Bumped to version v0.2.8

Add MOV GS:od, AL

.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-18.04
+          - platform: TEGRAX1
+            type: Trace
+            os: ubuntu-18.04
+          - platform: TEGRAX1ARM32
+            type: Release
+            os: ubuntu-18.04
+          - platform: TEGRAX1ARM32
+            type: Trace
+            os: ubuntu-18.04
+
+    runs-on: ${{ matrix.os }}
+    steps:
+      - name: "Checkout Box86 Repository"
+        uses: actions/checkout@v2
+
+      - name: "Environment preparation"
+        run: |
+          sudo apt-get update
+
+          if [[ ${{ matrix.platform }} != 'I386' && ${{ matrix.platform }} != 'OTHER_ARM' ]]; then
+            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@v2
+        with:
+          name: box86-${{ matrix.platform }}-${{ matrix.type }}
+          path: build/box86
+
+
+

.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

CMakeLists.txt

@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 3.0)
+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})
@@ -6,17 +6,26 @@ option(RPI2 "Set to ON if targeting an RaspberryPI2 device" ${RPI2})
 option(RPI3 "Set to ON if targeting an RaspberryPI3 device" ${RPI3})
 option(RPI4 "Set to ON if targeting an RaspberryPI4 device" ${RPI4})
 option(RPI4ARM64 "Set to ON if targeting an RaspberryPI4 device with multiarch arm64 and armhf" ${RPI4ARM64})
+option(GOA_CLONE "Set to ON if targeting GO Advance clones, like RG351p/v, Gameforce Chi, RGB10..." ${GOA_CLONE})
+option(RK3288 "Set to ON if targeting an Rockchip RK3288 based device" ${RK3288})
 option(RK3399 "Set to ON if targeting an Rockchip RK3399 based device" ${RK3399})
+option(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(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})
-if(PANDORA OR PYRA OR RPI2 OR RPI3 OR RPI4 OR GAMESHELL OR ODROID OR RK3399 OR RPI4ARM64)
+if(PANDORA OR PYRA OR RPI2 OR RPI3 OR RPI4 OR GAMESHELL OR ODROID OR GOA_CLONE OR RK3288 OR RK3399 OR ODROIDN2 OR TEGRAX1 OR TEGRAX1ARM32 OR RPI4ARM64 OR PHYTIUM OR SD845 OR A64 OR LX2160A)
     set(LD80BITS OFF CACHE BOOL "")
     set(NOALIGN OFF CACHE BOOL "")
     set(ARM_DYNAREC ON CACHE BOOL "")
@@ -46,7 +55,13 @@ else()
     set(PYTHON_EXECUTABLE ${Python3_EXECUTABLE} CACHE INTERNAL "The Python3 executable" FORCE)
 endif()
 
-if(RK3399 OR RPI4ARM64)
+cmake_host_system_information(RESULT CMAKE_HOST_SYSTEM_PROCESSOR QUERY OS_PLATFORM)
+string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "i686"  _x86)
+string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "x86_64"  _x86_64)
+string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "aarch64"  _aarch64)
+string(COMPARE EQUAL "${CMAKE_HOST_SYSTEM_PROCESSOR}" "armhf"  _armhf)
+
+if(_aarch64 AND (RK3399 OR ODROIDN2 OR TEGRAX1 OR RPI4ARM64 OR PHYTIUM OR SD845 OR A64 OR LX2160A))
     set(CMAKE_C_COMPILER arm-linux-gnueabihf-gcc)
     #set(CMAKE_ASM_COMPILER arm-linux-gnueabihf-as) #will use gcc in fact
     set(CMAKE_C_COMPILER_TARGET arm-linux-gnueabihf)
@@ -66,50 +81,83 @@ endif()
 if(PANDORA)
     add_definitions(-DPANDORA)
     #add_definitions(-mcpu=cortex-a8 -mfpu=neon -mfloat-abi=softfp -ftree-vectorize -fsingle-precision-constant -ffast-math)
-endif()
-if(GAMESHELL)
-    add_definitions(-DGAMEHELL)
+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")
-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")
-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")
-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")
-endif()
-if(RPI4)
+elseif(RPI4)
     add_definitions(-DRPI)
     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")
-endif()
-if(RPI4ARM64)
+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")
-endif()
-if(ODROID)
+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")
-endif()
-if(RK3399)
+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)
-    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crc+simd+crypto -mcpu=cortex-a72+crypto")
+    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(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()
 
 if(POWER9)
@@ -123,7 +171,7 @@ endif()
 if(BOX86LIB)
     set(BOX86 box86)
 elseif(HAVE_TRACE)
-    set(BOX86 box86t)
+    set(BOX86 box86)
 else()
     set(BOX86 box86)
 endif()
@@ -137,6 +185,7 @@ endif()
 
 if(NOALIGN)
     add_definitions(-DNOALIGN)
+    add_definitions(-pipe -march=core2 -msse2)
 endif()
 
 if(HAVE_TRACE)
@@ -203,6 +252,7 @@ set(ELFLOADER_SRC
     "${BOX86_ROOT}/src/box86context.c"
     "${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"
     "${BOX86_ROOT}/src/tools/fileutils.c"
@@ -216,6 +266,9 @@ set(ELFLOADER_SRC
     "${BOX86_ROOT}/src/elfs/elfload_dump.c"
     "${BOX86_ROOT}/src/librarian/library.c"
     "${BOX86_ROOT}/src/librarian/librarian.c"
+    "${BOX86_ROOT}/src/librarian/dictionnary.c"
+    "${BOX86_ROOT}/src/librarian/symbols.c"
+    "${BOX86_ROOT}/src/librarian/globalsymbols.c"
     "${BOX86_ROOT}/src/emu/x86emu.c"
     "${BOX86_ROOT}/src/emu/x86run.c"
     "${BOX86_ROOT}/src/emu/x86run66.c"
@@ -332,8 +385,11 @@ 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"
     "${BOX86_ROOT}/src/wrapped/wrappedgtk3.c"
@@ -343,11 +399,19 @@ set(WRAPPEDS
     "${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/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"
@@ -376,8 +440,34 @@ set(WRAPPEDS
     "${BOX86_ROOT}/src/wrapped/wrappedlibtiff.c"
     "${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"
+    "${BOX86_ROOT}/src/wrapped/wrappedtbbmallocproxy.c"
 )
-if(NOT ANDROID)
+if(ANDROID)
+    list(APPEND WRAPPEDS
+        "${BOX86_ROOT}/src/wrapped/wrappedandroidshmem.c"
+    )
+else()
     list(APPEND WRAPPEDS
         "${BOX86_ROOT}/src/wrapped/wrappedlibjpeg.c"
         "${BOX86_ROOT}/src/wrapped/wrappedlibjpeg62.c"
@@ -387,7 +477,7 @@ endif()
 
 # If BOX86_ROOT contains a ".c", the build breaks...
 string(REPLACE ".c" "_private.h" MODROOT ${BOX86_ROOT})
-set(WRAPPEDS_HEAD "")
+set(WRAPPEDS_HEAD "${BOX86_ROOT}/src/wrapped/wrappedd3dadapter9_genvate.h")
 foreach(A ${WRAPPEDS})
     string(REPLACE ".c" "_private.h" C ${A})
     string(REPLACE "${MODROOT}" "${BOX86_ROOT}" B ${C})
@@ -456,6 +546,7 @@ if(ARM_DYNAREC)
     "${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"
     "${BOX86_ROOT}/src/dynarec/dynarec_arm_f30f.c"
     )
@@ -513,7 +604,7 @@ if(ARM_DYNAREC)
     if(ANDROID)
         target_link_libraries(${BOX86} dynarec m dl)
     else()
-        target_link_libraries(${BOX86} dynarec m dl rt pthread)
+        target_link_libraries(${BOX86} dynarec m dl rt pthread resolv)
     endif()
 endif()
 if(${CMAKE_VERSION} VERSION_LESS "3.13")
@@ -533,9 +624,10 @@ add_executable(${BOX86} ${ELFLOADER_SRC} ${WRAPPEDS} "${BOX86_ROOT}/src/git_head
 add_dependencies(${BOX86} WRAPPERS)
 add_dependencies(${BOX86} PRINTER)
 if(ANDROID)
-    target_link_libraries(${BOX86} m dl)
+    target_link_libraries(${BOX86} c m dl)
 else()
-    target_link_libraries(${BOX86} m dl rt pthread)
+    #target_link_libraries(${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)
@@ -554,9 +646,6 @@ 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)
@@ -565,6 +654,7 @@ if(NOT _x86 AND NOT _x86_64)
   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()
 
 if(NOT TARGET uninstall)
@@ -577,92 +667,145 @@ if(NOT TARGET uninstall)
         COMMAND ${CMAKE_COMMAND} -P ${CMAKE_CURRENT_BINARY_DIR}/cmake_uninstall.cmake)
 endif()
 
+set(CPACK_GENERATOR "DEB")
+set(CPACK_DEBIAN_PACKAGE_MAINTAINER "ptitSeb")
+set(CPACK_PACKAGE_CONTACT "[email protected]")
+set(CPACK_PACKAGE_DESCRIPTION "Box86 - Linux Userspace x86 Emulator with a twist")
+set(CPACK_DEBIAN_PACKAGE_DEPENDS "libc6")
+set(CPACK_PACKAGE_HOMEPAGE_URL, "https://box86.org")
+file(STRINGS "${BOX86_ROOT}/src/box86version.h" TEMP_STRING REGEX "BOX86_MAJOR\\s*")
+string(REGEX REPLACE "BOX86_MAJOR" "" TEMP_STRING ${TEMP_STRING})
+string(REGEX MATCH "[0-9]" BOX86_MAJOR ${TEMP_STRING})
+file(STRINGS "${BOX86_ROOT}/src/box86version.h" TEMP_STRING REGEX "BOX86_MINOR\\s*")
+string(REGEX REPLACE "BOX86_MINOR" "" TEMP_STRING ${TEMP_STRING})
+string(REGEX MATCH "[0-9]" BOX86_MINOR ${TEMP_STRING})
+file(STRINGS "${BOX86_ROOT}/src/box86version.h" TEMP_STRING REGEX "BOX86_REVISION\\s*")
+string(REGEX REPLACE "BOX86_REVISION" "" TEMP_STRING ${TEMP_STRING})
+string(REGEX MATCH "[0-9]" BOX86_REVISION ${TEMP_STRING})
+set(CPACK_PACKAGE_VERSION_MAJOR ${BOX86_MAJOR})
+set(CPACK_PACKAGE_VERSION_MINOR ${BOX86_MINOR})
+set(CPACK_PACKAGE_VERSION_PATCH ${BOX86_REVISION})
+set(CPACK_DEBIAN_PACKAGE_CONTROL_EXTRA "${CMAKE_CURRENT_SOURCE_DIR}/postinst")
+if(_aarch64 OR _armhf)
+    set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "armhf")
+elseif(_x86_64 OR _x86)
+    set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "i386")
+else()
+    #probably wrong...
+    execute_process(COMMAND "dpkg --print-architecture" OUTPUT_VARIABLE CPACK_DEBIAN_PACKAGE_ARCHITECTURE)
+endif()
+set(CPACK_DEBIAN_FILE_NAME "${BOX86}-${BOX86_MAJOR}.${BOX86_MINOR}.${BOX86_REVISION}_Linux-${CPACK_DEBIAN_PACKAGE_ARCHITECTURE}.deb")
+INCLUDE(CPack)
+
 add_test(NAME bootSyscall COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test01 -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_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(NAME bootSyscallC COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test02 -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_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(NAME printf COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test03 -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_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(NAME args COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test04 -D TEST_ARGS2=yeah -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_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(NAME maths1 COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test05 -D TEST_ARGS2=7 -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_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(NAME threadsStart COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test06 -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_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(NAME trig COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test07 -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_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(NAME pi COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test08 -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_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(NAME fork COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test09 -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_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(NAME cppThreads COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test10 -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_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(NAME tlsData COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test11 -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_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(NAME fpu COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test12 -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_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(NAME contexts COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test13 -D TEST_OUTPUT=tmpfile.txt 
+    -D TEST_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(NAME conditionalThreads COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test14 -D TEST_OUTPUT=tmpfile.txt 
-    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref14.txt
+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 )
 
-# WIP
-#add_test(NAME linkingIndirectNoversion COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
-#    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test15 -D TEST_OUTPUT=tmpfile.txt 
-#    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref15.txt
-#    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
-#
-#add_test(NAME linkingIndirectVersion COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX86} 
-#    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test16 -D TEST_OUTPUT=tmpfile.txt 
-#    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref16.txt
-#    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+add_test(NAME 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")
+
+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()

archlinux/PKGBUILD-arm

@@ -1,44 +0,0 @@
-# Maintainer: SpacingBat3 <[email protected]>
-# Contributor: Sebastien Chevalier <[email protected]>
-pkgname=box86-arm-git
-pkgver=r2798.0edc0374
-pkgrel=1
-pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
-arch=('armv7h')
-url="https://github.com/ptitSeb/box86"
-license=('MIT')
-optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
-makedepends=('git' 'cmake' 'make')
-provides=("${pkgname%-git}" "${_pkgname%-arm-git}-git" "${_pkgname%-arm-git}")
-conflicts=("${pkgname%-git}" "${_pkgname%-arm-git}-git" "${_pkgname%-arm-git}")
-source=('git+https://github.com/ptitSeb/box86')
-md5sums=('SKIP')
-
-pkgver() {
-	cd "$srcdir/${pkgname%-arm-git}"
-	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
-}
-
-prepare() {
-	cd "${srcdir}/${pkgname%-arm-git}"
-	if [[ ! -d ./build ]]; then
-		mkdir build && cd build
-		cmake .. -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo
-	fi
-}
-
-build() {
-	cd "$srcdir/${pkgname%-arm-git}/build"
-	make -j$(nproc)
-}
-
-package() {
-	cd "$srcdir/${pkgname%-arm-git}/build"
-	make DESTDIR="${pkgdir}/" install
-	# /usr/local/bin isn't in PATH by the default,
-	# we should move it to /usr/bin
-	cd ${pkgdir}
-	mv usr/local/bin/ usr/bin/
-	# cleanup when dir is empty
-	rmdir usr/local || exit 0
-}

archlinux/PKGBUILD-odroid

@@ -1,44 +0,0 @@
-# Maintainer: SpacingBat3 <[email protected]>
-# Contributor: Sebastien Chevalier <[email protected]>
-pkgname=box86-odroid-git
-pkgver=r2798.0edc0374
-pkgrel=1
-pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
-arch=('armv7h')
-url="https://github.com/ptitSeb/box86"
-license=('MIT')
-optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
-makedepends=('git' 'cmake' 'make')
-provides=("${pkgname%-git}" "${_pkgname%-odroid-git}-git" "${_pkgname%-odroid-git}")
-conflicts=("${pkgname%-git}" "${_pkgname%-odroid-git}-git" "${_pkgname%-odroid-git}")
-source=('git+https://github.com/ptitSeb/box86')
-md5sums=('SKIP')
-
-pkgver() {
-	cd "$srcdir/${pkgname%-odroid-git}"
-	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
-}
-
-prepare() {
-	cd "${srcdir}/${pkgname%-odroid-git}"
-	if [[ ! -d ./build ]]; then
-		mkdir build && cd build
-		cmake .. -DODROID=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
-	fi
-}
-
-build() {
-	cd "$srcdir/${pkgname%-odroid-git}/build"
-	make -j$(nproc)
-}
-
-package() {
-	cd "$srcdir/${pkgname%-odroid-git}/build"
-	make DESTDIR="${pkgdir}/" install
-	# /usr/local/bin isn't in PATH by the default,
-	# we should move it to /usr/bin
-	cd ${pkgdir}
-	mv usr/local/bin/ usr/bin/
-	# cleanup when dir is empty
-	rmdir usr/local || exit 0
-}

archlinux/PKGBUILD-x86

@@ -1,44 +0,0 @@
-# Maintainer: SpacingBat3 <[email protected]>
-# Contributor: Sebastien Chevalier <[email protected]>
-pkgname=box86-x86-git
-_pkgname=box86-git
-pkgver=r2798.0edc0374
-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=("${pkgname%-x86-git}")
-conflicts=("${pkgname%-x86-git}")
-source=('git+https://github.com/ptitSeb/box86')
-md5sums=('SKIP')
-
-pkgver() {
-	cd "$srcdir/${pkgname%-x86-git}"
-	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
-}
-
-prepare() {
-	cd "${srcdir}/${pkgname%-x86-git}"
-	if [[ ! -d ./build ]]; then
-		mkdir build && cd build
-		cmake .. -DLD80BITS=1 -DNOALIGN=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
-	fi
-}
-
-build() {
-	cd "$srcdir/${pkgname%-x86-git}/build"
-	make -j$(nproc)
-}
-
-package() {
-	cd "$srcdir/${pkgname%-x86-git}/build"
-	make DESTDIR="${pkgdir}/" install
-	# /usr/local/bin isn't in PATH by the default,
-	# we should move it to /usr/bin
-	cd ${pkgdir}
-	mv usr/local/bin/ usr/bin/
-	# cleanup when dir is empty
-	rmdir usr/local || exit 0
-}

debian/changelog

@@ -0,0 +1,23 @@
+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 +1,12 @@
+#!/usr/bin/make -f
+
+%:
+	dh $@  
+
+override_dh_auto_configure:
+	dh_auto_configure -- \
+	      -DCMAKE_BUILD_TYPE=RelWithDebInfo \
+	      -DARM_DYNAREC=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

@@ -1,3 +1,66 @@
+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

docs/COMPILE.md

@@ -1,5 +1,17 @@
-Compiling
+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 apt-key add -
+sudo apt update && sudo apt install box86 -y
+```
+
+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`
@@ -14,68 +26,142 @@ Compiling
 
 #### 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)_
+ _A build for model 2, 3 and 4 can be done. Model 1 and 0 cannot (at least not with Dynarec, as they lack NEON support)_
  
