Hotfix for steam not launching when libudev1 is wrapped

(overwritten warning fix)

.github/FUNDING.yml

@@ -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

.github/workflows/manual.yml

@@ -0,0 +1,100 @@
+### Manual build Box86 with Github Action
+name: Custom build Box86
+
+on:
+  workflow_dispatch:
+    inputs:
+      platform:
+        description: 'Target platform name'
+        required: true
+        default: 'GENERIC_ARM'
+        type: choice
+        options:
+        - I386
+        - GENERIC_ARM
+        - A64
+        - GAMESHELL
+        - ODROID
+        - PANDORA
+        - PHYTIUM
+        - PYRA
+        - RK3288
+        - RK3399
+        - RPI4
+        - RPI4ARM64
+        - SD845
+      build_type:
+        description: 'Build type'
+        required: true
+        default: 'RelWithDebInfo'
+        type: choice
+        options:
+        - RelWithDebInfo
+        - Release
+        - Debug
+        - MinSizeRel
+      dynarec:
+        description: 'Enable dynarec'
+        required: false
+        default: 'true'
+        type: boolean
+      trace:
+        description: 'Enable trace'
+        required: false
+        default: 'false'
+        type: boolean
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - name: "Checkout Box86 Repository"
+        uses: actions/checkout@v2
+
+      - name: "Environment preparation"
+        run: |
+          sudo apt-get update
+
+          if [[ ${{ github.event.inputs.platform }} != 'I386' && ${{ github.event.inputs.platform }} != 'GENERIC_ARM' ]]; then
+            echo "BOX86_PLATFORM_MARCRO=-D${{ github.event.inputs.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 [[ ${{ github.event.inputs.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
+            else
+              echo "BOX86_PLATFORM_MARCRO=-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
+
+      - name: "Display Build info"
+        run: |
+          echo "CMake Platform Macro: ${{ env.BOX86_PLATFORM_MARCRO }}"
+          echo "CMake C Compiler: ${{ env.BOX86_COMPILER }}"
+          echo "Build type: ${{ github.event.inputs.build_type }}"
+          echo "Dynarec Enabled: ${{ github.event.inputs.dynarec }}"
+          echo "Trace Enabled: ${{ github.event.inputs.trace }}"
+
+      - name: "Build Box86"
+        run: |
+          mkdir build
+          cd build
+          cmake .. -DCMAKE_C_COMPILER=${{ env.BOX86_COMPILER }} ${{ env.BOX86_PLATFORM_MARCRO }}\
+                   -DCMAKE_BUILD_TYPE=${{ github.event.inputs.build_type }}\
+                   -DARM_DYNAREC:BOOL=${{ github.event.inputs.dynarec }}\
+                   -DHAVE_TRACE:BOOL=${{ github.event.inputs.trace }}\
+                   -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON
+          make -j$(nproc) VERBOSE=1
+
+      - name: "Upload Artifact"
+        uses: actions/upload-artifact@v2
+        with:
+          name: box86-${{ github.event.inputs.platform }}-${{ github.event.inputs.build_type }}
+          path: build/box86
+
+
+

.github/workflows/release.yml

@@ -0,0 +1,120 @@
+### 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"
+      - "CMakeLists.txt"
+      - "**/*.yml"
+  pull_request:
+    types: [assigned, opened, synchronize, reopened]
+    paths:
+      - "**/*.c"
+      - "**/*.h"
+      - "**/*.S"
+      - "**/*.py"
+      - "CMakeLists.txt"
+      - "**/*.yml"
+
+jobs:
+  build:
+    strategy:
+      fail-fast: false
+      matrix:
+        platform: [I386, OTHER_ARM, PANDORA, PYRA, GAMESHELL, RPI4, RPI4ARM64, ODROID, RK3399, RK3288, A64, SD845, PHYTIUM, ANDROID]
+        type: [Release, Trace]
+        os: [ubuntu-latest]
+        
+        include:
+          - platform: TEGRAX1
+            type: Release
+            os: ubuntu-20.04
+          - platform: TEGRAX1
+            type: Trace
+            os: ubuntu-20.04
+          - platform: TEGRAX1ARM32
+            type: Release
+            os: ubuntu-20.04
+          - platform: TEGRAX1ARM32
+            type: Trace
+            os: ubuntu-20.04
+
+    runs-on: ${{ matrix.os }}
+    steps:
+      - name: "Checkout Box86 Repository"
+        uses: actions/checkout@v3
+
+      - name: "Environment preparation"
+        run: |
+          sudo apt-get update
+
+          if [[ ${{ matrix.platform }} != 'I386' && ${{ matrix.platform }} != 'OTHER_ARM' ]]; then
+            sudo apt-get -y install git cmake make python3
+            if [[ ${{ matrix.platform }} == 'ANDROID' ]]; then
+              wget https://dl.google.com/android/repository/android-ndk-r23b-linux.zip
+              unzip android-ndk-r23b-linux.zip
+              echo "BOX86_COMPILER=$PWD/android-ndk-r23b/toolchains/llvm/prebuilt/linux-x86_64/bin/armv7a-linux-androideabi31-clang" >> $GITHUB_ENV
+              echo "BOX86_PLATFORM_MARCRO=-DANDROID=1 -DNOLOADADDR=1" >> $GITHUB_ENV
+            else
+              sudo apt-get -y install gcc-arm-linux-gnueabihf
+              echo "BOX86_COMPILER=arm-linux-gnueabihf-gcc" >> $GITHUB_ENV
+              echo "BOX86_PLATFORM_MARCRO=-D${{ matrix.platform }}=1" >> $GITHUB_ENV
+            fi
+          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
+            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: "Test Box86"
+        if: ${{ matrix.platform == 'I386' }}
+        run: |
+          cd build
+          ctest -j$(nproc)
+
+      - name: "Upload Artifact"
+        uses: actions/upload-artifact@v3
+        with:
+          name: box86-${{ matrix.platform }}-${{ matrix.type }}
+          path: build/box86
+
+
+

.gitignore

@@ -3,3 +3,7 @@ build*/
 .vscode
 .gdb_history
 src/git_head.h
+backup/
+box86_update.sh
+restore_backup.sh
+.cache

.travis.yml

@@ -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

CHANGELOG.md

@@ -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
-

CMakeLists.txt

@@ -1,25 +1,53 @@
-cmake_minimum_required(VERSION 2.6)
-
-project(box86)
-
-enable_testing()
-enable_language(C)
+cmake_minimum_required(VERSION 3.10)
 
 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(RPI3ARM64 "Set to ON if targeting an RaspberryPI3 device with multiarch arm64 and armhf" ${RPI3ARM64})
+option(RPI4ARM64 "Set to ON if targeting an RaspberryPI4 device with multiarch arm64 and armhf" ${RPI4ARM64})
+option(GOA_CLONE "Set to ON if targeting GO Advance clones, like RG351p/v, Gameforce Chi, RGB10..." ${GOA_CLONE})
+option(RK3288 "Set to ON if targeting an Rockchip RK3288 based device" ${RK3288})
+option(RK3399 "Set to ON if targeting an Rockchip RK3399 based device" ${RK3399})
+option(RK3588 "Set to ON if targeting an Rockchip RK3588 / RK3588S based device" ${RK3588})
+option(ODROIDN2 "Set to ON if targeting an Odroid-N2 device" ${ODROIDN2})
+option(TEGRAX1 "Set to ON if targeting an Tegra X1 (ARM64) based device" ${TEGRAX1})
+option(TEGRAX1ARM32 "Set to ON if targeting an Tegra X1 (ARM32) based device" ${TEGRAX1ARM32})
+option(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(ARM64 "Set to ON if targeting a generic ARM64 based device" ${ARM64})
+option(BAD_SIGNAL "Set to ON to activate the workaround for incoherent si_info on SIGSEGV" ${BAD_SIGNAL})
+option(LX2160A "Set to ON if targeting an LX2160A based device" ${LX2160A})
+option(ANDROID "Set to ON if targeting an Android device" ${ANDROID})
 option(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})
+option(NO_LIB_INSTALL "Set ON to not install a few x86 libs that are used by many program" ${NO_LIB_INSTALL})
+option(NO_CONF_INSTALL "Set ON to not install config files" ${NO_CONF_INSTALL})
+
+if(PANDORA OR PYRA OR RPI2 OR RPI3 OR RPI4 OR GAMESHELL OR ODROID OR GOA_CLONE OR RK3288 OR RK3399 OR ODROIDN2 OR TEGRAX1 OR TEGRAX1ARM32 OR RPI3ARM64 OR RPI4ARM64 OR PHYTIUM OR SD845 OR A64 OR LX2160A OR ARM64 OR RK3588)
+    set(LD80BITS OFF CACHE BOOL "")
+    set(NOALIGN OFF CACHE BOOL "")
+    set(ARM_DYNAREC ON CACHE BOOL "")
+    if(RK3588)
+        set(BAD_SIGNAL ON CACHE BOOL "")
+    endif()
+endif()
+if(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 +65,22 @@ else()
     set(PYTHON_EXECUTABLE ${Python3_EXECUTABLE} CACHE INTERNAL "The Python3 executable" FORCE)
 endif()
 
+cmake_host_system_information(RESULT CMAKE_HOST_SYSTEM_PROCESSOR QUERY OS_PLATFORM)
+string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "i686"  _x86)
+string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "x86_64"  _x86_64)
+string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "aarch64"  _aarch64)
+string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "armhf"  _armhf)
+
+if(_aarch64 AND (RK3399 OR ODROIDN2 OR TEGRAX1 OR RPI3ARM64 OR RPI4ARM64 OR PHYTIUM OR SD845 OR A64 OR LX2160A OR RK3588 OR ARM64))
+    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 +88,125 @@ 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)
+    set(CMAKE_ASM_FLAGS  "-marm -mcpu=cortex-a8 -mfpu=neon -mfloat-abi=softfp")
+elseif(GAMESHELL)
+    add_definitions(-DGAMESHELL)
     add_definitions(-marm -mcpu=cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard)
     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(RPI3ARM64)
+    add_definitions(-DRPI)
+    add_definitions(-DRPI3ARM64)
+    add_definitions(-marm -march=armv8-a+crc -mtune=cortex-a53 -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-marm -march=armv8-a+crc -mtune=cortex-a53 -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(RPI4ARM64)
+    add_definitions(-DRPI)
+    add_definitions(-DRPI4ARM64)
+    add_definitions(-marm -march=armv8-a+crc -mtune=cortex-a72 -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-marm -march=armv8-a+crc -mtune=cortex-a72 -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(ODROID)
+    add_definitions(-DODROID)
     add_definitions(-mcpu=cortex-a9 -mfpu=neon-vfpv4 -mfloat-abi=hard -marm)
     set(CMAKE_ASM_FLAGS  "-marm -mcpu=cortex-a9 -mfpu=neon-vfpv4 -mfloat-abi=hard")
-    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(RK3588)
+    add_definitions(-DRK3588)
+    add_definitions(-pipe -mcpu=cortex-a76.cortex-a55+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    add_definitions(-DBAD_SIGNAL)
+    set(CMAKE_ASM_FLAGS  "-pipe -mcpu=cortex-a76.cortex-a55+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(ARM64)
+    add_definitions(-DARM64)
+    add_definitions(-pipe -march=armv8-a+crc+simd+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crc+simd+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(ODROIDN2)
+    add_definitions(-DODROIDN2)
+    add_definitions(-pipe -march=armv8-a+crc+simd+crypto -mcpu=cortex-a73.cortex-a53+crypto -mfpu=crypto-neon-fp-armv8 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crc+simd+crypto -mcpu=cortex-a73.cortex-a53+crypto -mfpu=crypto-neon-fp-armv8 -mfloat-abi=hard")
+elseif(TEGRAX1)
+    add_definitions(-DTEGRAX1)
+    add_definitions(-marm -pipe -march=armv8-a+crc -mcpu=cortex-a57 -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-marm -pipe -march=armv8-a+crc -mcpu=cortex-a57 -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(TEGRAX1ARM32)
+    add_definitions(-DTEGRAX1ARM32)
+    add_definitions(-marm -pipe -march=armv8-a+crc -mcpu=cortex-a57 -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-marm -pipe -march=armv8-a+crc -mcpu=cortex-a57 -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(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(LX2160A)
+    add_definitions(-DLX2160A)
+    add_definitions(-pipe -march=armv8-a+crypto+crc -mcpu=cortex-a72+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crypto+crc -mcpu=cortex-a72+crypto -mfpu=neon-fp-armv8 -mfloat-abi=hard")
+elseif(ANDROID)
+    add_definitions(-DANDROID)
+elseif(ARM_DYNAREC)
+    #if DYNAREC is selected alone, without any arch
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv7-a+simd -mfpu=neon")
 endif()
 
-set(BOX86_ELF_ADDRESS "0x02800000")
+if(BAD_SIGNAL)
+    add_definitions(-DBAD_SIGNAL)
+endif()
+
+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)
@@ -114,16 +214,13 @@ endif()
 
 if(NOALIGN)
     add_definitions(-DNOALIGN)
+    add_definitions(-pipe -march=core2 -msse2)
 endif()
 
 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 +228,14 @@ 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 +250,11 @@ 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 +270,18 @@ 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/mallochook.c"
     "${BOX86_ROOT}/src/tools/bridge.c"
     "${BOX86_ROOT}/src/tools/pathcoll.c"
     "${BOX86_ROOT}/src/tools/fileutils.c"
@@ -184,15 +289,17 @@ set(ELFLOADER_SRC
     "${BOX86_ROOT}/src/tools/box86stack.c"
     "${BOX86_ROOT}/src/tools/my_cpuid.c"
     "${BOX86_ROOT}/src/tools/gtkclass.c"
+    "${BOX86_ROOT}/src/tools/rcfile.c"
     "${BOX86_ROOT}/src/tools/wine_tools.c"
     "${BOX86_ROOT}/src/elfs/elfloader.c"
     "${BOX86_ROOT}/src/elfs/elfparser.c"
     "${BOX86_ROOT}/src/elfs/elfload_dump.c"
     "${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/x86run_private.c"
     "${BOX86_ROOT}/src/emu/x86syscall.c"
     "${BOX86_ROOT}/src/emu/x86primop.c"
@@ -200,19 +307,48 @@ set(ELFLOADER_SRC
     "${BOX86_ROOT}/src/emu/x86int3.c"
     "${BOX86_ROOT}/src/emu/x86tls.c"
     "${BOX86_ROOT}/src/emu/x87emu_private.c"
-    "${BOX86_ROOT}/src/emu/x87run.c"
     "${BOX86_ROOT}/src/libtools/auxval.c"
     "${BOX86_ROOT}/src/libtools/myalign.c"
     "${BOX86_ROOT}/src/libtools/myalign64.c"
     "${BOX86_ROOT}/src/libtools/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 "${BOX86_ROOT}/src/libtools/obstack.c")
+endif()
+
+set(INTERPRETER
+    "${BOX86_ROOT}/src/emu/x86run.c"
+    "${BOX86_ROOT}/src/emu/x86run0f.c"
+    "${BOX86_ROOT}/src/emu/x86run64.c"
+    "${BOX86_ROOT}/src/emu/x86run640f.c"
+    "${BOX86_ROOT}/src/emu/x86run6466.c"
+    "${BOX86_ROOT}/src/emu/x86run6467.c"
+    "${BOX86_ROOT}/src/emu/x86run66.c"
+    "${BOX86_ROOT}/src/emu/x86run660f.c"
+    "${BOX86_ROOT}/src/emu/x86run66d9.c"
+    "${BOX86_ROOT}/src/emu/x86run66dd.c"
+    "${BOX86_ROOT}/src/emu/x86run67.c"
+    "${BOX86_ROOT}/src/emu/x86run6766.c"
+    "${BOX86_ROOT}/src/emu/x86rund8.c"
+    "${BOX86_ROOT}/src/emu/x86rund9.c"
+    "${BOX86_ROOT}/src/emu/x86runda.c"
+    "${BOX86_ROOT}/src/emu/x86rundb.c"
+    "${BOX86_ROOT}/src/emu/x86rundc.c"
+    "${BOX86_ROOT}/src/emu/x86rundd.c"
+    "${BOX86_ROOT}/src/emu/x86runde.c"
+    "${BOX86_ROOT}/src/emu/x86rundf.c"
+    "${BOX86_ROOT}/src/emu/x86runf0.c"
+    "${BOX86_ROOT}/src/emu/x86runf066.c"
+    "${BOX86_ROOT}/src/emu/x86runf20f.c"
+    "${BOX86_ROOT}/src/emu/x86runf30f.c"
+)
 
 set(WRAPPEDS
     "${BOX86_ROOT}/src/wrapped/wrappedlibc.c"
@@ -226,11 +362,15 @@ set(WRAPPEDS
     "${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"
@@ -240,11 +380,13 @@ set(WRAPPEDS
     "${BOX86_ROOT}/src/wrapped/wrappedpng16.c"
     "${BOX86_ROOT}/src/wrapped/wrappedlibxxf86vm.c"
     "${BOX86_ROOT}/src/wrapped/wrappedlibxrandr.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibxpresent.c"
     "${BOX86_ROOT}/src/wrapped/wrappedlibxext.c"
     "${BOX86_ROOT}/src/wrapped/wrappedxinerama.c"
     "${BOX86_ROOT}/src/wrapped/wrappedlibxfixes.c"
     "${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"
     "${BOX86_ROOT}/src/wrapped/wrappedlibxpm.c"
@@ -252,6 +394,16 @@ set(WRAPPEDS
     "${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"
     "${BOX86_ROOT}/src/wrapped/wrappedlibxcomposite.c"
@@ -259,8 +411,8 @@ set(WRAPPEDS
     "${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 +421,7 @@ 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"
     "${BOX86_ROOT}/src/wrapped/wrappedlibresolv.c"
@@ -278,6 +431,7 @@ set(WRAPPEDS
     "${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"
@@ -285,24 +439,41 @@ 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"
+    "${BOX86_ROOT}/src/wrapped/wrappedgmp.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedgcrypt.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"
+    "${BOX86_ROOT}/src/wrapped/wrappedgstreamer.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedgstaudio.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedgstvideo.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedgsttag.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedgstbase.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedgstapp.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibibus.c"
     "${BOX86_ROOT}/src/wrapped/wrappeddbusmenugtk.c"
     "${BOX86_ROOT}/src/wrapped/wrappeddbusmenuglib.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedappindicator.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedappindicator3.c"
     "${BOX86_ROOT}/src/wrapped/wrappedatk.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedatkbridge.c"
     "${BOX86_ROOT}/src/wrapped/wrappedpangoft2.c"
     "${BOX86_ROOT}/src/wrapped/wrappedpangocairo.c"
     "${BOX86_ROOT}/src/wrapped/wrappedgthread2.c"
     "${BOX86_ROOT}/src/wrapped/wrappedcairo.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedcairogobject.c"
     "${BOX86_ROOT}/src/wrapped/wrappedpango.c"
     "${BOX86_ROOT}/src/wrapped/wrappeddbusglib1.c"
     "${BOX86_ROOT}/src/wrapped/wrappedgudev1.c"
@@ -310,12 +481,63 @@ set(WRAPPEDS
     "${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/wrappedsoftokn3.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedfreebl3.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedsecret1.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedtbbmalloc.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedmimalloc.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedselinux.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibharfbuzz.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibcairogobject.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibvkd3d.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedgomp.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedcap.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedpcap.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedfaudio.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedtbbmallocproxy.c"
+    "${BOX86_ROOT}/src/wrapped/wrappedlibfuse.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 ${B})
 endforeach()
@@ -324,7 +546,10 @@ set(WRAPPER "${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 ${WRAPPER}
@@ -341,21 +566,20 @@ add_custom_command(
 add_custom_target(WRAPPERS DEPENDS "${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"
 
         "${BOX86_ROOT}/src/dynarec/dynarec_arm.c"
         "${BOX86_ROOT}/src/dynarec/dynarec_arm_functions.c"
+        "${BOX86_ROOT}/src/dynarec/dynarec_arm_jmpnext.c"
         "${BOX86_ROOT}/src/dynarec/arm_printer.c"
+        "${BOX86_ROOT}/src/emu/x86test.c"
 
         "${BOX86_ROOT}/src/dynarec/arm_prolog.S"
         "${BOX86_ROOT}/src/dynarec/arm_epilog.S"
-        "${BOX86_ROOT}/src/dynarec/arm_linker.S"
-        "${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
@@ -381,6 +605,7 @@ if(ARM_DYNAREC)
     "${BOX86_ROOT}/src/dynarec/dynarec_arm_df.c"
     "${BOX86_ROOT}/src/dynarec/dynarec_arm_f0.c"
     "${BOX86_ROOT}/src/dynarec/dynarec_arm_660f.c"
+    "${BOX86_ROOT}/src/dynarec/dynarec_arm_66f0.c"
     "${BOX86_ROOT}/src/dynarec/dynarec_arm_f20f.c"
     "${BOX86_ROOT}/src/dynarec/dynarec_arm_f30f.c"
     )
@@ -397,6 +622,8 @@ if(ARM_DYNAREC)
     set_target_properties(arm_pass2 PROPERTIES COMPILE_FLAGS "-DSTEP=2")
     add_library(arm_pass3 OBJECT ${DYNAREC_PASS})
     set_target_properties(arm_pass3 PROPERTIES COMPILE_FLAGS "-DSTEP=3")
+    add_library(test_interpreter OBJECT ${INTERPRETER})
+    set_target_properties(test_interpreter PROPERTIES COMPILE_FLAGS "-DTEST_INTERPRETER")
     add_dependencies(arm_pass0 WRAPPERS)
     add_dependencies(arm_pass1 WRAPPERS)
     add_dependencies(arm_pass2 WRAPPERS)
@@ -408,43 +635,68 @@ if(ARM_DYNAREC)
         $<TARGET_OBJECTS:arm_pass1>
         $<TARGET_OBJECTS:arm_pass2>
         $<TARGET_OBJECTS:arm_pass3>
+        $<TARGET_OBJECTS:test_interpreter>
     )
+
 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} ${INTERPRETER} ${WRAPPEDS}
+        VERBATIM)
+else()
+    add_custom_command(
+        OUTPUT "${BOX86_ROOT}/src/git_head.h"
+        COMMAND bash -c "echo \\\#define GITREV \\\"$(git rev-parse --short HEAD)\\\">\"${BOX86_ROOT}/src/git_head.h\""
+        DEPENDS ${ELFLOADER_SRC} ${INTERPRETER} ${WRAPPEDS}
+        VERBATIM)
+endif()
+
+add_library(interpreter OBJECT ${INTERPRETER})
+
 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(${BOX86} dynarec interpreter m dl)
+    else()
+        target_link_libraries(${BOX86} dynarec interpreter m dl rt pthread resolv)
+    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} c m dl)
+else()
+    #target_link_libraries(${BOX86} c m dl rt pthread resolv)
+    set_target_properties(${BOX86} PROPERTIES LINK_FLAGS "-Wl,--no-as-needed -lc -lm -ldl -lrt -lpthread -lresolv -Wl,--as-needed -pthread")
+endif()
 if(ARM_DYNAREC)
-    target_link_libraries(${BOX86} dynarec)
+    target_link_libraries(${BOX86} dynarec interpreter)
+else()
+    target_link_libraries(${BOX86} interpreter)
 endif()
 
 if(${CMAKE_VERSION} VERSION_LESS "3.13")
@@ -460,16 +712,20 @@ else()
     endif()
 endif()
 
-string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "i686"  _x86)
-string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "x86_64"  _x86_64)
-
 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/)
-  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/)
+  if(NOT NO_CONF_INSTALL)
+    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}/system/box86.box86rc DESTINATION /etc/)
+  endif()
+  if(NOT NO_LIB_INSTALL)
+    install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libstdc++.so.6 DESTINATION /usr/lib/i386-linux-gnu/)
+    install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libstdc++.so.5 DESTINATION /usr/lib/i386-linux-gnu/)
+    install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libgcc_s.so.1 DESTINATION /usr/lib/i386-linux-gnu/)
+    install(FILES ${CMAKE_SOURCE_DIR}/x86lib/libpng12.so.0 DESTINATION /usr/lib/i386-linux-gnu/)
+  endif()
 endif()
 
 if(NOT TARGET uninstall)
@@ -482,83 +738,154 @@ 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}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test01 -D TEST_OUTPUT=tmpfile.txt 
+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=tmpfile01.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}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test02 -D TEST_OUTPUT=tmpfile.txt 
+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=tmpfile02.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}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test03 -D TEST_OUTPUT=tmpfile.txt 
+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=tmpfile03.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}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test04 -D TEST_ARGS2=yeah -D TEST_OUTPUT=tmpfile.txt 
+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=tmpfile04.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}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test05 -D TEST_ARGS2=7 -D TEST_OUTPUT=tmpfile.txt 
+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=tmpfile05.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}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test06 -D TEST_OUTPUT=tmpfile.txt 
+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=tmpfile06.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}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test07 -D TEST_OUTPUT=tmpfile.txt 
+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=tmpfile07.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}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test08 -D TEST_OUTPUT=tmpfile.txt 
+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=tmpfile08.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}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test09 -D TEST_OUTPUT=tmpfile.txt 
+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=tmpfile09.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}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test10 -D TEST_OUTPUT=tmpfile.txt 
+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=tmpfile10.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}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test11 -D TEST_OUTPUT=tmpfile.txt 
+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=tmpfile11.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}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test12 -D TEST_OUTPUT=tmpfile.txt 
+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=tmpfile12.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}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test13 -D TEST_OUTPUT=tmpfile.txt 
+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=tmpfile13.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}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test14 -D TEST_OUTPUT=tmpfile.txt 
-    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref14.txt
+if(NOT LD80BITS)
+    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=tmpfile14.txt 
+        -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref14.txt
+        -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+endif()
+
+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=tmpfile15.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=tmpfile16.txt 
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref16.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(NAME sse COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test17 -D TEST_OUTPUT=tmpfile17.txt 
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref17.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+set_tests_properties(sse PROPERTIES ENVIRONMENT "BOX86_DYNAREC_FASTNAN=0;BOX86_DYNAREC_FASTROUND=0")
+
+add_test(NAME longjumpInSignals COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test18 -D TEST_OUTPUT=tmpfile18.txt 
+    -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=tmpfile19.txt 
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref19.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(NAME idiv COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test20 -D TEST_OUTPUT=tmpfile20.txt 
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref20.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
 file(GLOB extension_tests "${CMAKE_SOURCE_DIR}/tests/extensions/*.c")
 foreach(file ${extension_tests})
     get_filename_component(testname "${file}" NAME_WE)
     add_test(NAME "${testname}" COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86}
-        -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/extensions/${testname} -D TEST_OUTPUT=tmpfile.txt
+        -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/extensions/${testname} -D TEST_OUTPUT=tmpfile-${testname}.txt
         -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/extensions/${testname}.txt
         -P ${CMAKE_SOURCE_DIR}/runTest.cmake)
 endforeach()
 
+add_test(NAME sse_optimized COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test17_o2 -D TEST_OUTPUT=tmpfile17_o2.txt 
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref17_o2.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+set_tests_properties(sse_optimized PROPERTIES ENVIRONMENT "BOX86_DYNAREC_FASTNAN=0;BOX86_DYNAREC_FASTROUND=0")
+
 endif(BOX86LIB)

COMPILE.md

@@ -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.
-

LICENSE

@@ -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

README.md

@@ -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)

USAGE.md

@@ -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.
-

_config.yml

@@ -0,0 +1 @@
+theme: jekyll-theme-hacker

debian/changelog

@@ -0,0 +1,84 @@
+box86 (0.3.0) unstable; urgency=low
+
+ * Added rcfile handling: box86 now read "/etc/box86.box86rc" and "~/.box86rc" and apply per process parameters
+ * Added BOX86_DYNAREC_FORWARD to allow bigger Dynarec block to be build
+ * Refactored Strong Memory Model emulation (faster and more accurate)
+ * Box86 now return a custom name for the CPU instal of P4, build with Box86 and the cpu name it's running on
+ * Improved memory protection tracking and signal handling, again
+ * More wrapped libs (especially on the gtk familly)
+ * More wrapped functions
+ * Added support for CreateDeviceEx in d3dadapter9, for better gallium nine support
+ * Improved dladdr/dladdr1 wrapped function
+ * Some fixes to flags handling in the Dynarec on some special cases
+ * Added specific suppport for RK3588 board (needed some hack in signal handling, probably due to the use of a non mainlined kernel)
+ * Lots of improvment in the Rounding of float & double (Dynarec and Interpretor), and proper handling of denormals when converting to/from 80 bits long double
+ * Added specific suppport for RK3588 board (needed some hack in signal handling, probably due to the use of a non mainlined kernel)
+
+ -- Sebastien Chevalier <[email protected]>  Thu, 10 Mar 2023 10:20:33 -0500
+
+box86 (0.2.8) unstable; urgency=low
+
+  * Wrapped more Vulkan functions, dxvk2.0 now works
+  * Added support for overridden malloc/free functions
+  * Improved elf loader
+  * Improved dlsym handling with RTLD_NEXT
+  * Added BOX86_DYNAREC_SAFEFLAGS to allow finetuning Dynarec generated code handling of flags accross function calls and returns.
+  * Added BOX86_BASH to setup x86 bash binary, so shell script can be run in x86 world
+  * Added BOX86_ROLLING_LOG to have details log only on crash
+  * Improved TLS Size handling (Unity3D games now works)
+  * Improved execv family of function to have better box86 integration
+  * Added a few more wrapped libs, like FAudio
+  * Improved gtk, SDL2, GL, vulkan (and more) wrapping
+  * Improved Signal Handling and memory protection traking
+  * Added some basic support for statically linked program. Thread are not working yet, but memory allocation and file access does.
+  * add LX2160A config by @lizthegrey in #612
+  * Grammar and formatting COMPILE.md by @Radiicall in #615
+  * Update README.md by @everypizza1 in #617
+  * docs: add instructions for Debian-based Linux by @ryanfortner in #622
+  * Fix typoes in readme by @ryanfortner in #621
+  * Various improvements to install_steam.sh by @ryanfortner in #620
+  * docs: remove unnecessary newline by @ryanfortner in #624
+  * fix: the first parameter of realpath may be null by @ye-yeshun in #630
+  * Add 6 wrapped function to libdrm, and 2 to libexpat. by @cjy880927 in #632
+  * Fix #610 by @rajdakin in #634
+  * some floating point instructions by @ye-yeshun in #637
+  * Fix Typo in readme x86_64 -> x86 by @Jai-JAP in #642
+  * Fix grammar by @Jai-JAP in #645
+  * Test wine preload by @ptitSeb in #649
+  * Align address when call mprotect by @wannacu in #651
+  * Added 66 0F 7D HSUBPD opcode by @wannacu in #652
+  * add TEGRAX1 and TEGRAX1ARM32 targets on ubuntu-18.04 by @theofficialgman in #654
+  * automatically install 32 bit libraries if 64 bit OS is detected in install_steam.sh by @Moldytzu in #659
+  * mention gcc-arm-linux-gnueabihf in compile.md by @Moldytzu in #660
+  * Add libc6:armhf install to RPi4 aarch64 build info by @WheezyE in #685
+  * Updated wine install for RPiOS bullseye 32/64bit by @WheezyE in #689
+  * Add atomic add opcodes by @Trass3r in #695
+  * steam: disable modern login and pass arguments to client by @Moldytzu in #700
+  * Modify sizeof(instsize_t) to 1 byte by @liuli0217 in #721
+  * Add MOV GS:od, AL by @liuli0217 in #722
+
+ -- Sebastien Chevalier <[email protected]>  Thu, 17 Nov 2022 11:20:33 -0500
+
+box86 (0.2.6) unstable; urgency=low
+
+  * Added an option to get generated -NAN on SSE2 operation
+  * Many small fixes and improvement to get steam and SteamPlay working (along with box64)
+  * Added some workaround for program calling uname to mimic x86 system
+  * Added some more Vulkan extensions
+  * Added some granularity to mmap64, and generic memory protection tracking improvment, to mimic x86 system (for wine stagging and proton)
+  * Better clone syscall and libc implementation
+  * Improved wrapping of GTK libs
+  * More wrapped libs (more gtk libs, like gstreamer-1.0 among other non-gtk libs too)
+  * More wrapped functions(with some tricky one, like dladdr1)
+  * Some fixes and improvment to some wrapped functions
+  * Refactor Dynarec:
+  *  - better handling of SSE/x87 registers
+  *  - optimisation to limit float/double conversion
+  *  - better handling of "internal" jump inside a dynablock
+  * Improved elfloader
+  * Improved some tests
+  * A few fixes in some Dynarec opcodes
+  * Don't quit on unsupported syscall, return -1 / ENOSYS error
+  * A bit more optimisation in some Dynarec opcodes
+
+ -- Sebastien Chevalier <[email protected]>  Tue, 18 Jan 2022 11:20:33 -0500

debian/compat

@@ -0,0 +1 @@
+11

debian/control

@@ -0,0 +1,14 @@
+Source: box86
+Section: utils
+Priority: optional
+Maintainer: Ryan Fortner <[email protected]>
+Build-Depends: cmake, debhelper (>=11~)
+Standards-Version: 4.1.4
+Homepage: https://github.com/ptitSeb/box86
+
+Package: box86
+Architecture: any
+Multi-Arch: foreign
+Depends: ${misc:Depends}, ${shlibs:Depends}
+Description: Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices 
+ 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).

debian/copyright

@@ -0,0 +1,26 @@
+Format-Specification: http://dep.debian.net/deps/dep5/
+
+Files: *
+Copyright: © 2022 Sebastien Chevalier
+License: MIT
+
+License: MIT
+ 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
+ in the Software without restriction, including without limitation the rights
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ copies of the Software, and to permit persons to whom the Software is
+ furnished to do so, subject to the following conditions:
+ .
+ The above copyright notice and this permission notice shall be included in
+ all copies or substantial portions of the Software.
+ .
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ THE SOFTWARE.

debian/lintian-override

@@ -0,0 +1 @@
+box86: no-manual-page

debian/rules

@@ -0,0 +1,13 @@
+#!/usr/bin/make -f
+
+%:
+	dh $@  
+
+override_dh_auto_configure:
+	dh_auto_configure -- \
+	      -DCMAKE_BUILD_TYPE=RelWithDebInfo \
+	      -DARM_DYNAREC=1 \
+		  -DNOGIT=1
+
+override_dh_shlibdeps:
+	dh_shlibdeps --exclude=libgcc_s.so.1 --exclude=libpng12.so.0 --exclude=libstdc++.so.5 --exclude=libstdc++.so.6 

debian/source/format

@@ -0,0 +1 @@
+3.0 (quilt)

debian/upstream/metadata

@@ -0,0 +1,5 @@
+Bug-Database: https://github.com/ptitSeb/box86/issues
+Bug-Submit: https://github.com/ptitSeb/box86/issues/new
+Repository: https://github.com/ptitSeb/box86.git
+Repository-Browse: https://github.com/ptitSeb/box86
+Security-Contact: [email protected]

debian/watch

@@ -0,0 +1 @@
+version=3

docs/CHANGELOG.md

@@ -0,0 +1,225 @@
+v0.3.2
+======
+* Better syscall handling
+* Fixed some wrapped functions, added some others
+* Introduced another wrapper helper
+* Added some more opcodes
+* Added a cosimulation mode (to compare the behaviour of the dynarec against the interpreter)
+* Added support of FASTNAN for some more opcodes
+* Reduced memory footprint
+* Added BOX86_NOSANDBOX (adds an argument to the command line)
+* Added an ukrainian translation
+=> Mostly a stability update with improved elfloader, fixed opcodes and refined profiles
+
+v0.3.0
+======
+* Added rcfile handling: box86 now read `/etc/box86.box86rc` and `~/.box86rc` and apply per process parameters
+    => Every program can be finetuned, for speed or just to make them work
+* Added BOX86_DYNAREC_FORWARD to allow bigger Dynarec block to be build
+* Refactored Strong Memory Model emulation (faster and more accurate)
+* Box86 now return a custom name for the CPU instal of P4, build with Box86 and the cpu name it's running on
+* Improved memory protection tracking and signal handling, again
+* More wrapped libs (especially on the gtk familly)
+* More wrapped functions
+* Added support for CreateDeviceEx in d3dadapter9, for better gallium nine support
+* Improved dladdr/dladdr1 wrapped function
+* Some fixes to flags handling in the Dynarec on some special cases
+* Added specific suppport for RK3588 board (needed some hack in signal handling, probably due to the use of a non mainlined kernel)
+* Lots of improvment in the Rounding of float & double (Dynarec and Interpretor), and proper handling of denormals when converting to/from 80 bits long double
+* Added specific suppport for RK3588 board (needed some hack in signal handling, probably due to the use of a non mainlined kernel)
+
+
+v0.2.8
+======
+* Wrapped more vulkan function, dxvk 2.0 now works
+* Added support for overriden malloc/free functions
+* Improved elf loader
+* Improved dlsym handling with RTLD_NEXT
+* Added BOX86_DYNAREC_SAFEFLAGS to allow finetuning Dynarec generated code handling of flags accross function calls and returns.
+* Added BOX86_BASH to setup x86 bash binary, so shell script can be run in x86 world 
+* Added BOX86_ROLLING_LOG to have details log only on crash
+* Improved TLS Size handling (Unity3D games now works)
+* Improved `execv` family of function to have better box86 integration
+* Added a few more wrapped libs, like FAudio
+* Improved gtk, SDL2, GL, vulkan (and more) wrapping
+* Improved Signal Handling and memory protection traking
+* Added some basic support for statically linked program. Thread are not working yet, but memory allocation and file access does.
+=> with the use of the new SafeFlags, some C# Windows program now runs fine (it's forced when the Vara family of software is detected)
+=> With the basic support of statcaly linked program, old emulator like modeler for linux works. Program with threads like mugen for linux still doesn't
+=> The bash integration is usefull for device with no binfmt integration or for script that check the machine it's running on
+
+v0.2.6
+======
+* Added an option to get generated -NAN on SSE2 operation
+* Many small fixes and improvement to get steam and SteamPlay working (along with box64)
+* Added some workaround for program calling uname to mimic x86 system
+* Added some more Vulkan extensions
+* Added some granularity to mmap64, and generic memory protection tracking improvment, to mimic x86 system (for wine stagging and proton)
+* Better clone syscall and libc implementation
+* Improved wrapping of GTK libs
+* More wrapped libs (more gtk libs, like gstreamer-1.0 among other non-gtk libs too)
+* More wrapped functions(with some tricky one, like dladdr1)
+* Some fixes and improvment to some wrapped functions
+* Refactor Dynarec:
+    * better handling of SSE/x87 registers
+    * optimisation to limit float/double conversion
+    * better handling of "internal" jump inside a dynablock
+* Improved elfloader
+* Improved some tests
+* A few fixes in some Dynarec opcodes
+* Don't quit on unsupported syscall, return -1 / ENOSYS error
+* A bit more optimisation in some Dynarec opcodes
+
+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.
+

