } else {
        return 0; //Form [reg1 + reg2<<N + XXXXXX]
    }
+}
+
+// Do the GETED, but don't emit anything...
+uintptr_t fakeed(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop) 
+{
+    if((nextop&0xC0)==0xC0)
+        return addr;
+    if(!(nextop&0xC0)) {
+        if((nextop&7)==4) {
+            uint8_t sib = F8;
+            if((sib&0x7)==5) {
+                addr+=4;
+            }
+        } else if((nextop&7)==5) {
+            addr+=4;
+        }
+    } else {
+        if((nextop&7)==4) {
+            ++addr;
+        }
+        if(nextop&0x80) {
+            addr+=4;
+        } else {
+            ++addr;
+        }
+    }
+    return addr;
+}
 #undef F8
 #undef F32
+
+int isNativeCall(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t* calladdress, int* retn)
+{
+#define PK(a)       *(uint8_t*)(addr+a)
+#define PK32(a)     *(uint32_t*)(addr+a)
+
+    if(!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
+        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) {
+        // found !
+        if(retn) *retn = (b->C3==0xC2)?b->N:0;
+        if(calladdress) *calladdress = addr+1;
+        return 1;
+    }
+    return 0;
+#undef PK32
+#undef PK
 }
+
@@ -3,35 +3,30 @@

 typedef struct x86emu_s x86emu_t;

-void arm_popf(x86emu_t* emu, uint32_t f);
 void arm_fstp(x86emu_t* emu, void* p);

 void arm_print_armreg(x86emu_t* emu, uintptr_t reg, uintptr_t n);

 void arm_f2xm1(x86emu_t* emu);
-void arm_fyl2x(x86emu_t* emu);
-void arm_ftan(x86emu_t* emu);
-void arm_fpatan(x86emu_t* emu);
 void arm_fxtract(x86emu_t* emu);
 void arm_fprem(x86emu_t* emu);
 void arm_fyl2xp1(x86emu_t* emu);
-void arm_fsincos(x86emu_t* emu);
 void arm_frndint(x86emu_t* emu);
 void arm_fscale(x86emu_t* emu);
-void arm_fsin(x86emu_t* emu);
-void arm_fcos(x86emu_t* emu);
 void arm_fbld(x86emu_t* emu, uint8_t* ed);
 void arm_fild64(x86emu_t* emu, int64_t* ed);
-void arm_fbstp(x86emu_t* emu, uint8_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_fxsave(x86emu_t* emu, uint8_t* ed);
-void arm_fxrstor(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_ud(x86emu_t* emu);

+#define FPUFIRST    8
+
 // Get an FPU single scratch reg
 int fpu_get_scratch_single(dynarec_arm_t* dyn);
 // Get an FPU double scratch reg
@@ -40,18 +35,34 @@ int fpu_get_scratch_double(dynarec_arm_t* dyn);
 int fpu_get_scratch_quad(dynarec_arm_t* dyn);
 // Reset scratch regs counter
 void fpu_reset_scratch(dynarec_arm_t* dyn);
-// Get an FPU double reg
-int fpu_get_reg_double(dynarec_arm_t* dyn);
+// Get an FPU double reg, with NEON_CACHE_TYPE
+int fpu_get_reg_double(dynarec_arm_t* dyn, unsigned int t, unsigned int n);
 // Free a FPU double reg
 void fpu_free_reg_double(dynarec_arm_t* dyn, int reg);
 // Get an FPU quad reg
-int fpu_get_reg_quad(dynarec_arm_t* dyn);
+int fpu_get_reg_quad(dynarec_arm_t* dyn, unsigned int t, unsigned int n);
 // Free a FPU quad reg
 void fpu_free_reg_quad(dynarec_arm_t* dyn, int reg);
 // Reset fpu regs counter
 void fpu_reset_reg(dynarec_arm_t* dyn);

+// ---- Neon cache functions
+// Get type for STx
+int neoncache_get_st(dynarec_arm_t* dyn, int ninst, int a);
+// Get if STx is FLOAT or DOUBLE
+int neoncache_get_st_f(dynarec_arm_t* dyn, int ninst, int a, int st);
+// Get actual type for STx
+int neoncache_get_current_st(dynarec_arm_t* dyn, int ninst, int a);
+// Back-propagate a change float->double
+void neoncache_promote_double(dynarec_arm_t* dyn, int ninst, int a, int st);
+// Combine and propagate if needed (pass 1 only)
+int neoncache_combine_st(dynarec_arm_t* dyn, int ninst, int a, int b);  // with stack current dyn->n_stack*
 // Get if ED will have the correct parity. Not emiting anything. Parity is 2 for DWORD or 3 for QWORD
 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);
+
+// 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);

 #endif //__DYNAREC_ARM_FUNCTIONS_H__
@@ -22,9 +22,10 @@
 #define F32     *(uint32_t*)(addr+=4, addr-4)
 #define F32S    *(int32_t*)(addr+=4, addr-4)
 #define PK(a)   *(uint8_t*)(addr+a)
-#define PK32(a)   *(uint32_t*)(addr+a)
-#define PK16(a)   *(uint16_t*)(addr+a)
-#define PKip(a)   *(uint8_t*)(ip+a)
+#define PK32(a) *(uint32_t*)(addr+a)
+#define PK16(a) *(uint16_t*)(addr+a)
+#define PKip(a) *(uint8_t*)(ip+a)
+#define PKa(a)  *(uint8_t*)(a)

 // GETGD    get x86 register in gd
 #define GETGD   gd = xEAX+((nextop&0x38)>>3)
@@ -33,7 +34,7 @@
                    ed = xEAX+(nextop&7);   \
                    wback = 0;              \
                } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0, 0); \
                    LDR_IMM9(x1, wback, fixedaddress); \
                    ed = x1;                \
                }
@@ -42,7 +43,7 @@
                    ed = xEAX+(nextop&7);   \
                    wback = 0;              \
                } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, (hint==x2)?x1:x2, &fixedaddress, 4095, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, (hint==x2)?x1:x2, &fixedaddress, 4095, 0, 0); \
                    LDR_IMM9(hint, wback, fixedaddress); \
                    ed = hint;              \
                }
@@ -52,7 +53,7 @@
                    MOV_REG(ret, ed);       \
                    wback = 0;              \
                } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, hint, &fixedaddress, 4095, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, hint, &fixedaddress, 4095, 0, 0); \
                    ed = ret;               \
                    LDR_IMM9(ed, wback, fixedaddress); \
                }
@@ -65,10 +66,20 @@
                    ed = xEAX+(nextop&7);   \
                    wback = 0;              \
                } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 0); \
                    LDR_REG_LSL_IMM5(x1, wback, O, 0);  \
                    ed = x1;                 \
                }
+//GETEDO can use r1 for ed, and r2 for wback. wback is 0 if ed is xEAX..xEDI, else O is added to wback
+#define GETEDO2(O) if((nextop&0xC0)==0xC0) {            \
+                    ed = xEAX+(nextop&7);               \
+                    wback = 0;                          \
+                } else {                                \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 0); \
+                    ADD_REG_LSL_IMM5(wback, wback, O, 0);\
+                    LDR_IMM9(x1, wback, 0);             \
+                    ed = x1;                            \
+                }
 #define WBACKO(O)   if(wback) {STR_REG_LSL_IMM5(ed, wback, O, 0);}
 //FAKEELike GETED, but doesn't get anything
 #define FAKEED  if((nextop&0xC0)!=0xC0) {   \
@@ -83,7 +94,7 @@
                    ed = i;                 \
                    wb1 = 0;                \
                } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, w, &fixedaddress, 255, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, w, &fixedaddress, 255, 0, 0); \
                    LDRH_IMM8(i, wback, fixedaddress); \
                    ed = i;                 \
                    wb1 = 1;                \
@@ -95,7 +106,7 @@
                    ed = i;                 \
                    wb1 = 0;                \
                } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0); \
                    LDRH_IMM8(i, wback, fixedaddress); \
                    ed = i;                 \
                    wb1 = 1;                \
@@ -107,7 +118,7 @@
                    ed = i;                 \
                    wb1 = 0;                \
                } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0); \
                    LDRSH_IMM8(i, wback, fixedaddress);\
                    ed = i;                 \
                    wb1 = 1;                \
@@ -127,11 +138,26 @@
                    wb1 = 0;                \
                    ed = i;                 \
                } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095, 0, 0); \
                    LDRB_IMM9(i, wback, fixedaddress); \
                    wb1 = 1;                \
                    ed = i;                 \
                }
+//GETEBO will use i for ed, i is also Offset, and can use r3 for wback.
+#define GETEBO(i) if((nextop&0xC0)==0xC0) {  \
+                    wback = (nextop&7);     \
+                    wb2 = (wback>>2);       \
+                    wback = xEAX+(wback&3); \
+                    UXTB(i, wback, wb2);    \
+                    wb1 = 0;                \
+                    ed = i;                 \
+                } else {                    \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0, 0, 0); \
+                    ADD_REG_LSL_IMM5(wback, wback, i, 0);   \
+                    LDRB_IMM9(i, wback, fixedaddress);      \
+                    wb1 = 1;                \
+                    ed = i;                 \
+                }
 //GETSEB sign extend EB, will use i for ed, and can use r3 for wback.
 #define GETSEB(i) if((nextop&0xC0)==0xC0) { \
                    wback = (nextop&7);     \
@@ -141,7 +167,7 @@
                    wb1 = 0;                \
                    ed = i;                 \
                } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0); \
                    LDRSB_IMM8(i, wback, fixedaddress);\
                    wb1 = 1;                \
                    ed = i;                 \
@@ -164,18 +190,28 @@
 #define GBBACK   BFI(gb1, gd, gb2*8, 8);

 // Get Direction with size Z and based of F_DF flag, on register r ready for LDR/STR fetching
+// F_DF is 1<<10, so 1 ROR 11*2 (so F_OF)
 #define GETDIR(r, A)    \
-    LDR_IMM9(r, xEmu, offsetof(x86emu_t, flags[F_DF]));     \
-    CMPS_IMM8(r, 1);                                        \
-    MOVW(r, A);                                             \
-    RSB_COND_IMM8(cEQ, r, r, 0)
+    TSTS_IMM8_ROR(xFlags, 1, 0x0b);         \
+    MOVW(r, A);                             \
+    RSB_COND_IMM8(cNE, r, r, 0)

-// CALL will use x12 for the call address. Return value can be put in ret (unless ret is -1)
+// CALL will use x14 for the call address. Return value can be put in ret (unless ret is -1)
 // R0 will not be pushed/popd if ret is -2
-#define CALL(F, ret, M) call_c(dyn, ninst, F, x12, ret, M)
+#define CALL(F, ret, M) call_c(dyn, ninst, F, x14, ret, M, 1)
 // CALL_ will use x3 for the call address. Return value can be put in ret (unless ret is -1)
 // R0 will not be pushed/popd if ret is -2
-#define CALL_(F, ret, M) call_c(dyn, ninst, F, x3, ret, M)
+#define CALL_(F, ret, M) call_c(dyn, ninst, F, x3, ret, M, 1)
+// CALL_S will use x3 for the call address. Return value can be put in ret (unless ret is -1)
+// R0 will not be pushed/popd if ret is -2. Flags are not save/restored
+#define CALL_S(F, ret, M) call_c(dyn, ninst, F, x3, ret, M, 0)
+// CALL_1D will use x3 for the call address. Return value in D0, 1 ARG in D0
+#define CALL_1D(F, M) call_d(dyn, ninst, F, NULL, 1, x3, -1, M, 0)
+// CALL_2D will use x3 for the call address. Return value in D0, 2 ARGs in D0, D1
+#define CALL_2D(F, M) call_d(dyn, ninst, F, NULL, 2, x3, -1, M, 0)
+// CALL_1DD will use x3 for the call address. Return value in D0, 1 ARG in D0, cals 2 functions in a row
+#define CALL_1DD(F, F2, M) call_d(dyn, ninst, F, F2, 1, x3, -1, 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;}
@@ -186,101 +222,97 @@
 #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 GETMARKLOCK ((dyn->insts)?dyn->insts[ninst].marklock:(dyn->arm_size+4))

-// Branch to MARK if cond (use i32)
+// Branch to MARK if cond (use j32)
 #define B_MARK(cond)    \
-    i32 = GETMARK-(dyn->arm_size+8);    \
-    Bcond(cond, i32)
-// Branch to MARK2 if cond (use i32)
+    j32 = GETMARK-(dyn->arm_size+8);    \
+    Bcond(cond, j32)
+// Branch to MARK2 if cond (use j32)
 #define B_MARK2(cond)    \
-    i32 = GETMARK2-(dyn->arm_size+8);   \
-    Bcond(cond, i32)
-// Branch to MARK3 if cond (use i32)
+    j32 = GETMARK2-(dyn->arm_size+8);   \
+    Bcond(cond, j32)
+// Branch to MARK3 if cond (use j32)
 #define B_MARK3(cond)    \
-    i32 = GETMARK3-(dyn->arm_size+8);   \
-    Bcond(cond, i32)
-// Branch to next instruction if cond (use i32)
+    j32 = GETMARK3-(dyn->arm_size+8);   \
+    Bcond(cond, j32)
+// Branch to next instruction if cond (use j32)
 #define B_NEXT(cond)     \
-    i32 = (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->arm_size+8)):0; \
-    Bcond(cond, i32)
-// Branch to MARKSEG if cond (use i32)
+    j32 = (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->arm_size+8)):0; \
+    Bcond(cond, j32)
+// Branch to MARKSEG if cond (use j32)
 #define B_MARKSEG(cond)    \
-    i32 = GETMARKSEG-(dyn->arm_size+8);   \
-    Bcond(cond, i32)
+    j32 = GETMARKSEG-(dyn->arm_size+8);   \
+    Bcond(cond, j32)
+// Branch to MARKLOCK if cond (use j32)
+#define B_MARKLOCK(cond)    \
+    j32 = GETMARKLOCK-(dyn->arm_size+8);   \
+    Bcond(cond, j32)

-#define IFX(A)  if(dyn->insts && (dyn->insts[ninst].x86.need_flags&(A)))
-#define IFXX(A) if(dyn->insts && (dyn->insts[ninst].x86.need_flags==(A)))
-#define IFX2X(A, B) if(dyn->insts && (dyn->insts[ninst].x86.need_flags==(A) || dyn->insts[ninst].x86.need_flags==(B) || dyn->insts[ninst].x86.need_flags==((A)|(B))))
+#define IFX(A)  if((dyn->insts[ninst].x86.need_flags&(A)))
+#define IFX_PENDOR0  if((dyn->insts[ninst].x86.need_flags&(X_PEND) || !dyn->insts[ninst].x86.need_flags))
+#define IFXX(A) if((dyn->insts[ninst].x86.need_flags==(A)))
+#define IFX2X(A, B) if((dyn->insts[ninst].x86.need_flags==(A) || dyn->insts[ninst].x86.need_flags==(B) || dyn->insts[ninst].x86.need_flags==((A)|(B))))
+#define IFXN(A, B)  if((dyn->insts[ninst].x86.need_flags&(A) && !(dyn->insts[ninst].x86.need_flags&(B))))

 // Generate FCOM with s1 and s2 scratch regs (the VCMP is already done)
 #define FCOM(s1, s2)    \
    VMRS_APSR();    /* 0b0100011100000000 */                                    \
-    LDRH_IMM8(s2, xEmu, offsetof(x86emu_t, sw));   /*offset is 8bits right?*/   \
-    MOVW(s1, 0b0100011100000000);                                               \
-    BIC_REG_LSL_IMM5(s2, s2, s1, 0);                                            \
-    MOVW_COND(cVS, s1, 0b0100010100000000); /* unordered */                     \
-    MOVW_COND(cEQ, s1, 0b0100000000000000); /* zero */                          \
-    MOVW_COND(cGT, s1, 0b0000000000000000); /* greater than */                  \
-    MOVW_COND(cLO, s1, 0b0000000100000000); /* less than */                     \
-    ORR_REG_LSL_IMM5(s2, s2, s1, 0);                                            \
-    STRH_IMM8(s2, xEmu, offsetof(x86emu_t, sw))
+    LDRH_IMM8(s1, xEmu, offsetof(x86emu_t, sw));   /*offset is 8bits right?*/   \
+    BIC_IMM8(s1, s1, 0b01000111, 12);                                           \
+    ORR_IMM8_COND(cVS, s1, s1, 0b01000101, 12); /* unordered */                 \
+    ORR_IMM8_COND(cEQ, s1, s1, 0b01000000, 12); /* equal */                     \
+    ORR_IMM8_COND(cMI, s1, s1, 0b00000001, 12); /* less than */                 \
+    /* greater than leave 0 */                                                  \
+    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)    \
-    VMRS_APSR();    /* 0b111 */                             \
-    XOR_REG_LSL_IMM5(s2, s2, s2, 0);                        \
-    MOVW_COND(cVS, s1, 0b111); /* unordered */              \
-    MOVW_COND(cEQ, s1, 0b100); /* zero */                   \
-    MOVW_COND(cGT, s1, 0b000); /* greater than */           \
-    MOVW_COND(cLO, s1, 0b001); /* less than */              \
-    IFX(X_CF|X_PEND) {                                      \
-        UBFX(s2, s1, 0, 1);                                 \
-        STR_IMM9(s2, xEmu, offsetof(x86emu_t, flags[F_CF]));\
-    }                                                       \
-    IFX(X_PF|X_PEND) {                                      \
-        UBFX(s2, s1, 1, 1);                                 \
-        STR_IMM9(s2, xEmu, offsetof(x86emu_t, flags[F_PF]));\
-    }                                                       \
-    IFX(X_ZF|X_PEND) {                                      \
-        UBFX(s2, s1, 2, 1);                                 \
-        STR_IMM9(s2, xEmu, offsetof(x86emu_t, flags[F_ZF]));\
-    }                                                       \
-    MOVW(s2, d_none);                                       \
-    STR_IMM9(s2, xEmu, offsetof(x86emu_t, df));             \
-    IFX(X_OF|X_PEND) {                                      \
-        STR_IMM9(s2, xEmu, offsetof(x86emu_t, flags[F_OF]));\
-    }                                                       \
-    IFX(X_AF|X_PEND) {                                      \
-        STR_IMM9(s2, xEmu, offsetof(x86emu_t, flags[F_AF]));\
-    }                                                       \
-    IFX(X_SF|X_PEND) {                                      \
-        STR_IMM9(s2, xEmu, offsetof(x86emu_t, flags[F_SF]));\
-    }                                                       \
+    IFX(X_CF|X_PF|X_ZF|X_PEND) {                                            \
+        VMRS_APSR();    /* 0b111 */                                         \
+        BIC_IMM8(xFlags, xFlags, 0b1000101, 0);                             \
+        ORR_IMM8_COND(cVS, xFlags, xFlags, 0b01000101, 0); /* unordered */  \
+        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);                           \
+    }                                                                       \


+#define SET_DFNONE(S)    if(!dyn->dfnone) {MOVW(S, d_none); STR_IMM9(S, xEmu, offsetof(x86emu_t, df)); dyn->dfnone=1;}
+#define SET_DF(S, N)     if(N) {MOVW(S, N); STR_IMM9(S, xEmu, offsetof(x86emu_t, df)); dyn->dfnone=0;} else SET_DFNONE(S)
+#define SET_NODF()          dyn->dfnone = 0
+#define SET_DFOK()          dyn->dfnone = 1

 #ifndef READFLAGS
 #define READFLAGS(A) \
-    if(dyn->state_flags!=SF_SET) {                      \
+    if(((A)!=X_PEND) && dyn->state_flags!=SF_SET && dyn->state_flags!=SF_SET_PENDING) { \
        if(dyn->state_flags!=SF_PENDING) {              \
            LDR_IMM9(x3, xEmu, offsetof(x86emu_t, df)); \
            TSTS_REG_LSL_IMM5(x3, x3, 0);               \
-            i32 = (GETMARKF)-(dyn->arm_size+8);         \
-            Bcond(cEQ, i32);                            \
+            j32 = (GETMARKF)-(dyn->arm_size+8);         \
+            Bcond(cEQ, j32);                            \
        }                                               \
        CALL_(UpdateFlags, -1, 0);                      \
        MARKF;                                          \
        dyn->state_flags = SF_SET;                      \
+        SET_DFOK();                                     \
    }
 #endif
 #ifndef SETFLAGS
 #define SETFLAGS(A, B)  \
-    if(dyn->state_flags!=SF_SET && B==SF_SUBSET && (dyn->insts[ninst].x86.need_flags&(~((A)|X_PEND)))) \
-        READFLAGS(dyn->insts[ninst].x86.need_flags&(~(A)));    \
-    dyn->state_flags = (B==SF_SUBSET)?SF_SET:B
+    if(dyn->state_flags!=SF_SET && B==SF_SUBSET && (dyn->insts[ninst].x86.need_flags&(~((A)/*|X_PEND*/)))) \
+        READFLAGS(dyn->insts[ninst].x86.need_flags&(~(A)|X_PEND));  \
+    dyn->state_flags = (B==SF_SUBSET)?SF_SET:                       \
+        ((B==SF_SET_PENDING && !(dyn->insts[ninst].x86.need_flags&X_PEND)?SF_SET:B))
 #endif
 #ifndef JUMP
-#define JUMP(A) 
+#define JUMP(A, C) 
 #endif
 #ifndef BARRIER
 #define BARRIER(A) 
@@ -288,12 +320,12 @@
 #ifndef BARRIER_NEXT
 #define BARRIER_NEXT(A)
 #endif
-#define UFLAG_OP1(A) if(dyn->insts && dyn->insts[ninst].x86.need_flags) {STR_IMM9(A, 0, offsetof(x86emu_t, op1));}
-#define UFLAG_OP2(A) if(dyn->insts && dyn->insts[ninst].x86.need_flags) {STR_IMM9(A, 0, offsetof(x86emu_t, op2));}
-#define UFLAG_OP12(A1, A2) if(dyn->insts && dyn->insts[ninst].x86.need_flags) {STR_IMM9(A1, 0, offsetof(x86emu_t, op1));STR_IMM9(A2, 0, offsetof(x86emu_t, op2));}
-#define UFLAG_RES(A) if(dyn->insts && dyn->insts[ninst].x86.need_flags) {STR_IMM9(A, 0, offsetof(x86emu_t, res));}
-#define UFLAG_DF(r, A) if(dyn->insts && dyn->insts[ninst].x86.need_flags) {MOVW(r, A); STR_IMM9(r, 0, offsetof(x86emu_t, df));}
-#define UFLAG_IF if(dyn->insts && dyn->insts[ninst].x86.need_flags)
+#define UFLAG_OP1(A) if(dyn->insts[ninst].x86.need_flags) {STR_IMM9(A, xEmu, offsetof(x86emu_t, op1));}
+#define UFLAG_OP2(A) if(dyn->insts[ninst].x86.need_flags) {STR_IMM9(A, xEmu, offsetof(x86emu_t, op2));}
+#define UFLAG_OP12(A1, A2) if(dyn->insts[ninst].x86.need_flags) {STR_IMM9(A1, xEmu, offsetof(x86emu_t, op1));STR_IMM9(A2, 0, offsetof(x86emu_t, op2));}
+#define UFLAG_RES(A) if(dyn->insts[ninst].x86.need_flags) {STR_IMM9(A, xEmu, offsetof(x86emu_t, res));}
+#define UFLAG_DF(r, A) if(dyn->insts[ninst].x86.need_flags) {SET_DF(r, A)}
+#define UFLAG_IF if(dyn->insts[ninst].x86.need_flags)
 #ifndef DEFAULT
 #define DEFAULT      *ok = -1; BARRIER(2)
 #endif
@@ -301,14 +333,16 @@
 #define NEW_BARRIER_INST
 #endif

-// Emit the LOCK mutex (x1, x2 and x3 are lost)
-#define LOCK        emit_lock(dyn, addr, ninst)
-// Emit the UNLOCK mutex (x1, x2 and x3 are lost)
-#define UNLOCK      emit_unlock(dyn, addr, ninst)
-
+#if STEP < 2
+#define PASS2IF(A, B) if(A)
+#elif STEP == 2
+#define PASS2IF(A, B) if(A) dyn->insts[ninst].pass2choice = B; if(dyn->insts[ninst].pass2choice == B)
+#else
+#define PASS2IF(A, B) if(dyn->insts[ninst].pass2choice == B)
+#endif

 void arm_epilog();
-void* arm_linker(x86emu_t* emu, void** table, uintptr_t addr);
+void* arm_next(x86emu_t* emu, uintptr_t addr);

 #ifndef STEPNAME
 #define STEPNAME3(N,M) N##M
@@ -338,18 +372,16 @@ void* arm_linker(x86emu_t* emu, void** table, uintptr_t addr);
 #define dynarecF30F     STEPNAME(dynarecF30F)

 #define geted           STEPNAME(geted_)
-#define fakeed          STEPNAME(fakeed_)
+#define geted16         STEPNAME(geted16_)
 #define jump_to_epilog  STEPNAME(jump_to_epilog_)
-#define jump_to_linker  STEPNAME(jump_to_linker_)
+#define jump_to_next    STEPNAME(jump_to_next_)
 #define ret_to_epilog   STEPNAME(ret_to_epilog_)
 #define retn_to_epilog  STEPNAME(retn_to_epilog_)
 #define iret_to_epilog  STEPNAME(iret_to_epilog_)
 #define call_c          STEPNAME(call_c_)
+#define call_d          STEPNAME(call_d_)
 #define grab_fsdata     STEPNAME(grab_fsdata_)
 #define grab_tlsdata    STEPNAME(grab_tlsdata_)
-#define isNativeCall    STEPNAME(isNativeCall_)
-#define emit_lock       STEPNAME(emit_lock)
-#define emit_unlock     STEPNAME(emit_unlock)
 #define emit_cmp8       STEPNAME(emit_cmp8)
 #define emit_cmp16      STEPNAME(emit_cmp16)
 #define emit_cmp32      STEPNAME(emit_cmp32)
@@ -410,23 +442,36 @@ void* arm_linker(x86emu_t* emu, void** table, uintptr_t addr);
 #define emit_neg32      STEPNAME(emit_neg32)
 #define emit_neg16      STEPNAME(emit_neg16)
 #define emit_neg8       STEPNAME(emit_neg8)
+#define emit_shl32      STEPNAME(emit_shl32)
 #define emit_shl32c     STEPNAME(emit_shl32c)
+#define emit_shr32      STEPNAME(emit_shr32)
 #define emit_shr32c     STEPNAME(emit_shr32c)
 #define emit_sar32c     STEPNAME(emit_sar32c)
 #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_shrd32c    STEPNAME(emit_shrd32c)
+#define emit_shld32c    STEPNAME(emit_shld32c)
+#define emit_shrd32     STEPNAME(emit_shrd32)
+
+#define emit_pf         STEPNAME(emit_pf)