 ```
 git clone https://github.com/ptitSeb/box86
 cd box86
-mkdir build; cd build; cmake .. -DRPI4=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
-make -j4
+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 Pi4. Change to RPI2 or RPI3 for other models.  Change to RPI4ARM64 for compiling on arm64. (armhf multiarch or chroot required alongside armhf gcc. Install it with 'sudo apt install gcc-arm-linux-gnueabihf'.)_
+#### 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 `-DRPI2ARM64=1` etc for other models)
+make -j2
+sudo make install
+sudo systemctl restart systemd-binfmt
+```
 
 #### for ODROID
 
-`mkdir build; cd build; cmake .. -DODROID=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j$(nproc)`
+`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 -j4`
+`mkdir build; cd build; cmake .. -DRK3399=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
 
-As most RK3399 device run AARCH64 OS, you'll need an `armhf` multiarch environment, and an armhf gcc: On debian, install it with `sudo apt install gcc-arm-linux-gnueabihf`.
+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 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 ARM Linux platforms
 
- `mkdir build; cd build; cmake .. -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j$(nproc)`
+ `mkdir build; cd build; cmake .. -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
 
 #### for x86 Linux
 
- `mkdir build; cd build; cmake .. -DLD80BITS=1 -DNOALIGN=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j$(nproc)`
+ `mkdir build; cd build; cmake .. -DCMAKE_C_FLAGS="-m32" -DLD80BITS=1 -DNOALIGN=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j3`
 
-#### use ccmake
+### use ccmake
 
 Alternatively, you can use the curses-bases ccmake (or any other gui frontend for cmake) to select wich platform to use interactively.
 
-#### Customize your build
+### Customize your build
 
-*use ccache if you have it* 
+#### 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* 
+#### 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
 
-To have a trace enabled build ( ***it will be slower***), add `-DHAVE_TRACE=1` but you will need, at runtime, to have the [Zydis library](https://github.com/zyantific/zydis) library in your `LD_LIBRARY_PATH` or in the system library folders.
+To have 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*
+#### to have ARM Dynarec
 
-The Dynarec is only available on the ARM architecture(Right now, anyways.). Notes also that VFPv3 and NEON are required for the Dynarec. Activate it by using `-DARM_DYNAREC=1`. Also, be sure to use `-marm` in compilation flags (because many compileur use Thumb as default, and the dynarec will not work in this mode).
+###### *Note: VFPv3 and NEON are required for Dynarec.*
 
-*not building from a git clone*
+###### *Note: Compiling with `ARM_DYNAREC` without selecting a hardware profile is not advised.*
 
-If you are not building from a git clone (for example, downloading a release source zipped from github), you need to activate `-DNOGIT=1` from cmake to be able to build (normal process include git sha1 of HEAD in the version that box86 print).
+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.
-They can be launched with `ctest`
-They are very basic and don't test much for now.
+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 with `DEB_BUILD_OPTIONS=nostrip dpkg-buildpackage -us -uc -nc`.
 

docs/README.md

@@ -1,39 +1,45 @@
-# box86
-
 ![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) | [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)
 
-Compiling/Installation
 ----
-> Compilation instructions can be found [here](COMPILE.md)  
-> Instructions for installing Wine for Box86 can be found [here](X86WINE.md)  
 
 Box86 lets you run x86 Linux programs (such as games) on non-x86 Linux systems, like ARM (host system needs to be 32bit little-endian).
 
-You *NEED* a 32-bit subsystem to run and build Box86. Box86 is useless on 64-bit only systems. Also, you *NEED* a 32-bit toolchain to build Box86. A toolchain that only supports 64-bit will not compile Box86, and you'll get errors (typically on aarch64, you get "-marm" not recognized, and you'll need a multiarch or chroot environnement).
+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 some cases.
+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 an solution like [gl4es](https://github.com/ptitSeb/gl4es) is usually necessary. (Most ARM platforms only support OpenGL ES and/or their OpenGL implementation is dodgy (see OpenGL on Android).)
+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.
+Box86 now integrates a DynaRec (dynamic recompiler) for the ARM platform, providing a speed boost between 5 to 10 times faster than only using the interpreter. Some high level information on how the Dynarec work can be found [here](https://box86.org/2021/07/inner-workings-a-high%e2%80%91level-view-of-box86-and-a-low%e2%80%91level-view-of-the-dynarec/).
 
-Many games already work without much tweaking, for example: WorldOfGoo, Airline Tycoon Deluxe, and FTL. Many of the GameMaker Linux games also run fine. (there's a long list, among them are UNDERTALE, A Risk of Rain, and Cook Serve Delicious)
+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
 
-Here's  6 videos, the first 2 videos are videos of "Airline Tycoon Deluxe" and "Heretic 2" running on a GigaHertz OpenPandora (the second one is using  the dynarec), and the next 2 videos are videos of of "Bit.Trip.Runner" and "Neverwinter Night" running on an ODroid XU4 (without dynarec), and the last 2  videos are on on a Pi4: Shovel Knight (video from @ITotalJustice) and Freedom Planet (video from @djazz), also without dynarec.
+----
+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 video on Youtube in the [Pi Labs Channel](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g) or [The Byteman Channel](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA)
+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 there: https://github.com/ptitSeb/box86-compatibility-list/issues
+Compatibility list is here: https://github.com/ptitSeb/box86-compatibility-list/issues
 
 <img src="img/Box86Icon.png" width="96" height="96">
 
@@ -43,13 +49,6 @@ Note that this project is not to be mistaken with [86box](https://github.com/86B
 
 ----
 
-Discord
-----
-
-If you want to discuss (or have any problems) about box86, feel free to join the [Pi Labs Discord Server](https://discord.gg/Fh8sjmu)!
-
-----
-
 Usage
 ----
 
@@ -57,7 +56,7 @@ There are a few environment variables to control the behaviour of Box86.
 
 See [here](USAGE.md) for all environment variables and what they do.
 
-Note: Box86's Dynarec uses a mechanism with Memory Protection and a SegFault signal handler to handle JIT code. In simpler terms, if you want to use GDB to debug a running program that use JIT'd code (like mono/Unity3D), you will still have many "normal" segfaults triggering. It is suggested to use something like `handle SIGSEGV nostop` in GDB to not stop at each segfault, and maybe put a breakpoint inside `my_memprotectionhandler` in `signals.c` if you want to trap SegFaults.
+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.
 
 ----
 
@@ -73,17 +72,15 @@ 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.
 
-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 a 32-bit chroot) to run x86 binaries (in fact, the same is the case on actual x86_64 Linux).
+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).
 
 ----
 
-A note about Unity game emulation
+Notes about Unity game emulation
 ----
 
-Running Unity games is a hit or miss for now. Unity uses Mono (which uses signals that are not well emulated enough), and a runtime embedded in the main binary. A solution would be to use a native version of the libmono library used by Unity (it can be found here: https://github.com/Unity-Technologies/mono and needs to be built from source). But the wrapping of this lib is tricky, and not done for now, so the only solution is to emulate everything. The tricky part is to emulate the "JIT" code emitted by Mono, however with the new "protected memory" mechanism implemented it should be running with correct performance now.
-You should also note that some Unity3D games require OpenGL 3+ which can be tricky to provide on ARM SBC (single-board computers) for now.
-
-TL;DR: Not all Unity games work and can require a high OpenGL profile, but the speed, for the ones running, should be correct now.
+Running Unity games shoudl generaly work now, but you should also note that many Unity3D games require OpenGL 3+ which can be tricky to provide on ARM SBC (single-board computers) for now.
+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.
 
 ----
 
@@ -92,10 +89,11 @@ Notes about Steam
 
 Linux Steam's can run now with box86. But it's still a bit unstable, and not everything works:
 - First problem is Steam crashing after the sign-in window, if you encounter this issue, you may need to add libappindicator. To install it on Debian, run `sudo apt install libappindicator1`.
-- If you select to "Remember password", Steam is crashing on subsequent starts, unless you have libnm intalled. To install it on Debian, run `sudo apt install libnm0`.
+- If you select to "Remember password", Steam is crashing on subsequent starts, unless you have libnm installed. To install it on Debian, run `sudo apt install libnm0`.
+- If you have trouble installing Steam, you can find `install_steam.sh` in the root folder of the box86 repo. This simple script will download and install steam in your home folder, and then create a shortcut to steam in `/usr/local/bin` (and for this it will ask for sudo permission). Simply use `steam` to launch once it's installed. Note that the installation, being in the Home folder, will only work for a single user. Don't use this script if you need a multi-user installation.
 - Once open, Steam will only work on "Small Mode" and in "Big Picture", not in the regular "Large Mode". This is because some Steam components used in the browser view are only 64-bit now. So go in the "View" menu and switch to "Small view", else the list will stay empty. Alternatively, Steam can be started in small mode directly by using `+open steam://open/minigameslist` command line arguments.
-- To avoid the "libc.so.6 is absent" message, you can use `STEAMOS=1` and `STEAM_RUNTIME=1` as environment variables. 
-- Some Steam games (most Source engine games, like "Portal" or "Half-Life 2") use libtcmalloc. Box86 will detect it and will try to LD_PRELOAD it, for better compatibility. While it should work without the aformentionned feature, it is safer to add it to your system if you intend to play those game. To install it on Debian, run `sudo apt install libtcmalloc-minimal4`.
+- To avoid the "libc.so.6 is absent" message, you can use `STEAMOS=1` and `STEAM_RUNTIME=1` as environment variables (it's automatically there if you used the `install_steam.sh` script)
+- Some Steam games (most Source engine games, like "Portal" or "Half-Life 2") use libtcmalloc. Box86 will detect it and will try to LD_PRELOAD it, for better compatibility. While it should work without the aforementioned feature, it is safer to add it to your system if you intend to play those games. To install it on Debian, run `sudo apt install libtcmalloc-minimal4`.
 
 Steam for Windows installs fine but doesn't work yet.
 
@@ -104,11 +102,28 @@ Steam for Windows installs fine but doesn't work yet.
 Notes about Wine
 ----
 
-Wine is now partly supported. Wine integrated program all runs, and some windows programs and games now runs fine. Don't forget most Windows games use Direct3D, this 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.
+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.
+
 ----
 
-Final words
+Notes about Vulkan
 ----
 
-(If you use Box86 in your project, please don't forget to mention Box86!)
+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) | [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/USAGE.md

@@ -9,7 +9,18 @@ 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.
+ * 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
@@ -31,6 +42,7 @@ Enables/Disables the logging of `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
@@ -109,6 +121,34 @@ Enables/Disables trace for generated code.
 Forbid dynablock creation in the interval specified (helpfull for debugging behaviour difference between Dynarec and Interpretor)
  * 0xXXXXXXXX-0xYYYYYYYY : define the interval where dynablock cannot start (inclusive-exclusive)
 
+#### BOX86_DYNAREC_BIGBLOCK
+Enables/Disables Box86's Dynarec building BigBlock.
+ * 0 : Don't try to build block as big as possible (can help program using lots of thread and a JIT, like C#/Unity) (Default when libmonobdwgc-2.0.so is loaded)
+ * 1 : Build Dynarec block as big as possible (Default.)
+ * 2 : Build Dynarec block bigger (don't stop when block overlaps)
+
+#### 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_SAFEFLAGS
+Handling of flags on CALL/RET opcodes
+* 0 : Treat CALL/RET as if it never needs any flags (faster but may have side-effects)
+* 1 : most of RET will need flags, most of CALLS will not (Default)
+* 2 : All CALL/RET will need flags (slower, but might be needed. Automatically enabled for Vara.exe)
+
 #### BOX86_LIBGL
  * libXXXX set the name for libGL (defaults to libGL.so.1).
  * /PATH/TO/libGLXXX : Sets the name and path for libGL
@@ -123,11 +163,26 @@ Forbid dynablock creation in the interval specified (helpfull for debugging beha
 Some games uses an old version of some libraries, with an ABI incompatible with native version.
 Note that LittleInferno for example is auto detected, and libvorbis.so.0 is automatical added to emulated libs, and same for Don't Starve (and Together / Server variant) that use an old SDL2 too
 
-#### BOX86_ALLOWMISSINGLIBS
-Allow Box86 to continue even if a library is missing.
+#### 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.)
@@ -136,18 +191,45 @@ Disables the load of pulseaudio libraries.
 #### BOX86_NOGTK
 Disables the loading of wrapped GTK libraries.
  * 0 : Load wrapped GTK libraries if found. (Default.)
- * 1 : Disables loading wrapped GTK libraries (can be used with Steam, along with STEAM_RUNTIME=1 to use the i386 version of GTK).
+ * 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_BACKTRACE
+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)

docs/X86WINE.md

@@ -1,73 +1,31 @@
 # Installing Wine (and winetricks)
-Using Wine with Box86 allows (x86) Windows programs to run on ARM Linux computers (x64 is not yet implemented).  
+_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).
 
-_TwisterOS users, take note: Wine, winetricks, and Box86 are pre-installed in TwisterOS. TwisterOS users do not have to install anything._
+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 actually 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 Windows and Wine i386 libraries (`.so` or `.dll` files) so that they will work with your ARM device's 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.
+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) versions of Wine install files (even though we are installing it on an ARM processor). Box86 "wraps" many of Wine's core Linux i386 libraries so that their calls are interpretable by Linux ARM libraries. Below are examples of how to install Wine from each of those places.
-
-_Raspberry Pi users: Wine requires a 3G/1G split memory kernel. Raspberry Pi OS for the Pi 4 has a 3G/1G split kernel, but **the Pi 3 and earlier models have a 2G/2G kernel by default and will need you to install a custom-compiled 3G/1G kernel to get Wine to work.**
+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
-_These examples were all tested on Raspberry Pi 4.  Note the similarities between the different install methods._
-
-### Installing wine-devel, wine-stable, or wine-staging for Raspberry Pi OS Buster from WineHQ deb files
-_Links from the [WineHQ repo](https://dl.winehq.org/wine-builds/debian/dists/buster/main/binary-i386/)_
-```
-# Backup any old wine installations
-sudo mv ~/wine ~/wine-old
-sudo mv ~/.wine ~/.wine-old
-sudo mv /usr/local/bin/wine /usr/local/bin/wine-old
-sudo mv /usr/local/bin/wineboot /usr/local/bin/wineboot-old
-sudo mv /usr/local/bin/winecfg /usr/local/bin/winecfg-old
-sudo mv /usr/local/bin/wineserver /usr/local/bin/wineserver-old
-
-# Download, extract wine, and install wine
-# (Replace the links/versions below with links/versions from the WineHQ site for the version of wine you wish to install. Note that we need the i386 version for Box86 even though we're installing it on our ARM processor.)
-# (Pick an i386 version of wine-devel, wine-staging, or wine-stable)
-cd ~/Downloads
-wget https://dl.winehq.org/wine-builds/debian/dists/buster/main/binary-i386/wine-devel-i386_5.21~buster_i386.deb # NOTE: Replace this link with the version you want
-wget https://dl.winehq.org/wine-builds/debian/dists/buster/main/binary-i386/wine-devel_5.21~buster_i386.deb  # NOTE: Also replace this link with the version you want
-dpkg-deb -xv wine-devel-i386_5.21~buster_i386.deb wine-installer # NOTE: Make sure these dpkg command matches the filename of the deb package you just downloaded
-dpkg-deb -xv wine-devel_5.21~buster_i386.deb wine-installer
-mv ~/Downloads/wine-installer/opt/wine* ~/wine
-rm wine*.deb # clean up
-rm -rf wine-installer # clean up
-
-# Install shortcuts (make 32bit launcher & symlinks. Credits: grayduck, Botspot)
-echo -e '#!/bin/bash\nsetarch linux32 -L '"$HOME/wine/bin/wine "'"$@"' | sudo tee -a /usr/local/bin/wine >/dev/null # Create a script to launch wine programs as 32bit only
-#sudo ln -s ~/wine/bin/wine /usr/local/bin/wine # You could aslo just make a symlink, but box86 only works for 32bit apps at the moment
-sudo ln -s ~/wine/bin/wineboot /usr/local/bin/wineboot
-sudo ln -s ~/wine/bin/winecfg /usr/local/bin/winecfg
-sudo ln -s ~/wine/bin/wineserver /usr/local/bin/wineserver
-sudo chmod +x /usr/local/bin/wine /usr/local/bin/wineboot /usr/local/bin/winecfg /usr/local/bin/wineserver
-
-# These packages are needed for running wine-staging on RPi 4 (Credits: chills340)
-sudo apt install libstb0 -y
-cd ~/Downloads
-wget -r -l1 -np -nd -A "libfaudio0_*~bpo10+1_i386.deb" http://ftp.us.debian.org/debian/pool/main/f/faudio/ # Download libfaudio i386 no matter its version number
-dpkg-deb -xv libfaudio0_*~bpo10+1_i386.deb libfaudio
-sudo cp -TRv libfaudio/usr/ /usr/
-rm libfaudio0_*~bpo10+1_i386.deb # clean up
-rm -rf libfaudio # clean up
-
-# Boot wine (make fresh wineprefix in ~/.wine )
-wine wineboot
-```
-
-### Installing Wine for Raspberry Pi OS from the Twister OS FAQ page
+### 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
@@ -93,26 +51,101 @@ sudo chmod +x /usr/local/bin/wine /usr/local/bin/wineboot /usr/local/bin/winecfg
 # Boot wine (make fresh wineprefix in ~/.wine )
 wine wineboot
 ```
-## Installing winetricks
-Winetricks is a script which makes it easier to install & configure any desired Windows core system software packages which may be dependencies for certain Windows programs.
 
+### 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.
 ```
-# Backup old winetricks
-sudo mv /usr/local/bin/winetricks /usr/local/bin/winetricks-old
+### 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/
 
-# Download
+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://raw.githubusercontent.com/Winetricks/winetricks/master/src/winetricks
+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
-sudo chmod +x winetricks
-sudo mv winetricks /usr/local/bin/
+# 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
 
-# winetricks needs this installed
-sudo apt-get install cabextract -y
+# 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
 ```
 
-If you have a desktop menu itm for wintricks, you may need to edit it to add the nobanner env var. Do so by edition `/usr/share/applications/winetricks.desktop` with your favorite text editor (not that you'll need `sudo` to save the file) and change from `Exec=winetricks --gui` to `Exec=env BOX86_NOBANNER=1 winetricks --gui`
+## 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:  
@@ -126,13 +159,11 @@ For a list of all the different Windows packages & libraries that winetricks can
 ## Other notes
 
 ### Wineprefixes (and Wine initialization)
-The first time Wine is run (`wine wineboot`), it will create a fresh wineprefix for you (by default, located in the hidden folder `~/.wine`).  Think of a wineprefix as Wine's virtual 'harddrive' where it installs software and saves settings. Wineprefixes are portable and deletable.
+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)
-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). For more Wine documentation, see [WineHQ](https://www.winehq.org/documentation).
-
 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
@@ -140,4 +171,4 @@ Some versions of Wine work better with certain software. It is best to install a
 
 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 make a fresh wineprefix with your new version of Wine. You can check which version of Wine you're running with the `wine --version` command.
+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.

install_steam.sh

@@ -0,0 +1,40 @@
+#!/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
+~/steam/bin/steam -noreactlogin 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 
+MACHINE_TYPE=`uname -m`
+if [ ${MACHINE_TYPE} == 'aarch64' ] && [ -f '/etc/debian_version' ]; then
+ echo "Detected 64 bit ARM Debian. Installing 32 bit libraries"
+ sudo dpkg --add-architecture armhf # enable installation of armhf libraries
+ sudo apt update # update package lists with the newly added arch
+ sudo apt install libc6:armhf libncurses5:armhf libsdl2*:armhf libopenal*:armhf libpng*:armhf libfontconfig*:armhf libXcomposite*:armhf libbz2-dev:armhf libXtst*:armhf # install the libraries that Steam requires
+ echo "Don't forget to compile/install Box64!"
+fi
+
+echo "Script complete."

pkgbuilds/PKGBUILD-arm

@@ -0,0 +1,29 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-arm-git
+pkgver=r3858.5368285
+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)"
+}
+
+build() {
+	cmake -B build -S "${pkgname%-arm-git}" -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+	make -C build
+}
+
+package() {
+	make -C build DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-gameshell

@@ -1,7 +1,7 @@
-# Maintainer: SpacingBat3 <[email protected]>
-# Contributor: Sebastien Chevalier <[email protected]>
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-gameshell-git
-pkgver=r2798.0edc0374
+pkgver=r3858.5368285
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -19,26 +19,11 @@ pkgver() {
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }
 
-prepare() {
-	cd "${srcdir}/${pkgname%-gameshell-git}"
-	if [[ ! -d ./build ]]; then
-		mkdir build && cd build
-		cmake .. -DGAMESHELL=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
-	fi
-}
-
 build() {
-	cd "$srcdir/${pkgname%-gameshell-git}/build"
-	make -j$(nproc)
+	cmake -B build -S "${pkgname%-gameshell-git}" -DGAMESHELL=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+	make -C build
 }
 
 package() {
-	cd "$srcdir/${pkgname%-gameshell-git}/build"
-	make DESTDIR="${pkgdir}/" install
-	# /usr/local/bin isn't in PATH by the default,
-	# we should move it to /usr/bin
-	cd ${pkgdir}
-	mv usr/local/bin/ usr/bin/
-	# cleanup when dir is empty
-	rmdir usr/local || exit 0
+	make -C build DESTDIR="${pkgdir}/" install
 }

pkgbuilds/PKGBUILD-odroid

@@ -0,0 +1,29 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-odroid-git
+pkgver=r3858.5368285
+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)"
+}
+
+build() {
+	cmake -B build -S "${pkgname%-odroid-git}" -DODROID=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+	make -C build
+}
+
+package() {
+	make -C build DESTDIR="${pkgdir}/" install
+}

pkgbuilds/PKGBUILD-pandora

@@ -1,8 +1,7 @@
-# Maintainer: SpacingBat3 <[email protected]>
-# Contributor: Sebastien Chevalier <[email protected]>
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-pandora-git
-_pkgname=box86-git
-pkgver=r2798.0edc0374
+pkgver=r3858.5368285
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -20,26 +19,11 @@ pkgver() {
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }
 
-prepare() {
-	cd "${srcdir}/${pkgname%-pandora-git}"
-	if [[ ! -d ./build ]]; then
-		mkdir build && cd build
-		cmake .. -DPANDORA=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
-	fi
-}
-
 build() {
-	cd "$srcdir/${pkgname%-pandora-git}/build"
-	make -j$(nproc)
+	cmake -B build -S "${pkgname%-pandora-git}" -DPANDORA=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+	make -C build
 }
 
 package() {
-	cd "$srcdir/${pkgname%-git}/build"
-	make DESTDIR="${pkgdir}/" install
-	# /usr/local/bin isn't in PATH by the default,
-	# we should move it to /usr/bin
-	cd ${pkgdir}
-	mv usr/local/bin/ usr/bin/
-	# cleanup when dir is empty
-	rmdir usr/local || exit 0
+	make -C build DESTDIR="${pkgdir}/" install
 }

pkgbuilds/PKGBUILD-pyra

@@ -1,8 +1,7 @@
-# Maintainer: SpacingBat3 <[email protected]>
-# Contributor: Sebastien Chevalier <[email protected]>
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-pyra-git
-_pkgname=box86-git
-pkgver=r2798.0edc0374
+pkgver=r3858.5368285
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -20,26 +19,11 @@ pkgver() {
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }
 
-prepare() {
-	cd "${srcdir}/${pkgname%-pyra-git}"
-	if [[ ! -d ./build ]]; then
-		mkdir build && cd build
-		cmake .. -DPYRA=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
-	fi
-}
-
 build() {
-	cd "$srcdir/${pkgname%-pyra-git}/build"
-	make -j$(nproc)
+	cmake -B build -S "${pkgname%-pyra-git}" -DPYRA=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+	make -C build
 }
 
 package() {
-	cd "$srcdir/${pkgname%-pyra-git}/build"
-	make DESTDIR="${pkgdir}/" install
-	# /usr/local/bin isn't in PATH by the default,
-	# we should move it to /usr/bin
-	cd ${pkgdir}
-	mv usr/local/bin/ usr/bin/
-	# cleanup when dir is empty
-	rmdir usr/local || exit 0
+	make -C build DESTDIR="${pkgdir}/" install
 }

pkgbuilds/PKGBUILD-raspberrypi

@@ -1,8 +1,7 @@
-# Maintainer: SpacingBat3 <[email protected]>
-# Contributor: Sebastien Chevalier <[email protected]>
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-rpi-git
-_pkgname=box86-git
-pkgver=r2798.0edc0374
+pkgver=r3858.5368285
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -10,8 +9,8 @@ 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}")
+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`
@@ -21,8 +20,7 @@ pkgver() {
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }
 
-prepare() {
-	cd "${srcdir}/${pkgname%-rpi-git}"
+build() {
 	# Check for Pi version number
 	if [[ ${_pimodel} =~ "Raspberry Pi 4" ]]; then
 		_piversion=4
@@ -37,24 +35,11 @@ prepare() {
 		exit 1
 	fi
 	echo -e "\033[01m\033[01;34m==> INFO:\033[00;01m Your detected Pi verion number is: \033[01;31m${_piversion}\033[01m.\033[00m"
-	if [[ ! -d ./build ]]; then
-		mkdir build && cd build
-		cmake .. -DRPI${_piversion}=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
-	fi
-}
 
-build() {
-	cd "$srcdir/${pkgname%-rpi-git}/build"
-	make -j$(nproc)
+	cmake -B build -S "${pkgname%-rpi-git}" -DRPI${_piversion}=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+	make -C build
 }
 
 package() {
-	cd "$srcdir/${pkgname%-rpi-git}/build"
-	make DESTDIR="${pkgdir}/" install
-	# /usr/local/bin isn't in PATH by the default,
-	# we should move it to /usr/bin
-	cd ${pkgdir}
-	mv usr/local/bin/ usr/bin/
-	# cleanup when dir is empty
-	rmdir usr/local || exit 0
+	make -C build DESTDIR="${pkgdir}/" install
 }

pkgbuilds/PKGBUILD-rk3399

@@ -1,8 +1,7 @@
-# Maintainer: SpacingBat3 <[email protected]>
-# Contributor: Sebastien Chevalier <[email protected]>
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-rk3399-git
-_pkgname=box86-git
-pkgver=r2798.0edc0374
+pkgver=r3858.5368285
 pkgrel=1
 pkgdesc="Linux Userspace x86 Emulator with a twist, targeted at ARM Linux devices."
 arch=('armv7h')
@@ -20,20 +19,11 @@ pkgver() {
 	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
 }
 
-prepare() {
-	cd "${srcdir}/${pkgname%-rk3399-git}"
-	if [[ ! -d ./build ]]; then
-		mkdir build && cd build
-		cmake .. -DRK3399=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
-	fi
-}
-
 build() {
-	cd "$srcdir/${pkgname%-rk3399-git}/build"
-	make -j$(nproc)
+	cmake -B build -S "${pkgname%-rk3399-git}" -DRK3399=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+	make -C build
 }
 
 package() {
-	cd "$srcdir/${pkgname%-rk3399-git}/build"
-	make DESTDIR="${pkgdir}/" install
+	make -C build DESTDIR="${pkgdir}/" install
 }

pkgbuilds/PKGBUILD-x86

@@ -0,0 +1,28 @@
+# Maintainer: Jai-JAP <[email protected]>, SpacingBat3 <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box86-x86-git
+pkgver=r3858.5368285
+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)"
+}
+
+build() {
+	cmake -B build -S "${pkgname%-x86-git}" -DLD80BITS=1 -DNOALIGN=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo -DCMAKE_INSTALL_PREFIX=/usr
+	make -C build
+}
+
+package() {
+	make -C build DESTDIR="${pkgdir}/" install
+}

pkgbuilds/README.md

@@ -12,10 +12,13 @@ This will copy the PKGBUILD for your platform (where you type your platform name
 In the future, I might try to do an unified PKGBUILD for all platforms that will recognize your platform automatically.
 
 ### 2. What platforms are currently supported.
-All specified in [COMPILE.md](../COMPILE.md).
+All specified in [COMPILE.md](../docs/COMPILE.md).
 