docs/COMPILE.md

@@ -0,0 +1,198 @@
+Compiling/Installing
+----
+
+#### Debian-based Linux 
+You can use [@Itai-Nelken](https://github.com/Itai-Nelken)'s apt repository to install precompiled box86 debs, updated weekly. 
+
+```
+sudo wget https://itai-nelken.github.io/weekly-box86-debs/debian/box86.list -O /etc/apt/sources.list.d/box86.list
+wget -qO- https://itai-nelken.github.io/weekly-box86-debs/debian/KEY.gpg | sudo gpg --dearmor -o /etc/apt/trusted.gpg.d/box86-debs-archive-keyring.gpg
+sudo apt update && sudo apt install box86 -y
+```
+
+**Note:** On a 64bit OS, install the `box86:armhf` package.
+
+Alternatively, you can generate your own package using the [instructions below](https://github.com/ptitSeb/box86/blob/master/docs/COMPILE.md#debian-packaging). 
+
+#### 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 # -DRPI4=1 for Pi4 (use `-DRPI2=1` etc for other models)
+make # can use `make -j2`, `make -j3`, etc to speed up 1st build (but beware of running out of memory if going too high)
+sudo make install
+sudo systemctl restart systemd-binfmt
+```
+ 
+#### for Raspberry Pi on 64bit OS
+
+_armhf multiarch or chroot required for running box86 (armhf) on aarch64_
+
+```
+# example of enabling multiarch & installing libc6:armhf (to run box86:armhf on aarch64)
+#  (running i386-wine on aarch64 requires more armhf libraries)
+sudo dpkg --add-architecture armhf && sudo apt-get update
+sudo apt-get install libc6:armhf -y
+```
+_Build box86:armhf on RPiOS 64-bit ARM (aarch64)_
+```
+sudo apt install gcc-arm-linux-gnueabihf # building 32-bit ARM code on aarch64 requires this armhf gcc cross-compiler toolchain 
+git clone https://github.com/ptitSeb/box86
+cd box86
+mkdir build; cd build; cmake .. -DRPI4ARM64=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo # -DRPI4ARM64=1 for Pi4 aarch64 (use `-DRPI3ARM64=1` for a PI3 model)
+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 devices run an AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On debian, install it with `sudo apt install gcc-arm-linux-gnueabihf`. 
+
+Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.
+
+#### for RK3588 / RK3588S
+
+`mkdir build; cd build; cmake .. -DRK3588=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j4`
+
+As most RK3588 devices run an AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On debian, install it with `sudo apt install gcc-arm-linux-gnueabihf`. 
+
+Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.
+
+#### for ODROID N2/N2+
+
+`mkdir build; cd build; cmake .. -DODROIDN2=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+
+As most ODROID N2/N2+ devices run an AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On debian, install it with `sudo apt install gcc-arm-linux-gnueabihf`. 
+
+Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.
+
+#### for Tinker Board (1/1S) or RK3288
+
+`mkdir build; cd build; cmake .. -DRK3288=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+
+#### for Allwinner A64
+
+`mkdir build; cd build; cmake .. -DA64=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+
+As most Allwinner A64 devices run an AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On debian, install it with `sudo apt install gcc-arm-linux-gnueabihf`. 
+
+Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.
+
+#### for Snapdragon 845
+
+`mkdir build; cd build; cmake .. -DSD845=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j4`
+
+As most Snapdragon 845 devices run an AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On mobian, install it with `sudo apt install gcc-arm-linux-gnueabihf`.
+
+#### for Phytium
+
+`mkdir build; cd build; cmake .. -DPHYTIUM=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j4`
+
+As most Phytium (D2000 or FT2000/4) devices run an AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On debian, install it with `sudo apt install gcc-arm-linux-gnueabihf`. 
+
+
+Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.
+
+#### for Other ARM64 64bits Linux platform
+
+`mkdir build; cd build; cmake .. -DARM64=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+
+You can also add `-DBAD_SIGNAL=ON` to the cmake command if you are on Linux Kernel mixed with Android, like on RK3588 or maybe Termux
+
+You'll need an `armhf` multiarch environment, and an armhf gcc: On debian, install it with `sudo apt install gcc-arm-linux-gnueabihf`. 
+
+Also, on armbian, you may need to install `libc6-dev-armhf-cross` or you may have an issue with `crt1.o` and a few other files not included with box86.
+
+#### for Other ARM Linux platforms
+
+ `mkdir build; cd build; cmake .. -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+
+You can also add `-DBAD_SIGNAL=ON` to the cmake command if you are on Linux Kernel mixed with Android, like on RK3588 or maybe Termux
+
+#### for x86 Linux
+
+ `mkdir build; cd build; cmake .. -DCMAKE_C_FLAGS="-m32" -DLD80BITS=1 -DNOALIGN=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
+
+### 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)
+
+#### To 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 (***the interpreter will be slightly slower***), add `-DHAVE_TRACE=1`. But you will need the [Zydis library](https://github.com/zyantific/zydis) in your `LD_LIBRARY_PATH` or in the system library folders at runtime.
+
+#### to have ARM Dynarec
+
+###### *Note: VFPv3 and NEON are required for Dynarec.*
+
+###### *Note: Compiling with `ARM_DYNAREC` without selecting a hardware profile is not advised.*
+
+Dynarec is only available on the ARM architecture (for the meantime anyways). Activate it by using `-DARM_DYNAREC=1`.
+
+You will most likely need `-marm` in compilation flags (Many compilers default to the Thumb instruction set and Dynarec does not support this).
+
+###### *Note: If you get error building that "target CPU does not support ARM mode", then try to pick a hardware profile (like ODROID for armv7 or PI4 for armv8).*
+
+##### 64bit OS with Dynarec
+
+If you are using a 64bit OS with armhf multiarch, it's much easier to pick one of the hardware profiles.
+
+`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 code zip from github), you need to use `-DNOGIT=1` from cmake to be able to build (box86 uses git SHA1 to show last commit in version number).
+
+----
+
+Testing
+----
+A few tests are included with box86.
+
+They can be launched using the `ctest` command.
+
+The tests are very basic and only tests some functionality for now.
+
+----
+
+Note about devices with Tegra X1 and newer.
+
+NVIDIA doesn't provide armhf libraries for their GPU drivers at this time. There is no special variable to compile box86 for them, as it would be misleading to many people. If you still want to use it wihout GPU acceleration, building it with RPI4 configuration should work just fine. Installation of Mesa can break the NVIDIA driver, so the safest option is to use a chroot environment.
+
+----
+
+Debian Packaging
+----
+Box86 can also be packaged into a .deb file ***using the source code zip from the releases page*** with `DEB_BUILD_OPTIONS=nostrip dpkg-buildpackage -us -uc -nc`. Configure any additional cmake options you might want in `debian/rules`.
+

docs/README.md

@@ -0,0 +1,143 @@
+![Official logo](img/Box86Logo.png "Official Logo")
+
+Linux Userspace x86 Emulator with a twist
+
+[View changelog](https://github.com/ptitSeb/box86/blob/master/docs/CHANGELOG.md) | [中文](https://github.com/ptitSeb/box86/blob/master/docs/README_CN.md) | [Українська](https://github.com/ptitSeb/box86/blob/master/docs/README_UK.md) | [Report an error](https://github.com/ptitSeb/box86/issues/new)
+
+![build](https://app.travis-ci.com/ptitSeb/box86.svg?branch=master) ![stars](https://img.shields.io/github/stars/ptitSeb/box86) ![forks](https://img.shields.io/github/forks/ptitSeb/box86) ![contributors](https://img.shields.io/github/contributors/ptitSeb/box86) ![prs](https://img.shields.io/github/issues-pr/ptitSeb/box86) ![issues](https://img.shields.io/github/issues/ptitSeb/box86)
+
+----
+
+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 environment).
+
+Because Box86 uses the native versions of some "system" libraries, like libc, libm, SDL, and OpenGL, it's easy to integrate and use with most applications, and performance can be surprisingly high in many cases. Take a look at those bench analysis for an example [here](https://box86.org/index.php/2021/06/game-performances/). That also means that you will need a 32bits userspace on 64bits OS, like `armhf` on top an `aarch64` 64bits OS.
+
+Most x86 Games need OpenGL, so on ARM platforms a solution like [gl4es](https://github.com/ptitSeb/gl4es) might be necessary. (Some ARM platforms only support OpenGL ES and/or their OpenGL implementation is dodgy. (see OpenGL on Android))
+
+Box86 now integrates a DynaRec (dynamic recompiler) for the ARM platform, providing a speed boost between 5 to 10 times faster than only using the interpreter. Some high level information on how the Dynarec work can be found [here](https://box86.org/2021/07/inner-workings-a-high%e2%80%91level-view-of-box86-and-a-low%e2%80%91level-view-of-the-dynarec/).
+
+Many games just 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). Unity3D games also works fine, but the OpenGL requirement might be an issue on some ARM platforms.
+
+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
+
+----
+Compiling/Installation
+----
+> Compilation instructions can be found [here](COMPILE.md)  
+> Instructions for installing Wine for Box86 can be found [here](X86WINE.md)  
+
+----
+
+Here are 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 videos on 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) YouTube channels.
+
+Compatibility list is here: 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_box86signalhandler` 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 Box86 configuration
+----
+
+Box86 now have configurations files. There are 2 files loaded. `/etc/box4.box86rc` and `~/.box86rc`. Both files have the same syntax, and is basicaly an ini files. Section in square brakets define the process name, and the rest is the env. var. to set. Looke at [Usage](USAGE.md) for detail on what parameters can be put. Box86 comes with a default file that should be installed for better stability. The file in in `system/box86.box86rc` and should be installed to `/etc/box86.box86rc` If, for some reasons, you don't want to install that file here, at least copy it to `~/.box86rc` or some game may not function correctly.
+Note that the priority is: `~/.bashrc` > `/etc/box86.box86rc` > command line
+So, your settings in `~/.bashrc` may override the setting from your command line...
+
+----
+
+Notes about Unity game emulation
+----
+
+Running Unity games should generaly work now, but you should also note that many Unity3D games require OpenGL 3+ which can be tricky to provide on ARM SBC (single-board computers) for now.
+Hint: on Pi4, use `MESA_GL_VERSION_OVERRIDE=3.2` and with Panfrost use `PAN_MESA_DEBUG=gl3` to use higher profile if the game starts then quits before showing anything.
+
+----
+
+Notes about Steam
+----
+
+Linux Steam's can run now with box86. but you also need box64 for it to completly usable.
+It is advised to run steam in the Small mode, as it use the less memory, but steamwebhelper (a 64bits process) will still be loaded even if not used.
+The login screen cannot run without steamwebhelper, and will just show a blank windows without box64 properly setup in the system.
+File note, Steam will use a lot of memory, and barely fit on system with 4GB of RAM. It will not work anymore on system with less memory (as a workaround, create a swap file, login and check "remember me", the use box64rc to disable steamwebhelper and run only in small mode without swap after the 1st login)
+Final note: the Steam BigPicture will work, but also need steamwebhelper (and so box64), and lots of memory. It will not start on system with only 4GB of RAM without swap.
+- If you have trouble installing Steam, you can find `install_steam.sh` in the root folder of the box86 repo. This simple script will download and install steam in your home folder, and then create a shortcut to steam in `/usr/local/bin` (and for this it will ask for sudo permission). Simply use `steam` to launch once it's installed. Note that the installation, being in the Home folder, will only work for a single user. Don't use this script if you need a multi-user installation.
+- To avoid the "libc.so.6 is absent" message, you can use `STEAMOS=1` and `STEAM_RUNTIME=1` as environment variables (it's automatically there if you used the `install_steam.sh` script)
+
+to disable `steamwebhelper` when box64 installed and running, create (or edit) `~/.box64rc` and add:
+```
+[steamwebhelper]
+BOX64_EXIT=1
+```
+in it and it will be disable. Comment with a `#` those 2 lines to enable it again
+
+----
+
+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 Vulkan. If your system has a 32bits Vulkan driver, box86 will use it when needed. Profile 1.0, 1.1, 1.2 and 1.3, with some extensions, should be OK. DXVK, including the 2.0, works. 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 (hardware 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 development.
+There are many ways to contribute: code, financial, hardware and advertisement!
+So, in no particular order, I want to thank:
+ * For their major code contribution: rajdakin, icecream95, M-HT
+ * For their major financial contribution: FlyingFathead, stormchaser3000, dennis1248, sll00, [libre-computer-project](https://libre.computer/)
+ * For their hardware contribution: [Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/), [DragonBox](https://pyra-handheld.com/), [Novaspirit](https://www.youtube.com/channel/UCrjKdwxaQMSV_NDywgKXVmw), [HardKernel](https://www.hardkernel.com/), [TwisterOS team](https://twisteros.com/)
+ * For their continuous advertisement of box86 project: salva ([microLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ)), [PILab](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g)/[TwisterOS team](https://twisteros.com/), [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA), [NicoD](https://www.youtube.com/channel/UCpv7NFr0-9AB5xoklh3Snhg)
+
+And I also thank the many other people who participated even once in the project.
+
+(If you use Box86 in your project, please don't forget to mention it!)

docs/README_CN.md

@@ -0,0 +1,135 @@
+![Official logo](img/Box86Logo.png "Official Logo")
+
+Linux用户空间的x86应用模拟器 
+
+[更新历史](https://github.com/ptitSeb/box86/blob/master/docs/CHANGELOG.md) | [English](https://github.com/ptitSeb/box86/blob/master/docs/README.md) | [Українська](https://github.com/ptitSeb/box86/blob/master/docs/README_UK.md) | [报告错误](https://github.com/ptitSeb/box86/issues/new)
+
+![build](https://app.travis-ci.com/ptitSeb/box86.svg?branch=master) ![stars](https://img.shields.io/github/stars/ptitSeb/box86) ![forks](https://img.shields.io/github/forks/ptitSeb/box86) ![contributors](https://img.shields.io/github/contributors/ptitSeb/box86) ![prs](https://img.shields.io/github/issues-pr/ptitSeb/box86) ![issues](https://img.shields.io/github/issues/ptitSeb/box86)
+
+----
+
+使用Box86可以在ARM或其他非x86架构的Linux操作系统上运行x86 Linux程序（例如游戏），请注意，主机系统需要是**32位小端**的。
+
+你需要一个**32位**的子系统来运行Box86，因为Box86在纯64位系统上无法使用，同时你也需要一个**32位**的Linux编译工具链来编译Box86，例如**arm-linux-gnueabihf-gcc**， 否则你将无法编译Box86 （一般的64位ARM编译工具链如 aarch64-linux-gnu-gcc 不支持"-marm"编译选项，会编译失败并报错）。
+
+因为Box86使用了一些“系统”库的原生版本，例如libc、libm、SDL和OpenGL，所以在很多情况下性能很不错，也可以方便地与其他应用集成使用。点击[此处](https://box86.org/index.php/2021/06/game-performances/)查看Benchmark示例。
+
+大多数x86游戏都用到了OpenGL，所以在ARM平台上类似于[gl4es](https://github.com/ptitSeb/gl4es)的解决方案通常是必要的。（许多ARM平台仅提供OpenGL ES支持，又或者他们的OpenGL实现很不可靠（例如Android上的OpenGL）。）
+
+Box86为ARM平台集成了DynaRec（动态重编译器），比仅使用解释器提升5到10倍性能。关于Dynarec如何工作的更多信息，可以在[这里](https://box86.org/2021/07/inner-workings-a-high%e2%80%91level-view-of-box86-and-a-low%e2%80%91level-view-of-the-dynarec/)找到。
+
+许多游戏已可在不做多少改动的情况下直接运行，例如：《粘粘世界》(WorldOfGoo)、《航空大亨豪华版》(Airline Tycoon Deluxe)、《超越光速》(FTL)等、基于GameMaker的游戏（《传说之下》(UNDERTALE)等）。
+
+如果你想开发Box86，最好安装ccache，以缓存编译结果。
+如果你需要启用Trace跟踪（例如输出寄存器值与各条被执行的x86指令），你也需要安装[Zydis](https://github.com/zyantific/zydis)。
+
+一些x86内部操作码使用“Realmode X86 Emulator Library”的部分内容，请参阅[x86primop.c](../src/emu/x86primop.c)了解版权详情 
+
+----
+
+编译与安装
+----
+> [Box86编译指南](COMPILE.md)
+> 
+> [Box86 Wine安装指南](X86WINE.md)
+
+一些示例视频
+----
+
+[![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)
+
+在这几个Youtube频道可以找到更多Box86的相关视频
+
+[MicroLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ)
+
+[Pi Labs](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g)
+
+[The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA)
+
+程序兼容性列表: https://github.com/ptitSeb/box86-compatibility-list/issues
+
+<img src="img/Box86Icon.png" width="96" height="96">
+
+感谢制作Logo和图标的grayduck!
+
+注意：
+不要把本项目和[86box](https://github.com/86Box/86Box)搞混了，后者是一个"**全量**"的x86 PC模拟器。
+
+----
+
+使用方法
+----
+
+有若干环境变量可以设置Box86的行为。
+
+点击[这里](USAGE.md)了解所有的环境变量和它们的作用。
+
+请注意：Box86的Dynarec提供了一种包含"内存保护"和段错误信号处理程序的机制来处理JIT代码。简单来说，如果你想用GDB来调试一个使用JIT代码的运行程序（比如mono/Unity3D），就会触发很多“正常”的段错误。建议在GDB中使用类似于`handle SIGSEGV nostop`的命令，来防止在每个段错误时都停止程序运行，如果你想捕获段错误，可以在`signals.c`中的`my_memprotectionhandler`中放置一个断点。
+
+----
+
+关于64位平台的注意事项
+----
+
+Box86的工作原理是直接将函数调用从x86_转换到主机系统，主机系统（Box86运行的系统）需要有32位的库。Box86并不包括任何64位<->32位的转换。所以如果你需要在64位平台上运行Box86，那么你需要建立一个32位的库环境，并以32位ARM为目标系统编译Box86（例如使用arm-linux-gnueabihf-gcc工具链交叉编译或直接在armhf chroot环境中编译运行）。
+
+如果你需要运行64位应用，请考虑使用[Box64](https://github.com/ptitSeb/box64)，它可以在64位平台上运行x86_64程序，但请再一次注意运行x86程序需要Box86与32位库环境（就像在x86_64的Linux上运行x86程序需要x86 multiarch库一样）。
+
+----
+
+关于Unity游戏的注意事项
+----
+
+当前对于Unity游戏的模拟仍不完善。Unity使用Mono，但它所使用的信号机制还没有被完美模拟。理论上可以使用原生Mono库（在[这里](https://github.com/Unity-Technologies/mono)可以找到），但Mono的封装比较麻烦，现在还没完成，所以我们使用了模拟的Mono库，目前Mono的JIT动态编译代码已由"内存保护"(Protected memory)机制实现，应该能正常运行。
+
+同时，一些Unity3D游戏需要OpenGL 3+，目前在大部分Arm开发板上提供OpenGL 3+是很困难的。
+
+所以，并不是所有的Unity游戏都能运行，尤其是那些需要提供很高的OpenGL配置（3.2+）的游戏更有可能不行，但除此之外应该问题不大，而且能运行的游戏的性能应该不错。
+
+提示：如果游戏开始后在显示任何东西之前退出，在Raspberry Pi4上使用`MESA_GL_VERSION_OVERRIDE=3.2`，在Panfrost上使用`PAN_MESA_DEBUG=gl3`来使用更高的配置文件。
+
+----
+
+关于Steam的注意事项
+----
+
+Linux Steam 现在可以在Box86上运行，但还是有点不稳定，而且有些功能可能不正常：
+- 如果Steam在显示登录界面后崩溃，你可能需要安装libappindicator库，例如运行`sudo apt install libappindicator1`。
+- 如果你选择记住密码，而Steam在下次启动时崩溃，你可能需要安装libnm库，例如运行`sudo apt install libnm0`。
+- Steam不能在正常的大界面模式下工作，因为Steam在浏览器模式使用的一些组件只能在64位平台上运行。所以请在查看选项卡中选择"小模式"，否则Steam将无法正常显示列表。或者在启动Steam添加`+open steam://open/minigameslist`参数，这样Steam将会直接在小模式下启动。
+- 为了避免"libc.so.6 is absent"的错误，请添加环境变量`STEAMOS=1` 与 `STEAM_RUNTIME=1`。
+- 一些游戏（例如基于起源引擎的游戏：《传送门》、《半衰期2》）使用了libtcmalloc库，Box86会尝试检测它们并通过LD_PRELOAD预加载它们来获得更好的兼容性，虽然应该可以在没有上述功能的情况下工作，但保险起见，我们还是建议你安装它，例如运行`sudo apt install libtcmalloc-minimal4`。
+
+Windows Steam 可以安装但目前不能运行。
+
+----
+
+关于Wine的注意事项
+----
+
+Box86现在已经支持Wine。Wine的内建程序全部能运行，许多Windows 程序和游戏也运行良好。但是不要忘记大多数 Windows 游戏都使用 Direct3D，这可能需要完整的 OpenGL 驱动程序和尽可能高的配置文件（基于 ES2 后端的 gl4es 现在不能运行 Wine 的Direct3D组件）。
+注意：如果你想在 Raspberry Pi 3 或更早版本上用 Box86 运行 Wine，由于这些设备的内核默认使用 2G/2G 虚拟内存分配（2G 空间用于用户态程序，2G 空间用于内核），与需要访问内存地址 > 2G 的 Wine 程序不兼容。 因此，你需要重新配置内核以使用 3G/1G 的内存分配方式。
+
+----
+
+关于Vulkan的注意事项
+----
+
+Box86已经封装了Vulkan。如果设备的系统有32位的Vulkan驱动程序，Box86会在需要时使用它。但请注意，由于作者目前拥有的硬件上的Vulkan支持有限，因此Vulkan封装库尚未经过太多测试。 Vulkan 1.0、1.1与一些扩展应该能工作。 但Vulkan 1.2还没有被实现。已经有一些在树莓派4上的Vulkan演示 （Sascha Willems 的演示中，为x86构建的demo与与直接为armhf构建的demo工作均正常）。但请注意，Pi4的Vulkan驱动程序目前不支持 dxvk（wine DirectX->Vulkan 兼容层），这不是Box86的问题，而是由于缺少扩展（硬件支持）与一些其他驱动问题。而在Panfrost上，PanVK尚不成熟，暂时没有用它测试dxvk。
+
+----
+
+结语
+----
+
+我要感谢所有为 box86 开发做出贡献的人。
+贡献方式有很多种：代码贡献、财务帮助、捐赠捐赠和广告合作！
+以下感谢不分先后： 
+ * 代码贡献：rajdakin, icecream95, M-HT
+ * 财务帮助：FlyingFathead, stormchaser3000, dennis1248, sll00, [libre-computer-project](https://libre.computer/)
+ * 硬件捐赠：[Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/), [DragonBox](https://pyra-handheld.com/), [Novaspirit](https://www.youtube.com/channel/UCrjKdwxaQMSV_NDywgKXVmw), [HardKernel](https://www.hardkernel.com/), [TwisterOS team](https://twisteros.com/)
+ * 为本项目持续的宣传：([microLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ)), [PILab](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g)/[TwisterOS team](https://twisteros.com/), [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA), [NicoD](https://www.youtube.com/channel/UCpv7NFr0-9AB5xoklh3Snhg)
+
+我还要感谢很多人，即使他们只参与了一次本项目。
+
+(如果你的项目中使用了Box86，请不要忘记提及！)

docs/README_UK.md

@@ -0,0 +1,143 @@
+![Official logo](img/Box86Logo.png "Official Logo")
+
+Емулятор простіра користувача x86_64 із родзинкою
+
+[Дивитись журнал змін](https://github.com/ptitSeb/box86/blob/master/docs/CHANGELOG.md) | [中文](https://github.com/ptitSeb/box86/blob/master/docs/README_CN.md) | [Українська](https://github.com/ptitSeb/box86/blob/master/docs/README_UK.md) | [Повідомити про помилку](https://github.com/ptitSeb/box86/issues/new)
+
+![build](https://app.travis-ci.com/ptitSeb/box86.svg?branch=main) ![stars](https://img.shields.io/github/stars/ptitSeb/box86) ![forks](https://img.shields.io/github/forks/ptitSeb/box86) ![contributors](https://img.shields.io/github/contributors/ptitSeb/box86) ![prs](https://img.shields.io/github/issues-pr/ptitSeb/box86) ![issues](https://img.shields.io/github/issues/ptitSeb/box86)
+
+----
+
+Box86 дозволяє запускати x86 Linux програми (наприклад, ігри) на системах Linux відмінних від x86, наприклад ARM (хост-система має мати 32-розрядний порядок редагування).
+
+Вам *ПОТРІБНА* 32-розрядна підсистема для запуску та створення Box86. Box86 марний лише на 64-розрядних системах. Крім того, вам *ПОТРІБЕН* 32-розрядний інструментарій для створення Box86. Ланцюжок інструментів, який підтримує лише 64-розрядні версії, не компілює Box86, і ви отримаєте помилки (зазвичай на aarch64 ви отримуєте що «-marm» не розпізнається, і вам знадобиться середовище з мультиархітектурою або chroot).
+
+Оскільки Box86 використовує рідні версії деяких «системних» бібліотек, таких як libc, libm, SDL і OpenGL, його легко інтегрувати та використовувати з більшістю програм, а продуктивність у багатьох випадках може бути напрочуд високою. Подивіться на цей стендовий аналіз для прикладу [тут](https://box86.org/index.php/2021/06/game-performances/). Це також означає, що вам знадобиться 32-бітний простір користувача на 64-бітній ОС, наприклад `armhf` поверх `aarch64` 64-бітної ОС.
+
+Більшість x86-ігор потребують OpenGL, тому на платформах ARM може знадобитися таке рішення, як [gl4es](https://github.com/ptitSeb/gl4es). (Деякі платформи ARM підтримують лише OpenGL ES і/або їх реалізація OpenGL хитра. (див. OpenGL на Android))
+
+Box86 інтегрується з DynaRec (динамічний рекомпілятор) для платформ ARM64 і RV64 забезпечуючи підвищення швидкості в 5-10 разів, ніж використання лише інтерпретатора. Деякі високорівневі відомості про те, як працює DynaRec, можна знайти [тут](https://box86.org/2021/07/inner-workings-a-high%e2%80%91level-view-of-box86-and-a-low%e2%80%91level-view-of-the-dynarec/).
+
+Багато ігор просто працюють без особливих налаштувань, наприклад: WorldOfGoo, Airline Tycoon Deluxe і FTL. Багато ігор GameMaker Linux також працюють нормально. (є довгий список, серед них UNDERTALE, A Risk of Rain і Cook Serve Delicious). Ігри Unity3D також працюють добре, але вимога OpenGL може бути проблемою на деяких платформах ARM.
+
+Якщо ви серйозно ставитеся до розробки Box86, вам слід встановити ccache та створити Box86 з ним. (Використовуйте, наприклад, ccmake.)
+Щоб увімкнути TRACE (тобто скидання в stdout усіх окремих виконаних інструкцій x86 із дампом регістрів), вам також знадобиться [бібліотека Zydis](https://github.com/zyantific/zydis), доступна у вашій системі.
+
+Деякі внутрішні коди операцій x86 використовують частини «Бібліотеки емулятора Realmode X86», дивіться [x86primop.c](../src/emu/x86primop.c) для детальної інформації про авторські права.
+
+----
+Компіляція/Встановлення
+----
+> Інструкцію зі складання Box86 можна знайти [тут](COMPILE.md).\
+> Інструкцію з встановлення Wine для Box86 можна знайти [тут](X86WINE.md).
+
+----
+
+Ось 6 відео, перші 2 відео – це відео "Airline Tycoon Deluxe" і "Heretic 2", які працюють на GigaHertz OpenPandora (друге використовує Dynarec), а наступні 2 відео – це відео "Bit.Trip.Runner" і "Neverwinter Night", які працюють на ODroid XU4 (без dynarec), а останні 2 відео на Pi4: Shovel Knight (відео від @ITotalJustice) і Freedom Planet (відео від @djazz), також без Dynarec.
+
+
+[![Play on Youtube](https://img.youtube.com/vi/bLt0hMoFDLk/3.jpg)](https://www.youtube.com/watch?v=bLt0hMoFDLk) [![Play on Youtube](https://img.youtube.com/vi/MM7kWYts7IA/3.jpg)](https://www.youtube.com/watch?v=MM7kWYts7IA) [![Play on Youtube](https://img.youtube.com/vi/8hr71S029Hg/1.jpg)](https://www.youtube.com/watch?v=8hr71S029Hg) [![Play on Youtube](https://img.youtube.com/vi/B4YN37z3-ws/1.jpg)](https://www.youtube.com/watch?v=B4YN37z3-ws) [![Play on Youtube](https://img.youtube.com/vi/xk8Q30mxqPg/1.jpg)](https://www.youtube.com/watch?v=xk8Q30mxqPg) [![Play on Youtube](https://img.youtube.com/vi/_QMRMVvYrqU/1.jpg)](https://www.youtube.com/watch?v=_QMRMVvYrqU)
+
+Ви можете знайти багато інших відео про Box86 у YouTube каналах [MicroLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ), [Pi Labs](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g) або [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA).
+
+Список сумісності доступний тут: https://github.com/ptitSeb/box86-compatibility-list/issues
+
+<img src="img/Box86Icon.png" width="96" height="96">
+
+Логотип і піктограма зроблені @grayduck, дякую!
+
+Зауважте, що цей проект не слід плутати з [86box](https://github.com/86Box/86Box), гарним емулятором "повної системи", що спеціалізується на ранньому (до відносно недавньому) апаратному забезпеченні ПК.
+
+----
+
+Використання
+----
+
+Є кілька змінних середовища для керування поведінкою Box86.
+
+Дивіться [тут](USAGE.md) для перегляду всіх змінних середовища та того, що вони роблять.
+
+Примітка: Dynarec від Box86 використовує механізм із захистом пам’яті та обробником сигналу помилки сегментів для обробки JIT коду. Простіше кажучи, якщо ви хочете використовувати GDB для налагодження запущеної програми, яка використовує JIT-код (наприклад, mono/Unity3D), ви все одно матимете багато «звичайних» помилок сегментів. Пропонується використовувати щось на зразок `handle SIGSEGV nostop` в GDB, щоб не зупинятися на кожній помилці сегмента, і, можливо, поставити точку зупину всередині `my_memprotectionhandler` в `signals.c`, якщо ви хочете перехопити помилки сегментів.
+
+----
+
+Історія версій/Журнал змін
+----
+
+Історія версій доступна [тут](docs/CHANGELOG.md).
+
+----
+
+Примітки щодо 64-розрядних платформ
+----
+
+Оскільки Box86 працює шляхом прямого перекладу викликів функцій із x86 на хост-систему, хост-система (на якій працює Box86) повинна мати 32-розрядні бібліотеки. Box86 не містить 32-біт <-> 64-біт перекладу. Отже, щоб запустити Box86, наприклад, на платформі ARM64, вам потрібно буде зібрати Box86 для 32-бітної ARM, а також потрібно мати chroot з 32-бітними бібліотеками.
+
+Якщо ви дивитесь на 64-розрядну версію box86, подивіться на [Box64](https://github.com/ptitSeb/box64): він може запускати двійкові файли x86_64 на 64-розрядних платформах. Але зауважте, що вам все одно потрібен Box86 (і 32-розрядний chroot) для запуску бінарних файлів x86 (як це також трапляється в реальному Linux x86_64, для запуску якого потрібні бібліотеки x86 і двійковий файл в multiarch).
+
+----
+
+Примітки щодо конфігурації Box86
+----
+
+Box86 тепер має конфігураційні файли. Завантажуються 2 файли: `/etc/box4.box86rc` і `~/.box86rc`. Обидва файли мають однаковий синтаксис і в основному є файлами ini. Розділ у квадратних дужках визначає назву процесу, а решта — встановлють змінну середовища. Подивіться на [використання](docs/USAGE.md), щоб дізнатися, які параметри можна використовувати. Box86 поставляється з файлом за замовчуванням, який слід інсталювати для кращої стабільності. Файл знаходиться в `system/box86.box86rc` і має бути встановлено в `/etc/box86.box86rc`. Якщо з певних причин ви не хочете встановлювати цей файл у `/etc`, принаймні скопіюйте його в `~/.box86rc` або гра може не працювати належним чином. Зауважте, що пріоритетом є: `~/.bashrc` > `/etc/box86.box86rc` > командний рядок
+Таким чином, ваші налаштування в `~/.bashrc` можуть замінити налаштування з вашого командного рядка...
+
+----
+
+Примітки щодо емуляції ігор Unity
+----
+
+Запуск ігор Unity має працювати, але ви також повинні зауважити, що для багатьох ігор Unity3D потрібен OpenGL 3+, який може бути складно забезпечити на ARM SBC (одноплатних комп’ютерах). Крім того, багато нових ігор Unity3D (наприклад, KSP) використовують стиснуті текстури BC7, які не підтримуються багатьма інтегрованими GPU ARM.
+Підказка: в Pi4 використовуйте `MESA_GL_VERSION_OVERRIDE=3.2`, а в Panfrost — `PAN_MESA_DEBUG=gl3`, щоб використовувати вищий профіль, якщо гра запускається, а потім завершується, перш ніж щось показує.
+
+----
+
+Примітки щодо Steam
+----
+
+Linux версія Steam тепер може працювати з Box86, але вам також потрібен Box64, щоб його можна було повністю використовувати.
+Рекомендується запускати Steam у small режимі, оскільки він використовує менше пам’яті, але steamwebhelper (64-бітний процес) усе одно завантажуватиметься, навіть якщо він не використовується.
+Екран входу не може працювати без steamwebhelper, і буде показувати лише порожні вікна без належного налаштування box64 у системі.
+Зауважте, що Steam використовуватиме багато пам’яті та ледве поміститься в систему з 4 ГБ оперативної пам’яті. Він більше не працюватиме в системі з меншим об’ємом пам’яті (як обхідний шлях, створіть файл підкачки, увійдіть і поставте прапорець «запам’ятати мене», використовуйте box64rc, щоб вимкнути steamwebhelper і запускати лише в малому режимі без підкачки після першого входу в систему)
+Останнє зауваження: Steam BigPicture працюватиме, але також потрібен steamwebhelper (і box64) і багато пам’яті. Він не запуститься в системі лише з 4 ГБ оперативної пам’яті без файлу підкачки.
+- Якщо у вас виникли проблеми зі встановленням Steam, ви можете знайти `install_steam.sh` у кореневій папці сховища box86. Цей простий скрипт завантажить і встановить Steam у вашу домашню папку, а потім створить ярлик steam у `/usr/local/bin` (для цього він запитає дозвіл sudo). Просто використовуйте `steam` для запуску після встановлення. Зауважте, що інсталяція, яка знаходиться в папці Home, працюватиме лише для одного користувача. Не використовуйте цей скрипт, якщо вам потрібна багатокористувацька інсталяція.
+- Щоб уникнути повідомлення «libc.so.6 відсутній», ви можете використовувати `STEAMOS=1` і `STEAM_RUNTIME=1` як змінні середовища (воно автоматично з’являється, якщо ви використовували сценарій `install_steam.sh`)
+
+Щоб вимкнути `steamwebhelper`, коли Box64 встановлений і працює, створіть (або відредагуйте) `~/.box64rc` і додайте у ньому:
+```
+[steamwebhelper]
+BOX64_EXIT=1
+```
+І тоді `steamwebhelper` буде вимкнено. Прокоментуйте `#` ці 2 рядки, щоб увімкнути його знову.
+
+----
+
+Примітки щодо Wine
+----
+
+Wine тепер підтримується. Інтегровани програми Wine працює як і багато програм та ігор для Windows. Не забувайте, що більшість ігор для Windows використовують Direct3D, для цього може знадобитися повний драйвер OpenGL і якомога більш високий профіль (і gl4es з бекендом ES2 наразі мають проблеми з Wine).
+Примітка: якщо ви плануєте використовувати box86 з Wine на Raspberry Pi 3 або ранішої версії, ці моделі використовують ОС за замовчуванням, яка має ядро ​​з розділенням 2/2 (тобто 2G місця для програми користувача та 2 Г місця для ядра). Це несумісно з програмами Wine, яким потрібен доступ до пам’яті > 2 Гб адреси. Тож вам потрібно буде переналаштувати ядро ​​для розділення 3G/1G.
+
+----
+
+Примітки щодо Vulkan
+----
+
+Box86 вже використовує Vulkan. Якщо у вашій системі є 32-розрядний драйвер Vulkan, Box86 використовуватиме його за потреби. Профіль 1.0, 1.1, 1.2 і 1.3, з деякими розширеннями, має бути в порядку. DXVK, включаючи 2.0, працює. Я знаю, що деякі демонстраційні версії працюють на Pi4 (демонстраційні версії Sascha Willems для x86 працюють так само, як якщо б створювати безпосередньо на armhf). Зауважте, що драйвер Vulkan для Pi4 наразі НЕ підтримує DXVK (обгортка Wine DirectX->Vulkan). Це не проблема Box86, у драйвері відсутні розширення (підтримка апаратного забезпечення) та кілька інших речей, через які DXVK не працює на pi4. З боку Panfrost, PanVK трохи молодий, і я ще не тестував DXVK з ним.
+
+----
+Останнє слово
+----
+
+Я хочу подякувати всім, хто зробив внесок у розвиток box86.
+Є багато способів зробити свій внесок: код, фінанси, апаратне забезпечення та реклама!
+Отже, без особливого порядку, я хочу подякувати:
+ * За основний внесок у код: rajdakin, icecream95, M-HT
+ * За великий фінансовий внесок: FlyingFathead, stormchaser3000, dennis1248, sll00, [libre-computer-project](https://libre.computer/)
+ * За внесок у апаратне забезпечення: [Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/), [DragonBox](https://pyra-handheld.com/), [Novaspirit](https://www.youtube.com/channel/UCrjKdwxaQMSV_NDywgKXVmw), [HardKernel](https://www.hardkernel.com/), [Команда TwisterOS](https://twisteros.com/)
+ * За постійну рекламу Box86: salva ([microLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ)), [PILab](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g)/[Команда TwisterOS](https://twisteros.com/), [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA), [NicoD](https://www.youtube.com/channel/UCpv7NFr0-9AB5xoklh3Snhg)
+
+І я також дякую багатьом іншим людям, які хоч раз брали участь у проекті.
+
+(Якщо ви використовуєте Box86 у своєму проекті, будь ласка, не забудьте згадати про це!)