 #define x87_do_push     STEPNAME(x87_do_push)
 #define x87_do_push_empty STEPNAME(x87_do_push_empty)
 #define x87_do_pop      STEPNAME(x87_do_pop)
+#define x87_get_current_cache   STEPNAME(x87_get_current_cache)
 #define x87_get_cache   STEPNAME(x87_get_cache)
+#define x87_get_neoncache STEPNAME(x87_get_neoncache)
 #define x87_get_st      STEPNAME(x87_get_st)
+#define x87_get_st_empty  STEPNAME(x87_get_st)
 #define x87_refresh     STEPNAME(x87_refresh)
 #define x87_forget      STEPNAME(x87_forget)
 #define x87_reget_st    STEPNAME(x87_reget_st)
 #define x87_stackcount  STEPNAME(x87_stackcount)
 #define x87_setround    STEPNAME(x87_setround)
 #define x87_restoreround STEPNAME(x87_restoreround)
+#define sse_setround    STEPNAME(sse_setround)
 #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)
@@ -436,12 +481,11 @@ void* arm_linker(x86emu_t* emu, void** table, uintptr_t addr);
 #define fpu_popcache    STEPNAME(fpu_popcache)
 #define fpu_reset       STEPNAME(fpu_reset)
 #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 cstack_push     STEPNAME(cstack_push)
-#define cstack_pop      STEPNAME(cstack_pop)
-

 // get the single reg that from the double "reg" (so Dx[idx])
 #define fpu_get_single_reg      STEPNAME(fpu_get_single_reg)
@@ -449,21 +493,22 @@ void* arm_linker(x86emu_t* emu, void** table, uintptr_t addr);
 #define fpu_putback_single_reg  STEPNAME(fpu_putback_single_reg)

 /* setup r2 to address pointed by */
-uintptr_t geted(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop, uint8_t* ed, uint8_t hint, int* fixedaddress, uint32_t absmax, uint32_t mask);
+uintptr_t geted(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop, uint8_t* ed, uint8_t hint, int* fixedaddress, uint32_t absmax, uint32_t mask, int getfixonly);
+
+/* setup r2 to address pointed by */
+uintptr_t geted16(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop, uint8_t* ed, uint8_t hint, int* fixedaddress, uint32_t absmax, uint32_t mask);

-// Do the GETED, but don't emit anything...
-uintptr_t fakeed(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop);

 // generic x86 helper
 void jump_to_epilog(dynarec_arm_t* dyn, uintptr_t ip, int reg, int ninst);
-void jump_to_linker(dynarec_arm_t* dyn, uintptr_t ip, int reg, int ninst);
+void jump_to_next(dynarec_arm_t* dyn, uintptr_t ip, int reg, int ninst);
 void ret_to_epilog(dynarec_arm_t* dyn, int ninst);
 void retn_to_epilog(dynarec_arm_t* dyn, int ninst, int n);
 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);
+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 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);
-int isNativeCall(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t* calladdress, int* retn);
 void emit_lock(dynarec_arm_t* dyn, uintptr_t addr, int ninst);
 void emit_unlock(dynarec_arm_t* dyn, uintptr_t addr, int ninst);
 void emit_cmp8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
@@ -489,21 +534,21 @@ void emit_xor32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
 void emit_xor32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
 void emit_and32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
 void emit_and32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
-void emit_or8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, int save_s4);
+void emit_or8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
 void emit_or8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
-void emit_xor8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, int save_s4);
+void emit_xor8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
 void emit_xor8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
-void emit_and8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, int save_s4);
+void emit_and8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
 void emit_and8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
 void emit_add16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, int save_s4);
 void emit_add16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
 void emit_sub16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, int save_s4);
 void emit_sub16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
-void emit_or16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, int save_s4);
+void emit_or16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
 void emit_or16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
-void emit_xor16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, int save_s4);
+void emit_xor16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
 void emit_xor16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
-void emit_and16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, int save_s4);
+void emit_and16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
 void emit_and16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
 void emit_inc32(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4);
 void emit_inc16(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4);
@@ -526,25 +571,40 @@ void emit_sbb16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
 void emit_neg32(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4);
 void emit_neg16(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4);
 void emit_neg8(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4);
+void emit_shl32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
 void emit_shl32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4);
+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_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_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_pf(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4);

 // x87 helper
 // cache of the local stack counter, to avoid upadte at every call
 void x87_stackcount(dynarec_arm_t* dyn, int ninst, int scratch);
 // fpu push. Return the Dd value to be used
-int x87_do_push(dynarec_arm_t* dyn, int ninst);
+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)
 void x87_do_push_empty(dynarec_arm_t* dyn, int ninst, int s1);
 // fpu pop. All previous returned Dd should be considered invalid
-void x87_do_pop(dynarec_arm_t* dyn, int ninst);
+void x87_do_pop(dynarec_arm_t* dyn, int ninst, int s1);
+// get cache index for a x87 reg, return -1 if cache doesn't exist
+int x87_get_current_cache(dynarec_arm_t* dyn, int ninst, int st, int t);
 // get cache index for a x87 reg, create the entry if needed
-int x87_get_cache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a);
+int x87_get_cache(dynarec_arm_t* dyn, int ninst, int populate, int s1, int s2, int a, int t);
+// get neoncache index for a x87 reg
+int x87_get_neoncache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a);
 // get vfpu register for a x87 reg, create the entry if needed
-int x87_get_st(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a);
+int x87_get_st(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a, int t);
+// get vfpu register for a x87 reg, create the entry if needed. Do not fetch the Stx if not already in cache
+int x87_get_st_empty(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a, int t);
 // refresh a value from the cache ->emu (nothing done if value is not cached)
 void x87_refresh(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st);
 // refresh a value from the cache ->emu and then forget the cache (nothing done if value is not cached)
@@ -555,24 +615,52 @@ void x87_reget_st(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st);
 int x87_setround(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);
+// 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);
+
+#define neoncache_st_coherency STEPNAME(neoncache_st_coherency)
+int neoncache_st_coherency(dynarec_arm_t* dyn, int ninst, int a, int b);
+
+#if STEP == 0
+#define ST_IS_F(A)          0
+#define X87_COMBINE(A, B)   NEON_CACHE_ST_D
+#define X87_ST0             NEON_CACHE_ST_D
+#define X87_ST(A)           NEON_CACHE_ST_D
+#elif STEP == 1
+#define ST_IS_F(A) (neoncache_get_current_st(dyn, ninst, A)==NEON_CACHE_ST_F)
+#define X87_COMBINE(A, B) neoncache_combine_st(dyn, ninst, A, B)
+#define X87_ST0     neoncache_get_current_st(dyn, ninst, 0)
+#define X87_ST(A)   neoncache_get_current_st(dyn, ninst, A)
+#else
+#define ST_IS_F(A) (neoncache_get_st(dyn, ninst, A)==NEON_CACHE_ST_F)
+#if STEP == 3
+#define X87_COMBINE(A, B) neoncache_st_coherency(dyn, ninst, A, B)
+#else
+#define X87_COMBINE(A, B) neoncache_get_st(dyn, ninst, A)
+#endif
+#define X87_ST0     neoncache_get_st(dyn, ninst, 0)
+#define X87_ST(A)   neoncache_get_st(dyn, ninst, A)
+#endif

 //MMX helpers
 // get neon register for a MMX reg, create the entry if needed
-int mmx_get_reg(dynarec_arm_t* dyn, int ninst, int s1, int a);
+int mmx_get_reg(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int a);
 // get neon register for a MMX reg, but don't try to synch it if it needed to be created
-int mmx_get_reg_empty(dynarec_arm_t* dyn, int ninst, int s1, int a);
+int mmx_get_reg_empty(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int a);

 //SSE/SSE2 helpers
 // get neon register for a SSE reg, create the entry if needed
-int sse_get_reg(dynarec_arm_t* dyn, int ninst, int s1, int a);
+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);

 // common coproc helpers
 // reset the cache
 void fpu_reset(dynarec_arm_t* dyn, int ninst);
-// purge the FPU cache (needs 3 scratch registers)
-void fpu_purgecache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
+// 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
@@ -603,11 +691,6 @@ uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
 uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog);
 uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog);

-// PUSH a x86/native couple of address in cstack, using s1, s2, s2+1
-void cstack_push(dynarec_arm_t* dyn, int ninst, uintptr_t x86ip, uintptr_t armip, int s1, int s2);
-// POP a x86/native couple of address from cstack, and generate the jump is x86 address is s0, use s1, s2 and s2+1 as scratch
-void cstack_pop(dynarec_arm_t* dyn, int ninst, int s0, int s1, int s2);
-
 #if STEP < 2
 #define PASS2(A)
 #else
@@ -3,6 +3,7 @@
 #include <stddef.h>
 #include <pthread.h>
 #include <errno.h>
+#include <string.h>

 #include "debug.h"
 #include "box86context.h"
@@ -19,42 +20,56 @@
 #include "arm_printer.h"
 #include "dynarec_arm_functions.h"
 #include "dynarec_arm_helper.h"
+#include "custommem.h"

 #ifndef STEP
 #error No STEP defined
 #endif

-void arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
+uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
 {
    int ok = 1;
    int ninst = 0;
    uintptr_t ip = addr;
+    uintptr_t init_addr = addr;
+    MAYUSE(init_addr);
    int need_epilog = 1;
-    dyn->tablei = 0;
    dyn->sons_size = 0;
    // Clean up (because there are multiple passes)
    dyn->state_flags = 0;
+    dyn->dfnone = 0;
    fpu_reset(dyn, ninst);
    // ok, go now
    INIT;
    while(ok) {
-        if(dyn->insts && (ninst>dyn->size)) {dynarec_log(LOG_NONE, "Warning, too many inst treated (%d / %d)\n",ninst, dyn->size);}
        ip = addr;
-        if(dyn->insts && (dyn->insts[ninst].x86.barrier==1)) {
+        if((dyn->insts[ninst].x86.barrier==1)) {
            NEW_BARRIER_INST;
        }
+        // 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;
        NEW_INST;
        fpu_reset_scratch(dyn);
 #ifdef HAVE_TRACE
        if(my_context->dec && box86_dynarec_trace) {
        if((trace_end == 0) 
-            || ((ip >= trace_start) && (ip <= trace_end)))  {
+            || ((ip >= trace_start) && (ip < trace_end)))  {
                MESSAGE(LOG_DUMP, "TRACE ----\n");
                fpu_reflectcache(dyn, ninst, x1, x2, x3);
-                MOV32(1, ip);
-                STM(0, (1<<4)|(1<<5)|(1<<6)|(1<<7)|(1<<8)|(1<<9)|(1<<10)|(1<<11));
-                STR_IMM9(1, 0, offsetof(x86emu_t, ip));
-                MOV32(2, 1);
+                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(PrintTrace, -1, 0);
                MESSAGE(LOG_DUMP, "----------\n");
            }
@@ -65,30 +80,53 @@ void arm_pass(dynarec_arm_t* dyn, uintptr_t addr)

        INST_EPILOG;

-        if(dyn->insts && dyn->insts[ninst+1].x86.barrier) {
-            fpu_purgecache(dyn, ninst, x1, x2, x3);
-            if(dyn->insts[ninst+1].x86.barrier!=2)
+        if(dyn->insts[ninst+1].x86.barrier) {
+            fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+            if(dyn->insts[ninst+1].x86.barrier!=2) {
                dyn->state_flags = 0;
-        }
-        if(!ok && !need_epilog && dyn->insts && (addr < (dyn->start+dyn->isize))) {
-            ok = 1;
-        }
-        if(!ok && !need_epilog && !dyn->insts) {   // check if need to continue
-            uintptr_t next = get_closest_next(dyn, addr);
-            if(next && ((next-addr)<15) && is_nops(dyn, addr, next-addr)) {
-                dynarec_log(LOG_DEBUG, "Extend block %p, %p -> %p (ninst=%d)\n", dyn, (void*)addr, (void*)next, ninst);
-                ok = 1;
-            } else if(next && (next-addr)<15) {
-                dynarec_log(LOG_DEBUG, "Cannot extend block %p -> %p (%02X %02X %02X %02X %02X)\n", (void*)addr, (void*)next, PK(0), PK(1), PK(2), PK(3), PK(4));
+                dyn->dfnone = 0;
            }
        }
+        if(!ok && !need_epilog && (addr < (dyn->start+dyn->isize))) {
+            ok = 1;
+        }
+        #if STEP == 0
+        if(!ok && !need_epilog && box86_dynarec_bigblock && getProtection(addr+3))
+            if(*(uint32_t*)addr!=0) {   // check if need to continue (but is next 4 bytes are 0, stop)
+                uintptr_t next = get_closest_next(dyn, addr);
+                if(next && (
+                    (((next-addr)<15) && is_nops(dyn, addr, next-addr)) 
+                    ||(((next-addr)<30) && is_instructions(dyn, addr, next-addr)) ))
+                {
+                    dynarec_log(LOG_DEBUG, "Extend block %p, %p -> %p (ninst=%d)\n", dyn, (void*)addr, (void*)next, ninst);
+                    ok = 1;
+                } else if(next && (next-addr)<30) {
+                    dynarec_log(LOG_DEBUG, "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;}
        ++ninst;
+        #if STEP == 0
+        if(ok && !isJumpTableDefault((void*)addr))
+        #else
+        if(ok && (ninst==dyn->size))
+        #endif
+        {
+            #if STEP == 3
+            dynarec_log(LOG_DEBUG, "Stopping block %p (%d / %d)\n",(void*)init_addr, ninst, dyn->size); 
+            #endif
+            BARRIER(2);
+            fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+            jump_to_next(dyn, addr, 0, ninst);
+            ok=0; need_epilog=0;
+        }
    }
    if(need_epilog) {
-        fpu_purgecache(dyn, ninst, x1, x2, 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);
+    return addr;
 }
@@ -1,22 +1,39 @@

 #define INIT    uintptr_t sav_addr=addr
-#define FINI    dyn->isize = addr-sav_addr
+#define FINI    \
+    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 SETFLAGS(A, B)  
-#define READFLAGS(A)    
+#define READFLAGS(A)    dyn->insts[ninst].x86.use_flags = A
+#define SETFLAGS(A,B)   {dyn->insts[ninst].x86.set_flags = A; dyn->insts[ninst].x86.state_flags = B;}
 #define EMIT(A)     
-#define JUMP(A)         add_next(dyn, (uintptr_t)A)
-#define NEW_INST        ++dyn->size
+#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 BARRIER(A)      dyn->insts[ninst].x86.barrier = A
+#define BARRIER_NEXT(A) if(ninst<dyn->size) dyn->insts[ninst+1].x86.barrier = A
+#define NEW_INST \
+        ++dyn->size;                            \
+        if(dyn->size+3>=dyn->cap) {             \
+                dyn->insts = (instruction_arm_t*)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;        \
+        if(ninst) {dyn->insts[ninst-1].x86.size = dyn->insts[ninst].x86.addr - dyn->insts[ninst-1].x86.addr;}
+
 #define INST_EPILOG         
 #define INST_NAME(name) 
 #define DEFAULT                         \
        --dyn->size;                    \
        *ok = -1;                       \
-        dynarec_log(LOG_INFO, "%p: Dynarec stopped because of Opcode %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X (%s)\n", \
+        if(box86_dynarec_log>=LOG_INFO) {\
+        dynarec_log(LOG_NONE, "%p: Dynarec stopped because of Opcode %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X", \
        (void*)ip, PKip(0),             \
        PKip(1), PKip(2), PKip(3),      \
        PKip(4), PKip(5), PKip(6),      \
        PKip(7), PKip(8), PKip(9),      \
        PKip(10),PKip(11),PKip(12),     \
-        PKip(13),PKip(14),              \
-        GetNativeName((void*)addr))
+        PKip(13),PKip(14));             \
+        printFunctionAddr(ip, " => ");  \
+        dynarec_log(LOG_NONE, "\n");    \
+        }
@@ -1,17 +1,12 @@
 #define INIT    
-#define FINI     \
-    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 FINI
 #define MESSAGE(A, ...)  
 #define EMIT(A)     
-#define READFLAGS(A)    dyn->insts[ninst].x86.use_flags = A
-#define SETFLAGS(A,B)   {dyn->insts[ninst].x86.set_flags = A; dyn->insts[ninst].x86.state_flags = B;}
-#define JUMP(A)         dyn->insts[ninst].x86.jmp = A
-#define BARRIER(A)      dyn->insts[ninst].x86.barrier = A
-#define BARRIER_NEXT(A) if(ninst+1<dyn->size) dyn->insts[ninst+1].x86.barrier = A
+#define READFLAGS(A)
+#define SETFLAGS(A,B)
+
+#define NEW_INST        dyn->n.combined1 = dyn->n.combined2 = 0
+
+#define INST_EPILOG dyn->insts[ninst].n = dyn->n

-#define NEW_INST \
-    dyn->insts[ninst].x86.addr = ip; \
-    if(ninst) dyn->insts[ninst-1].x86.size = dyn->insts[ninst].x86.addr - dyn->insts[ninst-1].x86.addr;
-#define INST_EPILOG 
 #define INST_NAME(name)  
@@ -2,8 +2,9 @@
 #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
+    *(uint32_t*)(dyn->block) = A;       \
+    dyn->block += 4; dyn->arm_size += 4;\
+    dyn->insts[ninst].size2 += 4

 #define MESSAGE(A, ...)  if(box86_dynarec_dump) dynarec_log(LOG_NONE, __VA_ARGS__)
 #define NEW_INST        
@@ -11,16 +12,34 @@
 #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, state=%d/%d, set=%X, use=%X, need=%X%s\n", \
+        dynarec_log(LOG_NONE, "%s%p: %d emited opcodes, inst=%d, barrier=%d state=%d/%d, set=%X, use=%X, need=%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->state_flags,                   \
            dyn->insts[ninst].x86.set_flags,    \
            dyn->insts[ninst].x86.use_flags,    \
-            dyn->insts[ninst].x86.need_flags,   \
-            (box86_dynarec_dump>1)?"\e[m":"");  \
+            dyn->insts[ninst].x86.need_flags);  \
+        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:ST%d", ii+8, dyn->insts[ninst].n.neoncache[ii].n); break;              \
+                case NEON_CACHE_ST_F: dynarec_log(LOG_NONE, " S%d:st%d", (ii+8)*2, dyn->insts[ninst].n.neoncache[ii].n); break;          \
+                case NEON_CACHE_MM: dynarec_log(LOG_NONE, " D%d:MM%d", ii+8, dyn->insts[ninst].n.neoncache[ii].n); break;                \
+                case NEON_CACHE_XMMW: dynarec_log(LOG_NONE, " Q%d:XMM%d", (ii+8)/2, dyn->insts[ninst].n.neoncache[ii].n); ++ii; break;   \
+                case NEON_CACHE_XMMR: dynarec_log(LOG_NONE, " Q%d:xmm%d", (ii+8)/2, dyn->insts[ninst].n.neoncache[ii].n); ++ii; break;   \
+                case NEON_CACHE_SCR: dynarec_log(LOG_NONE, " D%d:Scratch", ii+8); break;                                    \
+                case NEON_CACHE_NONE:           \
+                default:    break;              \
+            }                                   \
+        }                                       \
+        if(dyn->n.stack || dyn->insts[ninst].n.stack_next)                                   \
+            printf_log(LOG_NONE, " X87:%d/%d", dyn->n.stack, dyn->insts[ninst].n.stack_next);\
+        if(dyn->insts[ninst].n.combined1 || dyn->insts[ninst].n.combined2)                   \
+            printf_log(LOG_NONE, " CMB:%d/%d", dyn->insts[ninst].n.combined1, dyn->insts[ninst].n.combined2);   \
+        dynarec_log(LOG_NONE, "%s\n", (box86_dynarec_dump>1)?"\e[m":"");                \
    }

 #define NEW_BARRIER_INST                            \
@@ -4,17 +4,55 @@
 #include "dynarec_private.h"

 typedef struct x86emu_s x86emu_t;
+typedef struct dynablock_s dynablock_t;
+
+#define NEON_CACHE_NONE 0
+#define NEON_CACHE_ST_D 1
+#define NEON_CACHE_ST_F 2
+#define NEON_CACHE_MM   3
+#define NEON_CACHE_XMMW 4
+#define NEON_CACHE_XMMR 5
+#define NEON_CACHE_SCR  6
+typedef union neon_cache_s {
+    int8_t           v;
+    struct {
+        unsigned int t:4;   // reg type
+        unsigned int n:4;   // reg number
+    };
+} neon_cache_t;
+typedef struct neoncache_s {
+    neon_cache_t neoncache[24];
+    int8_t       stack;
+    int8_t       stack_next;
+    int8_t       stack_pop;
+    uint8_t      combined1;
+    uint8_t      combined2;
+} neoncache_t;

 typedef struct instruction_arm_s {
    instruction_x86_t   x86;
    uintptr_t           address;    // (start) address of the arm emited instruction
    uintptr_t           epilog;     // epilog of current instruction (can be start of next, of barrier stuff)
    int                 size;       // size of the arm emited instruction
+    int                 size2;      // size of the arm emited instrucion after pass2
    uintptr_t           mark, mark2, mark3;
    uintptr_t           markf;
    uintptr_t           markseg;
+    uintptr_t           marklock;
+    int                 pass2choice;// value for choices that are fixed on pass2 for pass3
+    uintptr_t           natcall;
+    int                 retn;
+    neoncache_t         n;
 } instruction_arm_t;

+typedef union sse_cache_s {
+    int     v;
+    struct {
+        uint8_t reg;
+        uint8_t write;
+    };
+} sse_cache_t;
+
 typedef struct dynarec_arm_s {
    instruction_arm_t   *insts;
    int32_t             size;
@@ -24,29 +62,31 @@ 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*          table;      // jump table
-    int                 tablesz;    // size of the jump table
-    int                 tablei;     // index
    int                 state_flags;// actual state for on-demand flags
    int                 x87cache[8];// cache status for the 8 x87 register behind the fpu stack
    int                 x87reg[8];  // reg used for x87cache entry
    int                 mmxcache[8];// cache status for the 8 MMX registers
-    int                 ssecache[8];// cache status for the 8 SSE(2) registers
+    sse_cache_t         ssecache[8];// cache status for the 8 SSE(2) registers
    int                 fpuused[24];// all 8..31 double reg from fpu, used by x87, sse and mmx
+    neoncache_t         n;          // cache for the 8..31 double reg from fpu, plus x87 stack delta
    int                 x87stack;   // cache stack counter
+    int                 mmxcount;   // number of mmx register used (not both mmx and x87 at the same time)
    int                 fpu_scratch;// scratch counter
+    int                 fpu_extra_qscratch; // some opcode need an extra quad scratch register
    int                 fpu_reg;    // x87/sse/mmx reg counter
-    int                 nolinker;   // disable use of (smart) linker in the block
+    int                 dfnone;     // if defered flags is already set to df_none
    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
+    dynablock_t*        dynablock;
 } 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);
 int is_nops(dynarec_arm_t *dyn, uintptr_t addr, int n);
+int is_instructions(dynarec_arm_t *dyn, uintptr_t addr, int n);

 #endif //__DYNAREC_ARM_PRIVATE_H_
@@ -15,14 +15,16 @@
 #define SF_UNKNOWN  0
 #define SF_SET      1
 #define SF_PENDING  2
-#define SF_MAYSET   3
-#define SF_SUBSET   4
+#define SF_SET_PENDING 3
+#define SF_MAYSET   4
+#define SF_SUBSET   5

 typedef struct instruction_x86_s {
    uintptr_t   addr;   //address of the instruction
    int32_t     size;   // size of the instruction
    int         barrier; // next instruction is a jump point, so no optim allowed
    uintptr_t   jmp;    // offset to jump to, even if conditionnal (0 if not), no relative offset here
+    int         jmp_cond;   // 1 of conditionnal jump
    int         jmp_insts;  // instuction to jump to (-1 if out of the block)
    uint32_t    use_flags;  // 0 or combination of X_?F
    uint32_t    set_flags;  // 0 or combination of X_?F
@@ -31,7 +33,5 @@ typedef struct instruction_x86_s {
 } instruction_x86_t;

 void printf_x86_instruction(zydis_dec_t* dec, instruction_x86_t* inst, const char* name);
-void tableupdate(void* jumpto, uintptr_t ref, void** table);
-void resettable(void** table);