 ### 3. Any advantages with that sollution?
 This might give you some advantages like those:
 - automatic updates (with some AUR frontends like [`pamac-aur`](https://aur.archlinux.org/packages/pamac-aur/) or [`yay`](https://aur.archlinux.org/packages/yay));
 - easily uninstall with `pacman -R` when no needed (no need other scripts nor Makefile to do this);
 - your `box86` installation is recognized as a package (where it wasn't when installing with `make install`).
+
+
+### Note: You can use [makedeb](https://github.com/makedeb/makedeb) (or) [pkgbuild2spec](https://github.com/prozum/pkgbuild2spec) to compile box86 to .deb or .rpm installer files.

postinst

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

rebuild_printer.py

@@ -106,7 +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:
@@ -119,7 +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
@@ -134,7 +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:
@@ -605,7 +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):
@@ -621,7 +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":
@@ -858,7 +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):
@@ -1156,7 +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):
@@ -1456,7 +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
@@ -1675,300 +1675,300 @@ 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) >> (2*(imm12 >> 8))) | ((imm12 & 0xFF) << (32 - 2*(imm12 >> 8))))
-
-const char* print_register_list(int list, int size) {
-	int last = -2;
-	int cnt = 0;
-	static __thread char ret[68];
-	
-	ret[0] = '{';
-	ret[1] = '\\0';
-	for (int i = 0; i < size; ++i) {
-		if (list & (1 << i)) {
-			if (last >= 0) {
-				++cnt;
+			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[6];
+		for (int cur = start; cur < start + size - 1; ++cur) {
+			sprintf(tmp, "%c%d, ", regChr, cur);
+			strcat(ret, tmp);
+		}
+		sprintf(tmp, "%c%d}", regChr, start + size - 1);
 		strcat(ret, tmp);
+		
+		return ret;
 	}
-	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", 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", "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", "last_run.txt"), 'w') as file:
 		file.write('\n'.join(insts))
@@ -1980,6 +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

@@ -52,5 +52,10 @@ if( TEST_RESULT )
   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

@@ -21,6 +21,7 @@
 #include "signals.h"
 #include <sys/mman.h>
 #include "custommem.h"
+#include "dictionnary.h"
 #ifdef DYNAREC
 #include "dynablock.h"
 #include "dynarec/arm_lock_helper.h"
@@ -36,6 +37,7 @@ void initAllHelpers(box86context_t* context)
         return;
     my_context = context;
     init_pthread_helper();
+    init_bridge_helper();
     init_signal_helper(context);
     inited = 1;
 }
@@ -48,7 +50,7 @@ void finiAllHelpers(box86context_t* context)
         return;
     fini_pthread_helper(context);
     fini_signal_helper();
-    cleanAlternate();
+    fini_bridge_helper();
     fini_custommem_helper(context);
     finied = 1;
 }
@@ -72,8 +74,8 @@ 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()
@@ -96,6 +98,7 @@ int unlockMutex()
     #endif
     GO(my_context->mutex_tls, 9)
     GO(my_context->mutex_thread, 10)
+    GO(my_context->mutex_bridge, 11)
     #undef GO
 
     return ret;
@@ -103,7 +106,6 @@ int unlockMutex()
 
 void relockMutex(int locks)
 {
-    relockCustommemMutex(locks);
     #define GO(A, B)                    \
         if(locks&(1<<B))                \
             pthread_mutex_lock(&A);     \
@@ -118,7 +120,9 @@ void relockMutex(int locks)
     #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)
@@ -129,13 +133,14 @@ static void init_mutexes(box86context_t* context)
     pthread_mutex_init(&context->mutex_once, &attr);
     pthread_mutex_init(&context->mutex_once2, &attr);
     pthread_mutex_init(&context->mutex_trace, &attr);
-#ifndef DYNAREC
-    pthread_mutex_init(&context->mutex_lock, &attr);
-#else
+#ifdef DYNAREC
     pthread_mutex_init(&context->mutex_dyndump, &attr);
+#else
+    pthread_mutex_init(&context->mutex_lock, &attr);
 #endif
     pthread_mutex_init(&context->mutex_tls, &attr);
     pthread_mutex_init(&context->mutex_thread, &attr);
+    pthread_mutex_init(&context->mutex_bridge, &attr);
 
     pthread_mutexattr_destroy(&attr);
 }
@@ -146,6 +151,31 @@ static void atfork_child_box64context(void)
     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
 box86context_t *NewBox86Context(int argc)
 {
@@ -156,7 +186,9 @@ box86context_t *NewBox86Context(int argc)
     }
 #endif
     // init and put default values
-    box86context_t *context = my_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;
@@ -169,9 +201,13 @@ box86context_t *NewBox86Context(int argc)
 
     context->maplib = NewLibrarian(context, 1);
     context->local_maplib = NewLibrarian(context, 1);
+    context->versym = NewDictionnary();
     context->system = NewBridge();
+    context->globaldefver = NewDefaultVersion();
+    context->weakdefver = NewDefaultVersion();
     // create vsyscall
-    context->vsyscall = AddBridge(context->system, vFv, x86Syscall, 0);
+    context->vsyscall = AddBridge(context->system, iFEv, x86Syscall, 0, NULL);
+    addAlternate((void*)0xffffe400, (void*)context->vsyscall);
 #ifdef BUILD_LIB
     context->box86lib = RTLD_DEFAULT;   // not ideal
 #else
@@ -180,7 +216,7 @@ box86context_t *NewBox86Context(int argc)
     context->dlprivate = NewDLPrivate();
 
     context->argc = argc;
-    context->argv = (char**)calloc(context->argc+1, sizeof(char*));
+    context->argv = (char**)box_calloc(context->argc+1, sizeof(char*));
 
     init_mutexes(context);
     pthread_atfork(NULL, NULL, atfork_child_box64context);
@@ -217,11 +253,12 @@ void FreeBox86Context(box86context_t** context)
         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);
 
@@ -235,16 +272,16 @@ void FreeBox86Context(box86context_t** context)
         DeleteX86Trace(ctx);
 
     if(ctx->deferedInitList)
-        free(ctx->deferedInitList);
+        box_free(ctx->deferedInitList);
 
-    free(ctx->argv);
+    /*box_free(ctx->argv);*/
     
-    for (int i=0; i<ctx->envc; ++i)
-        free(ctx->envv[i]);
-    free(ctx->envv);
+    /*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;
     }
@@ -265,16 +302,20 @@ void FreeBox86Context(box86context_t** context)
 
     FreeDLPrivate(&ctx->dlprivate);
 
-//    free(ctx->stack); // don't free the stack, it's owned by main emu!
+//    box_free(ctx->stack); // don't free the stack, it's owned by main emu!
 
-    free(ctx->fullpath);
-    free(ctx->box86path);
+    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);
@@ -283,7 +324,7 @@ void FreeBox86Context(box86context_t** context)
     pthread_key_delete(ctx->tlskey);
 
     if(ctx->tlsdata)
-        free(ctx->tlsdata);
+        box_free(ctx->tlsdata);
 
     for(int i=0; i<3; ++i) {
         if(ctx->segtls[i].present) {
@@ -292,25 +333,30 @@ void FreeBox86Context(box86context_t** context)
         }
     }
 
-    pthread_mutex_destroy(&ctx->mutex_once);
-    pthread_mutex_destroy(&ctx->mutex_once2);
-    pthread_mutex_destroy(&ctx->mutex_trace);
-#ifndef DYNAREC
-    pthread_mutex_destroy(&ctx->mutex_lock);
-#else
-    pthread_mutex_destroy(&ctx->mutex_dyndump);
-#endif
-    pthread_mutex_destroy(&ctx->mutex_tls);
-    pthread_mutex_destroy(&ctx->mutex_thread);
-
     free_neededlib(&ctx->neededlibs);
 
     if(ctx->emu_sig)
         FreeX86Emu(&ctx->emu_sig);
+    FreeDefaultVersion(&ctx->globaldefver);
+    FreeDefaultVersion(&ctx->weakdefver);
 
     finiAllHelpers(ctx);
 
-    free(ctx);
+    pthread_mutex_destroy(&ctx->mutex_once);
+    pthread_mutex_destroy(&ctx->mutex_once2);
+    pthread_mutex_destroy(&ctx->mutex_trace);
+#ifdef DYNAREC
+    pthread_mutex_destroy(&ctx->mutex_dyndump);
+#else
+    pthread_mutex_destroy(&ctx->mutex_lock);
+#endif
+    pthread_mutex_destroy(&ctx->mutex_tls);
+    pthread_mutex_destroy(&ctx->mutex_thread);
+    pthread_mutex_destroy(&ctx->mutex_bridge);
+
+    freeCycleLog(ctx);
+
+    box_free(ctx);
 }
 
 int AddElfHeader(box86context_t* ctx, elfheader_t* head) {
@@ -318,7 +364,7 @@ int AddElfHeader(box86context_t* ctx, elfheader_t* head) {
     if(idx==ctx->elfcap) {
         // resize...
         ctx->elfcap += 16;
-        ctx->elfs = (elfheader_t**)realloc(ctx->elfs, sizeof(elfheader_t*) * ctx->elfcap);
+        ctx->elfs = (elfheader_t**)box_realloc(ctx->elfs, sizeof(elfheader_t*) * ctx->elfcap);
     }
     ctx->elfs[idx] = head;
     ctx->elfsize++;
@@ -329,7 +375,7 @@ int AddElfHeader(box86context_t* ctx, elfheader_t* head) {
 int AddTLSPartition(box86context_t* context, int tlssize) {
     int oldsize = context->tlssize;
     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

src/box86version.h

@@ -3,6 +3,6 @@
 
 #define BOX86_MAJOR 0
 #define BOX86_MINOR 2
-#define BOX86_REVISION 2
+#define BOX86_REVISION 8
 
 #endif //__BOX86_VERSION_H_

src/custommem.c

@@ -28,22 +28,33 @@
 #include "khash.h"
 
 #define USE_MMAP
+//#define USE_MMAP_MORE
 
 // init inside dynablocks.c
 KHASH_MAP_INIT_INT(dynablocks, dynablock_t*)
-static dynablocklist_t*    dynmap[DYNAMAP_SIZE];     // 4G of memory mapped by 4K block
-static pthread_mutex_t     mutex_mmap;
-static mmaplist_t          *mmaplist;
-static int                 mmapsize;
+static dynablocklist_t*    dynmap[DYNAMAP_SIZE] = {0};     // 4G of memory mapped by 4K block
+static mmaplist_t          *mmaplist = NULL;
+static int                 mmapsize = 0;
+static int                 mmapcap = 0;
 static kh_dynablocks_t     *dblist_oversized;      // store the list of oversized dynablocks (normal sized are inside mmaplist)
 static uintptr_t           *box86_jumptable[JMPTABL_SIZE];
 static uintptr_t           box86_jmptbl_default[1<<JMPTABL_SHIFT];
+// lock addresses
+KHASH_SET_INIT_INT(lockaddress)
+static kh_lockaddress_t    *lockaddress = NULL;
 #endif
 #define MEMPROT_SHIFT 12
 #define MEMPROT_SIZE (1<<(32-MEMPROT_SHIFT))
 static uint8_t             memprot[MEMPROT_SIZE] = {0};    // protection flags by 4K block
 static int inited = 0;
 
+typedef struct mapmem_s {
+    uintptr_t begin, end;
+    struct mapmem_s *next;
+} mapmem_t;
+
+static mapmem_t *mapmem = NULL;
+
 typedef struct blocklist_s {
     void*               block;
     int                 maxfree;
@@ -58,8 +69,8 @@ static blocklist_t*        p_blocks = NULL;    // actual blocks for custom mallo
 
 typedef union mark_s {
     struct {
-        unsigned int    fill:1;
         unsigned int    size:31;
+        unsigned int    fill:1;
     };
     uint32_t            x32;
 } mark_t;
@@ -68,98 +79,146 @@ typedef struct blockmark_s {
     mark_t  next;
 } blockmark_t;
 
+#define NEXT_BLOCK(b) (blockmark_t*)((uintptr_t)(b) + (b)->next.size + sizeof(blockmark_t))
+#define PREV_BLOCK(b) (blockmark_t*)(((uintptr_t)(b) - (b)->prev.size) - sizeof(blockmark_t))
+#define LAST_BLOCK(b, s) (blockmark_t*)(((uintptr_t)(b)+(s))-sizeof(blockmark_t))
+
+static void printBlock(blockmark_t* b, void* start)
+{
+    printf_log(LOG_INFO, "========== Block is:\n");
+    do {
+        printf_log(LOG_INFO, "%c%p, fill=%d, size=0x%x (prev=%d/0x%x)\n", b==start?'*':' ', b, b->next.fill, b->next.size, b->prev.fill, b->prev.size);
+        b = NEXT_BLOCK(b);
+    } while(b->next.x32);
+    printf_log(LOG_INFO, "===================\n");
+}
 
 // get first subblock free in block. Return NULL if no block, else first subblock free (mark included), filling size
-static void* getFirstBlock(void* block, int maxsize, int* size)
+static void* getFirstBlock(void* block, size_t maxsize, size_t* size, void* start)
 {
     // get start of block
-    blockmark_t *m = (blockmark_t*)block;
+    blockmark_t *m = (blockmark_t*)((start)?start:block);
     while(m->next.x32) {    // while there is a subblock
-        if(!m->next.fill && m->next.size>=maxsize+sizeof(blockmark_t)) {
-            *size = m->next.size-sizeof(blockmark_t);
+        if(!m->next.fill && m->next.size>=maxsize) {
+            *size = m->next.size;
             return m;
         }
-        m = (blockmark_t*)((uintptr_t)m + m->next.size);
+        m = NEXT_BLOCK(m);
     }
 
     return NULL;
 }
 
-static int getMaxFreeBlock(void* block, size_t block_size)
+static size_t getMaxFreeBlock(void* block, size_t block_size, void* start)
 {
     // get start of block
-    blockmark_t *m = (blockmark_t*)((uintptr_t)block+block_size-sizeof(blockmark_t)); // start with the end
-    int maxsize = 0;
-    while(m->prev.x32) {    // while there is a subblock
-        if(!m->prev.fill && m->prev.size>maxsize) {
-            maxsize = m->prev.size;
-            if((uintptr_t)block+maxsize>(uintptr_t)m)
-                return (maxsize>=sizeof(blockmark_t))?(maxsize-sizeof(blockmark_t)):0; // no block large enough left...
+    if(start) {
+        blockmark_t *m = (blockmark_t*)start;
+        int maxsize = 0;
+        while(m->next.x32) {    // while there is a subblock
+            if(!m->next.fill && m->next.size>maxsize) {
+                maxsize = m->next.size;
+            }
+            m = NEXT_BLOCK(m);
         }
-        m = (blockmark_t*)((uintptr_t)m - m->prev.size);
+        return (maxsize>=sizeof(blockmark_t))?maxsize:0;
+    } else {
+        blockmark_t *m = LAST_BLOCK(block, block_size); // start with the end
+        int maxsize = 0;
+        while(m->prev.x32) {    // while there is a subblock
+            if(!m->prev.fill && m->prev.size>maxsize) {
+                maxsize = m->prev.size;
+                if((uintptr_t)block+maxsize>(uintptr_t)m)
+                    return (maxsize>=sizeof(blockmark_t))?maxsize:0; // no block large enough left...
+            }
+            m = PREV_BLOCK(m);
+        }
+        return (maxsize>=sizeof(blockmark_t))?maxsize:0;
     }
-    return (maxsize>=sizeof(blockmark_t))?(maxsize-sizeof(blockmark_t)):0;
 }
 
-static void* allocBlock(void* block, void *sub, int size)
+static void* allocBlock(void* block, void *sub, size_t size, void** pstart)
 {
+    (void)block;
+
     blockmark_t *s = (blockmark_t*)sub;
-    blockmark_t *n = (blockmark_t*)((uintptr_t)s + s->next.size);
+    blockmark_t *n = NEXT_BLOCK(s);
 
     s->next.fill = 1;
-    s->next.size = size+sizeof(blockmark_t);
-    blockmark_t *m = (blockmark_t*)((uintptr_t)s + s->next.size);   // this is new n
+    // check if a new mark is worth it
+    if(s->next.size>size+2*sizeof(blockmark_t))
+        s->next.size = size;
+    blockmark_t *m = NEXT_BLOCK(s);   // this is new n
     m->prev.fill = 1;
-    m->prev.size = s->next.size;
     if(n!=m) {
         // new mark
-        m->prev.fill = 1;
         m->prev.size = s->next.size;
         m->next.fill = 0;
-        m->next.size = (uintptr_t)n - (uintptr_t)m;
+        m->next.size = ((uintptr_t)n - (uintptr_t)m) - sizeof(blockmark_t);
         n->prev.fill = 0;
         n->prev.size = m->next.size;
     }
 
+    if(pstart && sub==*pstart) {
+        // get the next free block
+        m = (blockmark_t*)*pstart;
+        while(m->next.fill)
+            m = NEXT_BLOCK(m);
+        *pstart = (void*)m;
+    }
     return (void*)((uintptr_t)sub + sizeof(blockmark_t));
 }
-static void freeBlock(void *block, void* sub)
+static size_t freeBlock(void *block, void* sub, void** pstart)
 {
     blockmark_t *m = (blockmark_t*)block;
     blockmark_t *s = (blockmark_t*)sub;
-    blockmark_t *n = (blockmark_t*)((uintptr_t)s + s->next.size);
+    blockmark_t *n = NEXT_BLOCK(s);
     if(block!=sub)
-        m = (blockmark_t*)((uintptr_t)s - s->prev.size);
+        m = PREV_BLOCK(s);
     s->next.fill = 0;
     n->prev.fill = 0;
     // check if merge with previous
     if (s->prev.x32 && !s->prev.fill) {
         // remove s...
-        m->next.size += s->next.size;
+        m->next.size += s->next.size + sizeof(blockmark_t);
         n->prev.size = m->next.size;
         s = m;
     }
     // check if merge with next
     if(n->next.x32 && !n->next.fill) {
-        blockmark_t *n2 = (blockmark_t*)((uintptr_t)n + n->next.size);
+        blockmark_t *n2 = NEXT_BLOCK(n);
         //remove n
-        s->next.size += n->next.size;
+        s->next.size += n->next.size + sizeof(blockmark_t);
         n2->prev.size = s->next.size;
     }
+    if(pstart && (uintptr_t)*pstart>(uintptr_t)sub) {
+        *pstart = (void*)s;
+    }
+    // return free size at current block (might be bigger)
+    return s->next.size;
 }
 // return 1 if block has been expanded to new size, 0 if not
-static int expandBlock(void* block, void* sub, int newsize)
+static int expandBlock(void* block, void* sub, size_t newsize)
 {
+    (void)block;
+
     newsize = (newsize+3)&~3;
     blockmark_t *s = (blockmark_t*)sub;
-    blockmark_t *n = (blockmark_t*)((uintptr_t)s + s->next.size);
+    blockmark_t *n = NEXT_BLOCK(s);
+    if(s->next.size>=newsize)
+        // big enough, no shrinking...
+        return 1;
     if(s->next.fill)
         return 0;   // next block is filled
-    if(s->next.size + n->next.size < newsize)
+    // unsigned bitfield of this length gets "promoted" to *signed* int...
+    if((size_t)(s->next.size + n->next.size + sizeof(blockmark_t)) < newsize)
         return 0;   // free space too short
     // ok, doing the alloc!
-    s->next.size = newsize+sizeof(blockmark_t);
-    blockmark_t *m = (blockmark_t*)((uintptr_t)s + s->next.size);   // this is new n
+    if((s->next.size+n->next.size+sizeof(blockmark_t))-newsize<2*sizeof(blockmark_t))
+        s->next.size += n->next.size+sizeof(blockmark_t);
+    else
+        s->next.size = newsize+sizeof(blockmark_t);
+    blockmark_t *m = NEXT_BLOCK(s);   // this is new n
     m->prev.fill = 1;
     m->prev.size = s->next.size;
     if(n!=m) {
@@ -174,7 +233,7 @@ static int expandBlock(void* block, void* sub, int newsize)
     return 1;
 }
 // return size of block
-static int sizeBlock(void* sub)
+static size_t sizeBlock(void* sub)
 {
     blockmark_t *s = (blockmark_t*)sub;
     return s->next.size;
@@ -187,12 +246,12 @@ void* customMalloc(size_t size)
     pthread_mutex_lock(&mutex_blocks);
     for(int i=0; i<n_blocks; ++i) {
         if(p_blocks[i].maxfree>=size) {
-            int rsize = 0;
-            sub = getFirstBlock(p_blocks[i].block, size, &rsize);
+            size_t rsize = 0;
+            sub = getFirstBlock(p_blocks[i].block, size, &rsize, NULL);
             if(sub) {
-                void* ret = allocBlock(p_blocks[i].block, sub, size);
+                void* ret = allocBlock(p_blocks[i].block, sub, size, NULL);
                 if(rsize==p_blocks[i].maxfree)
-                    p_blocks[i].maxfree = getMaxFreeBlock(p_blocks[i].block, p_blocks[i].size);
+                    p_blocks[i].maxfree = getMaxFreeBlock(p_blocks[i].block, p_blocks[i].size, NULL);
                 pthread_mutex_unlock(&mutex_blocks);
                 return ret;
             }
@@ -200,15 +259,16 @@ void* customMalloc(size_t size)
     }
     // add a new block
     int i = n_blocks++;
-    p_blocks = (blocklist_t*)realloc(p_blocks, n_blocks*sizeof(blocklist_t));
+    p_blocks = (blocklist_t*)box_realloc(p_blocks, n_blocks*sizeof(blocklist_t));
     size_t allocsize = MMAPSIZE;
     if(size+2*sizeof(blockmark_t)>allocsize)
         allocsize = size+2*sizeof(blockmark_t);
-    #ifdef USE_MMAP
+    #ifdef USE_MMAP_MORE
     void* p = mmap(NULL, allocsize, PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
     memset(p, 0, allocsize);
+    setProtection((uintptr_t)p, allocsize, PROT_READ|PROT_WRITE);
     #else
-    void* p = calloc(1, allocsize);
+    void* p = box_calloc(1, allocsize);
     #endif
     p_blocks[i].block = p;
     p_blocks[i].size = allocsize;
@@ -216,14 +276,14 @@ void* customMalloc(size_t size)
     blockmark_t* m = (blockmark_t*)p;
     m->prev.x32 = 0;
     m->next.fill = 0;
-    m->next.size = allocsize-sizeof(blockmark_t);
-    m = (blockmark_t*)(p+allocsize-sizeof(blockmark_t));
-    m->next.x32 = 0;
-    m->prev.fill = 0;
-    m->prev.size = allocsize-sizeof(blockmark_t);
+    m->next.size = allocsize-2*sizeof(blockmark_t);
+    blockmark_t* n = NEXT_BLOCK(m);
+    n->next.x32 = 0;
+    n->prev.fill = 0;
+    n->prev.size = m->next.size;
     // alloc 1st block
-    void* ret  = allocBlock(p_blocks[i].block, p, size);
-    p_blocks[i].maxfree = getMaxFreeBlock(p_blocks[i].block, p_blocks[i].size);
+    void* ret  = allocBlock(p_blocks[i].block, p, size, NULL);
+    p_blocks[i].maxfree = getMaxFreeBlock(p_blocks[i].block, p_blocks[i].size, NULL);
     pthread_mutex_unlock(&mutex_blocks);
     return ret;
 }
@@ -245,21 +305,20 @@ void* customRealloc(void* p, size_t size)
          && (addr<((uintptr_t)p_blocks[i].block+p_blocks[i].size))) {
             void* sub = (void*)(addr-sizeof(blockmark_t));
             if(expandBlock(p_blocks[i].block, sub, size)) {
-                p_blocks[i].maxfree = getMaxFreeBlock(p_blocks[i].block, p_blocks[i].size);
+                p_blocks[i].maxfree = getMaxFreeBlock(p_blocks[i].block, p_blocks[i].size, NULL);
                 pthread_mutex_unlock(&mutex_blocks);
                 return p;
             }
-                pthread_mutex_unlock(&mutex_blocks);
-                void* newp = customMalloc(size);
-                memcpy(newp, p, sizeBlock(sub));
-                customFree(p);
-                return newp;
-            
+            pthread_mutex_unlock(&mutex_blocks);
+            void* newp = customMalloc(size);
+            memcpy(newp, p, sizeBlock(sub));
+            customFree(p);
+            return newp;
         }
     }
     pthread_mutex_unlock(&mutex_blocks);
     if(n_blocks)
-        dynarec_log(LOG_NONE, "Warning, block %p not found in p_blocks for Realloc, Malloc'ng again without free\n", (void*)addr);
+        dynarec_log(LOG_NONE, "Warning, block %p not found in p_blocks for realloc, malloc'ing again without free\n", (void*)addr);
     return customMalloc(size);
 }
 void customFree(void* p)
@@ -272,8 +331,8 @@ void customFree(void* p)
         if ((addr>(uintptr_t)p_blocks[i].block) 
          && (addr<((uintptr_t)p_blocks[i].block+p_blocks[i].size))) {
             void* sub = (void*)(addr-sizeof(blockmark_t));
-            freeBlock(p_blocks[i].block, sub);
-            p_blocks[i].maxfree = getMaxFreeBlock(p_blocks[i].block, p_blocks[i].size);
+            size_t newfree = freeBlock(p_blocks[i].block, sub, NULL);
+            if(p_blocks[i].maxfree < newfree) p_blocks[i].maxfree = newfree;
             pthread_mutex_unlock(&mutex_blocks);
             return;
         }
@@ -290,6 +349,8 @@ typedef struct mmaplist_s {
     size_t              size;
     kh_dynablocks_t*    dblist;
     uint8_t*            helper;
+    void*               first;  // first free block, to speed up things
+    int                 locked; // don't try to add stuff on locked block
 } mmaplist_t;
 
 uintptr_t FindFreeDynarecMap(dynablock_t* db, int size)
@@ -297,13 +358,15 @@ uintptr_t FindFreeDynarecMap(dynablock_t* db, int size)
     // look for free space
     void* sub = NULL;
     for(int i=0; i<mmapsize; ++i) {
-        if(mmaplist[i].maxfree>=size) {
-            int rsize = 0;
-            sub = getFirstBlock(mmaplist[i].block, size, &rsize);
+        if(mmaplist[i].maxfree>=size+sizeof(blockmark_t) && !mmaplist[i].locked) {
+            mmaplist[i].locked = 1;
+            size_t rsize = 0;
+            sub = getFirstBlock(mmaplist[i].block, size, &rsize, mmaplist[i].first);
             if(sub) {
-                uintptr_t ret = (uintptr_t)allocBlock(mmaplist[i].block, sub, size);
-                if(rsize==mmaplist[i].maxfree)
-                    mmaplist[i].maxfree = getMaxFreeBlock(mmaplist[i].block, mmaplist[i].size);
+                uintptr_t ret = (uintptr_t)allocBlock(mmaplist[i].block, sub, size, &mmaplist[i].first);
+                if(rsize==mmaplist[i].maxfree) {
+                    mmaplist[i].maxfree = getMaxFreeBlock(mmaplist[i].block, mmaplist[i].size, mmaplist[i].first);
+                }
                 kh_dynablocks_t *blocks = mmaplist[i].dblist;
                 if(!blocks) {
                     blocks = mmaplist[i].dblist = kh_init(dynablocks);
@@ -313,9 +376,17 @@ uintptr_t FindFreeDynarecMap(dynablock_t* db, int size)
                 int r;
                 k = kh_put(dynablocks, blocks, (uintptr_t)ret, &r);
                 kh_value(blocks, k) = db;
-                for(int j=0; j<size; ++j)
-                    mmaplist[i].helper[(uintptr_t)ret-(uintptr_t)mmaplist[i].block+j] = (j<256)?j:255;
+                int size255=(size<256)?size:255;
+                for(size_t j=0; j<size255; ++j)
+                    mmaplist[i].helper[(uintptr_t)ret-(uintptr_t)mmaplist[i].block+j] = j;
+                if(size!=size255)
+                    memset(&mmaplist[i].helper[(uintptr_t)ret-(uintptr_t)mmaplist[i].block+256], -1, size-255);
+                mmaplist[i].locked = 0;
                 return ret;
+            } else {
+                printf_log(LOG_INFO, "BOX86: Warning, sub not found, corrupted mmaplist[%d] info?\n", i);
+                if(box86_log >= LOG_DEBUG)
+                    printBlock(mmaplist[i].block, mmaplist[i].first);
             }
         }
     }
@@ -325,8 +396,11 @@ uintptr_t FindFreeDynarecMap(dynablock_t* db, int size)
 uintptr_t AddNewDynarecMap(dynablock_t* db, int size)
 {
     int i = mmapsize++;    // yeah, useful post incrementation
-    dynarec_log(LOG_DEBUG, "Ask for DynaRec Block Alloc #%d\n", mmapsize);
-    mmaplist = (mmaplist_t*)realloc(mmaplist, mmapsize*sizeof(mmaplist_t));
+    dynarec_log(LOG_DEBUG, "Ask for DynaRec Block Alloc #%zu/%zu\n", mmapsize, mmapcap);
+    if(mmapsize>mmapcap) {
+        mmapcap += 32;
+        mmaplist = (mmaplist_t*)box_realloc(mmaplist, mmapcap*sizeof(mmaplist_t));
+    }
     #ifndef USE_MMAP
     void *p = NULL;
     if(posix_memalign(&p, box86_pagesize, MMAPSIZE)) {
@@ -345,21 +419,23 @@ uintptr_t AddNewDynarecMap(dynablock_t* db, int size)
     #endif
     setProtection((uintptr_t)p, MMAPSIZE, PROT_READ | PROT_WRITE | PROT_EXEC);
 