docs/USAGE.md

@@ -0,0 +1,313 @@
+
+Usage
+----
+
+There are many environment variables to control Box86 behaviour. 
+Env. var with * can also be put inside box64rc files
+
+#### 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_ROLLING_LOG *
+Show last few wrapped function call when a Signal is caught
+ * 0: No last function call printed (Default.)
+ * 1: Last 16 wrapped functions calls printed when a signal is printed. Incompatible with BOX86_LOG>1 (may need BOX86_SHOWSEGV=1 also)
+ * N: Last N wrapped functions calls printed when a signal is printed. Incompatible with BOX86_LOG>1 (may need BOX86_SHOWSEGV=1 also)
+
+#### 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_TRACE_EMM *
+Only on builds with trace enabled.
+ * 0 : The EMM (i.e. MMX) register will not be logged with the general and x86 registers. (Default.)
+ * 1 : Dump the EMM registers.
+
+#### BOX86_LOAD_ADDR *
+Try to load at 0xXXXXXX main binary (if binary is a PIE)
+ * 0xXXXXXXXX : The load address . (Only active on PIE programs.)
+
+#### BOX86_NOSIGSEGV *
+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_TEST *
+Dynarec will compare it's execution with the interpreter (super slow, only for testing)
+ * 0 : No comparison
+ * 1 : Each opcode runs on interepter and on Dynarec, and regs and memory are compared and print if different
+
+#### BOX86_DYNAREC_BIGBLOCK *
+Enables/Disables Box86's Dynarec building BigBlock.
+ * 0 : Don't try to build block as big as possible (can help program using lots of thread and a JIT, like C#/Unity) (Default when libmonobdwgc-2.0.so is loaded)
+ * 1 : Build Dynarec block as big as possible (Default.)
+ * 2 : Build Dynarec block bigger (don't stop when block overlaps)
+
+#### BOX86_DYNAREC_FORWARD *
+Define Box86's Dynarec max allowed forward value when building Block.
+ * 0 : No forward value. When current block end, don't try to go further even if there are previous forward jumps
+ * XXX : Allow up to XXXX bytes of gap when building a Block after the block end to next forward jump (Default: 128)
+ 
+#### BOX86_DYNAREC_STRONGMEM *
+Enable/Disable simulation of Strong Memory model
+* 0 : Don't try anything special (Default.)
+* 1 : Enable some Memory Barrier when reading from memory (on some MOV opcode) to simulate Strong Memory Model while trying to limit performance impact (Default when libmonobdwgc-2.0.so is loaded)
+* 2 : Enable some Memory Barrier when reading from memory (on some MOV opcode) to simulate Strong Memory Model
+
+#### BOX86_DYNAREC_X87DOUBLE *
+Force the use of Double for x87 emulation
+* 0 : Try to use float when possible for x87 emulation (faster, Default)
+* 1 : Only use Double for x87 emulation (slower, may be needed for some specific games, like Crysis were it's enable automatically)
+
+#### BOX86_DYNAREC_FASTNAN *
+Enable/Disable generation of -NAN
+* 0 : Generate -NAN like on x86 (slower, more x86 accurate)
+* 1 : Don't do anything special with NAN, to go as fast as possible (Default)
+
+#### BOX86_DYNAREC_FASTROUND *
+Enable/Disable generation of precise x86 rounding
+* 0 : Generate float/double -> int rounding like on x86
+* 1 : Don't do anything special with edge case Rounding, to go as fast as possible (no INF/NAN/Overflow -> MIN_INT conversion) (faster, Default)
+
+#### BOX86_DYNAREC_SAFEFLAGS *
+Handling of flags on CALL/RET opcodes
+* 0 : Treat CALL/RET as if it never needs any flags (faster but may have side-effects)
+* 1 : most of RET will need flags, most of CALLS will not (Default)
+* 2 : All CALL/RET will need flags (slower, but might be needed. Automatically enabled for Vara.exe)
+
+#### BOX86_DYNAREC_HOTPAGE *
+Handling of HotPage (Page beeing both executed and writen)
+* 0 : Don't track hotpage
+* 1-255 : Trak HotPage, and disable execution of a page beeing writen for N attempts (default is 16)
+
+#### BOX86_DYNAREC_FASTPAGE *
+Will use a faster handling of HotPage (Page being both executed and written)
+* 0 : use regular hotpage (Default)
+* 1 : Use faster hotpage, taking the risk of running obsolete JIT code (might be faster, but more prone to crash)
+
+#### BOX86_DYNAREC_BLEEDING_EDGE *
+Detect MonoBleedingEdge and apply conservative settings
+* 0 : Don't detect MonoBleedingEdge
+* 1 : Detect MonoBleedingEdge, and apply BIGBLOCK=0 STRONGMEM=1 if detected (Default)
+
+#### BOX86_DYNAREC_JVM *
+Detect libjvm and apply conservative settings
+* 0 : Don't detect libjvm
+* 1 : Detect libjvm, and apply BIGBLOCK=0 STRONGMEM=1 if detected (Default)
+
+#### BOX86_DYNAREC_WAIT *
+Behavior with FillBlock is not availble (FillBlock build Dynarec blocks and is not multithreaded)
+* 0 : Dynarec will not wait for FillBlock to ready and use Interpreter instead (might speedup a bit massive multithread or JIT programs)
+* 1 : Dynarec will wait for FillBlock to be ready (Default)
+
+#### BOX86_SSE_FLUSHTO0 *
+Handling of SSE Flush to 0 flags
+* 0 : Just track the flag (Default)
+* 1 : Direct apply of SSE Flush to 0 flag
+
+#### BOX86_X87_NO80BITS *
+Handling of x87 80bits long double
+* 0 : Try to handle 80bits long double as precise as possible (Default)
+* 1 : Handle them as double
+
+#### BOX86_LIBCEF *
+Detect libcef and apply malloc_hack settings
+* 0 : Don't detect libcef
+* 1 : Detect libcef, and apply MALLOC_HACK=2 if detected (Default)
+
+#### BOX86_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 (not that using this inside a rcfile might not work if the process needs to be renamed, for example for Wine process)
+ 
+#### BOX86_EMULATED_LIBS *
+ * XXXX[:YYYYY] force lib XXXX (and YYYY...) to be emulated (and not wrapped)
+Some games uses an old version of some libraries, with an ABI incompatible with native version.
+Note that LittleInferno for example is auto detected, and libvorbis.so.0 is automatical added to emulated libs, and same for Don't Starve (and Together / Server variant) that use an old SDL2 too
+
+#### BOX86_ALLOWMISSING_LIBS *
+Allow Box86 to continue even if a library is missing (renamed from BOX86_ALLOWMISSINGLIBS). 
+ * 0 : Box86 will stop if a library cannot be loaded. (Default.)
+ * 1 : Continue even if a needed library cannot be loaded. Unadvised, this will, in most cases, crash later on.
+
+#### BOX86_ALLOWMISSING_SYMBOLS *
+Allow dlopen with RTLD_NOW flag to not resolve all symbols. 
+ * 0 : Box86 will stop if a library have unresolved symbols. (Default.)
+ * 1 : Continue even if a needed library have unresolved symbols (like symbols not yet wrapped).
+
+#### BOX86_PREFER_WRAPPED *
+Box86 will use wrapped libs even if the lib is specified with absolute path
+ * 0 : Try to use emulated libs when they are defined with absolute path  (Default.)
+ * 1 : Use Wrapped native libs even if path is absolute
+
+#### BOX86_PREFER_EMULATED *
+Box86 will prefer emulated libs first (execpt for glibc, alsa, pulse, GL, vulkan and X11
+ * 0 : Native libs are prefered (Default.)
+ * 1 : Emulated libs are prefered (Default for program running inside pressure-vessel)
+
+#### BOX86_NOPULSE *
+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.
+
+#### 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_ENV
+ * XXX=yyyy
+ will add XXX=yyyy env. var.
+
+#### BOX86_ENV1
+ * XXX=yyyy
+ will add XXX=yyyy env. var. and continue with BOX86_ENV2 ... until var doesn't exist
+
+#### BOX86_FIX_64BIT_INODES *
+Fix/Don't fix 64bit inodes
+ * 0 : Don't fix 64bit inodes. (Default.)
+ * 1 : Fix 64bit inodes. Helps when running on Filesystems with 64bit inodes. Is useful when a program uses API functions which doesn't support it and the program doesn't use inodes information.
+
+#### BOX86_SHOWSEGV *
+Show Segfault signal even if a signal handler is present
+ * 0 : Don't force segfault details when a signal handler is install (Default, to limit message spamming)
+ * 1 : Show segfault details
+
+#### BOX86_SHOWBT *
+Show a backtrace (native and emulated) on Signal
+ * 0 : Don't show any backtrace (Default.)
+ * 1 : Show native backtrace when a signal is dumped (not available on Andoid build)
+
+#### BOX86_BASH *
+Define x86 bash to launch script
+ * yyyy
+Will use yyyy as x86 bash to launch script. yyyy needs to be a full path to a valid x86 version of bash. Note that is Box64 is available, an x86_64 version of bash can also be used. Also, BOX64_BASH env. var. will also be read if BOX86_BASH is not defined.
+
+#### BOX86_BOX64
+Define where to find box64 binary
+ * yyyy
+Will use yyyy as x86_64 interpretor, to launch x64_64 binaries
+
+#### BOX86_JITGDB *
+
+ * 0 : Just print the Segfault message on segfault (default)
+ * 1 : Launch `gdb` when a segfault, bus error or illegal instruction signal is trapped, attached to the offending process and go in an endless loop, waiting.
+ 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)
+
+#### BOX86_NORCFILES
+If the env var exist, no rc files (like /etc/box86.box86rc and ~/.box86rc) will be loaded
+
+#### BOX86_NOSANDBOX
+ * 0 : Nothing special
+ * 1 : Added "--no-sandbox" to command line arguments (usefull for chrome based programs)
+

docs/X86WINE.md

@@ -0,0 +1,174 @@
+# 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.**_
+
+See installation steps below (in the [Examples](#examples) section).
+
+Using Wine with Box86 allows (x86) Windows programs to run on ARM Linux computers (for x64, use [box64](https://github.com/ptitSeb/box64) & wine-amd64 with an `aarch64` processor).
+
+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 device's 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)
+ - If you are running box86/i386-wine on a 64bit ARM OS (aarch64), you will also have to install some extra armhf libraries.
+ - 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 also just make a symlink, but box86 only works for 32bit apps at the moment
+sudo ln -s ~/wine/bin/wineboot /usr/local/bin/wineboot
+sudo ln -s ~/wine/bin/winecfg /usr/local/bin/winecfg
+sudo ln -s ~/wine/bin/wineserver /usr/local/bin/wineserver
+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 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/branch of Wine that you wish.
+```
+### User-defined Wine version variables ################
+# - Replace the variables below with your system's info.
+# - Note that we need the i386 version for Box86 even though we're installing it on our ARM processor.
+# - Wine download links from WineHQ: https://dl.winehq.org/wine-builds/
+
+wbranch="devel" #example: devel, staging, or stable (wine-staging 4.5+ requires libfaudio0:i386 - see below)
+wversion="7.1" #example: 7.1
+wid="debian" #example: debian, ubuntu
+wdist="bullseye" #example (for debian): bullseye, buster, jessie, wheezy, etc
+wtag="-1" #example: -1 (some wine .deb files have -1 tag on the end and some don't)
+
+########################################################
+
+# Clean up any old wine instances
+wineserver -k # stop any old wine installations from running
+rm -rf ~/.cache/wine # remove old wine-mono/wine-gecko install files
+rm -rf ~/.local/share/applications/wine # remove old program shortcuts
+
+# 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
+cd ~/Downloads
+wget https://dl.winehq.org/wine-builds/${wid}/dists/${wdist}/main/binary-i386/wine-${wbranch}-i386_${wversion}~${wdist}${wtag}_i386.deb # download
+wget https://dl.winehq.org/wine-builds/${wid}/dists/${wdist}/main/binary-i386/wine-${wbranch}_${wversion}~${wdist}${wtag}_i386.deb # (required for wine_i386 if no wine64 / CONFLICTS WITH wine64 support files)
+dpkg-deb -x wine-${wbranch}-i386_${wversion}~${wdist}${wtag}_i386.deb wine-installer # extract
+dpkg-deb -x wine-${wbranch}_${wversion}~${wdist}${wtag}_i386.deb wine-installer
+mv wine-installer/opt/wine* ~/wine # install
+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 on a 64-bit RPiOS via multiarch
+karch=$(uname -m)
+if [ "$karch" = "aarch64" ] || [ "$karch" = "aarch64-linux-gnu" ] || [ "$karch" = "arm64" ] || [ "$karch" = "aarch64_be" ]; then
+    sudo dpkg --add-architecture armhf && sudo apt-get update # enable multi-arch
+    sudo apt-get install -y libasound2:armhf libc6:armhf libglib2.0-0:armhf libgphoto2-6:armhf libgphoto2-port12:armhf \
+        libgstreamer-plugins-base1.0-0:armhf libgstreamer1.0-0:armhf libldap-2.4-2:armhf libopenal1:armhf libpcap0.8:armhf \
+        libpulse0:armhf libsane1:armhf libudev1:armhf libusb-1.0-0:armhf libvkd3d1:armhf libx11-6:armhf libxext6:armhf \
+        libasound2-plugins:armhf ocl-icd-libopencl1:armhf libncurses6:armhf libncurses5:armhf libcap2-bin:armhf libcups2:armhf \
+        libdbus-1-3:armhf libfontconfig1:armhf libfreetype6:armhf libglu1-mesa:armhf libglu1:armhf libgnutls30:armhf \
+        libgssapi-krb5-2:armhf libkrb5-3:armhf libodbc1:armhf libosmesa6:armhf libsdl2-2.0-0:armhf libv4l-0:armhf \
+        libxcomposite1:armhf libxcursor1:armhf libxfixes3:armhf libxi6:armhf libxinerama1:armhf libxrandr2:armhf \
+        libxrender1:armhf libxxf86vm1 libc6:armhf libcap2-bin:armhf # to run wine-i386 through box86:armhf on aarch64
+        # This list found by downloading...
+        #	wget https://dl.winehq.org/wine-builds/debian/dists/bullseye/main/binary-i386/wine-devel-i386_7.1~bullseye-1_i386.deb
+        #	wget https://dl.winehq.org/wine-builds/debian/dists/bullseye/main/binary-i386/winehq-devel_7.1~bullseye-1_i386.deb
+        #	wget https://dl.winehq.org/wine-builds/debian/dists/bullseye/main/binary-i386/wine-devel_7.1~bullseye-1_i386.deb
+        # then `dpkg-deb -I package.deb`. Read output, add `:armhf` to packages in dep list, then try installing them on Pi aarch64.
+fi
+
+# These packages are needed for running wine-staging on RPiOS (Credits: chills340)
+sudo apt install libstb0 -y
+cd ~/Downloads
+wget -r -l1 -np -nd -A "libfaudio0_*~bpo10+1_i386.deb" http://ftp.us.debian.org/debian/pool/main/f/faudio/ # Download libfaudio i386 no matter its version number
+dpkg-deb -xv libfaudio0_*~bpo10+1_i386.deb libfaudio
+sudo cp -TRv libfaudio/usr/ /usr/
+rm libfaudio0_*~bpo10+1_i386.deb # clean up
+rm -rf libfaudio # clean up
+
+# 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.

docs/_config.yml

@@ -0,0 +1 @@
+theme: jekyll-theme-hacker

install_steam.sh

@@ -0,0 +1,49 @@
+#!/bin/bash
+
+# create necessary directories
+mkdir -p ~/steam
+mkdir -p ~/steam/tmp
+cd ~/steam/tmp
+
+# download latest deb and unpack
+wget https://cdn.cloudflare.steamstatic.com/client/installer/steam.deb
+ar x steam.deb
+tar xf data.tar.xz
+
+# remove deb archives, not needed anymore
+rm ./*.tar.xz ./steam.deb
+
+# move deb contents to steam folder
+mv ./usr/* ../
+cd ../ && rm -rf ./tmp/
+
+# create run script
+echo "#!/bin/bash
+export STEAMOS=1
+export STEAM_RUNTIME=1
+export DBUS_FATAL_WARNINGS=0
+~/steam/bin/steam steam://open/minigameslist $@" > steam
+
+# make script executable and move
+chmod +x steam
+sudo mv steam /usr/local/bin/
+
+# detect if we're running on 64 bit Debian (maybe this can be moved in another script then mentioned in documentation for the people that don't use steam but don't want a headache)
+MACHINE_TYPE=`uname -m`
+if [ ${MACHINE_TYPE} == 'aarch64' ] && [ -f '/etc/debian_version' ]; then
+ echo "Detected 64 bit ARM Debian. Installing 32 bit libraries"
+ sudo dpkg --add-architecture armhf # enable installation of armhf libraries
+ sudo apt update # update package lists with the newly added arch
+ # install the libraries that Steam requires
+ sudo apt install libc6:armhf libncurses5:armhf libsdl2-2.0-0:armhf libsdl2-image-2.0-0:armhf libsdl2-mixer-2.0-0:armhf libsdl2-ttf-2.0-0:armhf libopenal1:armhf libpng16-16:armhf libfontconfig1:armhf libxcomposite1:armhf libbz2-1.0:armhf libxtst6:armhf libsm6:armhf libice6:armhf libgl1:armhf libxinerama1:armhf libxdamage1:armhf
+
+ # install mesa for armhf if already installed
+ if dpkg-query -W libgl1-mesa-dri 2>/dev/null; then
+  echo "Detected mesa driver for 64 bit ARM. Installing complementary 32 bit one"
+  sudo apt install libgl1-mesa-dri:armhf
+ fi
+
+ echo "Don't forget to compile/install Box64!"
+fi
+
+echo "Script complete."

pkgbuilds/PKGBUILD-a64

@@ -0,0 +1,36 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-a64-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-a64-git}-git" "${pkgname%-a64-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-a64-git}-git" "${pkgname%-a64-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "${srcdir}/${pkgname%-a64-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-a64-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DA64=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-a64-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-a64-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-arm

@@ -0,0 +1,36 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-arm-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-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}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-arm-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-arm-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-arm64

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

pkgbuilds/PKGBUILD-gameshell

@@ -0,0 +1,36 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-gameshell-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-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}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DGAMESHELL=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-gameshell-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-gameshell-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-goaclone

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

pkgbuilds/PKGBUILD-lx2610a

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

pkgbuilds/PKGBUILD-odroid

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

pkgbuilds/PKGBUILD-odroidn2

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

pkgbuilds/PKGBUILD-pandora

@@ -0,0 +1,36 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-pandora-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-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}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DPANDORA=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-pandora-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-pandora-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-phytium

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

pkgbuilds/PKGBUILD-pyra

@@ -0,0 +1,36 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-pyra-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-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}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DPYRA=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-pyra-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-pyra-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-raspberrypi

@@ -0,0 +1,51 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-rpi-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-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 [[ ${_pimodel} =~ "Raspberry Pi 2" ]]; then
+		_piversion=2
+	else
+		# This RPI is not supported or this device is not a pi.
+		# We should cancel then the build.
+		echo -e "\033[01m\033[01;31m==> ERROR:\033[00;01m Your device is not supported! This script is for RPi model 4/3/2 (1/0 isn't supported).\033[00m"
+		exit 1
+	fi
+	echo -e "\033[01m\033[01;34m==> INFO:\033[00;01m Your detected Pi verion number is: \033[01;31m${_piversion}\033[01m.\033[00m"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DRPI${_piversion}=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-rpi-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-rpi-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-rk3xxx

@@ -0,0 +1,50 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-rk3xxx-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-rk3xxx-git}-git" "${pkgname%-rk3xxx-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-rk3xxx-git}-git" "${pkgname%-rk3xxx-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "$srcdir/${pkgname%-rk3xxx-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-rk3xxx-git}"
+	# Check for RockChip version number
+	if [[ ${_rkmodel} =~ "rk3288"]]; then
+		_rksuffix=3288
+	elif [[ ${_rkmodel} =~ "rk3399" ]]; then
+		_rksuffix=3399	
+	elif [[ ${_rkmodel} =~ "rk3588" ]]; then
+		_rksuffix=3588	
+	else
+		# This RKxxxx is not supported or this device is not RockChip based.
+		# We should cancel then the build.
+		echo -e "\033[01m\033[01;31m==> ERROR:\033[00;01m Your device is not supported! This script is for RockChip model 3399/3588.\033[00m"
+		exit 1
+	fi
+	echo -e "\033[01m\033[01;34m==> INFO:\033[00;01m Your detected RockChip verion number is: \033[01;31m${_rksuffix}\033[01m.\033[00m"
+	[[ ! -d ./build ]] && mkdir build
+    cmake -B build -DRK${_rksuffix}=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-rk3xxx-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-rk3xxx-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-sd845

@@ -0,0 +1,36 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-sd845-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-sd845-git}-git" "${pkgname%-sd845-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-sd845-git}-git" "${pkgname%-sd845-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "${srcdir}/${pkgname%-sd845-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-sd845-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DSD845=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-sd845-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-sd845-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-tegra

@@ -0,0 +1,36 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-tegra-git
+pkgver=r4581.d698861
+pkgrel=1
+pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
+arch=('armv7h')
+url="https://github.com/ptitSeb/box86"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-tegra-git}"-git "${pkgname%-tegra-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-tegra-git}"-git "${pkgname%-tegra-git}")
+source=('git+https://github.com/ptitSeb/box86')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "${srcdir}/${pkgname%-tegra-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-tegra-git}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DTEGRAX1ARM32=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-tegra-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-tegra-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-x86

@@ -0,0 +1,35 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-x86-git
+pkgver=r4581.d698861
+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}"
+	[[ ! -d ./build ]] && mkdir build
+	cmake -B build -DLD80BITS=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+}
+
+build() {
+	cd "${srcdir}/${pkgname%-x86-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "${srcdir}/${pkgname%-x86-git}/build"
+	make DESTDIR="${pkgdir}/" install
+}

pkgbuilds/README.md

@@ -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.
+Most specified in [COMPILE.md](../docs/COMPILE.md). Except PowerPC based.
+
+### 3. Any advantages with that sollution?
+This might give you some advantages like those:
+- 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.

postinst

@@ -0,0 +1,4 @@
+#!/bin/sh
+if [ -d '/run/systemd/system' ]; then
+    systemctl restart systemd-binfmt || :
+fi

rebuild_printer.py

@@ -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):
@@ -99,7 +106,7 @@ def main(root, ver, __debug_forceAllDebugging=False):
 					alreadyChanged = 0
 					csp = 0
 					while csp < curSplt:
-						if spltln[csp] == '\x00':
+						if spltln[csp] == '\x01':
 							alreadyChanged = 1
 							curSplt = curSplt + 1
 						else:
@@ -112,7 +119,7 @@ def main(root, ver, __debug_forceAllDebugging=False):
 								line = line + spltln[csp] + ">"
 								alreadyChanged = 1
 						csp = csp + 1
-					if spltln[csp] == '\x00':
+					if spltln[csp] == '\x01':
 						alreadyChanged = 1
 						curSplt = curSplt + 1
 						csp = csp + 1
@@ -127,7 +134,7 @@ def main(root, ver, __debug_forceAllDebugging=False):
 						alreadyChanged = 1
 					csp = csp + 1
 					while csp < len(spltln):
-						if spltln[csp] == '\x00':
+						if spltln[csp] == '\x01':
 							alreadyChanged = 1
 							curSplt = curSplt + 1
 						else:
@@ -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 != []:
@@ -604,7 +605,7 @@ def main(root, ver, __debug_forceAllDebugging=False):
 				instText = [itp for itp2 in instText for itp in itp2]
 				
 				# Make the failures nicer
-				spltln = spltln[:curSplt] + ["\x00"] + instText
+				spltln = spltln[:curSplt] + ["\x01"] + instText
 				
 				printf_args = ""
 				for idx, text in enumerate(instText):
@@ -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":
@@ -857,7 +858,7 @@ def main(root, ver, __debug_forceAllDebugging=False):
 				instText = [itp for itp2 in instText for itp in itp2]
 				
 				# Make the failures nicer
-				spltln = spltln[:curSplt] + ["\x00"] + instText
+				spltln = spltln[:curSplt] + ["\x01"] + instText
 				
 				printf_args = ""
 				for idx, text in enumerate(instText):
@@ -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")
@@ -1155,7 +1156,7 @@ def main(root, ver, __debug_forceAllDebugging=False):
 				instText = [itp for itp2 in instText for itp in itp2]
 				
 				# Make the failures nicer
-				spltln = spltln[:curSplt] + ["\x00"] + instText
+				spltln = spltln[:curSplt] + ["\x01"] + instText
 				
 				printf_args = ""
 				for idx, text in enumerate(instText):
@@ -1455,7 +1456,7 @@ def main(root, ver, __debug_forceAllDebugging=False):
 				instText = [itp for itp2 in instText for itp in itp2]
 				
 				# Make the failures nicer
-				spltln = spltln[:curSplt] + ["\x00"] + instText
+				spltln = spltln[:curSplt] + ["\x01"] + instText
 				
 				# inside '{, '...'}'?
 				hasOpened = False
@@ -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 +1675,301 @@ 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';
+	header = """
+	#include <stdint.h>
+	#include <stddef.h>
+	#include <string.h>
+	#include <stdio.h>
 	
-	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;
-			}
+	#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 {
-			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;
+			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 {
-				if (last == -1)
-					strcat(ret, ", ");
-				strcat(ret, regname[i]);
-				last = i;
+				sprintf(ret, "%s%s#%u", (comma ? ", " : ""), shift_type[sh_op], amount);
 			}
-		} else {
-			if (cnt) {
-				strcat(ret, "-");
-				strcat(ret, regname[i - 1]);
-			}
-			if (last != -2) last = -1;
-			cnt = 0;
 		}
+		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];
+	#define print_modified_imm_ARM(imm12) (((imm12 & 0xFF) >> (2*(imm12 >> 8))) | ((imm12 & 0xFF) << (32 - 2*(imm12 >> 8))))
 	
-	// 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 {
+	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;
+			}
+		}
+		if (cnt) {
+			strcat(ret, "-");
+			strcat(ret, regname[size - 1]);
+		}
+		strcat(ret, "}");
+		
+		return ret;
 	}
 	
-	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);
+	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[7];
+		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;
 	}
-	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]
+	static const char* _print_modified_imm_I64_bytes[2] = {"00", "FF"};
 	
-	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) {
+	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%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) {
+			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%02X00", imm);
+			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;
 		}
-		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;
 	}
 	
-	return ret;
-}
-
-const char* arm_print(uint32_t opcode) {
-	static __thread char ret[100];
-	memset(ret, 0, sizeof(ret));
-	
+	const char* arm_print(uint32_t opcode) {
+		static __thread char ret[100];
+		memset(ret, 0, sizeof(ret));
+		
+		"""
+	footer = """
+	{
+			strcpy(ret, "???");
+		}
+		
+		return ret;
+	}
 	"""
-	}
-	files_guard = {"arm_printer.c": """{
-		strcpy(ret, "???");
-	}
+	banner = "/*******************************************************" + ('*'*len(ver)) + "***\n" \
+	         " * File automatically generated by rebuild_printer.py (v" + ver + ") *\n" \
+	         " *******************************************************" + ('*'*len(ver)) + "***/\n"
+	trim = lambda string: '\n'.join(line[1:] for line in string.splitlines())[1:]
+	header = banner + trim(header)
+	footer = trim(footer)
 	
-	return ret;
-}
-"""}
-	
-	for f in files_header:
+	with open(os.path.join(root, "src", "dynarec", "arm_printer.c"), 'w') as file:
+		file.write(header)
+		file.write(output)
+		file.write(footer)
 	# 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])
 	with open(os.path.join(root, "src", "dynarec", "last_run.txt"), 'w') as file:
 		file.write('\n'.join(insts))
 	
@@ -1979,6 +1980,6 @@ 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], "1.2.0.06") != 0:
 		exit(2)
 	exit(0)

runTest.cmake

@@ -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)
@@ -31,19 +32,30 @@ 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 )
 
+# remove the temporary files if they exist
+if( EXISTS ${TEST_OUTPUT} )
+  file(REMOVE "${TEST_OUTPUT}")
+endif()
+
 # everything went fine...
 message( "Passed: The output of ${TEST_PROGRAM} matches ${TEST_REFERENCE}" )

src/box86context.c

@@ -19,201 +19,18 @@
 #include "threads.h"
 #include "x86trace.h"
 #include "signals.h"
-#ifdef DYNAREC
 #include <sys/mman.h>
+#include "custommem.h"
+#include "dictionnary.h"
+#include "rcfile.h"
+#include "gltools.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 +39,22 @@ 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();
+    DeleteParams();
+    fini_pthread_helper(context);
     fini_signal_helper();
+    fini_bridge_helper();
+    fini_custommem_helper(context);
     finied = 1;
 }
 
@@ -244,11 +65,11 @@ 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)
@@ -256,8 +77,122 @@ void free_tlsdatasize(void* p)
     if(!p)
         return;
     tlsdatasize_t *data = (tlsdatasize_t*)p;
-    free(data->tlsdata);
-    free(p);
+    box_free(data->ptr);
+    box_free(p);
+}
+
+int unlockMutex()
+{
+    int ret = unlockCustommemMutex();
+    int i;
+    #ifdef DYNAREC
+    void* tid = (void*)GetTID();
+    #define GO(A, B)                    \
+        i = (arm_lock_storeifref2(&A, NULL, tid)==tid);  \
+        if(i) {                         \
+            ret|=(1<<B);                \
+        }
+    #else
+    #define GO(A, B)                    \
+        i = checkUnlockMutex(&A);       \
+        if(i) {                         \
+            ret|=(1<<B);                \
+        }
+    #endif
+
+    GO(my_context->mutex_once, 5)
+    GO(my_context->mutex_once2, 6)
+    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))                \
+            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)
+{
+
+#ifdef DYNAREC
+    arm_lock_stored(&context->mutex_dyndump, 0);
+    arm_lock_stored(&context->mutex_once, 0);
+    arm_lock_stored(&context->mutex_once2, 0);
+    arm_lock_stored(&context->mutex_trace, 0);
+    arm_lock_stored(&context->mutex_tls, 0);
+    arm_lock_stored(&context->mutex_thread, 0);
+    arm_lock_stored(&context->mutex_bridge, 0);
+    pthread_mutex_init(&context->mutex_lock, NULL);
+#else
+    pthread_mutexattr_t attr;
+    pthread_mutexattr_init(&attr);
+    pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_ERRORCHECK);
+    pthread_mutex_init(&context->mutex_once, &attr);
+    pthread_mutex_init(&context->mutex_once2, &attr);
+    pthread_mutex_init(&context->mutex_trace, &attr);
+    pthread_mutex_init(&context->mutex_tls, &attr);
+    pthread_mutex_init(&context->mutex_thread, &attr);
+    pthread_mutex_init(&context->mutex_bridge, &attr);
+    pthread_mutex_init(&context->mutex_lock, &attr);
+
+    pthread_mutexattr_destroy(&attr);
+#endif
+}
+
+static void atfork_child_box64context(void)
+{
+    // (re)init mutex if it was lock before the fork
+    init_mutexes(my_context);
+}
+
+void freeCycleLog(box86context_t* ctx)
+{
+    if(cycle_log) {
+        for(int i=0; i<cycle_log; ++i) {
+            box_free(ctx->log_call[i]);
+            box_free(ctx->log_ret[i]);
+        }
+        box_free(ctx->log_call);
+        box_free(ctx->log_ret);
+        ctx->log_call = NULL;
+        ctx->log_ret = NULL;
+    }
+}
+void initCycleLog(box86context_t* context)
+{
+    if(cycle_log) {
+        context->log_call = (char**)box_calloc(cycle_log, sizeof(char*));
+        context->log_ret = (char**)box_calloc(cycle_log, sizeof(char*));
+        for(int i=0; i<cycle_log; ++i) {
+            context->log_call[i] = (char*)box_calloc(256, 1);
+            context->log_ret[i] = (char*)box_calloc(128, 1);
+        }
+    }
 }
 
 EXPORTDYN
@@ -270,18 +205,29 @@ 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*)box_calloc(1, sizeof(box86context_t));
+
+    initCycleLog(context);
 
 #ifdef BUILD_LIB
     context->deferedInit = 0;
 #else
-    context->deferedInit = 1;
+    context->deferredInit = 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();
+    // Cannot use Bridge name as the map is not initialized yet
     // 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);
+    // create exit bridge
+    context->exit_bridge = AddBridge(context->system, NULL, NULL, 0, NULL);
 #ifdef BUILD_LIB
     context->box86lib = RTLD_DEFAULT;   // not ideal
 #else
@@ -289,35 +235,27 @@ box86context_t *NewBox86Context(int argc)
 #endif
     context->dlprivate = NewDLPrivate();
 
-    context->callbacks = NewCallbackList();
-
     context->argc = argc;
-    context->argv = (char**)calloc(context->argc+1, sizeof(char*));
+    context->argv = (char**)box_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);
 
+    context->globdata = NewMapSymbols();
+
     initAllHelpers(context);
 
     return context;
 }
-
+void freeALProcWrapper(box86context_t* context);
 EXPORTDYN
 void FreeBox86Context(box86context_t** context)
 {
@@ -333,36 +271,19 @@ 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]);
     }
-    free(ctx->elfs);
+    box_free(ctx->elfs);
 
     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);
@@ -372,61 +293,17 @@ void FreeBox86Context(box86context_t** context)
     if(ctx->zydis)
         DeleteX86Trace(ctx);
 
-    if(ctx->deferedInitList)
-        free(ctx->deferedInitList);
+    if(ctx->deferredInitList)
+        box_free(ctx->deferredInitList);
 
-#ifndef BUILD_LIB
-    if(ctx->box86lib)
-        dlclose(ctx->box86lib);
-#endif
-
-    FreeDLPrivate(&ctx->dlprivate);
-
-    free(ctx->argv);
+    /*box_free(ctx->argv);*/
     
-    for (int i=0; i<ctx->envc; ++i)
-        free(ctx->envv[i]);
-    free(ctx->envv);
-
-    free(ctx->stack);
-
-    free(ctx->fullpath);
-    free(ctx->box86path);
-
-    FreeBridge(&ctx->system);
-
-    freeGLProcWrapper(ctx);
-    freeALProcWrapper(ctx);
-
-    FreeCallbackList(&ctx->callbacks);
-
-    void* ptr;
-    if ((ptr = pthread_getspecific(ctx->tlskey)) != NULL) {
-        free_tlsdatasize(ptr);
-        pthread_setspecific(ctx->tlskey, NULL);
-    }
-    pthread_key_delete(ctx->tlskey);
-
-    if(ctx->tlsdata)
-        free(ctx->tlsdata);
-
-    for(int i=0; i<3; ++i) {
-        if(ctx->segtls[i].present) {
-            // key content not handled by box86, so not doing anything with it
-            pthread_key_delete(ctx->segtls[i].key);
-        }
-    }
-
-    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);
+    /*for (int i=0; i<ctx->envc; ++i)
+        box_free(ctx->envv[i]);
+    box_free(ctx->envv);*/
 
     if(ctx->atfork_sz) {
-        free(ctx->atforks);
+        box_free(ctx->atforks);
         ctx->atforks = NULL;
         ctx->atfork_sz = ctx->atfork_cap = 0;
     }
@@ -436,54 +313,125 @@ void FreeBox86Context(box86context_t** context)
             signal(i, SIG_DFL);
         }
 
+    *context = NULL;                // bye bye my_context
+
+    CleanStackSize(ctx);
+
+#ifndef BUILD_LIB
+    if(ctx->box86lib)
+        dlclose(ctx->box86lib);
+#endif
+
+    FreeDLPrivate(&ctx->dlprivate);
+
+//    box_free(ctx->stack); // don't free the stack, it's owned by main emu!
+
+    box_free(ctx->fullpath);
+    box_free(ctx->box86path);
+    box_free(ctx->bashpath);
+
+    FreeBridge(&ctx->system);
+
+    freeGLProcWrapper(ctx);
+    freeALProcWrapper(ctx);
+
+    if(ctx->stack_clone)
+        box_free(ctx->stack_clone);
+
+    void* ptr;
+    if ((ptr = pthread_getspecific(ctx->tlskey)) != NULL) {
+        free_tlsdatasize(ptr);
+        pthread_setspecific(ctx->tlskey, NULL);
+    }
+    pthread_key_delete(ctx->tlskey);
+
+    if(ctx->tlsdata)
+        box_free(ctx->tlsdata);
+
+    for(int i=0; i<3; ++i) {
+        if(ctx->segtls[i].present) {
+            // key content not handled by box86, so not doing anything with it
+            pthread_key_delete(ctx->segtls[i].key);
+        }
+    }
+
+    free_neededlib(ctx->neededlibs);
+    ctx->neededlibs = NULL;
+
     if(ctx->emu_sig)
         FreeX86Emu(&ctx->emu_sig);
 
-    finiAllHelpers();
+    FreeMapSymbols(&ctx->globdata);
 
-    free(ctx);
+    finiAllHelpers(ctx);
+
+#ifdef DYNAREC
+    pthread_mutex_destroy(&ctx->mutex_lock);
+#else
+    pthread_mutex_destroy(&ctx->mutex_once);
+    pthread_mutex_destroy(&ctx->mutex_once2);
+    pthread_mutex_destroy(&ctx->mutex_trace);
+    pthread_mutex_destroy(&ctx->mutex_tls);
+    pthread_mutex_destroy(&ctx->mutex_thread);
+    pthread_mutex_destroy(&ctx->mutex_bridge);
+    pthread_mutex_destroy(&ctx->mutex_lock);
+#endif
+
+    freeCycleLog(ctx);
+
+    box_free(ctx);
 }
 
 int AddElfHeader(box86context_t* ctx, elfheader_t* head) {
-    int idx = ctx->elfsize;
-    if(idx==ctx->elfcap) {
-        // resize...
-        ctx->elfcap += 16;
-        ctx->elfs = (elfheader_t**)realloc(ctx->elfs, sizeof(elfheader_t*) * ctx->elfcap);
+    int idx = 0;
+    while(idx<ctx->elfsize && ctx->elfs[idx]) idx++;
+    if(idx == ctx->elfsize) {
+        if(idx==ctx->elfcap) {
+            // resize...
+            ctx->elfcap += 16;
+            ctx->elfs = (elfheader_t**)box_realloc(ctx->elfs, sizeof(elfheader_t*) * ctx->elfcap);
+        }
+        ctx->elfs[idx] = head;
+        ctx->elfsize++;
+    } else {
+        ctx->elfs[idx] = head;
     }
-    ctx->elfs[idx] = head;
-    ctx->elfsize++;
     printf_log(LOG_DEBUG, "Adding \"%s\" as #%d in elf collection\n", ElfName(head), idx);
     return idx;
 }
 
+void RemoveElfHeader(box86context_t* ctx, elfheader_t* head) {
+    if(GetTLSBase(head)) {
+        // should remove the tls info
+        int tlsbase = GetTLSBase(head);
+        /*if(tlsbase == -ctx->tlssize) {
+            // not really correct, but will do for now
+            ctx->tlssize -= GetTLSSize(head);
+            if(!(++ctx->sel_serial))
+                ++ctx->sel_serial;
+        }*/
+    }
+    for(int i=0; i<ctx->elfsize; ++i)
+        if(ctx->elfs[i] == head) {
+            ctx->elfs[i] = NULL;
+            return;
+        }
+}
+
+
 int AddTLSPartition(box86context_t* context, int tlssize) {
     int oldsize = context->tlssize;
+    // should in fact first try to map a hole, but rewinding all elfs and checking filled space, like with the mapmem utilities
     context->tlssize += tlssize;
-    context->tlsdata = realloc(context->tlsdata, context->tlssize);
+    context->tlsdata = box_realloc(context->tlsdata, context->tlssize);
     memmove(context->tlsdata+tlssize, context->tlsdata, oldsize);   // move to the top, using memmove as regions will probably overlap
     memset(context->tlsdata, 0, tlssize);           // fill new space with 0 (not mandatory)
+    // clean GS segment for current emu
+    if(my_context) {
+        ResetSegmentsCache(thread_get_emu());
+        if(!(++context->sel_serial))
+            ++context->sel_serial;
+    }
+
     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;
-}

src/box86version.h

@@ -2,7 +2,7 @@
 #define __BOX86_VERSION_H_
 
 #define BOX86_MAJOR 0
-#define BOX86_MINOR 1
+#define BOX86_MINOR 3
 #define BOX86_REVISION 2
 
 #endif //__BOX86_VERSION_H_

src/build_info.c

@@ -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,

src/dynarec/arm_emitter.h

@@ -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
@@ -65,7 +66,7 @@ Op is 20-27
 #define c__ (0b1110<<28)    // means all
 
 //  nop
-#define NOP     EMIT(0xe1a00000)
+#define NOP     EMIT(0xe320f000)
 
 // mov dst, src
 #define MOV_REG(dst, src) EMIT(0xe1a00000 | ((dst) << 12) | (src) )
@@ -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) )
@@ -187,6 +213,9 @@ Op is 20-27
 // add dst, src1, src2, lsl #imm
 #define ADD_REG_LSL_IMM5(dst, src1, src2, imm5) \
     EMIT(0xe0800000 | ((dst) << 12) | ((src1) << 16) | brLSL(imm5, src2) )
+// add.c dst, src1, src2, lsl #imm
+#define ADD_REG_LSL_IMM5_COND(cond, dst, src1, src2, imm5) \
+    EMIT((cond) | 0x00800000 | ((dst) << 12) | ((src1) << 16) | brLSL(imm5, src2) )
 // add.s dst, src1, src2, lsl #imm
 #define ADDS_REG_LSL_IMM5(dst, src1, src2, imm5) \
     EMIT(0xe0900000 | ((dst) << 12) | ((src1) << 16) | brLSL(imm5, src2) )
@@ -207,9 +236,21 @@ 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.cond dst, src, #imm
+#define CMNS_IMM8_COND(cond, src, imm8) \
+    EMIT((cond) | 0x03700000 | ((0) << 12) | ((src) << 16) | brIMM(imm8) )
+// cmn.s dst, src, #imm
+#define CMNS_IMM8(src, imm8) \
+    EMIT(c__ | 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) )
@@ -219,21 +260,48 @@ Op is 20-27
 // tst.s dst, src1, #imm ror rot*2
 #define TSTS_IMM8_ROR(src, imm8, rot) \
     EMIT(0xe3100000 | ((0) << 12) | ((src) << 16) | ((rot)<<8) | brIMM(imm8) )
+// tst.s.cond dst, src1, #imm ror rot*2
+#define TSTS_IMM8_ROR_COND(cond, src, imm8, rot) \
+    EMIT((cond) | 0x03100000 | ((0) << 12) | ((src) << 16) | ((rot)<<8) | brIMM(imm8) )
 // orr dst, src1, src2, lsl #imm
 #define ORR_REG_LSL_IMM5(dst, src1, src2, imm5) \
     EMIT(0xe1800000 | ((dst) << 12) | ((src1) << 16) | brLSL(imm5, src2) )
+// orr.cond dst, src1, src2, lsl #imm
+#define ORR_REG_LSL_IMM5_COND(cond, dst, src1, src2, imm5) \
+    EMIT(cond | 0x01800000 | ((dst) << 12) | ((src1) << 16) | brLSL(imm5, src2) )
 // 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.cond dst, src1, src2, lsr #imm
+#define ORR_REG_LSR_IMM5_COND(cond, dst, src1, src2, imm5) \
+    EMIT(cond | 0x01800000 | ((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.s dst, src1, src2, lsr rs
+#define ORRS_REG_LSR_REG(dst, src1, src2, rs) \
+    EMIT(0xe1900000 | ((dst) << 12) | ((src1) << 16) | brRLSR(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) )
+// orr dst, src1, src2, lsr rs
+#define ORR_REG_LSR_REG(dst, src1, src2, rs) \
+    EMIT(0xe1800000 | ((dst) << 12) | ((src1) << 16) | brRLSR(rs, src2) )
 // xor dst, src1, src2, lsl #imm
 #define XOR_REG_LSL_IMM5(dst, src1, src2, imm5) \
     EMIT(0xe0200000 | ((dst) << 12) | ((src1) << 16) | brLSL(imm5, src2) )
@@ -247,6 +315,9 @@ 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 +327,21 @@ 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 +352,8 @@ 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 +364,6 @@ 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]
@@ -313,9 +390,15 @@ Op is 20-27
 #define LDRD_IMM8(reg, addr, imm8) EMIT(c__ | 0b000<<25 | 1<<24  | (((imm8)<0)?0:1)<<23 | 1<<22 | 0<<21 | 0<<20 | ((reg) << 12) | ((addr) << 16) | ((abs(imm8))&0xf0)<<(8-4) | (0b1101<<4) | ((abs(imm8))&0x0f) )
 // ldrd reg, reg+1, [addr, +rm], reg must be even, reg+1 is implicit
 #define LDRD_REG(reg, addr, rm) EMIT(c__ | 0b000<<25 | 1<<24  | 1<<23 | 0<<22 | 0<<21 | 0<<20 | ((reg) << 12) | ((addr) << 16) | 0b1101<<4 | (rm) )
+// ldr reg, [pc+imm]
+#define LDR_literal(reg, imm12) EMIT(c__ | 0b010<<25 | 1<<24 | (((imm12)>=0)?1:0)<<23 | 0<<21 | 1<<20 | 0b1111<<16 | (reg)<<12 | (abs(imm12)))
 