 #endif //__DYNAREC_PRIVATE_H_
@@ -1,3 +1,4 @@
+ARM_ 1 1 1 1 0 0 0 1 0 0 0 0 (0) (0) (0) 1 (0) (0) (0) (0) (0) (0) @<1> (0) 0 0 0 0 (0) (0) (0) (0) SETEND \%cparam ? 'B' : 'L'\E
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 0 op 0 N Q M 0 Vm VH\(op ? "SUB" : "ADD")\<c>.<dt> <Qd>, <Qn>, <Qm>
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 0 0 0 N Q M 1 Vm VQADD<c>.<dt> <Qd>, <Qn>, <Qm>
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 0 0 1 N Q M 0 Vm VRHADD<c>.<dt> <Qd>, <Qn>, <Qm>
@@ -8,10 +9,10 @@ NEON 1 1 1 1 0 0 1 0 0 D 1 1 Vn Vd 0 0 0 1 N Q M 1 Vm VORN<c> <Qd>, <Qn>, <Qm>
 NEON 1 1 1 1 0 0 1 1 0 D @<2> Vn Vd 0 0 0 1 N Q M 1 Vm @ param=0b00=0b01=0b10 const char* op op,"EOR","BSL","BIT","BIF"@ @ V\op\<c> <Qd>, <Qn>, <Qm>
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 0 1 0 N Q M 1 Vm VQSUB<c>.<dt> <Qd>, <Qn>, <Qm>
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 0 1 1 N Q M op Vm @ op=0 const char* strict strict,"T","E"@ @ VCG\strict\<c>.<dt> <Qd>, <Qn>, <Qm>
-NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 0 0 N Q M 0 Vm VSHL<c>.<dt> <Qd>, <Qn>, <Qm>
-NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 0 0 N Q M 1 Vm VQSHL<c>.<dt> <Qd>, <Qn>, <Qm>
-NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 0 1 N Q M 0 Vm VRSHL<c>.<dt> <Qd>, <Qn>, <Qm>
-NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 0 1 N Q M 1 Vm VQRSHL<c>.<dt> <Qd>, <Qn>, <Qm>
+NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 0 0 N Q M 0 Vm VSHL<c>.<dt> <Qd>, <Qm>, <Qn>
+NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 0 0 N Q M 1 Vm VQSHL<c>.<dt> <Qd>, <Qm>, <Qn>
+NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 0 1 N Q M 0 Vm VRSHL<c>.<dt> <Qd>, <Qm>, <Qn>
+NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 0 1 N Q M 1 Vm VQRSHL<c>.<dt> <Qd>, <Qm>, <Qn>
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 1 0 N Q M op Vm @ op=0 const char* opname opname,"MAX","MIN"@ @ V\opname\<c>.<dt> <Qd>, <Qn>, <Qm>
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 1 1 N Q M 0 Vm VABD<c>.<dt> <Qd>, <Qn>, <Qm>
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 1 1 1 N Q M 1 Vm VABA<c>.<dt> <Qd>, <Qn>, <Qm>
@@ -50,12 +51,12 @@ NEON 1 1 1 1 0 0 1 i 1 D 0 0 0 imm3 Vd cmode/=10x1 0 Q op/=1 1 imm4 BIC<c>.<dt>
 NEON 1 1 1 1 0 0 1 i 1 D 0 0 0 imm3 Vd cmode/=110x 0 Q op/=1 1 imm4 VMVN<c>.<dt> <Qd>, #<imm>
 NEON 1 1 1 1 0 0 1 i 1 D 0 0 0 imm3 Vd cmode/=1110 0 Q op/=1 1 imm4 VMOV<c>.<dt> <Qd>, #<imm>
 INVALIDATE 1 1 1 1 0 0 1 <1> 1 <1> 0 0 0 <11> 0 <2> 1 <4>
-NEON 1 1 1 1 0 0 1 0 1 D 0 0 1 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.S8 <Qd>, <Qm>
-NEON 1 1 1 1 0 0 1 0 1 D 0 1 0 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.S16 <Qd>, <Qm>
-NEON 1 1 1 1 0 0 1 0 1 D 1 0 0 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.S32 <Qd>, <Qm>
-NEON 1 1 1 1 0 0 1 1 1 D 0 0 1 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.U8 <Qd>, <Qm>
-NEON 1 1 1 1 0 0 1 1 1 D 0 1 0 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.U16 <Qd>, <Qm>
-NEON 1 1 1 1 0 0 1 1 1 D 1 0 0 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.U32 <Qd>, <Qm>
+NEON 1 1 1 1 0 0 1 0 1 D 0 0 1 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.S8 <Qd>, <Dm>
+NEON 1 1 1 1 0 0 1 0 1 D 0 1 0 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.S16 <Qd>, <Dm>
+NEON 1 1 1 1 0 0 1 0 1 D 1 0 0 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.S32 <Qd>, <Dm>
+NEON 1 1 1 1 0 0 1 1 1 D 0 0 1 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.U8 <Qd>, <Dm>
+NEON 1 1 1 1 0 0 1 1 1 D 0 1 0 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.U16 <Qd>, <Dm>
+NEON 1 1 1 1 0 0 1 1 1 D 1 0 0 0 0 0 Vd 1 0 1 0 0 0 M 1 Vm VMOVL<c>.U32 <Qd>, <Dm>
 NEON 1 1 1 1 0 0 1 U 1 D imm6 Vd 0 0 0 0 L Q M 1 Vm VSHR<c>.<dt> <Qd>, <Qm>, #<imm>
 NEON 1 1 1 1 0 0 1 U 1 D imm6 Vd 0 0 0 1 L Q M 1 Vm VSRA<c>.<dt> <Qd>, <Qm>, #<imm>
 NEON 1 1 1 1 0 0 1 U 1 D imm6 Vd 0 0 1 0 L Q M 1 Vm VRSHR<c>.<dt> <Qd>, <Qm>, #<imm>
@@ -64,7 +65,7 @@ NEON 1 1 1 1 0 0 1 1 1 D imm6 Vd 0 1 0 0 L Q M 1 Vm VSRI<c>.<size> <Qd>, <Qm>, #
 NEON 1 1 1 1 0 0 1 0 1 D Imm6 Vd 0 1 0 1 L Q M 1 Vm VSHL<c>.I<size> <Qd>, <Qm>, #<imm>
 NEON 1 1 1 1 0 0 1 1 1 D Imm6 Vd 0 1 0 1 L Q M 1 Vm VSLI<c>.<size> <Qd>, <Qm>, #<imm>
 NEON 1 1 1 1 0 0 1 U 1 D Imm6 Vd 0 1 1 op L Q M 1 Vm VQSHL\((u & op) ? "U" : "")\<c>.<dt> <Qd>, <Qm>, #<imm>
-NEON 1 1 1 1 0 0 1 0 1 D imm6 Vd 1 0 0 0 0 0 M 1 Vm VSHRN<c>.I<size> <Dd>, <Qm>, #<imm>
+NEON 1 1 1 1 0 0 1 0 1 D imm6 Vd 1 0 0 0 0 0 M 1 Vm @ set ++size@ @ VSHRN<c>.I<size> <Dd>, <Qm>, #<imm>
 NEON 1 1 1 1 0 0 1 0 1 D imm6 Vd 1 0 0 0 0 1 M 1 Vm VRSHRN<c>.I<size> <Dd>, <Qm>, #<imm>
 NEON 1 1 1 1 0 0 1 U 1 D imm6 Vd 1 0 0 op 0 0 M 1 Vm VQSHR\((u & op) ? "U" : "")\N<c>.<dt> <Dd>, <Qm>, #<imm>
 NEON 1 1 1 1 0 0 1 U 1 D imm6 Vd 1 0 0 op 0 1 M 1 Vm VQRSHR\((u & op) ? "U" : "")\N<c>.<dt> <Dd>, <Qm>, #<imm>
@@ -104,7 +105,7 @@ NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 1 0 Vd 0 0 0 1 0 Q M 0 Vm VUZP<c>.\%d8 &l&l si
 NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 1 0 Vd 0 0 0 1 1 Q M 0 Vm VZIP<c>.\%d8 &l&l size\ <Qd>, <Qm>
 NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 1 0 Vd 0 0 1 0 0 0 M 0 Vm VMOVN<c>.<dt> <Dd>, <Qm>
 NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 1 0 Vd 0 0 1 0 0 1 M 0 Vm VQMOVUN<c>.S\%d16 &l&l size\ <Dd>, <Qm>
-NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 1 0 Vd 0 0 1 0 1 U M 0 Vm @ set u = 1 - u@ set ++size@ @ VQMOVN<c>.<dt> <Dd>, <Qm>
+NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 1 0 Vd 0 0 1 0 1 U M 0 Vm @ set ++size@ @ VQMOVN<c>.<dt> <Dd>, <Qm>
 NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 1 0 Vd 0 0 1 1 0 0 M 0 Vm VSHLL<c>.I\%d8 &l&l size\ <Qd>, <Dm>, #\%d8 &l&l size\
 NEON 1 1 1 1 0 0 1 1 1 D 1 1 size/=01 1 0 Vd 0 1 1 0 0 0 M 0 Vm VCVT<c>.F16.F32 <Dd>, <Qm>
 NEON 1 1 1 1 0 0 1 1 1 D 1 1 size/=01 1 0 Vd 0 1 1 1 0 0 M 0 Vm VCVT<c>.F32.F16 <Qd>, <Dm>
@@ -141,99 +142,128 @@ NEON 1 1 1 1 0 0 1 0 1 D size Vn Vd 1 0 1 1 N 1 M 0 Vm @ size=0b01=0b10 unsigned
 NEON 1 1 1 1 0 0 1 Q 1 D size Vn Vd 1 1 0 0 N 1 M 0 Vm @ size=0b01=0b10 unsigned int decodedImm !@ decodedImm = m >> %; m %@ 3@ &= 0x7@ 4@ &= 0xF@ 5@ = 0x21@ @ VQDMULH<c>.<dt> <Qd>, <Qn>, <Dm[x]>
 NEON 1 1 1 1 0 0 1 Q 1 D size Vn Vd 1 1 0 1 N 1 M 0 Vm @ size=0b01=0b10 unsigned int decodedImm !@ decodedImm = m >> %; m %@ 3@ &= 0x7@ 4@ &= 0xF@ 5@ = 0x21@ @ VQRDMULH<c>.<dt> <Qd>, <Qn>, <Dm[x]>
 INVALIDATE 1 1 1 1 0 0 1 <1> 1 <16> 0 <1> 0 <4>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 0 size @<2> Rm @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 1 size @<2> Rm @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\], <Rm>
-INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 0 0 <8> 1 1 <10>
-NEON 1 1 1 1 0 1 0 0 1 D 0 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)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]
-NEON 1 1 1 1 0 1 0 0 1 D 0 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)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]!
-NEON 1 1 1 1 0 1 0 0 1 D 0 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)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 0 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@ @ VST2<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 0 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@ @ VST2<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 0 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@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]!
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> Rm @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 0 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=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]
-NEON 1 1 1 1 0 1 0 0 1 D 0 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=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]!
-NEON 1 1 1 1 0 1 0 0 1 D 0 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=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 0 size @<2> Rm @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 1 size @<2> Rm @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\], <Rm>
-INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <8> 1 1 <10>
-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\]
+INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 0 0 <8> 1 1 <10>
+NEON 1 1 1 1 0 1 0 0 1 D 0 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)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 0 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)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 0 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)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 0 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@ @ VST2<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 0 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@ @ VST2<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 0 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@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 0 @<4> 1 1 0 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]!
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 0 @<4> 1 1 1 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 0 @<4> Rm @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
+INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <8> 1 1 <2> 1 1 <6>
+INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <8> 1 1 <2> 0 0 <1> 1 <4>
+INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <8> 1 1 1 0 <3> 1 <4>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 0 @<1> 1 1 0 1 @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 0 @<1> 1 1 1 1 @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 0 @<1> Rm @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 1 @<1> 1 1 0 1 @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 1 @<1> 1 1 1 1 @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + d]\[]}, [<Rn>\align\]
+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] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + d]\[]}, [<Rn>\align\], <Rm>
 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 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 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 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>
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]!
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> Rm @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>], <Rm>
-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=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*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 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*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=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 0 size @<1> @<1> 1 1 0 1 VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[]}, [<Rn>]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 0 size @<1> @<1> 1 1 1 1 VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[]}, [<Rn>]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 0 size @<1> @<1> Rm VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[]}, [<Rn>], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 0 @<4> 1 1 0 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 0 @<4> 1 1 1 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 0 @<4> Rm @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 1 size @<1> @<1> 1 1 0 1 @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 32 << size)@ align[0] = '\0'@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[], \vecname[0x23 + param1_1+param1_1+param1_1 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 1 size @<1> @<1> 1 1 1 1 @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 32 << size)@ align[0] = '\0'@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[], \vecname[0x23 + param1_1+param1_1+param1_1 + d]\[]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 1 size @<1> @<1> Rm @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 32 << size)@ align[0] = '\0'@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[], \vecname[0x23 + param1_1+param1_1+param1_1 + d]\[]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
 INVALIDATE 1 1 1 1 0 1 0 0 <3> 0 <20>
+ARM_ 1 1 1 1 0 1 0 0 @<1> 0 0 1 Rn @<8> @<4> @<4> 'NOP'
+ARM_ 1 1 1 1 0 1 0 0 U 1 0 1 Rn (1) (1) (1) (1) imm12 PLI [<Rn>, #+/-<imm12>]
+ARM_ 1 1 1 1 0 1 0 1 U @<1> 0 1 Rn (1) (1) (1) (1) imm12 PLD\param ? "W" : ""\ [<Rn>, #+/-<imm12>]
+ARM_ 1 1 1 1 0 1 0 1 U (1) 0 1 1 1 1 1 (1) (1) (1) (1) imm12 PLD [PC, #+/-<imm>]
+ARM_ 1 1 1 1 0 1 0 1 0 1 1 1 (1) (1) (1) (1) (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 1 (1) (1) (1) (1) CLREX
+ARM_ 1 1 1 1 0 1 0 1 0 1 1 1 (1) (1) (1) (1) (1) (1) (1) (1) (0) (0) (0) (0) 0 1 0 0 @<4> @ param=0b1111=0b1110=0b1011=0b1010=0b0111=0b0110=0b0011=0b0010 const char* option option,"SY","ST","ISH","ISHST","NSH","NSHST","OSH","OSHST","!!!"@ @ DSB \option\
+ARM_ 1 1 1 1 0 1 0 1 0 1 1 1 (1) (1) (1) (1) (1) (1) (1) (1) (0) (0) (0) (0) 0 1 0 1 @<4> @ param=0b1111=0b1110=0b1011=0b1010=0b0111=0b0110=0b0011=0b0010 const char* option option,"SY","ST","ISH","ISHST","NSH","NSHST","OSH","OSHST","!!!"@ @ DMB \option\
+ARM_ 1 1 1 1 0 1 0 1 0 1 1 1 (1) (1) (1) (1) (1) (1) (1) (1) (0) (0) (0) (0) 0 1 1 0 @<4> @ param=0b1111 const char* option option,"SY","!!!"@ @ ISB \option\
+ARM_ 1 1 1 1 0 1 1 0 @<1> 0 0 1 Rn @<8> @<3> 0 @<4> 'NOP'
+ARMs 1 1 1 1 0 1 1 0 U 1 0 1 Rn (1) (1) (1) (1) imm5 type 0 Rm PLI [<Rn>, +/-<Rm>{, <shift>}]
+ARMs 1 1 1 1 0 1 1 1 U @<1> 0 1 Rn (1) (1) (1) (1) imm5 type 0 Rm PLD\param ? "W" : ""\ [<Rn>, +/-<Rm>{, <shift>}]
+INVALIDATE 1 1 1 1 0 1 1 1 1 1 1 1 <4> <8> 1 1 1 1 <4>
 INVALIDATE 1 1 1 1 <28>
 ARMS cond 0 0 0 0 0 0 0 S Rn Rd imm5 type 0 Rm AND{S}<c> <Rd>, <Rn>, <Rm>{, <shift>}
 ARM_ cond 0 0 0 0 0 0 0 S Rd (0) (0) (0) (0) Rm 1 0 0 1 Rn MUL{S}<c> <Rd>, <Rn>, <Rm>
@@ -283,12 +313,15 @@ ARMS cond 0 0 0 1 0 0 0 1 Rn (0) (0) (0) (0) imm5 type 0 Rm TST<c> <Rn>, <Rm>{,
 ARM_ cond 0 0 0 1 U 0 W 1 Rn Rt (0) (0) (0) (0) 1 0 1 1 Rm LDRH<c> <Rt>, [<Rn>, +/-<Rm>]{!}
 ARM_ cond 0 0 0 1 U 0 W 1 Rn Rt (0) (0) (0) (0) 1 1 0 1 Rm LDRSB<c> <Rt>, [<Rn>, +/-<Rm>]{!}
 ARM_ cond 0 0 0 1 U 0 W 1 Rn Rt (0) (0) (0) (0) 1 1 1 1 Rm LDRSH<c> <Rt>, [<Rn>, +/-<Rm>]{!}
+ARM_ cond 0 0 0 1 0 0 0 0 (1) (1) (1) (1) Rd (0) (0) 0 (0) 0 0 0 0 (0) (0) (0) (0) MRS<c> <Rd>, APSR
+ARM_ cond 0 0 0 1 0 0 1 0 @<2> 0 0 (1) (1) (1) (1) (0) (0) 0 (0) 0 0 0 0 Rn @ param=0b01=0b10 const char* spec_reg spec_reg,"APSR_g","APSR_nzcvq","APSR_nzcvqg"@ @ MSR<c> \spec_reg\, <Rn>
 ARM_ cond 0 0 0 1 0 0 1 0 (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) 0 0 0 1 Rm BX<c> <Rm>
 ARM_ cond 0 0 0 1 0 0 1 0 (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) 0 0 1 0 Rm BXJ<c> <Rm>
 ARM_ cond 0 0 0 1 0 0 1 0 (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) 0 0 1 1 Rm BLX<c> <Rm>
 ARM_ cond/=1110 0 0 0 1 0 0 1 0 imm12 0 1 1 1 imm4 BKPT #<imm16>
 ARMS cond 0 0 0 1 0 0 1 1 Rn (0) (0) (0) (0) imm5 type 0 Rm TEQ<c> <Rn>, <Rm>{, <shift>}
 ARM_ cond 0 0 0 1 0 1 0 0 Rn Rt (0) (0) (0) (0) 1 0 0 1 Rm SWPB<c> <Rt>, <Rm>, [<Rn>]
+ARM_ cond 0 0 0 1 0 1 1 0 (1) (1) (1) (1) Rd (1) (1) (1) (1) 0 0 0 1 Rm CLZ<c> <Rd>, <Rm>
 ARM_ cond 0 0 0 1 U 1 W 0 Rn Rt imm4H 1 0 1 1 imm4L STRH<c> <Rt>, [<Rn>, #+/-<imm8>]{!}
 ARM_ cond 0 0 0 1 U 1 W 0 Rn Rt imm4H 1 1 0 1 imm4L LDRD<c> <Rt>, <Rt2>, [<Rn>, #+/-<imm8>]{!}
 ARM_ cond 0 0 0 1 U 1 W 0 Rn Rt imm4H 1 1 1 1 imm4L STRD<c> <Rt>, <Rt2>, [<Rn>, #+/-<imm8>]{!}
@@ -329,6 +362,17 @@ ARM_ cond 0 0 1 0 1 1 1 S Rn Rd imm12 RSC{S}<c> <Rd>, <Rn>, #<const>
 ARM_ cond 0 0 1 1 0 0 0 0 imm4 Rd imm12 MOVW<c> <Rd>, #<imm16>
 ARM_ cond 0 0 1 1 0 0 0 1 Rn (0) (0) (0) (0) imm12 TST<c> <Rn>, #<const>
 ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 0 0 0 0 0 NOP<c>
+ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 0 0 0 0 1 YIELD<c>
+ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 0 0 0 1 0 WFE<c>
+ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 0 0 0 1 1 WFI<c>
+ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 0 0 1 0 0 SEV<c>
+ARM_ cond/=1110 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 1 0 1 0 0 CSDB<c>
+ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 1 1 1 1 imm4 DBG<c> #<imm>
+INVALIDATE <4> 0 0 1 1 0 0 1 0 0 0 0 0 <8> <8>
+ARM_ cond 0 0 1 1 0 0 1 0 @<2> 0 0 (1) (1) (1) (1) imm12 @ param=0b01=0b10 const char* spec_reg spec_reg,"APSR_g","APSR_nzcvq","APSR_nzcvqg"@ @ MSR<c> \spec_reg\, #<imm>
+INVALIDATE <4> 0 0 1 1 0 0 1 0 <2> 0 1 <8> <8>
+INVALIDATE <4> 0 0 1 1 0 0 1 0 <2> 1 <1> <8> <8>
+INVALIDATE <4> 0 0 1 1 0 1 1 0 <4> <8> <8>
 ARM_ cond 0 0 1 1 0 0 1 1 Rn (0) (0) (0) (0) imm12 TEQ<c> <Rn>, #<const>
 ARM_ cond 0 0 1 1 0 1 0 0 imm4 Rd imm12 MOVT<c> <Rd>, #<imm16>
 ARM_ cond 0 0 1 1 0 1 0 1 Rn (0) (0) (0) (0) imm12 CMP<c> <Rn>, #<const>
@@ -375,6 +419,7 @@ ARMs cond 0 1 1 0 1 1 1 sat_imm Rd imm5 sh 0 1 Rn USAT<c> <Rd>, #<imm>, <Rn>{, <
 ARM_ cond 0 1 1 0 1 1 1 0 sat_imm Rd (1) (1) (1) (1) 0 0 1 1 Rn USAT16<c> <Rd>, #<imm>, <Rn>
 ARM_ cond 0 1 1 0 1 1 1 0 1 1 1 1 Rd rotate (0) (0) 0 1 1 1 Rm UXTB<c> <Rd>, <Rm>{, <rotation>}
 ARM_ cond 0 1 1 0 1 1 1 0 Rn Rd rotate (0) (0) 0 1 1 1 Rm UXTAB<c> <Rd>, <Rn>, <Rm>{, <rotation>}
+ARM_ cond 0 1 1 0 1 1 1 1 (1) (1) (1) (1) Rd (1) (1) (1) (1) 0 0 1 1 Rm RBIT<c> <Rd>, <Rm>
 ARM_ cond 0 1 1 0 1 1 1 1 1 1 1 1 Rd rotate (0) (0) 0 1 1 1 Rm UXTH<c> <Rd>, <Rm>{, <rotation>}
 ARM_ cond 0 1 1 0 1 1 1 1 (1) (1) (1) (1) Rd (1) (1) (1) (1) 1 0 1 1 Rm REVSH<c> <Rd>, <Rm>
 ARM_ cond 0 1 1 0 1 1 1 1 Rn Rd rotate (0) (0) 0 1 1 1 Rm UXTAH<c> <Rd>, <Rn>, <Rm>{, <rotation>}
@@ -386,6 +431,8 @@ ARMs cond 0 1 1 0 U 1 0 0 Rn Rt imm5 type 0 Rm STRB<c> <Rt>, [<Rn>], +/-<Rm>{, <
 ARMs cond 0 1 1 0 U 1 0 1 Rn Rt imm5 type 0 Rm LDRB<c> <Rt>, [<Rn>], +/-<Rm>{, <shift>}
 ARMs cond 0 1 1 0 U 1 1 0 Rn Rt imm5 type 0 Rm STRBT<c> <Rt>, [<Rn>], +/-<Rm>{, <shift>}
 ARMs cond 0 1 1 0 U 1 1 1 Rn Rt imm5 type 0 Rm LDRBT<c> <Rt>, [<Rn>], +/-<Rm>{, <shift>}
+ARM_ cond 0 1 1 1 0 0 0 1 Rd (1) (1) (1) (1) Rm 0 0 0 1 Rn SDIV<c> <Rd>, <Rn>, <Rm>
+ARM_ cond 0 1 1 1 0 0 1 1 Rd (1) (1) (1) (1) Rm 0 0 0 1 Rn UDIV<c> <Rd>, <Rn>, <Rm>
 ARMs cond 0 1 1 1 U 0 W 0 Rn Rt imm5 type 0 Rm STR<c> <Rt>, [<Rn>, +/-<Rm>{, <shift>}]{!}
 ARMs cond 0 1 1 1 U 0 W 1 Rn Rt imm5 type 0 Rm LDR<c> <Rt>, [<Rn>, +/-<Rm>{, <shift>}]{!}
 ARMs cond 0 1 1 1 U 1 W 0 Rn Rt imm5 type 0 Rm STRB<c> <Rt>, [<Rn>, +/-<Rm>{, <shift>}]{!}
@@ -395,6 +442,7 @@ ARM_ cond 0 1 1 1 1 0 0 0 Rd Ra Rm 0 0 0 1 Rn USADA8<c> <Rd>, <Rn>, <Rm>, <Ra>
 ARM_ cond 0 1 1 1 1 1 0 msb Rd lsb 0 0 1 1 1 1 1 BFC<c> <Rd>, #<lsb>, #<width>
 ARM_ cond 0 1 1 1 1 1 0 msb Rd lsb 0 0 1 Rn BFI<c> <Rd>, <Rn>, #<lsb>, #<width>
 ARM_ cond 0 1 1 1 1 1 1 widthm1 Rd lsb 1 0 1 Rn UBFX<c> <Rd>, <Rn>, #<lsb>, #<width>
+ARM_ cond/=1110 0 1 1 1 1 1 1 1 imm12 1 1 1 1 imm4 UDF<c> #<imm16>
 ARM_ cond 1 0 0 0 0 0 W 0 Rn/=1101 register_list STMED<c> <Rn>{!}, <registers>
 ARM_ cond 1 0 0 0 0 0 W 0 Rn register_list STMDA<c> <Rn>{!}, <registers>
 ARM_ cond 1 0 0 0 0 0 W 1 Rn/=1101 register_list LDMED<c> <Rn>{!}, <registers>
@@ -425,14 +473,14 @@ INVALIDATE <4> 1 1 0 1 1 <1> 1 <9> 1 0 1 <9>
 INVALIDATE <4> 1 1 0 1 1 <1> 1 <9> 1 0 1 <9>
 VFPU cond 1 1 0 1 0 D 1 0 1 1 0 1 Vd 1 0 1 1 imm8/01 @ set --imm8@ @ FSTMX<c> <LIst>
 VFPU cond 1 1 0 1 0 D 1 0 1 1 0 1 Vd 1 0 1 sz imm8 VPUSH<c> <list>
-VFPU cond 1 1 0 1 U D 0 0 Rn Vd 1 0 1 sz imm8 VSTR<c> <Dd>, [<Rn>{, #+/-<imm>}]
+VFPU cond 1 1 0 1 U D 0 0 Rn Vd 1 0 1 sz imm8 @ set imm8*=4@ @ VSTR<c> <Dd>, [<Rn>{, #+/-<imm>}]
 VFPU cond 1 1 0 @<2> D W 0 Rn Vd 1 0 1 1 imm8/01 @ set --imm8@ @ FSTM\(param == 0b01 ? "IA" : "DB")\X<c> <Rn>{!}, <LIst>
 VFPU cond 1 1 0 @<2> D W 0 Rn Vd 1 0 1 sz imm8 VSTM\(param == 0b01 ? "IA" : "DB")\<c> <Rn>{!}, <list>
 VFPU cond 1 1 0 0 1 D 1 1 1 1 0 1 Vd 1 0 1 1 imm8/01 @ set --imm8@ @ FLDMX<c> <LIst>
 VFPU cond 1 1 0 0 1 D 1 1 1 1 0 1 Vd 1 0 1 sz imm8 VPOP<c> <list>
 VFPU cond 1 1 0 1 0 D 0 1 1 1 1 1 Vd 1 0 1 sz 0 0 0 0 0 0 0 0 VLDR<c> <Dd>, [PC, #-0]
 VFPU cond 1 1 0 1 U D 0 1 1 1 1 1 Vd 1 0 1 sz imm8 VLDR<c> <Dd>, <label>
-VFPU cond 1 1 0 1 U D 0 1 Rn Vd 1 0 1 sz imm8 VLDR<c> <Dd>, [<Rn>{, #+/-<imm>}]
+VFPU cond 1 1 0 1 U D 0 1 Rn Vd 1 0 1 sz imm8 @ set imm8*=4@ @ VLDR<c> <Dd>, [<Rn>{, #+/-<imm>}]
 VFPU cond 1 1 0 @<2> D W 1 Rn Vd 1 0 1 1 imm8/01 @ set --imm8@ @ FLDM\(param == 0b01 ? "IA" : "DB")\X<c> <Rn>{!}, <LIst>
 VFPU cond 1 1 0 @<2> D W 1 Rn Vd 1 0 1 sz imm8 VLDM\(param == 0b01 ? "IA" : "DB")\<c> <Rn>{!}, <list>
 INVALIDATE <4> 1 1 0 <25>
@@ -444,8 +492,9 @@ VFPU cond 1 1 1 0 1 0 Q 0 Vd Rt 1 0 1 1 D 0 0 1 (0) (0) (0) (0) VDUP<c>.32 <Qd>,
 VFPU cond 1 1 1 0 1 0 Q 0 Vd Rt 1 0 1 1 D 0 1 1 (0) (0) (0) (0) VDUP<c>.16 <Qd>, <Rt>
 VFPU cond 1 1 1 0 1 1 Q 0 Vd Rt 1 0 1 1 D 0 0 1 (0) (0) (0) (0) VDUP<c>.8 <Qd>, <Rt>
 VFPU cond 1 1 1 0 0 0 0 1 Vn Rt 1 0 1 0 N (0) (0) 1 (0) (0) (0) (0) VMOV<c> <Rt>, <Sn>
+VFPU cond 1 1 1 0 1 1 1 1 0 0 0 1 Rt/=1111 1 0 1 0 (0) (0) (0) 1 (0) (0) (0) (0) VMRS<c> APSR_nzcv, FPSCR
 VFPU cond 1 1 1 0 1 1 1 1 0 0 0 1 Rt 1 0 1 0 (0) (0) (0) 1 (0) (0) (0) (0) VMRS<c> <Rt>, FPSCR
-VFPU cond 1 1 1 0 U @<2> 1 Vn Rt 1 0 1 1 N @<2> 1 (0) (0) (0) (0) @ set uint8_t opc = (u << 3) + (param1_2 << 2) + param2_2@ = uint8_t shift, size @@ (opc & 0b01000) == 0b01000@ @@ (opc & 0b01001) == 0b00001@ @@ (opc & 0b11011) == 0b00000@ shift,0,1,2,0xFF;size,0,1,2,4@ set uint8_t decodedImm = (opc & 0x7) >> shift@ @ VMOV<c>.<dt> <Rt>, <Dn[x]>
+VFPU cond 1 1 1 0 U @<2> 1 Vn Rt 1 0 1 1 N @<2> 1 (0) (0) (0) (0) @ set uint8_t opc = (u << 4) + (param1_2 << 2) + param2_2@ = uint8_t shift, size @@ (opc & 0b01000) == 0b01000@ @@ (opc & 0b01001) == 0b00001@ @@ (opc & 0b11011) == 0b00000@ shift,0,1,2,0xFF;size,0,1,2,4@ set uint8_t decodedImm = (opc & 0x7) >> shift@ @ VMOV<c>.<dt> <Rt>, <Dn[x]>
 INVALIDATE <4> 1 1 1 0 <12> 1 0 1 <4> 1 <4>
 VFPU cond 1 1 1 0 0 D 0 0 Vn Vd 1 0 1 sz N op M 0 Vm VML\(op ? "S" : "A")\<c>.F\(size ? "64" : "32")\ <Dd>, <Dn>, <Dm>
 VFPU cond 1 1 1 0 0 D 0 1 Vn Vd 1 0 1 sz N op M 0 Vm VNML\(op ? "S" : "A")\<c>.F\(size ? "64" : "32")\ <Dd>, <Dn>, <Dm>
@@ -470,6 +519,6 @@ VFPU cond 1 1 1 0 1 D 1 1 0 1 1 1 Vd 1 0 1 0 1 1 M 0 Vm VCVT<c>.F64.F32 <Dd>, <S
 VFPU cond 1 1 1 0 1 D 1 1 0 1 1 1 Vd 1 0 1 1 1 1 M 0 Vm VCVT<c>.F32.F64 <Sd>, <Dm>
 VFPU cond 1 1 1 0 1 D 1 1 1 0 0 0 Vd 1 0 1 sz op 1 M 0 Vm VCVT<c>.F\(size ? "64" : "32")\.\(op ? "S" : "U")\32 <Dd>, <Sm>
 VFPU cond 1 1 1 0 1 D 1 1 1 0 1 U Vd 1 0 1 sz op 1 i 0 imm4 @ set uint8_t td = (u << 2) + op + 1@ td=0b001=0b010=0b101 const char* Td Td,"S16","U16","S32","U32"@ set imm5 = (16 << op) - (imm5 >> 4) - ((imm5 & 0xF) << 1)@ @ VCVT<c>.F\(size ? "64" : "32")\.\Td\ <Dd>, <Dd>, #<imm>
-VFPU cond 1 1 1 0 1 D 1 1 1 1 0 U Vd 1 0 1 sz op 1 M 0 Vm VCVT\(op ? "" : "R")\<c>.\(u ? "U" : "S")\32.F\(size ? "64" : "32")\ <Sd>, <Dm>
+VFPU cond 1 1 1 0 1 D 1 1 1 1 U 1 Vd 1 0 1 sz op 1 M 0 Vm VCVT\(op ? "" : "R")\<c>.\(u ? "U" : "S")\32.F\(size ? "64" : "32")\ <Sd>, <Dm>
 VFPU cond 1 1 1 0 1 D 1 1 1 1 1 U Vd 1 0 1 sz op 1 i 0 imm4 @ set uint8_t td = (u << 2) + op + 1@ td=0b001=0b010=0b101 const char* Td Td,"S16","U16","S32","U32"@ set imm5 = (16 << op) - (imm5 >> 4) - ((imm5 & 0xF) << 1)@ @ VCVT<c>.\Td\.F\(size ? "64" : "32")\ <Dd>, <Dd>, #<imm>
 INVALIDATE <32>
@@ -40,6 +40,7 @@ const char* DumpSection(Elf32_Shdr *s, char* SST) {
        GO(SHT_HIPROC);
        GO(SHT_LOUSER);
        GO(SHT_HIUSER);
+        #if defined(SHT_GNU_versym) && defined(SHT_GNU_ATTRIBUTES)
        GO(SHT_GNU_versym);
        GO(SHT_GNU_ATTRIBUTES);
        GO(SHT_GNU_HASH);
@@ -51,6 +52,7 @@ const char* DumpSection(Elf32_Shdr *s, char* SST) {
        GO(SHT_SUNW_syminfo);
        GO(SHT_GNU_verdef);
        GO(SHT_GNU_verneed);
+        #endif
        #undef GO
        default:
            sprintf(buff, "0x%X unknown type", s->sh_type);
@@ -98,6 +100,7 @@ const char* DumpDynamic(Elf32_Dyn *s) {
            GO(DT_RUNPATH, 0);
            GO(DT_FLAGS, 0);
            GO(DT_ENCODING, 0);
+            #if defined(DT_NUM) && defined(DT_TLSDESC_PLT)
            GO(DT_NUM, 0);
            GO(DT_VALRNGLO, 0);
            GO(DT_GNU_PRELINKED, 0);
@@ -133,6 +136,7 @@ const char* DumpDynamic(Elf32_Dyn *s) {
            GO(DT_VERNEEDNUM, 0);
            GO(DT_AUXILIARY, 0);
            GO(DT_FILTER, 0);
+            #endif
        #undef GO
        default:
            sprintf(buff, "0x%X unknown type", s->d_tag);
@@ -154,11 +158,13 @@ const char* DumpPHEntry(Elf32_Phdr *e)
        GO(PT_SHLIB);
        GO(PT_PHDR);
        GO(PT_TLS);
+        #ifdef PT_NUM
        GO(PT_NUM);
        GO(PT_LOOS);
        GO(PT_GNU_EH_FRAME);
        GO(PT_GNU_STACK);
        GO(PT_GNU_RELRO);
+        #endif
        #undef GO
        default: sprintf(buff, "type: %x, Off=%x vaddr=%p paddr=%p filesz=%u memsz=%u flags=%x align=%u", e->p_type, e->p_offset, (void*)e->p_vaddr, (void*)e->p_paddr, e->p_filesz, e->p_memsz, e->p_flags, e->p_align); break;
    }
@@ -187,19 +193,21 @@ const char* DumpRelType(int t)
        GO(R_386_TLS_GD_32);
        GO(R_386_TLS_DTPMOD32);
        GO(R_386_TLS_DTPOFF32);
+        GO(R_386_TLS_TPOFF32);
        #undef GO
        default: sprintf(buff, "type: 0x%x (unknown)", t); break;
    }
    return buff;
 }