+    mmaplist[i].locked = 1;
     mmaplist[i].block = p;
     mmaplist[i].size = MMAPSIZE;
-    mmaplist[i].helper = (uint8_t*)calloc(1, MMAPSIZE);
+    mmaplist[i].helper = (uint8_t*)box_calloc(1, MMAPSIZE);
+    mmaplist[i].first = p;
     // setup marks
     blockmark_t* m = (blockmark_t*)p;
     m->prev.x32 = 0;
     m->next.fill = 0;
-    m->next.size = MMAPSIZE-sizeof(blockmark_t);
-    m = (blockmark_t*)(p+MMAPSIZE-sizeof(blockmark_t));
-    m->next.x32 = 0;
-    m->prev.fill = 0;
-    m->prev.size = MMAPSIZE-sizeof(blockmark_t);
+    m->next.size = MMAPSIZE-2*sizeof(blockmark_t);
+    blockmark_t* n = NEXT_BLOCK(m);
+    n->next.x32 = 0;
+    n->prev.fill = 0;
+    n->prev.size = m->next.size;
     // alloc 1st block
-    uintptr_t sub  = (uintptr_t)allocBlock(mmaplist[i].block, p, size);
-    mmaplist[i].maxfree = getMaxFreeBlock(mmaplist[i].block, mmaplist[i].size);
+    uintptr_t sub  = (uintptr_t)allocBlock(mmaplist[i].block, p, size, &mmaplist[i].first);
+    mmaplist[i].maxfree = getMaxFreeBlock(mmaplist[i].block, mmaplist[i].size, mmaplist[i].first);
     kh_dynablocks_t *blocks = mmaplist[i].dblist = kh_init(dynablocks);
     kh_resize(dynablocks, blocks, 64);
     khint_t k;
@@ -368,6 +444,7 @@ uintptr_t AddNewDynarecMap(dynablock_t* db, int size)
     kh_value(blocks, k) = db;
     for(int j=0; j<size; ++j)
         mmaplist[i].helper[(uintptr_t)sub-(uintptr_t)mmaplist[i].block + j] = (j<256)?j:255;
+    mmaplist[i].locked = 0;
     return sub;
 }
 
@@ -379,15 +456,18 @@ void ActuallyFreeDynarecMap(dynablock_t* db, uintptr_t addr, int size)
         if ((addr>(uintptr_t)mmaplist[i].block) 
          && (addr<((uintptr_t)mmaplist[i].block+mmaplist[i].size))) {
             void* sub = (void*)(addr-sizeof(blockmark_t));
-            freeBlock(mmaplist[i].block, sub);
-            mmaplist[i].maxfree = getMaxFreeBlock(mmaplist[i].block, mmaplist[i].size);
+            size_t newfree = freeBlock(mmaplist[i].block, sub, &mmaplist[i].first);
+            if(mmaplist[i].maxfree < newfree) mmaplist[i].maxfree = newfree;
             kh_dynablocks_t *blocks = mmaplist[i].dblist;
             if(blocks) {
                 khint_t k = kh_get(dynablocks, blocks, (uintptr_t)sub);
                 if(k!=kh_end(blocks))
                     kh_del(dynablocks, blocks, k);
-                for(int j=0; j<size; ++j)
-                    mmaplist[i].helper[(uintptr_t)sub-(uintptr_t)mmaplist[i].block+j] = 0;
+                memset(&mmaplist[i].helper[(uintptr_t)sub-(uintptr_t)mmaplist[i].block], 0, size);
+            }
+            if(mmaplist[i].locked) {
+                printf_log(LOG_INFO, "BOX86: Warning, Free a chunk in a locked mmaplist[%d]\n", i);
+                ++mmaplist[i].locked;
             }
             return;
         }
@@ -424,7 +504,6 @@ uintptr_t AllocDynarecMap(dynablock_t* db, int size)
         return 0;
     if(size>MMAPSIZE-2*sizeof(blockmark_t)) {
         #ifndef USE_MMAP
-        pthread_mutex_lock(&mutex_mmap);
         void *p = NULL;
         if(posix_memalign(&p, box86_pagesize, size)) {
             dynarec_log(LOG_INFO, "Cannot create dynamic map of %d bytes\n", size);
@@ -448,18 +527,13 @@ uintptr_t AllocDynarecMap(dynablock_t* db, int size)
         int ret;
         k = kh_put(dynablocks, blocks, (uintptr_t)p, &ret);
         kh_value(blocks, k) = db;
-        pthread_mutex_unlock(&mutex_mmap);
         return (uintptr_t)p;
     }
     
-    pthread_mutex_lock(&mutex_mmap);
-
     uintptr_t ret = FindFreeDynarecMap(db, size);
     if(!ret)
         ret = AddNewDynarecMap(db, size);
 
-    pthread_mutex_unlock(&mutex_mmap);
-
     return ret;
 }
 
@@ -468,9 +542,8 @@ void FreeDynarecMap(dynablock_t* db, uintptr_t addr, uint32_t size)
     if(!addr || !size)
         return;
     if(size>MMAPSIZE-2*sizeof(blockmark_t)) {
-        pthread_mutex_lock(&mutex_mmap);
         #ifndef USE_MMAP
-        free((void*)addr);
+        box_free((void*)addr);
         #else
         munmap((void*)addr, size);
         #endif
@@ -480,12 +553,9 @@ void FreeDynarecMap(dynablock_t* db, uintptr_t addr, uint32_t size)
             if(k!=kh_end(blocks))
                 kh_del(dynablocks, blocks, k);
         }
-        pthread_mutex_unlock(&mutex_mmap);
         return;
     }
-    pthread_mutex_lock(&mutex_mmap);
     ActuallyFreeDynarecMap(db, addr, size);
-    pthread_mutex_unlock(&mutex_mmap);
 }
 
 dynablocklist_t* getDB(uintptr_t idx)
@@ -533,11 +603,11 @@ void addJumpTableIfDefault(void* addr, void* jmp)
 {
     const uintptr_t idx = ((uintptr_t)addr>>JMPTABL_SHIFT);
     if(box86_jumptable[idx] == box86_jmptbl_default) {
-        uintptr_t* tbl = (uintptr_t*)malloc((1<<JMPTABL_SHIFT)*sizeof(uintptr_t));
+        uintptr_t* tbl = (uintptr_t*)box_malloc((1<<JMPTABL_SHIFT)*sizeof(uintptr_t));
         for(int i=0; i<(1<<JMPTABL_SHIFT); ++i)
             tbl[i] = (uintptr_t)arm_next;
         if(arm_lock_storeifref(&box86_jumptable[idx], tbl, box86_jmptbl_default)!=tbl)
-            free(tbl);
+            box_free(tbl);
     }
     const uintptr_t off = (uintptr_t)addr&((1<<JMPTABL_SHIFT)-1);
     arm_lock_storeifref(&box86_jumptable[idx][off], jmp, arm_next);
@@ -569,11 +639,11 @@ uintptr_t getJumpTableAddress(uintptr_t addr)
 {
     const uintptr_t idx = ((uintptr_t)addr>>JMPTABL_SHIFT);
     if(box86_jumptable[idx] == box86_jmptbl_default) {
-        uintptr_t* tbl = (uintptr_t*)malloc((1<<JMPTABL_SHIFT)*sizeof(uintptr_t));
+        uintptr_t* tbl = (uintptr_t*)box_malloc((1<<JMPTABL_SHIFT)*sizeof(uintptr_t));
         for(int i=0; i<(1<<JMPTABL_SHIFT); ++i)
             tbl[i] = (uintptr_t)arm_next;
         if(arm_lock_storeifref(&box86_jumptable[idx], tbl, box86_jmptbl_default)!=tbl)
-            free(tbl);
+            box_free(tbl);
     }
     const uintptr_t off = (uintptr_t)addr&((1<<JMPTABL_SHIFT)-1);
     return (uintptr_t)&box86_jumptable[idx][off];
@@ -588,47 +658,166 @@ void protectDB(uintptr_t addr, uintptr_t size)
     uintptr_t end = ((addr+size-1)>>MEMPROT_SHIFT);
     for (uintptr_t i=idx; i<=end; ++i) {
         uint32_t prot = memprot[i];
+        uint32_t dyn = prot&PROT_CUSTOM;
+        prot&=~PROT_CUSTOM;
         if(!prot)
-            prot = PROT_READ | PROT_WRITE;    // comes from malloc & co, so should not be able to execute
-        memprot[i] = prot|PROT_DYNAREC;
-        if(!(prot&PROT_DYNAREC))
-            mprotect((void*)(i<<MEMPROT_SHIFT), 1<<MEMPROT_SHIFT, prot&~PROT_WRITE);
+            prot = PROT_READ | PROT_WRITE | PROT_EXEC;      // comes from malloc & co, so should not be able to execute
+        if(prot&PROT_WRITE) {
+            memprot[i] = prot|PROT_DYNAREC;
+            if(!dyn) mprotect((void*)(i<<MEMPROT_SHIFT), 1<<MEMPROT_SHIFT, prot&~PROT_WRITE);
+        } else
+            memprot[i] = prot|PROT_DYNAREC_R;
     }
 }
 
 // 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)
+void unprotectDB(uintptr_t addr, uintptr_t size, int mark)
 {
-    dynarec_log(LOG_DEBUG, "unprotectDB %p -> %p\n", (void*)addr, (void*)(addr+size-1));
+    dynarec_log(LOG_DEBUG, "unprotectDB %p -> %p (mark=%d)\n", (void*)addr, (void*)(addr+size-1), mark);
     uintptr_t idx = (addr>>MEMPROT_SHIFT);
     uintptr_t end = ((addr+size-1)>>MEMPROT_SHIFT);
     for (uintptr_t i=idx; i<=end; ++i) {
         uint32_t prot = memprot[i];
-        memprot[i] = prot&~PROT_DYNAREC;
         if(prot&PROT_DYNAREC) {
+            memprot[i] = prot&~PROT_DYNAREC;
             mprotect((void*)(i<<MEMPROT_SHIFT), 1<<MEMPROT_SHIFT, prot&~PROT_DYNAREC);
-            cleanDBFromAddressRange((i<<MEMPROT_SHIFT), 1<<MEMPROT_SHIFT, 0);
-        }
+            if(mark)
+                cleanDBFromAddressRange((i<<MEMPROT_SHIFT), 1<<MEMPROT_SHIFT, 0);
+        } else if(prot&PROT_DYNAREC_R)
+            memprot[i] = prot&~(PROT_CUSTOM);
     }
 }
 
 #endif
 
+void printMapMem()
+{
+    mapmem_t* m = mapmem;
+    while(m) {
+        printf_log(LOG_NONE, " %p-%p\n", (void*)m->begin, (void*)m->end);
+        m = m->next;
+    }
+}
+
+void addMapMem(uintptr_t begin, uintptr_t end)
+{
+    begin &=~0xfff;
+    end = (end&~0xfff)+0xfff; // full page
+    // sanitize values
+    if(end<0x10000) return;
+    if(!begin) begin = 0x10000;
+    // find attach point (cannot be the 1st one by construction)
+    mapmem_t* m = mapmem;
+    while(m->next && begin>m->next->begin) {
+        m = m->next;
+    }
+    // attach at the end of m
+    mapmem_t* newm;
+    if(m->end>=begin-1) {
+        if(end<=m->end)
+            return; // zone completly inside current block, nothing to do
+        m->end = end;   // enlarge block
+        newm = m;
+    } else {
+    // create a new block
+        newm = (mapmem_t*)box_calloc(1, sizeof(mapmem_t));
+        newm->next = m->next;
+        newm->begin = begin;
+        newm->end = end;
+        m->next = newm;
+    }
+    while(newm && newm->next && (newm->next->begin-1)<=newm->end) {
+        // fuse with next
+        if(newm->next->end>newm->end)
+            newm->end = newm->next->end;
+        mapmem_t* tmp = newm->next;
+        newm->next = tmp->next;
+        box_free(tmp);
+    }
+    // all done!
+}
+void removeMapMem(uintptr_t begin, uintptr_t end)
+{
+    begin &=~0xfff;
+    end = (end&~0xfff)+0xfff; // full page
+    // sanitize values
+    if(end<0x10000) return;
+    if(!begin) begin = 0x10000;
+    mapmem_t* m = mapmem, *prev = NULL;
+    while(m) {
+        // check if block is beyond the zone to free
+        if(m->begin > end)
+            return;
+        // check if the block is completly inside the zone to free
+        if(m->begin>=begin && m->end<=end) {
+            // just free the block
+            mapmem_t *tmp = m;
+            if(prev) {
+                prev->next = m->next;
+                m = prev;
+            } else {
+                mapmem = m->next; // change attach, but should never happens
+                m = mapmem;
+                prev = NULL;
+            }
+            box_free(tmp);
+        } else if(begin>m->begin && end<m->end) { // the zone is totaly inside the block => split it!
+            mapmem_t* newm = (mapmem_t*)box_calloc(1, sizeof(mapmem_t));    // create a new "next"
+            newm->end = m->end;
+            m->end = begin - 1;
+            newm->begin = end + 1;
+            newm->next = m->next;
+            m->next = newm;
+            // nothing more to free
+            return;
+        } else if(begin>m->begin && begin<m->end) { // free the tail of the block
+            m->end = begin - 1;
+        } else if(end>m->begin && end<m->end) { // free the head of the block
+            m->begin = end + 1;
+        }
+        prev = m;
+        m = m->next;
+    }
+}
+
 void updateProtection(uintptr_t addr, uintptr_t size, uint32_t prot)
 {
+    //dynarec_log(LOG_DEBUG, "updateProtection %p -> %p to 0x%02x\n", (void*)addr, (void*)(addr+size-1), prot);
+    addMapMem(addr, addr+size-1);
     const uintptr_t idx = (addr>>MEMPROT_SHIFT);
     const uintptr_t end = ((addr+size-1)>>MEMPROT_SHIFT);
     for (uintptr_t i=idx; i<=end; ++i) {
-        uint32_t dyn=(memprot[i]&PROT_DYNAREC);
-        if(dyn && (prot&PROT_WRITE))    // need to remove the write protection from this block
+        uint32_t dyn=(memprot[i]&(PROT_DYNAREC|PROT_DYNAREC_R));
+        if(dyn && (prot&PROT_WRITE)) {   // need to remove the write protection from this block
+            dyn = PROT_DYNAREC;
             mprotect((void*)(i<<MEMPROT_SHIFT), 1<<MEMPROT_SHIFT, prot&~PROT_WRITE);
+        } else if(dyn && !(prot&PROT_WRITE)) {
+            dyn = PROT_DYNAREC_R;
+        }
+
         memprot[i] = prot|dyn;
     }
 }
 
+void forceProtection(uintptr_t addr, uintptr_t size, uint32_t prot)
+{
+    //dynarec_log(LOG_DEBUG, "forceProtection %p -> %p to 0x%02x\n", (void*)addr, (void*)(addr+size-1), prot);
+    addMapMem(addr, addr+size-1);
+    const uintptr_t idx = (addr>>MEMPROT_SHIFT);
+    const uintptr_t end = ((addr+size-1)>>MEMPROT_SHIFT);
+    for (uintptr_t i=idx; i<=end; ++i) {
+        mprotect((void*)(i<<MEMPROT_SHIFT), 1<<MEMPROT_SHIFT, prot&~PROT_CUSTOM);
+        memprot[i] = prot;
+    }
+}
+
 void setProtection(uintptr_t addr, uintptr_t size, uint32_t prot)
 {
+    //dynarec_log(LOG_DEBUG, "setProtection %p -> %p to 0x%02x\n", (void*)addr, (void*)(addr+size-1), prot);
+    if(!mapmem)
+        return;
+    addMapMem(addr, addr+size-1);
     const uintptr_t idx = (addr>>MEMPROT_SHIFT);
     const uintptr_t end = ((addr+size-1)>>MEMPROT_SHIFT);
     for (uintptr_t i=idx; i<=end; ++i) {
@@ -636,6 +825,16 @@ void setProtection(uintptr_t addr, uintptr_t size, uint32_t prot)
     }
 }
 
+void freeProtection(uintptr_t addr, uintptr_t size)
+{
+    removeMapMem(addr, addr+size-1);
+    const uintptr_t idx = (addr>>MEMPROT_SHIFT);
+    const uintptr_t end = ((addr+size-1)>>MEMPROT_SHIFT);
+    for (uintptr_t i=idx; i<=end; ++i) {
+        memprot[i] = 0;
+    }
+}
+
 uint32_t getProtection(uintptr_t addr)
 {
     const uintptr_t idx = (addr>>MEMPROT_SHIFT);
@@ -643,6 +842,82 @@ uint32_t getProtection(uintptr_t addr)
     return ret;
 }
 
+void allocProtection(uintptr_t addr, uintptr_t size, uint32_t prot)
+{
+    addMapMem(addr, addr+size-1);
+    const uintptr_t idx = (addr>>MEMPROT_SHIFT);
+    const uintptr_t end = ((addr+size-1)>>MEMPROT_SHIFT);
+    for (uintptr_t i=idx; i<=end; ++i) {
+        if(!memprot[i])
+            memprot[i] = prot;
+    }
+}
+
+void loadProtectionFromMap()
+{
+    if(box86_mapclean)
+        return;
+    char buf[500];
+    FILE *f = fopen("/proc/self/maps", "r");
+    uintptr_t current = 0x0;
+    if(box86_log>=LOG_DEBUG || box86_dynarec_log>=LOG_DEBUG) {printf_log(LOG_NONE, "Refresh mmap allocated block start =============\n");}
+    if(!f)
+        return;
+    while(!feof(f)) {
+        char* ret = fgets(buf, sizeof(buf), f);
+        (void)ret;
+        char r, w, x;
+        uintptr_t s, e;
+        if(box86_log>=LOG_DEBUG || box86_dynarec_log>=LOG_DEBUG) {printf_log(LOG_NONE, "\t%s", buf);}
+        if(sscanf(buf, "%lx-%lx %c%c%c", &s, &e, &r, &w, &x)==5) {
+            if(current<s) {
+                removeMapMem(current, s-1);
+                current = e;
+            }
+            int prot = ((r=='r')?PROT_READ:0)|((w=='w')?PROT_WRITE:0)|((x=='x')?PROT_EXEC:0);
+            allocProtection(s, e-s, prot);
+        }
+    }
+    fclose(f);
+    if(box86_log>=LOG_DEBUG || box86_dynarec_log>=LOG_DEBUG) {
+        printf_log(LOG_NONE, "Refresh mmap allocated block done =============\n");
+        printMapMem();
+    }
+    box86_mapclean = 1;
+}
+
+#define LOWEST (void*)0x10000
+#define MEDIAN (void*)0x40000000
+static void* findBlockHinted(void* hint, size_t size)
+{
+    mapmem_t* m = mapmem;
+    uintptr_t h = (uintptr_t)hint;
+    while(m) {
+        // granularity 0x10000
+        uintptr_t addr = (m->end+1+0xffff)&~0xffff;
+        uintptr_t end = (m->next)?(m->next->begin-1):0xffffffff;
+        // check hint and available size
+        if(addr<=h && end>=h && end-h+1>=size)
+            return hint;
+        if(addr>=h && end-addr+1>=size)
+            return (void*)addr;
+        if(end>=0xc0000000 && h<0xc0000000)
+            return NULL;
+        m = m->next;
+    }
+    return NULL;
+}
+void* findBlockNearHint(void* hint, size_t size)
+{   void* ret = findBlockHinted(hint, size);
+    return ret?ret:hint;
+}
+void* find32bitBlock(size_t size)
+{
+    void* ret = findBlockHinted(MEDIAN, size);
+    if(!ret) ret = findBlockHinted(LOWEST, size);
+    return ret;
+}
+
 int unlockCustommemMutex()
 {
     int ret = 0;
@@ -653,9 +928,6 @@ int unlockCustommemMutex()
             ret|=(1<<B);                \
         }
     GO(mutex_blocks, 0)
-    #ifdef DYNAREC
-    GO(mutex_mmap, 1)
-    #endif
     #undef GO
     return ret;
 }
@@ -667,9 +939,6 @@ void relockCustommemMutex(int locks)
             pthread_mutex_lock(&A);     \
 
     GO(mutex_blocks, 0)
-    #ifdef DYNAREC
-    GO(mutex_mmap, 1)
-    #endif
     #undef GO
 }
 