 // str reg, [addr, #+/-imm9]
 #define STR_IMM9(reg, addr, imm9) EMIT(0xe5000000 | (((imm9)<0)?0:1)<<23 | ((reg) << 12) | ((addr) << 16) | brIMM(imm9) )
+// 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 +436,16 @@ 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))
@@ -377,13 +466,18 @@ Op is 20-27
 #define LDRH_IMM8(reg, addr, imm8) EMIT(HWS_OFF(c__, 1,  (((imm8)<0)?0:1), 0, 1, addr, reg, 0, 1, abs(imm8)))
 #define LDRH_IMM8_COND(cond, reg, addr, imm8) EMIT(HWS_OFF(cond, 1, (((imm8)<0)?0:1), 0, 1, addr, reg, 0, 1, abs(imm8)))
 #define STRH_IMM8(reg, addr, imm8) EMIT(HWS_OFF(c__, 1, (((imm8)<0)?0:1), 0, 0, addr, reg, 0, 1, abs(imm8)))
-#define STRSH_IMM8(reg, addr, imm8) EMIT(HWS_OFF(c__, 1, (((imm8)<0)?0:1), 0, 0, addr, reg, 1, 1, abs(imm8)))
 
 #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))
@@ -394,6 +488,11 @@ Op is 20-27
 // Mul and MulA
 #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 MLA(Rd, Rn, Rm, Ra) EMIT(MULMULA(c__, 1, 0, (Rd), (Ra), (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
@@ -410,18 +509,32 @@ 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))
+// UBFX: Unsigned Bit Field Extract: extract any number of bits from Rn, zero extend and put in Rd
+#define UBFX_COND(cond, rd, rn, lsb, width)    EMIT(cond | (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 +551,43 @@ 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))
+
+#define LDREXH_gen(cond, Rn, Rt)        (cond | 0b0001111<<21 | 1<<20 | (Rn)<<16 | (Rt)<<12 | 0b1111<<8 | 0b1001<<4 | 0b1111)
+// Load Exclusive Half-Word Rt from Rn (tagging the memory)
+#define LDREXH(Rt, Rn)                  EMIT(LDREXH_gen(c__, Rn, Rt))
+#define LDREXH_COND(cond, Rt, Rn)       EMIT(LDREXH_gen(cond, Rn, Rt))
+
+#define STREXH_gen(cond, Rd, Rn, Rt)    (cond | 0b0001111<<21 | 0<<20 | (Rn)<<16 | (Rd)<<12 | 0b1111<<8 | 0b1001<<4 | (Rt))
+// Store Exclusive Half-Word Rt to Rn, with result in Rd=0 if tag is ok, Rd==1 if store failed (Rd!=Rn && Rd!=Rt)
+#define STREXH(Rd, Rt, Rn)              EMIT(STREXH_gen(c__, Rd, Rn, Rt))
+#define STREXH_COND(cond, Rd, Rt, Rn)   EMIT(STREXH_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 +600,10 @@ 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 +612,17 @@ 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 +656,8 @@ 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]
@@ -535,6 +689,10 @@ Op is 20-27
 // Store to memory to single  VSTR Sd, [Rn, #+/-imm8], imm8&3 ignored!
 #define VSTR_32(Sd, Rn, Imm8)    EMIT(c__ | (0b1101<<24) | (((Imm8)<0)?0:1)<<23 | (((Sd)&1)<<22) | (0<<20) | ((Rn)<<16) | ((((Sd)>>1)&15)<<12) | (0b1010<<8) | ((abs(Imm8)>>2)&255))
 
+#define VSTM_64_gen(cond, P, U, D, W, Rn, Vd, imm8)	(cond | 0b110<<25 | (P)<<24 | (U)<<23 | (D)<<22 | (W)<<21 | (Rn)<<16 | (Vd)<<12 | 0b1011<<8 | (imm8))
+// Store a sery of Dx, post imcrementing Rn
+#define VSTM_64_W(Dd, Rn)	    EMIT(VSTM_64_gen(c__, 0, 1, ((Dd)>>4)&1, 1, Rn, (Dd)&15, 2))
+
 // Convert from single Sm to double Dd
 #define VCVT_F64_F32(Dd, Sm)    EMIT(c__ | (0b1110<<24) | (1<<23) |  ((((Dd)>>4)&1)<<22) | (0b11<<20) | (0b0111<<16) | (((Dd)&15)<<12) | (0b101<<9) | (0<<8) | (0b11<<6) | (((Sm)&1)<<5) | (0<<4) | (((Sm)>>1)&15))
 // Convert from double Dm to single Sd
@@ -542,10 +700,14 @@ 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 +715,45 @@ 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))
+
+#define VRINT_FP_gen(D, RM, Vd, size, M, Vm)    (0b111111101<<23 | (D)<<22 | 0b111<<19 | (RM)<<16 | (Vd)<<12 | 0b10<<10 | (size)<<8 | 1<<6 | (M)<<4 | (Vm))
+// Round floating-point to integer to Nearest with Ties to Away
+#define VRINTA_F32(Sd, Sm)  EMIT(VRINT_FP_gen(((Sd)&1), 0b00, ((Sd)>>1)&15, 0b10, (Sm)&1, ((Sm)>>1)&15))
+// Round floating-point to integer to Nearest with Ties to Away
+#define VRINTA_F64(Dd, Dm)  EMIT(VRINT_FP_gen(((Dd)>>4)&1, 0b00, (Dd)&15, 0b11, ((Dm)>>4)&1, (Dm)&15))
+// Round floating-point to integer towards -Infinity
+#define VRINTM_F32(Sd, Sm)  EMIT(VRINT_FP_gen(((Sd)&1), 0b11, ((Sd)>>1)&15, 0b10, (Sm)&1, ((Sm)>>1)&15))
+// Round floating-point to integer towards -Infinity
+#define VRINTM_F64(Dd, Dm)  EMIT(VRINT_FP_gen(((Dd)>>4)&1, 0b11, (Dd)&15, 0b11, ((Dm)>>4)&1, (Dm)&15))
+// Round floating-point to integer to Nearest
+#define VRINTN_F32(Sd, Sm)  EMIT(VRINT_FP_gen(((Sd)&1), 0b01, ((Sd)>>1)&15, 0b10, (Sm)&1, ((Sm)>>1)&15))
+// Round floating-point to integer to Nearest
+#define VRINTN_F64(Dd, Dm)  EMIT(VRINT_FP_gen(((Dd)>>4)&1, 0b01, (Dd)&15, 0b11, ((Dm)>>4)&1, (Dm)&15))
+// Round floating-point to integer towards +Infinity
+#define VRINTP_F32(Sd, Sm)  EMIT(VRINT_FP_gen(((Sd)&1), 0b10, ((Sd)>>1)&15, 0b10, (Sm)&1, ((Sm)>>1)&15))
+// Round floating-point to integer towards +Infinity
+#define VRINTP_F64(Dd, Dm)  EMIT(VRINT_FP_gen(((Dd)>>4)&1, 0b10, (Dd)&15, 0b11, ((Dm)>>4)&1, (Dm)&15))
+
+#define VRINTR_gen(cond, D, Vd, size, M, Vm)    (cond | 0b11101<<23 | (D)<<22 | 0b11<<20 | 0b110<<16 | (Vd)<<12 | 0b10<<10 | (size)<<8 | 1<<6 | (M)<<5 | (Vm))
+// Round floating-point to integer rounds a floating-point value to an integral floating-point value of the same size using the rounding mode specified in the FPSCR
+#define VRINTR_F32(Sd, Sm)  EMIT(VRINTR_gen(c__, (Sd)&1, ((Sd)>>1)&15, 0b10, (Sm)&1, ((Sm)>>1)&15))
+// Round floating-point to integer rounds a floating-point value to an integral floating-point value of the same size using the rounding mode specified in the FPSCR
+#define VRINTR_F64(Dd, Dm)  EMIT(VRINTR_gen(c__, ((Dd)>>4)&1, (Dd)&15, 0b11, ((Dm)>>4)&1, (Dm)&15))
+
+#define VRINTX_gen(cond, D, Vd, size, M, Vm)    (cond | 0b11101<<23 | (D)<<22 | 0b11<<20 | 0b111<<16 | (Vd)<<12 | 0b10<<10 | (size)<<8 | 1<<6 | (M)<<5 | (Vm))
+// Round floating-point to integer rounds a floating-point value to an integral floating-point value of the same size using the rounding mode specified in the FPSCR with eXecptions if inexact
+#define VRINTX_F32(Sd, Sm)  EMIT(VRINTX_gen(c__, (Sd)&1, ((Sd)>>1)&15, 0b10, (Sm)&1, ((Sm)>>1)&15))
+// Round floating-point to integer rounds a floating-point value to an integral floating-point value of the same size using the rounding mode specified in the FPSCR with eXecptions if inexact
+#define VRINTX_F64(Dd, Dm)  EMIT(VRINTX_gen(c__, ((Dd)>>4)&1, (Dd)&15, 0b11, ((Dm)>>4)&1, (Dm)&15))
+
+#define VRINTZ_gen(cond, D, Vd, size, M, Vm)    (cond | 0b11101<<23 | (D)<<22 | 0b11<<20 | 0b110<<16 | (Vd)<<12 | 0b10<<10 | (size)<<8 | 1<<7 | 1<<6 | (M)<<5 | (Vm))
+// Round floating-point to integer towards Zero
+#define VRINTZ_F32(Sd, Sm)  EMIT(VRINTZ_gen(c__, (Sd)&1, ((Sd)>>1)&15, 0b10, (Sm)&1, ((Sm)>>1)&15))
+// Round floating-point to integer towards Zero
+#define VRINTZ_F64(Dd, Dm)  EMIT(VRINTZ_gen(c__, ((Dd)>>4)&1, (Dd)&15, 0b11, ((Dm)>>4)&1, (Dm)&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 +790,12 @@ 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) )
+// Neg F32 Dd = - Dm
+#define VNEG_F32_cond(cond, Sd, Sm)    EMIT((cond) | (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 +804,13 @@ 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
 
@@ -611,6 +822,8 @@ Op is 20-27
 #define VLD1_32(Dd, Rn) EMIT(Vxx1gen(1, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b0111, 2, 0, 15))
 // Load [Rn]! => Dd. Align is 4
 #define VLD1_32_W(Dd, Rn) EMIT(Vxx1gen(1, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b0111, 2, 0, 13))
+// Load [Rn]! => Dd. Align is 8
+#define VLD1_64_W(Dd, Rn) EMIT(Vxx1gen(1, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b0111, 3, 0, 13))
 // Load [Rn]! => Dd/Dd+1. Align is 4
 #define VLD1Q_32_W(Dd, Rn) EMIT(Vxx1gen(1, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b1010, 2, 0, 13))
 // Load [Rn, Rm]! => Dd/Dd+1. If Rm==15, no writeback, Rm ignored, else writeback Rn <- Rn+Rm. Align is 4
@@ -632,6 +845,8 @@ Op is 20-27
 
 // Store [Rn]! => Dd. Align is 4
 #define VST1_32_W(Dd, Rn) EMIT(Vxx1gen(0, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b0111, 2, 0, 13))
+// Store [Rn]! => Dd. Align is 8
+#define VST1_64_W(Dd, Rn) EMIT(Vxx1gen(0, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b0111, 3, 0, 13))
 // Store [Rn] => Dd/Dd+1. Align is 4
 #define VST1Q_32(Dd, Rn) EMIT(Vxx1gen(0, ((Dd)>>4)&1, Rn, ((Dd)&0x0f), 0b1010, 2, 0, 15))
 // Store [Rn]! => Dd/Dd+1. Align is 4
@@ -658,31 +873,41 @@ 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))
-#define VLD1LANE_16(Dd, Rn, index)   EMIT(VLD1LANE_gen(((Dd)>>4)&1, Rn, (Dd)&15, 1, (index)<<2, 15))
-#define VLD1LANE_32(Dd, Rn, index)   EMIT(VLD1LANE_gen(((Dd)>>4)&1, Rn, (Dd)&15, 2, (index)<<3, 15))
+#define VLD1LANE_8(Dd, Rn, index)    EMIT(VLD1LANE_gen(((Dd)>>4)&1, (Rn), (Dd)&15, 0, (index)<<1, 15))
+#define VLD1LANE_16(Dd, Rn, index)   EMIT(VLD1LANE_gen(((Dd)>>4)&1, (Rn), (Dd)&15, 1, (index)<<2, 15))
+#define VLD1LANE_32(Dd, Rn, index)   EMIT(VLD1LANE_gen(((Dd)>>4)&1, (Rn), (Dd)&15, 2, (index)<<3, 15))
 
 #define VLD1ALL_gen(D, Rn, Vd, size, T, a, Rm) (0b1111<<28 | 0b0100<<24 | 1<<23 | (D)<<22 | 0b10<<20 | (Rn)<<16 | (Vd)<<12 | 0b11<<10| (size)<<6 | (T)<<5 | (a)<<4 | (Rm))
-#define VLD1ALL_16(Dd, Rn)      EMIT(VLD1ALL_gen(((Dd)>>4)&1, Rn, (Dd)&15, 1, 0, 0, 15))
-#define VLD1ALL_32(Dd, Rn)      EMIT(VLD1ALL_gen(((Dd)>>4)&1, Rn, (Dd)&15, 2, 0, 0, 15))
-#define VLD1QALL_32(Dd, Rn)     EMIT(VLD1ALL_gen(((Dd)>>4)&1, Rn, (Dd)&15, 2, 1, 0, 15))
+#define VLD1ALL_16(Dd, Rn)      EMIT(VLD1ALL_gen(((Dd)>>4)&1, (Rn), (Dd)&15, 1, 0, 0, 15))
+#define VLD1ALL_32(Dd, Rn)      EMIT(VLD1ALL_gen(((Dd)>>4)&1, (Rn), (Dd)&15, 2, 0, 0, 15))
+#define VLD1QALL_32(Dd, Rn)     EMIT(VLD1ALL_gen(((Dd)>>4)&1, (Rn), (Dd)&15, 2, 1, 0, 15))
 
 #define VST1LANE_gen(D, Rn, Vd, size, index_align, Rm) (0b1111<<28 | 0b0100<<24 | 1<<23 | (D)<<22 | 0b00<<20 | (Rn)<<16 | (Vd)<<12 | (size)<<10 | (index_align)<<4 | (Rm))
-#define VST1LANE_8(Dd, Rn, index)    EMIT(VST1LANE_gen(((Dd)>>4)&1, Rn, (Dd)&15, 0, (index)<<1, 15))
-#define VST1LANE_16(Dd, Rn, index)   EMIT(VST1LANE_gen(((Dd)>>4)&1, Rn, (Dd)&15, 1, (index)<<2, 15))
-#define VST1LANE_32(Dd, Rn, index)   EMIT(VST1LANE_gen(((Dd)>>4)&1, Rn, (Dd)&15, 2, (index)<<3, 15))
+#define VST1LANE_8(Dd, Rn, index)    EMIT(VST1LANE_gen(((Dd)>>4)&1, (Rn), (Dd)&15, 0, (index)<<1, 15))
+#define VST1LANE_16(Dd, Rn, index)   EMIT(VST1LANE_gen(((Dd)>>4)&1, (Rn), (Dd)&15, 1, (index)<<2, 15))
+#define VST1LANE_32(Dd, Rn, index)   EMIT(VST1LANE_gen(((Dd)>>4)&1, (Rn), (Dd)&15, 2, (index)<<3, 15))
 
 #define VADD_gen(size, D, Vn, Vd, N, Q, M, Vm) (0b1111<<28 | 0b0010<<24 | 0<<23 | (D)<<22 | (size)<<20 | (Vn)<<16 | (Vd)<<12 | 0b1000<<8 | (N)<<7 | (Q)<<6 | (M)<<5 | 0<<4 | (Vm))
 #define VADD_8(Dd, Dn, Dm)     EMIT(VADD_gen(0, ((Dd)>>4)&1, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 0, ((Dm)>>4)&1, (Dm)&15))
@@ -726,12 +951,21 @@ Op is 20-27
 #define VCEQQ_16(Dd, Dn, Dm)   EMIT(VCEQ_I_gen(1, ((Dd)>>4)&1, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 1, ((Dm)>>4)&1, (Dm)&15))
 #define VCEQ_32(Dd, Dn, Dm)    EMIT(VCEQ_I_gen(2, ((Dd)>>4)&1, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 0, ((Dm)>>4)&1, (Dm)&15))
 #define VCEQQ_32(Dd, Dn, Dm)   EMIT(VCEQ_I_gen(2, ((Dd)>>4)&1, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 1, ((Dm)>>4)&1, (Dm)&15))
-#define VCEQ_64(Dd, Dn, Dm)    EMIT(VCEQ_I_gen(3, ((Dd)>>4)&1, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 0, ((Dm)>>4)&1, (Dm)&15))
+//#define VCEQ_64(Dd, Dn, Dm)    EMIT(VCEQ_I_gen(3, ((Dd)>>4)&1, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 0, ((Dm)>>4)&1, (Dm)&15))   // doesn't exist
 #define VCEQQ_64(Dd, Dn, Dm)   EMIT(VCEQ_I_gen(3, ((Dd)>>4)&1, (Dn)&15, (Dd)&15, ((Dn)>>4)&1, 1, ((Dm)>>4)&1, (Dm)&15))
 #define VCEQ_F_gen(sz, D, Vn, Vd, N, Q, M, Vm)  (0b1111<<28 | 0b0010<<24 | 0<<23 | (D)<<22 | (sz)<<20 | (Vn)<<16 | (Vd)<<12 | 0b1110<<8 | (N)<<7 | (Q)<<6 | (M)<<5 | 0<<4 | (Vm))
 #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 VCEQQ_0_F32(Dd, Dm)     EMIT(VCEQ_0_gen(((Dd)>>4)&1, 2, (Dd)&15, 1, 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 +991,15 @@ 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 +1046,11 @@ 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 +1103,8 @@ 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 +1310,7 @@ 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 +1361,35 @@ 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_8(Dd, Dm, x)       EMIT(VDUP_sgen(((Dd)>>4)&1, (((x)<<1)|0b1), (Dd)&15, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VDUPQ_8(Dd, Dm, x)      EMIT(VDUP_sgen(((Dd)>>4)&1, (((x)<<1)|0b1), (Dd)&15, 1, ((Dm)>>4)&1, (Dm)&15))
+#define VDUP_16(Dd, Dm, x)      EMIT(VDUP_sgen(((Dd)>>4)&1, (((x)<<2)|0b10), (Dd)&15, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VDUPQ_16(Dd, Dm, x)     EMIT(VDUP_sgen(((Dd)>>4)&1, (((x)<<2)|0b10), (Dd)&15, 1, ((Dm)>>4)&1, (Dm)&15))
+#define VDUP_32(Dd, Dm, x)      EMIT(VDUP_sgen(((Dd)>>4)&1, (((x)<<3)|0b100), (Dd)&15, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VDUPQ_32(Dd, Dm, x)     EMIT(VDUP_sgen(((Dd)>>4)&1, (((x)<<3)|0b100), (Dd)&15, 1, ((Dm)>>4)&1, (Dm)&15))
+#define VDUPQ_64(Dd, Dm, x)     EMIT(VDUP_sgen(((Dd)>>4)&1, (((x)<<4)|0b1000), (Dd)&15, 1, ((Dm)>>4)&1, (Dm)&15))
+
+#define VREV64_gen(D, size, Vd, Q, M, Vm)   (0b1111<<28 | 0b0011<<24 | 1<<23 | (D)<<22 | 0b11<<20 | (size)<<18 | (Vd)<<12 | (Q)<<6 | (M)<<5 | (Vm))
+#define VREV64_32(Dd, Dm)       EMIT(VREV64_gen(((Dd)>>4)&1, 0b10, (Dd)&15, 0, ((Dm)>>4)&1, (Dm)&15))
+#define VREV64Q_32(Dd, Dm)      EMIT(VREV64_gen(((Dd)>>4)&1, 0b10, (Dd)&15, 1, ((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__

src/dynarec/arm_epilog.S

@@ -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

src/dynarec/arm_instructions.txt

@@ -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>
@@ -172,7 +179,7 @@ INVALIDATE 1 1 1 1 0 0 1 1 1 <1> 1 1 <2>/=10 0 0 <4> 0 0 0 <2>/11 <2> 0 <4>
 INVALIDATE 1 1 1 1 0 0 1 1 1 <1> 1 1 <2>/=10 0 0 <4> 0 0 0 <2>/01 <2> 0 <4>
 INVALIDATE 1 1 1 1 0 0 1 1 1 <1> 1 1 <2>/=01 0 0 <4> 0 0 0 <2>/11 <2> 0 <4>
 INVALIDATE 1 1 1 1 0 0 1 1 1 <1> 1 1 <2>/=00 0 0 <4> 0 0 0 <2>/=11 <2> 0 <4>
-NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 0 0 Vd 0 0 0 @<2> Q M 0 Vm VREV\%d16 &l&l param\<c>.\%d8 &l&l size\ <Qd>, <Qm>
+NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 0 0 Vd 0 0 0 @<2> Q M 0 Vm VREV\%d64 &g&g param\<c>.\%d8 &l&l size\ <Qd>, <Qm>
 NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 0 0 Vd 0 0 1 0 U Q M 0 Vm VPADDL<c>.<dt> <Qd>, <Qm>
 NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 0 0 Vd 0 1 0 0 0 Q M 0 Vm VCLS<c>.<dt> <Qd>, <Qm>
 NEON 1 1 1 1 0 0 1 1 1 D 1 1 size 0 0 Vd 0 1 0 0 1 Q M 0 Vm VCLZ<c>.<dt> <Qd>, <Qm>
@@ -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[5] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> 1 1 1 1 @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> Rm @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\], <Rm>
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
-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>

src/dynarec/arm_linker.S

@@ -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
-

src/dynarec/arm_lock_helper.S

@@ -0,0 +1,206 @@
+// 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
+.global arm_lock_storeifref2
+.global arm_lock_decifnot0b
+.global arm_lock_incb
+.global arm_lock_incif0b
+.global arm_lock_stored
+.global arm_lock_storeb
+.global arm_lock_storeif0b
+.global arm_lock_readb
+
+arm_lock_read_b:
+    // address is r0, return is r0
+    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 old [r0] value
+    // r0 needs to be aligned
+    dmb     ish
+    ldrex   r2, [r0]
+    cmp     r2, #0
+    bne     arm_lock_storeifnull_exit
+    strex   r3, r1, [r0]
+    cmp     r3, #1
+    beq     arm_lock_storeifnull
+arm_lock_storeifnull_exit:
+    mov     r0, r2
+    bx      lr
+
+arm_lock_storeifref2:
+    dmb     ish
+arm_lock_storeifref2_loop:
+    // address is r0, value is r1, r1 store to r0 only if [r0] is r2. return old [r0] value (so r2 or old value)
+    ldrex   r3, [r0]
+    cmp     r3, r2
+    movne   r0, r3
+    bne     arm_lock_storeifref2_exit
+    strex   r3, r1, [r0]
+    cmp     r3, #1
+    beq     arm_lock_storeifref_loop
+    mov     r0, r2
+arm_lock_storeifref2_exit:
+    bx      lr
+
+arm_lock_storeifref:
+    dmb     ish
+arm_lock_storeifref_loop:
+    // address is r0, value is r1, r1 store to r0 only if [r0] is r2. return new [r0] value (so r1 or old value)
+    ldrex   r3, [r0]
+    cmp     r3, r2
+    movne   r0, r3
+    bne     arm_lock_storeifref_exit
+    strex   r3, r1, [r0]
+    cmp     r3, #1
+    beq     arm_lock_storeifref_loop
+    mov     r0, r1
+arm_lock_storeifref_exit:
+    bx      lr
+
+arm_lock_decifnot0b:
+    dmb     ish
+arm_lock_decifnot0b_0:
+    ldrexb  r1, [r0]
+    cmp     r1, #0
+    subne   r3, r1, #1
+    moveq   r3, r1
+    strexb  r2, r3, [r0]
+    cmp     r2, #1
+    beq     arm_lock_decifnot0b_0
+    dmb     ish
+    mov     r0, r1
+    bx      lr
+
+arm_lock_incb:
+    dmb     ish
+arm_lock_incb_0:
+    ldrexb  r1, [r0]
+    add     r1, r1, #1
+    strexb  r2, r1, [r0]
+    cmp     r2, #1
+    beq     arm_lock_incb_0
+arm_lock_incb_exit:
+    dmb     ish
+    bx      lr
+
+arm_lock_incif0b:
+    dmb     ish
+arm_lock_incif0b_0:
+    ldrexb  r1, [r0]
+    cmp     r1, #0
+    addeq   r3, r1, #1
+    movne   r3, r1
+    strexb  r2, r3, [r0]
+    cmp     r2, #1
+    beq     arm_lock_incif0b_0
+    dmb     ish
+    mov     r0, r1
+    bx      lr
+
+arm_lock_stored:
+    dmb     ish
+    str     r1, [r0]
+    dmb     ish
+    bx      lr
+
+arm_lock_storeb:
+    dmb     ish
+    strb    r1, [r0]
+    dmb     ish
+    bx      lr
+
+arm_lock_storeif0b:
+    dmb     ish
+arm_lock_storeif0b_0:
+    ldrexb  r2, [r0]
+    cmp     r2, #0
+    bne     arm_lock_storeif0b_exit
+    strexb  r3, r1, [r0]
+    cmp     r3, #1
+    beq     arm_lock_storeif0b_0
+    mov     r2, #0
+    dmb     ish
+arm_lock_storeif0b_exit:
+    mov     r0, r2
+    bx      lr
+
+arm_lock_readb:
+    dmb     ish
+    ldrb    r0, [r0]
+    bx      lr

src/dynarec/arm_lock_helper.h

@@ -0,0 +1,58 @@
+#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 old [p] value
+extern void* arm_lock_storeifnull(void* p, void* val);
+
+// Atomically store value to [p] only if [p] is ref. Return new [p] value (so val or old)
+extern void* arm_lock_storeifref(void* p, void* val, void* ref);
+
+// Atomically store value to [p] only if [p] is ref. Return old [p] value
+extern void* arm_lock_storeifref2(void* p, void* val, void* ref);
+
+// decrement atomicaly the byte at [p] (but only if p not 0). Return old [p] value
+extern int arm_lock_decifnot0b(void*p);
+
+// increment atomicaly the byte at [p]
+extern void arm_lock_incb(void*p);
+
+// increment atomicaly the byte at [p] only if it was 0. Return the old value of [p]
+extern int arm_lock_incif0b(void*p);
+
+// atomic store (with memory barrier)
+extern void arm_lock_stored(void*p, uint32_t d);
+
+// atomic store (with memory barrier)
+extern void arm_lock_storeb(void*p, uint8_t b);
+
+// atomic store (with memory barrier) only if it was 0. Return the old value of [p]
+extern int arm_lock_storeif0b(void*p, uint8_t b);
+
+// atomic load (with memory barrier)
+extern int arm_lock_readb(void*p);
+
+#endif  //__ARM_LOCK_HELPER__H__

src/dynarec/arm_next.S

@@ -0,0 +1,24 @@
+//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
+    .4byte  0   // NULL pointer before arm64_next, for getDB
+arm_next:
+    // emu is r0
+    //stm     r0, {r4-r12,r14}   // don't put put back reg value in emu, faster but more tricky to debug
+    // 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
+

src/dynarec/arm_prolog.S

@@ -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

src/dynarec/arm_table.S

@@ -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
-

src/dynarec/dynablock.c

@@ -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"
@@ -18,18 +20,18 @@
 #include "dynablock_private.h"
 #include "dynarec_private.h"
 #include "elfloader.h"
+#include "bridge.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;
@@ -39,549 +41,304 @@ 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* 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);}
-    }
-    return ret;
-}
-
-void FreeDynablock(dynablock_t* db)
+dynablock_t* InvalidDynablock(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 NULL; // already in the process of deletion!
+        dynarec_log(LOG_DEBUG, "InvalidDynablock(%p), db->block=%p x86=%p:%p already gone=%d\n", db, db->block, db->x86_addr, db->x86_addr+db->x86_size-1, db->gone);
+        if(need_lock)
+            mutex_lock(&my_context->mutex_dyndump);
+        // remove jumptable
+        setJumpTableDefault(db->x86_addr);
         db->done = 0;
-        // remove from direct if there
-        uintptr_t startdb = db->parent->text;
-        uintptr_t enddb = db->parent->text + db->parent->textsz;
-        if(db->parent->direct) {
-            uintptr_t addr = (uintptr_t)db->x86_addr;
-            if(addr>=startdb && addr<enddb)
-                db->parent->direct[addr-startdb] = NULL;
-        }
-       // 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 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);
-        }
-        // 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);
-        }
-        free(db->sons);
-        free(db->table);
-        free(db);
+        db->gone = 1;
+        if(need_lock)
+            mutex_unlock(&my_context->mutex_dyndump);
     }
+    return db;
 }
 
-void FreeDynablockList(dynablocklist_t** dynablocks)
+void FreeInvalidDynablock(dynablock_t* db, int need_lock)
 {
-    if(!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);
+    if(db) {
+        if(!db->gone)
+            return; // already in the process of deletion!
+        dynarec_log(LOG_DEBUG, "FreeInvalidDynablock(%p), db->block=%p x86=%p:%p already gone=%d\n", db, db->block, db->x86_addr, db->x86_addr+db->x86_size-1, db->gone);
+        if(need_lock)
+            mutex_lock(&my_context->mutex_dyndump);
+        FreeDynarecMap((uintptr_t)db->actual_block);
+        customFree(db);
+        if(need_lock)
+            mutex_unlock(&my_context->mutex_dyndump);
     }
-    if((*dynablocks)->direct) {
-        for (int i=0; i<(*dynablocks)->textsz; ++i) {
-            if((*dynablocks)->direct[i] && !(*dynablocks)->direct[i]->father) 
-                FreeDynablock((*dynablocks)->direct[i]);
-        }
-        free((*dynablocks)->direct);
-    }
-    (*dynablocks)->direct = 0;
-
-    pthread_rwlock_destroy(&(*dynablocks)->rwlock_blocks);
-
-    free(*dynablocks);
-    *dynablocks = NULL;
 }
 
+void FreeDynablock(dynablock_t* db, int need_lock)
+{
+    if(db) {
+        if(db->gone)
+            return; // already in the process of deletion!
+        dynarec_log(LOG_DEBUG, "FreeDynablock(%p), db->block=%p x86=%p:%p already gone=%d\n", db, db->block, db->x86_addr, db->x86_addr+db->x86_size-1, db->gone);
+        if(need_lock)
+            mutex_lock(&my_context->mutex_dyndump);
+        // remove jumptable
+        setJumpTableDefault(db->x86_addr);
+        dynarec_log(LOG_DEBUG, " -- FreeDyrecMap(%p, %d)\n", db->actual_block, db->size);
+        db->done = 0;
+        db->gone = 1;
+        if(db->previous)
+            FreeInvalidDynablock(db->previous, 0);
+        FreeDynarecMap((uintptr_t)db->actual_block);
+        customFree(db);
+        if(need_lock)
+            mutex_unlock(&my_context->mutex_dyndump);
+    }
+}
+
+
+
 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);
+        dynarec_log(LOG_DEBUG, "MarkDynablock %p %p-%p\n", db, db->x86_addr, db->x86_addr+db->x86_size-1);
+        if(!setJumpTableIfRef(db->x86_addr, db->jmpnext, db->block)) {
+            dynablock_t* old = db;
+            db = getDB((uintptr_t)old->x86_addr);
+            if(!old->gone && db!=old) {
+                printf_log(LOG_INFO, "Warning, couldn't mark block as dirty for %p, block=%p, current_block=%p\n", old->x86_addr, old, db);
+                // the block is lost, need to invalidate it...
+                old->gone = 1;
+                old->done = 0;
+                if(!db || db->previous)
+                    FreeInvalidDynablock(old, 1);
+                else
+                    db->previous = old;
+            }
         }
-        db->need_test = 1;
     }
 }
 
-void ProtectDynablock(dynablock_t* db)
+static int IntervalIntersects(uintptr_t start1, uintptr_t end1, uintptr_t start2, uintptr_t end2)
+{
+    if(start1 > end2 || start2 > end1)
+        return 0;
+    return 1;
+}
+
+static int MarkedDynablock(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);
+        if(getNeedTest((uintptr_t)db->x86_addr))
+            return 1; // already done
     }
-}
-
-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]);
-        }
-    }
-}
-
-uintptr_t StartDynablockList(dynablocklist_t* db)
-{
-    if(db)
-        return db->text;
     return 0;
 }
-uintptr_t EndDynablockList(dynablocklist_t* db)
+
+void MarkRangeDynablock(dynablock_t* db, uintptr_t addr, uintptr_t size)
 {
-    if(db)
-        return db->text+db->textsz;
-    return 0;
-}
-void FreeDirectDynablock(dynablocklist_t* dynablocks, uintptr_t addr, uintptr_t size)
-{
-    if(!dynablocks)
+    // Mark will try to find *any* blocks that intersect the range to mark
+    if(!db)
         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]);
-            }
-}
-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]);
-            }
+    dynarec_log(LOG_DEBUG, "MarkRangeDynablock %p-%p .. startdb=%p, sizedb=%p\n", (void*)addr, (void*)addr+size-1, (void*)db->x86_addr, (void*)db->x86_size);
+    if(IntervalIntersects((uintptr_t)db->x86_addr, (uintptr_t)db->x86_addr+db->x86_size-1, addr, addr+size+1))
+        MarkDynablock(db);
 }
 
-void ProtectDirectDynablock(dynablocklist_t* dynablocks, uintptr_t addr, uintptr_t size)
+int FreeRangeDynablock(dynablock_t* db, 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)
-        protectDB(start, end-start); //no +1; as end/enddb is exclusive and not inclusive
-}
+    if(!db)
+        return 1;
 
-void FreeRangeDynablock(dynablocklist_t* dynablocks, uintptr_t addr, uintptr_t size)
-{
-    if(!dynablocks)
-        return;
-    if(dynablocks->direct)
-        FreeDirectDynablock(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)))
-                ProtectDynablock(db);
-        );
-    }
-}
-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);
-        );
+    int need_lock = my_context?1:0;
+    if(IntervalIntersects((uintptr_t)db->x86_addr, (uintptr_t)db->x86_addr+db->x86_size-1, addr, addr+size+1)) {
+        FreeDynablock(db, need_lock);
+        return 0;
     }
+    return 1;
 }
 
-dynablock_t* FindDynablockDynablocklist(void* addr, dynablocklist_t* dynablocks)
+dynablock_t *AddNewDynablock(uintptr_t addr)
 {
-    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;
-        );
-    }
-    return NULL;
-}
-
-dynablock_t* FindDynablockFromNativeAddress(void* 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;
-}
-
-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)
-{
-    if(!dynablocks) {
-        dynarec_log(LOG_INFO, "Warning: Ask to create a dynblock with a NULL dynablocklist (addr=%p)\n", (void*)addr);
-        *created = 0;
+    dynablock_t* block;
+    #if 0
+    // check if memory as the correct flags
+    int prot = getProtection(addr);
+    if(!(prot&(PROT_EXEC|PROT_DYNAREC|PROT_DYNAREC_R))) {
+        dynarec_log(LOG_VERBOSE, "Block asked on a memory with no execution flags 0x%02X\n", prot);
         return NULL;
     }
-    khint_t k;
-    int ret;
-    dynablock_t* block = NULL;
-    // first, check if it exist in direct access mode
-    if(dynablocks->direct && (addr>=dynablocks->text) && (addr<(dynablocks->text+dynablocks->textsz))) {
-        block = dynablocks->direct[addr-dynablocks->text];
-        if(block) {
-            dynarec_log(LOG_DUMP, "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);
+    #endif
     // 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);
-    }
-    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);
-
-    *created = 1;
+    dynarec_log(LOG_VERBOSE, "Ask for DynaRec Block creation @%p\n", (void*)addr);
+    block = (dynablock_t*)customCalloc(1, sizeof(dynablock_t));
     return block;
 }
 
+//TODO: move this to dynrec_arm.c and track allocated structure to avoid memory leak
+static __thread struct __jmp_buf_tag dynarec_jmpbuf;
+
+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, 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 = 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)
+    if(hasAlternate((void*)addr))
         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)
+    dynablock_t* block = getDB(addr);
+    if(block || !create)
         return block;
 
-    int created = create;
-    block = AddNewDynablock(dynablocks, addr, (dynablocks->nolinker)&&(dynablocks->direct), &created);
-    if(!created || !create)
-        return block;   // existing block...
+    if(need_lock) {
+        if(box86_dynarec_wait) {
+            mutex_lock(&my_context->mutex_dyndump);
+        } else {
+            if(mutex_trylock(&my_context->mutex_dyndump))   // FillBlock not available for now
+                return NULL;
+        }
+        block = getDB(addr);    // just in case
+        if(block) {
+            mutex_unlock(&my_context->mutex_dyndump);
+            return block;
+        }
+    }
+    
+    block = AddNewDynablock(addr);
 
-    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(sigsetjmp(&dynarec_jmpbuf, 1)) {
+        printf_log(LOG_INFO, "FillBlock at %p triggered a segfault, cancelling\n", (void*)addr);
+        FreeDynablock(block, 0);
+        if(need_lock)
+            mutex_unlock(&my_context->mutex_dyndump);
+        return NULL;
+    }
+    void* ret = FillBlock(block, filladdr);
+    if(!ret) {
+        dynarec_log(LOG_DEBUG, "Fillblock of block %p for %p returned an error\n", block, (void*)addr);
+        customFree(block);
+        block = NULL;
+    }
+    // check size
+    if(block) {
+        int blocksz = block->x86_size;
+        if(blocksz>my_context->max_db_size)
+            my_context->max_db_size = blocksz;
+        // fill-in jumptable
+        if(!addJumpTableIfDefault(block->x86_addr, block->dirty?block->jmpnext:block->block)) {
+            FreeDynablock(block, 0);
+            block = getDB(addr);
+            MarkDynablock(block);   // just in case...
+        } else {
+            if(block->x86_size)
+                block->done = 1;    // don't validate the block if the size is null, but keep the block
+        }
+    }
+    if(need_lock)
+        mutex_unlock(&my_context->mutex_dyndump);
 