-const char* DumpSym(elfheader_t *h, Elf32_Sym* sym)
+const char* DumpSym(elfheader_t *h, Elf32_Sym* sym, int version)
 {
    static char buff[4096];
+    const char* vername = (version==-1)?"(none)":((version==0)?"*local*":((version==1)?"*global*":GetSymbolVersion(h, version)));
    memset(buff, 0, sizeof(buff));
-    sprintf(buff, "\"%s\", value=%p, size=%d, info/other=%d/%d index=%d", 
+    sprintf(buff, "\"%s\", value=%p, size=%d, info/other=%d/%d index=%d (ver=%d/%s)", 
        h->DynStr+sym->st_name, (void*)sym->st_value, sym->st_size,
-        sym->st_info, sym->st_other, sym->st_shndx);
+        sym->st_info, sym->st_other, sym->st_shndx, version, vername);
    return buff;
 }

@@ -214,116 +222,118 @@ const char* IdxSymName(elfheader_t *h, int sym)

 void DumpMainHeader(Elf32_Ehdr *header, elfheader_t *h)
 {
-    if(box86_log>=LOG_DUMP) {
-        printf_log(LOG_DUMP, "ELF Dump main header\n");
-        printf_log(LOG_DUMP, "  Entry point = %p\n", (void*)header->e_entry);
-        printf_log(LOG_DUMP, "  Program Header table offset = %p\n", (void*)header->e_phoff);
-        printf_log(LOG_DUMP, "  Section Header table offset = %p\n", (void*)header->e_shoff);
-        printf_log(LOG_DUMP, "  Flags = 0x%X\n", header->e_flags);
-        printf_log(LOG_DUMP, "  ELF Header size = %d\n", header->e_ehsize);
-        printf_log(LOG_DUMP, "  Program Header Entry num/size = %d(%d)/%d\n", h->numPHEntries, header->e_phnum, header->e_phentsize);
-        printf_log(LOG_DUMP, "  Section Header Entry num/size = %d(%d)/%d\n", h->numSHEntries, header->e_shnum, header->e_shentsize);
-        printf_log(LOG_DUMP, "  Section Header index num = %d(%d)\n", h->SHIdx, header->e_shstrndx);
-        printf_log(LOG_DUMP, "ELF Dump ==========\n");
+    if(box86_dump) {
+        printf_dump(LOG_NEVER, "ELF Dump main header\n");
+        printf_dump(LOG_NEVER, "  Entry point = %p\n", (void*)header->e_entry);
+        printf_dump(LOG_NEVER, "  Program Header table offset = %p\n", (void*)header->e_phoff);
+        printf_dump(LOG_NEVER, "  Section Header table offset = %p\n", (void*)header->e_shoff);
+        printf_dump(LOG_NEVER, "  Flags = 0x%X\n", header->e_flags);
+        printf_dump(LOG_NEVER, "  ELF Header size = %d\n", header->e_ehsize);
+        printf_dump(LOG_NEVER, "  Program Header Entry num/size = %d(%d)/%d\n", h->numPHEntries, header->e_phnum, header->e_phentsize);
+        printf_dump(LOG_NEVER, "  Section Header Entry num/size = %d(%d)/%d\n", h->numSHEntries, header->e_shnum, header->e_shentsize);
+        printf_dump(LOG_NEVER, "  Section Header index num = %d(%d)\n", h->SHIdx, header->e_shstrndx);
+        printf_dump(LOG_NEVER, "ELF Dump ==========\n");

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

-        printf_log(LOG_DUMP, "ELF Dump Sections (%d)\n", h->numSHEntries);
+        printf_dump(LOG_NEVER, "ELF Dump Sections (%d)\n", h->numSHEntries);
        for (int i=0; i<h->numSHEntries; ++i)
-            printf_log(LOG_DUMP, "  Section %04d : %s\n", i, DumpSection(h->SHEntries+i, h->SHStrTab));
-        printf_log(LOG_DUMP, "ELF Dump Sections ====\n");
+            printf_dump(LOG_NEVER, "  Section %04d : %s\n", i, DumpSection(h->SHEntries+i, h->SHStrTab));
+        printf_dump(LOG_NEVER, "ELF Dump Sections ====\n");
    }
 }

 void DumpSymTab(elfheader_t *h)
 {
-    if(box86_log>=LOG_DUMP && h->SymTab) {
+    if(box86_dump && h->SymTab) {
        const char* name = ElfName(h);
-        printf_log(LOG_DUMP, "ELF Dump SymTab(%d)=\n", h->numSymTab);
+        printf_dump(LOG_NEVER, "ELF Dump SymTab(%d)=\n", h->numSymTab);
        for (int i=0; i<h->numSymTab; ++i)
-            printf_log(LOG_DUMP, "  %s:SymTab[%d] = \"%s\", value=%p, size=%d, info/other=%d/%d index=%d\n", name, 
+            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);
-        printf_log(LOG_DUMP, "ELF Dump SymTab=====\n");
+        printf_dump(LOG_NEVER, "ELF Dump SymTab=====\n");
    }
 }

 void DumpDynamicSections(elfheader_t *h)
 {
-    if(box86_log>=LOG_DUMP && h->Dynamic) {
-        printf_log(LOG_DUMP, "ELF Dump Dynamic(%d)=\n", h->numDynamic);
+    if(box86_dump && h->Dynamic) {
+        printf_dump(LOG_NEVER, "ELF Dump Dynamic(%d)=\n", h->numDynamic);
        for (int i=0; i<h->numDynamic; ++i)
-            printf_log(LOG_DUMP, "  Dynamic %04d : %s\n", i, DumpDynamic(h->Dynamic+i));
-        printf_log(LOG_DUMP, "ELF Dump Dynamic=====\n");
+            printf_dump(LOG_NEVER, "  Dynamic %04d : %s\n", i, DumpDynamic(h->Dynamic+i));
+        printf_dump(LOG_NEVER, "ELF Dump Dynamic=====\n");
    }
 }

 void DumpDynSym(elfheader_t *h)
 {
-    if(box86_log>=LOG_DUMP && h->DynSym) {
+    if(box86_dump && h->DynSym) {
        const char* name = ElfName(h);
-        printf_log(LOG_DUMP, "ELF Dump DynSym(%d)=\n", h->numDynSym);
-        for (int i=0; i<h->numDynSym; ++i)
-            printf_log(LOG_DUMP, "  %s:DynSym[%d] = %s\n", name, i, DumpSym(h, h->DynSym+i));
-        printf_log(LOG_DUMP, "ELF Dump DynSym=====\n");
+        printf_dump(LOG_NEVER, "ELF Dump DynSym(%d)=\n", h->numDynSym);
+        for (int 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));
+        }
+        printf_dump(LOG_NEVER, "ELF Dump DynSym=====\n");
    }
 }

 void DumpDynamicNeeded(elfheader_t *h)
 {
-    if(box86_log>=LOG_DUMP && h->DynStrTab) {
-        printf_log(LOG_DUMP, "ELF Dump DT_NEEDED=====\n");
+    if(box86_dump && h->DynStrTab) {
+        printf_dump(LOG_NEVER, "ELF Dump DT_NEEDED=====\n");
        for (int i=0; i<h->numDynamic; ++i)
            if(h->Dynamic[i].d_tag==DT_NEEDED) {
-                printf_log(LOG_DUMP, "  Needed : %s\n", h->DynStrTab+h->Dynamic[i].d_un.d_val + h->delta);
+                printf_dump(LOG_NEVER, "  Needed : %s\n", h->DynStrTab+h->Dynamic[i].d_un.d_val + h->delta);
            }
-        printf_log(LOG_DUMP, "ELF Dump DT_NEEDED=====\n");
+        printf_dump(LOG_NEVER, "ELF Dump DT_NEEDED=====\n");
    }
 }

 void DumpDynamicRPath(elfheader_t *h)
 {
-    if(box86_log>=LOG_DUMP && h->DynStrTab) {
-        printf_log(LOG_DUMP, "ELF Dump DT_RPATH/DT_RUNPATH=====\n");
+    if(box86_dump && h->DynStrTab) {
+        printf_dump(LOG_NEVER, "ELF Dump DT_RPATH/DT_RUNPATH=====\n");
        for (int i=0; i<h->numDynamic; ++i) {
            if(h->Dynamic[i].d_tag==DT_RPATH) {
-                printf_log(LOG_DUMP, "   RPATH : %s\n", h->DynStrTab+h->Dynamic[i].d_un.d_val + h->delta);
+                printf_dump(LOG_NEVER, "   RPATH : %s\n", h->DynStrTab+h->Dynamic[i].d_un.d_val + h->delta);
            }
            if(h->Dynamic[i].d_tag==DT_RUNPATH) {
-                printf_log(LOG_DUMP, " RUNPATH : %s\n", h->DynStrTab+h->Dynamic[i].d_un.d_val + h->delta);
+                printf_dump(LOG_NEVER, " RUNPATH : %s\n", h->DynStrTab+h->Dynamic[i].d_un.d_val + h->delta);
            }
        }
-        printf_log(LOG_DUMP, "=====ELF Dump DT_RPATH/DT_RUNPATH\n");
+        printf_dump(LOG_NEVER, "=====ELF Dump DT_RPATH/DT_RUNPATH\n");
    }
 }

 void DumpRelTable(elfheader_t *h, int cnt, Elf32_Rel *rel, const char* name)
 {
-    if(box86_log>=LOG_DUMP) {
+    if(box86_dump) {
        const char* elfname = ElfName(h);
-        printf_log(LOG_DUMP, "ELF Dump %s Table(%d) @%p\n", name, cnt, rel);
+        printf_dump(LOG_NEVER, "ELF Dump %s Table(%d) @%p\n", name, cnt, rel);
        for (int i = 0; i<cnt; ++i)
-            printf_log(LOG_DUMP, "  %s:Rel[%d] = %p (0x%X: %s, sym=0x%0X/%s)\n", elfname,
+            printf_dump(LOG_NEVER, "  %s:Rel[%d] = %p (0x%X: %s, sym=0x%0X/%s)\n", elfname,
                i, (void*)rel[i].r_offset, rel[i].r_info, DumpRelType(ELF32_R_TYPE(rel[i].r_info)), 
                ELF32_R_SYM(rel[i].r_info), IdxSymName(h, ELF32_R_SYM(rel[i].r_info)));
-        printf_log(LOG_DUMP, "ELF Dump Rel Table=====\n");
+        printf_dump(LOG_NEVER, "ELF Dump Rel Table=====\n");
    }
 }

 void DumpRelATable(elfheader_t *h, int cnt, Elf32_Rela *rela, const char* name)
 {
-    if(box86_log>=LOG_DUMP && h->rela) {
+    if(box86_dump && h->rela) {
        const char* elfname = ElfName(h);
-        printf_log(LOG_DUMP, "ELF Dump %s Table(%d) @%p\n", name, cnt, rela);
+        printf_dump(LOG_NEVER, "ELF Dump %s Table(%d) @%p\n", name, cnt, rela);
        for (int i = 0; i<cnt; ++i)
-            printf_log(LOG_DUMP, "  %s:RelA[%d] = %p (0x%X: %s, sym=0x%X/%s) Addend=%d\n", elfname,
+            printf_dump(LOG_NEVER, "  %s:RelA[%d] = %p (0x%X: %s, sym=0x%X/%s) Addend=%d\n", elfname,
                i, (void*)rela[i].r_offset, rela[i].r_info, DumpRelType(ELF32_R_TYPE(rela[i].r_info)), 
                ELF32_R_SYM(rela[i].r_info), IdxSymName(h, ELF32_R_SYM(rela[i].r_info)), 
                rela[i].r_addend);
-        printf_log(LOG_DUMP, "ELF Dump RelA Table=====\n");
+        printf_dump(LOG_NEVER, "ELF Dump RelA Table=====\n");
    }
 }

@@ -334,24 +344,24 @@ void DumpBinary(char* p, int sz)
    unsigned char* d = (unsigned char*)p;
    int delta = ((uintptr_t)p)&0xf;
    for (int i = 0; sz; ++i) {
-        printf_log(LOG_DUMP, "%p ", (void*)(((uintptr_t)d)&~0xf));
+        printf_dump(LOG_NEVER, "%p ", (void*)(((uintptr_t)d)&~0xf));
        int n = 16 - delta;
        if (n>sz) n = sz;
        int fill = 16-sz;
        for (int j = 0; j<delta; ++j)
-            printf_log(LOG_DUMP, "   ");
+            printf_dump(LOG_NEVER, "   ");
        for (int j = 0; j<n; ++j)
-            printf_log(LOG_DUMP, "%02X ", d[j]);
+            printf_dump(LOG_NEVER, "%02X ", d[j]);
        for (int j = 0; j<fill; ++j)
-            printf_log(LOG_DUMP, "   ");
-        printf_log(LOG_DUMP, " | ");
+            printf_dump(LOG_NEVER, "   ");
+        printf_dump(LOG_NEVER, " | ");
        for (int j = 0; j<delta; ++j)
-            printf_log(LOG_DUMP, " ");
+            printf_dump(LOG_NEVER, " ");
        for (int j = 0; j<n; ++j)
-            printf_log(LOG_DUMP, "%c", (d[j]<32 || d[j]>127)?'.':d[j]);
+            printf_dump(LOG_NEVER, "%c", (d[j]<32 || d[j]>127)?'.':d[j]);
        for (int j = 0; j<fill; ++j)
-            printf_log(LOG_DUMP, " ");
-        printf_log(LOG_DUMP, "\n");
+            printf_dump(LOG_NEVER, " ");
+        printf_dump(LOG_NEVER, "\n");
        d+=n;
        sz-=n;
        delta=0;
@@ -30,6 +30,12 @@ struct elfheader_s {
    int         numDynamic;
    char*       DynStrTab;
    int         szDynStrTab;
+    Elf32_Half* VerSym;
+    Elf32_Verneed*  VerNeed;
+    int         szVerNeed;
+    Elf32_Verdef*   VerDef;
+    int         szVerDef;
+    int         e_type;

    intptr_t    delta;  // should be 0

@@ -52,8 +58,12 @@ struct elfheader_s {
    int         pltent;
    uint32_t    pltrel;
    uintptr_t   gotplt;
+    uintptr_t   gotplt_end;
    uintptr_t   pltgot;
    uintptr_t   got;
+    uintptr_t   got_end;
+    uintptr_t   plt;
+    uintptr_t   plt_end;
    uintptr_t   text;
    int         textsz;

@@ -63,7 +73,9 @@ struct elfheader_s {
    uint32_t    memsz;
    uint32_t    stacksz;
    int         stackalign;
+    uintptr_t   tlsaddr;
    uint32_t    tlssize;
+    uint32_t    tlsfilesize;
    int         tlsalign;

    int32_t     tlsbase;    // the base of the tlsdata in the global tlsdata (always negative)
@@ -76,9 +88,7 @@ struct elfheader_s {
    uintptr_t*  multiblock_offs;
    uint32_t*   multiblock_size;
    int         multiblock_n;
-#ifdef DYNAREC
-    dynablocklist_t *blocks;
-#endif
+
    library_t   *lib;
    needed_libs_t *neededlibs;
 };
@@ -61,8 +61,9 @@ void LoadNamedSection(FILE *f, Elf32_Shdr *s, int size, char* SHStrTab, const ch
 elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
 {
    Elf32_Ehdr header;
+    int level = (exec)?LOG_INFO:LOG_DEBUG;
    if(fread(&header, sizeof(Elf32_Ehdr), 1, f)!=1) {
-        printf_log(LOG_INFO, "Cannot read ELF Header\n");
+        printf_log(level, "Cannot read ELF Header\n");
        return NULL;
    }
    if(memcmp(header.e_ident, ELFMAG, SELFMAG)!=0) {
@@ -96,7 +97,7 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
    }

    if(header.e_machine != EM_386) {
-        printf_log(LOG_INFO, "Not an i386 ELF (%d)\n", header.e_machine);
+        printf_log(level, "Not an i386 ELF (%d)\n", header.e_machine);
        return NULL;
    }

@@ -119,7 +120,8 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
    h->numPHEntries = header.e_phnum;
    h->numSHEntries = header.e_shnum;
    h->SHIdx = header.e_shstrndx;
-    if(header.e_shentsize!=0) {
+    h->e_type = header.e_type;
+    if(header.e_shentsize && header.e_shnum) {
        // special cases for nums
        if(h->numSHEntries == 0) {
            printf_log(LOG_DEBUG, "Read number of Sections in 1st Section\n");
@@ -159,7 +161,7 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
            return NULL;
    }

-    if(header.e_shentsize!=0) {
+    if(header.e_shentsize && header.e_shnum) {
        if(h->SHIdx == SHN_XINDEX) {
            printf_log(LOG_DEBUG, "Read number of String Table in 1st Section\n");
            h->SHIdx = h->SHEntries[0].sh_link;
@@ -175,42 +177,99 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
            FreeElfHeader(&h);
            return NULL;
        }
-        if(box86_log>=LOG_DUMP) DumpMainHeader(&header, h);
+        if(box86_dump) DumpMainHeader(&header, h);