@@ -679,9 +948,6 @@ static void init_mutexes(void)
     pthread_mutexattr_init(&attr);
     pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_ERRORCHECK);
     pthread_mutex_init(&mutex_blocks, &attr);
-#ifdef DYNAREC
-    pthread_mutex_init(&mutex_mmap, &attr);
-#endif
 
     pthread_mutexattr_destroy(&attr);
 }
@@ -707,8 +973,14 @@ void init_custommem_helper(box86context_t* ctx)
 #else
 #error Unsupported architecture!
 #endif
+    lockaddress = kh_init(lockaddress);
 #endif
     pthread_atfork(NULL, NULL, atfork_child_custommem);
+    // init mapmem list
+    mapmem = (mapmem_t*)box_calloc(1, sizeof(mapmem_t));
+    mapmem->begin = 0x0;
+    mapmem->end = (uintptr_t)LOWEST - 1;
+    loadProtectionFromMap();
 }
 
 void fini_custommem_helper(box86context_t *ctx)
@@ -717,58 +989,72 @@ void fini_custommem_helper(box86context_t *ctx)
         return;
     inited = 0;
 #ifdef DYNAREC
-    dynarec_log(LOG_DEBUG, "Free global Dynarecblocks\n");
-    for (int i=0; i<mmapsize; ++i) {
-        if(mmaplist[i].block)
-            #ifdef USE_MMAP
-            munmap(mmaplist[i].block, mmaplist[i].size);
-            #else
-            free(mmaplist[i].block);
-            #endif
-        if(mmaplist[i].dblist) {
-            kh_destroy(dynablocks, mmaplist[i].dblist);
-            mmaplist[i].dblist = NULL;
+    if(box86_dynarec) {
+        dynarec_log(LOG_DEBUG, "Free global Dynarecblocks\n");
+        for (int i=0; i<mmapsize; ++i) {
+            if(mmaplist[i].block)
+                #ifdef USE_MMAP
+                munmap(mmaplist[i].block, mmaplist[i].size);
+                #else
+                box_free(mmaplist[i].block);
+                #endif
+            if(mmaplist[i].dblist) {
+                kh_destroy(dynablocks, mmaplist[i].dblist);
+                mmaplist[i].dblist = NULL;
+            }
+            if(mmaplist[i].helper) {
+                box_free(mmaplist[i].helper);
+                mmaplist[i].helper = NULL;
+            }
         }
-        if(mmaplist[i].helper) {
-            free(mmaplist[i].helper);
-            mmaplist[i].helper = NULL;
+        if(dblist_oversized) {
+            kh_destroy(dynablocks, dblist_oversized);
+            dblist_oversized = NULL;
         }
+        mmapsize = 0;
+        mmapcap = 0;
+        dynarec_log(LOG_DEBUG, "Free dynamic Dynarecblocks\n");
+        uintptr_t idx = 0;
+        uintptr_t end = ((0xFFFFFFFF)>>DYNAMAP_SHIFT);
+        for (uintptr_t i=idx; i<=end; ++i)
+            if(dynmap[i])
+                FreeDynablockList(&dynmap[i]);
+        box_free(mmaplist);
+        for (int i=0; i<DYNAMAP_SIZE; ++i)
+            if(box86_jumptable[i]!=box86_jmptbl_default)
+                box_free(box86_jumptable[i]);
     }
-    if(dblist_oversized) {
-        kh_destroy(dynablocks, dblist_oversized);
-        dblist_oversized = NULL;
-    }
-    mmapsize = 0;
-    dynarec_log(LOG_DEBUG, "Free dynamic Dynarecblocks\n");
-    uintptr_t idx = 0;
-    uintptr_t end = ((0xFFFFFFFF)>>DYNAMAP_SHIFT);
-    for (uintptr_t i=idx; i<=end; ++i) {
-        dynablocklist_t* dblist = dynmap[i];
-        if(dblist) {
-            uintptr_t startdb = StartDynablockList(dblist);
-            uintptr_t enddb = EndDynablockList(dblist);
-            uintptr_t startaddr = 0;
-            if(startaddr<startdb) startaddr = startdb;
-            uintptr_t endaddr = 0xFFFFFFFF;
-            if(endaddr>enddb) endaddr = enddb;
-            FreeRangeDynablock(dblist, startaddr, endaddr-startaddr+1);
-        }
-    }
-    for (uintptr_t i=idx; i<=end; ++i)
-        if(dynmap[i])
-            FreeDynablockList(&dynmap[i]);
-    pthread_mutex_destroy(&mutex_mmap);
-    free(mmaplist);
-    for (int i=0; i<DYNAMAP_SIZE; ++i)
-        if(box86_jumptable[i]!=box86_jmptbl_default)
-            free(box86_jumptable[i]);
+    kh_destroy(lockaddress, lockaddress);
+    lockaddress = NULL;
 #endif
     for(int i=0; i<n_blocks; ++i)
-        #ifdef USE_MMAP
+        #ifdef USE_MMAP_MORE
         munmap(p_blocks[i].block, p_blocks[i].size);
         #else
-        free(p_blocks[i].block);
+        box_free(p_blocks[i].block);
         #endif
-    free(p_blocks);
+    box_free(p_blocks);
     pthread_mutex_destroy(&mutex_blocks);
+    while(mapmem) {
+        mapmem_t *tmp = mapmem;
+        mapmem = mapmem->next;
+        box_free(tmp);
+    }
 }
+
+#ifdef DYNAREC
+// add an address to the list of "LOCK"able
+void addLockAddress(uintptr_t addr)
+{
+    int ret;
+    kh_put(lockaddress, lockaddress, addr, &ret);
+}
+
+// return 1 is the address is used as a LOCK, 0 else
+int isLockAddress(uintptr_t addr)
+{
+    khint_t k = kh_get(lockaddress, lockaddress, addr);
+    return (k==kh_end(lockaddress))?0:1;
+}
+
+#endif

src/dynarec/arm_emitter.h

@@ -66,7 +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) )
@@ -107,6 +107,8 @@ Op is 20-27
 #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
@@ -115,6 +117,8 @@ 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))
@@ -169,6 +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) \
@@ -182,11 +189,14 @@ 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) \
@@ -203,6 +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) )
@@ -232,6 +245,12 @@ Op is 20-27
 // 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) )
@@ -241,18 +260,33 @@ 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 )
@@ -265,6 +299,9 @@ Op is 20-27
 // 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) )
@@ -302,6 +339,9 @@ Op is 20-27
 // 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) )
@@ -324,9 +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]
@@ -427,7 +464,6 @@ 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))
@@ -450,6 +486,11 @@ 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
@@ -466,6 +507,10 @@ 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))
@@ -532,6 +577,16 @@ Op is 20-27
 #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))
 // Reverse bits of Rm, store result in Rd
@@ -543,6 +598,10 @@ 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))
@@ -551,17 +610,17 @@ 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))
 
 // Yield
 #define YIELD(cond) EMIT(cond | 0b00110010<<20 | 0b1111<<12 | 1)
 
-// VFPU
-#define TRANSFERT64(C, op) ((0b1100<<24) | (0b010<<21) | (0b101<<9) | ((C)<<8) | ((op)<<4))
+// 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))
@@ -595,6 +654,8 @@ 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]
@@ -626,6 +687,10 @@ 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
@@ -633,10 +698,14 @@ 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
@@ -648,6 +717,42 @@ Op is 20-27
 // Inplace convert from F64 to S16. Rounding is not selectable
 #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) )
 
@@ -683,6 +788,12 @@ 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) )
@@ -691,6 +802,13 @@ 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
 
@@ -771,19 +889,19 @@ Op is 20-27
 #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))
@@ -827,7 +945,7 @@ 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))
@@ -840,6 +958,7 @@ Op is 20-27
 #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))
@@ -1245,7 +1364,17 @@ Op is 20-27
 
 #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))

src/dynarec/arm_epilog.S

@@ -10,10 +10,10 @@
 arm_epilog:
     //update register -> emu
     //pop     {r0}
-    stm     r0, {r4-r12,r14}   // put back reg value in emu, including EIP (so r14 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}
+    pop     {r4-r12, pc}
     //end, return...
     //bx      lr
 
@@ -22,6 +22,6 @@ arm_epilog:
 arm_epilog_fast:
     //restore all used register
     vpop     {d8-d15}
-    pop     {r4-r11, pc}
+    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>
@@ -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>
@@ -237,107 +244,141 @@ NEON 1 1 1 1 0 0 1 Q 1 D size Vn Vd 1 1 0 0 N 1 M 0 Vm @ size=0b01=0b10 unsigned
 NEON 1 1 1 1 0 0 1 Q 1 D size Vn Vd 1 1 0 1 N 1 M 0 Vm @ size=0b01=0b10 unsigned int decodedImm !@ decodedImm = m >> %; m %@ 3@ &= 0x7@ 4@ &= 0xF@ 5@ = 0x21@ @ VQRDMULH<c>.<dt> <Qd>, <Qn>, <Dm[x]>
 INVALIDATE 1 1 1 1 0 0 1 <1> 1 <16> 0 <1> 0 <4>
 
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 0 size @<2> Rm @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 1 size @<2> Rm @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\], <Rm>
-;INVALIDATE 1 1 1 1 0 1 0 0 0 <1> 0 0 <22>
-INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 0 0 <8> 1 1 <10>
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 0 @<4> 1 1 1 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 0 @<4> 1 1 0 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]!
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 0 @<4> Rm @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]!
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]!
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> Rm @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]!
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
-;INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 0 0 <20>
-;INVALIDATE 1 1 1 1 0 1 0 0 <2> 0 0 <20>
-
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 0 size @<2> Rm @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 1 size @<2> Rm @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\], <Rm>
-;INVALIDATE 1 1 1 1 0 1 0 0 0 <1> 1 0 <22>
-INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <8> 1 1 <10>
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 0 @<4> 1 1 1 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]
+INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 0 0 <8> 1 1 <10>
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 0 @<4> 1 1 0 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 0 @<4> 1 1 1 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 0 @<4> Rm @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 0 @<4> 1 1 0 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]!
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 0 @<4> 1 1 1 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 0 @<4> Rm @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
+INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <8> 1 1 <2> 1 1 <6>
+INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <8> 1 1 <2> 0 0 <1> 1 <4>
+INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <8> 1 1 1 0 <3> 1 <4>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 0 @<1> 1 1 0 1 @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 0 @<1> 1 1 1 1 @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 0 @<1> Rm @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 1 @<1> 1 1 0 1 @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 1 @<1> 1 1 1 1 @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + d]\[]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 1 @<1> Rm @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + d]\[]}, [<Rn>\align\], <Rm>
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 0 @<4> 1 1 0 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 0 @<4> 1 1 1 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 0 @<4> Rm @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> 1 1 0 1 @ param2_1=1 char align[4] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> 1 1 1 1 @ param2_1=1 char align[4] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> Rm @ param2_1=1 char align[4] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\], <Rm>
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]!
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> Rm @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]!
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
-;INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <20>
-;INVALIDATE 1 1 1 1 0 1 0 0 <2> 1 0 <20>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 0 size @<1> @<1> 1 1 0 1 VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[]}, [<Rn>]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 0 size @<1> @<1> 1 1 1 1 VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[]}, [<Rn>]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 0 size @<1> @<1> Rm VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[]}, [<Rn>], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 0 @<4> 1 1 0 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 0 @<4> 1 1 1 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 0 @<4> Rm @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 1 size @<1> @<1> 1 1 0 1 @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 32 << size)@ align[0] = '\0'@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[], \vecname[0x23 + param1_1+param1_1+param1_1 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 1 size @<1> @<1> 1 1 1 1 @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 32 << size)@ align[0] = '\0'@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[], \vecname[0x23 + param1_1+param1_1+param1_1 + d]\[]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 1 size @<1> @<1> Rm @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 32 << size)@ align[0] = '\0'@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[], \vecname[0x23 + param1_1+param1_1+param1_1 + d]\[]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
 INVALIDATE 1 1 1 1 0 1 0 0 <3> 0 <20>
 
+ARM_ 1 1 1 1 0 1 0 0 @<1> 0 0 1 Rn @<8> @<4> @<4> 'NOP'
+ARM_ 1 1 1 1 0 1 0 0 U 1 0 1 Rn (1) (1) (1) (1) imm12 PLI [<Rn>, #+/-<imm12>]
+;INVALIDATE 1 1 1 1 0 1 0 0 <2> 1 1 <4> <8> <4> <4>
+ARM_ 1 1 1 1 0 1 0 1 U @<1> 0 1 Rn (1) (1) (1) (1) imm12 PLD\param ? "W" : ""\ [<Rn>, #+/-<imm12>]
+ARM_ 1 1 1 1 0 1 0 1 U (1) 0 1 1 1 1 1 (1) (1) (1) (1) imm12 PLD [PC, #+/-<imm>]
+;INVALIDATE 1 1 1 1 0 1 0 1 0 0 1 1 <4> <8> <4> <4>
+;INVALIDATE 1 1 1 1 0 1 0 1 0 1 1 1 <4> <8> 0 0 0 0 <4>
+ARM_ 1 1 1 1 0 1 0 1 0 1 1 1 (1) (1) (1) (1) (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 1 (1) (1) (1) (1) CLREX
+;INVALIDATE 1 1 1 1 0 1 0 1 0 1 1 1 <4> <8> 0 0 1 <1> <4>
+ARM_ 1 1 1 1 0 1 0 1 0 1 1 1 (1) (1) (1) (1) (1) (1) (1) (1) (0) (0) (0) (0) 0 1 0 0 @<4> @ param=0b1111=0b1110=0b1011=0b1010=0b0111=0b0110=0b0011=0b0010 const char* option option,"SY","ST","ISH","ISHST","NSH","NSHST","OSH","OSHST","!!!"@ @ DSB \option\
+ARM_ 1 1 1 1 0 1 0 1 0 1 1 1 (1) (1) (1) (1) (1) (1) (1) (1) (0) (0) (0) (0) 0 1 0 1 @<4> @ param=0b1111=0b1110=0b1011=0b1010=0b0111=0b0110=0b0011=0b0010 const char* option option,"SY","ST","ISH","ISHST","NSH","NSHST","OSH","OSHST","!!!"@ @ DMB \option\
+ARM_ 1 1 1 1 0 1 0 1 0 1 1 1 (1) (1) (1) (1) (1) (1) (1) (1) (0) (0) (0) (0) 0 1 1 0 @<4> @ param=0b1111 const char* option option,"SY","!!!"@ @ ISB \option\
+;INVALIDATE 1 1 1 1 0 1 0 1 0 1 1 1 <4> <8> 0 1 1 1 <4>
+;INVALIDATE 1 1 1 1 0 1 0 1 0 1 1 1 <4> <8> 1 <3> <4>
+;INVALIDATE 1 1 1 1 0 1 0 1 0 1 1 1 <4> <8> <4> <4>
+;INVALIDATE 1 1 1 1 0 1 0 1 1 <1> 1 1 <4> <8> <4> <4>
+ARM_ 1 1 1 1 0 1 1 0 @<1> 0 0 1 Rn @<8> @<3> 0 @<4> 'NOP'
+ARMs 1 1 1 1 0 1 1 0 U 1 0 1 Rn (1) (1) (1) (1) imm5 type 0 Rm PLI [<Rn>, +/-<Rm>{, <shift>}]
+ARMs 1 1 1 1 0 1 1 1 U @<1> 0 1 Rn (1) (1) (1) (1) imm5 type 0 Rm PLD\param ? "W" : ""\ [<Rn>, +/-<Rm>{, <shift>}]
+;INVALIDATE 1 1 1 1 0 1 1 <3> 1 1 <4> <8> <3> 0 <4>
+# Permanently undefined
+INVALIDATE 1 1 1 1 0 1 1 1 1 1 1 1 <4> <8> 1 1 1 1 <4>
+;INVALIDATE 1 1 1 1 0 <7> <4> <8> <4> <4>
+
 ; ===== No more 1 1 1 1 ..., so invalidate all of them ====
 INVALIDATE 1 1 1 1 <28>
 
@@ -390,6 +431,8 @@ 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>
@@ -437,6 +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>
@@ -495,6 +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>}]{!}
@@ -504,6 +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>
@@ -537,14 +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>
@@ -558,6 +616,7 @@ 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 << 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>
@@ -587,7 +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_lock_helper.S

@@ -20,6 +20,7 @@
 
 arm_lock_read_b:
     // address is r0, return is r0
+    dmb     ish
     ldrexb  r0, [r0]
     bx      lr
 
@@ -31,6 +32,7 @@ arm_lock_write_b:
 
 arm_lock_read_h:
     // address is r0, return is r0
+    dmb     ish
     ldrexh  r0, [r0]
     bx      lr
 
@@ -43,6 +45,7 @@ arm_lock_write_h:
 arm_lock_read_d:
     // address is r0, return is r0
     // r0 needs to be aligned
+    dmb     ish
     ldrex   r0, [r0]
     bx      lr
 
@@ -55,6 +58,7 @@ arm_lock_write_d:
 
 arm_lock_read_dd:
     // address is r2, return is r0, r1
+    dmb     ish
     ldrexd  r2, r3, [r2]
     str     r2, [r0]
     str     r3, [r1]
@@ -70,6 +74,7 @@ arm_lock_write_dd:
 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
@@ -80,6 +85,7 @@ arm_lock_xchg:
 arm_lock_storeifnull:
     // address is r0, value is r1, r1 store to r0 only if [r0] is 0. return new [r0] value (so r1 or old value)
     // r0 needs to be aligned
+    dmb     ish
     ldrex   r2, [r0]
     cmp     r2, #0
     bne     arm_lock_storeifnull_exit
@@ -93,6 +99,7 @@ arm_lock_storeifnull_exit:
 
 arm_lock_storeifref:
     push    {r4}
+    dmb     ish
 arm_lock_storeifref_loop:
     // address is r0, value is r1, r1 store to r0 only if [r0] is r2. return new [r0] value (so r1 or old value)
     ldrex   r3, [r0]

src/dynarec/arm_prolog.S

@@ -9,7 +9,7 @@
 .global arm_prolog
 arm_prolog:
     //save all used register
-    push     {r4-r11, lr}
+    push     {r4-r12, lr}
     vpush     {d8-d15}
     //setup emu -> register
     //push     {r0}

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"
@@ -18,6 +20,7 @@
 #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"
@@ -44,23 +47,25 @@ dynablocklist_t* NewDynablockList(uintptr_t text, int textsz, int direct)
         printf_log(LOG_NONE, "Error, creating a NULL sized Dynablock\n");
         return NULL;
     }
-    dynablocklist_t* ret = (dynablocklist_t*)calloc(1, sizeof(dynablocklist_t));
+    dynablocklist_t* ret = (dynablocklist_t*)box_calloc(1, sizeof(dynablocklist_t));
     ret->text = text;
     ret->textsz = textsz;
     if(direct && textsz) {
-        ret->direct = (dynablock_t**)calloc(textsz, sizeof(dynablock_t*));
+        ret->direct = (dynablock_t**)box_calloc(textsz, sizeof(dynablock_t*));
         if(!ret->direct) {printf_log(LOG_NONE, "Warning, fail to create direct block for dynablock @%p\n", (void*)text);}
     }
     dynarec_log(LOG_DEBUG, "New Dynablocklist %p, from %p->%p\n", ret, (void*)text, (void*)(text+textsz));
     return ret;
 }
 
-void FreeDynablock(dynablock_t* db)
+void FreeDynablock(dynablock_t* db, int need_lock)
 {
     if(db) {
-        dynarec_log(LOG_DEBUG, "FreeDynablock(%p), db->block=%p x86=%p:%p father=%p, with %d son(s) already gone=%d\n", db, db->block, db->x86_addr, db->x86_addr+db->x86_size, db->father, db->sons_size, db->gone);
         if(db->gone)
             return; // already in the process of deletion!
+        dynarec_log(LOG_DEBUG, "FreeDynablock(%p), db->block=%p x86=%p:%p father=%p, with %d son(s) already gone=%d\n", db, db->block, db->x86_addr, db->x86_addr+db->x86_size, db->father, db->sons_size, db->gone);
+        if(need_lock)
+            pthread_mutex_lock(&my_context->mutex_dyndump);
         db->done = 0;
         db->gone = 1;
         // remove from direct if there
@@ -76,16 +81,18 @@ void FreeDynablock(dynablock_t* db)
         // remove and free the sons
         for (int i=0; i<db->sons_size; ++i) {
             dynablock_t *son = (dynablock_t*)arm_lock_xchg(&db->sons[i], 0);
-            FreeDynablock(son);
+            FreeDynablock(son, 0);
         }
         // only the father free the DynarecMap
         if(!db->father) {
             dynarec_log(LOG_DEBUG, " -- FreeDyrecMap(%p, %d)\n", db->block, db->size);
             FreeDynarecMap(db, (uintptr_t)db->block, db->size);
+            box_free(db->sons);
+            box_free(db->instsize);
         }
-        free(db->sons);
-        free(db->instsize);
-        free(db);
+        box_free(db);
+        if(need_lock)
+            pthread_mutex_unlock(&my_context->mutex_dyndump);
     }
 }
 
@@ -99,13 +106,13 @@ void FreeDynablockList(dynablocklist_t** dynablocks)
     if((*dynablocks)->direct) {
         for (int i=0; i<(*dynablocks)->textsz; ++i) {
             if((*dynablocks)->direct[i] && !(*dynablocks)->direct[i]->father) 
-                FreeDynablock((*dynablocks)->direct[i]);
+                FreeDynablock((*dynablocks)->direct[i], my_context?1:0);
         }
-        free((*dynablocks)->direct);
+        box_free((*dynablocks)->direct);
     }
     (*dynablocks)->direct = NULL;
 
-    free(*dynablocks);
+    box_free(*dynablocks);
     *dynablocks = NULL;
 }
 
@@ -116,7 +123,7 @@ void MarkDynablock(dynablock_t* db)
             db = db->father;    // mark only father
         if(db->need_test)
             return; // already done
-        db->need_test = 1;  // test only blocks that can be marked (and so deleted)
+        db->need_test = 1;
         setJumpTableDefault(db->x86_addr);
         for(int i=0; i<db->sons_size; ++i)
             setJumpTableDefault(db->sons[i]->x86_addr);
@@ -171,6 +178,7 @@ void FreeRangeDynablock(dynablocklist_t* dynablocks, uintptr_t addr, uintptr_t s
         return;
 
     if(dynablocks->direct) {
+        int need_lock = my_context?1:0;
         dynablock_t* db;
         int ret;
         khint_t k;
@@ -200,7 +208,7 @@ void FreeRangeDynablock(dynablocklist_t* dynablocks, uintptr_t addr, uintptr_t s
         }
         // purge the list
         kh_foreach_value(blocks, db,
-            FreeDynablock(db);
+            FreeDynablock(db, need_lock);
         );
         kh_destroy(dynablocks, blocks);
     }
@@ -258,7 +266,7 @@ dynablock_t *AddNewDynablock(dynablocklist_t* dynablocks, uintptr_t addr, int* c
     if(dynablocks->direct) {
         block = dynablocks->direct[addr-dynablocks->text];
         if(block) {
-            dynarec_log(LOG_DUMP, "Block already exist in Direct Map\n");
+            dynarec_log(LOG_VERBOSE, "Block already exist in Direct Map\n");
             *created = 0;
             return block;
         }
@@ -267,39 +275,50 @@ dynablock_t *AddNewDynablock(dynablocklist_t* dynablocks, uintptr_t addr, int* c
     if (!*created)
         return block;
     
+    pthread_mutex_lock(&my_context->mutex_dyndump);
     if(!dynablocks->direct) {
-        dynablock_t** p = (dynablock_t**)calloc(dynablocks->textsz, sizeof(dynablock_t*));
+        dynablock_t** p = (dynablock_t**)box_calloc(dynablocks->textsz, sizeof(dynablock_t*));
         if(arm_lock_storeifnull(&dynablocks->direct, p)!=p)
-            free(p);    // someone already create the direct array, too late...
+            box_free(p);    // someone already create the direct array, too late...
     }
 
     // create and add new block
-    dynarec_log(LOG_DUMP, "Ask for DynaRec Block creation @%p\n", (void*)addr);
+    dynarec_log(LOG_VERBOSE, "Ask for DynaRec Block creation @%p\n", (void*)addr);
 
-    block = (dynablock_t*)calloc(1, sizeof(dynablock_t));
+    block = (dynablock_t*)box_calloc(1, sizeof(dynablock_t));
     block->parent = dynablocks; 
     dynablock_t* tmp = (dynablock_t*)arm_lock_storeifnull(&dynablocks->direct[addr-dynablocks->text], block);
+    pthread_mutex_unlock(&my_context->mutex_dyndump);
     if(tmp !=  block) {
         // a block appeard!
-        free(block);
+        box_free(block);
         *created = 0;
         return tmp;
     }
-
     *created = 1;
     return block;
 }
 
+//TODO: move this to dynrec_arm.c and track allocated structure to avoid memory leak
+static __thread struct __jmp_buf_tag dynarec_jmpbuf;
+
+void cancelFillBlock()
+{
+    longjmp(&dynarec_jmpbuf, 1);
+}
+
 /* 
     return NULL if block is not found / cannot be created. 
     Don't create if create==0
 */
-static dynablock_t* internalDBGetBlock(x86emu_t* emu, uintptr_t addr, uintptr_t filladdr, int create, dynablock_t* current)
+static dynablock_t* internalDBGetBlock(x86emu_t* emu, uintptr_t addr, uintptr_t filladdr, int create, dynablock_t* current, int need_lock)
 {
+    if(hasAlternate((void*)addr))
+        return NULL;
     // try the quickest way first: get parent of current and check if ok!
     dynablocklist_t *dynablocks = NULL;
     dynablock_t* block = NULL;
-    if(current) {
+    if(current && current->done) {
         dynablocks = current->parent;
         if(dynablocks && !(addr>=dynablocks->text && addr<(dynablocks->text+dynablocks->textsz)))
             dynablocks = NULL;
@@ -323,15 +342,19 @@ static dynablock_t* internalDBGetBlock(x86emu_t* emu, uintptr_t addr, uintptr_t
     if(!created)
         return block;   // existing block...
 