-    dynarec_log(LOG_DEBUG, " --- DynaRec Block %s @%p:%p (%p, 0x%x bytes, %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, "%04d| --- DynaRec Block created @%p:%p (%p, 0x%x bytes)\n", GetTID(), (void*)addr, (void*)(addr+((block)?block->x86_size:1)-1), (block)?block->block:0, (block)?block->size:0);
 
     return block;
 }
 
-dynablock_t* DBGetBlock(x86emu_t* emu, uintptr_t addr, int create, dynablock_t** current)
+dynablock_t* DBGetBlock(x86emu_t* emu, uintptr_t addr, int create)
 {
-    dynablock_t *db = internalDBGetBlock(emu, addr, create, *current);
-    if(db && (db->need_test || (db->father && db->father->need_test))) {
-        dynablock_t *father = db->father?db->father:db;
-        uint32_t hash = X31_hash_code(father->x86_addr, father->x86_size);
-        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);
-            // 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);
-            // start again... (will create a new block)
-            db = internalDBGetBlock(emu, addr, create, *current);
-        } else {
-            father->need_test = 0;
-            protectDB((uintptr_t)father->x86_addr, father->x86_size);
+    dynablock_t *db = internalDBGetBlock(emu, addr, addr, create, 1);
+    if(db && db->done && db->block && getNeedTest(addr)) {
+        if(db->always_test)
+            sched_yield();  // just calm down...
+        if(AreaInHotPage((uintptr_t)db->x86_addr, (uintptr_t)db->x86_addr + db->x86_size - 1)) {
+            emu->test.test = 0;
+            if(box86_dynarec_fastpage) {
+                uint32_t hash = X31_hash_code(db->x86_addr, db->x86_size);
+                if(hash==db->hash) { // seems ok, run it without reprotecting it
+                    setJumpTableIfRef(db->x86_addr, db->block, db->jmpnext);
+                    return db;
+                }
+                db->done = 0;   // invalidating the block, it's already not good
+                dynarec_log(LOG_DEBUG, "Invalidating block %p from %p:%p (hash:%X/%X) for %p\n", db, db->x86_addr, db->x86_addr+db->x86_size-1, hash, db->hash, (void*)addr);
+                // Free db, it's now invalid!
+                FreeDynablock(db, 1);
+                return NULL;    // not building a new one, it's still a hotpage
+            } else {
+                dynarec_log(LOG_INFO, "Not running block %p from %p:%p with for %p because it's in a hotpage\n", db, db->x86_addr, db->x86_addr+db->x86_size-1, (void*)addr);
+                return NULL;
+            }
         }
+        uint32_t hash = X31_hash_code(db->x86_addr, db->x86_size);
+        int need_lock = mutex_trylock(&my_context->mutex_dyndump);
+        if(hash!=db->hash) {
+            db->done = 0;   // invalidating the block
+            dynarec_log(LOG_DEBUG, "Invalidating block %p from %p:%p (hash:%X/%X) for %p\n", db, db->x86_addr, db->x86_addr+db->x86_size-1, hash, db->hash, (void*)addr);
+            // Free db, it's now invalid!
+            dynablock_t* old = InvalidDynablock(db, need_lock);
+            // start again... (will create a new block)
+            db = internalDBGetBlock(emu, addr, addr, create, need_lock);
+            if(db) {
+                if(db->previous)
+                    FreeInvalidDynablock(db->previous, need_lock);
+                db->previous = old;
+            } else
+                FreeInvalidDynablock(old, need_lock);
+        } else {
+            dynarec_log(LOG_DEBUG, "Validating block %p from %p:%p (hash:%X) for %p\n", db, db->x86_addr, db->x86_addr+db->x86_size-1, db->hash, (void*)addr);
+            protectDB((uintptr_t)db->x86_addr, db->x86_size);
+            // fill back jumptable
+            if(isprotectedDB((uintptr_t)db->x86_addr, db->x86_size) && !db->always_test) {
+                setJumpTableIfRef(db->x86_addr, db->block, db->jmpnext);
+            }
+        }
+        if(!need_lock)
+            mutex_unlock(&my_context->mutex_dyndump);
     } 
+    if(!db || !db->block || !db->done)
+        emu->test.test = 0;
+    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, 1);
+    if(db && db->done && db->block && getNeedTest(filladdr)) {
+        if(db->always_test)
+            sched_yield();  // just calm down...
+        int need_lock = mutex_trylock(&my_context->mutex_dyndump);
+        uint32_t hash = X31_hash_code(db->x86_addr, db->x86_size);
+        if(hash!=db->hash) {
+            db->done = 0;   // invalidating the block
+            dynarec_log(LOG_DEBUG, "Invalidating alt block %p from %p:%p (hash:%X/%X) for %p\n", db, db->x86_addr, db->x86_addr+db->x86_size, hash, db->hash, (void*)addr);
+            // Free db, it's now invalid!
+            dynablock_t* old = InvalidDynablock(db, need_lock);
+            // start again... (will create a new block)
+            db = internalDBGetBlock(emu, addr, filladdr, create, need_lock);
+            if(db) {
+                if(db->previous)
+                    FreeInvalidDynablock(db->previous, need_lock);
+                db->previous = old;
+            } else
+                FreeInvalidDynablock(old, need_lock);
+        } else {
+            protectDB((uintptr_t)db->x86_addr, db->x86_size);
+            // fill back jumptable
+            if(isprotectedDB((uintptr_t)db->x86_addr, db->x86_size) && !db->always_test) {
+                setJumpTableIfRef(db->x86_addr, db->block, db->jmpnext);
+            }
+        }
+        if(!need_lock)
+            mutex_unlock(&my_context->mutex_dyndump);
+    } 
+    if(!db || !db->block || !db->done)
+        emu->test.test = 0;
     return db;
 }

src/dynarec/dynablock_private.h

@@ -1,36 +1,26 @@
 #ifndef __DYNABLOCK_PRIVATE_H_
 #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 char x86:4;
+    unsigned char nat:4;
+} instsize_t;
 
 typedef struct dynablock_s {
-    dynablocklist_t* parent;
-    kh_mark_t*      marks; // List of blocks that marked this block
-    void*           block;
+    void*           block;  // block-sizeof(void*) == self
+    void*           actual_block;   // the actual start of the block (so block-sizeof(void*))
+    struct dynablock_s*    previous;   // a previous block that might need to be freed
     int             size;
     void*           x86_addr;
-    int             x86_size;
+    uintptr_t       x86_size;
     uint32_t        hash;
-    int             need_test;
-    uintptr_t*      table;
-    int             tablesz;
-    int             done;
+    uint8_t         done;
+    uint8_t         gone;
+    uint8_t         always_test;
+    uint8_t         dirty;      // if need to be tested as soon as it's created
     int             isize;
-    dynablock_t**   sons;   // sons (kind-of dummy dynablock...)
-    int             sons_size;
-    dynablock_t*    father; // set only in the case of a son
+    instsize_t*     instsize;
+    void*           jmpnext;    // a branch jmpnext code when block is marked
 } dynablock_t;
 
-typedef struct dynablocklist_s {
-    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)
-} dynablocklist_t;
-
 #endif //__DYNABLOCK_PRIVATE_H_

src/dynarec/dynarec.c

@@ -2,6 +2,7 @@
 #include <stdlib.h>
 #include <pthread.h>
 #include <errno.h>
+#include <setjmp.h>
 
 #include "debug.h"
 #include "box86context.h"
@@ -14,84 +15,99 @@
 #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"
+void x86test_check(x86emu_t* ref, uintptr_t ip);
 #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
+#include "elfloader.h"
+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;
-    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);
+    #ifdef HAVE_TRACE
+    if(!addr) {
+        x2-=8;  // actual PC is 2 instructions ahead
+        dynablock_t* db = FindDynablockFromNativeAddress(x2);
+        void* x86addr = db?(void*)getX86Address(db, (uintptr_t)x2):NULL;
+        const char* pcname = getAddrFunctionName((uintptr_t)x86addr);
+        printf_log(LOG_NONE, "Warning, jumping to NULL address from %p (db=%p, x86addr=%p/%s)\n", x2, db, x86addr, pcname);
     }
+    #endif
+    void * jblock;
+    dynablock_t* block = NULL;
+    if(hasAlternate((void*)addr)) {
+        printf_log(LOG_INFO, "Jmp address has alternate: %p", (void*)addr);
+        addr = (uintptr_t)getAlternate((void*)addr);
+        R_EIP = addr;
+        printf_log(LOG_INFO, " -> %p\n", (void*)addr);
+        block = DBGetBlock(emu, addr, 1);
+    } else
+        block = DBGetBlock(emu, addr, 1);
     if(!block) {
-        // no block, don't try again, ever
-        tableupdate(arm_epilog, addr, table);
-        return arm_epilog_fast;
+        // no block, let link table as is...
+        #ifdef HAVE_TRACE
+        if(LOG_INFO<=box86_dynarec_log) {
+            dynablock_t* db = FindDynablockFromNativeAddress(x2);
+            elfheader_t* h = FindElfAddress(my_context, (uintptr_t)x2);
+            dynarec_log(LOG_INFO, "Warning, jumping to a no-block address %p from %p (db=%p, x64addr=%p(elf=%s))\n", (void*)addr, x2, db, db?(void*)getX86Address(db, (uintptr_t)x2-4):NULL, h?ElfName(h):"(none)");
+        }
+        #endif
+        //tableupdate(arm_epilog, addr, table);
+        return arm_epilog;
     }
     if(!block->done) {
         // not finished yet... leave linker
         //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;
+    int skip = 0;
+    if(emu->type == EMUTYPE_MAIN) {
+        ejb = GetJmpBuf();
+        if(!ejb->jmpbuf_ok) {
+            ejb->emu = emu;
+            ejb->jmpbuf_ok = 1;
+            jmpbuf_reset = 1;
+            if((skip=sigsetjmp((struct __jmp_buf_tag*)ejb->jmpbuf, 1))) {
+                printf_log(LOG_DEBUG, "Setjmp DynaCall %d, fs=0x%x\n", skip, ejb->emu->segs[_FS]);
+                addr = R_EIP;   // not sure if it should still be inside DynaCall!
+                #ifdef DYNAREC
+                if(box86_dynarec_test) {
+                    if(emu->test.clean)
+                        x86test_check(emu, R_EIP);
+                    emu->test.clean = 0;
+                }
+                #endif
+                if(skip!=2)
+                    skip = 0;
+            }
+        }
+    }
 #ifdef DYNAREC
     if(!box86_dynarec)
 #endif
@@ -110,15 +126,18 @@ void DynaCall(x86emu_t* emu, uintptr_t addr)
         dynablock_t* block = NULL;
         dynablock_t* current = NULL;
         while(!emu->quit) {
-            block = DBGetBlock(emu, R_EIP, 1, &current);
-            current = (block && !block->parent->nolinker)?block:NULL;
+            block = (skip==2)?NULL:DBGetBlock(emu, R_EIP, 1);
+            current = block;
             if(!block || !block->block || !block->done) {
+                skip = 0;
                 // no block, of block doesn't have DynaRec content (yet, temp is not null)
                 // Use interpreter (should use single instruction step...)
-                dynarec_log(LOG_DEBUG, "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);
+                if(box86_dynarec_test)
+                    emu->test.clean = 0;
                 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 emu=%p\n", GetTID(), (void*)R_EIP, block->block, block->isize ,emu);
                 CHECK_FLAGS(emu);
                 // block is here, let's run it!
                 #ifdef ARM
@@ -130,6 +149,16 @@ 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 +175,43 @@ void DynaCall(x86emu_t* emu, uintptr_t addr)
         }
     }
 #endif
+    // clear the setjmp
+    if(ejb && jmpbuf_reset)
+        ejb->jmpbuf_ok = 0;
 }
 
+int my_setcontext(x86emu_t* emu, void* ucp);
 int DynaRun(x86emu_t* emu)
 {
+    // prepare setjump for signal handling
+    emu_jmpbuf_t *ejb = NULL;
+    int skip = 0;
+#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((skip=sigsetjmp((struct __jmp_buf_tag*)ejb->jmpbuf, 1))) {
+                printf_log(LOG_DEBUG, "Setjmp DynaRun %d, fs=0x%x\n", skip, ejb->emu->segs[_FS]);
+                #ifdef DYNAREC
+                if(box86_dynarec_test) {
+                    if(emu->test.clean)
+                        x86test_check(emu, R_EIP);
+                    emu->test.clean = 0;
+                }
+                #endif
+                if(skip!=2)
+                    skip = 0;
+            }
+        }
+    }
 #ifdef DYNAREC
     if(!box86_dynarec)
 #endif
@@ -159,15 +221,18 @@ int DynaRun(x86emu_t* emu)
         dynablock_t* block = NULL;
         dynablock_t* current = NULL;
         while(!emu->quit) {
-            block = DBGetBlock(emu, R_EIP, 1, &current);
-            current = (block && !block->parent->nolinker)?block:NULL;
+            block = (skip==2)?NULL:DBGetBlock(emu, R_EIP, 1);
+            current = block;
             if(!block || !block->block || !block->done) {
+                skip = 0;
                 // no block, of block doesn't have DynaRec content (yet, temp is not null)
                 // Use interpreter (should use single instruction step...)
-                dynarec_log(LOG_DEBUG, "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);
+                if(box86_dynarec_test)
+                    emu->test.clean = 0;
                 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 emu=%p\n", GetTID(), (void*)R_EIP, block->block, block->isize, emu);
                 // block is here, let's run it!
                 #ifdef ARM
                 arm_prolog(emu, block->block);
@@ -178,9 +243,24 @@ 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]);
+                }
+            }
+            else if(emu->quit && emu->uc_link) {
+                emu->quit = 0;
+                my_setcontext(emu, emu->uc_link);
             }
         }
     }
+    // clear the setjmp
+    if(ejb && jmpbuf_reset)
+        ejb->jmpbuf_ok = 0;
     return 0;
 #endif
 }

src/dynarec/dynarec_arm.c

@@ -3,9 +3,11 @@
 #include <pthread.h>
 #include <errno.h>
 #include <string.h>
+#include <assert.h>
 
 #include "debug.h"
 #include "box86context.h"
+#include "custommem.h"
 #include "dynarec.h"
 #include "emu/x86emu_private.h"
 #include "tools/bridge_private.h"
@@ -19,6 +21,7 @@
 #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) {
@@ -26,15 +29,15 @@ void printf_x86_instruction(zydis_dec_t* dec, instruction_x86_t* inst, const cha
     if(ip[0]==0xcc && ip[1]=='S' && ip[2]=='C') {
         uint32_t a = *(uint32_t*)(ip+3);
         if(a==0) {
-            dynarec_log(LOG_NONE, "%s%p: Exit x86emu%s\n", (box86_dynarec_dump>1)?"\e[1m":"", (void*)ip, (box86_dynarec_dump>1)?"\e[m":"");
+            dynarec_log(LOG_NONE, "%s%p: Exit x86emu%s\n", (box86_dynarec_dump>1)?"\e[01;33m":"", (void*)ip, (box86_dynarec_dump>1)?"\e[m":"");
         } else {
-            dynarec_log(LOG_NONE, "%s%p: Native call to %p%s\n", (box86_dynarec_dump>1)?"\e[1m":"", (void*)ip, (void*)a, (box86_dynarec_dump>1)?"\e[m":"");
+            dynarec_log(LOG_NONE, "%s%p: Native call to %p%s\n", (box86_dynarec_dump>1)?"\e[01;33m":"", (void*)ip, (void*)a, (box86_dynarec_dump>1)?"\e[m":"");
         }
     } else {
         if(dec) {
-            dynarec_log(LOG_NONE, "%s%p: %s", (box86_dynarec_dump>1)?"\e[1m":"", ip, DecodeX86Trace(dec, inst->addr));
+            dynarec_log(LOG_NONE, "%s%p: %s", (box86_dynarec_dump>1)?"\e[01;33m":"", ip, DecodeX86Trace(dec, inst->addr));
         } else {
-            dynarec_log(LOG_NONE, "%s%p: ", (box86_dynarec_dump>1)?"\e[1m":"", ip);
+            dynarec_log(LOG_NONE, "%s%p: ", (box86_dynarec_dump>1)?"\e[01;33m":"", ip);
             for(int i=0; i<inst->size; ++i) {
                 dynarec_log(LOG_NONE, "%02X ", ip[i]);
             }
@@ -56,13 +59,23 @@ 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(dyn->next_sz == dyn->next_cap) {
-        dyn->next_cap += 16;
-        dyn->next = (uintptr_t*)realloc(dyn->next, dyn->next_cap*sizeof(uintptr_t));
-    }
+    if(!box86_dynarec_bigblock)
+        return;
+    // exist?
     for(int i=0; i<dyn->next_sz; ++i)
         if(dyn->next[i]==addr)
             return;
+    // put in a free slot
+    for(int i=0; i<dyn->next_sz; ++i)
+        if(!dyn->next[i]) {
+            dyn->next[i] = addr;
+            return;
+        }
+    // add slots
+    if(dyn->next_sz == dyn->next_cap) {
+        dyn->next_cap += 64;
+        dyn->next = (uintptr_t*)dynaRealloc(dyn->next, dyn->next_cap*sizeof(uintptr_t));
+    }
     dyn->next[dyn->next_sz++] = addr;
 }
 uintptr_t get_closest_next(dynarec_arm_t *dyn, uintptr_t addr) {
@@ -70,25 +83,28 @@ uintptr_t get_closest_next(dynarec_arm_t *dyn, uintptr_t addr) {
     uintptr_t best = 0;
     int i = 0;
     while((i<dyn->next_sz) && (best!=addr)) {
-        if(dyn->next[i]<addr) { // remove the address, it's before current address
-            memmove(dyn->next+i, dyn->next+i+1, (dyn->next_sz-i-1)*sizeof(uintptr_t));
-            --dyn->next_sz;
-        } else {
-            if((dyn->next[i]<best) || !best)
-                best = dyn->next[i];
-            ++i;
+        if(dyn->next[i]) {
+            if(dyn->next[i]<addr) { // remove the address, it's before current address
+                dyn->next[i] = 0;
+            } else {
+                if((dyn->next[i]<best) || !best)
+                    best = dyn->next[i];
+            }
         }
+        ++i;
     }
     return best;
 }
+#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)
         return is_nops(dyn, addr+1, n-1);
-    if(n>1 && PK(0)==0x66)  // if opcode start with 0x66, and there is more after, than is *can* be a NOP
+    if(n>1 && PK(0)==0x2E)  // if opcode start with 0x2E, and there is more after, than this *can* be a NOP
+        return is_nops(dyn, addr+1, n-1);
+    if(n>1 && PK(0)==0x66)  // if opcode start with 0x66, and there is more after, than this *can* be a NOP
         return is_nops(dyn, addr+1, n-1);
     if(n>2 && PK(0)==0x0f && PK(1)==0x1f && PK(2)==0x00)
         return is_nops(dyn, addr+3, n-3);
@@ -109,173 +125,469 @@ 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);
+void addInst(instsize_t* insts, size_t* size, int x86_size, int native_size)
+{
+    // x86 instruction is <16 bytes
+    int toadd;
+    if(x86_size>native_size)
+        toadd = 1 + x86_size/15;
+    else
+        toadd = 1 + native_size/15;
+    while(toadd) {
+        if(x86_size>15)
+            insts[*size].x86 = 15;    
+        else
+            insts[*size].x86 = x86_size;
+        x86_size -= insts[*size].x86;
+        if(native_size>15)
+            insts[*size].nat = 15;
+        else
+            insts[*size].nat = native_size;
+        native_size -= insts[*size].nat;
+        ++(*size);
+        --toadd;
+    }
+}
 
-void* FillBlock(dynablock_t* block) {
+
+static void fillPredecessors(dynarec_arm_t* dyn)
+{
+    int pred_sz = 1;    // to be safe
+    // compute total size of predecessor to alocate the array
+    // first compute the jumps
+    for(int i=0; i<dyn->size; ++i) {
+        if(dyn->insts[i].x86.jmp && dyn->insts[i].x86.jmp_insts!=-1) {
+            ++pred_sz;
+            ++dyn->insts[dyn->insts[i].x86.jmp_insts].pred_sz;
+        }
+    }
+    // remove "has_next" from orphean branch
+    for(int i=0; i<dyn->size-1; ++i) {
+        if(!dyn->insts[i].x86.has_next) {
+            if(dyn->insts[i+1].x86.has_next && !dyn->insts[i+1].pred_sz)
+                dyn->insts[i+1].x86.has_next = 0;
+        }
+    }
+    // second the "has_next"
+    for(int i=0; i<dyn->size-1; ++i) {
+        if(dyn->insts[i].x86.has_next) {
+            ++pred_sz;
+            ++dyn->insts[i+1].pred_sz;
+        }
+    }
+    dyn->predecessor = (int*)dynaMalloc(pred_sz*sizeof(int));
+    // fill pred pointer
+    int* p = dyn->predecessor;
+    for(int i=0; i<dyn->size; ++i) {
+        dyn->insts[i].pred = p;
+        p += dyn->insts[i].pred_sz;
+        dyn->insts[i].pred_sz=0;  // reset size, it's reused to actually fill pred[]
+    }
+    // fill pred
+    for(int i=0; i<dyn->size; ++i) {
+        if((i!=dyn->size-1) && dyn->insts[i].x86.has_next)
+            dyn->insts[i+1].pred[dyn->insts[i+1].pred_sz++] = i;
+        if(dyn->insts[i].x86.jmp && (dyn->insts[i].x86.jmp_insts!=-1)) {
+            int j = dyn->insts[i].x86.jmp_insts;
+            dyn->insts[j].pred[dyn->insts[j].pred_sz++] = i;
+        }
+    }
+}
+
+// updateNeed goes backward, from last intruction to top
+static int updateNeed(dynarec_arm_t* dyn, int ninst, uint8_t need) {
+    while (ninst>=0) {
+        // need pending but instruction is only a subset: remove pend and use an X_ALL instead
+        need |= dyn->insts[ninst].x86.need_after;
+        if((need&X_PEND) && (dyn->insts[ninst].x86.state_flags==SF_SUBSET)) {
+            need &=~X_PEND;
+            need |= X_ALL;
+        }
+        if((need&X_PEND) && (dyn->insts[ninst].x86.state_flags==SF_SET)) {
+            need &=~X_PEND;
+            need |= dyn->insts[ninst].x86.set_flags;    // SF_SET will compute all flags, it's not SUBSET!
+        }
+        if((need&X_PEND) && dyn->insts[ninst].x86.state_flags==SF_SUBSET_PENDING) {
+            need |= X_ALL&~(dyn->insts[ninst].x86.set_flags);
+        }
+        dyn->insts[ninst].x86.gen_flags = need&dyn->insts[ninst].x86.set_flags;
+        if((need&X_PEND) && (dyn->insts[ninst].x86.state_flags&SF_PENDING))
+            dyn->insts[ninst].x86.gen_flags |= X_PEND;
+        dyn->insts[ninst].x86.need_after = need;
+        need = dyn->insts[ninst].x86.need_after&~dyn->insts[ninst].x86.gen_flags;
+        if(dyn->insts[ninst].x86.may_set)
+            need |= dyn->insts[ninst].x86.gen_flags;    // forward the flags
+        // Consume X_PEND if relevant
+        if((need&X_PEND) && (dyn->insts[ninst].x86.set_flags&SF_PENDING))
+            need &=~X_PEND;
+        need |= dyn->insts[ninst].x86.use_flags;
+        if(dyn->insts[ninst].x86.need_before == need)
+            return ninst - 1;
+        dyn->insts[ninst].x86.need_before = need;
+        if(dyn->insts[ninst].x86.barrier&BARRIER_FLAGS) {
+            need = need?X_PEND:0;
+        }
+        int ok = 0;
+        for(int i=0; i<dyn->insts[ninst].pred_sz; ++i) {
+            if(dyn->insts[ninst].pred[i] == ninst-1)
+                ok = 1;
+            else
+                updateNeed(dyn, dyn->insts[ninst].pred[i], need);
+        }
+        --ninst;
+        if(!ok)
+            return ninst;
+    }
+    return ninst;
+}
+
+void* current_helper = NULL;
+
+void CancelBlock(int need_lock)
+{
+    if(need_lock)
+        mutex_lock(&my_context->mutex_dyndump);
+    dynarec_arm_t* helper = (dynarec_arm_t*)current_helper;
+    current_helper = NULL;
+    if(helper) {
+        dynaFree(helper->next);
+        dynaFree(helper->insts);
+        dynaFree(helper->predecessor);
+        if(helper->dynablock && helper->dynablock->actual_block) {
+            FreeDynarecMap((uintptr_t)helper->dynablock->actual_block);
+            helper->dynablock->actual_block = NULL;
+        }
+    }
+    if(need_lock)
+        mutex_unlock(&my_context->mutex_dyndump);
+}
+
+uintptr_t arm_pass0(dynarec_arm_t* dyn, uintptr_t addr);
+uintptr_t arm_pass1(dynarec_arm_t* dyn, uintptr_t addr);
+uintptr_t arm_pass2(dynarec_arm_t* dyn, uintptr_t addr);
+uintptr_t arm_pass3(dynarec_arm_t* dyn, uintptr_t addr);
+void arm_epilog();
+void arm_next();
+
+void* CreateEmptyBlock(dynablock_t* block, uintptr_t addr) {
+    block->isize = 0;
+    block->done = 0;
+    size_t sz = 8*sizeof(void*);
+    void* actual_p = (void*)AllocDynarecMap(sz);
+    void* p = actual_p + sizeof(void*);
+    if(actual_p==NULL) {
+        dynarec_log(LOG_INFO, "AllocDynarecMap(%p, %zu) failed, cancelling block\n", block, sz);
+        CancelBlock(0);
+        return NULL;
+    }
+    block->size = sz;
+    block->actual_block = actual_p;
+    block->block = p;
+    block->jmpnext = p;
+    *(dynablock_t**)actual_p = block;
+    *(void**)(p+4*sizeof(void*)) = arm_epilog;
+    CreateJmpNext(block->jmpnext, p+4*sizeof(void*));
+    // all done...
+    __clear_cache(actual_p, actual_p+sz);   // need to clear the cache before execution...
+    return block;
+}
+
+void* FillBlock(dynablock_t* block, uintptr_t addr) {
+    /*
+        A Block must have this layout:
+
+        0x0000..0x0003  : dynablock_t* : self
+        0x0004..4+4*n   : actual Native instructions, (n is the total number)
+        B ..    B+3     : dynablock_t* : self (as part of JmpNext, that simulate another block)
+        B+4 ..  B+15    : 3 Native code for jmpnext (or jmp epilog in case of empty block)
+        B+16 .. B+19    : jmpnext (or jmp_epilog) address
+        B+20 .. B+31    : empty (in case an architecture needs more than 3 opcodes)
+        B+32 .. B+32+sz : instsize (compressed array with each instruction lenght on x86 and native side)
+
+    */
+dynarec_log(LOG_DEBUG, "Asked to Fill block %p with %p\n", block, (void*)addr);
+    if(IsInHotPage(addr)) {
+        dynarec_log(LOG_DEBUG, "Cancelling dynarec FillBlock on hotpage for %p\n", (void*)addr);
+        return NULL;
+    }
+    if(addr>=box86_nodynarec_start && addr<box86_nodynarec_end) {
+        dynarec_log(LOG_DEBUG, "Asked to fill a block in fobidden zone\n");
+        return CreateEmptyBlock(block, addr);
+    }
+    if(current_helper) {
+        dynarec_log(LOG_DEBUG, "Cancelling dynarec FillBlock at %p as anothor one is going on\n", (void*)addr);
+        return NULL;
+    }
+    // protect the 1st page
+    protectDB(addr, 1);
     // 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*)dynaCalloc(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
+    dynaFree(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);
-        return (void*)block;
+        CancelBlock(0);
+        return CreateEmptyBlock(block, addr);
     }
-    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);
+    if(!isprotectedDB(addr, 1)) {
+        dynarec_log(LOG_INFO, "Warning, write on current page on pass0, aborting dynablock creation (%p)\n", (void*)addr);
+        AddHotPage(addr);
+        CancelBlock(0);
+        return NULL;
+    }
+    // already protect the block and compute hash signature
+    // protect the block of it goes over the 1st page
+    if((addr&~box86_pagesize)!=(end&~box86_pagesize)) // need to protect some other pages too
+        protectDB(addr, end-addr);  //end is 1byte after actual end
+    uint32_t hash = X31_hash_code((void*)addr, end-addr);
     // calculate barriers
-    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 || j==helper.insts[i].x86.addr) {
+                if(j==helper.insts[i].x86.addr) // if there is a loop on some opcode, make the block "always to tested"
+                    helper.always_test = 1;
                 helper.insts[i].x86.jmp_insts = -1;
-            else {
+                helper.insts[i].x86.need_after |= X_PEND;
+            } else {
                 // find jump address instruction
                 int k=-1;
                 for(int i2=0; i2<helper.size && k==-1; ++i2) {
                     if(helper.insts[i2].x86.addr==j)
                         k=i2;
                 }
-                if(k!=-1)   // -1 if not found, mmm, probably wrong, exit anyway
-                    helper.insts[k].x86.barrier = 1;
+                if(k!=-1 && !helper.insts[i].barrier_maybe)
+                    helper.insts[k].x86.barrier |= BARRIER_FULL;
                 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;
-        }
-    
+    // fill predecessors with the jump address
+    fillPredecessors(&helper);
+
+    int pos = helper.size;
+    while (pos>=0)
+        pos = updateNeed(&helper, pos, 0);
+
+    // pass 1, float optimisations, first pass for flags
+    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);
-    if(p==NULL) {
-        free(helper.insts);
-        free(helper.next);
-        return (void*)block;
+    size_t insts_rsize = (helper.insts_size+2)*sizeof(instsize_t);
+    insts_rsize = (insts_rsize+7)&~7;   // round the size...
+    size_t arm_size = (helper.arm_size+7)&~7;   // round the size...
+    // ok, now allocate mapped memory, with executable flag on
+    size_t sz = sizeof(void*) + arm_size + 8*sizeof(void*) + insts_rsize;
+    //           dynablock_t* block (arm insts) jmpnext code    instsize
+    void* actual_p = (void*)AllocDynarecMap(sz);
+    void* p = (void*)(((uintptr_t)actual_p) + sizeof(void*));
+    void* next = p + arm_size;
+    void* instsize = next + 8*sizeof(void*);
+    if(actual_p==NULL) {
+        dynarec_log(LOG_INFO, "AllocDynarecMap(%p, %zu) failed, cancelling block\n", block, sz);
+        CancelBlock(0);
+        return NULL;
     }
     helper.block = p;
+    block->actual_block = actual_p;
     helper.arm_start = (uintptr_t)p;
-    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)
+    helper.jmp_next = (uintptr_t)next+sizeof(void*);
+    helper.instsize = (instsize_t*)instsize;
+    *(dynablock_t**)actual_p = block;
+    // pass 3, emit (log emit native opcode)
     if(box86_dynarec_dump) {
-        dynarec_log(LOG_NONE, "%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":"");
     }
+    size_t oldarmsize = helper.arm_size;
+    size_t oldinstsize = helper.insts_size;
     helper.arm_size = 0;
+    helper.insts_size = 0;  // reset
     arm_pass3(&helper, addr);
-    if(sz!=helper.arm_size) {printf_log(LOG_NONE, "BOX86: Warning, size difference in block between pass2 (%d) & pass3 (%d)!\n", sz, helper.arm_size);}
-    // all done...
-    __builtin___clear_cache(p, p+sz);   // need to clear the cache before execution...
-    free(helper.insts);
-    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;
+    // keep size of instructions for signal handling
+    block->instsize = instsize;
+    // ok, free the helper now
+    dynaFree(helper.insts);
+    helper.insts = NULL;
+    helper.instsize = NULL;
+    dynaFree(helper.predecessor);
+    helper.predecessor = NULL;
     block->size = sz;
     block->isize = helper.size;
     block->block = p;
-    block->need_test = 0;
-    block->x86_addr = (void*)start;
+    block->jmpnext = next+sizeof(void*);
+    block->always_test = helper.always_test;
+    block->dirty = block->always_test;
+    *(dynablock_t**)next = block;
+    *(void**)(next+5*sizeof(void*)) = arm_next;
+    CreateJmpNext(block->jmpnext, next+5*sizeof(void*));
+    //block->x86_addr = (void*)start;
     block->x86_size = end-start;
-    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;
-    // fill sons if any
-    dynablock_t** sons = NULL;
-    int sons_size = 0;
-    if(helper.sons_size) {
-        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...
-                son->block = helper.sons_arm[i];
-                son->x86_addr = (void*)helper.sons_x86[i];
-                son->x86_size = end-helper.sons_x86[i];
-                if(!son->x86_size) {printf_log(LOG_NONE, "Warning, son with null x86 size! (@%p / ARM=%p)", son->x86_addr, son->block);}
-                son->father = block;
-                son->done = 1;
-                sons[sons_size++] = son;
-            }
-        }
-        if(sons_size) {
-            block->sons = sons;
-            block->sons_size = sons_size;
-        } else
-            free(sons);
+    // all done...
+    __clear_cache(actual_p, actual_p+sz);   // need to clear the cache before execution...
+    block->hash = X31_hash_code(block->x86_addr, block->x86_size);
+    // Check if something changed, to abbort if it as
+    if((block->hash != hash)) {
+        dynarec_log(LOG_DEBUG, "Warning, a block changed while being processed hash(%p:%zu)=%x/%x\n", block->x86_addr, block->x86_size, block->hash, hash);
+        AddHotPage(addr);
+        CancelBlock(0);
+        return NULL;
     }
-    if(block->parent->nolinker)
-        protectDB((uintptr_t)block->x86_addr, block->x86_size);
-    block->done = 1;
+    if((oldarmsize!=helper.arm_size)) {
+        printf_log(LOG_NONE, "BOx86: Warning, size difference in block between pass2 (%zu) & pass3 (%zu)!\n", sz, helper.arm_size);
+        uint8_t *dump = (uint8_t*)helper.start;
+        printf_log(LOG_NONE, "Dump of %d x86 opcodes:\n", helper.size);
+        for(int i=0; i<helper.size; ++i) {
+            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");
+        CancelBlock(0);
+        return NULL;
+    }
+    if(insts_rsize/sizeof(instsize_t)<helper.insts_size) {
+        printf_log(LOG_NONE, "BOX86: Warning, ists_size difference in block between pass2 (%zu) and pass3 (%zu), allocated: %zu\n", oldinstsize, helper.insts_size, insts_rsize/sizeof(instsize_t));
+    }
+    if(!isprotectedDB(addr, end-addr)) {
+        dynarec_log(LOG_DEBUG, "Warning, block unprotected while being processed %p:%zu, marking as need_test\n", block->x86_addr, block->x86_size);
+        AddHotPage(addr);
+        block->dirty = 1;
+        //protectDB(addr, end-addr);
+    }
+    current_helper = NULL;
+    //block->done = 1;
     return (void*)block;
-}
+}

src/dynarec/dynarec_arm_64.c

@@ -19,77 +19,224 @@
 #include "dynarec_arm_private.h"
 #include "arm_printer.h"
 
+#include "dynarec_arm_functions.h"
 #include "dynarec_arm_helper.h"
 
 uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog)
 {
     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, u8;
     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);
+                        SMWRITE2();
+                    }
+                    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
+                        SMREAD();
+                        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);
+                        SMWRITE2();
+                    }
+                    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);
+                    }
+                    SMREAD();
+                    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);
+                    SMWRITE2();
+                    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
+                                SMREAD();
+                                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, NULL);
+                STR_REG_LSL_IMM5(gd, ed, x14, 0);
+                SMWRITE2();
             }
             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);
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
+                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, NULL);
+                POP1(x2);
+                STR_REG_LSL_IMM5(x2, ed, x14, 0);
+                SMWRITE2();
             }
             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);
+                SMREAD();
+                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 +245,104 @@ 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);
+                SMREAD();
+                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);
+                SMWRITE2();
             }
             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);
+                SMWRITE2();
+            } else {
+                MOV32(x2, i32);
+                STR_REG_LSL_IMM5(xEAX, x2, x1, 0);
+                SMWRITE2();
+            }
+            break;
+
+        case 0xA8:
+            INST_NAME("TEST AL, Ib");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            UXTB(x1, xEAX, 0);
+            u8 = F8;
+            MOVW(x2, u8);
+            emit_test8(dyn, ninst, x1, x2, x3, x14);
             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, NULL);
                 i32 = F32S;
                 MOV32(x3, i32);
-                STR_REG_LSL_IMM5(x3, ed, x12, 0);
+                STR_REG_LSL_IMM5(x3, ed, x14, 0);
+                SMWRITE2();
             }
             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((box86_dynarec_safeflags>1) ||
+                        ((ninst && dyn->insts[ninst-1].x86.set_flags)
+                        || ((ninst>1) && dyn->insts[ninst-2].x86.set_flags)), 1)
+                    {
+                        READFLAGS(X_PEND);          // that's suspicious
+                    } else {
+                        SETFLAGS(X_ALL, SF_SET);    //Hack to put flag in "don't care" state
+                    }
+                    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, NULL);
+                        LDR_REG_LSL_IMM5(xEIP, ed, x14, 0);
+                    }
+                    BARRIER(BARRIER_FULL);
+                    *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);
+                        SMREAD();
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
+                        LDR_REG_LSL_IMM5(x3, ed, x14, 0);
+                        PUSH1(x3);
                     }
                     break;
                 default:

src/dynarec/dynarec_arm_65.c

@@ -19,71 +19,390 @@
 #include "dynarec_arm_private.h"
 #include "arm_printer.h"
 
+#include "dynarec_arm_functions.h"
 #include "dynarec_arm_helper.h"
 
 uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog)
 {
     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, eb1, eb2;
+    uint8_t u8;
     int fixedaddress;
+
+    MAYUSE(j32);
+    MAYUSE(wback);
+    MAYUSE(wb2);
+    MAYUSE(wb);
+    MAYUSE(eb1);
+    MAYUSE(eb2);
+    
     switch(opcode) {
-        case 0x03:
-            INST_NAME("ADD Gd, GS:Ed");
-            SETFLAGS(X_ALL, SF_SET);
-            grab_tlsdata(dyn, addr, ninst, x12);
+
+        case 0x01:
+            INST_NAME("ADD GS:Ed, Gd");
+            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);
+            GETEDO2(x14);
+            emit_add32(dyn, ninst, ed, gd, x3, x14);
+            WBACK2;
+            break;
+
+        case 0x03:
+            INST_NAME("ADD Gd, GS:Ed");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_tlsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGD;
+            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);
+            WBACK2;
+            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);
+            WBACK2;
             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);
+            WBACK2;
             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 0x39:
+            INST_NAME("CMP GS:Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_tlsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGD;
+            GETEDO(x14);
+            emit_cmp32(dyn, ninst, ed, gd, 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);
+                    EBBACK2;
+                    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);
+                    EBBACK2;
+                    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);
+                    EBBACK2;
+                    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);
+                    EBBACK2;
+                    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);
+                    EBBACK2;
+                    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);
+                    EBBACK2;
+                    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);
+                    EBBACK2;
+                    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);
+                    WBACK2;
+                    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);
+                    WBACK2;
+                    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);
+                    WBACK2;
+                    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);
+                    WBACK2;
+                    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);
+                    WBACK2;
+                    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);
+                    WBACK2;
+                    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);
+                    WBACK2;
+                    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, NULL);
+                STR_REG_LSL_IMM5(gd, ed, x14, 0);
+                SMWRITE2();
             }
             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);
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
+                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, NULL);
+                POP1(x2);
+                STR_REG_LSL_IMM5(x2, ed, x14, 0);
+                SMWRITE2();
             }
             break;
 
@@ -91,6 +410,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             INST_NAME("MOV EAX, GS:Id");
             grab_tlsdata(dyn, addr, ninst, x1);
             i32 = F32S;
+            SMREAD();
             if(i32>-4096 && i32<4096) {
                 LDR_IMM9(xEAX, x1, i32);
             } else {
@@ -100,6 +420,16 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             }
             break;
 
+        case 0xA2:
+            INST_NAME("MOV GS:Od, AL");
+            grab_tlsdata(dyn, addr, ninst, x1);
+            u32 = F32;
+            MOV32(x2, u32);
+            ADD_REG_LSL_IMM5(x2, x1, x2, 0);
+            STRB_IMM9(xEAX, x2, 0);
+            SMWRITE2();
+            break;
+
         case 0xA3:
             INST_NAME("MOV GS:Od, EAX");
             grab_tlsdata(dyn, addr, ninst, x1);
@@ -107,21 +437,230 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             MOV32(x2, u32);
             ADD_REG_LSL_IMM5(x2, x1, x2, 0);
             STR_IMM9(xEAX, x2, 0);
+            SMWRITE2();
             break;
 