        LoadNamedSection(f, h->SHEntries, h->numSHEntries, h->SHStrTab, ".strtab", "SymTab Strings", SHT_STRTAB, (void**)&h->StrTab, NULL);
        LoadNamedSection(f, h->SHEntries, h->numSHEntries, h->SHStrTab, ".symtab", "SymTab", SHT_SYMTAB, (void**)&h->SymTab, &h->numSymTab);
-        if(box86_log>=LOG_DUMP && h->SymTab) DumpSymTab(h);
+        if(box86_dump && h->SymTab) DumpSymTab(h);

        LoadNamedSection(f, h->SHEntries, h->numSHEntries, h->SHStrTab, ".dynamic", "Dynamic", SHT_DYNAMIC, (void**)&h->Dynamic, &h->numDynamic);
-        if(box86_log>=LOG_DUMP && h->Dynamic) DumpDynamicSections(h);
+        if(box86_dump && h->Dynamic) DumpDynamicSections(h);
        // grab DT_REL & DT_RELA stuffs
        // also grab the DT_STRTAB string table
        {
            for (int i=0; i<h->numDynamic; ++i) {
-                if(h->Dynamic[i].d_tag == DT_REL)
-                    h->rel = h->Dynamic[i].d_un.d_ptr;
-                else if(h->Dynamic[i].d_tag == DT_RELSZ)
-                    h->relsz = h->Dynamic[i].d_un.d_val;
-                else if(h->Dynamic[i].d_tag == DT_RELENT)
-                    h->relent = h->Dynamic[i].d_un.d_val;
-                else if(h->Dynamic[i].d_tag == DT_RELA)
-                    h->rela = h->Dynamic[i].d_un.d_ptr;
-                else if(h->Dynamic[i].d_tag == DT_RELASZ)
-                    h->relasz = h->Dynamic[i].d_un.d_val;
-                else if(h->Dynamic[i].d_tag == DT_RELAENT)
-                    h->relaent = h->Dynamic[i].d_un.d_val;
-                else if(h->Dynamic[i].d_tag == DT_PLTGOT)
-                    h->pltgot = h->Dynamic[i].d_un.d_val;
-                else if(h->Dynamic[i].d_tag == DT_PLTREL)
-                    h->pltrel = h->Dynamic[i].d_un.d_val;
-                else if(h->Dynamic[i].d_tag == DT_PLTRELSZ)
-                    h->pltsz = h->Dynamic[i].d_un.d_val;
-                else if(h->Dynamic[i].d_tag == DT_JMPREL)
-                    h->jmprel = h->Dynamic[i].d_un.d_val;
-                else if(h->Dynamic[i].d_tag == DT_STRTAB)
-                    h->DynStrTab = (char*)(h->Dynamic[i].d_un.d_ptr);
-                else if(h->Dynamic[i].d_tag == DT_STRSZ)
-                    h->szDynStrTab = h->Dynamic[i].d_un.d_val;
+                Elf32_Dyn d = h->Dynamic[i];
+                Elf32_Word val = d.d_un.d_val;
+                Elf32_Addr ptr = d.d_un.d_ptr;
+                switch (d.d_tag) {
+                case DT_REL:
+                    h->rel = ptr;
+                    break;
+                case DT_RELSZ:
+                    h->relsz = val;
+                    break;
+                case DT_RELENT:
+                    h->relent = val;
+                    break;
+                case DT_RELA:
+                    h->rela = ptr;
+                    break;
+                case DT_RELASZ:
+                    h->relasz = val;
+                    break;
+                case DT_RELAENT:
+                    h->relaent = val;
+                    break;
+                case DT_PLTGOT:
+                    h->pltgot = ptr;
+                    break;
+                case DT_PLTREL:
+                    h->pltrel = val;
+                    break;
+                case DT_PLTRELSZ:
+                    h->pltsz = val;
+                    break;
+                case DT_JMPREL:
+                    h->jmprel = ptr;
+                    break;
+                case DT_STRTAB:
+                    h->DynStrTab = (char *)(ptr);
+                    break;
+                case DT_STRSZ:
+                    h->szDynStrTab = val;
+                    break;
+                case DT_INIT: // Entry point
+                    h->initentry = ptr;
+                    printf_log(LOG_DEBUG, "The DT_INIT is at address %p\n", (void*)h->initentry);
+                    break;
+                case DT_INIT_ARRAY:
+                    h->initarray = ptr;
+                    printf_log(LOG_DEBUG, "The DT_INIT_ARRAY is at address %p\n", (void*)h->initarray);
+                    break;
+                case DT_INIT_ARRAYSZ:
+                    h->initarray_sz = val / sizeof(Elf32_Addr);
+                    printf_log(LOG_DEBUG, "The DT_INIT_ARRAYSZ is %d\n", h->initarray_sz);
+                    break;
+                case DT_FINI: // Exit hook
+                    h->finientry = ptr;
+                    printf_log(LOG_DEBUG, "The DT_FINI is at address %p\n", (void*)h->finientry);
+                    break;
+                case DT_FINI_ARRAY:
+                    h->finiarray = ptr;
+                    printf_log(LOG_DEBUG, "The DT_FINI_ARRAY is at address %p\n", (void*)h->finiarray);
+                    break;
+                case DT_FINI_ARRAYSZ:
+                    h->finiarray_sz = val / sizeof(Elf32_Addr);
+                    printf_log(LOG_DEBUG, "The DT_FINI_ARRAYSZ is %d\n", h->finiarray_sz);
+                    break;
+                case DT_VERNEEDNUM:
+                    h->szVerNeed = val;
+                    printf_log(LOG_DEBUG, "The DT_VERNEEDNUM is %d\n", h->szVerNeed);
+                    break;
+                case DT_VERNEED:
+                    h->VerNeed = (Elf32_Verneed*)ptr;
+                    printf_log(LOG_DEBUG, "The DT_VERNEED is at address %p\n", h->VerNeed);
+                    break;
+                case DT_VERDEFNUM:
+                    h->szVerDef = val;
+                    printf_log(LOG_DEBUG, "The DT_VERDEFNUM is %d\n", h->szVerDef);
+                    break;
+                case DT_VERDEF:
+                    h->VerDef = (Elf32_Verdef*)ptr;
+                    printf_log(LOG_DEBUG, "The DT_VERDEF is at address %p\n", h->VerDef);
+                    break;
+                }
            }
            if(h->rel) {
                if(h->relent != sizeof(Elf32_Rel)) {
@@ -253,38 +312,26 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
        int ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".got.plt");
        if(ii) {
            h->gotplt = h->SHEntries[ii].sh_addr;
+            h->gotplt_end = h->gotplt + h->SHEntries[ii].sh_size;
            printf_log(LOG_DEBUG, "The GOT.PLT Table is at address %p\n", (void*)h->gotplt);
        }
        ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".got");
        if(ii) {
            h->got = h->SHEntries[ii].sh_addr;
-            printf_log(LOG_DEBUG, "The GOT Table is at address %p\n", (void*)h->got);
+            h->got_end = h->got + h->SHEntries[ii].sh_size;
+            printf_log(LOG_DEBUG, "The GOT Table is at address %p..%p\n", (void*)h->got, (void*)h->got_end);
        }
-        // look for .init entry point
-        ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".init");
+        ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".plt");
        if(ii) {
-            h->initentry = h->SHEntries[ii].sh_addr;
-            printf_log(LOG_DEBUG, "The .init is at address %p\n", (void*)h->initentry);
+            h->plt = h->SHEntries[ii].sh_addr;
+            h->plt_end = h->plt + h->SHEntries[ii].sh_size;
+            printf_log(LOG_DEBUG, "The PLT Table is at address %p..%p\n", (void*)h->plt, (void*)h->plt_end);
        }
-        // and .init_array
-        ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".init_array");
+        // grab version of symbols
+        ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".gnu.version");
        if(ii) {
-            h->initarray_sz = h->SHEntries[ii].sh_size / sizeof(Elf32_Addr);
-            h->initarray = (uintptr_t)(h->SHEntries[ii].sh_addr);
-            printf_log(LOG_DEBUG, "The .init_array is at address %p, and have %d elements\n", (void*)h->initarray, h->initarray_sz);
-        }
-        // look for .fini entry point
-        ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".fini");
-        if(ii) {
-            h->finientry = h->SHEntries[ii].sh_addr;
-            printf_log(LOG_DEBUG, "The .fini is at address %p\n", (void*)h->finientry);
-        }
-        // and .fini_array
-        ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".fini_array");
-        if(ii) {
-            h->finiarray_sz = h->SHEntries[ii].sh_size / sizeof(Elf32_Addr);
-            h->finiarray = (uintptr_t)(h->SHEntries[ii].sh_addr);
-            printf_log(LOG_DEBUG, "The .fini_array is at address %p, and have %d elements\n", (void*)h->finiarray, h->finiarray_sz);
+            h->VerSym = (Elf32_Half*)(h->SHEntries[ii].sh_addr);
+            printf_log(LOG_DEBUG, "The .gnu.version is at address %p\n", h->VerSym);
        }
        // grab .text for main code
        ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".text");
@@ -296,8 +343,45 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)

        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);
-        if(box86_log>=LOG_DUMP && h->DynSym) DumpDynSym(h);
    }
    
    return h;
 }
+
+const char* GetSymbolVersion(elfheader_t* h, int version)
+{
+    if(!h->VerNeed || (version<2))
+        return NULL;
+    /*if(version==1)
+        return "*";*/
+    Elf32_Verneed *ver = (Elf32_Verneed*)((uintptr_t)h->VerNeed + h->delta);
+    while(ver) {
+        Elf32_Vernaux *aux = (Elf32_Vernaux*)((uintptr_t)ver + ver->vn_aux);
+        for(int j=0; j<ver->vn_cnt; ++j) {
+            if(aux->vna_other==version)
+                return h->DynStr+aux->vna_name;
+            aux = (Elf32_Vernaux*)((uintptr_t)aux + aux->vna_next);
+        }
+        ver = ver->vn_next?((Elf32_Verneed*)((uintptr_t)ver + ver->vn_next)):NULL;
+    }
+    return GetParentSymbolVersion(h, version);  // if symbol is "internal", use Def table instead
+}
+
+const char* GetParentSymbolVersion(elfheader_t* h, int index)
+{
+    if(!h->VerDef || (index<1))
+        return NULL;
+    Elf32_Verdef *def = (Elf32_Verdef*)((uintptr_t)h->VerDef + h->delta);
+    while(def) {
+        if(def->vd_ndx==index) {
+            if(def->vd_cnt<1)
+                return NULL;
+            /*if(def->vd_flags&VER_FLG_BASE)
+                return NULL;*/
+            Elf32_Verdaux *aux = (Elf32_Verdaux*)((uintptr_t)def + def->vd_aux);
+            return h->DynStr+aux->vda_name; // return Parent, so 1st aux
+        }
+        def = def->vd_next?((Elf32_Verdef*)((uintptr_t)def + def->vd_next)):NULL;
+    }
+    return NULL;
+}
@@ -47,6 +47,52 @@
        base+=(nextop&0x80)?(F32S):(F8S); \
        A = (T*)(base+O); \
    }
+
+#define  getecommon16(A, T)                     \
+    {                                           \
+        uint32_t m = nextop&0xC7;               \
+        uint32_t base = 0;                      \
+        switch(m&7) {                           \
+            case 0: base = R_BX+R_SI; break;    \
+            case 1: base = R_BX+R_DI; break;    \
+            case 2: base = R_BP+R_SI; break;    \
+            case 3: base = R_BP+R_DI; break;    \
+            case 4: base =      R_SI; break;    \
+            case 5: base =      R_DI; break;    \
+            case 6: base =      R_BP; break;    \
+            case 7: base =      R_BX; break;    \
+        }                                       \
+        switch((m>>6)&3) {                      \
+            case 0: if(m==6) base = F16; break; \
+            case 1: base += F8S; break;         \
+            case 2: base += F16S; break;        \
+        }                                       \
+        A =  (T*)base;                          \
+    }
+
+#define  getecommon16o(A, T, O)                 \
+    {                                           \
+        uint32_t m = nextop&0xC7;               \
+        uint32_t base = 0;                      \
+        switch(m&7) {                           \
+            case 0: base = R_BX+R_SI; break;    \
+            case 1: base = R_BX+R_DI; break;    \
+            case 2: base = R_BP+R_SI; break;    \
+            case 3: base = R_BP+R_DI; break;    \
+            case 4: base =      R_SI; break;    \
+            case 5: base =      R_DI; break;    \
+            case 6: base =      R_BP; break;    \
+            case 7: base =      R_BX; break;    \
+        }                                       \
+        switch((m>>6)&3) {                      \
+            case 0: if(m==6) base = F16; break; \
+            case 1: base += F8S; break;         \
+            case 2: base += F16S; break;        \
+        }                                       \
+        A =  (T*)(base+O);                      \
+    }
+
+
 #define geteb(A) \
    if((nextop&0xC0)==0xC0) { \
        A = (reg32_t*)&emu->regs[(nextop&3)].byte[((nextop&0x4)>>2)]; \
@@ -66,11 +112,20 @@
 #define getem(A) \
    if((nextop&0xC0)==0xC0) { \
        A = &emu->mmx[(nextop&7)]; \
-    } else getecommon(A, mmx_regs_t)
+    } else getecommon(A, mmx87_regs_t)
 #define getex(A) \
    if((nextop&0xC0)==0xC0) { \
        A = &emu->xmm[(nextop&7)]; \
    } else getecommon(A, sse_regs_t)
+#define getew16(A)  \
+    if((nextop&0xC0)==0xC0) { \
+        A = &emu->regs[(nextop&7)]; \
+    } else getecommon16(A, reg32_t)
+#define getew16o(A, O)              \
+    if((nextop&0xC0)==0xC0) {       \
+        A = &emu->regs[(nextop&7)]; \
+    } else getecommon16o(A, reg32_t, O)
+
 // Macros for ModR/M gets
 #define GET_EB      geteb(oped)
 #define GET_ED      geted(oped)
@@ -78,6 +133,8 @@
 #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)
 #define EB          oped
 #define ED          oped
 #define EM          opem
@@ -89,5 +146,6 @@

 // Alias
 #define GET_EW      GET_ED
+#define GET_EW_OFFS GET_ED_OFFS
 #define EW          ED
 #define GW          GD
@@ -3,6 +3,18 @@
        _0f_0x00:                       /* VERx Ed */
            nextop = F8;
            switch((nextop>>3)&7) {
+                case 0:                 /* SLDT Ew */
+                    GET_EW;
+                    EW->word[0] = 0;
+                    if((nextop&0xC0)==0xC0)
+                        EW->word[1] = 0;
+                    break;
+                case 1:                 /* STR Ew */
+                    GET_EW;
+                    EW->word[0] = 0x7f; // dummy return
+                    if((nextop&0xC0)==0xC0)
+                        EW->word[1] = 0;
+                    break;
                case 4: //VERR
                case 5: //VERW
                    GET_EW;
@@ -15,9 +27,38 @@
                    goto _default;
            }
            NEXT;
+        _0f_0x01:                      
+            nextop = F8;
+            switch((nextop>>3)&7) {
+                case 0:                 /* SGDT Ed */
+                    GET_ED;
+                    ED->word[0] = 0x7f;    // dummy return...
+                    ED->word[1] = 0x000c;
+                    ED->word[2] = 0xd000;
+                    break;
+                case 1:                 /* SIDT Ed */
+                    GET_ED;
+                    ED->word[0] = 0xfff;    // dummy return, like "disabled"
+                    ED->word[1] = 0;
+                    ED->word[2] = 0;
+                    break;
+                case 4:                 /* SMSW Ew */
+                    GET_ED;
+                    // dummy for now... Do I need to track CR0 state?
+                    ED->word[0] = (1<<0) | (1<<4); // only PE and ET set...
+                    break;
+                default:
+                    goto _default;
+            }
+            NEXT;

        _0f_0x0B:                      /* UD2 */
-            kill(getpid(), SIGILL);              // this is undefined instruction
+            emu->old_ip = R_EIP;
+            R_EIP = ip-2;
+            emit_signal(emu, SIGILL, (void*)R_EIP, 0);
+            ip = R_EIP;
+            if(emu->quit) goto fini;
+            STEP;
            NEXT;
        _0f_0x10:                      /* MOVUPS Gx,Ex */
            nextop = F8;
@@ -84,6 +125,15 @@
            }
            NEXT;

+        _0f_0x1A:                      /* NOP (multi-byte) / ignored BNDLDX */
+            nextop = F8;
+            GET_ED;
+            NEXT;
+        _0f_0x1B:                      /* NOP (multi-byte) / ignored BNDSTX */
+            nextop = F8;
+            GET_ED;
+            NEXT;
+
        _0f_0x1F:                      /* NOP (multi-byte) */
            nextop = F8;
            GET_ED;
@@ -123,7 +173,7 @@
            // rounding should be done; and indefinite integer should also be assigned if overflow or NaN/Inf
            nextop = F8;
            GET_EX;
-            switch(emu->round) {
+            switch((emu->mxcsr>>13)&3) {
                case ROUND_Nearest:
                    GM.sd[1] = floorf(EX->f[1]+0.5f);
                    GM.sd[0] = floorf(EX->f[0]+0.5f);
@@ -169,18 +219,105 @@
        _0f_0x38:  // these are some SSE3 opcodes
            opcode = F8;
            switch(opcode) {
+                case 0x00:  /* PSHUFB */
+                    nextop = F8;
+                    GET_EM;
+                    eam1 = GM;
+                    for (int i=0; i<8; ++i) {
+                        if(EM->ub[i]&128)
+                            GM.ub[i] = 0;
+                        else
+                            GM.ub[i] = eax1.ub[EM->ub[i]&7];
+                    }
+                    break;
                case 0x04:  /* PMADDUBSW Gm,Em */
                    nextop = F8;
                    GET_EM;
                    for (int i=0; i<4; ++i) {
-                        tmp32s = (int32_t)(GM.ub[i*2+0])*EM->ub[i*2+0] + (int32_t)(GM.ub[i*2+1])*EM->ub[i*2+1];
-                        GM.sw[i] = (tmp32s>32767)?32767:tmp32s; // no negative value to test
+                        tmp32s = (int32_t)(GM.ub[i*2+0])*EM->sb[i*2+0] + (int32_t)(GM.ub[i*2+1])*EM->sb[i*2+1];
+                        GM.sw[i] = (tmp32s>32767)?32767:((tmp32s<-32768)?-32768:tmp32s);
                    }
                    break;
+                case 0x09:  /* PSIGNW Gm, Em */
+                    nextop = F8;
+                    GET_EM;
+                    for (int i=0; i<4; ++i) {
+                        if (EM->sw[i] < 0) {
+                            GM.sw[i] = -GM.sw[i];
+                        } else if (EM->sw[i] == 0) {
+                            GM.sw[i] = 0;
+                        }
+                    }
+                    break;
+                case 0x0B:  /* PMULHRSW Gm, Em */
+                    nextop = F8;
+                    GET_EM;
+                    for (int i=0; i<4; ++i) {
+                        tmp32s = ((((int32_t)(GM.sw[i])*(int32_t)(EM->sw[i]))>>14) + 1)>>1;
+                        GM.uw[i] = tmp32s&0xffff;
+                    }
+                    break;
+
+                case 0x1C:  /* PABSB Gm, Em */
+                    nextop = F8;
+                    GET_EM;
+                    for (int i=0; i<8; ++i) {
+                        GM.sb[i] = abs(EM->sb[i]);
+                    }
+                    break;
+                case 0x1D:  /* PABSW Gm, Em */
+                    nextop = F8;
+                    GET_EM;
+                    for (int i=0; i<4; ++i) {
+                        GM.sw[i] = abs(EM->sw[i]);
+                    }
+                    break;
+                case 0x1E:  /* PABSD Gm, Em */
+                    nextop = F8;
+                    GET_EM;
+                    for (int i=0; i<2; ++i) {
+                        GM.sd[i] = abs(EM->sd[i]);
+                    }
+                    break;
+
                default:
                    goto _default;
            }
            NEXT;
+
+        _0f_0x3A:
+            opcode = F8;
+            switch(opcode) {
+                case 0xF:  /* palignr */
+                    nextop = F8;
+                    GET_EM;
+                    tmp8u = F8;
+                    if (tmp8u >= 16) {
+                        GM.q = 0;
+                    } else if (tmp8u > 8) {
+                        tmp8u -= 8;
+                        GM.q >>= tmp8u*8;
+                    } else if (tmp8u == 8 || tmp8u == 0) {
+
+                    } else {
+                        GM.q <<= (8-tmp8u)*8;
+                        GM.q |= (EM->q >> tmp8u*8);
+                    }
+                    break;
+
+                default:
+                    goto _default;
+            }
+            NEXT;
+
+        _0f_0x3F:
+            emu->old_ip = R_EIP;
+            R_EIP = ip-2;
+            emit_signal(emu, SIGILL, (void*)R_EIP, 0);
+            ip = R_EIP;
+            if(emu->quit) goto fini;
+            STEP;
+            NEXT;
            
        #define GOCOND(BASE, PREFIX, CONDITIONAL) \
        _0f_##BASE##_0:                          \
@@ -415,10 +552,14 @@
            GM.sb[1] = (GM.sw[1] > 127) ? 127 : ((GM.sw[1] < -128) ? -128 : GM.sw[1]);
            GM.sb[2] = (GM.sw[2] > 127) ? 127 : ((GM.sw[2] < -128) ? -128 : GM.sw[2]);
            GM.sb[3] = (GM.sw[3] > 127) ? 127 : ((GM.sw[3] < -128) ? -128 : GM.sw[3]);
-            GM.sb[4] = (EM->sw[0] > 127) ? 127 : ((EM->sw[0] < -128) ? -128 : EM->sw[0]);
-            GM.sb[5] = (EM->sw[1] > 127) ? 127 : ((EM->sw[1] < -128) ? -128 : EM->sw[1]);
-            GM.sb[6] = (EM->sw[2] > 127) ? 127 : ((EM->sw[2] < -128) ? -128 : EM->sw[2]);
-            GM.sb[7] = (EM->sw[3] > 127) ? 127 : ((EM->sw[3] < -128) ? -128 : EM->sw[3]);
+            if(EM==&GM)
+                GM.ud[1] = GM.ud[0];
+            else {
+                GM.sb[4] = (EM->sw[0] > 127) ? 127 : ((EM->sw[0] < -128) ? -128 : EM->sw[0]);
+                GM.sb[5] = (EM->sw[1] > 127) ? 127 : ((EM->sw[1] < -128) ? -128 : EM->sw[1]);
+                GM.sb[6] = (EM->sw[2] > 127) ? 127 : ((EM->sw[2] < -128) ? -128 : EM->sw[2]);
+                GM.sb[7] = (EM->sw[3] > 127) ? 127 : ((EM->sw[3] < -128) ? -128 : EM->sw[3]);
+            }
            NEXT;
        _0f_0x64:                       /* PCMPGTB Gm,Em */
            nextop = F8;
@@ -446,33 +587,42 @@
            GET_EM;
            for(int i=0; i<4; ++i)
                GM.ub[i] = (GM.sw[i]<0)?0:((GM.sw[i]>0xff)?0xff:GM.sw[i]);
-            for(int i=0; i<4; ++i)
-                GM.ub[4+i] = (EM->sw[i]<0)?0:((EM->sw[i]>0xff)?0xff:EM->sw[i]);
+            if(EM==&GM)
+                GM.ud[1] = GM.ud[0];
+            else
+                for(int i=0; i<4; ++i)
+                    GM.ub[4+i] = (EM->sw[i]<0)?0:((EM->sw[i]>0xff)?0xff:EM->sw[i]);
            NEXT;
        _0f_0x68:                       /* PUNPCKHBW Gm,Em */
            nextop = F8;
            GET_EM;
-            if(EM==&GM) {eam1 = GM; EM = &eam1;}   // copy is needed
            for(int i=0; i<4; ++i)
                GM.ub[2 * i] = GM.ub[i + 4];
-            for(int i=0; i<4; ++i)
-                GM.ub[2 * i + 1] = EM->ub[i + 4];
+            if(EM==&GM)
+                for(int i=0; i<4; ++i)
+                    GM.ub[2 * i + 1] = GM.ub[2 * i];
+            else
+                for(int i=0; i<4; ++i)
+                    GM.ub[2 * i + 1] = EM->ub[i + 4];
            NEXT;
        _0f_0x69:                       /* PUNPCKHWD Gm,Em */
            nextop = F8;
            GET_EM;
-            if(EM==&GM) {eam1 = GM; EM = &eam1;}   // copy is needed
            for(int i=0; i<2; ++i)
                GM.uw[2 * i] = GM.uw[i + 2];
-            for(int i=0; i<2; ++i)
-                GM.uw[2 * i + 1] = EM->uw[i + 2];
+            if(EM==&GM)
+                for(int i=0; i<2; ++i)
+                    GM.uw[2 * i + 1] = GM.uw[2 * i];
+            else
+                for(int i=0; i<2; ++i)
+                    GM.uw[2 * i + 1] = EM->uw[i + 2];
            NEXT;
        _0f_0x6A:                       /* PUNPCKHDQ Gm,Em */
            nextop = F8;
            GET_EM;
-            if(EM==&GM) {eam1 = GM; EM = &eam1;}   // copy is needed
            GM.ud[0] = GM.ud[1];
-            GM.ud[1] = EM->ud[1];
+            if(EM!=&GM)
+                GM.ud[1] = EM->ud[1];
            NEXT;
        _0f_0x6B:                       /* PACKSSDW Gm,Em */
            nextop = F8;
@@ -480,8 +630,11 @@
            if(EM==&GM) {eam1 = GM; EM = &eam1;}   // copy is needed
            for(int i=0; i<2; ++i)
                GM.sw[i] = (GM.sd[i]<-32768)?-32768:((GM.sd[i]>32767)?32767:GM.sd[i]);
-            for(int i=0; i<2; ++i)
-                GM.sw[2+i] = (EM->sd[i]<-32768)?-32768:((EM->sd[i]>32767)?32767:EM->sd[i]);
+            if(EM==&GM)
+                GM.ud[1] = GM.ud[0];
+            else
+                for(int i=0; i<2; ++i)
+                    GM.sw[2+i] = (EM->sd[i]<-32768)?-32768:((EM->sd[i]>32767)?32767:EM->sd[i]);
            NEXT;