-    #if 0
-    if(box86_dynarec_dump)
-        pthread_mutex_lock(&my_context->mutex_dyndump);
-    #endif
     // fill the block
     block->x86_addr = (void*)addr;
-    pthread_mutex_lock(&my_context->mutex_dyndump);
+    if(need_lock)
+        pthread_mutex_lock(&my_context->mutex_dyndump);
+    if(sigsetjmp(&dynarec_jmpbuf, 1)) {
+        printf_log(LOG_INFO, "FillBlock at %p triggered a segfault, cancelling\n", (void*)addr);
+        if(need_lock)
+            pthread_mutex_unlock(&my_context->mutex_dyndump);
+        return NULL;
+    }
     void* ret = FillBlock(block, filladdr);
-    pthread_mutex_unlock(&my_context->mutex_dyndump);
+    if(need_lock)
+        pthread_mutex_unlock(&my_context->mutex_dyndump);
     if(!ret) {
         dynarec_log(LOG_DEBUG, "Fillblock of block %p for %p returned an error\n", block, (void*)addr);
         void* old = (void*)arm_lock_storeifref(&dynablocks->direct[addr-dynablocks->text], 0, block);
@@ -339,13 +362,9 @@ static dynablock_t* internalDBGetBlock(x86emu_t* emu, uintptr_t addr, uintptr_t
             dynarec_log(LOG_INFO, "Warning, a wild block appeared at %p: %p\n", (void*)addr, old);
             // Doing nothing else, the block has not been written
         }
-        free(block);
+        box_free(block);
         block = NULL;
     }
-    #if 0
-    if(box86_dynarec_dump)
-        pthread_mutex_unlock(&my_context->mutex_dyndump);
-    #endif
     // check size
     if(block && block->x86_size) {
         int blocksz = block->x86_size;
@@ -358,7 +377,10 @@ static dynablock_t* internalDBGetBlock(x86emu_t* emu, uintptr_t addr, uintptr_t
                         dblist->maxsz = blocksz;
             }
         }
-        protectDB((uintptr_t)block->x86_addr, block->x86_size);
+        //protectDB((uintptr_t)block->x86_addr, block->x86_size);   // already protected in FillBlock
+        // but check if it already need test
+        if(block->need_test || (block->father && block->father->need_test))
+            return NULL;
         // fill-in jumptable
         addJumpTableIfDefault(block->x86_addr, block->block);
         for(int i=0; i<block->sons_size; ++i)
@@ -372,19 +394,21 @@ static dynablock_t* internalDBGetBlock(x86emu_t* emu, uintptr_t addr, uintptr_t
 
 dynablock_t* DBGetBlock(x86emu_t* emu, uintptr_t addr, int create, dynablock_t** current)
 {
-    dynablock_t *db = internalDBGetBlock(emu, addr, addr, create, *current);
+    dynablock_t *db = internalDBGetBlock(emu, addr, addr, create, *current, 1);
     if(db && db->done && db->block && (db->need_test || (db->father && db->father->need_test))) {
+        if(pthread_mutex_trylock(&my_context->mutex_dyndump))
+            return NULL;
         dynablock_t *father = db->father?db->father:db;
-        uint32_t hash = X31_hash_code(father->x86_addr, father->x86_size);
+        uint32_t hash = (getProtection((uintptr_t)father->x86_addr)&PROT_READ)?X31_hash_code(father->x86_addr, father->x86_size):0;
         if(hash!=father->hash) {
             dynarec_log(LOG_DEBUG, "Invalidating block %p from %p:%p (hash:%X/%X) with %d son(s) for %p\n", father, father->x86_addr, father->x86_addr+father->x86_size, hash, father->hash, father->sons_size, (void*)addr);
             // no more current if it gets invalidated too
             if(*current && father->x86_addr>=(*current)->x86_addr && (father->x86_addr+father->x86_size)<(*current)->x86_addr)
                 *current = NULL;
             // Free father, it's now invalid!
-            FreeDynablock(father);
+            FreeDynablock(father, 0);
             // start again... (will create a new block)
-            db = internalDBGetBlock(emu, addr, addr, create, *current);
+            db = internalDBGetBlock(emu, addr, addr, create, *current, 0);
         } else {
             father->need_test = 0;
             protectDB((uintptr_t)father->x86_addr, father->x86_size);
@@ -393,6 +417,7 @@ dynablock_t* DBGetBlock(x86emu_t* emu, uintptr_t addr, int create, dynablock_t**
             for(int i=0; i<father->sons_size; ++i)
                 addJumpTableIfDefault(father->sons[i]->x86_addr, father->sons[i]->block);
         }
+        pthread_mutex_unlock(&my_context->mutex_dyndump);
     } 
     return db;
 }
@@ -401,16 +426,18 @@ dynablock_t* DBAlternateBlock(x86emu_t* emu, uintptr_t addr, uintptr_t filladdr)
 {
     dynarec_log(LOG_DEBUG, "Creating AlternateBlock at %p for %p\n", (void*)addr, (void*)filladdr);
     int create = 1;
-    dynablock_t *db = internalDBGetBlock(emu, addr, filladdr, create, NULL);
+    dynablock_t *db = internalDBGetBlock(emu, addr, filladdr, create, NULL, 1);
     if(db && db->done && db->block && (db->need_test || (db->father && db->father->need_test))) {
+        if(pthread_mutex_trylock(&my_context->mutex_dyndump))
+            return NULL;
         dynablock_t *father = db->father?db->father:db;
-        uint32_t hash = X31_hash_code(father->x86_addr, father->x86_size);
+        uint32_t hash = (getProtection((uintptr_t)father->x86_addr)&PROT_READ)?X31_hash_code(father->x86_addr, father->x86_size):0;
         if(hash!=father->hash) {
             dynarec_log(LOG_DEBUG, "Invalidating alt block %p from %p:%p (hash:%X/%X) with %d son(s) for %p\n", father, father->x86_addr, father->x86_addr+father->x86_size, hash, father->hash, father->sons_size, (void*)addr);
             // Free father, it's now invalid!
-            FreeDynablock(father);
+            FreeDynablock(father, 0);
             // start again... (will create a new block)
-            db = internalDBGetBlock(emu, addr, filladdr, create, NULL);
+            db = internalDBGetBlock(emu, addr, filladdr, create, NULL, 0);
         } else {
             father->need_test = 0;
             protectDB((uintptr_t)father->x86_addr, father->x86_size);
@@ -419,6 +446,7 @@ dynablock_t* DBAlternateBlock(x86emu_t* emu, uintptr_t addr, uintptr_t filladdr)
             for(int i=0; i<father->sons_size; ++i)
                 addJumpTableIfDefault(father->sons[i]->x86_addr, father->sons[i]->block);
         }
+        pthread_mutex_unlock(&my_context->mutex_dyndump);
     } 
     return db;
 }

src/dynarec/dynablock_private.h

@@ -4,8 +4,8 @@
 typedef struct dynablocklist_s  dynablocklist_t;
 
 typedef struct instsize_s {
-    unsigned int x86:4;
-    unsigned int nat:4;
+    unsigned char x86:4;
+    unsigned char nat:4;
 } instsize_t;
 
 typedef struct dynablock_s {

src/dynarec/dynarec.c

@@ -19,6 +19,7 @@
 #ifdef DYNAREC
 #include "dynablock.h"
 #include "dynablock_private.h"
+#include "bridge.h"
 #endif
 
 #ifdef ARM
@@ -29,6 +30,7 @@ void arm_epilog_fast() EXPORTDYN;
 
 #ifdef DYNAREC
 #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)
@@ -37,7 +39,9 @@ void* LinkNext(x86emu_t* emu, uintptr_t addr, void* x2)
     if(!addr) {
         x2-=8;  // actual PC is 2 instructions ahead
         dynablock_t* db = FindDynablockFromNativeAddress(x2);
-        printf_log(LOG_NONE, "Warning, jumping to NULL address from %p (db=%p, x86addr=%p)\n", x2, db, db?(void*)getX86Address(db, (uintptr_t)x2):NULL);
+        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
     dynablock_t* current = NULL;
@@ -45,12 +49,21 @@ void* LinkNext(x86emu_t* emu, uintptr_t addr, void* x2)
     dynablock_t* block = DBGetBlock(emu, addr, 1, &current);
     if(!block) {
         // no block, let link table as is...
-        #ifdef HAVE_TRACE
-        dynablock_t* db = FindDynablockFromNativeAddress(x2);
-        printf_log(LOG_INFO, "Warning, jumping to a no-block address from %p (db=%p, x86addr=%p)\n", x2, db, db?(void*)getX86Address(db, (uintptr_t)x2):NULL);
-        #endif
-        //tableupdate(arm_epilog, addr, table);
-        return arm_epilog;
+        if(hasAlternate((void*)addr)) {
+            printf_log(LOG_INFO, "Jmp address has alternate: %p", (void*)addr);
+            addr = (uintptr_t)getAlternate((void*)addr);
+            R_EIP = addr;
+            printf_log(LOG_INFO, " -> %p\n", (void*)addr);
+            block = DBGetBlock(emu, addr, 1, &current);
+        }
+        if(!block) {
+            #ifdef HAVE_TRACE
+            dynablock_t* db = FindDynablockFromNativeAddress(x2);
+            dynarec_log(LOG_INFO, "Warning, jumping to a no-block address %p from %p (db=%p, x86addr=%p)\n", (void*)addr, x2, db, db?(void*)getX86Address(db, (uintptr_t)x2):NULL);
+            #endif
+            //tableupdate(arm_epilog, addr, table);
+            return arm_epilog;
+        }
     }
     if(!block->done) {
         // not finished yet... leave linker
@@ -155,6 +168,7 @@ void DynaCall(x86emu_t* emu, uintptr_t addr)
         ejb->jmpbuf_ok = 0;
 }
 
+int my_setcontext(x86emu_t* emu, void* ucp);
 int DynaRun(x86emu_t* emu)
 {
     // prepare setjump for signal handling
@@ -216,6 +230,10 @@ int DynaRun(x86emu_t* emu)
                         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

src/dynarec/dynarec_arm.c

@@ -3,6 +3,7 @@
 #include <pthread.h>
 #include <errno.h>
 #include <string.h>
+#include <assert.h>
 
 #include "debug.h"
 #include "box86context.h"
@@ -28,15 +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]);
             }
@@ -58,13 +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 += 16;
+        dyn->next = (uintptr_t*)box_realloc(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) {
@@ -73,13 +84,12 @@ uintptr_t get_closest_next(dynarec_arm_t *dyn, uintptr_t addr) {
     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;
+            dyn->next[i] = 0;
         } else {
             if((dyn->next[i]<best) || !best)
                 best = dyn->next[i];
-            ++i;
         }
+        ++i;
     }
     return best;
 }
@@ -90,7 +100,9 @@ int is_nops(dynarec_arm_t *dyn, uintptr_t addr, int 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);
@@ -204,7 +216,7 @@ int next_instruction(dynarec_arm_t *dyn, uintptr_t addr)
             nextop = PK(1);
             switch((nextop>>3)&7) {
                 case 0: // INC Ed
-                case 1: //DEC Ed
+                case 1: // DEC Ed
                 case 2: // CALL Ed
                 case 4: // JMP Ed
                 case 6: // Push Ed
@@ -229,54 +241,7 @@ int is_instructions(dynarec_arm_t *dyn, uintptr_t addr, int n)
     return (i==n)?1:0;
 }
 
-uint32_t needed_flags(dynarec_arm_t *dyn, int ninst, uint32_t setf, int recurse)
-{
-    if(recurse == 10)
-        return X_PEND;
-    if(ninst == dyn->size)
-        return X_PEND; // 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;
-    }
-
-    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;
-    }
-
-    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;
-}
-
-instsize_t* addInst(instsize_t* insts, size_t* size, size_t* cap, int x86_size, int arm_size)
+static instsize_t* addInst(instsize_t* insts, size_t* size, size_t* cap, int x86_size, int arm_size)
 {
     // x86 instruction is <16 bytes
     int toadd;
@@ -286,7 +251,7 @@ instsize_t* addInst(instsize_t* insts, size_t* size, size_t* cap, int x86_size,
         toadd = 1 + arm_size/15;
     if((*size)+toadd>(*cap)) {
         *cap = (*size)+toadd;
-        insts = (instsize_t*)realloc(insts, (*cap)*sizeof(instsize_t));
+        insts = (instsize_t*)box_realloc(insts, (*cap)*sizeof(instsize_t));
     }
     while(toadd) {
         if(x86_size>15)
@@ -305,11 +270,109 @@ instsize_t* addInst(instsize_t* insts, size_t* size, size_t* cap, int x86_size,
     return insts;
 }
 
+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;
+        }
+    }
+    // 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*)box_malloc(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 && (!i || dyn->insts[i].pred_sz))
+            dyn->insts[i+1].pred[dyn->insts[i+1].pred_sz++] = i;
+        if(dyn->insts[i].x86.jmp && dyn->insts[i].x86.jmp_insts!=-1)
+            dyn->insts[dyn->insts[i].x86.jmp_insts].pred[dyn->insts[dyn->insts[i].x86.jmp_insts].pred_sz++] = i;
+    }
+
+}
+
+// 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);
+        }
+        if(!ok)
+            return ninst - 1;
+        --ninst;
+    }
+    return ninst;
+}
+
 uintptr_t arm_pass0(dynarec_arm_t* dyn, uintptr_t addr);
 uintptr_t arm_pass1(dynarec_arm_t* dyn, uintptr_t addr);
 uintptr_t arm_pass2(dynarec_arm_t* dyn, uintptr_t addr);
 uintptr_t arm_pass3(dynarec_arm_t* dyn, uintptr_t addr);
 
+__thread void* current_helper = NULL;
+
+void CancelBlock()
+{
+    dynarec_arm_t* helper = (dynarec_arm_t*)current_helper;
+    current_helper = NULL;
+    if(!helper)
+        return;
+    box_free(helper->next);
+    box_free(helper->insts);
+    box_free(helper->predecessor);
+    if(helper->dynablock && helper->dynablock->block)
+        FreeDynarecMap(helper->dynablock, (uintptr_t)helper->dynablock->block, helper->dynablock->size);
+}
+
 void* FillBlock(dynablock_t* block, uintptr_t addr) {
 dynarec_log(LOG_DEBUG, "Asked to Fill block %p with %p\n", block, (void*)addr);
     if(addr>=box86_nodynarec_start && addr<box86_nodynarec_end) {
@@ -322,47 +385,93 @@ dynarec_log(LOG_DEBUG, "Asked to Fill block %p with %p\n", block, (void*)addr);
     }
     // init the helper
     dynarec_arm_t helper = {0};
+    current_helper = &helper;
+    helper.dynablock = block;
     helper.start = addr;
     uintptr_t start = addr;
+    helper.cap = 64; // needs epilog handling
+    helper.insts = (instruction_arm_t*)box_calloc(helper.cap, sizeof(instruction_arm_t));
+    // pass 0, addresses, x86 jump addresses, overall size of the block
     uintptr_t end = arm_pass0(&helper, addr);
+    // no need for next anymore
+    box_free(helper.next);
+    helper.next_sz = helper.next_cap = 0;
+    helper.next = NULL;
+    // basic checks
     if(!helper.size) {
         dynarec_log(LOG_DEBUG, "Warning, null-sized dynarec block (%p)\n", (void*)addr);
-        block->done = 1;
-        free(helper.next);
+        CancelBlock();
         return (void*)block;
     }
-    helper.cap = helper.size+3; // needs epilog handling
-    helper.insts = (instruction_arm_t*)calloc(helper.cap, sizeof(instruction_arm_t));
     // already protect the block and compute hash signature
-    protectDB(addr, end-addr+1);
-    uint32_t hash = X31_hash_code((void*)addr, end-addr+1);
-    // pass 1, addresses, x86 jump addresses, flags
-    arm_pass1(&helper, addr);
+    protectDB(addr, end-addr);  //end is 1byte after actual end
+    uint32_t hash = X31_hash_code((void*)addr, end-addr);
     // calculate barriers
     for(int i=0; i<helper.size; ++i)
         if(helper.insts[i].x86.jmp) {
             uintptr_t j = helper.insts[i].x86.jmp;
-            if(j<start || j>=end)
+            if(j<start || j>=end) {
                 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);
+    // check for the optionnal barriers now
+    /*for(int i=helper.size-1; i>=0; --i) {
+        if(helper.insts[i].barrier_maybe) {
+            // out-of-block jump
+            if(helper.insts[i].x86.jmp_insts == -1) {
+                // nothing for now
+            } else {
+                // inside block jump
+                int k = helper.insts[i].x86.jmp_insts;
+                if(k>i) {
+                    // jump in the future
+                    if(helper.insts[k].pred_sz>1) {
+                        // with multiple flow, put a barrier
+                        helper.insts[k].x86.barrier|=BARRIER_FLAGS;
+                    }
+                } else {
+                    // jump back
+                    helper.insts[k].x86.barrier|=BARRIER_FLAGS;
+                }
+            }
+	}
+    }*/
+    // check to remove useless barrier, in case of jump when destination doesn't needs flags
+    /*for(int i=helper.size-1; i>=0; --i) {
+        int k;
+        if(helper.insts[i].x86.jmp
+        && ((k=helper.insts[i].x86.jmp_insts)>=0)
+        && helper.insts[k].x86.barrier&BARRIER_FLAGS) {
+            //TODO: optimize FPU barrier too
+            if((!helper.insts[k].x86.need_flags)
+             ||(helper.insts[k].x86.set_flags==X_ALL
+                  && helper.insts[k].x86.state_flags==SF_SET)
+             ||(helper.insts[k].x86.state_flags==SF_SET_PENDING)) {
+                //if(box86_dynarec_dump) dynarec_log(LOG_NONE, "Removed barrier for inst %d\n", k);
+                helper.insts[k].x86.barrier &= ~BARRIER_FLAGS; // remove flag barrier
+             }
         }
-    
+    }*/
+    int pos = helper.size;
+    while (pos>=0)
+        pos = updateNeed(&helper, pos, 0);
+
+    // 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
@@ -370,15 +479,14 @@ dynarec_log(LOG_DEBUG, "Asked to Fill block %p with %p\n", block, (void*)addr);
     void* p = (void*)AllocDynarecMap(block, sz);
     if(p==NULL) {
         dynarec_log(LOG_DEBUG, "AllocDynarecMap(%p, %d) failed, cancelling block\n", block, sz);
-        free(helper.insts);
-        free(helper.next);
+        CancelBlock();
         return NULL;
     }
     helper.block = p;
     helper.arm_start = (uintptr_t)p;
     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*));
+        helper.sons_x86 = (uintptr_t*)alloca(helper.sons_size*sizeof(uintptr_t));
+        helper.sons_arm = (void**)alloca(helper.sons_size*sizeof(void*));
     }
     // pass 3, emit (log emit arm opcode)
     if(box86_dynarec_dump) {
@@ -408,43 +516,43 @@ dynarec_log(LOG_DEBUG, "Asked to Fill block %p with %p\n", block, (void*)addr);
         for(int i=0; i<helper.size; ++i)
             cap += 1 + ((helper.insts[i].x86.size>helper.insts[i].size)?helper.insts[i].x86.size:helper.insts[i].size)/15;
         size_t size = 0;
-        block->instsize = (instsize_t*)calloc(cap, sizeof(instsize_t));
+        block->instsize = (instsize_t*)box_calloc(cap, sizeof(instsize_t));
         for(int i=0; i<helper.size; ++i)
             block->instsize = addInst(block->instsize, &size, &cap, helper.insts[i].x86.size, helper.insts[i].size/4);
         block->instsize = addInst(block->instsize, &size, &cap, 0, 0);    // add a "end of block" mark, just in case
     }
     // ok, free the helper now
-    free(helper.insts);
-    free(helper.next);
+    box_free(helper.insts);
+    helper.insts = NULL;
+    box_free(helper.next);
+    helper.next = NULL;
     block->size = sz;
     block->isize = helper.size;
     block->block = p;
     block->need_test = 0;
     //block->x86_addr = (void*)start;
-    block->x86_size = end-start+1;
+    block->x86_size = end-start;
     if(box86_dynarec_largest<block->x86_size)
         box86_dynarec_largest = block->x86_size;
     block->hash = X31_hash_code(block->x86_addr, block->x86_size);
     // Check if something changed, to abbort if it as
     if(block->hash != hash) {
         dynarec_log(LOG_INFO, "Warning, a block changed while beeing processed hash(%p:%d)=%x/%x\n", block->x86_addr, block->x86_size, block->hash, hash);
-        free(helper.sons_x86);
-        free(helper.sons_arm);
-        FreeDynarecMap(block, (uintptr_t)p, sz);
+        CancelBlock();
         return NULL;
     }
     // 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*));
+        sons = (dynablock_t**)box_calloc(helper.sons_size, sizeof(dynablock_t*));
         for (int i=0; i<helper.sons_size; ++i) {
             int created = 1;
             dynablock_t *son = AddNewDynablock(block->parent, helper.sons_x86[i], &created);
             if(created) {    // avoid breaking a working block!
                 son->block = helper.sons_arm[i];
                 son->x86_addr = (void*)helper.sons_x86[i];
-                son->x86_size = end-helper.sons_x86[i]+1;
+                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;
@@ -457,10 +565,11 @@ dynarec_log(LOG_DEBUG, "Asked to Fill block %p with %p\n", block, (void*)addr);
             block->sons = sons;
             block->sons_size = sons_size;
         } else
-            free(sons);
+            box_free(sons);
     }
-    free(helper.sons_x86);
-    free(helper.sons_arm);
+    box_free(helper.predecessor);
+    helper.predecessor = NULL;
+    current_helper = NULL;
     block->done = 1;
     return (void*)block;
-}
+}

src/dynarec/dynarec_arm_64.c

@@ -19,6 +19,7 @@
 #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)
@@ -28,7 +29,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
     int32_t i32, j32;
     uint32_t u32;
     uint8_t gd, ed;
-    uint8_t wback;
+    uint8_t wback, u8;
     int fixedaddress;
 