+        case 0xC6:
+            INST_NAME("MOV GS:Eb, Ib");
+            grab_tlsdata(dyn, addr, ninst, x14);
+            nextop=F8;
+            if((nextop&0xC0)==0xC0) {   // reg <= u8
+                u8 = F8;
+                ed = (nextop&7);
+                eb1 = xEAX+(ed&3);  // Ax, Cx, Dx or Bx
+                eb2 = (ed&4)>>2;    // L or H
+                MOVW(x3, u8);
+                BFI(eb1, x3, eb2*8, 8);
+            } else {                    // mem <= u8
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
+                u8 = F8;
+                MOVW(x3, u8);
+                STRB_REG_LSL_IMM5(x3, ed, x14, 0);
+                SMWRITE2();
+            }
+            break;
         case 0xC7:
             INST_NAME("MOV GS:Ed, Id");
-            grab_tlsdata(dyn, addr, ninst, 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, NULL);
                 i32 = F32S;
                 MOV32(x3, i32);
-                STR_REG_LSL_IMM5(x3, ed, x12, 0);
+                STR_REG_LSL_IMM5(x3, ed, x14, 0);
+                SMWRITE2();
+            }
+            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_PENDING);
+                    GETEDO2(x14);
+                    emit_rol32c(dyn, ninst, ed, 1, x3, x14);
+                    WBACK2;
+                    break;
+                case 1:
+                    INST_NAME("ROR Ed, 1");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
+                    GETEDO2(x14);
+                    emit_ror32c(dyn, ninst, ed, 1, x3, x14);
+                    WBACK2;
+                    break;
+                case 2:
+                    INST_NAME("RCL Ed, 1");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    MOVW(x2, 1);
+                    GETEDO2(x14);
+                    if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
+                    CALL_(rcl32, ed, (1<<x2));
+                    WBACK2;
+                    break;
+                case 3:
+                    INST_NAME("RCR Ed, 1");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    MOVW(x2, 1);
+                    if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
+                    CALL_(rcr32, ed, (1<<x2));
+                    WBACK2;
+                    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);
+                    WBACK2;
+                    break;
+                case 5:
+                    INST_NAME("SHR Ed, 1");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                    GETEDO2(x14);
+                    emit_shr32c(dyn, ninst, ed, 1, x3, x14);
+                    WBACK2;
+                    break;
+                case 7:
+                    INST_NAME("SAR Ed, 1");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                    GETEDO2(x14);
+                    emit_sar32c(dyn, ninst, ed, 1, x3, x14);
+                    WBACK2;
+                    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);
+                    WBACK2;
+                    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);
+                    WBACK2;
+                    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);
+                    if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
+                    CALL_(rcl32, ed, (1<<x2));
+                    WBACK2;
+                    break;
+                case 3:
+                    INST_NAME("RCR Ed, CL");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    AND_IMM8(x2, xECX, 0x1f);
+                    if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
+                    CALL_(rcr32, ed, (1<<x14));
+                    WBACK2;
+                    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);
+                    WBACK2;
+                    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);
+                    WBACK2;
+                    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);
+                    WBACK2;
+                    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
+                        SMREAD();
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
+                        LDR_REG_LSL_IMM5(x3, ed, x14, 0);
+                        PUSH1(x3);
+                    }
+                    break;
+                default:
+                    DEFAULT;
             }
             break;
 

src/dynarec/dynarec_arm_66.c

@@ -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)
@@ -27,42 +28,45 @@ 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;
     uint8_t wback, wb1;
+    int lock;
     int fixedaddress;
     MAYUSE(u16);
     MAYUSE(u8);
+    MAYUSE(j32);
+    MAYUSE(lock);
     while(opcode==0x66) opcode = F8;    // "unlimited" 0x66 as prefix for variable sized NOP
-    if(opcode==0x2E) opcode = F8;       // cs: is ignored
+    if((opcode==0x26) || (opcode==0x2E) || (opcode==0x36) || (opcode==0x3E)) opcode = F8;       // segment prefix are 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 +85,33 @@ 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 +121,93 @@ 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 0x1E:
+            INST_NAME("PUSH DS");
+            MOVW(x1, offsetof(x86emu_t, segs[_DS]));
+            LDRH_REG(x2, xEmu, x1);
+            PUSH16(x2, xESP);
+            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 +217,81 @@ 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:
@@ -294,10 +303,10 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
         case 0x46:
         case 0x47:
             INST_NAME("INC Reg16");
-            SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
+            SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
             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:
@@ -309,10 +318,10 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
         case 0x4E:
         case 0x4F:
             INST_NAME("DEC Reg16");
-            SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
+            SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
             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 +333,7 @@ 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 +344,45 @@ 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 +400,7 @@ 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 +425,116 @@ 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("(LOCK) 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);
+                SMDMB();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, LOCK_LOCK);
+                #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
+                SMDMB();
+                BFI(gd, x1, 0, 16);
+            }
             break;
 
         case 0x89:
@@ -479,8 +547,9 @@ 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, &lock);
                 STRH_IMM8(gd, ed, fixedaddress);
+                SMWRITELOCK(lock);
             }
             break;
         case 0x8B:
@@ -493,7 +562,8 @@ 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, &lock);
+                SMREADLOCK(lock);
                 LDRH_IMM8(x1, ed, fixedaddress);
                 BFI(gd, x1, 0, 16);
             }
@@ -501,14 +571,15 @@ 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, NULL);
                 STRH_IMM8(x1, ed, fixedaddress);
+                SMWRITE2();
             }
             break;
 
@@ -517,16 +588,17 @@ 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);
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0, NULL);
                 LDRH_IMM8(x1, ed, fixedaddress);
-                MOV32(x2, offsetof(x86emu_t, segs[(nextop&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:
@@ -545,17 +617,39 @@ 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");
             u32 = F32;
             MOV32(x2, u32);
+            SMREAD();
             LDRH_IMM8(x2, x2, 0);
             BFI(xEAX, x2, 0, 16);
             break;
@@ -564,6 +658,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             u32 = F32;
             MOV32(x2, u32);
             STRH_IMM8(xEAX, x2, 0);
+            SMWRITE();
             break;
         case 0xA5:
             INST_NAME("MOVSW");
@@ -573,20 +668,20 @@ 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:
@@ -602,10 +697,11 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             break;
         case 0xAF:
             INST_NAME("SCASW");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
             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 +808,7 @@ 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, NULL);
                 u16 = F16;
                 MOVW(x1, u16);
                 STRH_IMM8(x1, ed, fixedaddress);
@@ -771,15 +867,15 @@ 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 +883,15 @@ 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 +899,15 @@ 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);
@@ -819,6 +915,10 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             }
             break;
 
+        case 0xF0:
+            return dynarec66F0(dyn, addr, ip, ninst, ok, need_epilog);
+            break;
+
         case 0xF2:
         case 0xF3:
             nextop = F8;
@@ -849,7 +949,8 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
 
                 case 0xA7:
                     if(opcode==0xF2) {INST_NAME("REPNZ CMPSW");} else {INST_NAME("REPZ CMPSW");}
-                    SETFLAGS(X_ALL, SF_MAYSET);
+                    MAYSETFLAGS();
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     TSTS_REG_LSL_IMM5(xECX, xECX, 0);
                     B_NEXT(cEQ);    // end of loop
                     GETDIR(x3, 2);
@@ -869,7 +970,7 @@ 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:
@@ -885,7 +986,8 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
 
                 case 0xAF:
                     if(opcode==0xF2) {INST_NAME("REPNZ SCASW");} else {INST_NAME("REPZ SCASW");}
-                    SETFLAGS(X_ALL, SF_MAYSET);
+                    MAYSETFLAGS();
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     TSTS_REG_LSL_IMM5(xECX, xECX, 0);
                     B_NEXT(cEQ);    // end of loop
                     GETDIR(x3, 2);
@@ -905,24 +1007,31 @@ 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;
             }
             break;
 
+        case 0xF5:
+            INST_NAME("CMC");
+            READFLAGS(X_CF);
+            SETFLAGS(X_CF, SF_SUBSET);
+            XOR_IMM8(xFlags, xFlags, 1<<F_CF);
+            break;
+
         case 0xF7:
             nextop = F8;
             switch((nextop>>3)&7) {
                 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 +1041,14 @@ 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 +1068,27 @@ 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, 0, 16);
+                        BFI(xEDX, x14, 0, 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));
@@ -974,21 +1097,35 @@ 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) {
                 case 0:
                     INST_NAME("INC Ew");
-                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
+                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
                     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_PENDING);
                     GETEW(x1);
-                    emit_dec16(dyn, ninst, x1, x2, x12);
+                    emit_dec16(dyn, ninst, x1, x2, x14);
                     EWBACK;
                     break;
                 default:

src/dynarec/dynarec_arm_66f0.c

@@ -0,0 +1,188 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <pthread.h>
+#include <errno.h>
+
+#include "debug.h"
+#include "box86context.h"
+#include "dynarec.h"
+#include "emu/x86emu_private.h"
+#include "emu/x86run_private.h"
+#include "x86run.h"
+#include "x86emu.h"
+#include "box86stack.h"
+#include "callback.h"
+#include "emu/x86run_private.h"
+#include "x86trace.h"
+#include "dynarec_arm.h"
+#include "dynarec_arm_private.h"
+#include "arm_printer.h"
+
+#include "dynarec_arm_functions.h"
+#include "dynarec_arm_helper.h"
+
+
+uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog)
+{
+    uint8_t nextop, opcode = F8;
+    uint8_t wback, wb1, wb2, gb1, gb2;
+    uint8_t ed, gd, u8;
+    int fixedaddress;
+    int32_t i32, j32;
+    int16_t i16;
+    MAYUSE(i32);
+    MAYUSE(j32);
+    MAYUSE(gb1);
+    MAYUSE(gb2);
+    MAYUSE(wb1);
+    MAYUSE(wb2);
+    switch(opcode) {
+        case 0x81:
+        case 0x83:
+            nextop = F8;
+            SMDMB();
+            switch((nextop>>3)&7) {
+                case 0: //ADD
+                    if(opcode==0x81) {
+                        INST_NAME("LOCK ADD Ew, Iw");
+                    } else {
+                        INST_NAME("LOCK ADD Ew, Ib");
+                    }
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    SMDMB();
+                    if((nextop&0xC0)==0xC0) {
+                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
+                        ed = xEAX+(nextop&7);
+                        emit_add16c(dyn, ninst, ed, i16, x3, x14);
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 1, LOCK_LOCK);
+                        if(opcode==0x81) i16 = F16S; else i16 = F8S;
+                        if(!fixedaddress) {
+                            TSTS_IMM8(wback, 0x1);
+                            B_MARK(cNE);
+                        }
+                        if(!fixedaddress || (fixedaddress && !(fixedaddress&1))) {
+                            MARKLOCK;
+                            LDREXH(x1, wback);
+                            emit_add16c(dyn, ninst, x1, i16, x3, x14);
+                            STREXH(x3, x1, wback);
+                            CMPS_IMM8(x3, 0);
+                            B_MARKLOCK(cNE);
+                        }
+                        if(!fixedaddress) {
+                            SMDMB();
+                            B_NEXT(c__);
+                        }
+                        if(!fixedaddress || (fixedaddress && (fixedaddress&1))) {
+                            MARK;   // unaligned! also, not enough
+                            LDRH_IMM8(x1, wback, 0);
+                            LDREXB(x14, wback);
+                            BFI(x1, x14, 0, 8); // re-inject
+                            emit_add16c(dyn, ninst, x1, i16, x3, x14);
+                            STREXB(x3, x1, wback);
+                            CMPS_IMM8(x3, 0);
+                            B_MARK(cNE);
+                            STRH_IMM8(x1, wback, 0); // put the whole value
+                        }
+                        SMDMB();
+                    }
+                    break;
+                default:
+                  DEFAULT;
+            }
+            SMDMB();
+            break;
+        case 0xFF:
+            nextop = F8;
+            SMDMB();
+            switch((nextop>>3)&7)
+            {
+                case 0: // INC Ew
+                    INST_NAME("INC Ew");
+                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
+                    if((nextop&0xC0)==0xC0) {
+                        GETEW(x1);
+                        emit_inc16(dyn, ninst, ed, x3, x14);
+                        EWBACK(x1);
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 1, LOCK_LOCK);
+                        if(!fixedaddress) {
+                            TSTS_IMM8(wback, 1);
+                            B_MARK(cNE);    // unaligned
+                        }
+                        if(!fixedaddress || (fixedaddress && !(fixedaddress&1))) {
+                            MARKLOCK;
+                            LDREXH(x1, wback);
+                            emit_inc16(dyn, ninst, x1, x3, x14);
+                            STREXH(x3, x1, wback);
+                            CMPS_IMM8(x3, 0);
+                            B_MARKLOCK(cNE);
+                        }
+                        if(!fixedaddress) {
+                            SMDMB();
+                            B_NEXT(c__);
+                        }
+                        if(!fixedaddress || (fixedaddress && (fixedaddress&1))) {
+                            MARK;
+                            LDRH_IMM8(x1, wback, 0);
+                            LDREXB(x3, wback);
+                            BFI(x1, x3, 0, 8);
+                            emit_inc16(dyn, ninst, x1, x3, x14);
+                            STREXB(x3, x1, wback);
+                            CMPS_IMM8(x3, 0);
+                            B_MARK(cNE);
+                            STRH_IMM8(x1, wback, 0);
+                        }
+                    }
+                    break;
+                case 1: //DEC Ew
+                    INST_NAME("DEC Ew");
+                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
+                    if((nextop&0xC0)==0xC0) {
+                        GETEW(x1);
+                        emit_dec32(dyn, ninst, ed, x3, x14);
+                        EWBACK(x1);
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 1, LOCK_LOCK);
+                        if(!fixedaddress) {
+                            TSTS_IMM8(wback, 1);
+                            B_MARK(cNE);    // unaligned
+                        }
+                        if(!fixedaddress || (fixedaddress && !(fixedaddress&1))) {
+                            MARKLOCK;
+                            LDREXH(x1, wback);
+                            emit_dec16(dyn, ninst, x1, x3, x14);
+                            STREXH(x3, x1, wback);
+                            CMPS_IMM8(x3, 0);
+                            B_MARKLOCK(cNE);
+                        }
+                        if(!fixedaddress) {
+                            SMDMB();
+                            B_NEXT(c__);
+                        }
+                        if(!fixedaddress || (fixedaddress && (fixedaddress&3))) {
+                            MARK;
+                            LDRH_IMM8(x1, wback, 0);
+                            LDREXB(x3, wback);
+                            BFI(x1, x3, 0, 8);
+                            emit_dec16(dyn, ninst, x1, x3, x14);
+                            STREXB(x3, x1, wback);
+                            CMPS_IMM8(x3, 0);
+                            B_MARK(cNE);
+                            STRH_IMM8(x1, wback, 0);
+                        }
+                    }
+                    break;
+                default:
+                    // default to NO LOCK
+                    addr-=2;
+            }
+            SMDMB();
+            break;
+        default:
+            DEFAULT;
+    }
+
+    return addr;
+}

src/dynarec/dynarec_arm_67.c

@@ -19,33 +19,149 @@
 #include "dynarec_arm_private.h"
 #include "arm_printer.h"
 
+#include "dynarec_arm_functions.h"
 #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;
+    int cacheupd;
+
     MAYUSE(i32);
+    MAYUSE(j32);
+    MAYUSE(cacheupd);
+    
     nextop = F8;
+
     switch(nextop) {
        
 
-        #define GO(NO, YES)   \
-            BARRIER(2); \
-            JUMP(addr+i8);\
-            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); \
-                } else {    \
-                    /* inside the block */  \
-                    i32 = dyn->insts[dyn->insts[ninst].x86.jmp_insts].address-(dyn->arm_size+8);    \
-                    Bcond(YES, i32);    \
-                }   \
+        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);
+                        SMWRITE2();
+                    }
+                    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
+                        SMREAD();
+                        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);
+                        SMWRITE2();
+                    }
+                    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);
+                    }
+                    SMREAD();
+                    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);
+                    SMWRITE2();
+                    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
+                                SMREAD();
+                                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(BARRIER_MAYBE);                                     \
+            JUMP(addr+i8, 1);                                           \
+            if(dyn->insts[ninst].x86.jmp_insts==-1 ||                   \
+                CHECK_CACHE()) {                                        \
+                /* out of the block */                                  \
+                i32 = dyn->insts[ninst].epilog-(dyn->arm_size+8);       \
+                Bcond(NO, i32);                                         \
+                if(dyn->insts[ninst].x86.jmp_insts==-1) {               \
+                    if(!(dyn->insts[ninst].x86.barrier&BARRIER_FLOAT))  \
+                        fpu_purgecache(dyn, ninst, 1, x1, x2, x3);      \
+                    jump_to_next(dyn, addr+i8, 0, ninst);               \
+                } else {                                                \
+                    CacheTransform(dyn, ninst, cacheupd, x1, x2, x3);   \
+                    i32 = dyn->insts[dyn->insts[ninst].x86.jmp_insts].address-(dyn->arm_size+8);\
+                    Bcond(c__, i32);                                    \
+                }                                                       \
+            } else {                                                    \
+                /* inside the block */                                  \
+                i32 = dyn->insts[dyn->insts[ninst].x86.jmp_insts].address-(dyn->arm_size+8);    \
+                Bcond(YES, i32);                                        \
             }
         case 0xE0:
             INST_NAME("LOOPNZ (16bits)");
@@ -55,8 +171,7 @@ 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 +182,7 @@ 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:

src/dynarec/dynarec_arm_d8.c

@@ -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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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;

src/dynarec/dynarec_arm_d9.c

@@ -23,28 +23,38 @@
 #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;
 
+extern int arm_v8;
+
+static const void* round_map[4] = {nearbyint, floor, ceil, trunc};
 uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog)
 {
     uint8_t nextop = F8;
     uint8_t ed;
     uint8_t wback, wb1;
+    uint8_t u8;
     int fixedaddress;
     int v1, v2;
     int s0;
     int i1, i2, i3;
+    int j32;
     int parity;
 
     MAYUSE(s0);
     MAYUSE(v2);
     MAYUSE(v1);
+    MAYUSE(j32);
+    (void)u8;
 
     switch(nextop) {
 
@@ -57,12 +67,19 @@ 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);
+            }
+            // should set C1 to 0
             break;
 
         case 0xC8:
+            INST_NAME("FXCH ST0");
+            break;
         case 0xC9:
         case 0xCA:
         case 0xCB:
@@ -72,11 +89,8 @@ 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);
-            i3 = dyn->x87cache[i1];
-            dyn->x87cache[i1] = dyn->x87cache[i2];
-            dyn->x87cache[i2] = i3;
+            x87_swapreg(dyn, ninst, x1, x2, 0, nextop&7);
+            // should set C1 to 0
             break;
 
         case 0xD0:
@@ -85,160 +99,447 @@ 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);
+            }
+            // should set C1 to 0
             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);
+            }
+            // should set C1 to 0
             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->n.x87stack;
+                if(i2) {
+                    if(i2<0) {
+                        SUB_IMM8(x14, x14, -i2);
+                    } else {
+                        ADD_IMM8(x14, x14, i2);
+                    }
+                    AND_IMM8(x14, x14, 7);    // (emu->top + i)&7
+                }
+                ADD_REG_LSL_IMM5(x1, xEmu, x14, 3);
+                LDRD_IMM8(x2, x1, offsetof(x86emu_t, x87)); // load r2/r3 with ST0 anyway, for sign extraction
+                ADD_REG_LSL_IMM5(x1, xEmu, x14, 2);
+                LDR_IMM9(x1, x1, offsetof(x86emu_t, p_regs));
+                CMPS_IMM8(x1, 0b11);
+                MOVW_COND(cEQ, x14, 0b100000100000000); // empty: C3,C2,C0 = 101
+                B_MARK3(cEQ);
+            } else {
+                // simply move from cache reg to r2/r3
+                v1 = dyn->n.x87reg[i1];
+                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, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                VMOV_i_32(v1, 0b01110000);
+            } else {
+                VMOV_i_64(v1, 0b01110000);
+            }
+            // should set C1 to 0
             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, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                MOV32(x2, (&f_l2t));
+                VLDR_32(v1, x2, 0);
+            } else {
+                MOV32(x2, (&d_l2t));
+                VLDR_64(v1, x2, 0);
+            }
+            // should set C1 to 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, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                MOV32(x2, (&f_l2e));
+                VLDR_32(v1, x2, 0);
+            } else {
+                MOV32(x2, (&d_l2e));
+                VLDR_64(v1, x2, 0);
+            }
+            // should set C1 to 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, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                MOV32(x2, (&f_pi));
+                VLDR_32(v1, x2, 0);
+            } else {
+                MOV32(x2, (&d_pi));
+                VLDR_64(v1, x2, 0);
+            }
+            // should set C1 to 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, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                MOV32(x2, (&f_lg2));
+                VLDR_32(v1, x2, 0);
+            } else {
+                MOV32(x2, (&d_lg2));
+                VLDR_64(v1, x2, 0);
+            }
+            // should set C1 to 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, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                MOV32(x2, (&f_ln2));
+                VLDR_32(v1, x2, 0);
+            } else {
+                MOV32(x2, (&d_ln2));
+                VLDR_64(v1, x2, 0);
+            }
+            // should set C1 to 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, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                VMOV_8(v1/2, 0);  // float is *2...
+            } else {
+                VMOV_8(v1, 0);  // cannot use VMOV_i_F64
+            }
+            // should set C1 to 0
             break;
 
-        case 0xFA:
-            INST_NAME("FSQRT");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            VSQRT_F64(v1, v1);
-            break;
-
-        case 0xFC:
-            INST_NAME("FRNDINT");
-            // use C helper for now, nothing staightforward is available
-            x87_forget(dyn, ninst, x1, x2, 0);
-            CALL(arm_frndint, -1, 0);
-            /*
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            VCMP_F64_0(v1);
-            VMRS_APSR();
-            B_NEXT(cVS);    // Unordered, skip
-            B_NEXT(cEQ);    // Zero, skip
-            u8 = x87_setround(dyn, ninst, x1, x2, x3);
-            VCVT_S32_F64(x1, v1);   // limit to 32bits....
-            VCVT_F64_S32(v1, x1);
-            x87_restoreround(dyn, ninst, u8);
-            */
-            break;
         case 0xF0:
             INST_NAME("F2XM1");
+            #if 0
+            MESSAGE(LOG_DUMP, "Need Optimization\n");
             x87_forget(dyn, ninst, x1, x2, 0);
             CALL(arm_f2xm1, -1, 0);
+            #else
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            //if(ST0.d!=0.0)
+            //    ST0.d = exp2(ST0.d)-1.0;
+            VMOV_64(0, v1);
+            CALL_1D(exp2, 0);   // return is d0
+            if((PK(0)==0xD9 && PK(1)==0xE8) && // next inst is FLD1
+            (PK(2)==0xDE && PK(3)==0xC1)) {
+                MESSAGE(LOG_DUMP, "Hack for fld1 / faddp st1, st0\n");
+                VMOV_64(v1, 0);
+                addr+=4;
+            } else {
+                VMOV_i_64(v1, 0b01110000);  // 1.0
+                VSUB_F64(v1, 0, v1);
+            }
+            #endif
+            // should set C1 to 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);
+            // should set C1 to 0
             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);
+            INST_NAME("FPTAN");
+            v2 = x87_do_push(dyn, ninst, x1, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
+            VMOV_64(0, v1);    // prepare call to tan
+            CALL_1D(tan, 0);
+            VMOV_64(v1, 0);
+            //emu->sw.f.F87_C2 = 0; 
+            //emu->sw.f.F87_C1 = 0; 
+            LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
+            BFC(x1, 9, 2); //C1 C2 = 0 0
+            STRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
+            // so here: F64: Imm8 = abcd efgh that gives => aBbbbbbb bbcdefgh 0000000 00000000 00000000...
+            // and want 1.0 = 0x3ff0000000000000
+            // so 00111111 11110000 00000000 00000000....
+            // a = 0, b = 1, c = 1, d = 1, efgh=0
+            // 0b01110000
+            if(ST_IS_F(0)) {
+                VMOV_i_32(v2, 0b01110000);
+            } else {
+                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);
+            // should set C1 to 0
             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);
+            // C1 set only if stack under/overflow occurs
+            break;
+        case 0xF5:
+            INST_NAME("FPREM1");
+            #if 0
+            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);
+            #else
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            v2 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
+            s0 = fpu_get_scratch_double(dyn);
+            VDIV_F64(s0, v1, v2);
+            // set rounding to Nearest
+                MOVW(x2, 1);    // Nearest
+                VMRS(x1);               // get fpscr
+                MOV_REG(x14, x1);
+                BFI(x1, x2, 22, 2);     // inject new round
+                VMSR(x1);               // put new fpscr
+            VCVTR_S32_F64(s0*2, s0); // => Q
+            // set back rounding
+                VMSR(x14);
+            // get Q
+            VMOVfrV(x14, s0*2);
+            // Put back Q in a double
+            VCVT_F64_S32(s0, s0*2);
+            VMUL_F64(s0, s0, v2);
+            VSUB_F64(v1, v1, s0);
+            LDR_IMM9(x1, xEmu, offsetof(x86emu_t, sw));
+            // set C2 = 0
+            BFC(x1, 10, 1);
+            // set C1 = Q0
+            BFI(x1, x14, 9, 1);
+            // set C3 = Q1
+            MOV_REG_LSR_IMM5(x14, x14, 1);
+            BFI(x1, x14, 14, 1);
+            // Set C0 = Q2
+            MOV_REG_LSR_IMM5(x14, x14, 1);
+            BFI(x1, x14, 8, 1);
+            STR_IMM9(x1, xEmu, offsetof(x86emu_t, sw));
+            #endif
+            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));
+            // should set C1 to 0
+            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));
+            // should set C1 to 0
             break;
         case 0xF8:
             INST_NAME("FPREM");
+            #if 0
+            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);
+            #else
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            v2 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
+            s0 = fpu_get_scratch_double(dyn);
+            VDIV_F64(s0, v1, v2);
+            VCVT_S32_F64(s0*2, s0); // => Q
+            // get Q
+            VMOVfrV(x14, s0*2);
+            // Put back Q in a double
+            VCVT_F64_S32(s0, s0*2);
+            VMUL_F64(s0, s0, v2);
+            VSUB_F64(v1, v1, s0);
+            LDR_IMM9(x1, xEmu, offsetof(x86emu_t, sw));
+            // set C2 = 0
+            BFC(x1, 10, 1);
+            // set C1 = Q0
+            BFI(x1, x14, 9, 1);
+            // set C3 = Q1
+            MOV_REG_LSR_IMM5(x14, x14, 1);
+            BFI(x1, x14, 14, 1);
+            // Set C0 = Q2
+            MOV_REG_LSR_IMM5(x14, x14, 1);
+            BFI(x1, x14, 8, 1);
+            STR_IMM9(x1, xEmu, offsetof(x86emu_t, sw));
+            #endif
             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);
+            // should set C1 to 0
+            break;
+        case 0xFA:
+            INST_NAME("FSQRT");
+            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);
+            }
+            // should set C1 to 0
             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);
+            v2 = x87_do_push(dyn, ninst, x3, NEON_CACHE_ST_D);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
+            VMOV_64(0, v1);
+            CALL_1D(sin, 0);
+            VSWP(v1, 0);
+            CALL_1D(cos, 0);    // would it be faster to do sqrt(1-sin()²) ???
+            VMOV_64(v2, 0);
+            //emu->sw.f.F87_C2 = 0; C1 too
+            LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
+            BFC(x1, 9, 2); //C2 C1 = 0 0
+            STRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
+            break;
+        case 0xFC:
+            INST_NAME("FRNDINT");
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            if(!arm_v8) {
+                // check if finite first
+                VCMP_F64_0(v1);
+                VMRS_APSR();
+                B_NEXT(cVS);    // Unordered, skip
+                B_NEXT(cEQ);    // Zero, skip
+                // load round mode
+                LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, cw));    // hopefully cw is not too far for an imm8
+                UBFX(x2, x1, 10, 2);    // extract round...
+                MOV32(x1, round_map);
+                LDR_REG_LSL_IMM5(x2, x1, x2, 2);
+                VMOV_64(0, v1);    // prepare call to fpu_round
+                CALL_1DR(x2, x3, 0);
+                VMOV_64(v1, 0);
+            } else {
+                u8 = x87_setround(dyn, ninst, x1, x2, x14);
+                VRINTR_F64(v1, v1);
+                x87_restoreround(dyn, ninst, u8);
+            }
+            // should set C1 to 0
             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);
+            if(!arm_v8) {
+                VMOV_64(0, v2);
+                CALL_1DD(trunc, exp2, 0);
+            } else {
+                VRINTZ_F64(0, v2);
+                CALL_1D(exp2, 0);
+            }
+            VMUL_F64(v1, v1, 0);
+            // should set C1 to 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; C1 too
+            LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
+            BFC(x1, 9, 2); //C2 C1 = 0 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; C1 too
+            LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
+            BFC(x1, 9, 2); //C2 C1 = 0 0
+            STRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
             break;
 
 
@@ -260,9 +561,6 @@ 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 +568,69 @@ 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, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+                    if(ST_IS_F(0))
+                        s0 = v1;
+                    else
+                        s0 = fpu_get_scratch_single(dyn);
                     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, NULL);
                         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, NULL);
                         LDR_IMM9(x2, ed, fixedaddress);
                         VMOVtoV(s0, x2);
                     }
-                    VCVT_F64_F32(v1, s0);
+                    if(!ST_IS_F(0)) {
+                        VCVT_F64_F32(v1, s0);
+                    }
+                    // should set C1 to 0
                     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, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+                    if(ST_IS_F(0))
+                        s0 = v1;
+                    else {
+                        s0 = fpu_get_scratch_single(dyn);
+                        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, NULL);
                         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, NULL);
                         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, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+                    if(ST_IS_F(0))
+                        s0 = v1;
+                    else {
+                        s0 = fpu_get_scratch_single(dyn);
+                        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, NULL);
                         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, NULL);
                         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, NULL);
                     if(ed!=x1) {
                         MOV_REG(x1, ed);
                     }
@@ -328,12 +641,21 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     INST_NAME("FLDCW Ew");
                     GETEW(x1);
                     STRH_IMM8(x1, xEmu, offsetof(x86emu_t, cw));    // hopefully cw is not too far for an imm8
-                    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, NULL);
+                    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, NULL);
                     ed = x1;
                     wb1 = 1;
                     LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, cw));

src/dynarec/dynarec_arm_da.c

@@ -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);

src/dynarec/dynarec_arm_db.c

@@ -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 +107,29 @@ 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));
-            BFI(x2, x1, 0, 8);  // IE .. PE, SF, ES
-            BFI(x2, x1, 15, 1); // B
+            BFC(x2, 0, 8);  // IE .. PE, SF, ES
+            BFC(x2, 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 +142,13 @@ 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 +161,17 @@ 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 +183,11 @@ 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, NULL);
                         VLDR_32(s0, ed, fixedaddress);
                     } else {
                         GETED;
@@ -176,18 +197,18 @@ 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, NULL);
                         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, 0b010, 9); // enable exceptions
                     BIC_IMM8(x3, x3, 0b10011111, 0);
                     VMSR(x3);
                     VCVT_S32_F64(s0, v1);
@@ -196,26 +217,22 @@ 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_reset(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);
-                        ed = x1;
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0, 0, NULL);
+                        ed = x1; // will not be writen immediatly, so x1 is safe for now
                     }
                     s0 = fpu_get_scratch_single(dyn);
                     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);
                     VCVTR_S32_F64(s0, v1);
                     VMRS(x3);   // get the FPCSR reg and test FPU execption (invalid operation only)
                     VMOVfrV(ed, s0);
@@ -226,48 +243,171 @@ 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_reset(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, NULL);
                         ed = x1;
                     }
                     s0 = fpu_get_scratch_single(dyn);
                     MSR_nzcvq_0();
-                    // The FPCSR reg to clear exceptions flags
-                    ORR_IMM8(x3, x1, 0b001, 6); // enable exceptions
-                    BIC_IMM8(x3, x3, 0b10011111, 0);
-                    VMSR(x3);
                     VCVTR_S32_F64(s0, v1);
                     VMRS(x3);   // get the FPCSR reg and test FPU execption (denormal and overflow only)
                     VMOVfrV(ed, s0);
                     TSTS_IMM8_ROR(x3, 0b00000001, 0);
                     MOV_IMM_COND(cNE, ed, 0b10, 1);   // 0x80000000
                     WBACK;
-                    x87_do_pop(dyn, ninst);
+                    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, NULL);
+                    if(PK(0)==0xDB && ((PK(1)>>3)&7)==7) {
+                        // the FLD is immediatly followed by an FSTP
+                        LDR_IMM9(x2, ed, 0);
+                        LDR_IMM9(x3, ed, 4);
+                        LDRH_IMM8(x14, ed, 8);
+                        // no persistant scratch register, so unrool both instruction here...
+                        MESSAGE(LOG_DUMP, "\tHack: FSTP tbyte\n");
+                        nextop = F8;    //0xDB
+                        nextop = F8;    //modrm
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
+                        STR_IMM9(x2, ed, 0);
+                        STR_IMM9(x3, ed, 4);
+                        STRH_IMM8(x14, ed, 8);
+                    } else {
+                        if(box86_x87_no80bits) {
+                            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, NULL);
+                                VLDR_64(v1, ed, fixedaddress);
+                            } else {
+                                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
+                                LDR_IMM9(x2, ed, fixedaddress);
+                                LDR_IMM9(x3, ed, fixedaddress + 4);
+                                VMOVtoV_D(v1, x2, x3);
+                            }
+                        } else {
+                            #if 0
+                            MESSAGE(LOG_DUMP, "Need Optimization\n");
+                            if(ed!=x1) {
+                                MOV_REG(x1, ed);
+                            }
+                            x87_do_push_empty(dyn, ninst, x3);
+                            CALL(arm_fld, -1, 0);
+                            #else
+                            v1 = x87_do_push(dyn, ninst, x2, NEON_CACHE_ST_D);
+                            // copy 10bytes of *ED to STld(0)
+                            LDR_IMM9(x3, xEmu, offsetof(x86emu_t, top));
+                            int a = -dyn->n.x87stack;
+                            if(a) {
+                                if(a<0) {
+                                    SUB_IMM8(x3, x3, -a);
+                                } else {
+                                    ADD_IMM8(x3, x3, a);
+                                }
+                                AND_IMM8(x3, x3, 7);
+                            }
+                            ADD_IMM8(x2, xEmu, offsetof(x86emu_t, fpu_ld)+offsetof(fpu_ld_t, ld));
+                            MOVW(x14, sizeof(fpu_ld_t));
+                            MLA(x2, x3, x14, x2);
+                            LDR_IMM9(x14, ed, 0);
+                            STR_IMM9(x14, x2, 0);
+                            LDR_IMM9(x14, ed, 4);
+                            STR_IMM9(x14, x2, 4);
+                            LDRH_IMM8(x14, ed, 8);
+                            STRH_IMM8(x14, x2, 8);
+                            // convert 10bytes -> double
+                            CALL_1RD(FromLD, ed, x2, (1<<x3)|(1<<xFlags)|(1<<x2));
+                            // store ref to STld(0).uref, but can be unaligned...
+                            VMOVfrV_D(x3, x14, 0);
+                            // also, store result to ST0
+                            VMOVD(v1, 0);
+                            STR_IMM9(x3, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld));
+                            STR_IMM9(x14, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld)+4);
+                            #endif
+                        }
                     }
-                    CALL(arm_fld, -1, 0);
                     break;
                 case 7:
                     INST_NAME("FSTP tbyte");
-                    x87_forget(dyn, ninst, x1, x3, 0);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
-                    if(ed!=x1) {
-                        MOV_REG(x1, ed);
+                        if(box86_x87_no80bits) {
+                        v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                        parity = getedparity(dyn, ninst, addr, nextop, 3);
+                        if(parity) {
+                            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
+                            VSTR_64(v1, ed, fixedaddress);
+                        } else {
+                            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
+                            VMOVfrV_D(x2, x3, v1);
+                            STR_IMM9(x2, ed, fixedaddress);
+                            STR_IMM9(x3, ed, fixedaddress+4);
+                        }
+                    } else {
+                        #if 0
+                        MESSAGE(LOG_DUMP, "Need Optimization\n");
+                        x87_forget(dyn, ninst, x1, x3, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
+                        if(ed!=x1) {
+                            MOV_REG(x1, ed);
+                        }
+                        CALL(arm_fstp, -1, 0);
+                        #else
+                        v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                        v2 = fpu_get_scratch_double(dyn);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
+                        // get top in x3
+                        LDR_IMM9(x3, xEmu, offsetof(x86emu_t, top));
+                        int a = -dyn->n.x87stack;
+                        if(a) {
+                            if(a<0) {
+                                SUB_IMM8(x3, x3, -a);
+                            } else {
+                                ADD_IMM8(x3, x3, a);
+                            }
+                            AND_IMM8(x3, x3, 7);
+                        }
+                        // get STld(0)
+                        ADD_IMM8(x2, xEmu, offsetof(x86emu_t, fpu_ld)+offsetof(fpu_ld_t, ld));
+                        MOVW(x14, sizeof(fpu_ld_t));
+                        MLA(x2, x3, x14, x2);
+                        // compare uref with actual value, top is not used anymore, so x3 is free
+                        LDR_IMM9(x14, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld));
+                        LDR_IMM9(x3, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld)+4);
+                        VMOVtoV_D(v2, x14, x3);
+                        VCEQ_32(v2, v2, v1);    // same?
+                        VMOVfrV_D(x14, x3, v2);
+                        TSTS_REG_LSL_IMM5(x14, x3, 0);
+                        B_MARK(cEQ);    // different, do the conversion
+                        // same... copy STld(0).ld
+                        LDR_IMM9(x14, x2, 0);
+                        LDR_IMM9(x3, x2, 4);
+                        STR_IMM9(x14, ed, 0);
+                        STR_IMM9(x3, ed, 4);
+                        LDRH_IMM8(x14, x2, 8);
+                        STRH_IMM8(x14, ed, 8);
+                        B_MARK2(c__);
+                        MARK;
+                        // different, call D2LD
+                        SUB_IMM8(xSP, xSP, 8);
+                        VSTR_64(v1, xSP, 0);
+                        PUSH(xSP, (1<<xEmu)|(1<<xFlags));
+                        ADD_IMM8(0, xSP, 8);    //1st arg is &d
+                        if(ed!=x1) {
+                            MOV_REG_LSL_IMM5(x1, ed, 0);    //2nd arg is ed
+                        }
+                        CALL_S(D2LD, -2, 0);
+                        POP(xSP, (1<<xEmu)|(1<<xFlags));
+                        ADD_IMM8(xSP, xSP, 8);
+                        MARK2;
+                        #endif
                     }
-                    CALL(arm_fstp, -1, 0);
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                     break;
                 default:
                     DEFAULT;

src/dynarec/dynarec_arm_dc.c

@@ -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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         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, NULL);
                         LDR_IMM9(x2, wback, fixedaddress);
                         LDR_IMM9(x3, wback, fixedaddress+4);
                         VMOVtoV_D(d1, x2, x3);

src/dynarec/dynarec_arm_dd.c

@@ -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 +79,9 @@ 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
+            x87_swapreg(dyn, ninst, x1, x2, 0, nextop&7);
+            x87_do_pop(dyn, ninst, x3);
             break;
 
         case 0xE0:
@@ -86,9 +93,13 @@ 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 +111,15 @@ 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 +153,35 @@ 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, NULL);
                         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, NULL);
+                        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, NULL);
                     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, NULL);
                         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, NULL);
                         VMOVfrV_D(x2, x3, v1);
                         STR_IMM9(x2, ed, fixedaddress);
                         STR_IMM9(x3, ed, fixedaddress+4);
@@ -180,33 +189,50 @@ 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, NULL);
                         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, NULL);
                         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, NULL);
                     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, NULL);
                     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, NULL);
+                    LDR_IMM9(x2, xEmu, offsetof(x86emu_t, top));
+                    LDRH_IMM8(x1, xEmu, offsetof(x86emu_t, sw));
+                    if(dyn->n.x87stack) {
+                        // update top
+                        if(dyn->n.x87stack>0) {
+                            SUB_IMM8(x2, x2, dyn->n.x87stack);
+                        } else {
+                            ADD_IMM8(x2, x2, -dyn->n.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;
             }

src/dynarec/dynarec_arm_de.c

@@ -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 +164,32 @@ 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(!box86_dynarec_fastnan) {
+                VMRS(x14);               // get fpscr
+                ORR_IMM8(x3, x14, 0b010, 9); // enable exceptions
+                BIC_IMM8(x3, x3, 0b10011111, 0);
+                VMSR(x3);
+            }
+            if(ST_IS_F(0)) {
+                VDIV_F32(v2, v2, v1);
+            } else {
+                VDIV_F64(v2, v2, v1);
+            }
+            if(!box86_dynarec_fastnan) {
+                VMRS(x3);   // get the FPCSR reg and test FPU execption (invalid operation only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                if(ST_IS_F(0)) {
+                    VNEG_F32_cond(cNE, v2, v2);
+                } else {
+                    VNEG_F64_cond(cNE, v2, v2);
+                }
+                VMSR(x14);  // restore fpscr
+            }
+            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:

src/dynarec/dynarec_arm_df.c

@@ -23,29 +23,52 @@
 #include "dynarec_arm_functions.h"
 #include "dynarec_arm_helper.h"
 