        _0f_0x6E:                      /* MOVD Gm, Ed */
@@ -619,7 +772,7 @@
            NEXT;
        _0f_0xA1:                      /* POP FS */
            emu->segs[_FS] = Pop(emu);    // no check, no use....
-            emu->segs_clean[_FS] = 0;
+            emu->segs_serial[_FS] = 0;
            NEXT;
        _0f_0xA2:                      /* CPUID */
            tmp32u = R_EAX;
@@ -629,12 +782,13 @@
            CHECK_FLAGS(emu);
            nextop = F8;
            GET_ED;
-            tmp8u = GD.byte[0];
+            tmp32s = GD.sdword[0];
+            tmp8u=tmp32s&31;
+            tmp32s >>= 5;
            if((nextop&0xC0)!=0xC0)
            {
-                ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
+                ED=(reg32_t*)(((uint32_t*)(ED))+tmp32s);
            }
-            tmp8u&=31;
            if(ED->dword[0] & (1<<tmp8u))
                SET_FLAG(F_CF);
            else
@@ -653,7 +807,7 @@
            NEXT;
        _0f_0xA9:                      /* POP GS */
            emu->segs[_GS] = Pop(emu);
-            emu->segs_clean[_GS] = 0;
+            emu->segs_serial[_GS] = 0;
            NEXT;


@@ -661,12 +815,13 @@
            CHECK_FLAGS(emu);
            nextop = F8;
            GET_ED;
-            tmp8u = GD.byte[0];
+            tmp32s = GD.sdword[0];
+            tmp8u=tmp32s&31;
+            tmp32s >>= 5;
            if((nextop&0xC0)!=0xC0)
            {
-                ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
+                ED=(reg32_t*)(((uint32_t*)(ED))+tmp32s);
            }
-            tmp8u&=31;
            if(ED->dword[0] & (1<<tmp8u))
                SET_FLAG(F_CF);
            else {
@@ -696,25 +851,10 @@
            GET_ED;
            switch((nextop>>3)&7) {
                case 0:                 /* FXSAVE m512byte */
-                    // should save flags & all
-                    // copy MMX regs...
-                    for(int i=0; i<8; ++i)
-                        memcpy(ED->dword+32+i*16, &emu->mmx[0], sizeof(emu->mmx[0]));
-                    // copy SSE regs
-                    memcpy(ED->dword+160, &emu->xmm[0], sizeof(emu->xmm));
-                    // put also FPU regs in a reserved area...
-                    for(int i=0; i<8; ++i)
-                        memcpy(ED->dword+416+i*8, &emu->fpu[0], sizeof(emu->fpu[0]));
+                    fpu_fxsave(emu, ED);
                    break;
                case 1:                 /* FXRSTOR m512byte */
-                    // should restore flags & all
-                    // copy back MMX regs...
-                    for(int i=0; i<8; ++i)
-                        memcpy(&emu->mmx[i], ED->dword+32+i*16, sizeof(emu->mmx[0]));
-                    // copy SSE regs
-                    memcpy(&emu->xmm[0], ED->dword+160/4, sizeof(emu->xmm));
-                    for(int i=0; i<8; ++i)
-                        memcpy(&emu->fpu[0], ED->dword+416+i*8, sizeof(emu->fpu[0]));
+                    fpu_fxrstor(emu, ED);
                    break;
                case 2:                 /* LDMXCSR Md */
                    emu->mxcsr = ED->dword[0];
@@ -732,7 +872,6 @@
            GD.dword[0] = imul32(emu, GD.dword[0], ED->dword[0]);
            NEXT;
        _0f_0xB0:                      /* CMPXCHG Eb,Gb */
-            CHECK_FLAGS(emu);
            nextop = F8;
            GET_EB;
            cmp8(emu, R_AL, EB->byte[0]);
@@ -757,12 +896,13 @@
            CHECK_FLAGS(emu);
            nextop = F8;
            GET_ED;
-            tmp8u = GD.byte[0];
+            tmp32s = GD.sdword[0];
+            tmp8u=tmp32s&31;
+            tmp32s >>= 5;
            if((nextop&0xC0)!=0xC0)
            {
-                ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
+                ED=(reg32_t*)(((uint32_t*)(ED))+tmp32s);
            }
-            tmp8u&=31;
            if(ED->dword[0] & (1<<tmp8u)) {
                SET_FLAG(F_CF);
                ED->dword[0] ^= (1<<tmp8u);
@@ -788,10 +928,6 @@
                    CHECK_FLAGS(emu);
                    GET_ED;
                    tmp8u = F8;
-                    if((nextop&0xC0)!=0xC0)
-                    {
-                        ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
-                    }
                    tmp8u&=31;
                    if(ED->dword[0] & (1<<tmp8u))
                        SET_FLAG(F_CF);
@@ -802,25 +938,18 @@
                    CHECK_FLAGS(emu);
                    GET_ED;
                    tmp8u = F8;
-                    if((nextop&0xC0)!=0xC0)
-                    {
-                        ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
-                    }
                    tmp8u&=31;
                    if(ED->dword[0] & (1<<tmp8u)) {
                        SET_FLAG(F_CF);
-                    } else
+                    } else {
                        ED->dword[0] ^= (1<<tmp8u);
                        CLEAR_FLAG(F_CF);
+                    }
                    break;
                case 6:             /* BTR Ed, Ib */
                    CHECK_FLAGS(emu);
                    GET_ED;
                    tmp8u = F8;
-                    if((nextop&0xC0)!=0xC0)
-                    {
-                        ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
-                    }
                    tmp8u&=31;
                    if(ED->dword[0] & (1<<tmp8u)) {
                        SET_FLAG(F_CF);
@@ -832,10 +961,6 @@
                    CHECK_FLAGS(emu);
                    GET_ED;
                    tmp8u = F8;
-                    if((nextop&0xC0)!=0xC0)
-                    {
-                        ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
-                    }
                    tmp8u&=31;
                    if(ED->dword[0] & (1<<tmp8u))
                        SET_FLAG(F_CF);
@@ -852,12 +977,13 @@
            CHECK_FLAGS(emu);
            nextop = F8;
            GET_ED;
-            tmp8u = GD.byte[0];
+            tmp32s = GD.sdword[0];
+            tmp8u=tmp32s&31;
+            tmp32s >>= 5;
            if((nextop&0xC0)!=0xC0)
            {
-                ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
+                ED=(reg32_t*)(((uint32_t*)(ED))+tmp32s);
            }
-            tmp8u&=31;
            if(ED->dword[0] & (1<<tmp8u))
                SET_FLAG(F_CF);
            else
@@ -936,7 +1062,11 @@
                GX.ud[i]=(tmp8s)?0xffffffff:0;
            }
            NEXT;
-
+        _0f_0xC3:                       /* MOVNTI Ed,Gd */
+            nextop = F8;
+            GET_ED;
+            ED->dword[0] = GD.dword[0];
+            NEXT;
        _0f_0xC4:                       /* PINSRW Gm,Ew,Ib */
            nextop = F8;
            GET_ED;
@@ -1033,12 +1163,15 @@

        _0f_0xD7:                   /* PMOVMSKB Gd,Em */
            nextop = F8;
-            GET_EM;
-            GD.dword[0] = 0;
-            for (int i=0; i<8; ++i)
-                if(EM->ub[i]&0x80)
-                    GD.dword[0] |= (1<<i);
-            NEXT;
+            if((nextop&0xC0)==0xC0) {
+                GET_EM;
+                GD.dword[0] = 0;
+                for (int i=0; i<8; ++i)
+                    if(EM->ub[i]&0x80)
+                        GD.dword[0] |= (1<<i);
+                NEXT;
+            } else
+                goto _default;
        _0f_0xD8:                   /* PSUBUSB Gm,Em */
            nextop = F8;
            GET_EM;
@@ -1055,7 +1188,12 @@
                GM.uw[i] = (tmp32s < 0) ? 0 : tmp32s;
            }
            NEXT;
-
+        _0f_0xDA:                   /* PMINUB Gm,Em */
+            nextop = F8;
+            GET_EM;
+            for (int i=0; i<8; ++i)
+                GM.ub[i] = (GM.ub[i]<EM->ub[i])?GM.ub[i]:EM->ub[i];
+            NEXT;
        _0f_0xDB:                   /* PAND Gm,Em */
            nextop = F8;
            GET_EM;
@@ -1077,7 +1215,12 @@
                GM.uw[i] = (tmp32u>65535) ? 65535 : tmp32u;
            }
            NEXT;
-
+        _0f_0xDE:                   /* PMAXUB Gm,Em */
+            nextop = F8;
+            GET_EM;
+            for (int i=0; i<8; ++i)
+                GM.ub[i] = (GM.ub[i]>EM->ub[i])?GM.ub[i]:EM->ub[i];
+            NEXT;
        _0f_0xDF:                   /* PANDN Gm,Em */
            nextop = F8;
            GET_EM;
@@ -1111,20 +1254,17 @@
                    GM.sd[i] >>= tmp8u;
            }
            NEXT;
-        _0f_0xE3:                   /* PSRAQ Gm, Em */
+        _0f_0xE3:                   /* PAVGW Gm, Em */
            nextop = F8;
            GET_EM;
-            if(EM->q>63)
-                tmp8u = 64;
-            else
-                tmp8u = EM->ub[0];
-            GM.sq >>= tmp8u;
+            for(int i=0; i<4; ++i)
+                GM.uw[i] = ((uint32_t)GM.uw[i]+EM->uw[i]+1)>>1;
            NEXT;
        _0f_0xE4:                   /* PMULHUW Gm, Em */
            nextop = F8;
            GET_EM;
            for(int i=0; i<4; ++i) {
-                tmp32u = (int32_t)GM.uw[i] * EM->uw[i];
+                tmp32u = (uint32_t)GM.uw[i] * EM->uw[i];
                GM.uw[i] = (tmp32u>>16)&0xffff;
            }
            NEXT;
@@ -1277,7 +1417,11 @@
            for(int i=0; i<2; ++i)
                GM.sd[i] -= EM->sd[i];
            NEXT;
-
+        _0f_0xFB:                   /* PSUBQ Gm,Em */
+            nextop = F8;
+            GET_EM;
+            GM.sq -= EM->sq;
+            NEXT;
        _0f_0xFC:                   /* PADDB Gm, Em */
            nextop = F8;
            GET_EM;
@@ -2,16 +2,29 @@
    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;
@@ -30,11 +43,22 @@
        NEXT;
    _66_0x07:                      /* POP ES */
        emu->segs[_ES] = Pop16(emu);    // no check, no use....
-        emu->segs_clean[_ES] = 0;
+        emu->segs_serial[_ES] = 0;
        NEXT;

    _66_0x0F:                      /* 66 0f prefix */
-        #include "run660f.h"
+        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_0x1F:                      /* POP DS */
+        emu->segs[_DS] = Pop16(emu);    // no check, no use....
+        emu->segs_serial[_DS] = 0;
+        NEXT;
        
    _66_0x26:                      /* ES: */
        // ignored
@@ -129,6 +153,13 @@
        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

@@ -203,14 +234,14 @@
    _66_0x8C:                              /* MOV Ew,Seg */
        nextop = F8;
        GET_EW;
-        EW->word[0] = emu->segs[(nextop&38)>>3];
+        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_clean[((nextop&0x38)>>3)] = 0;
+        emu->segs_serial[((nextop&0x38)>>3)] = 0;
        NEXT;
    _66_0x8F:                              /* POP Ew */
        nextop = F8;
@@ -241,15 +272,12 @@

    _66_0x9C:                              /* PUSHFW */
        CHECK_FLAGS(emu);
-        PackFlags(emu);
-        Push16(emu, (uint16_t)emu->packed_eflags.x32);
+        Push16(emu, (uint16_t)emu->eflags.x32);
        NEXT;
    _66_0x9D:                              /* POPFW */
        CHECK_FLAGS(emu);
-        PackFlags(emu);
-        emu->packed_eflags.x32 &=0xffff0000;
-        emu->packed_eflags.x32 |= (Pop16(emu) & 0x3F7FD7) | 0x2;
-        UnpackFlags(emu);
+        emu->eflags.x32 &=0xffff0000;
+        emu->eflags.x32 |= (Pop16(emu) & 0x3F7FD7) | 0x2;
        NEXT;

    _66_0xA1:                              /* MOV AX,Ow */
@@ -327,6 +355,22 @@
        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;
@@ -350,6 +394,7 @@
        Run66D9(emu);
        ip = R_EIP;
        if(emu->quit) goto fini;
+        STEP
        NEXT;

    _66_0xDD:
@@ -358,8 +403,17 @@
        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 */
@@ -10,11 +10,7 @@
        case 0xC5:
        case 0xC6:
        case 0xC7:  /* FADD */
-            #ifdef USE_FLOAT
-            ST0.f += ST(nextop&7).f;
-            #else
            ST0.d += ST(nextop&7).d;
-            #endif
            break;
        case 0xC8:
        case 0xC9:
@@ -24,11 +20,7 @@
        case 0xCD:
        case 0xCE:
        case 0xCF:  /* FMUL */
-            #ifdef USE_FLOAT
-            ST0.f *= ST(nextop&7).f;
-            #else
            ST0.d *= ST(nextop&7).d;
-            #endif
            break;
        case 0xD0:
        case 0xD1:
@@ -38,11 +30,7 @@
        case 0xD5:
        case 0xD6:
        case 0xD7:  /* FCOM */
-            #ifdef USE_FLOAT
-            fpu_fcom(emu, ST(nextop&7).f);
-            #else
            fpu_fcom(emu, ST(nextop&7).d);
-            #endif
            break;
        case 0xD8:
        case 0xD9:
@@ -52,11 +40,7 @@
        case 0xDD:
        case 0xDE:
        case 0xDF:  /* FCOMP */
-            #ifdef USE_FLOAT
-            fpu_fcom(emu, ST(nextop&7).f);
-            #else
            fpu_fcom(emu, ST(nextop&7).d);
-            #endif
            fpu_do_pop(emu);
            break;
        case 0xE0:
@@ -67,11 +51,7 @@
        case 0xE5:
        case 0xE6:
        case 0xE7:  /* FSUB */
-            #ifdef USE_FLOAT
-            ST0.f -= ST(nextop&7).f;
-            #else
            ST0.d -= ST(nextop&7).d;
-            #endif
            break;
        case 0xE8:
        case 0xE9:
@@ -81,11 +61,7 @@
        case 0xED:
        case 0xEE:
        case 0xEF:  /* FSUBR */
-            #ifdef USE_FLOAT
-            ST0.f = ST(nextop&7).f - ST0.f;
-            #else
            ST0.d = ST(nextop&7).d - ST0.d;
-            #endif
            break;
        case 0xF0:
        case 0xF1:
@@ -95,11 +71,7 @@
        case 0xF5:
        case 0xF6:
        case 0xF7:  /* FDIV */
-            #ifdef USE_FLOAT
-            ST0.f /= ST(nextop&7).f;
-            #else
            ST0.d /= ST(nextop&7).d;
-            #endif
            break;
        case 0xF8:
        case 0xF9:
@@ -109,46 +81,26 @@
        case 0xFD:
        case 0xFE:
        case 0xFF:  /* FDIVR */
-            #ifdef USE_FLOAT
-            ST0.f = ST(nextop&7).f / ST0.f;
-            #else
            ST0.d = ST(nextop&7).d / ST0.d;
-            #endif
            break;
        default:
        switch((nextop>>3)&7) {
            case 0:         /* FADD ST0, float */
                GET_ED;
                if(!(((uintptr_t)ED)&3))
-                    #ifdef USE_FLOAT
-                    ST0.f += *(float*)ED;
-                    #else
                    ST0.d += *(float*)ED;
-                    #endif
                else {
-                    *(uint32_t*)&f = ED->dword[0];
-                    #ifdef USE_FLOAT
-                    ST0.f += f;
-                    #else
+                    memcpy(&f, ED, sizeof(float));
                    ST0.d += f;
-                    #endif
                }
                break;
            case 1:         /* FMUL ST0, float */
                GET_ED;
                if(!(((uintptr_t)ED)&3))
-                    #ifdef USE_FLOAT
-                    ST0.f *= *(float*)ED;
-                    #else
                    ST0.d *= *(float*)ED;
-                    #endif
                else {
-                    *(uint32_t*)&f = ED->dword[0];
-                    #ifdef USE_FLOAT
-                    ST0.f *= f;
-                    #else
+                    memcpy(&f, ED, sizeof(float));
                    ST0.d *= f;
-                    #endif
                }
                break;
            case 2:      /* FCOM ST0, float */
@@ -156,7 +108,7 @@
                if(!(((uintptr_t)ED)&3))
                    fpu_fcom(emu, *(float*)ED);
                else {
-                    *(uint32_t*)&f = ED->dword[0];
+                    memcpy(&f, ED, sizeof(float));
                    fpu_fcom(emu, f);
                }
                break;
@@ -165,7 +117,7 @@
                if(!(((uintptr_t)ED)&3))
                    fpu_fcom(emu, *(float*)ED);
                else {
-                    *(uint32_t*)&f = ED->dword[0];
+                    memcpy(&f, ED, sizeof(float));
                    fpu_fcom(emu, f);
                }
                fpu_do_pop(emu);
@@ -173,69 +125,37 @@
            case 4:         /* FSUB ST0, float */
                GET_ED;
                if(!(((uintptr_t)ED)&3))
-                    #ifdef USE_FLOAT
-                    ST0.f -= *(float*)ED;
-                    #else
                    ST0.d -= *(float*)ED;
-                    #endif
                else {
-                    *(uint32_t*)&f = ED->dword[0];
-                    #ifdef USE_FLOAT
-                    ST0.f -= f;
-                    #else
+                    memcpy(&f, ED, sizeof(float));
                    ST0.d -= f;
-                    #endif
                }
                break;
            case 5:         /* FSUBR ST0, float */
                GET_ED;
                if(!(((uintptr_t)ED)&3))
-                    #ifdef USE_FLOAT
-                    ST0.f = *(float*)ED - ST0.f;
-                    #else
                    ST0.d = *(float*)ED - ST0.d;
-                    #endif
                else {
-                    *(uint32_t*)&f = ED->dword[0];
-                    #ifdef USE_FLOAT
-                    ST0.f = f - ST0.f;
-                    #else
+                    memcpy(&f, ED, sizeof(float));
                    ST0.d = f - ST0.d;
-                    #endif
                }
                break;
            case 6:         /* FDIV ST0, float */
                GET_ED;
                if(!(((uintptr_t)ED)&3))
-                    #ifdef USE_FLOAT
-                    ST0.f /= *(float*)ED;
-                    #else
                    ST0.d /= *(float*)ED;
-                    #endif
                else {
-                    *(uint32_t*)&f = ED->dword[0];
-                    #ifdef USE_FLOAT
-                    ST0.f /= f;
-                    #else
+                    memcpy(&f, ED, sizeof(float));
                    ST0.d /= f;
-                    #endif
                }
                break;
            case 7:         /* FDIVR ST0, float */
                GET_ED;
                if(!(((uintptr_t)ED)&3))
-                    #ifdef USE_FLOAT
-                    ST0.f = *(float*)ED / ST0.f;
-                    #else
                    ST0.d = *(float*)ED / ST0.d;
-                    #endif
                else {
-                    *(uint32_t*)&f = ED->dword[0];
-                    #ifdef USE_FLOAT
-                    ST0.f = f / ST0.f;
-                    #else
+                    memcpy(&f, ED, sizeof(float));
                    ST0.d = f / ST0.d;
-                    #endif
                }
                break;
            default:
@@ -1,5 +1,3 @@
-#pragma GCC diagnostic push
-#pragma GCC diagnostic ignored "-Wstrict-aliasing"
    nextop = F8;
    switch (nextop) {
        case 0xC0:
@@ -10,9 +8,9 @@
        case 0xC5:
        case 0xC6:
        case 0xC7:  /* FLD STx */
-            ll = ST(nextop&7).ll;
+            ll = ST(nextop&7).sq;
            fpu_do_push(emu);
-            ST0.ll = ll;
+            ST0.sq = ll;
            break;
        case 0xC8:
        case 0xC9:
@@ -22,27 +20,19 @@
        case 0xCD:
        case 0xCE:
        case 0xCF:  /* FXCH STx */
-            ll = ST(nextop&7).ll;
-            ST(nextop&7).ll = ST0.ll;
-            ST0.ll = ll;
+            ll = ST(nextop&7).sq;
+            ST(nextop&7).sq = ST0.sq;
+            ST0.sq = ll;
            break;

        case 0xD0:  /* FNOP */
            break;

        case 0xE0:  /* FCHS */
-            #ifdef USE_FLOAT
-            ST0.f = -ST0.f;
-            #else
            ST0.d = -ST0.d;
-            #endif
            break;
        case 0xE1:  /* FABS */
-            #ifdef USE_FLOAT
-            ST0.f = fabsf(ST0.f);
-            #else
            ST0.d = fabs(ST0.d);
-            #endif
            break;
        
        case 0xE4:  /* FTST */
@@ -54,153 +44,96 @@

        case 0xE8:  /* FLD1 */
            fpu_do_push(emu);
-            #ifdef USE_FLOAT
-            ST0.f = 1.0f;
-            #else
            ST0.d = 1.0;
-            #endif
            break;
        case 0xE9:  /* FLDL2T */
            fpu_do_push(emu);
-            #ifdef USE_FLOAT
-            ST0.f = L2T;
-            #else
            ST0.d = L2T;
-            #endif
            break;
        case 0xEA:  /* FLDL2E */
            fpu_do_push(emu);
-            #ifdef USE_FLOAT
-            ST0.f = L2E;
-            #else
            ST0.d = L2E;
-            #endif
            break;
        case 0xEB:  /* FLDPI */
            fpu_do_push(emu);
-            #ifdef USE_FLOAT
-            ST0.f = PI;
-            #else
            ST0.d = PI;
-            #endif
            break;
        case 0xEC:  /* FLDLG2 */
            fpu_do_push(emu);
-            #ifdef USE_FLOAT
-            ST0.f = LG2;
-            #else
            ST0.d = LG2;
-            #endif
            break;
        case 0xED:  /* FLDLN2 */
            fpu_do_push(emu);
-            #ifdef USE_FLOAT
-            ST0.f = LN2;
-            #else
            ST0.d = LN2;
-            #endif
            break;
        case 0xEE:  /* FLDZ */
            fpu_do_push(emu);
-            #ifdef USE_FLOAT
-            ST0.f = 0.0f;
-            #else
            ST0.d = 0.0;
-            #endif
            break;

        case 0xF0:  /* F2XM1 */
-            #ifdef USE_FLOAT
-            ST0.f = exp2f(ST0.f) - 1.0f;
-            #else
            ST0.d = exp2(ST0.d) - 1.0;
-            #endif
            break;
        case 0xF1:  /* FYL2X */
-            #ifdef USE_FLOAT
-            ST(1).f = log2f(ST0.f)*ST(1).f;
-            #else
-            ST(1).d = log2(ST0.d)*ST(1).d;
-            #endif
+            ST(1).d *= log2(ST0.d);
            fpu_do_pop(emu);
            break;
        case 0xF2:  /* FPTAN */
-            #ifdef USE_FLOAT
-            ST0.f = tanf(ST0.f);
-            fpu_do_push(emu);
-            ST0.f = 1.0f;
-            #else
            ST0.d = tan(ST0.d);
            fpu_do_push(emu);
            ST0.d = 1.0;
-            #endif
            emu->sw.f.F87_C2 = 0;
            break;
        case 0xF3:  /* FPATAN */
-            #ifdef USE_FLOAT
-            ST1.f = atan2f(ST1.f, ST0.f);
-            #else
            ST1.d = atan2(ST1.d, ST0.d);
-            #endif
            fpu_do_pop(emu);
            break;
        case 0xF4:  /* FXTRACT */
-            #ifdef USE_FLOAT
-            tmp32s = (ST0.ll&0x7f700000)>>23;
-            tmp32s -= 127;
-            ST0.f /= exp2f(tmp32s);
-            fpu_do_push(emu);
-            ST0.f = tmp32s;
-            #else
-            tmp32s = (ST0.ll&0x7ff0000000000000LL)>>52;
-            tmp32s -= 1023;
-            ST0.d /= exp2(tmp32s);
+            ST0.d = frexp(ST0.d, &tmp32s);
            fpu_do_push(emu);
            ST0.d = tmp32s;
-            #endif
            break;

        case 0xF8:  /* FPREM */
-            #ifdef USE_FLOAT
-            ll = ST0.f / ST1.f;
-            ST0.f -= ST1.f * ll;
-            #else
-            ll = ST0.d / ST1.d;
-            ST0.d -= ST1.d * ll;
-            #endif
-            emu->sw.f.F87_C2 = 0;
-            emu->sw.f.F87_C0 = (ll&1);
-            emu->sw.f.F87_C3 = ((ll>>1)&1);
-            emu->sw.f.F87_C1 = ((ll>>2)&1);
-            break;
-        case 0xF5:  /* FPREM1 */
-            // get exponant(ST(0))-exponant(ST(1)) in temp32s
-            #ifdef USE_FLOAT
-            tmp32s = ((ST0.ll&0x7f700000)>>23) - ((ST1.ll&0x7f700000)>>23);
-            #else
-            tmp32s = ((ST0.ll&0x7ff0000000000000LL)>>52) - ((ST1.ll&0x7ff0000000000000LL)>>52);
-            #endif
+            {
+                int e0, e1;
+                frexp(ST0.d, &e0);
+                frexp(ST1.d, &e1);
+                tmp32s = e0 - e1;
+            }
            if(tmp32s<64)
            {
-                #ifdef USE_FLOAT
-                ll = (int64_t)round(ST0.f/ST1.f);
-                ST0.f = ST0.f - (ST1.f*ll);
-                #else
-                ll = (int64_t)round(ST0.d/ST1.d);
+                ll = (int64_t)floor(ST0.d/ST1.d);
                ST0.d = ST0.d - (ST1.d*ll);
-                #endif
                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 {
-                #ifdef USE_FLOAT
-                ll = (int64_t)(floor((ST0.f/ST1.f))/powf(2, tmp32s - 32));
-                ST0.f = ST0.f - ST1.f*ll*powf(2, 32);
-                #else
-                ll = (int64_t)(floor((ST0.d/ST1.d))/pow(2, tmp32s - 32));
-                ST0.d = ST0.d - ST1.d*ll*pow(2, tmp32s - 32);
-                #endif
+                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;
+            }
+            break;
+        case 0xF5:  /* FPREM1 */
+            // get exponant(ST(0))-exponant(ST(1)) in temp32s
+            {
+                int e0, e1;
+                frexp(ST0.d, &e0);
+                frexp(ST1.d, &e1);
+                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;
            }
            break;
@@ -211,56 +144,32 @@
            emu->top=(emu->top+1)&7;    // this will probably break a few things
            break;
        case 0xF9:  /* FYL2XP1 */
-            #ifdef USE_FLOAT
-            ST(1).f = log2f(ST0.f + 1.0f)*ST(1).f;
-            #else
-            ST(1).d = log2(ST0.d + 1.0)*ST(1).d;
-            #endif
+            ST(1).d *= log2(ST0.d + 1.0);
            fpu_do_pop(emu);
            break;
        case 0xFA:  /* FSQRT */
-            #ifdef USE_FLOAT
-            ST0.f = sqrtf(ST0.f);
-            #else
            ST0.d = sqrt(ST0.d);
-            #endif
            break;
        case 0xFB:  /* FSINCOS */
            fpu_do_push(emu);
-            #ifdef USE_FLOAT
-            sincosf(ST1.f, &ST1.f, &ST0.f);
-            #else
            sincos(ST1.d, &ST1.d, &ST0.d);
-            #endif
+            emu->sw.f.F87_C2 = 0;
            break;
        case 0xFC:  /* FRNDINT */
-            #ifdef USE_FLOAT
-            ST0.f = fpu_round(emu, ST0.f);
-            #else
            ST0.d = fpu_round(emu, ST0.d);
-            #endif
            break;
        case 0xFD:  /* FSCALE */
            // this could probably be done by just altering the exponant part of the float...
-            #ifdef USE_FLOAT
-            ST0.f *= exp2f(truncf(ST1.f));
-            #else
-            ST0.d *= exp2(trunc(ST1.d));
-            #endif
+            if(ST0.d!=0.0)
+                ST0.d *= exp2(trunc(ST1.d));
            break;
        case 0xFE:  /* FSIN */
-            #ifdef USE_FLOAT
-            ST0.f = sinf(ST0.f);
-            #else
            ST0.d = sin(ST0.d);
-            #endif
+            emu->sw.f.F87_C2 = 0;
            break;
        case 0xFF:  /* FCOS */
-            #ifdef USE_FLOAT
-            ST0.f = cosf(ST0.f);
-            #else
            ST0.d = cos(ST0.d);
-            #endif
+            emu->sw.f.F87_C2 = 0;
            break;