     MAYUSE(j32);
@@ -37,7 +38,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
 
         case 0x03:
             INST_NAME("ADD Gd, FS:Ed");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
             grab_fsdata(dyn, addr, ninst, x14);
             nextop = F8;
             GETGD;
@@ -47,7 +48,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
 
         case 0x2B:
             INST_NAME("SUB Gd, FS:Ed");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
             grab_fsdata(dyn, addr, ninst, x14);
             nextop = F8;
             GETGD;
@@ -57,7 +58,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
 
         case 0x33:
             INST_NAME("XOR Gd, FS:Ed");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
             grab_fsdata(dyn, addr, ninst, x14);
             nextop = F8;
             GETGD;
@@ -65,9 +66,19 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
             grab_fsdata(dyn, addr, ninst, x14);
             nextop = F8;
             GETGD;
@@ -174,7 +185,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                 STR_REG_LSL_IMM5(gd, ed, x14, 0);
             }
             break;
@@ -187,7 +198,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                 LDR_REG_LSL_IMM5(gd, ed, x14, 0);
             }
             break;
@@ -199,7 +210,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             if((nextop&0xC0)==0xC0) {
                 POP1(xEAX+(nextop&7));
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 POP1(x2);
                 STR_REG_LSL_IMM5(x2, ed, x14, 0);
             }
@@ -250,6 +261,15 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             }
             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, x14);
@@ -259,7 +279,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 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, x14, 0);
@@ -278,7 +298,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             switch((nextop>>3)&7) {
                 case 2: // CALL Ed
                     INST_NAME("CALL FS:Ed");
-                    PASS2IF(ninst && dyn->insts && dyn->insts[ninst-1].x86.set_flags, 1) {
+                    PASS2IF(ninst && dyn->insts[ninst-1].x86.set_flags, 1) {
                         READFLAGS(X_PEND);          // that's suspicious
                     } else {
                         SETFLAGS(X_ALL, SF_SET);    //Hack to put flag in "don't care" state
@@ -286,16 +306,12 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     if((nextop&0xC0)==0xC0) {   // reg
                         MOV_REG(xEIP, xEAX+(nextop&7));
                     } else {                    // mem
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                         LDR_REG_LSL_IMM5(xEIP, ed, x14, 0);
                     }
-                    BARRIER(1);
-                    BARRIER_NEXT(1);
-                    if(!dyn->insts || ninst!=dyn->size-1) {
-                    } else {
-                        *need_epilog = 0;
-                        *ok = 0;
-                    }
+                    BARRIER(BARRIER_FULL);
+                    *need_epilog = 0;
+                    *ok = 0;
                     MOV32(x2, addr);
                     PUSH1(x2);
                     jump_to_next(dyn, 0, xEIP, ninst);
@@ -305,7 +321,7 @@ uintptr_t dynarecFS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     if((nextop&0xC0)==0xC0) {   // reg
                         DEFAULT;
                     } 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);
                         LDR_REG_LSL_IMM5(x3, ed, x14, 0);
                         PUSH1(x3);
                     }

src/dynarec/dynarec_arm_65.c

@@ -19,6 +19,7 @@
 #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)
@@ -28,19 +29,33 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
     int32_t i32, j32;
     uint32_t u32;
     uint8_t gd, ed;
-    uint8_t wback, wb1, wb2;
+    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 0x01:
+            INST_NAME("ADD GS:Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_tlsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGD;
+            GETEDO2(x14);
+            emit_add32(dyn, ninst, ed, gd, x3, x14);
+            WBACK;
+            break;
+
         case 0x03:
             INST_NAME("ADD Gd, GS:Ed");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
             grab_tlsdata(dyn, addr, ninst, x14);
             nextop = F8;
             GETGD;
@@ -48,9 +63,67 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             emit_add32(dyn, ninst, gd, ed, x3, x14);
             break;
 
+        case 0x0F:  // more opcodes
+            opcode = F8;
+            switch(opcode) {
+                
+                case 0xAF:
+                    INST_NAME("IMUL Gd, Ed");
+                    SETFLAGS(X_ALL, SF_PENDING);
+                    grab_tlsdata(dyn, addr, ninst, x14);
+                    nextop = F8;
+                    GETGD;
+                    GETEDO(x14);
+                    UFLAG_IF {
+                        SMULL(x3, gd, gd, ed);
+                        UFLAG_OP1(x3);
+                        UFLAG_RES(gd);
+                        UFLAG_DF(x3, d_imul32);
+                    } else {
+                        MUL(gd, gd, ed);
+                    }
+                    break;
+
+                default:
+                    DEFAULT;
+            }
+            break;
+
+        case 0x21:
+            INST_NAME("AND Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_tlsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGD;
+            GETEDO2(x14);
+            emit_and32(dyn, ninst, ed, gd, x3, x14);
+            WBACK;
+            break;
+
+        case 0x23:
+            INST_NAME("AND Gd, GS:Ed");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_tlsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGD;
+            GETEDO(x14);
+            emit_and32(dyn, ninst, gd, ed, x3, x14);
+            break;
+
+        case 0x29:
+            INST_NAME("SUB GS:Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_tlsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGD;
+            GETEDO2(x14);
+            emit_sub32(dyn, ninst, ed, gd, x3, x14);
+            WBACK;
+            break;
+
         case 0x2B:
             INST_NAME("SUB Gd, GS:Ed");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
             grab_tlsdata(dyn, addr, ninst, x14);
             nextop = F8;
             GETGD;
@@ -58,9 +131,20 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             emit_sub32(dyn, ninst, gd, ed, x3, x14);
             break;
 
+        case 0x31:
+            INST_NAME("XOR GS:Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_tlsdata(dyn, addr, ninst, x14);
+            nextop = F8;
+            GETGD;
+            GETEDO2(x14);
+            emit_xor32(dyn, ninst, gd, ed, x3, x14);
+            WBACK;
+            break;
+
         case 0x33:
             INST_NAME("XOR Gd, GS:Ed");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
             grab_tlsdata(dyn, addr, ninst, x14);
             nextop = F8;
             GETGD;
@@ -68,9 +152,19 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
             grab_tlsdata(dyn, addr, ninst, x14);
             nextop = F8;
             GETGD;
@@ -103,7 +197,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             switch((nextop>>3)&7) {
                 case 0: //ADD
                     INST_NAME("ADD GS:Eb, Ib");
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     grab_tlsdata(dyn, addr, ninst, x1);
                     GETEBO(x1);
                     u8 = F8;
@@ -112,7 +206,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 1: //OR
                     INST_NAME("OR GS:Eb, Ib");
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     grab_tlsdata(dyn, addr, ninst, x1);
                     GETEBO(x1);
                     u8 = F8;
@@ -122,7 +216,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 case 2: //ADC
                     INST_NAME("ADC GS:Eb, Ib");
                     READFLAGS(X_CF);
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     grab_tlsdata(dyn, addr, ninst, x1);
                     GETEBO(x1);
                     u8 = F8;
@@ -132,7 +226,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 case 3: //SBB
                     INST_NAME("SBB GS:Eb, Ib");
                     READFLAGS(X_CF);
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     grab_tlsdata(dyn, addr, ninst, x1);
                     GETEBO(x1);
                     u8 = F8;
@@ -141,7 +235,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 4: //AND
                     INST_NAME("AND GS:Eb, Ib");
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     grab_tlsdata(dyn, addr, ninst, x1);
                     GETEBO(x1);
                     u8 = F8;
@@ -150,7 +244,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 5: //SUB
                     INST_NAME("SUB GS:Eb, Ib");
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     grab_tlsdata(dyn, addr, ninst, x1);
                     GETEBO(x1);
                     u8 = F8;
@@ -159,7 +253,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 6: //XOR
                     INST_NAME("XOR GS:Eb, Ib");
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     grab_tlsdata(dyn, addr, ninst, x1);
                     GETEBO(x1);
                     u8 = F8;
@@ -168,7 +262,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 7: //CMP
                     INST_NAME("CMP GS:Eb, Ib");
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     grab_tlsdata(dyn, addr, ninst, x1);
                     GETEBO(x1);
                     u8 = F8;
@@ -191,7 +285,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     } else {
                         INST_NAME("ADD Ed, Ib");
                     }
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     grab_tlsdata(dyn, addr, ninst, x14);
                     GETEDO2(x14);
                     if(opcode==0x81) i32 = F32S; else i32 = F8S;
@@ -200,7 +294,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 1: //OR
                     if(opcode==0x81) {INST_NAME("OR Ed, Id");} else {INST_NAME("OR Ed, Ib");}
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     grab_tlsdata(dyn, addr, ninst, x14);
                     GETEDO2(x14);
                     if(opcode==0x81) i32 = F32S; else i32 = F8S;
@@ -210,7 +304,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 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);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     grab_tlsdata(dyn, addr, ninst, x14);
                     GETEDO2(x14);
                     if(opcode==0x81) i32 = F32S; else i32 = F8S;
@@ -220,7 +314,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 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);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     grab_tlsdata(dyn, addr, ninst, x14);
                     GETEDO2(x14);
                     if(opcode==0x81) i32 = F32S; else i32 = F8S;
@@ -229,7 +323,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 4: //AND
                     if(opcode==0x81) {INST_NAME("AND Ed, Id");} else {INST_NAME("AND Ed, Ib");}
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     grab_tlsdata(dyn, addr, ninst, x14);
                     GETEDO2(x14);
                     if(opcode==0x81) i32 = F32S; else i32 = F8S;
@@ -238,7 +332,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 5: //SUB
                     if(opcode==0x81) {INST_NAME("SUB Ed, Id");} else {INST_NAME("SUB Ed, Ib");}
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     grab_tlsdata(dyn, addr, ninst, x14);
                     GETEDO2(x14);
                     if(opcode==0x81) i32 = F32S; else i32 = F8S;
@@ -247,7 +341,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 6: //XOR
                     if(opcode==0x81) {INST_NAME("XOR Ed, Id");} else {INST_NAME("XOR Ed, Ib");}
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     grab_tlsdata(dyn, addr, ninst, x14);
                     GETEDO2(x14);
                     if(opcode==0x81) i32 = F32S; else i32 = F8S;
@@ -256,7 +350,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 7: //CMP
                     if(opcode==0x81) {INST_NAME("CMP Ed, Id");} else {INST_NAME("CMP Ed, Ib");}
-                    SETFLAGS(X_ALL, SF_SET);
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
                     grab_tlsdata(dyn, addr, ninst, x14);
                     GETEDO2(x14);
                     if(opcode==0x81) i32 = F32S; else i32 = F8S;
@@ -278,7 +372,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                 STR_REG_LSL_IMM5(gd, ed, x14, 0);
             }
             break;
@@ -291,7 +385,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, NULL);
                 LDR_REG_LSL_IMM5(gd, ed, x14, 0);
             }
             break;
@@ -303,7 +397,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             if((nextop&0xC0)==0xC0) {
                 POP1((xEAX+(nextop&7)));
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 POP1(x2);
                 STR_REG_LSL_IMM5(x2, ed, x14, 0);
             }
@@ -322,6 +416,15 @@ 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);
+            break;
+
         case 0xA3:
             INST_NAME("MOV GS:Od, EAX");
             grab_tlsdata(dyn, addr, ninst, x1);
@@ -331,6 +434,24 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             STR_IMM9(xEAX, x2, 0);
             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);
+            }
+            break;
         case 0xC7:
             INST_NAME("MOV GS:Ed, Id");
             grab_tlsdata(dyn, addr, ninst, x14);
@@ -340,13 +461,179 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 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, x14, 0);
             }
             break;
 
+        case 0xD1:
+            nextop = F8;
+            grab_tlsdata(dyn, addr, ninst, x14);
+            switch((nextop>>3)&7) {
+                case 0:
+                    INST_NAME("ROL Ed, 1");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
+                    GETEDO2(x14);
+                    emit_rol32c(dyn, ninst, ed, 1, x3, x14);
+                    WBACK;
+                    break;
+                case 1:
+                    INST_NAME("ROR Ed, 1");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
+                    GETEDO2(x14);
+                    emit_ror32c(dyn, ninst, ed, 1, x3, x14);
+                    WBACK;
+                    break;
+                case 2:
+                    INST_NAME("RCL Ed, 1");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    MOVW(x2, 1);
+                    GETEDO2(x14);
+                    wb = ed;
+                    if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
+                    CALL_(rcl32, ed, (1<<x14));
+                    SBACK(wb);
+                    break;
+                case 3:
+                    INST_NAME("RCR Ed, 1");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    MOVW(x2, 1);
+                    wb = ed;
+                    if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
+                    CALL_(rcr32, ed, (1<<x14));
+                    SBACK(wb);
+                    break;
+                case 4:
+                case 6:
+                    INST_NAME("SHL Ed, 1");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                    GETEDO2(x14);
+                    emit_shl32c(dyn, ninst, ed, 1, x3, x14);
+                    WBACK;
+                    break;
+                case 5:
+                    INST_NAME("SHR Ed, 1");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                    GETEDO2(x14);
+                    emit_shr32c(dyn, ninst, ed, 1, x3, x14);
+                    WBACK;
+                    break;
+                case 7:
+                    INST_NAME("SAR Ed, 1");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                    GETEDO2(x14);
+                    emit_sar32c(dyn, ninst, ed, 1, x3, x14);
+                    WBACK;
+                    break;
+            }
+            break;
+        case 0xD3:
+            nextop = F8;
+            grab_tlsdata(dyn, addr, ninst, x14);
+            switch((nextop>>3)&7) {
+                case 0:
+                    INST_NAME("ROL Ed, CL");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    AND_IMM8(x3, xECX, 0x1f);
+                    TSTS_REG_LSL_IMM5(x3, x3, 0);
+                    B_MARK2(cEQ);
+                    RSB_IMM8(x3, x3, 0x20);
+                    GETEDO2(x14);
+                    MOV_REG_ROR_REG(ed, ed, x3);
+                    WBACK;
+                    UFLAG_IF {  // calculate flags directly
+                        CMPS_IMM8(x3, 31);
+                        B_MARK(cNE);
+                            MOV_REG_LSR_IMM5(x1, ed, 31);
+                            ADD_REG_LSL_IMM5(x1, x1, ed, 0);
+                            BFI(xFlags, x1, F_OF, 1);
+                        MARK;
+                        BFI(xFlags, ed, F_CF, 1);
+                        UFLAG_DF(x2, d_none);
+                    }
+                    MARK2;
+                    break;
+                case 1:
+                    INST_NAME("ROR Ed, CL");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    AND_IMM8(x3, xECX, 0x1f);
+                    TSTS_REG_LSL_IMM5(x3, x3, 0);
+                    B_MARK2(cEQ);
+                    GETEDO2(x14);
+                    MOV_REG_ROR_REG(ed, ed, x3);
+                    WBACK;
+                    UFLAG_IF {  // calculate flags directly
+                        CMPS_IMM8(x3, 1);
+                        B_MARK(cNE);
+                            MOV_REG_LSR_IMM5(x2, ed, 30); // x2 = d>>30
+                            XOR_REG_LSR_IMM8(x2, x2, x2, 1); // x2 = ((d>>30) ^ ((d>>30)>>1))
+                            BFI(xFlags, x2, F_OF, 1);
+                        MARK;
+                        MOV_REG_LSR_IMM5(x2, ed, 31);
+                        BFI(xFlags, x2, F_CF, 1);
+                        UFLAG_DF(x2, d_none);
+                    }
+                    MARK2;
+                    break;
+                case 2:
+                    INST_NAME("RCL Ed, CL");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    AND_IMM8(x2, xECX, 0x1f);
+                    wb = ed;
+                    if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
+                    CALL_(rcl32, ed, (1<<x14));
+                    SBACK(wb);
+                    break;
+                case 3:
+                    INST_NAME("RCR Ed, CL");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    AND_IMM8(x2, xECX, 0x1f);
+                    wb = ed;
+                    if(ed!=x1) {MOV_REG(x1, ed); wb = x1;}
+                    CALL_(rcr32, ed, (1<<x14));
+                    SBACK(wb);
+                    break;
+                case 4:
+                case 6:
+                    INST_NAME("SHL Ed, CL");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                    AND_IMM8(x3, xECX, 0x1f);
+                    GETEDO2(x14);
+                    emit_shl32(dyn, ninst, ed, x3, x3, x14);
+                    WBACK;
+                    break;
+                case 5:
+                    INST_NAME("SHR Ed, CL");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                    AND_IMM8(x3, xECX, 0x1f);
+                    GETEDO2(x14);
+                    emit_shr32(dyn, ninst, ed, x3, x3, x14);
+                    WBACK;
+                    break;
+                case 7:
+                    INST_NAME("SAR Ed, CL");
+                    SETFLAGS(X_ALL, SF_PENDING);
+                    AND_IMM8(x3, xECX, 0x1f);
+                    GETEDO2(x14);
+                    UFLAG_OP12(ed, x3);
+                    MOV_REG_ASR_REG(ed, ed, x3);
+                    WBACK;
+                    UFLAG_RES(ed);
+                    UFLAG_DF(x3, d_sar32);
+                    break;
+            }
+            break;
+
         case 0xE9:
         case 0xEB:
             INST_NAME("(ignored) GS:");
@@ -362,7 +649,7 @@ uintptr_t dynarecGS(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     if((nextop&0xC0)==0xC0) {   // reg
                         DEFAULT;
                     } 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);
                         LDR_REG_LSL_IMM5(x3, ed, x14, 0);
                         PUSH1(x3);
                     }

src/dynarec/dynarec_arm_66.c

@@ -33,17 +33,19 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
     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);
@@ -52,7 +54,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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);
@@ -61,7 +63,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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, x14);
@@ -88,7 +90,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
 
         case 0x09:
             INST_NAME("OR Ew, Gw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGW(x2);
             GETEW(x1);
@@ -97,7 +99,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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);
@@ -106,7 +108,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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, x14);
@@ -119,7 +121,7 @@ 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);
@@ -129,7 +131,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
         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);
@@ -139,7 +141,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
         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, x14);
@@ -149,7 +151,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
         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);
@@ -159,7 +161,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
         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);
@@ -169,16 +171,22 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
         case 0x1D:
             INST_NAME("SBB AX, Iw");
             READFLAGS(X_CF);
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
             i16 = F16;
             UXTH(x1, xEAX, 0);
             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);
@@ -187,7 +195,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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);
@@ -196,7 +204,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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, x14);
@@ -209,7 +217,7 @@ 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);
@@ -218,7 +226,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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);
@@ -227,7 +235,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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, x14);
@@ -236,7 +244,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
 
         case 0x31:
             INST_NAME("XOR Ew, Gw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGW(x2);
             GETEW(x1);
@@ -245,7 +253,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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);
@@ -254,7 +262,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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, x14);
@@ -263,7 +271,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
 
         case 0x39:
             INST_NAME("CMP Ew, Gw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGW(x2);
             GETEW(x1);
@@ -271,7 +279,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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);
@@ -279,7 +287,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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);
@@ -295,7 +303,7 @@ 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, x14);
@@ -310,7 +318,7 @@ 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, x14);
@@ -417,7 +425,7 @@ 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, x14);
@@ -425,7 +433,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     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, x14);
@@ -434,7 +442,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 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, x14);
@@ -443,7 +451,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 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, x14);
@@ -451,7 +459,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     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, x14);
@@ -459,7 +467,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     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, x14);
@@ -467,7 +475,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     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, x14);
@@ -475,7 +483,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     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) {
@@ -489,7 +497,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 
         case 0x85:
             INST_NAME("TEST Ew, Gw");
-            SETFLAGS(X_ALL, SF_SET);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETEW(x1);
             GETGW(x2);
@@ -500,15 +508,17 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             INST_NAME("XCHG Ew, Gw");
             nextop = F8;
             if((nextop&0xC0)==0xC0) {
-                GETGW(x14);
+                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 {
-                GETGW(x14);
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0);
+                GETGD;
+                UXTH(x14, gd, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, 0, LOCK_LOCK);
                 #if 0
                 MARKLOCK;
                 // do the swap with exclusive locking
@@ -535,8 +545,9 @@ 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);
+                if(lock) {DMB_ISH();}
             }
             break;
         case 0x8B:
@@ -549,7 +560,8 @@ 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);
+                if(lock) {DMB_ISH();}
                 LDRH_IMM8(x1, ed, fixedaddress);
                 BFI(gd, x1, 0, 16);
             }
@@ -563,7 +575,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 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);
             }
             break;
@@ -576,7 +588,7 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                 MOV32(x2, offsetof(x86emu_t, segs[(nextop&0x38)>>3]));
                 STRH_REG(ed, xEmu, x2);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 255, 0, 0, NULL);
                 LDRH_IMM8(x1, ed, fixedaddress);
                 MOV32(x2, offsetof(x86emu_t, segs[(nextop&0x38)>>3]));
                 STRH_REG(x1, xEmu, x2);
@@ -650,16 +662,16 @@ 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);
@@ -679,10 +691,11 @@ 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:
@@ -789,7 +802,7 @@ 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);
@@ -896,6 +909,10 @@ 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;
@@ -926,7 +943,8 @@ 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);
@@ -962,7 +980,8 @@ 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);
@@ -989,13 +1008,20 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             }
             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);
@@ -1009,13 +1035,14 @@ 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, 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));
@@ -1035,14 +1062,27 @@ 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));
@@ -1053,20 +1093,14 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             break;
         case 0xF8:
             INST_NAME("CLC");
-            i32 = dyn->state_flags; //TODO: find a better way to do this
             SETFLAGS(X_CF, SF_SUBSET);
-            if(i32!=SF_SET) {
-                SET_DFNONE(x1);
-            } 
+            SET_DFNONE(x1);
             BFC(xFlags, F_CF, 1);
             break;
         case 0xF9:
             INST_NAME("STC");
-            i32 = dyn->state_flags; //TODO: find a better way to do this
             SETFLAGS(X_CF, SF_SUBSET);
-            if(i32!=SF_SET) {
-                SET_DFNONE(x1);
-            }
+            SET_DFNONE(x1);
             MOVW(x1, 1);
             BFI(xFlags, x1, F_CF, 1);
             break;
@@ -1076,14 +1110,14 @@ uintptr_t dynarec66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             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, 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, x14);
                     EWBACK;

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;
+            DMB_ISH();
+            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);
+                    DMB_ISH();
+                    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) {
+                            DMB_ISH();
+                            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
+                        }
+                        DMB_ISH();
+                    }
+                    break;
+                default:
+                  DEFAULT;
+            }
+            DMB_ISH();
+            break;
+        case 0xFF:
+            nextop = F8;
+            DMB_ISH();
+            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) {
+                            DMB_ISH();
+                            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) {
+                            DMB_ISH();
+                            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;
+            }
+            DMB_ISH();
+            break;
+        default:
+            DEFAULT;
+    }
+
+    return addr;
+}

src/dynarec/dynarec_arm_67.c

@@ -19,6 +19,7 @@
 #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)
@@ -29,9 +30,11 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
     uint32_t u32;
     uint8_t gd, ed;
     int fixedaddress;
+    int cacheupd;
 
     MAYUSE(i32);
     MAYUSE(j32);
+    MAYUSE(cacheupd);
     
     nextop = F8;
 
@@ -132,20 +135,27 @@ uintptr_t dynarec67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             }
             break;
             
-        #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_next(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);    \
-                }   \
+        #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)");

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:
@@ -90,9 +102,13 @@ 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, x3);
             break;
@@ -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, 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 const double d_1   = 1.0;
 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 double d_0   = 0.0;
+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, x3);
-            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:
@@ -85,182 +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, x1);
-            MOV32(x2, (&d_1));
-            VLDR_64(v1, x2, 0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                VMOV_i_32(v1, 0b01110000);
+            } else {
+                VMOV_i_64(v1, 0b01110000);
+            }
+            // should set C1 to 0
             break;
         case 0xE9:
             INST_NAME("FLDL2T");
-            v1 = x87_do_push(dyn, ninst, x1);
-            MOV32(x2, (&d_l2t));
-            VLDR_64(v1, x2, 0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                MOV32(x2, (&f_l2t));
+                VLDR_32(v1, x2, 0);
+            } else {
+                MOV32(x2, (&d_l2t));
+                VLDR_64(v1, x2, 0);
+            }
+            // should set C1 to 0
             break;
         case 0xEA:     
             INST_NAME("FLDL2E");
-            v1 = x87_do_push(dyn, ninst, x1);
-            MOV32(x2, (&d_l2e));
-            VLDR_64(v1, x2, 0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                MOV32(x2, (&f_l2e));
+                VLDR_32(v1, x2, 0);
+            } else {
+                MOV32(x2, (&d_l2e));
+                VLDR_64(v1, x2, 0);
+            }
+            // should set C1 to 0
             break;
         case 0xEB:
             INST_NAME("FLDPI");
-            v1 = x87_do_push(dyn, ninst, x1);
-            MOV32(x2, (&d_pi));
-            VLDR_64(v1, x2, 0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                MOV32(x2, (&f_pi));
+                VLDR_32(v1, x2, 0);
+            } else {
+                MOV32(x2, (&d_pi));
+                VLDR_64(v1, x2, 0);
+            }
+            // should set C1 to 0
             break;
         case 0xEC:
             INST_NAME("FLDLG2");
-            v1 = x87_do_push(dyn, ninst, x1);
-            MOV32(x2, (&d_lg2));
-            VLDR_64(v1, x2, 0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                MOV32(x2, (&f_lg2));
+                VLDR_32(v1, x2, 0);
+            } else {
+                MOV32(x2, (&d_lg2));
+                VLDR_64(v1, x2, 0);
+            }
+            // should set C1 to 0
             break;
         case 0xED:
             INST_NAME("FLDLN2");
-            v1 = x87_do_push(dyn, ninst, x1);
-            MOV32(x2, (&d_ln2));
-            VLDR_64(v1, x2, 0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                MOV32(x2, (&f_ln2));
+                VLDR_32(v1, x2, 0);
+            } else {
+                MOV32(x2, (&d_ln2));
+                VLDR_64(v1, x2, 0);
+            }
+            // should set C1 to 0
             break;
         case 0xEE:
             INST_NAME("FLDZ");
-            v1 = x87_do_push(dyn, ninst, x1);
-            MOV32(x2, (&d_0));
-            VLDR_64(v1, x2, 0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                VMOV_8(v1/2, 0);  // float is *2...
+            } else {
+                VMOV_8(v1, 0);  // cannot use VMOV_i_F64
+            }
+            // 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);
+            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, x3);
-            MOV32(x2, (&d_1));
-            VLDR_64(v1, x2, 0);
+            INST_NAME("FPTAN");
+            v2 = x87_do_push(dyn, ninst, x1, 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);
+            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, x1, x2, x3);
+            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, x1, x2, x3);
+            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);
+            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;
 