-
+static const void* round_map[4] = {arm_fist64_0, arm_fist64_1, arm_fist64_2, arm_fist64_3};
 uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog)
 {
     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->n.x87stack) {
+                // update top
+                if(dyn->n.x87stack>0) {
+                    SUB_IMM8(x2, x2, dyn->n.x87stack);
+                } else {
+                    ADD_IMM8(x2, x2, -dyn->n.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 +161,286 @@ 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, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL);
                     LDRSH_IMM8(x1, wback, fixedaddress);
-                    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, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0, 0, NULL);
                     ed = x1;
                     s0 = fpu_get_scratch_single(dyn);
-                    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, 0b010, 9); // 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, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+                    u8 = x87_setround_reset(dyn, ninst, x1, x2, x14);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0, 0, NULL);
                     ed = x1;
                     s0 = fpu_get_scratch_single(dyn);
-                    VCVTR_S32_F64(s0, v1);
+                    MSR_nzcvq_0();
+                    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, box86_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+                    u8 = x87_setround_reset(dyn, ninst, x1, x2, x14);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0, 0, NULL);
                     ed = x1;
                     s0 = fpu_get_scratch_single(dyn);
-                    VCVTR_S32_F64(s0, v1);
+                    MSR_nzcvq_0();
+                    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, NULL);
                     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, NULL);
+                    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 {
+                        // 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(x1, xEmu, offsetof(x86emu_t, top));
+                        int a = 0 - dyn->n.x87stack;
+                        if(a<0) {
+                            SUB_IMM8(x1, x1, -a);
+                            AND_IMM8(x1, x1, 7);
+                        } else if(a>0) {
+                            ADD_IMM8(x1, x1, a);
+                            AND_IMM8(x1, x1, 7);
+                        }
+                        // ed (and so x1) is not use anymore.
+                        ADD_REG_LSL_IMM5(x1, xEmu, x1, 4);  // each fpu_ll is 2 int64: ref than ll
+                        ADD_IMM8_ROR(x1, x1, offsetof(x86emu_t, fpu_ll)>>2, 15);
+                        STRD_IMM8(x2, x1, 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);    // last time x3 is used
+                        VCVT_F64_U32(v1, s0);
+                        VEOR(v2, v2, v2);
+                        MOVW(x3, 0x41F0);
+                        VMOVtoDx_16(v2, 3, x3);
+                        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, x1);    // Store STll(0).ref
+                    }
                     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, NULL);
                     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, NULL);
+                    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 1
+                        v1 = x87_get_st(dyn, ninst, x2, x3, 0, NEON_CACHE_ST_D);
+                        //addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
+                        fpu_get_scratch_double(dyn); // to alocate v0
+                        v2 = fpu_get_scratch_double(dyn);
+                        #if 1
+                        //  get TOP
+                        LDR_IMM9(x14, xEmu, offsetof(x86emu_t, top));
+                        int a = 0 - dyn->n.x87stack;
+                        if(a<0) {
+                            SUB_IMM8(x14, x14, -a);
+                            AND_IMM8(x14, x14, 7);    // (emu->top + i)&7
+                        } else if(a>0) {
+                            ADD_IMM8(x14, x14, a);
+                            AND_IMM8(x14, x14, 7);    // (emu->top + i)&7
+                        }
+                        ADD_REG_LSL_IMM5(x14, xEmu, x14, 4);  // each fpu_ll is 2 int64: ref than ll
+                        MOVW(x2, offsetof(x86emu_t, fpu_ll));   //can be optimized?
+                        ADD_REG_LSL_IMM5(x14, x14, x2, offsetof(fpu_ll_t, sref));
+                        VLDR_64(v2, x14, 0);
+                        VCEQ_32(v2, v2, v1);    // compare
+                        VMOVfrV_D(x2, x3, v2);
+                        ANDS_REG_LSL_IMM5(x2, x2, x3, 0);   // if NE then values are the same!
+                        B_MARK(cEQ);    // do the i64 conversion
+                        // memcpy(ed, &STll(0).ll, sizeof(int64_t));
+                        LDRD_IMM8(x2, x14, offsetof(fpu_ll_t, sq));  // load ll
+                        B_MARK3(c__);
+                        MARK;
+                        // load round mode
+                        LDRH_IMM8(x14, xEmu, offsetof(x86emu_t, cw));    // hopefully cw is not too far for an imm8
+                        UBFX(x2, x14, 10, 2);    // extract round...
+                        MOV32(x14, round_map);
+                        LDR_REG_LSL_IMM5(x2, x14, x2, 2);
+                        VMOV_64(0, v1);    // prepare call to log2
+                        CALL_1DR_U64(x2, x2, x3, x14, (1<<x1));
+                        MARK3;
+                        STR_IMM9(x2, ed, 0);
+                        STR_IMM9(x3, ed, 4);
+                        #else
+                        v0 = fpu_get_scratch_double(dyn);
+                        s0 = fpu_get_scratch_single(dyn);
+                        // check STll(0).ref==ST(0).q so emu->fpu_ll[emu->top].ref == emu->mmx87[emu->top]
+                        //  get TOP
+                        LDR_IMM9(x14, xEmu, offsetof(x86emu_t, top));
+                        int a = 0 - dyn->n.x87stack;
+                        if(a<0) {
+                            SUB_IMM8(x14, x14, -a);
+                            AND_IMM8(x14, x14, 7);    // (emu->top + i)&7
+                        } else if(a>0) {
+                            ADD_IMM8(x14, x14, a);
+                            AND_IMM8(x14, x14, 7);    // (emu->top + i)&7
+                        }
+                        ADD_REG_LSL_IMM5(x14, xEmu, x14, 4);  // each fpu_ll is 2 int64: ref than ll
+                        MOVW(x2, offsetof(x86emu_t, fpu_ll));   //can be optimized?
+                        ADD_REG_LSL_IMM5(x14, x14, x2, 0);
+                        VLDR_64(v2, x14, 0);
+                        VCEQ_32(v2, v2, v1);    // compare
+                        VMOVfrV_D(x2, x3, v2);
+                        ANDS_REG_LSL_IMM5(x2, x2, x3, 0);   // if NE then values are the same!
+                        B_MARK(cEQ);    // do the i64 conversion
+                        // memcpy(ed, &STll(0).ll, sizeof(int64_t));
+                        LDRD_IMM8(x2, x14, 8);  // load ll
+                        B_MARK3(c__);
+                        MARK;
+                        VEOR(v0, v0, v0);
+                        MOVW(x2, 0x41F0);
+                        VMOVtoDx_16(v0, 3, x2); // V0 = (1<<32) as double
+                        VMOVfrDx_32(x14, v1, 1);    // get high part to extract sign
+                        VABS_F64(v1 ,v1);    //ST0 will be poped, so lost...
+                        VDIV_F64(v2, v1, v0);   // v2 = abs(ST0)/(1<<32) : so 32 bits high part
+                        MSR_nzcvq_0();
+                        VMRS(x2);               // get fpscr
+                        ORR_IMM8(x3, x2, 0b010, 9); // enable exceptions
+                        BIC_IMM8(x3, x3, 0b10011111, 0);
+                        VMSR(x3);
+                        VCVT_U32_F64(s0, v2);   // convert high part to U32
+                        VMRS(x3);   // get the FPCSR reg and test FPU execption (invalid operation only)
+                        VMSR(x2);               // put back fpscr
+                        TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                        B_MARK2(cEQ);   // not overflow...
+                        MARKLOCK;
+                        MOV_IMM(x3, 0b10, 1);   // 0x80000000
+                        MOVW(x2, 0);
+                        B_MARK3(c__);
+                        MARK2;  // continue conversion, it fits an int64!
+                        VCVT_F64_U32(v2, s0);   // int part now
+                        VMLS_F64(v1, v2, v0);   // compute low part
+                        VMOVfrV(x3, s0);        // transfert high path
+                        TSTS_IMM8_ROR(x14, 0b10, 1);    // test high part with 0x800000000
+                        B_MARKLOCK(cNE);        // int overflow...
+                        VCVT_U32_F64(s0, v1);   // convert low part
+                        VMOVfrV(x2, s0);        // transfert low part
+                        TSTS_IMM8_ROR(x14, 0b10, 1);    // 0x800000000
+                        B_MARK3(cEQ);
+                        RSBS_IMM8(x2, x2, 0);   // NEG(i64)
+                        RSC_IMM8(x3, x3, 0);
+                        MARK3;
+                        STR_IMM9(x2, x1, 0);
+                        STR_IMM9(x3, x1, 4);
+                        #endif
+                        #else
+                        MESSAGE(LOG_DUMP, "Need Optimization\n");
+                        x87_forget(dyn, ninst, x2, x3, 0);
+                        if(ed!=x1) {MOV_REG(x1, ed);}
+                        CALL(arm_fistp64, -1, 0);
+                        #endif
+                    }
+                    x87_do_pop(dyn, ninst, x3);
                     break;
                 default:
                     DEFAULT;

src/dynarec/dynarec_arm_emit_shift.c

@@ -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 +335,324 @@ 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;
+    MAYUSE(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 SHLD32 instruction, from s1, fill s2 , shift s3, store result in s1 using s4 and s3 as scratch
+void emit_shld32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
+{
+    int j32;
+    MAYUSE(j32);
+    IFX(X_PEND) {
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
+        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
+        // same flags calc as shl32
+        SET_DF(s4, d_shl32);
+    } else IFX(X_ALL) {
+        SET_DFNONE(s4);
+    }
+    CMPS_IMM8(s3, 0);
+        IFX(F_OF) {
+            BFC_COND(cEQ, xFlags, F_OF, 1);
+        }
+        IFX(X_PEND) {
+            STR_IMM9_COND(cEQ, s1, xEmu, offsetof(x86emu_t, res));
+        }
+        B_NEXT(cEQ);
+
+    IFX(X_CF) {
+        RSB_IMM8(s4, s3, 32);
+        MOV_REG_LSR_REG(s4, s1, s4);
+        BFI(xFlags, s4, F_CF, 1);
+    }
+    IFX(X_OF) {
+        MOVS_REG_LSL_REG(s1, s1, s3);
+    } else {
+        MOV_REG_LSL_REG(s1, s1, s3);
+    }
+    IFX(X_OF) {
+        MOVW_COND(cCS, s4, 1);
+        MOVW_COND(cCC, s4, 0);
+        CMPS_IMM8(s3, 1);
+            ORR_REG_LSR_IMM5_COND(cEQ, s4, s4, s1, 30);
+            BFI_COND(cEQ, xFlags, s4, F_OF, 1);
+        //else
+            BFC_COND(cNE, xFlags, F_OF, 1);
+    }
+    RSB_IMM8(s3, s3, 32);
+    IFX(X_ZF) {
+        ORRS_REG_LSR_REG(s1, s1, s2, s3);
+    } else {
+        ORR_REG_LSR_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);
+        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_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);
         }
     }
 }

src/dynarec/dynarec_arm_emit_tests.c

@@ -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 +238,174 @@ 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
+// emit TEST8 instruction, from test s1 , s2, using s3 and s4 as scratch. May destroy s2 also
 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));
-        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(s2, s3, 0xE0); // lsr 5 masking pre-applied
+        MOV32(s4, GetParityTab());
+        LDR_REG_LSR_IMM5(s4, s4, s2, 5-2);   // x/32 and then *4 because array is integer
         AND_IMM8(s3, s3, 31);
         MVN_REG_LSR_REG(s4, s4, s3);
-        AND_IMM8(s4, s4, 1);
-        STR_IMM9(s4, xEmu, offsetof(x86emu_t, flags[F_PF]));
+        BFI(xFlags, s4, F_PF, 1);
     }
 }

src/dynarec/dynarec_arm_f20f.c

@@ -23,23 +23,28 @@
 #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);            \
+        SMREAD();                                   \
         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, NULL); \
             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, NULL);\
             LDR_IMM9(x2, ed, fixedaddress+0);       \
             LDR_IMM9(x3, ed, fixedaddress+4);       \
             VMOVtoV_D(a, x2, x3);                   \
         }                                           \
     }
 
+#define GETGM(a)    \
+    gd = (nextop&0x38)>>3;  \
+    a = mmx_get_reg(dyn, ninst, x1, x2, x3, gd)
+
 uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog)
 {
     uint8_t nextop = F8;
@@ -58,17 +63,19 @@ 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);
+                SMREAD();
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
                 LDR_IMM9(x2, ed, fixedaddress+0);
                 LDR_IMM9(x3, ed, fixedaddress+4);
                 VMOVtoV_D(v0, x2, x3);
@@ -79,40 +86,42 @@ 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, NULL);
                 STR_IMM9(x2, ed, fixedaddress+0);
                 STR_IMM9(x3, ed, fixedaddress+4);
+                SMWRITE2();
             }
             break;
         case 0x12:
             INST_NAME("MOVDDUP Gx, Ex");
             nextop = F8;
             gd = (nextop&0x38)>>3;
-            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);
+                SMREAD();
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
                 LDR_IMM9(x2, ed, fixedaddress+0);
                 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);
@@ -125,30 +134,52 @@ 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);
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                 d0 = fpu_get_scratch_double(dyn);
                 VLDR_64(d0, ed, fixedaddress);
             }
+            if(!box86_dynarec_fastround) {
+                VMRS(x14);   // Get FPCSR reg to clear exceptions flags
+                ORR_IMM8(x3, x14, 0b010, 9); // enable exceptions
+                BIC_IMM8(x3, x3, 0b10011111, 0);
+                VMSR(x3);
+            }
             VCVT_S32_F64(s0, d0);
             VMOVfrV(gd, s0);
+            if(!box86_dynarec_fastround) {
+                VMRS(x3);   // get the FPCSR reg and test FPU exception (IO only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                MOV_IMM_COND(cNE, gd, 0b10, 1);   // 0x80000000
+                VMSR(x14);  // put back values
+            }
             break;
         case 0x2D:
             INST_NAME("CVTSD2SI Gd, Ex");
-            u8 = x87_setround(dyn, ninst, x1, x2, x12);
+            if(!box86_dynarec_fastround)
+                u8 = sse_setround_reset(dyn, ninst, x1, x2, x14);
+            else
+                u8 = sse_setround(dyn, ninst, x1, x2, x14);
             nextop = F8;
             GETGD;
             s0 = fpu_get_scratch_single(dyn);
             if((nextop&0xC0)==0xC0) {
-                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);
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                 d0 = fpu_get_scratch_double(dyn);
                 VLDR_64(d0, ed, fixedaddress);
             }
             VCVTR_S32_F64(s0, d0);
             VMOVfrV(gd, s0);
+            if(!box86_dynarec_fastround) {
+                VMRS(x3);   // get the FPCSR reg and test FPU exception (IO only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                MOV_IMM_COND(cNE, gd, 0b10, 1);   // 0x80000000
+            }
             x87_restoreround(dyn, ninst, u8);
             break;
 
@@ -156,89 +187,152 @@ 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);
+            if(!box86_dynarec_fastnan) {
+                VMRS(x14);               // get fpscr
+                ORR_IMM8(x3, x14, 0b010, 9); // enable exceptions
+                BIC_IMM8(x3, x3, 0b10011111, 0);
+                VMSR(x3);
+            }
             VSQRT_F64(v0, d0);
+            if(!box86_dynarec_fastnan) {
+                VMRS(x3);   // get the FPCSR reg and test FPU execption (invalid operation only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                VNEG_F64_cond(cNE, v0, v0);
+                VMSR(x14);  // restore FPSCR
+            }
             break;
 
         case 0x58:
             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);
+            if(!box86_dynarec_fastnan) {
+                VMRS(x14);               // get fpscr
+                ORR_IMM8(x3, x14, 0b010, 9); // enable exceptions
+                BIC_IMM8(x3, x3, 0b10011111, 0);
+                VMSR(x3);
+            }
             VADD_F64(v0, v0, d0);
+            if(!box86_dynarec_fastnan) {
+                VMRS(x3);   // get the FPCSR reg and test FPU execption (invalid operation only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                VNEG_F64_cond(cNE, v0, v0);
+                VMSR(x14);  // restore FPSCR
+            }
             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);
+            if(!box86_dynarec_fastnan) {
+                VMRS(x14);               // get fpscr
+                ORR_IMM8(x3, x14, 0b010, 9); // enable exceptions
+                BIC_IMM8(x3, x3, 0b10011111, 0);
+                VMSR(x3);
+            }
             VMUL_F64(v0, v0, d0);
+            if(!box86_dynarec_fastnan) {
+                VMRS(x3);   // get the FPCSR reg and test FPU execption (invalid operation only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                VNEG_F64_cond(cNE, v0, v0);
+                VMSR(x14);  // restore FPSCR
+            }
             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);
+            if(!box86_dynarec_fastnan) {
+                VMRS(x14);               // get fpscr
+                ORR_IMM8(x3, x14, 0b010, 9); // enable exceptions
+                BIC_IMM8(x3, x3, 0b10011111, 0);
+                VMSR(x3);
+            }
             VSUB_F64(v0, v0, d0);
+            if(!box86_dynarec_fastnan) {
+                VMRS(x3);   // get the FPCSR reg and test FPU execption (invalid operation only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                VNEG_F64_cond(cNE, v0, v0);
+                VMSR(x14);  // restore FPSCR
+            }
             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);
+            if(!box86_dynarec_fastnan) {
+                VMRS(x14);               // get fpscr
+                ORR_IMM8(x3, x14, 0b010, 9); // enable exceptions
+                BIC_IMM8(x3, x3, 0b10011111, 0);
+                VMSR(x3);
+            }
             VDIV_F64(v0, v0, d0);
+            if(!box86_dynarec_fastnan) {
+                VMRS(x3);   // get the FPCSR reg and test FPU execption (invalid operation only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                VNEG_F64_cond(cNE, v0, v0);
+                VMSR(x14);  // restore FPSCR
+            }
             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);
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 v1 = fpu_get_scratch_quad(dyn);
                 VLD1Q_64(v1, ed);
             }
@@ -269,11 +363,12 @@ 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);
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 v1 = fpu_get_scratch_quad(dyn);
                 VLD1Q_64(v1, ed);
             }
@@ -285,29 +380,103 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             }
             break;
 
+        case 0x80:
+        case 0x81:
+        case 0x82:
+        case 0x83:
+        case 0x84:
+        case 0x85:
+        case 0x86:
+        case 0x87:
+        case 0x88:
+        case 0x89:
+        case 0x8A:
+        case 0x8B:
+        case 0x8C:
+        case 0x8D:
+        case 0x8E:
+        case 0x8F:
+            return dynarec0F(dyn, addr-1, ip, ninst, ok, need_epilog);
+
         case 0xC2:
             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 0xD6:
+            INST_NAME("MOVDQ2Q Gm, Ex");
+            nextop = F8;
+            GETGM(v0);
+            GETEX(v1, 0);
+            VMOV_64(v0, v1);
+            break;
+
+        case 0xE6:
+            INST_NAME("CVTPD2DQ Gx, Ex");
+            nextop = F8;
+            gd = (nextop&0x38)>>3;
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+            if((nextop&0xC0)==0xC0) {
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
+                v1 = fpu_get_scratch_quad(dyn);
+                VLD1Q_8(v1, ed);
+            }
+            u8 = sse_setround(dyn, ninst, x1, x2, x14);
+            if(v0<16) {
+                VCVTR_S32_F64(v0*2, v1);
+                VCVTR_S32_F64(v0*2+1, v1+1);
+            } else {
+                d0 = fpu_get_scratch_double(dyn);
+                VCVTR_S32_F64(d0*2, v1);
+                VCVTR_S32_F64(d0*2+1, v1+1);
+                VMOVD(v0, d0);
+            }
+            x87_restoreround(dyn, ninst, u8);
+            VEOR(v0+1, v0+1, v0+1);
+            break;
+
+        case 0xF0:
+            INST_NAME("LDDQ Gx,Ex");
+            nextop = F8;
+            gd = (nextop&0x38)>>3;
+            if((nextop&0xC0)==0xC0) {
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                VMOVQ(v0, v1);
+            } else {
+                SMREAD();
+                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-12, 0, 0, NULL);
+                LDR_IMM9(x2, ed, fixedaddress+0);
+                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;
     }

src/dynarec/dynarec_arm_f30f.c

@@ -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 {                                      \
@@ -37,8 +37,9 @@
         b = fpu_get_scratch_double(dyn);            \
         a = b*2;                                    \
         parity = getedparity(dyn, ninst, addr, nextop, 2);  \
+        SMREAD();                                   \
         if(parity) {                                \
-            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 0); \
+            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 0, 0, NULL); \
             VLDR_32(a, ed, fixedaddress);           \
         } else {                                    \
             GETED;                                  \
@@ -46,16 +47,19 @@
         }                                           \
     }
 
-#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)
+
+#define GETGX_empty(a)  gd = ((nextop&0x38)>>3);    \
+                        a = sse_get_reg_empty(dyn, ninst, x1, gd)
 
 
 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 +67,13 @@ 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 +82,8 @@ 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 {
@@ -97,19 +102,27 @@ 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);
-                LDR_IMM9(x2, ed, fixedaddress);   // to avoid bus errors
-                VEORQ(v0, v0, v0);
-                VMOVtoDx_32(v0, 0, x2);
+                parity = getedparity(dyn, ninst, addr, nextop, 2);
+                SMREAD();
+                if(parity && (v0<16)) {
+                    VEORQ(v0, v0, v0);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 0, 0, NULL);
+                    VLDR_32(v0*2, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0, 0, NULL);
+                    LDR_IMM9(x2, ed, fixedaddress);   // to avoid bus errors
+                    VEORQ(v0, v0, v0);
+                    VMOVtoDx_32(v0, 0, x2);
+                }
             }
             break;
         case 0x11:
             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 {
@@ -127,11 +140,52 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                     VMOVD(q0, d1);
                 }
             } else {
-                VMOVfrDx_32(x2, v0, 0);
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0);
-                STR_IMM9(x2, ed, fixedaddress);
+                parity = getedparity(dyn, ninst, addr, nextop, 2);
+                if(parity && (v0<16)) {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 0, 0, NULL);
+                    VSTR_32(v0*2, ed, fixedaddress);
+                } else {
+                    VMOVfrDx_32(x2, v0, 0);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0, 0, NULL);
+                    STR_IMM9(x2, ed, fixedaddress);
+                }
+                SMWRITE2();
             }
             break;
+        case 0x12:
+            INST_NAME("MOVSLDUP Gx, Ex");
+            nextop = F8;
+            gd = (nextop&0x38)>>3;
+            if((nextop&0xC0)==0xC0) {
+                q1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
+                q1 = fpu_get_scratch_quad(dyn);
+                VLD1Q_64(q1, ed);
+            }
+            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 {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
+                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 +202,7 @@ 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 +224,7 @@ 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 +233,43 @@ 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);
+                SMREAD();
+                if(parity) {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
+                    VLDR_32(s0, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0, 0, NULL);
+                    LDR_IMM9(x2, ed, fixedaddress);
+                    VMOVtoV(s0, x2);
+                }
+            }
+            if(!box86_dynarec_fastround) {
+                VMRS(x14);   // Get FPCSR reg to clear exceptions flags
+                ORR_IMM8(x3, x14, 0b010, 9); // enable exceptions
+                BIC_IMM8(x3, x3, 0b10011111, 0);
+                VMSR(x3);
             }
             VCVT_S32_F32(s0, s0);
             VMOVfrV(gd, s0);
+            if(!box86_dynarec_fastround) {
+                VMRS(x3);   // get the FPCSR reg and test FPU exception (IO only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                MOV_IMM_COND(cNE, gd, 0b10, 1);   // 0x80000000
+                VMSR(x14);  // put back values
+            }
             break;
         case 0x2D:
             INST_NAME("CVTSS2SI Gd, Ex");
-            u8 = x87_setround(dyn, ninst, x1, x2, x12);
+            if(!box86_dynarec_fastround)
+                u8 = sse_setround_reset(dyn, ninst, x1, x2, x14);
+            else
+                u8 = sse_setround(dyn, ninst, x1, x2, x14);
             nextop = F8;
             GETGD;
             s0 = fpu_get_scratch_single(dyn);
             if((nextop&0xC0)==0xC0) {
-                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 {
@@ -201,18 +278,31 @@ 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);
+                SMREAD();
+                if(parity) {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
+                    VLDR_32(s0, ed, fixedaddress);
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095, 0, 0, NULL);
+                    LDR_IMM9(x2, ed, fixedaddress);
+                    VMOVtoV(s0, x2);
+                }
             }
             VCVTR_S32_F32(s0, s0);
             VMOVfrV(gd, s0);
+            if(!box86_dynarec_fastround) {
+                VMRS(x3);   // get the FPCSR reg and test FPU exception (IO only)
+                TSTS_IMM8_ROR(x3, 0b00000001, 0);
+                MOV_IMM_COND(cNE, gd, 0b10, 1);   // 0x80000000
+            }
             x87_restoreround(dyn, ninst, u8);
             break;
 
         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 +322,7 @@ 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 +350,7 @@ 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 +373,7 @@ 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 +393,7 @@ 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 +413,7 @@ 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 +421,10 @@ 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);
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 v1 = fpu_get_scratch_quad(dyn);
                 VLD1Q_64(v1, ed);
             }
@@ -344,7 +435,7 @@ 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 +455,7 @@ 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 +470,7 @@ 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 +478,7 @@ 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 +498,7 @@ 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 +513,7 @@ 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 +524,13 @@ 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 {
+                SMREAD();
                 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, NULL);
                 LDR_IMM9(x2, ed, fixedaddress+0);
                 LDR_IMM9(x3, ed, fixedaddress+4);
                 VMOVtoV_D(v0, x2, x3);
@@ -451,16 +543,17 @@ 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);
             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);
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 v1 = fpu_get_scratch_quad(dyn);
                 VLD1Q_64(v1, ed);
             }
+            v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
             u8 = F8;
-            // only high part need to be 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 +580,13 @@ 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 {
+                SMREAD();
                 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, NULL);
                 VLD1_64(v0, ed);
             }
             VEOR(v0+1, v0+1, v0+1);
@@ -501,44 +595,45 @@ 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, NULL);
                 VMOVfrV_D(x2, x3, v0);
                 STR_IMM9(x2, ed, fixedaddress+0);
                 STR_IMM9(x3, ed, fixedaddress+4);
                 VMOVfrV_D(x2, x3, v0+1);
                 STR_IMM9(x2, ed, fixedaddress+8);
                 STR_IMM9(x3, ed, fixedaddress+12);
+                SMWRITE2();
             }
             break;
 
         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 +645,77 @@ 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 0xD6:
+            INST_NAME("MOVQ2DQ Gx, Em");
+            nextop = F8;
+            GETGX_empty(v0);
+            if((nextop&0xC0)==0xC0) {
+                v1 = mmx_get_reg(dyn, ninst, x1, x2, x3, (nextop&7));
+                VEORQ(v0, v0, v0);  // usefull?
+                VMOV_64(v0, v1);
+            } else {
+                parity = getedparity(dyn, ninst, addr, nextop, 3);
+                SMREAD();
+                if(parity) {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
+                    VLDR_64(v0, ed, fixedaddress);    // vfpu opcode here
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-8, 0, 0, NULL);
+                    LDR_IMM9(x2, ed, fixedaddress);
+                    LDR_IMM9(x3, ed, fixedaddress+4);
+                    VMOVtoV_D(v0, x2, x3);
+                }
+            }
+            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);
+                SMREAD();
+                if(parity) {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
+                    VLDR_64(v0, ed, fixedaddress);    // vfpu opcode here
+                } else {
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-8, 0, 0, NULL);
+                    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;

src/dynarec/dynarec_arm_functions.c

@@ -8,6 +8,7 @@
 #include <signal.h>
 #include <sys/types.h>
 #include <unistd.h>
+#include <assert.h>
 
 #include "debug.h"
 #include "box86context.h"
@@ -21,20 +22,17 @@
 #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"
+#include "arm_printer.h"
+#include "custommem.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 +54,7 @@ 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)
 {
@@ -63,27 +62,19 @@ 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;
 }
 void arm_fprem(x86emu_t* emu)
 {
-    int32_t tmp32s = ST0.d / ST1.d;
-    ST0.d -= ST1.d * tmp32s;
+    int64_t tmp64s = ST0.d / ST1.d;
+    ST0.d -= ST1.d * tmp64s;
     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);
-}
-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);
+    emu->sw.f.F87_C1 = (tmp64s&1);
+    emu->sw.f.F87_C3 = ((tmp64s>>1)&1);
+    emu->sw.f.F87_C0 = ((tmp64s>>2)&1);
 }
 void arm_frndint(x86emu_t* emu)
 {
@@ -91,17 +82,9 @@ void arm_frndint(x86emu_t* emu)
 }
 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 +92,11 @@ 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 +106,86 @@ 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)(uint64_t)0x8000000000000000LL) || !isfinite(ST0.d))
+            tmp = 0x8000000000000000LL;
         else
-            *ed = (int64_t)ST0.d;
+            tmp = fpu_round(emu, ST0.d);
+        memcpy(ed, &tmp, sizeof(tmp));
     }
 }
 
+int64_t arm_fist64_0(double d)
+{
+    int64_t tmp;
+    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, -(double)(uint64_t)0x8000000000000000LL) || !isfinite(d))
+        tmp = 0x8000000000000000LL;
+    else
+        tmp = nearbyint(d);
+    return tmp;
+}
+int64_t arm_fist64_1(double d)
+{
+    int64_t tmp;
+    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, -(double)(uint64_t)0x8000000000000000LL) || !isfinite(d))
+        tmp = 0x8000000000000000LL;
+    else
+        tmp = floor(d);
+    return tmp;
+}
+int64_t arm_fist64_2(double d)
+{
+    int64_t tmp;
+    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, -(double)(uint64_t)0x8000000000000000LL) || !isfinite(d))
+        tmp = 0x8000000000000000LL;
+    else
+        tmp = ceil(d);
+    return tmp;
+}
+int64_t arm_fist64_3(double d)
+{
+    int64_t tmp;
+    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, -(double)(uint64_t)0x8000000000000000LL) || !isfinite(d))
+        tmp = 0x8000000000000000LL;
+    else
+        tmp = trunc(d);
+    return tmp;
+}
+
+void arm_fistt64(x86emu_t* emu, int64_t* ed)
+{
+    // used of memcpy to avoid aligments issues
+    int64_t tmp = ST0.d;
+    memcpy(ed, &tmp, sizeof(tmp));
+}
+
 void arm_fld(x86emu_t* emu, uint8_t* ed)
 {
     memcpy(&STld(0).ld, ed, 10);
-    LD2D(&STld(0), &ST(0).d);
-    STld(0).ref = ST0.ll;
+    LD2D(&STld(0).ld, &ST(0).d);
+    STld(0).uref = ST0.q;
 }
 
 void arm_ud(x86emu_t* emu)
 {
-    kill(getpid(), SIGILL);
+    emu->test.test = 0;
+    emit_signal(emu, SIGILL, (void*)R_EIP, 0);
 }
 
-void arm_fxsave(x86emu_t* emu, uint8_t* ed)
+void arm_priv(x86emu_t* emu)
 {
-    // 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]));
+    emu->test.test = 0;
+    emit_signal(emu, SIGSEGV, (void*)R_EIP, 0);
 }
 
-void arm_fxrstor(x86emu_t* emu, uint8_t* ed)
+void arm_singlestep(x86emu_t* emu)
 {
-    // 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]));
+    emu->test.test = 0;
+    emit_signal(emu, SIGTRAP, (void*)R_EIP, 1);
 }
 
 void arm_fsave(x86emu_t* emu, uint8_t* ed)
@@ -175,12 +195,7 @@ 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
+        D2LD(&ST(i).d, p);
         p+=10;
     }
 }
@@ -191,50 +206,75 @@ 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
+        LD2D(p, &ST(i).d);
         p+=10;
     }
 
 }
 
+void arm_fprem1(x86emu_t* emu)
+{
+    // simplified version
+    int64_t tmp64s = lrint(ST0.d / ST1.d);
+    ST0.d -= ST1.d*tmp64s;
+    emu->sw.f.F87_C2 = 0;
+    emu->sw.f.F87_C1 = (tmp64s&1);
+    emu->sw.f.F87_C3 = ((tmp64s>>1)&1);
+    emu->sw.f.F87_C0 = ((tmp64s>>2)&1);
+}
+
+void arm_clflush(x86emu_t* emu, void* p)
+{
+    cleanDBFromAddressRange((uintptr_t)p, 8, 0);
+}
 
 // Get a FPU single scratch reg
 int fpu_get_scratch_single(dynarec_arm_t* dyn)
 {
-    return dyn->fpu_scratch++;  // return an Sx
+    return dyn->n.fpu_scratch++;  // return an Sx
 }
 // Get a FPU double scratch reg
 int fpu_get_scratch_double(dynarec_arm_t* dyn)
 {
-    int i = (dyn->fpu_scratch+1)&(~1);
-    dyn->fpu_scratch = i+2;
+    int i = (dyn->n.fpu_scratch+1)&(~1);
+    dyn->n.fpu_scratch = i+2;
     return i/2; // return a Dx
 }
 // Get a FPU quad scratch reg
 int fpu_get_scratch_quad(dynarec_arm_t* dyn)
 {
-    int i = (dyn->fpu_scratch+3)&(~3);
-    dyn->fpu_scratch = i+4;
+    if(dyn->n.fpu_scratch>4) {
+        if(dyn->n.fpu_extra_qscratch) {
+            dynarec_log(LOG_NONE, "Warning, Extra QScratch slot taken and need another one!\n");
+        } else
+            dyn->n.fpu_extra_qscratch = fpu_get_reg_quad(dyn, NEON_CACHE_SCR, 0);
+        return dyn->n.fpu_extra_qscratch;
+    }
+    int i = (dyn->n.fpu_scratch+3)&(~3);
+    dyn->n.fpu_scratch = i+4;
     return i/2; // Return a Dx, not a Qx
 }
 // Reset scratch regs counter
 void fpu_reset_scratch(dynarec_arm_t* dyn)
 {
-    dyn->fpu_scratch = 0;
+    dyn->n.fpu_scratch = 0;
+    if(dyn->n.fpu_extra_qscratch) {
+        fpu_free_reg_quad(dyn, dyn->n.fpu_extra_qscratch);
+        dyn->n.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;
+    while (dyn->n.fpuused[i] && (i<24)) ++i;
+    assert(i<24);
+
+    dyn->n.fpuused[i] = 1;
+    dyn->n.neoncache[i].n = n;
+    dyn->n.neoncache[i].t = t;
+    dyn->n.news |= (1<<i);
     return i+FPUFIRST; // return a Dx
 }
 // Free a FPU double reg
@@ -242,35 +282,378 @@ void fpu_free_reg_double(dynarec_arm_t* dyn, int reg)
 {
     // TODO: check upper limit?
     int i=reg-FPUFIRST;
-    dyn->fpuused[i] = 0;
+    dyn->n.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;
+    while ((dyn->n.fpuused[i] || dyn->n.fpuused[i+1]) && (i<24)) i+=2;
+    assert(i<24);
+    dyn->n.fpuused[i] = dyn->n.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;
+    dyn->n.news |= (3<<i);
     return i+FPUFIRST; // Return a Dx, not a Qx
 }
 // Free a FPU quad reg
 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.fpuused[i] = dyn->n.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)
-        dyn->fpuused[i]=0;
+    dyn->n.fpu_reg = 0;
+    for (int i=0; i<24; ++i) {
+        dyn->n.fpuused[i]=0;
+        dyn->n.neoncache[i].v = 0;
+    }
 }
 