@@ -288,42 +197,30 @@
            case 0:     /* FLD ST0, Ed float */
                GET_ED;
                fpu_do_push(emu);
-                #ifdef USE_FLOAT
-                ST0.ll = ED->dword[0];
-                #else
                if(!(((uintptr_t)ED)&3))
                    ST0.d = *(float*)ED;
                else {
-                    memcpy(&f, ED->dword, sizeof(float));
+                    memcpy(&f, ED, sizeof(float));
                    ST0.d = f;
                }
-                #endif
                break;
            case 2:     /* FST Ed, ST0 */
                GET_ED;
-                #ifdef USE_FLOAT
-                *(float*)ED = ST0.f;
-                #else
                if(!(((uintptr_t)ED)&3))
                    *(float*)ED = ST0.d;
                else {
                    f = ST0.d;
-                    memcpy(ED->dword, &f, sizeof(float));
+                    memcpy(ED, &f, sizeof(float));
                }
-                #endif
                break;
            case 3:     /* FSTP Ed, ST0 */
                GET_ED;
-                #ifdef USE_FLOAT
-                ED->dword[0] = ST0.ll;
-                #else
                if(!(((uintptr_t)ED)&3))
                    *(float*)ED = ST0.d;
                else {
                    f = ST0.d;
-                    memcpy(ED->dword, &f, sizeof(float));
+                    memcpy(ED, &f, sizeof(float));
                }
-                #endif
                fpu_do_pop(emu);
                break;
            case 4:     /* FLDENV m */
@@ -341,12 +238,6 @@
                // warning, incomplete
                GET_ED;
                fpu_savenv(emu, (char*)ED, 0);
-                ED->dword[0] = emu->cw;
-                ED->dword[1] = emu->sw.x16;
-                // tagword: 2bits*8
-                ED->word[4] = 0;
-                for (int i=0; i<8; ++i)
-                    ED->word[4] |= (emu->p_regs[i].tag)<<(i*2);
                // intruction pointer: 48bits
                // data (operand) pointer: 48bits
                // last opcode: 11bits save: 16bits restaured (1st and 2nd opcode only)
@@ -359,4 +250,3 @@
                goto _default;
        }
    }
-#pragma GCC diagnostic pop
@@ -11,7 +11,7 @@
    case 0xC7:
        CHECK_FLAGS(emu);
        if(ACCESS_FLAG(F_CF))
-            ST0.ll = ST(nextop&7).ll;
+            ST0.q = ST(nextop&7).q;
        break;
    case 0xC8:      /* FCMOVE ST(0), ST(i) */
    case 0xC9:
@@ -23,7 +23,7 @@
    case 0xCF:
        CHECK_FLAGS(emu);
        if(ACCESS_FLAG(F_ZF))
-            ST0.ll = ST(nextop&7).ll;
+            ST0.q = ST(nextop&7).q;
        break;
    case 0xD0:      /* FCMOVBE ST(0), ST(i) */
    case 0xD1:
@@ -35,7 +35,7 @@
    case 0xD7:
        CHECK_FLAGS(emu);
        if(ACCESS_FLAG(F_CF) || ACCESS_FLAG(F_ZF))
-            ST0.ll = ST(nextop&7).ll;
+            ST0.q = ST(nextop&7).q;
        break;
    case 0xD8:      /* FCMOVU ST(0), ST(i) */
    case 0xD9:
@@ -47,15 +47,11 @@
    case 0xDF:
        CHECK_FLAGS(emu);
        if(ACCESS_FLAG(F_PF))
-            ST0.ll = ST(nextop&7).ll;
+            ST0.q = ST(nextop&7).q;
        break;
    
    case 0xE9:      /* FUCOMPP */
-        #ifdef USE_FLOAT
-        fpu_fcom(emu, ST1.f);
-        #else
        fpu_fcom(emu, ST1.d);   // bad, should handle QNaN and IA interrupt
-        #endif
        fpu_do_pop(emu);
        fpu_do_pop(emu);
        break;
@@ -76,19 +72,11 @@
        switch((nextop>>3)&7) {
            case 0:     /* FIADD ST0, Ed int */
                GET_ED;
-                #ifdef USE_FLOAT
-                ST0.f += ED->sdword[0];
-                #else
                ST0.d += ED->sdword[0];
-                #endif
                break;
            case 1:     /* FIMUL ST0, Ed int */
                GET_ED;
-                #ifdef USE_FLOAT
-                ST0.f *= ED->sdword[0];
-                #else
                ST0.d *= ED->sdword[0];
-                #endif
                break;
            case 2:     /* FICOM ST0, Ed int */
                GET_ED;
@@ -101,35 +89,19 @@
                break;
            case 4:     /* FISUB ST0, Ed int */
                GET_ED;
-                #ifdef USE_FLOAT
-                ST0.f -= ED->sdword[0];
-                #else
                ST0.d -= ED->sdword[0];
-                #endif
                break;
            case 5:     /* FISUBR ST0, Ed int */
                GET_ED;
-                #ifdef USE_FLOAT
-                ST0.f = ED->sdword[0] - ST0.f;
-                #else
-                ST0.d = ED->sdword[0] - ST0.d;
-                #endif
+                ST0.d = (double)ED->sdword[0] - ST0.d;
                break;
            case 6:     /* FIDIV ST0, Ed int */
                GET_ED;
-                #ifdef USE_FLOAT
-                ST0.f /= ED->sdword[0];
-                #else
                ST0.d /= ED->sdword[0];
-                #endif
                break;
            case 7:     /* FIDIVR ST0, Ed int */
                GET_ED;
-                #ifdef USE_FLOAT
-                ST0.f = ED->sdword[0] / ST0.f;
-                #else
-                ST0.d = ED->sdword[0] / ST0.d;
-                #endif
+                ST0.d = (double)ED->sdword[0] / ST0.d;
                break;
        }
    }
@@ -10,7 +10,7 @@
    case 0xC7:
        CHECK_FLAGS(emu);
        if(!ACCESS_FLAG(F_CF))
-            ST0.ll = ST(nextop&7).ll;
+            ST0.q = ST(nextop&7).q;
        break;
    case 0xC8:      /* FCMOVNE ST(0), ST(i) */
    case 0xC9:
@@ -22,7 +22,7 @@
    case 0xCF:
        CHECK_FLAGS(emu);
        if(!ACCESS_FLAG(F_ZF))
-            ST0.ll = ST(nextop&7).ll;
+            ST0.q = ST(nextop&7).q;
        break;
    case 0xD0:      /* FCMOVNBE ST(0), ST(i) */
    case 0xD1:
@@ -34,7 +34,7 @@
    case 0xD7:
        CHECK_FLAGS(emu);
        if(!(ACCESS_FLAG(F_CF) || ACCESS_FLAG(F_ZF)))
-            ST0.ll = ST(nextop&7).ll;
+            ST0.q = ST(nextop&7).q;
        break;
    case 0xD8:      /* FCMOVNU ST(0), ST(i) */
    case 0xD9:
@@ -46,9 +46,11 @@
    case 0xDF:
        CHECK_FLAGS(emu);
        if(!ACCESS_FLAG(F_PF))
-            ST0.ll = ST(nextop&7).ll;
+            ST0.q = ST(nextop&7).q;
        break;

+    case 0xE1:      /* FDISI8087_NOP */
+        break;
    case 0xE2:      /* FNCLEX */
        //Clears the floating-point exception flags (PE, UE, OE, ZE, DE, and IE), 
        // the exception summary status flag (ES), the stack fault flag (SF), and the busy flag (B) in the FPU status word
@@ -73,11 +75,7 @@
    case 0xED:
    case 0xEE:
    case 0xEF:
-        #ifdef USE_FLOAT
-        fpu_fcomi(emu, ST(nextop&7).f);
-        #else
        fpu_fcomi(emu, ST(nextop&7).d);   // bad, should handle QNaN and IA interrupt
-        #endif
        break;

    case 0xF0:  /* FCOMI ST0, STx */
@@ -88,14 +86,9 @@
    case 0xF5:
    case 0xF6:
    case 0xF7:
-        #ifdef USE_FLOAT
-        fpu_fcomi(emu, ST(nextop&7).f);
-        #else
        fpu_fcomi(emu, ST(nextop&7).d);
-        #endif
        break;
    case 0xE0:
-    case 0xE1:
    case 0xE4:
    case 0xE5:
    case 0xE6:
@@ -103,134 +96,49 @@
        goto _default;
    default:
        switch((nextop>>3)&7) {
-            case 0: /* FILD ST0, Gd */
+            case 0: /* FILD ST0, Ed */
                GET_ED;
                fpu_do_push(emu);
-                #ifdef USE_FLOAT
-                ST0.f = ED->sdword[0];
-                #else
                ST0.d = ED->sdword[0];
-                #endif
                break;
            case 1: /* FISTTP Ed, ST0 */
                GET_ED;
-                #ifdef USE_FLOAT
-                tmp32s = ST0.f; // TODO: Handling of FPU Exception
-                if(tmp32s==0x7fffffff && isgreater(ST0.f, (float)(int32_t)0x7fffffff))
-                    tmp32s = 0x80000000;
-                #else
                tmp32s = ST0.d; // TODO: Handling of FPU Exception
                if(tmp32s==0x7fffffff && isgreater(ST0.d, (double)(int32_t)0x7fffffff))
                    tmp32s = 0x80000000;
-                #endif
                fpu_do_pop(emu);
                ED->sdword[0] = tmp32s;
                break;
            case 2: /* FIST Ed, ST0 */
                GET_ED;
-                #ifdef USE_FLOAT
-                if(isgreater(ST0.f, (float)(int32_t)0x7fffffff) || isless(ST0.f, -(float)(int32_t)0x7fffffff))
+                if(isgreater(ST0.d, (double)(int32_t)0x7fffffff) || isless(ST0.d, -(double)(int32_t)0x7fffffff) || !isfinite(ST0.d))
                    ED->sdword[0] = 0x80000000;
                else {
-                    switch(emu->round) {
-                        case ROUND_Nearest:
-                            ED->sdword[0] = floorf(ST0.f+0.5);
-                            break;
-                        case ROUND_Down:
-                            ED->sdword[0] = floorf(ST0.f);
-                            break;
-                        case ROUND_Up:
-                            ED->sdword[0] = ceilf(ST0.f);
-                            break;
-                        case ROUND_Chop:
-                            ED->sdword[0] = ST0.f;
-                            break;
-                    }
+                    volatile int32_t tmp = fpu_round(emu, ST0.d);    // tmp to avoid BUS ERROR
+                    ED->sdword[0] = tmp;
                }
-                #else
-                if(isgreater(ST0.d, (double)(int32_t)0x7fffffff) || isless(ST0.d, -(double)(int32_t)0x7fffffff))
-                    ED->sdword[0] = 0x80000000;
-                else {
-                    switch(emu->round) {
-                        case ROUND_Nearest:
-                            ED->sdword[0] = floor(ST0.d+0.5);
-                            break;
-                        case ROUND_Down:
-                            ED->sdword[0] = floor(ST0.d);
-                            break;
-                        case ROUND_Up:
-                            ED->sdword[0] = ceil(ST0.d);
-                            break;
-                        case ROUND_Chop:
-                            ED->sdword[0] = ST0.d;
-                            break;
-                    }
-                }
-                #endif
                break;
            case 3: /* FISTP Ed, ST0 */
                GET_ED;
-                #ifdef USE_FLOAT
-                if(isgreater(ST0.f, (float)(int32_t)0x7fffffff) || isless(ST0.f, -(float)(int32_t)0x7fffffff))
+                if(isgreater(ST0.d, (double)(int32_t)0x7fffffff) || isless(ST0.d, -(double)(int32_t)0x7fffffff) || !isfinite(ST0.d))
                    ED->sdword[0] = 0x80000000;
                else {
-                    switch(emu->round) {
-                        case ROUND_Nearest:
-                            ED->sdword[0] = floorf(ST0.f+0.5);
-                            break;
-                        case ROUND_Down:
-                            ED->sdword[0] = floorf(ST0.f);
-                            break;
-                        case ROUND_Up:
-                            ED->sdword[0] = ceilf(ST0.f);
-                            break;
-                        case ROUND_Chop:
-                            ED->sdword[0] = ST0.f;
-                            break;
-                    }
+                    volatile int32_t tmp = fpu_round(emu, ST0.d);    // tmp to avoid BUS ERROR
+                    ED->sdword[0] = tmp;
                }
-                #else
-                if(isgreater(ST0.d, (double)(int32_t)0x7fffffff) || isless(ST0.d, -(double)(int32_t)0x7fffffff))
-                    ED->sdword[0] = 0x80000000;
-                else {
-                    switch(emu->round) {
-                        case ROUND_Nearest:
-                            ED->sdword[0] = floor(ST0.d+0.5);
-                            break;
-                        case ROUND_Down:
-                            ED->sdword[0] = floor(ST0.d);
-                            break;
-                        case ROUND_Up:
-                            ED->sdword[0] = ceil(ST0.d);
-                            break;
-                        case ROUND_Chop:
-                            ED->sdword[0] = ST0.d;
-                            break;
-                    }
-                }
-                #endif
                fpu_do_pop(emu);
                break;
-            case 5: /* FLD ST0, Gt */
+            case 5: /* FLD ST0, Et */
                GET_ED;
                fpu_do_push(emu);
                memcpy(&STld(0).ld, ED, 10);
-                #ifdef USE_FLOAT
-                LD2D(&STld(0), &d);
-                ST0.f = d;
-                #else
                LD2D(&STld(0), &ST(0).d);
-                #endif
-                STld(0).ref = ST0.ll;
+                STld(0).uref = ST0.q;
                break;
            case 7: /* FSTP tbyte */
                GET_ED;
-                if(ST0.ll!=STld(0).ref)
-                    #ifdef USE_FLOAT
-                    {d = ST0.f; D2LD(&d, ED);}
-                    #else
+                if(ST0.q!=STld(0).uref)
                    D2LD(&ST0.d, ED);
-                    #endif
                else
                    memcpy(ED, &STld(0).ld, 10);
                fpu_do_pop(emu);
@@ -1,5 +1,3 @@
-#pragma GCC diagnostic push
-#pragma GCC diagnostic ignored "-Wstrict-aliasing"
    nextop = F8;
    switch(nextop) {
        case 0xC0:
@@ -10,11 +8,7 @@
        case 0xC5:
        case 0xC6:
        case 0xC7:  /* FADD */
-            #ifdef USE_FLOAT
-            ST(nextop&7).f += ST0.f;
-            #else
            ST(nextop&7).d += ST0.d;
-            #endif
            break;
        case 0xC8:
        case 0xC9:
@@ -24,11 +18,7 @@
        case 0xCD:
        case 0xCE:
        case 0xCF:  /* FMUL */
-            #ifdef USE_FLOAT
-            ST(nextop&7).f *= ST0.f;
-            #else
            ST(nextop&7).d *= ST0.d;
-            #endif
            break;
        case 0xD0:
        case 0xD1:
@@ -38,11 +28,7 @@
        case 0xD5:
        case 0xD6:
        case 0xD7:  /* FCOM */
-            #ifdef USE_FLOAT
-            fpu_fcom(emu, ST(nextop&7).f);
-            #else
            fpu_fcom(emu, ST(nextop&7).d);
-            #endif
            break;
        case 0xD8:
        case 0xD9:
@@ -52,11 +38,7 @@
        case 0xDD:
        case 0xDE:
        case 0xDF:  /* FCOMP */
-            #ifdef USE_FLOAT
-            fpu_fcom(emu, ST(nextop&7).f);
-            #else
-            fpu_fcom(emu, ST(nextop&7).d);  // TODO: is this ok?
-            #endif
+            fpu_fcom(emu, ST(nextop&7).d);
            fpu_do_pop(emu);
            break;
        case 0xE0:
@@ -67,11 +49,7 @@
        case 0xE5:
        case 0xE6:
        case 0xE7:  /* FSUBR */
-            #ifdef USE_FLOAT
-            ST(nextop&7).f = ST0.f -ST(nextop&7).f;
-            #else
            ST(nextop&7).d = ST0.d -ST(nextop&7).d;
-            #endif
            break;
        case 0xE8:
        case 0xE9:
@@ -81,11 +59,7 @@
        case 0xED:
        case 0xEE:
        case 0xEF:  /* FSUB */
-            #ifdef USE_FLOAT
-            ST(nextop&7).f -= ST0.f;
-            #else
            ST(nextop&7).d -= ST0.d;
-            #endif
            break;
        case 0xF0:
        case 0xF1:
@@ -95,11 +69,7 @@
        case 0xF5:
        case 0xF6:
        case 0xF7:  /* FDIVR */
-            #ifdef USE_FLOAT
-            ST(nextop&7).f = ST0.f / ST(nextop&7).f;
-            #else
            ST(nextop&7).d = ST0.d / ST(nextop&7).d;
-            #endif
            break;
        case 0xF8:
        case 0xF9:
@@ -109,54 +79,32 @@
        case 0xFD:
        case 0xFE:
        case 0xFF:  /* FDIV */
-            #ifdef USE_FLOAT
-            ST(nextop&7).f /=  ST0.f;
-            #else
            ST(nextop&7).d /=  ST0.d;
-            #endif
            break;
        default:
            GET_ED;
            switch((nextop>>3)&7) {
            case 0:         /* FADD ST0, double */
-                #ifdef USE_FLOAT
-                if(!(((uintptr_t)ED)&7))
-                    ST0.f += *(double*)ED;
-                else {
-                    *(uint64_t*)&d = *(uint64_t*)ED;
-                    ST0.f += d;
-                }
-                #else
                if(!(((uintptr_t)ED)&7))
                    ST0.d += *(double*)ED;
                else {
-                    *(uint64_t*)&d = *(uint64_t*)ED;
+                    memcpy(&d, ED, sizeof(double));
                    ST0.d += d;
                }
-                #endif
                break;
            case 1:         /* FMUL ST0, double */
-                #ifdef USE_FLOAT
-                if(!(((uintptr_t)ED)&7))
-                    ST0.f *= *(double*)ED;
-                else {
-                    *(uint64_t*)&d = *(uint64_t*)ED;
-                    ST0.f *= d;
-                }
-                #else
                if(!(((uintptr_t)ED)&7))
                    ST0.d *= *(double*)ED;
                else {
-                    *(uint64_t*)&d = *(uint64_t*)ED;
+                    memcpy(&d, ED, sizeof(double));
                    ST0.d *= d;
                }
-                #endif
                break;
            case 2:      /* FCOM ST0, double */
                if(!(((uintptr_t)ED)&7))
                    fpu_fcom(emu, *(double*)ED);
                else {
-                    *(uint64_t*)&d = *(uint64_t*)ED;
+                    memcpy(&d, ED, sizeof(double));
                    fpu_fcom(emu, d);
                }
                break;
@@ -164,78 +112,42 @@
                if(!(((uintptr_t)ED)&7))
                    fpu_fcom(emu, *(double*)ED);
                else {
-                    *(uint64_t*)&d = *(uint64_t*)ED;
+                    memcpy(&d, ED, sizeof(double));
                    fpu_fcom(emu, d);
                }
                fpu_do_pop(emu);
                break;
            case 4:         /* FSUB ST0, double */
-                #ifdef USE_FLOAT
-                if(!(((uintptr_t)ED)&7))
-                    ST0.f -= *(double*)ED;
-                else {
-                    *(uint64_t*)&d = *(uint64_t*)ED;
-                    ST0.f -= d;
-                }
-                #else
                if(!(((uintptr_t)ED)&7))
                    ST0.d -= *(double*)ED;
                else {
-                    *(uint64_t*)&d = *(uint64_t*)ED;
+                    memcpy(&d, ED, sizeof(double));
                    ST0.d -= d;
                }
-                #endif
                break;
            case 5:         /* FSUBR ST0, double */
-                #ifdef USE_FLOAT
-                if(!(((uintptr_t)ED)&7))
-                    ST0.f = *(double*)ED - ST0.f;
-                else {
-                    *(uint64_t*)&d = *(uint64_t*)ED;
-                    ST0.f = d - ST0.f;
-                }
-                #else
                if(!(((uintptr_t)ED)&7))
                    ST0.d = *(double*)ED - ST0.d;
                else {
-                    *(uint64_t*)&d = *(uint64_t*)ED;
+                    memcpy(&d, ED, sizeof(double));
                    ST0.d = d - ST0.d;
                }
-                #endif
                break;
            case 6:         /* FDIV ST0, double */
-                #ifdef USE_FLOAT
-                if(!(((uintptr_t)ED)&7))
-                    ST0.f /= *(double*)ED;
-                else {
-                    *(uint64_t*)&d = *(uint64_t*)ED;
-                    ST0.f /= d;
-                }
-                #else
                if(!(((uintptr_t)ED)&7))
                    ST0.d /= *(double*)ED;
                else {
-                    *(uint64_t*)&d = *(uint64_t*)ED;
+                    memcpy(&d, ED, sizeof(double));
                    ST0.d /= d;
                }
-                #endif
                break;
            case 7:         /* FDIVR ST0, double */
-                #ifdef USE_FLOAT
-                if(!(((uintptr_t)ED)&7))
-                    ST0.f = *(double*)ED / ST0.f;
-                else {
-                    *(uint64_t*)&d = *(uint64_t*)ED;
-                    ST0.f = d / ST0.f;
-                }
-                #else
                if(!(((uintptr_t)ED)&7))
                    ST0.d = *(double*)ED / ST0.d;
                else {
-                    *(uint64_t*)&d = *(uint64_t*)ED;
+                    memcpy(&d, ED, sizeof(double));
                    ST0.d = d / ST0.d;
                }
-                #endif
                break;
            default:
                goto _default;
@@ -20,7 +20,7 @@
    case 0xD5:
    case 0xD6:
    case 0xD7:
-        ST(nextop&7).ll = ST0.ll;
+        ST(nextop&7).q = ST0.q;
        break;
    case 0xD8:  /* FSTP ST0, STx */
    case 0xD9:
@@ -30,7 +30,7 @@
    case 0xDD:
    case 0xDE:
    case 0xDF:
-        ST(nextop&7).ll = ST0.ll;
+        ST(nextop&7).q = ST0.q;
        fpu_do_pop(emu);
        break;
    case 0xE0:  /* FUCOM ST0, STx */
@@ -41,11 +41,7 @@
    case 0xE5:
    case 0xE6:
    case 0xE7:
-        #ifdef USE_FLOAT
-        fpu_fcom(emu, ST(nextop&7).f);
-        #else
        fpu_fcom(emu, ST(nextop&7).d);   // bad, should handle QNaN and IA interrupt
-        #endif
        break;
    case 0xE8:  /* FUCOMP ST0, STx */
    case 0xE9:
@@ -55,11 +51,7 @@
    case 0xED:
    case 0xEE:
    case 0xEF:
-        #ifdef USE_FLOAT
-        fpu_fcom(emu, ST(nextop&7).f);
-        #else
        fpu_fcom(emu, ST(nextop&7).d);   // bad, should handle QNaN and IA interrupt
-        #endif
        fpu_do_pop(emu);
        break;

@@ -94,39 +86,37 @@
            case 0: /* FLD double */
                GET_ED;
                fpu_do_push(emu);
-                #ifdef USE_FLOAT
-                *(uint64_t*)&d = *(int64_t*)ED;
-                ST0.f = d;
-                #else
-                ST0.ll = *(int64_t*)ED;
-                #endif
+                if(!(((uintptr_t)ED)&7))
+                    ST0.d = *(double*)ED;
+                else {
+                    memcpy(&ST0.d, ED, sizeof(double));
+                }
                break;
            case 1: /* FISTTP ED qword */
                GET_ED;
-                #ifdef USE_FLOAT
-                *(int64_t*)ED = ST0.f;
-                #else
-                *(int64_t*)ED = ST0.d;
-                #endif
+                if(!(((uintptr_t)ED)&7))
+                    *(int64_t*)ED = ST0.d;
+                else {
+                    int64_t i64 = ST0.d;
+                    memcpy(ED, &i64, sizeof(int64_t));
+                }
                fpu_do_pop(emu);
                break;
            case 2: /* FST double */
                GET_ED;
-                #ifdef USE_FLOAT
-                d = ST0.f;
-                *(int64_t*)ED = *(uint64_t*)&d;
-                #else
-                *(int64_t*)ED = ST0.ll;
-                #endif
+                if(!(((uintptr_t)ED)&7))
+                    *(double*)ED = ST0.d;
+                else {
+                    memcpy(ED, &ST0.d, sizeof(double));
+                }
                break;
            case 3: /* FSTP double */
                GET_ED;
-                #ifdef USE_FLOAT
-                d = ST0.f;
-                *(int64_t*)ED = *(uint64_t*)&d;
-                #else
-                *(int64_t*)ED = ST0.ll;
-                #endif
+                if(!(((uintptr_t)ED)&7))
+                    *(double*)ED = ST0.d;
+                else {
+                    memcpy(ED, &ST0.d, sizeof(double));
+                }
                fpu_do_pop(emu);
                break;
            case 4: /* FRSTOR m108byte */
@@ -137,12 +127,7 @@
                    char* p =(char*)ED;
                    p += 28;
                    for (int i=0; i<8; ++i) {
-                        #ifdef USE_FLOAT
-                        d = ST(i).f;
-                        LD2D(p, &d);
-                        #else
                        LD2D(p, &ST(i).d);
-                        #endif
                        p+=10;
                    }
                }
@@ -157,12 +142,7 @@
                    char* p =(char*)ED;
                    p += 28;
                    for (int i=0; i<8; ++i) {
-                        #ifdef USE_FLOAT
-                        D2LD(&d, p);
-                        ST(i).f = d;
-                        #else
                        D2LD(&ST(i).d, p);
-                        #endif
                        p+=10;
                    }
                }
@@ -1,55 +1,41 @@
    nextop = F8;
    switch (nextop) {
-    case 0xC1:  /* FADDP ST1, ST0 */
-        #ifdef USE_FLOAT
-        ST(1).f += ST0.f;
-        #else
-        ST(1).d += ST0.d;
-        #endif
-        fpu_do_pop(emu);
-        break;
    case 0xC0:  /* FADDP STx, ST0 */
+    case 0xC1:  /* FADDP ST1, ST0 */
    case 0xC2:
    case 0xC3:
    case 0xC4:
    case 0xC5:
    case 0xC6:
    case 0xC7:
-        #ifdef USE_FLOAT
-        ST(nextop&7).f += ST0.f;
-        #else
        ST(nextop&7).d += ST0.d;
-        #endif
-        fpu_do_pop(emu);
-        break;
-    case 0xC9:  /* FMULP ST1, ST0 */
-        #ifdef USE_FLOAT
-        ST1.f *= ST0.f;
-        #else
-        ST1.d *= ST0.d;
-        #endif
        fpu_do_pop(emu);
        break;
    case 0xC8:  /* FMULP STx, ST0 */
+    case 0xC9:  /* FMULP ST1, ST0 */
    case 0xCA:
    case 0xCB:
    case 0xCC:
    case 0xCD:
    case 0xCE:
    case 0xCF:
-        #ifdef USE_FLOAT
-        ST(nextop&7).f *= ST0.f;
-        #else
        ST(nextop&7).d *= ST0.d;
-        #endif
        fpu_do_pop(emu);
        break;
+    case 0xD0:
+    case 0xD1:
+    case 0xD2:
+    case 0xD3:
+    case 0xD4:
+    case 0xD5:
+    case 0xD6:
+    case 0xD7:  /* FCOMP */
+        fpu_fcom(emu, ST(nextop&7).d);
+        fpu_do_pop(emu);
+        break;
+
    case 0xD9:  /* FCOMPP */
-        #ifdef USE_FLOAT
-        fpu_fcom(emu, ST1.f);
-        #else
        fpu_fcom(emu, ST1.d);
-        #endif
        fpu_do_pop(emu);
        fpu_do_pop(emu);
        break;
@@ -65,33 +51,18 @@
    case 0xE5:
    case 0xE6:
    case 0xE7:
-        #ifdef USE_FLOAT
-        ST(nextop&7).f = ST0.f - ST(nextop&7).f;
-        #else
        ST(nextop&7).d = ST0.d - ST(nextop&7).d;
-        #endif
-        fpu_do_pop(emu);
-        break;
-   case 0xE9:  /* FSUBP ST1, ST0 */
-        #ifdef USE_FLOAT
-        ST1.f -= ST0.f;
-        #else
-        ST1.d -= ST0.d;
-        #endif
        fpu_do_pop(emu);
        break;
    case 0xE8:  /* FSUBP STx, ST0 */
+    case 0xE9:  /* FSUBP ST1, ST0 */
    case 0xEA:
    case 0xEB:
    case 0xEC:
    case 0xED:
    case 0xEE:
    case 0xEF:
-        #ifdef USE_FLOAT
-        ST(nextop&7).f -= ST0.f;
-        #else
        ST(nextop&7).d -= ST0.d;
-        #endif
        fpu_do_pop(emu);
        break;
    case 0xF0:  /* FDIVRP STx, ST0 */
@@ -102,11 +73,7 @@
    case 0xF5:
    case 0xF6:
    case 0xF7:
-        #ifdef USE_FLOAT
-        ST(nextop&7).f = ST0.f / ST(nextop&7).f;
-        #else
        ST(nextop&7).d = ST0.d / ST(nextop&7).d;
-        #endif
        fpu_do_pop(emu);
        break;
    case 0xF8:  /* FDIVP STx, ST0 */
@@ -117,23 +84,11 @@
    case 0xFD:
    case 0xFE:
    case 0xFF:
-        #ifdef USE_FLOAT
-        ST(nextop&7).f /= ST0.f;
-        #else
        ST(nextop&7).d /= ST0.d;
-        #endif
        fpu_do_pop(emu);
        break;

-    case 0xD0:
-    case 0xD1:
-    case 0xD2:
-    case 0xD3:
-    case 0xD4:
-    case 0xD5:
-    case 0xD6:
-    case 0xD7:
-    case 0xD8:
+    case 0xD8:  
    case 0xDA:
    case 0xDB:
    case 0xDC:
@@ -146,51 +101,27 @@
        switch((nextop>>3)&7) {
            case 0:     /* FIADD ST0, Ew int */
                GET_EW;
-                #ifdef USE_FLOAT
-                ST0.f += EW->sword[0];
-                #else
                ST0.d += EW->sword[0];
-                #endif
                break;
            case 1:     /* FIMUL ST0, Ew int */
                GET_EW;
-                #ifdef USE_FLOAT
-                ST0.f *= EW->sword[0];
-                #else
                ST0.d *= EW->sword[0];
-                #endif
                break;
            case 4:     /* FISUB ST0, Ew int */
                GET_EW;
-                #ifdef USE_FLOAT
-                ST0.f -= EW->sword[0];
-                #else
                ST0.d -= EW->sword[0];
-                #endif
                break;
            case 5:     /* FISUBR ST0, Ew int */
                GET_EW;
-                #ifdef USE_FLOAT
-                ST0.f = EW->sword[0] - ST0.f;
-                #else
-                ST0.d = EW->sword[0] - ST0.d;
-                #endif
+                ST0.d = (double)EW->sword[0] - ST0.d;
                break;
            case 6:     /* FIDIV ST0, Ew int */
                GET_EW;
-                #ifdef USE_FLOAT
-                ST0.f /= EW->sword[0];
-                #else
                ST0.d /= EW->sword[0];
-                #endif
                break;
            case 7:     /* FIDIVR ST0, Ew int */
                GET_EW;
-                #ifdef USE_FLOAT
-                ST0.f = EW->sword[0] / ST0.f;
-                #else
-                ST0.d = EW->sword[0] / ST0.d;
-                #endif
+                ST0.d = (double)EW->sword[0] / ST0.d;
                break;
        default:
            goto _default;
@@ -1,5 +1,29 @@
    nextop = F8;
    switch (nextop) {
+    case 0xC0:  /* FFREEP STx */
+    case 0xC1:
+    case 0xC2:
+    case 0xC3:
+    case 0xC4:
+    case 0xC5:
+    case 0xC6:
+    case 0xC7:
+        fpu_do_free(emu, nextop-0xC0);
+        fpu_do_pop(emu);
+        break;
+
+    case 0xD0:  /* FSTP STx, ST0 */
+    case 0xD1:
+    case 0xD2:
+    case 0xD3:
+    case 0xD4:
+    case 0xD5:
+    case 0xD6:
+    case 0xD7:
+        ST(nextop&7).q = ST0.q;
+        fpu_do_pop(emu);
+        break;
+
    case 0xE0:  /* FNSTSW AX */
        emu->sw.f.F87_TOP = emu->top&7;
        R_AX = emu->sw.x16;
@@ -13,11 +37,7 @@
    case 0xED:
    case 0xEE:
    case 0xEF:
-        #ifdef USE_FLOAT
-        fpu_fcomi(emu, ST(nextop&7).f);
-        #else
        fpu_fcomi(emu, ST(nextop&7).d);   // bad, should handle QNaN and IA interrupt
-        #endif
        fpu_do_pop(emu);
        break;

@@ -29,22 +49,10 @@
    case 0xF5:
    case 0xF6:
    case 0xF7:
-        #ifdef USE_FLOAT
-        fpu_fcomi(emu, ST(nextop&7).f);
-        #else
        fpu_fcomi(emu, ST(nextop&7).d);
-        #endif
        fpu_do_pop(emu);
        break;

-    case 0xC0:
-    case 0xC1:
-    case 0xC2:
-    case 0xC3:
-    case 0xC4:
-    case 0xC5:
-    case 0xC6:
-    case 0xC7:
    case 0xC8:
    case 0xC9:
    case 0xCA:
@@ -53,14 +61,6 @@
    case 0xCD:
    case 0xCE:
    case 0xCF:
-    case 0xD0:
-    case 0xD1:
-    case 0xD2:
-    case 0xD3:
-    case 0xD4:
-    case 0xD5:
-    case 0xD6:
-    case 0xD7:
    case 0xD8:
    case 0xD9:
    case 0xDA:
@@ -92,105 +92,27 @@
            GET_EW;
            tmp16s = EW->sword[0];
            fpu_do_push(emu);
-            #ifdef USE_FLOAT
-            ST0.f = tmp16s;
-            #else
            ST0.d = tmp16s;
-            #endif
            break;
        case 1: /* FISTTP Ew, ST0 */
            GET_EW;
-            #ifdef USE_FLOAT
-            tmp16s = ST0.f;
-            #else
            tmp16s = ST0.d;
-            #endif
            EW->sword[0] = tmp16s;
            fpu_do_pop(emu);
            break;
        case 2: /* FIST Ew, ST0 */
            GET_EW;
-            #ifdef USE_FLOAT
-            if(isgreater(ST0.f, (float)(int32_t)0x7fff) || isless(ST0.f, -(float)(int32_t)0x7fff))
+            if(isgreater(ST0.d, (double)(int32_t)0x7fff) || isless(ST0.d, -(double)(int32_t)0x7fff) || !isfinite(ST0.d))
                EW->sword[0] = 0x8000;
-            else {
-                switch(emu->round) {
-                    case ROUND_Nearest:
-                        EW->sword[0] = floorf(ST0.f+0.5);
-                        break;
-                    case ROUND_Down:
-                        EW->sword[0] = floorf(ST0.f);
-                        break;
-                    case ROUND_Up:
-                        EW->sword[0] = ceilf(ST0.f);
-                        break;
-                    case ROUND_Chop:
-                        EW->sword[0] = ST0.f;
-                        break;
-                }
-            }
-            #else
-            if(isgreater(ST0.d, (double)(int32_t)0x7fff) || isless(ST0.d, -(double)(int32_t)0x7fff))
-                EW->sword[0] = 0x8000;
-            else {
-                switch(emu->round) {
-                    case ROUND_Nearest:
-                        EW->sword[0] = floor(ST0.d+0.5);
-                        break;
-                    case ROUND_Down:
-                        EW->sword[0] = floor(ST0.d);
-                        break;
-                    case ROUND_Up:
-                        EW->sword[0] = ceil(ST0.d);
-                        break;
-                    case ROUND_Chop:
-                        EW->sword[0] = ST0.d;
-                        break;
-                }
-            }
-            #endif
+            else
+                EW->sword[0] = fpu_round(emu, ST0.d);
            break;
        case 3: /* FISTP Ew, ST0 */
            GET_EW;
-            #ifdef USE_FLOAT
-            if(isgreater(ST0.f, (float)(int32_t)0x7fff) || isless(ST0.f, -(float)(int32_t)0x7fff))
+            if(isgreater(ST0.d, (double)(int32_t)0x7fff) || isless(ST0.d, -(double)(int32_t)0x7fff) || !isfinite(ST0.d))
                EW->sword[0] = 0x8000;
-            else {
-                switch(emu->round) {
-                    case ROUND_Nearest:
-                        EW->sword[0] = floorf(ST0.f+0.5);
-                        break;
-                    case ROUND_Down:
-                        EW->sword[0] = floorf(ST0.f);
-                        break;
-                    case ROUND_Up:
-                        EW->sword[0] = ceilf(ST0.f);
-                        break;
-                    case ROUND_Chop:
-                        EW->sword[0] = ST0.f;
-                        break;
-                }
-            }
-            #else
-            if(isgreater(ST0.d, (double)(int32_t)0x7fff) || isless(ST0.d, -(double)(int32_t)0x7fff))
-                EW->sword[0] = 0x8000;
-            else {
-                switch(emu->round) {
-                    case ROUND_Nearest:
-                        EW->sword[0] = floor(ST0.d+0.5);
-                        break;
-                    case ROUND_Down:
-                        EW->sword[0] = floor(ST0.d);
-                        break;
-                    case ROUND_Up:
-                        EW->sword[0] = ceil(ST0.d);
-                        break;
-                    case ROUND_Chop:
-                        EW->sword[0] = ST0.d;
-                        break;
-                }
-            }
-            #endif
+            else
+                EW->sword[0] = fpu_round(emu, ST0.d);
            fpu_do_pop(emu);
            break;
        case 4: /* FBLD ST0, tbytes */
@@ -202,13 +124,9 @@
            GET_ED;
            tmp64s = *(int64_t*)ED;
            fpu_do_push(emu);
-            #ifdef USE_FLOAT
-            ST0.f = tmp64s;
-            #else
            ST0.d = tmp64s;
-            #endif
-            STll(0).ll = tmp64s;
-            STll(0).ref = ST0.ll;
+            STll(0).sq = tmp64s;
+            STll(0).sref = ST0.sq;
            break;
        case 6: /* FBSTP tbytes, ST0 */
            GET_ED;
@@ -217,20 +135,27 @@
            break;
        case 7: /* FISTP i64 */
            GET_ED;
-            if(STll(0).ref==ST(0).ll) {
-                *(int64_t*)ED = STll(0).ll;
+            if((uintptr_t)ED & 0x7) {
+                // un-aligned!
+                if(STll(0).sref==ST(0).sq)
+                    memcpy(ED, &STll(0).sq, sizeof(int64_t));
+                else {
+                    int64_t i64;
+                    if(isgreater(ST0.d, (double)(int64_t)0x7fffffffffffffffLL) || isless(ST0.d, -(double)(int64_t)0x7fffffffffffffffLL) || !isfinite(ST0.d))
+                        i64 = 0x8000000000000000LL;
+                    else
+                        i64 = fpu_round(emu, ST0.d);
+                    memcpy(ED, &i64, sizeof(int64_t));
+                }
            } else {
-                #ifdef USE_FLOAT
-                if(isgreater(ST0.f, (float)(int64_t)0x7fffffffffffffffLL) || isless(ST0.f, (float)(int64_t)0x8000000000000000LL))
-                    *(int64_t*)ED = 0x8000000000000000LL;
-                else
-                    *(int64_t*)ED = (int64_t)ST0.f;
-                #else
-                if(isgreater(ST0.d, (double)(int64_t)0x7fffffffffffffffLL) || isless(ST0.d, (double)(int64_t)0x8000000000000000LL))
-                    *(int64_t*)ED = 0x8000000000000000LL;
-                else
-                    *(int64_t*)ED = (int64_t)ST0.d;
-                #endif
+                if(STll(0).sref==ST(0).sq)
+                    *(int64_t*)ED = STll(0).sq;
+                else {
+                    if(isgreater(ST0.d, (double)(int64_t)0x7fffffffffffffffLL) || isless(ST0.d, -(double)(int64_t)0x7fffffffffffffffLL) || !isfinite(ST0.d))
+                        *(int64_t*)ED = 0x8000000000000000LL;
+                    else
+                        *(int64_t*)ED = fpu_round(emu, ST0.d);
+                }
            }
            fpu_do_pop(emu);
            break;
@@ -35,7 +35,7 @@
    case 0x2D:  /* CVTSD2SI Gd, Ex */
        nextop = F8;
        GET_EX;
-        switch(emu->round) {
+        switch((emu->mxcsr>>13)&3) {
            case ROUND_Nearest:
                GD.sdword[0] = floor(EX->d[0]+0.5);
                break;
@@ -66,7 +66,6 @@
        GET_EX;
        GX.d[0] *= EX->d[0];
        break;
-
    case 0x5A:  /* CVTSD2SS Gx, Ex */
        nextop = F8;
        GET_EX;
@@ -125,6 +124,93 @@
        }
        break;

+    #define GOCOND(BASE, PREFIX, CONDITIONAL)   \
+    case BASE+0:                                \
+        PREFIX                                  \
+        if(ACCESS_FLAG(F_OF))                   \
+            CONDITIONAL                         \
+        break;                                  \
+    case BASE+1:                                \
+        PREFIX                                  \
+        if(!ACCESS_FLAG(F_OF))                  \
+            CONDITIONAL                         \
+        break;                                  \
+    case BASE+2:                                \
+        PREFIX                                  \
+        if(ACCESS_FLAG(F_CF))                   \
+            CONDITIONAL                         \
+        break;                                  \
+    case BASE+3:                                \
+        PREFIX                                  \
+        if(!ACCESS_FLAG(F_CF))                  \
+            CONDITIONAL                         \
+        break;                                  \
+    case BASE+4:                                \
+        PREFIX                                  \
+        if(ACCESS_FLAG(F_ZF))                   \
+            CONDITIONAL                         \
+        break;                                  \
+    case BASE+5:                                \
+        PREFIX                                  \
+        if(!ACCESS_FLAG(F_ZF))                  \
+            CONDITIONAL                         \
+        break;                                  \
+    case BASE+6:                                \
+        PREFIX                                  \
+        if((ACCESS_FLAG(F_ZF) || ACCESS_FLAG(F_CF)))  \
+            CONDITIONAL                         \
+        break;                                  \
+    case BASE+7:                                \
+        PREFIX                                  \
+        if(!(ACCESS_FLAG(F_ZF) || ACCESS_FLAG(F_CF))) \
+            CONDITIONAL                         \
+        break;                                  \
+    case BASE+8:                                \
+        PREFIX                                  \
+        if(ACCESS_FLAG(F_SF))                   \
+            CONDITIONAL                         \
+        break;                                  \
+    case BASE+9:                                \
+        PREFIX                                  \
+        if(!ACCESS_FLAG(F_SF))                  \
+            CONDITIONAL                         \
+        break;                                  \
+    case BASE+0xA:                              \
+        PREFIX                                  \
+        if(ACCESS_FLAG(F_PF))                   \
+            CONDITIONAL                         \
+        break;                                  \
+    case BASE+0xB:                              \
+        PREFIX                                  \
+        if(!ACCESS_FLAG(F_PF))                  \
+            CONDITIONAL                         \
+        break;                                  \
+    case BASE+0xC:                              \
+        PREFIX                                  \
+        if(ACCESS_FLAG(F_SF) != ACCESS_FLAG(F_OF))  \
+            CONDITIONAL                         \
+        break;                                  \
+    case BASE+0xD:                              \
+        PREFIX                                  \
+        if(ACCESS_FLAG(F_SF) == ACCESS_FLAG(F_OF)) \
+            CONDITIONAL                         \
+        break;                                  \
+    case BASE+0xE:                              \
+        PREFIX                                  \
+        if(ACCESS_FLAG(F_ZF) || (ACCESS_FLAG(F_SF) != ACCESS_FLAG(F_OF))) \
+            CONDITIONAL                         \
+        break;                                  \
+    case BASE+0xF:                              \
+        PREFIX                                  \
+        if(!ACCESS_FLAG(F_ZF) && (ACCESS_FLAG(F_SF) == ACCESS_FLAG(F_OF))) \
+            CONDITIONAL                         \
+        break;
+        
+    GOCOND(0x80
+        , tmp32s = F32S; CHECK_FLAGS(emu);
+        , ip += tmp32s;
+    )                               /* 0x80 -> 0x8F Jxx */
+        
    case 0xC2:  /* CMPSD Gx, Ex, Ib */
        nextop = F8;
        GET_EX;
@@ -132,8 +218,8 @@
        tmp8s = 0;
        switch(tmp8u&7) {
            case 0: tmp8s=(GX.d[0] == EX->d[0]); break;
-            case 1: tmp8s=isless(GX.d[0], EX->d[0]); break;
-            case 2: tmp8s=islessequal(GX.d[0], EX->d[0]); break;
+            case 1: tmp8s=isless(GX.d[0], EX->d[0]) && !(isnan(GX.d[0]) || isnan(EX->d[0])); break;
+            case 2: tmp8s=islessequal(GX.d[0], EX->d[0]) && !(isnan(GX.d[0]) || isnan(EX->d[0])); break;
            case 3: tmp8s=isnan(GX.d[0]) || isnan(EX->d[0]); break;
            case 4: tmp8s=(GX.d[0] != EX->d[0]); break;
            case 5: tmp8s=isnan(GX.d[0]) || isnan(EX->d[0]) || isgreaterequal(GX.d[0], EX->d[0]); break;
@@ -143,6 +229,15 @@
        GX.q[0]=(tmp8s)?0xffffffffffffffffLL:0LL;
        break;

+    case 0xD0:  /* ADDSUBPS Gx, Ex */
+        nextop = F8;
+        GET_EX;
+        GX.f[0] -= EX->f[0];
+        GX.f[1] += EX->f[1];
+        GX.f[2] -= EX->f[2];
+        GX.f[3] += EX->f[3];
+        break;
+
    case 0xD6:  /* MOVDQ2Q Gm, Ex */
        nextop = F8;
        GET_EX;
@@ -152,11 +247,33 @@
    case 0xE6:  /* CVTPD2DQ Gx, Ex */
        nextop = F8;
        GET_EX;
-        GX.sd[0] = EX->d[0];
-        GX.sd[1] = EX->d[1];
+        switch((emu->mxcsr>>13)&3) {
+            case ROUND_Nearest:
+                GX.sd[0] = floor(EX->d[0]+0.5);
+                GX.sd[1] = floor(EX->d[1]+0.5);
+                break;
+            case ROUND_Down:
+                GX.sd[0] = floor(EX->d[0]);
+                GX.sd[1] = floor(EX->d[1]);
+                break;
+            case ROUND_Up:
+                GX.sd[0] = ceil(EX->d[0]);
+                GX.sd[1] = ceil(EX->d[1]);
+                break;
+            case ROUND_Chop:
+                GX.sd[0] = EX->d[0];
+                GX.sd[1] = EX->d[1];
+                break;
+        }
        GX.q[1] = 0;
        break;

+    case 0xF0:  /* LDDQU Gx, Ex */
+        nextop = F8;
+        GET_EX;
+        memcpy(&GX, EX, 16);
+        break;
+
    default:
        goto _default;
    }
@@ -15,6 +15,19 @@
        GET_EX;
        EX->ud[0] = GX.ud[0];
        break;
+    case 0x12:  /* MOVSLDUP Gx, Ex */
+        nextop = F8;
+        GET_EX;
+        GX.ud[1] = GX.ud[0] = EX->ud[0];
+        GX.ud[3] = GX.ud[2] = EX->ud[2];
+        break;
+    
+    case 0x16:  /* MOVSHDUP Gx, Ex */
+        nextop = F8;
+        GET_EX;
+        GX.ud[1] = GX.ud[0] = EX->ud[1];
+        GX.ud[3] = GX.ud[2] = EX->ud[3];
+        break;
    
    case 0x1E:  /* NOP / ENDBR32 / ENDBR64 */
        nextop = F8;
@@ -39,7 +52,7 @@
    case 0x2D:  /* CVTSS2SI Gd, Ex */
        nextop = F8;
        GET_EX;
-        switch(emu->round) {
+        switch((emu->mxcsr>>13)&3) {
            case ROUND_Nearest:
                GD.sdword[0] = floor(EX->f[0]+0.5);
                break;
@@ -169,6 +182,24 @@
        }
        break;

+    case 0xBD:                      /* LZCNT Ed,Gd */
+        CHECK_FLAGS(emu);
+        nextop = F8;
+        GET_ED;
+        tmp32u = ED->dword[0];
+        if(tmp32u) {
+            tmp8u = 0;
+            while(!(tmp32u&(1<<(31-tmp8u)))) ++tmp8u;
+            GD.dword[0] = tmp8u;
+            CONDITIONAL_SET_FLAG(tmp8u==0, F_ZF);
+            CLEAR_FLAG(F_CF);
+        } else {
+            SET_FLAG(F_CF);
+            CLEAR_FLAG(F_ZF);
+            GD.dword[0] = 32;
+        }
+        break;
+
    case 0xC2:  /* CMPSS Gx, Ex, Ib */
        nextop = F8;
        GET_EX;
@@ -176,8 +207,8 @@
        tmp8s = 0;
        switch(tmp8u&7) {
            case 0: tmp8s=(GX.f[0] == EX->f[0]); break;
-            case 1: tmp8s=isless(GX.f[0], EX->f[0]); break;
-            case 2: tmp8s=islessequal(GX.f[0], EX->f[0]); break;
+            case 1: tmp8s=isless(GX.f[0], EX->f[0]) && !(isnan(GX.f[0]) || isnan(EX->f[0])); break;
+            case 2: tmp8s=islessequal(GX.f[0], EX->f[0]) && !(isnan(GX.f[0]) || isnan(EX->f[0])); break;
            case 3: tmp8s=isnan(GX.f[0]) || isnan(EX->f[0]); break;
            case 4: tmp8s=(GX.f[0] != EX->f[0]); break;
            case 5: tmp8s=isnan(GX.f[0]) || isnan(EX->f[0]) || isgreaterequal(GX.f[0], EX->f[0]); break;