+int neoncache_get_st(dynarec_arm_t* dyn, int ninst, int a)
+{
+    if (dyn->insts[ninst].n.swapped) {
+        if(dyn->insts[ninst].n.combined1 == a)
+            a = dyn->insts[ninst].n.combined2;
+        else if(dyn->insts[ninst].n.combined2 == a)
+            a = dyn->insts[ninst].n.combined1;
+    }
+    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)
+{
+    /*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)
+            return i;
+    return -1;
+} 
+int neoncache_get_st_f_noback(dynarec_arm_t* dyn, int ninst, int a)
+{
+    for(int i=0; i<24; ++i)
+        if(dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_F
+         && dyn->insts[ninst].n.neoncache[i].n==a)
+            return i;
+    return -1;
+} 
+int neoncache_get_current_st_f(dynarec_arm_t* dyn, int a)
+{
+    for(int i=0; i<24; ++i)
+        if(dyn->n.neoncache[i].t==NEON_CACHE_ST_F
+         && dyn->n.neoncache[i].n==a)
+            return i;
+    return -1;
+} 
+static void neoncache_promote_double_forward(dynarec_arm_t* dyn, int ninst, int maxinst, int a);
+static void neoncache_promote_double_internal(dynarec_arm_t* dyn, int ninst, int maxinst, int a);
+static void neoncache_promote_double_combined(dynarec_arm_t* dyn, int ninst, int maxinst, int a)
+{
+    if(a == dyn->insts[ninst].n.combined1 || a == dyn->insts[ninst].n.combined2) {
+        if(a == dyn->insts[ninst].n.combined1) {
+            a = dyn->insts[ninst].n.combined2;
+        } else 
+            a = dyn->insts[ninst].n.combined1;
+        int i = neoncache_get_st_f_noback(dyn, ninst, a);
+        //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double_combined, ninst=%d combined%c %d i=%d (stack:%d/%d)\n", ninst, (a == dyn->insts[ninst].n.combined2)?'2':'1', a ,i, dyn->insts[ninst].n.stack_push, -dyn->insts[ninst].n.stack_pop);
+        if(i>=0) {
+            dyn->insts[ninst].n.neoncache[i].t = NEON_CACHE_ST_D;
+            if(!dyn->insts[ninst].n.barrier)
+                neoncache_promote_double_internal(dyn, ninst-1, maxinst, a-dyn->insts[ninst].n.stack_push);
+            // go forward is combined is not pop'd
+            if(a-dyn->insts[ninst].n.stack_pop>=0)
+                if(!dyn->insts[ninst+1].n.barrier)
+                    neoncache_promote_double_forward(dyn, ninst+1, maxinst, a-dyn->insts[ninst].n.stack_pop);
+        }
+    }
+}
+static void neoncache_promote_double_internal(dynarec_arm_t* dyn, int ninst, int maxinst, int a)
+{
+    if(dyn->insts[ninst+1].n.barrier)
+        return;
+    while(ninst>=0) {
+        a+=dyn->insts[ninst].n.stack_pop;    // adjust Stack depth: add pop'd ST (going backward)
+        int i = neoncache_get_st_f(dyn, ninst, a);
+        //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double_internal, ninst=%d, a=%d st=%d:%d, i=%d\n", ninst, a, 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;
+        // check combined propagation too
+        if(dyn->insts[ninst].n.combined1 || dyn->insts[ninst].n.combined2) {
+            if(dyn->insts[ninst].n.swapped) {
+                //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double_internal, ninst=%d swapped %d/%d vs %d with st %d\n", ninst, dyn->insts[ninst].n.combined1 ,dyn->insts[ninst].n.combined2, a, dyn->insts[ninst].n.stack);
+                if (a==dyn->insts[ninst].n.combined1)
+                    a = dyn->insts[ninst].n.combined2;
+                else if (a==dyn->insts[ninst].n.combined2)
+                    a = dyn->insts[ninst].n.combined1;
+            } else {
+                //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double_internal, ninst=%d combined %d/%d vs %d with st %d\n", ninst, dyn->insts[ninst].n.combined1 ,dyn->insts[ninst].n.combined2, a, dyn->insts[ninst].n.stack);
+                neoncache_promote_double_combined(dyn, ninst, maxinst, a);
+            }
+        }
+        a-=dyn->insts[ninst].n.stack_push;  // // adjust Stack depth: remove push'd ST (going backward)
+        --ninst;
+        if(ninst<0 || a<0 || dyn->insts[ninst].n.barrier)
+            return;
+    }
+}
+
+static void neoncache_promote_double_forward(dynarec_arm_t* dyn, int ninst, int maxinst, int a)
+{
+    while((ninst!=-1) && (ninst<maxinst) && (a>=0)) {
+        a+=dyn->insts[ninst].n.stack_push;  // // adjust Stack depth: add push'd ST (going forward)
+        if((dyn->insts[ninst].n.combined1 || dyn->insts[ninst].n.combined2) && dyn->insts[ninst].n.swapped) {
+            //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double_forward, ninst=%d swapped %d/%d vs %d with st %d\n", ninst, dyn->insts[ninst].n.combined1 ,dyn->insts[ninst].n.combined2, a, dyn->insts[ninst].n.stack);
+            if (a==dyn->insts[ninst].n.combined1)
+                a = dyn->insts[ninst].n.combined2;
+            else if (a==dyn->insts[ninst].n.combined2)
+                a = dyn->insts[ninst].n.combined1;
+        }
+        int i = neoncache_get_st_f_noback(dyn, ninst, a);
+        //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double_forward, ninst=%d, a=%d st=%d:%d(%d/%d), i=%d\n", ninst, a, dyn->insts[ninst].n.stack, dyn->insts[ninst].n.stack_next, dyn->insts[ninst].n.stack_push, -dyn->insts[ninst].n.stack_pop, i);
+        if(i<0) return;
+        dyn->insts[ninst].n.neoncache[i].t = NEON_CACHE_ST_D;
+        // check combined propagation too
+        if((dyn->insts[ninst].n.combined1 || dyn->insts[ninst].n.combined2) && !dyn->insts[ninst].n.swapped) {
+            //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double_forward, ninst=%d combined %d/%d vs %d with st %d\n", ninst, dyn->insts[ninst].n.combined1 ,dyn->insts[ninst].n.combined2, a, dyn->insts[ninst].n.stack);
+            neoncache_promote_double_combined(dyn, ninst, maxinst, a);
+        }
+        a-=dyn->insts[ninst].n.stack_pop;    // adjust Stack depth: remove pop'd ST (going forward)
+        if(dyn->insts[ninst].x86.has_next && !dyn->insts[ninst].n.barrier)
+            ++ninst;
+        else
+            ninst=-1;
+    }
+    if(ninst==maxinst)
+        neoncache_promote_double(dyn, ninst, a);
+}
+
+void neoncache_promote_double(dynarec_arm_t* dyn, int ninst, int a)
+{
+    int i = neoncache_get_current_st_f(dyn, a);
+    //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double, ninst=%d a=%d st=%d i=%d\n", ninst, a, dyn->n.stack, i);
+    if(i<0) return;
+    dyn->n.neoncache[i].t = NEON_CACHE_ST_D;
+    dyn->insts[ninst].n.neoncache[i].t = NEON_CACHE_ST_D;
+    // check combined propagation too
+    if(dyn->n.combined1 || dyn->n.combined2) {
+        if(dyn->n.swapped) {
+            //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double, ninst=%d swapped! %d/%d vs %d\n", ninst, dyn->n.combined1 ,dyn->n.combined2, a);
+            if(dyn->n.combined1 == a)
+                a = dyn->n.combined2;
+            else if(dyn->n.combined2 == a)
+                a = dyn->n.combined1;
+        } else {
+            //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);
+            else if(dyn->n.combined2 == a)
+                neoncache_promote_double(dyn, ninst, dyn->n.combined1);
+        }
+    }
+    a-=dyn->insts[ninst].n.stack_push;  // // adjust Stack depth: remove push'd ST (going backward)
+    if(!ninst || a<0) return;
+    neoncache_promote_double_internal(dyn, ninst-1, ninst, a);
+}
+
+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;
+}
+
+int isPred(dynarec_arm_t* dyn, int ninst, int pred) {
+    for(int i=0; i<dyn->insts[ninst].pred_sz; ++i)
+        if(dyn->insts[ninst].pred[i]==pred)
+            return pred;
+    return -1;
+}
+int getNominalPred(dynarec_arm_t* dyn, int ninst) {
+    if((ninst<=0) || !dyn->insts[ninst].pred_sz)
+        return -1;
+    if(isPred(dyn, ninst, ninst-1)!=-1)
+        return ninst-1;
+    return dyn->insts[ninst].pred[0];
+}
+
+static int fpuCacheNeedsTransform(dynarec_arm_t* dyn, int ninst) {
+    int i2 = dyn->insts[ninst].x86.jmp_insts;
+    if(i2<0)
+        return 1;
+    if((dyn->insts[i2].x86.barrier&BARRIER_FLOAT))
+        return ((dyn->insts[ninst].x86.barrier&BARRIER_FLOAT))?0:1; // if the barrier as already been apply, no transform needed
+    int ret = 0;
+    if(!i2) { // just purge
+        if(dyn->insts[ninst].n.stack_next)  {
+            return 1;
+        }
+        for(int i=0; i<24 && !ret; ++i)
+            if(dyn->insts[ninst].n.neoncache[i].v) {       // there is something at ninst for i
+                if(!(
+                (dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_F || dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_D)
+                && dyn->insts[ninst].n.neoncache[i].n<dyn->insts[ninst].n.stack_pop))
+                    ret = 1;
+            }
+        return ret;
+    }
+    // Check if ninst can be compatible to i2
+    if(dyn->insts[ninst].n.stack_next != dyn->insts[i2].n.stack-dyn->insts[i2].n.stack_push) {
+        return 1;
+    }
+    neoncache_t cache_i2 = dyn->insts[i2].n;
+    neoncacheUnwind(&cache_i2);
+
+    for(int i=0; i<24; ++i) {
+        if(dyn->insts[ninst].n.neoncache[i].v) {       // there is something at ninst for i
+            if(!cache_i2.neoncache[i].v) {    // but there is nothing at i2 for i
+                ret = 1;
+            } else if(dyn->insts[ninst].n.neoncache[i].v!=cache_i2.neoncache[i].v) {  // there is something different
+                if(dyn->insts[ninst].n.neoncache[i].n!=cache_i2.neoncache[i].n) {   // not the same x86 reg
+                    ret = 1;
+                }
+                else if(dyn->insts[ninst].n.neoncache[i].t == NEON_CACHE_XMMR && cache_i2.neoncache[i].t == NEON_CACHE_XMMW)
+                    {/* nothing */ }
+                else
+                    ret = 1;
+            }
+        } else if(cache_i2.neoncache[i].v)
+            ret = 1;
+    }
+    return ret;
+}
+
+void neoncacheUnwind(neoncache_t* cache)
+{
+    if(cache->swapped) {
+        // unswap
+        int a = -1; 
+        int b = -1;
+        for(int j=0; j<24 && ((a==-1) || (b==-1)); ++j)
+            if((cache->neoncache[j].t == NEON_CACHE_ST_D || cache->neoncache[j].t == NEON_CACHE_ST_F)) {
+                if(cache->neoncache[j].n == cache->combined1)
+                    a = j;
+                else if(cache->neoncache[j].n == cache->combined2)
+                    b = j;
+            }
+        if(a!=-1 && b!=-1) {
+            int tmp = cache->neoncache[a].n;
+            cache->neoncache[a].n = cache->neoncache[b].n;
+            cache->neoncache[b].n = tmp;
+        }
+        cache->swapped = 0;
+        cache->combined1 = cache->combined2 = 0;
+    }
+    if(cache->news) {
+        // reove the newly created neoncache
+        for(int i=0; i<24; ++i)
+            if(cache->news&(1<<i))
+                cache->neoncache[i].v = 0;
+        cache->news = 0;
+    }
+    if(cache->stack_push) {
+        // unpush
+        for(int j=0; j<24; ++j) {
+            if((cache->neoncache[j].t == NEON_CACHE_ST_D || cache->neoncache[j].t == NEON_CACHE_ST_F)) {
+                if(cache->neoncache[j].n<cache->stack_push)
+                    cache->neoncache[j].v = 0;
+                else
+                    cache->neoncache[j].n-=cache->stack_push;
+            }
+        }
+        cache->x87stack-=cache->stack_push;
+        cache->stack-=cache->stack_push;
+        cache->stack_push = 0;
+    }
+    cache->x87stack+=cache->stack_pop;
+    cache->stack_next = cache->stack;
+    cache->stack_pop = 0;
+    cache->barrier = 0;
+    // And now, rebuild the x87cache info with neoncache
+    cache->mmxcount = 0;
+    cache->fpu_scratch = 0;
+    cache->fpu_extra_qscratch = 0;
+    cache->fpu_reg = 0;
+    for(int i=0; i<8; ++i) {
+        cache->x87cache[i] = -1;
+        cache->mmxcache[i] = -1;
+        cache->ssecache[i].v = -1;
+        cache->x87reg[i] = 0;
+    }
+    int x87reg = 0;
+    for(int i=0; i<24; ++i) {
+        if(cache->neoncache[i].v) {
+            cache->fpuused[i] = 1;
+            switch (cache->neoncache[i].t) {
+                case NEON_CACHE_MM:
+                    cache->mmxcache[cache->neoncache[i].n] = i+FPUFIRST;
+                    ++cache->mmxcount;
+                    ++cache->fpu_reg;
+                    break;
+                case NEON_CACHE_XMMR:
+                case NEON_CACHE_XMMW:
+                    cache->ssecache[cache->neoncache[i].n].reg = i+FPUFIRST;
+                    cache->ssecache[cache->neoncache[i].n].write = (cache->neoncache[i].t==NEON_CACHE_XMMW)?1:0;
+                    cache->neoncache[i] = cache->neoncache[i+1];
+                    ++cache->fpu_reg;
+                    ++i;    // next next
+                    break;
+                case NEON_CACHE_ST_F:
+                case NEON_CACHE_ST_D:
+                    cache->x87cache[x87reg] = cache->neoncache[i].n;
+                    cache->x87reg[x87reg] = i+FPUFIRST;
+                    ++x87reg;
+                    ++cache->fpu_reg;
+                    break;
+                case NEON_CACHE_SCR:
+                    cache->fpuused[i] = 0;
+                    cache->neoncache[i].v = 0;
+                    break;
+            }
+        } else {
+            cache->fpuused[i] = 0;
+        }
+    }
+}
+
+#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 +672,7 @@ 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 +686,169 @@ 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 || !getProtection(addr))
+        return 0;
+    if(PK(0)==0xff && PK(1)==0x25) {  // absolute jump, maybe the GOT
+        uintptr_t a1 = (PK32(2));   // need to add a check to see if the address is from the GOT !
+        addr = (uintptr_t)getAlternate(*(void**)a1); 
+    }
+    if(addr<0x10000 || !getProtection(addr))    // too low, that is suspicious
+        return 0;
+    onebridge_t *b = (onebridge_t*)(addr);
+    if(b->CC==0xCC && b->S=='S' && b->C=='C' && b->w!=(wrapper_t)0 && b->f!=(uintptr_t)PltResolver) {
+        // found !
+        if(retn) *retn = (b->C3==0xC2)?b->N:0;
+        if(calladdress) *calladdress = addr+1;
+        return 1;
+    }
+    return 0;
+#undef PK32
+#undef PK
+}
+
+const char* getCacheName(int t, int n)
+{
+    static char buff[20];
+    switch(t) {
+        case NEON_CACHE_ST_D: sprintf(buff, "ST%d", n); break;
+        case NEON_CACHE_ST_F: sprintf(buff, "st%d", n); break;
+        case NEON_CACHE_MM: sprintf(buff, "MM%d", n); break;
+        case NEON_CACHE_XMMW: sprintf(buff, "XMM%d", n); break;
+        case NEON_CACHE_XMMR: sprintf(buff, "xmm%d", n); break;
+        case NEON_CACHE_SCR: sprintf(buff, "Scratch"); break;
+        case NEON_CACHE_NONE: buff[0]='\0'; break;
+    }
+    return buff;
+}
+
+static int flagsCacheNeedsTransform(dynarec_arm_t* dyn, int ninst) {
+    int jmp = dyn->insts[ninst].x86.jmp_insts;
+    if(jmp<0)
+        return 0;
+    if(dyn->insts[ninst].f_exit.dfnone)  // flags are fully known, nothing we can do more
+        return 0;
+/*    if((dyn->f.pending!=SF_SET)
+    && (dyn->f.pending!=SF_SET_PENDING)) {
+        if(dyn->f.pending!=SF_PENDING) {*/
+    switch (dyn->insts[jmp].f_entry.pending) {
+        case SF_UNKNOWN: return 0;
+        case SF_SET: 
+            if(dyn->insts[ninst].f_exit.pending!=SF_SET && dyn->insts[ninst].f_exit.pending!=SF_SET_PENDING) 
+                return 1; 
+            else 
+                return 0;
+        case SF_SET_PENDING:
+            if(dyn->insts[ninst].f_exit.pending!=SF_SET 
+            && dyn->insts[ninst].f_exit.pending!=SF_SET_PENDING
+            && dyn->insts[ninst].f_exit.pending!=SF_PENDING) 
+                return 1; 
+            else 
+                return 0;
+        case SF_PENDING:
+            if(dyn->insts[ninst].f_exit.pending!=SF_SET 
+            && dyn->insts[ninst].f_exit.pending!=SF_SET_PENDING
+            && dyn->insts[ninst].f_exit.pending!=SF_PENDING)
+                return 1;
+            else
+                return (dyn->insts[jmp].f_entry.dfnone  == dyn->insts[ninst].f_exit.dfnone)?0:1;
+    }
+    if(dyn->insts[jmp].f_entry.dfnone && !dyn->insts[ninst].f_exit.dfnone)
+        return 1;
+    return 0;
+}
+int CacheNeedsTransform(dynarec_arm_t* dyn, int ninst) {
+    int ret = 0;
+    if (fpuCacheNeedsTransform(dyn, ninst)) ret|=1;
+    if (flagsCacheNeedsTransform(dyn, ninst)) ret|=2;
+    return ret;
+}
+
+void inst_name_pass3(dynarec_arm_t* dyn, int ninst, const char* name)
+{
+    if(box86_dynarec_dump) {
+        printf_x86_instruction(my_context->dec, &dyn->insts[ninst].x86, name);
+        dynarec_log(LOG_NONE, "%s%p: %d emited opcodes, inst=%d, barrier=%d state=%d/%d(%d), %s=%X/%X, use=%X, need=%X/%X sm=%d/%d",
+            (box86_dynarec_dump>1)?"\e[32m":"",
+            (void*)(dyn->arm_start+dyn->insts[ninst].address),
+            dyn->insts[ninst].size/4,
+            ninst,
+            dyn->insts[ninst].x86.barrier,
+            dyn->insts[ninst].x86.state_flags,
+            dyn->f.pending,
+            dyn->f.dfnone,
+            dyn->insts[ninst].x86.may_set?"may":"set",
+            dyn->insts[ninst].x86.set_flags,
+            dyn->insts[ninst].x86.gen_flags,
+            dyn->insts[ninst].x86.use_flags,
+            dyn->insts[ninst].x86.need_before,
+            dyn->insts[ninst].x86.need_after,
+            dyn->smread, dyn->smwrite);
+        if(dyn->insts[ninst].pred_sz) {
+            dynarec_log(LOG_NONE, ", pred=");
+            for(int ii=0; ii<dyn->insts[ninst].pred_sz; ++ii)
+                dynarec_log(LOG_NONE, "%s%d", ii?"/":"", dyn->insts[ninst].pred[ii]);
+        }
+        if(dyn->insts[ninst].x86.jmp && dyn->insts[ninst].x86.jmp_insts>=0)
+            dynarec_log(LOG_NONE, ", jmp=%d", dyn->insts[ninst].x86.jmp_insts);
+        if(dyn->insts[ninst].x86.jmp && dyn->insts[ninst].x86.jmp_insts==-1)
+            dynarec_log(LOG_NONE, ", jmp=out");
+        for(int ii=0; ii<24; ++ii) {
+            switch(dyn->insts[ninst].n.neoncache[ii].t) {
+                case NEON_CACHE_ST_D: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;
+                case NEON_CACHE_ST_F: dynarec_log(LOG_NONE, " S%d:%s", (ii+8)*2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;
+                case NEON_CACHE_MM: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;
+                case NEON_CACHE_XMMW: dynarec_log(LOG_NONE, " Q%d:%s", (ii+8)/2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); ++ii; break;
+                case NEON_CACHE_XMMR: dynarec_log(LOG_NONE, " Q%d:%s", (ii+8)/2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); ++ii; break;
+                case NEON_CACHE_SCR: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;
+                case NEON_CACHE_NONE:
+                default:    break;
+            }
+        }
+        if(dyn->n.stack || dyn->insts[ninst].n.stack_next || dyn->insts[ninst].n.x87stack)
+            dynarec_log(LOG_NONE, " X87:%d/%d(+%d/-%d)%d", dyn->n.stack, dyn->insts[ninst].n.stack_next, dyn->insts[ninst].n.stack_push, dyn->insts[ninst].n.stack_pop, dyn->insts[ninst].n.x87stack);
+        if(dyn->insts[ninst].n.combined1 || dyn->insts[ninst].n.combined2)
+            dynarec_log(LOG_NONE, " %s:%d/%d", dyn->insts[ninst].n.swapped?"SWP":"CMB", dyn->insts[ninst].n.combined1, dyn->insts[ninst].n.combined2);
+        dynarec_log(LOG_NONE, "%s\n", (box86_dynarec_dump>1)?"\e[m":"");
+    }
+}
+
+void print_opcode(dynarec_arm_t* dyn, int ninst, uint32_t opcode)
+{
+    dynarec_log(LOG_NONE, "\t%08x\t%s\n", opcode, arm_print(opcode));
 }

src/dynarec/dynarec_arm_functions.h

@@ -3,34 +3,38 @@
 
 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_clflush(x86emu_t* emu, void* p);
+
+int64_t arm_fist64_0(double d);
+int64_t arm_fist64_1(double d);
+int64_t arm_fist64_2(double d);
+int64_t arm_fist64_3(double d);
 
 void arm_ud(x86emu_t* emu);
+void arm_priv(x86emu_t* emu);
+void arm_singlestep(x86emu_t* emu);
+
+#define FPUFIRST    8
 
 // Get an FPU single scratch reg
 int fpu_get_scratch_single(dynarec_arm_t* dyn);
@@ -40,18 +44,52 @@ 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);
+// Get actual type for STx
+int neoncache_get_current_st(dynarec_arm_t* dyn, int ninst, int a);
+// Get actual STx is FLOAT or DOUBLE
+int neoncache_get_current_st_f(dynarec_arm_t* dyn, int a);
+// Back-propagate a change float->double
+void neoncache_promote_double(dynarec_arm_t* dyn, int ninst, int a);
+// Combine and propagate if needed (pass 1 only)
+int neoncache_combine_st(dynarec_arm_t* dyn, int ninst, int a, int b);  // with stack current dyn->n_stack*
+
+// FPU Cache transformation (for loops)
+int CacheNeedsTransform(dynarec_arm_t* dyn, int i1);
+
+// Undo the changes of a neoncache to get the status before the instruction
+void neoncacheUnwind(neoncache_t* cache);
+
+// predecessor access
+int isPred(dynarec_arm_t* dyn, int ninst, int pred);
+int getNominalPred(dynarec_arm_t* dyn, int ninst);
+
 // 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);
+
+const char* getCacheName(int t, int n);
+
+void inst_name_pass3(dynarec_arm_t* dyn, int ninst, const char* name);
+void print_opcode(dynarec_arm_t* dyn, int ninst, uint32_t opcode);
 
 #endif //__DYNAREC_ARM_FUNCTIONS_H__

src/dynarec/dynarec_arm_helper.h

@@ -22,9 +22,37 @@
 #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)
+
+// Strong mem emulation helpers
+// Sequence of Read will trigger a DMB on "first" read if strongmem is 2
+// Squence of Write will trigger a DMB on "last" write if strongmem is 1
+// Opcode will read
+#define SMREAD()    if(!dyn->smread && box86_dynarec_strongmem>1) {SMDMB();}
+// Opcode will read with option forced lock
+#define SMREADLOCK(lock)    if(lock || (!dyn->smread && box86_dynarec_strongmem>1)) {SMDMB();}
+// Opcode migh read (depend on nextop)
+#define SMMIGHTREAD()   if(!MODREG) {SMREAD();}
+// Opcode has wrote
+#define SMWRITE()   dyn->smwrite=1
+// Opcode has wrote (strongmem>1 only)
+#define SMWRITE2()   if(box86_dynarec_strongmem>1) dyn->smwrite=1
+// Opcode has wrote with option forced lock
+#define SMWRITELOCK(lock)   if(lock) {SMDMB();} else dyn->smwrite=1
+// Opcode migh have wrote (depend on nextop)
+#define SMMIGHTWRITE()   if(!MODREG) {SMWRITE();}
+// Start of sequence
+#define SMSTART()   SMEND()
+// End of sequence
+#define SMEND()     if(dyn->smwrite && box86_dynarec_strongmem) {DMB_ISH();} dyn->smwrite=0; dyn->smread=0;
+// Force a Data memory barrier (for LOCK: prefix)
+#define SMDMB()     DMB_ISH(); dyn->smwrite=0; dyn->smread=1
+
+//LOCK_* define
+#define LOCK_LOCK   (int*)1
 
 // GETGD    get x86 register in gd
 #define GETGD   gd = xEAX+((nextop&0x38)>>3)
@@ -33,7 +61,8 @@
                     ed = xEAX+(nextop&7);   \
                     wback = 0;              \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0); \
+                    SMREAD();               \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0, 0, NULL); \
                     LDR_IMM9(x1, wback, fixedaddress); \
                     ed = x1;                \
                 }
@@ -42,7 +71,8 @@
                     ed = xEAX+(nextop&7);   \
                     wback = 0;              \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, (hint==x2)?x1:x2, &fixedaddress, 4095, 0); \
+                    SMREAD();               \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, (hint==x2)?x1:x2, &fixedaddress, 4095, 0, 0, NULL); \
                     LDR_IMM9(hint, wback, fixedaddress); \
                     ed = hint;              \
                 }
@@ -52,24 +82,38 @@
                     MOV_REG(ret, ed);       \
                     wback = 0;              \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, hint, &fixedaddress, 4095, 0); \
+                    SMREAD();               \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, hint, &fixedaddress, 4095, 0, 0, NULL); \
                     ed = ret;               \
                     LDR_IMM9(ed, wback, fixedaddress); \
                 }
 // Write back ed in wback (if wback not 0)
-#define WBACK       if(wback) {STR_IMM9(ed, wback, fixedaddress);}
-// Send back wb to either ed or wback
-#define SBACK(wb)   if(wback) {STR_IMM9(wb, wback, fixedaddress);} else {MOV_REG(ed, wb);}
+#define WBACK       if(wback) {STR_IMM9(ed, wback, fixedaddress); SMWRITE();}
+// Write back ed in wback (if wback not 0) (SMWRITE2 version)
+#define WBACK2      if(wback) {STR_IMM9(ed, wback, fixedaddress); SMWRITE2();}
 //GETEDO can use r1 for ed, and r2 for wback. wback is 0 if ed is xEAX..xEDI
 #define GETEDO(O)   if((nextop&0xC0)==0xC0) {   \
                     ed = xEAX+(nextop&7);   \
                     wback = 0;              \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0); \
+                    SMREAD();               \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 0, NULL); \
                     LDR_REG_LSL_IMM5(x1, wback, O, 0);  \
                     ed = x1;                 \
                 }
-#define WBACKO(O)   if(wback) {STR_REG_LSL_IMM5(ed, wback, O, 0);}
+//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 {                                \
+                    SMREAD();                           \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 0, NULL); \
+                    ADD_REG_LSL_IMM5(x2, wback, O, 0);  \
+                    if(wback != x2) wback = x2;         \
+                    LDR_IMM9(x1, wback, 0);             \
+                    ed = x1;                            \
+                }
+#define WBACKO(O)   if(wback) {STR_REG_LSL_IMM5(ed, wback, O, 0); SMWRITE();}
 //FAKEELike GETED, but doesn't get anything
 #define FAKEED  if((nextop&0xC0)!=0xC0) {   \
                     addr = fakeed(dyn, addr, ninst, nextop); \
@@ -83,7 +127,8 @@
                     ed = i;                 \
                     wb1 = 0;                \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, w, &fixedaddress, 255, 0); \
+                    SMREAD();               \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, w, &fixedaddress, 255, 0, 0, NULL); \
                     LDRH_IMM8(i, wback, fixedaddress); \
                     ed = i;                 \
                     wb1 = 1;                \
@@ -95,7 +140,8 @@
                     ed = i;                 \
                     wb1 = 0;                \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0); \
+                    SMREAD();               \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL); \
                     LDRH_IMM8(i, wback, fixedaddress); \
                     ed = i;                 \
                     wb1 = 1;                \
@@ -107,15 +153,16 @@
                     ed = i;                 \
                     wb1 = 0;                \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0); \
+                    SMREAD();               \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL); \
                     LDRSH_IMM8(i, wback, fixedaddress);\
                     ed = i;                 \
                     wb1 = 1;                \
                 }
 // Write ed back to original register / memory
-#define EWBACK   if(wb1) {STRH_IMM8(ed, wback, fixedaddress);} else {BFI(wback, ed, 0, 16);}
+#define EWBACK   if(wb1) {STRH_IMM8(ed, wback, fixedaddress); SMWRITE();} else {BFI(wback, ed, 0, 16);}
 // Write w back to original register / memory
-#define EWBACKW(w)   if(wb1) {STRH_IMM8(w, wback, fixedaddress);} else {BFI(wback, w, 0, 16);}
+#define EWBACKW(w)   if(wb1) {STRH_IMM8(w, wback, fixedaddress); SMWRITE();} else {BFI(wback, w, 0, 16);}
 // Write back gd in correct register
 #define GWBACK       BFI((xEAX+((nextop&0x38)>>3)), gd, 0, 16);
 //GETEB will use i for ed, and can use r3 for wback.
@@ -127,11 +174,29 @@
                     wb1 = 0;                \
                     ed = i;                 \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095, 0); \
+                    SMREAD();               \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095, 0, 0, NULL); \
                     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 {                    \
+                    SMREAD();               \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0, 0, 0, NULL); \
+                    ADD_REG_LSL_IMM5(x3, wback, i, 0);  \
+                    if(wback!=x3) wback = x3;           \
+                    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);     \
@@ -141,13 +206,16 @@
                     wb1 = 0;                \
                     ed = i;                 \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0); \
+                    SMREAD();               \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL); \
                     LDRSB_IMM8(i, wback, fixedaddress);\
                     wb1 = 1;                \
                     ed = i;                 \
                 }
 // Write eb (ed) back to original register / memory
-#define EBBACK   if(wb1) {STRB_IMM9(ed, wback, fixedaddress);} else {BFI(wback, ed, wb2*8, 8);}
+#define EBBACK   if(wb1) {STRB_IMM9(ed, wback, fixedaddress); SMWRITE();} else {BFI(wback, ed, wb2*8, 8);}
+// Write eb (ed) back to original register / memory (SMWRITE2 version)
+#define EBBACK2   if(wb1) {STRB_IMM9(ed, wback, fixedaddress); SMWRITE2();} else {BFI(wback, ed, wb2*8, 8);}
 //GETGB will use i for gd
 #define GETGB(i)    gd = (nextop&0x38)>>3;  \
                     gb2 = ((gd&4)>>2);      \
@@ -164,123 +232,152 @@
 #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 MARK    if(dyn->insts) {dyn->insts[ninst].mark = (uintptr_t)dyn->arm_size;}
-#define GETMARK ((dyn->insts)?dyn->insts[ninst].mark:(dyn->arm_size+4))
-#define MARK2   if(dyn->insts) {dyn->insts[ninst].mark2 = (uintptr_t)dyn->arm_size;}
-#define GETMARK2 ((dyn->insts)?dyn->insts[ninst].mark2:(dyn->arm_size+4))
-#define MARK3   if(dyn->insts) {dyn->insts[ninst].mark3 = (uintptr_t)dyn->arm_size;}
-#define GETMARK3 ((dyn->insts)?dyn->insts[ninst].mark3:(dyn->arm_size+4))
-#define MARKF   if(dyn->insts) {dyn->insts[ninst].markf = (uintptr_t)dyn->arm_size;}
-#define GETMARKF ((dyn->insts)?dyn->insts[ninst].markf:(dyn->arm_size+4))
-#define MARKSEG if(dyn->insts) {dyn->insts[ninst].markseg = (uintptr_t)dyn->arm_size;}
-#define GETMARKSEG ((dyn->insts)?dyn->insts[ninst].markseg:(dyn->arm_size+4))
+#define 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)
+// CALL_1D_U64 will use S as scratch. Return value in ret/ret2, 1 ARG in D0
+#define CALL_1DR_U64(R, ret, ret2, S, M) call_dr(dyn, ninst, R, 1, S, ret, ret2, M, 1)
+// CALL_1D will use S as scratch. Return value in D0, 1 ARG in D0
+#define CALL_1DR(R, S, M) call_dr(dyn, ninst, R, 1, S, -1, -1, M, 0)
+// CALL_1RD will use S as scratch. Return value in D0, 1 ARG in R
+#define CALL_1RD(F, R, S, M) call_rd(dyn, ninst, F, R, S, M, 0);
 
-// Branch to MARK if cond (use i32)
+#define MARK        if(dyn->insts) {dyn->insts[ninst].mark = (uintptr_t)dyn->arm_size;}
+#define GETMARK     ((dyn->insts)?dyn->insts[ninst].mark:(dyn->arm_size+4))
+#define MARK2       if(dyn->insts) {dyn->insts[ninst].mark2 = (uintptr_t)dyn->arm_size;}
+#define GETMARK2    ((dyn->insts)?dyn->insts[ninst].mark2:(dyn->arm_size+4))
+#define MARK3       if(dyn->insts) {dyn->insts[ninst].mark3 = (uintptr_t)dyn->arm_size;}
+#define GETMARK3    ((dyn->insts)?dyn->insts[ninst].mark3:(dyn->arm_size+4))
+#define MARKF       if(dyn->insts) {dyn->insts[ninst].markf = (uintptr_t)dyn->arm_size;}
+#define GETMARKF    ((dyn->insts)?dyn->insts[ninst].markf:(dyn->arm_size+4))
+#define MARKF2      if(dyn->insts) {dyn->insts[ninst].markf2 = (uintptr_t)dyn->arm_size;}
+#define GETMARKF2   ((dyn->insts)?dyn->insts[ninst].markf2:(dyn->arm_size+4))
+#define MARKSEG     if(dyn->insts) {dyn->insts[ninst].markseg = (uintptr_t)dyn->arm_size;}
+#define GETMARKSEG  ((dyn->insts)?dyn->insts[ninst].markseg:(dyn->arm_size+4))
+#define MARKLOCK    if(dyn->insts) {dyn->insts[ninst].marklock = (uintptr_t)dyn->arm_size;}
+#define GETMARKLOCK ((dyn->insts)?dyn->insts[ninst].marklock:(dyn->arm_size+4))
+
+// Branch to MARK if cond (use j32)
 #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[ninst].epilog-(dyn->arm_size+8); \
+    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.gen_flags&(A)))
+#define IFX_PENDOR0  if((dyn->insts[ninst].x86.gen_flags&(X_PEND) || !dyn->insts[ninst].x86.gen_flags))
+#define IFXX(A) if((dyn->insts[ninst].x86.gen_flags==(A)))
+#define IFX2X(A, B) if((dyn->insts[ninst].x86.gen_flags==(A) || dyn->insts[ninst].x86.gen_flags==(B) || dyn->insts[ninst].x86.gen_flags==((A)|(B))))
+#define IFXN(A, B)  if((dyn->insts[ninst].x86.gen_flags&(A) && !(dyn->insts[ninst].x86.gen_flags&(B))))
 
 // Generate FCOM with s1 and s2 scratch regs (the VCMP is already done)
 #define FCOM(s1, s2)    \
     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->f.dfnone) {MOVW(S, d_none); STR_IMM9(S, xEmu, offsetof(x86emu_t, df)); dyn->f.dfnone=1;}
+#define SET_DF(S, N)     if(N) {MOVW(S, N); STR_IMM9(S, xEmu, offsetof(x86emu_t, df)); dyn->f.dfnone=0;} else SET_DFNONE(S)
+#define SET_NODF()          dyn->f.dfnone = 0
+#define SET_DFOK()          dyn->f.dfnone = 1
+
+#ifndef MAYSETFLAGS
+#define MAYSETFLAGS()
+#endif
 
 #ifndef READFLAGS
 #define READFLAGS(A) \
-    if(dyn->state_flags!=SF_SET) {                      \
-        if(dyn->state_flags!=SF_PENDING) {              \
+    if(((A)!=X_PEND && dyn->f.pending!=SF_SET)          \
+    && (dyn->f.pending!=SF_SET_PENDING)) {              \
+        if(dyn->f.pending!=SF_PENDING) {                \
             LDR_IMM9(x3, xEmu, offsetof(x86emu_t, df)); \
             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;                      \
+        dyn->f.pending = 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
+#define SETFLAGS(A, B)                                                                          \
+    if(dyn->f.pending!=SF_SET                                                                   \
+    && ((B)&SF_SUB)                                                                             \
+    && (dyn->insts[ninst].x86.gen_flags&(~(A))))                                               \
+        READFLAGS(((dyn->insts[ninst].x86.gen_flags&X_PEND)?X_ALL:dyn->insts[ninst].x86.gen_flags)&(~(A)));\
+    if(dyn->insts[ninst].x86.gen_flags) switch(B) {                                            \
+        case SF_SUBSET:                                                                         \
+        case SF_SET: dyn->f.pending = SF_SET; break;                                            \
+        case SF_PENDING: dyn->f.pending = SF_PENDING; break;                                    \
+        case SF_SUBSET_PENDING:                                                                 \
+        case SF_SET_PENDING:                                                                    \
+            dyn->f.pending = (dyn->insts[ninst].x86.gen_flags&X_PEND)?SF_SET_PENDING:SF_SET;   \
+            break;                                                                              \
+    } else dyn->f.pending = SF_SET
 #endif
 #ifndef JUMP
-#define JUMP(A) 
+#define JUMP(A, C) 
 #endif
 #ifndef BARRIER
 #define BARRIER(A) 
@@ -288,27 +385,26 @@
 #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.gen_flags) {STR_IMM9(A, xEmu, offsetof(x86emu_t, op1));}
+#define UFLAG_OP2(A) if(dyn->insts[ninst].x86.gen_flags) {STR_IMM9(A, xEmu, offsetof(x86emu_t, op2));}
+#define UFLAG_OP12(A1, A2) if(dyn->insts[ninst].x86.gen_flags) {STR_IMM9(A1, xEmu, offsetof(x86emu_t, op1));STR_IMM9(A2, 0, offsetof(x86emu_t, op2));}
+#define UFLAG_RES(A) if(dyn->insts[ninst].x86.gen_flags) {STR_IMM9(A, xEmu, offsetof(x86emu_t, res));}
+#define UFLAG_DF(r, A) if(dyn->insts[ninst].x86.gen_flags) {SET_DF(r, A)}
+#define UFLAG_IF if(dyn->insts[ninst].x86.gen_flags)
 #ifndef DEFAULT
-#define DEFAULT      *ok = -1; BARRIER(2)
-#endif
-#ifndef NEW_BARRIER_INST
-#define NEW_BARRIER_INST
+#define DEFAULT      *ok = -1; BARRIER(BARRIER_NOFLAGS)
 #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
@@ -334,22 +430,23 @@ void* arm_linker(x86emu_t* emu, void** table, uintptr_t addr);
 #define dynarecDF       STEPNAME(dynarecDF)
 #define dynarecF0       STEPNAME(dynarecF0)
 #define dynarec660F     STEPNAME(dynarec660F)
+#define dynarec66F0     STEPNAME(dynarec66F0)
 #define dynarecF20F     STEPNAME(dynarecF20F)
 #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 call_dr         STEPNAME(call_dr_)
+#define call_rd         STEPNAME(call_rd_)
 #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 +507,40 @@ 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_shld32     STEPNAME(emit_shld32)
+
+#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_setround_reset    STEPNAME(x87_setround_reset)
 #define x87_restoreround STEPNAME(x87_restoreround)
+#define x87_swapreg     STEPNAME(x87_swapreg)
+#define sse_setround    STEPNAME(sse_setround)
+#define sse_setround_reset    STEPNAME(sse_setround_reset)
 #define mmx_get_reg     STEPNAME(mmx_get_reg)
 #define mmx_get_reg_empty STEPNAME(mmx_get_reg_empty)
 #define sse_get_reg     STEPNAME(sse_get_reg)
@@ -435,37 +549,40 @@ void* arm_linker(x86emu_t* emu, void** table, uintptr_t addr);
 #define fpu_pushcache   STEPNAME(fpu_pushcache)
 #define fpu_popcache    STEPNAME(fpu_popcache)
 #define fpu_reset       STEPNAME(fpu_reset)
+#define fpu_reset_cache STEPNAME(fpu_reset_cache)
+#define fpu_propagate_stack STEPNAME(fpu_propagate_stack)
 #define fpu_purgecache  STEPNAME(fpu_purgecache)
-#ifdef HAVE_TRACE
+#define x87_purgecache  STEPNAME(x87_purgecache)
+#define mmx_purgecache  STEPNAME(mmx_purgecache)
 #define fpu_reflectcache STEPNAME(fpu_reflectcache)
-#endif
-#define cstack_push     STEPNAME(cstack_push)
-#define cstack_pop      STEPNAME(cstack_pop)
-
+#define fpu_unreflectcache STEPNAME(fpu_unreflectcache)
 
 // get the single reg that from the double "reg" (so Dx[idx])
 #define fpu_get_single_reg      STEPNAME(fpu_get_single_reg)
 // put back (if needed) the single reg in place
 #define fpu_putback_single_reg  STEPNAME(fpu_putback_single_reg)
 
+#define CacheTransform       STEPNAME(CacheTransform)
+
 /* 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, int* l);
+
+/* 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 call_dr(dynarec_arm_t* dyn, int ninst, int reg, int n, int s1, int ret, int ret2, uint32_t mask, int saveflags);
+void call_rd(dynarec_arm_t* dyn, int ninst, void* fnc, int reg, int s1, uint32_t mask, int saveflags);
 void grab_fsdata(dynarec_arm_t* dyn, uintptr_t addr, int ninst, int reg);
 void grab_tlsdata(dynarec_arm_t* dyn, uintptr_t addr, int ninst, int reg);
-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);
 void emit_cmp16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
 void emit_cmp32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
@@ -489,21 +606,21 @@ 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 +643,41 @@ 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_shld32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
+
+void emit_pf(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4);
 
 // 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)
@@ -553,29 +686,74 @@ void x87_forget(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st);
 void x87_reget_st(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st);
 // Set rounding according to cw flags, return reg to restore flags
 int x87_setround(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
+// Set rounding according to cw flags, return reg to restore flags, also enable exceptions and reset counters
+int x87_setround_reset(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
 // Restore round flag
 void x87_restoreround(dynarec_arm_t* dyn, int ninst, int s1);
+// swap 2 x87 regs
+void x87_swapreg(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a, int b);
+// Set rounding according to mxcsr flags, return reg to restore flags
+int sse_setround(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
+// Set rounding according to mxcsr flags, return reg to restore flags, also enable exceptions and reset counters
+int sse_setround_reset(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
+
+void CacheTransform(dynarec_arm_t* dyn, int ninst, int cacheupd, int s1, int s2, int s3);
+
+#if STEP < 2
+#define CHECK_CACHE()   0
+#else
+#define CHECK_CACHE()   (cacheupd = CacheNeedsTransform(dyn, ninst))
+#endif
+
+#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);
-#ifdef HAVE_TRACE
+void fpu_reset(dynarec_arm_t* dyn);
+// reset the cache with n
+void fpu_reset_cache(dynarec_arm_t* dyn, int ninst, int reset_n);
+// propagate stack state
+void fpu_propagate_stack(dynarec_arm_t* dyn, int ninst);
+// purge the FPU cache (needs 3 scratch registers) next=1 if for a conditionnal branch jumping out of block (no tracking updated)
+void fpu_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1, int s2, int s3);
+void x87_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1, int s2, int s3);
+void mmx_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1);
 void fpu_reflectcache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
-#endif
+void fpu_unreflectcache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
 void fpu_pushcache(dynarec_arm_t* dyn, int ninst, int s1);
 void fpu_popcache(dynarec_arm_t* dyn, int ninst, int s1);
 
@@ -600,14 +778,10 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
 uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog);
 uintptr_t dynarecF0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog);
 uintptr_t dynarec660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog);
+uintptr_t dynarec66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, int* ok, int* need_epilog);
 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
@@ -626,4 +800,21 @@ void cstack_pop(dynarec_arm_t* dyn, int ninst, int s0, int s1, int s2);
 #define MAYUSE(A)   
 #endif
 
+#define NOTEST(s1)                                          \
+    if(box86_dynarec_test) {                                \
+        MOVW(s1, 0);                                        \
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, test.test));  \
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, test.clean)); \
+    }
+#define SKIPTEST(s1)                                        \
+    if(box86_dynarec_test) {                                \
+        MOVW(s1, 0);                                        \
+        STR_IMM9(s1, xEmu, offsetof(x86emu_t, test.clean)); \
+    }
+#define GOTEST(s1, s2)                                      \
+    if(box86_dynarec_test) {                                \
+        MOVW(s2, 1);                                        \
+        STR_IMM9(s2, xEmu, offsetof(x86emu_t, test.test));  \
+    }
+
 #endif //__DYNAREC_ARM_HELPER_H__