 
@@ -289,45 +568,59 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("FLD ST0, float[ED]");
-                    v1 = x87_do_push(dyn, ninst, x1);
-                    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, 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, 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);
                     }
@@ -335,8 +628,9 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     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);
                     }
@@ -347,13 +641,12 @@ 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");
-                    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);
                     }
@@ -362,7 +655,7 @@ uintptr_t dynarecD9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     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

@@ -52,10 +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);
+            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);
-            VMOVcond_64(cNE, v1, v2);   // F_CF==1
+            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:
@@ -67,10 +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);
+            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);
-            VMOVcond_64(cNE, v1, v2);   // F_ZF==1
+            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:
@@ -82,10 +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);
+            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));
-            VMOVcond_64(cNE, v1, v2);   // F_CF==1 | F_ZF==1
+            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:
@@ -97,16 +109,24 @@ 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);
+            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);
-            VMOVcond_64(cNE, v1, v2);   // F_PF==1
+            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, x3);
             x87_do_pop(dyn, ninst, x3);
@@ -129,7 +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);
@@ -139,7 +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);
@@ -149,7 +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);
@@ -160,7 +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);
@@ -172,7 +192,7 @@ uintptr_t dynarecDA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     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);
@@ -182,7 +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);
@@ -192,7 +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);
@@ -202,7 +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

@@ -50,10 +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);
+            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);
-            VMOVcond_64(cEQ, v1, v2);   // F_CF==0
+            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:
@@ -65,10 +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);
+            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);
-            VMOVcond_64(cEQ, v1, v2);   // F_ZF==0
+            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:
@@ -80,10 +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);
+            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));
-            VMOVcond_64(cEQ, v1, v2);   // F_CF==0 & F_ZF==0
+            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:
@@ -95,10 +107,14 @@ 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);
+            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);
-            VMOVcond_64(cEQ, v1, v2);   // F_PF==0
+            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?
@@ -106,14 +122,14 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
         case 0xE2:
             INST_NAME("FNCLEX");
             LDRH_IMM8(x2, xEmu, offsetof(x86emu_t, sw));
-            MOVW(x1, 0);
-            BFI(x2, x1, 0, 8);  // IE .. PE, SF, ES
-            BFI(x2, x1, 15, 1); // B
+            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:
@@ -126,9 +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:  
@@ -141,9 +161,13 @@ 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;
 
@@ -159,11 +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, x1);
+                    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;
@@ -173,18 +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(x14);   // Get FPCSR reg to clear exceptions flags
-                    ORR_IMM8(x3, x14, 0b001, 6); // enable exceptions
+                    ORR_IMM8(x3, x14, 0b010, 9); // enable exceptions
                     BIC_IMM8(x3, x3, 0b10011111, 0);
                     VMSR(x3);
                     VCVT_S32_F64(s0, v1);
@@ -198,19 +222,19 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 2:
                     INST_NAME("FIST Ed, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                     u8 = x87_setround(dyn, ninst, x1, x2, x14); // x1 have the modified RPSCR reg
                     if((nextop&0xC0)==0xC0) {
                         ed = xEAX+(nextop&7);
                         wback = 0;
                     } else {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0);
-                        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
+                    ORR_IMM8(x3, x1, 0b010, 9); // enable exceptions
                     BIC_IMM8(x3, x3, 0b10011111, 0);
                     VMSR(x3);
                     VCVTR_S32_F64(s0, v1);
@@ -223,19 +247,19 @@ 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);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                     u8 = x87_setround(dyn, ninst, x1, x2, x14); // x1 have the modified RPSCR reg
                     if((nextop&0xC0)==0xC0) {
                         ed = xEAX+(nextop&7);
                         wback = 0;
                     } else {
-                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0, 0, 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
+                    ORR_IMM8(x3, x1, 0b010, 9); // enable exceptions
                     BIC_IMM8(x3, x3, 0b10011111, 0);
                     VMSR(x3);
                     VCVTR_S32_F64(s0, v1);
@@ -249,7 +273,7 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 5:
                     INST_NAME("FLD tbyte");
-                    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(PK(0)==0xDB && ((PK(1)>>3)&7)==7) {
                         // the FLD is immediatly followed by an FSTP
                         LDR_IMM9(x2, ed, 0);
@@ -259,26 +283,110 @@ uintptr_t dynarecDB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                         MESSAGE(LOG_DUMP, "\tHack: FSTP tbyte\n");
                         nextop = F8;    //0xDB
                         nextop = F8;    //modrm
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                        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 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
                     }
                     break;
                 case 7:
                     INST_NAME("FSTP tbyte");
+                    #if 0
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
                     x87_forget(dyn, ninst, x1, x3, 0);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                     if(ed!=x1) {
                         MOV_REG(x1, ed);
                     }
                     CALL(arm_fstp, -1, 0);
+                    #else
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                    v2 = fpu_get_scratch_double(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
+                    // get top in x3
+                    LDR_IMM9(x3, xEmu, offsetof(x86emu_t, top));
+                    int a = -dyn->n.x87stack;
+                    if(a) {
+                        if(a<0) {
+                            SUB_IMM8(x3, x3, -a);
+                        } else {
+                            ADD_IMM8(x3, x3, a);
+                        }
+                        AND_IMM8(x3, x3, 7);
+                    }
+                    // get STld(0)
+                    ADD_IMM8(x2, xEmu, offsetof(x86emu_t, fpu_ld)+offsetof(fpu_ld_t, ld));
+                    MOVW(x14, sizeof(fpu_ld_t));
+                    MLA(x2, x3, x14, x2);
+                    // compare uref with actual value, top is not used anymore, so x3 is free
+                    LDR_IMM9(x14, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld));
+                    LDR_IMM9(x3, x2, offsetof(fpu_ld_t, uref)-offsetof(fpu_ld_t, ld)+4);
+                    VMOVtoV_D(v2, x14, x3);
+                    VCEQ_32(v2, v2, v1);    // same?
+                    VMOVfrV_D(x14, x3, v2);
+                    TSTS_REG_LSL_IMM5(x14, x3, 0);
+                    B_MARK(cEQ);    // different, do the conversion
+                    // same... copy STld(0).ld
+                    LDR_IMM9(x14, x2, 0);
+                    LDR_IMM9(x3, x2, 4);
+                    STR_IMM9(x14, ed, 0);
+                    STR_IMM9(x3, ed, 4);
+                    LDRH_IMM8(x14, x2, 8);
+                    STRH_IMM8(x14, ed, 8);
+                    B_MARK2(c__);
+                    MARK;
+                    // different, call D2LD
+                    SUB_IMM8(xSP, xSP, 8);
+                    VSTR_64(v1, xSP, 0);
+                    PUSH(xSP, (1<<xEmu)|(1<<xFlags));
+                    ADD_IMM8(0, xSP, 8);    //1st arg is &d
+                    if(ed!=x1) {
+                        MOV_REG_LSL_IMM5(x1, ed, 0);    //2nd arg is ed
+                    }
+                    CALL_S(D2LD, -2, 0);
+                    POP(xSP, (1<<xEmu)|(1<<xFlags));
+                    ADD_IMM8(xSP, xSP, 8);
+                    MARK2;
+                    #endif
                     x87_do_pop(dyn, ninst, x3);
                     break;
                 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:
@@ -86,9 +98,13 @@ 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, x3);
             break;
@@ -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);
@@ -197,14 +229,14 @@ 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);
@@ -215,14 +247,14 @@ uintptr_t dynarecDC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     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:
@@ -55,9 +59,13 @@ 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");
@@ -71,9 +79,8 @@ 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);
+            // copy the cache value for st0 to stx
+            x87_swapreg(dyn, ninst, x1, x2, 0, nextop&7);
             x87_do_pop(dyn, ninst, x3);
             break;
 
@@ -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:
@@ -100,9 +111,13 @@ 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, x3);
             break;
@@ -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, x1);
-                    #if 0
-                    // can bus error...
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3);
-                    VLDR_64(v1, ed, fixedaddress);
-                    #else
+                    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);
+                        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);
+                        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_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);
@@ -180,13 +189,13 @@ 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);
@@ -195,28 +204,34 @@ uintptr_t dynarecDD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     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");
-                    fpu_purgecache(dyn, ninst, x1, x2, x3);
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x14, &fixedaddress, 0, 0);
-                    LDR_IMM9(x1, xEmu, offsetof(x86emu_t, top));
-                    MOVW(x2, offsetof(x86emu_t, sw));
-                    ADD_REG_LSL_IMM5(x2, xEmu, x2, 0);
-                    LDRH_IMM8(x3, x2, 0);
-                    BFI(x3, x1, 11, 3); // inject TOP at bit 11 (3 bits)
-                    STRH_IMM8(x3, ed, 0);   // store whole sw flags
+                    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"
@@ -40,9 +41,13 @@ 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);
+            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:
@@ -54,9 +59,13 @@ 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);
+            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 0xD0:
@@ -68,18 +77,26 @@ uintptr_t dynarecDE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
         case 0xD6:
         case 0xD7:
             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, x3);
             break;
 
         case 0xD9: 
             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);
+            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);
@@ -93,9 +110,13 @@ 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);
+            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:
@@ -107,9 +128,13 @@ 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);
+            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:
@@ -121,9 +146,13 @@ 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);
+            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:
@@ -135,9 +164,29 @@ 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);
+            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;
 

src/dynarec/dynarec_arm_df.c

@@ -23,7 +23,7 @@
 #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;
@@ -31,13 +31,15 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
     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(v0);
     MAYUSE(j32);
 
     switch(nextop) {
@@ -58,6 +60,15 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             INST_NAME("FNSTSW AX");
             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
             BFI(xEAX, x1, 0, 16);
             break;
@@ -71,9 +82,14 @@ 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, x3);
             break;
@@ -87,9 +103,14 @@ 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, x3);
             break;
@@ -140,26 +161,35 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("FILD ST0, Ew");
-                    v1 = x87_do_push(dyn, ninst, x1);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0);
+                    v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, 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);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0, 0, NULL);
                     ed = x1;
                     s0 = fpu_get_scratch_single(dyn);
                     MSR_nzcvq_0();
                     VMRS(x14);               // get fpscr
                     // x1 already have FPCSR reg to clear exceptions flags
-                    ORR_IMM8(x3, x14, 0b001, 6); // enable exceptions
+                    ORR_IMM8(x3, x14, 0b010, 9); // enable exceptions
                     BIC_IMM8(x3, x3, 0b10011111, 0);
                     VMSR(x3);
-                    VCVTR_S32_F64(s0, v1);
+                    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);
                     TSTS_IMM8_ROR(x3, 0b00000001, 0);
@@ -172,17 +202,21 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 2:
                     INST_NAME("FIST Ew, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
                     u8 = x87_setround(dyn, ninst, x1, x2, x14);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0, 0, NULL);
                     ed = x1;
                     s0 = fpu_get_scratch_single(dyn);
                     MSR_nzcvq_0();
                     // x1 already have FPCSR reg to clear exceptions flags
-                    ORR_IMM8(x3, x1, 0b001, 6); // enable exceptions
+                    ORR_IMM8(x3, x1, 0b010, 9); // enable exceptions
                     BIC_IMM8(x3, x3, 0b10011111, 0);
                     VMSR(x3);
-                    VCVTR_S32_F64(s0, v1);
+                    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);
                     TSTS_IMM8_ROR(x3, 0b00000001, 0);
@@ -194,17 +228,21 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     break;
                 case 3:
                     INST_NAME("FISTP Ew, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
                     u8 = x87_setround(dyn, ninst, x1, x2, x14);
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 255, 0, 0, NULL);
                     ed = x1;
                     s0 = fpu_get_scratch_single(dyn);
                     MSR_nzcvq_0();
                     // x1 already have FPCSR reg to clear exceptions flags
-                    ORR_IMM8(x3, x1, 0b001, 6); // enable exceptions
+                    ORR_IMM8(x3, x1, 0b010, 9); // enable exceptions
                     BIC_IMM8(x3, x3, 0b10011111, 0);
                     VMSR(x3);
-                    VCVTR_S32_F64(s0, v1);
+                    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);
                     TSTS_IMM8_ROR(x3, 0b00000001, 0);
@@ -217,32 +255,199 @@ uintptr_t dynarecDF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst,
                     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, 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);
+                    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:

src/dynarec/dynarec_arm_emit_logic.c

@@ -27,8 +27,6 @@
 void emit_or32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 {
     IFX(X_PEND) {
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s4, d_or32);
     } else IFX(X_ALL) {
         SET_DFNONE(s4);
@@ -63,9 +61,6 @@ void emit_or32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 void emit_or32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
 {
     IFX(X_PEND) {
-        MOV32(s3, c);
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s4, d_or32);
     } else IFX(X_ALL) {
         SET_DFNONE(s4);
@@ -77,7 +72,7 @@ void emit_or32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4
             ORR_IMM8(s1, s1, c, 0);
         }
     } else {
-        IFX(X_PEND) {} else {MOV32(s3, c);}
+        MOV32(s3, c);
         IFX(X_ALL) {
             ORRS_REG_LSL_IMM5(s1, s1, s3, 0);
         } else {
@@ -109,8 +104,6 @@ void emit_or32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4
 void emit_xor32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 {
     IFX(X_PEND) {
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s4, d_xor32);
     } else IFX(X_ALL) {
         SET_DFNONE(s4);
@@ -145,9 +138,6 @@ 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)
 {
     IFX(X_PEND) {
-        MOV32(s3, c);
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s4, d_xor32);
     } else IFX(X_ALL) {
         SET_DFNONE(s4);
@@ -159,7 +149,7 @@ void emit_xor32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
             XOR_IMM8(s1, s1, c);
         }
     } else {
-        IFX(X_PEND) {} else {MOV32(s3, c);}
+        MOV32(s3, c);
         IFX(X_ALL) {
             XORS_REG_LSL_IMM5(s1, s1, s3, 0);
         } else {
@@ -191,8 +181,6 @@ void emit_xor32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
 void emit_and32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 {
     IFX(X_PEND) {
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s4, d_and32);
     } else IFX(X_ALL) {
         SET_DFNONE(s4);
@@ -227,9 +215,6 @@ 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)
 {
     IFX(X_PEND) {
-        MOV32(s3, c);
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s4, d_and32);
     } else IFX(X_ALL) {
         SET_DFNONE(s4);
@@ -248,7 +233,7 @@ void emit_and32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
                 BIC_IMM8(s1, s1, ~c, 0);
             }
         } else {
-            IFX(X_PEND) {} else {MOV32(s3, c);}
+            MOV32(s3, c);
             IFX(X_ALL) {
                 ANDS_REG_LSL_IMM5(s1, s1, s3, 0);
             } else {
@@ -281,8 +266,6 @@ void emit_and32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
 void emit_or8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 {
     IFX(X_PEND) {
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s3, d_or8);
     } else IFX(X_ALL) {
         SET_DFNONE(s3);
@@ -317,9 +300,6 @@ 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)
 {
     IFX(X_PEND) {
-        MOV32(s3, c&0xff);
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s4, d_or8);
     } else IFX(X_ALL) {
         SET_DFNONE(s4);
@@ -354,8 +334,6 @@ 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)
 {
     IFX(X_PEND) {
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s3, d_xor8);
     } else IFX(X_ALL) {
         SET_DFNONE(s3);
@@ -390,9 +368,6 @@ 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)
 {
     IFX(X_PEND) {
-        MOV32(s3, c&0xff);
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s4, d_xor8);
     } else IFX(X_ALL) {
         SET_DFNONE(s4);
@@ -427,8 +402,6 @@ 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)
 {
     IFX(X_PEND) {
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s3, d_and8);
     } else IFX(X_ALL) {
         SET_DFNONE(s3);
@@ -463,9 +436,6 @@ 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)
 {
     IFX(X_PEND) {
-        MOV32(s3, c&0xff);
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s4, d_and8);
     } else IFX(X_ALL) {
         SET_DFNONE(s4);
@@ -501,8 +471,6 @@ void emit_and8c(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)
 {
     IFX(X_PEND) {
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s3, d_or16);
     } else IFX(X_ALL) {
         SET_DFNONE(s3);
@@ -537,9 +505,6 @@ 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)
 {
     IFX(X_PEND) {
-        MOVW(s3, c);
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s4, d_or16);
     } else IFX(X_ALL) {
         SET_DFNONE(s4);
@@ -551,7 +516,7 @@ void emit_or16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4
             ORR_IMM8(s1, s1, c, 0);
         }
     } else {
-        IFX(X_PEND) {} else {MOVW(s3, c);}
+        MOVW(s3, c);
         IFX(X_ALL) {
             ORRS_REG_LSL_IMM5(s1, s1, s3, 0);
         } else {
@@ -583,8 +548,6 @@ 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)
 {
     IFX(X_PEND) {
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s3, d_xor16);
     } else IFX(X_ALL) {
         SET_DFNONE(s3);
@@ -619,9 +582,6 @@ 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)
 {
     IFX(X_PEND) {
-        MOVW(s3, c);
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s4, d_xor16);
     } else IFX(X_ALL) {
         SET_DFNONE(s4);
@@ -633,7 +593,7 @@ void emit_xor16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
             XOR_IMM8(s1, s1, c);
         }
     } else {
-        IFX(X_PEND) {} else {MOVW(s3, c);}
+        MOVW(s3, c);
         IFX(X_ALL) {
             XORS_REG_LSL_IMM5(s1, s1, s3, 0);
         } else {
@@ -665,8 +625,6 @@ void emit_xor16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
 void emit_and16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 {
     IFX(X_PEND) {
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s2, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s3, d_and16);
     } else IFX(X_ALL) {
         SET_DFNONE(s3);
@@ -701,9 +659,6 @@ 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)
 {
     IFX(X_PEND) {
-        MOVW(s3, c);
-        STR_IMM9(s1, xEmu, offsetof(x86emu_t, op1));
-        STR_IMM9(s3, xEmu, offsetof(x86emu_t, op2));
         SET_DF(s4, d_and16);
     } else IFX(X_ALL) {
         SET_DFNONE(s4);
@@ -715,7 +670,7 @@ void emit_and16c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
             AND_IMM8(s1, s1, c);
         }
     } else {
-        IFX(X_PEND) {} else {MOVW(s3, c);}
+        MOVW(s3, c);
         IFX(X_ALL) {
             ANDS_REG_LSL_IMM5(s1, s1, s3, 0);
         } else {

src/dynarec/dynarec_arm_emit_math.c

@@ -76,7 +76,7 @@ void emit_add32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 // emit ADD32 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
 void emit_add32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
 {
-    if(s1==xESP && (!dyn->insts || dyn->insts[ninst].x86.need_flags==X_PEND))
+    if(s1==xESP && (!dyn->insts || (dyn->insts[ninst].x86.gen_flags==X_PEND || dyn->insts[ninst].x86.gen_flags==X_ALL)))
     {
         // special case when doing math on ESP and only PEND is needed: ignoring it!
         if(c>=0 && c<256) {
@@ -210,7 +210,7 @@ void emit_sub32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 // emit SUB32 instruction, from s1 , constant c, store result in s1 using s3 and s4 as scratch
 void emit_sub32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
 {
-    if(s1==xESP && (!dyn->insts || dyn->insts[ninst].x86.need_flags==X_PEND))
+    if(s1==xESP && (!dyn->insts || (dyn->insts[ninst].x86.gen_flags==X_PEND || dyn->insts[ninst].x86.gen_flags==X_ALL)))
     {
         // special case when doing math on ESP and only PEND is needed: ignoring it!
         if(c>=0 && c<256) {

src/dynarec/dynarec_arm_emit_shift.c

@@ -282,6 +282,8 @@ void emit_sar32c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s
 // 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));
@@ -313,6 +315,8 @@ 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));
@@ -432,7 +436,7 @@ void emit_shld32c(dynarec_arm_t* dyn, int ninst, int s1, int s2, int32_t c, int
     }
     IFX(X_OF) {
         if(c==1) {
-            UBFX(s3, s2, 0, 1);
+            UBFX(s3, s1, (31-c), 1);
             XOR_IMM8_COND(cCS, s3, s3, 1);
             BFI(xFlags, s3, F_OF, 1);
         } else {
@@ -460,3 +464,195 @@ void emit_shld32c(dynarec_arm_t* dyn, int ninst, int s1, int s2, int32_t c, int
         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

@@ -326,8 +326,12 @@ void emit_test32(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
         AND_IMM8(s3, s3, 0xE0); // lsr 5 masking pre-applied
         MOV32(s4, GetParityTab());
         LDR_REG_LSR_IMM5(s4, s4, s3, 5-2);   // x/32 and then *4 because array is integer
-        AND_REG_LSL_IMM5(s3, s1, s2, 0);
-        AND_IMM8(s3, s3, 31);
+        if(s1==s2) {
+            AND_IMM8(s3, s1, 31);
+        } else {
+            AND_REG_LSL_IMM5(s3, s1, s2, 0);
+            AND_IMM8(s3, s3, 31);
+        }
         MVN_REG_LSR_REG(s4, s4, s3);
         BFI(xFlags, s4, F_PF, 1);
     }
@@ -370,7 +374,7 @@ void emit_test16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
     }
 }
 
-// emit TEST8 instruction, from test s1 , s2, using s3 and s4 as scratch
+// emit TEST8 instruction, from test s1 , s2, using s3 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)
 {
     IFX_PENDOR0 {
@@ -397,10 +401,9 @@ void emit_test8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
     }
     // PF: (((emu->x86emu_parity_tab[(res) / 32] >> ((res) % 32)) & 1) == 0)
     IFX(X_PF) {
-        AND_IMM8(s3, s3, 0xE0); // lsr 5 masking pre-applied
+        AND_IMM8(s2, s3, 0xE0); // lsr 5 masking pre-applied
         MOV32(s4, GetParityTab());
-        LDR_REG_LSR_IMM5(s4, s4, s3, 5-2);   // x/32 and then *4 because array is integer
-        AND_REG_LSL_IMM5(s3, s1, s2, 0);
+        LDR_REG_LSR_IMM5(s4, s4, s2, 5-2);   // x/32 and then *4 because array is integer
         AND_IMM8(s3, s3, 31);
         MVN_REG_LSR_REG(s4, s4, s3);
         BFI(xFlags, s4, F_PF, 1);

src/dynarec/dynarec_arm_f20f.c

@@ -23,23 +23,27 @@
 #include "dynarec_arm_helper.h"
 
 // Get Ex as a double, not a quad (warning, x2 and x3 may get used)
-#define GETEX(a) \
-    if((nextop&0xC0)==0xC0) { \
-        a = sse_get_reg(dyn, ninst, x1, nextop&7); \
+#define GETEX(a, w)                                 \
+    if((nextop&0xC0)==0xC0) {                       \
+        a = sse_get_reg(dyn, ninst, x1, nextop&7,w);\
     } else {    \
         parity = getedparity(dyn, ninst, addr, nextop, 3);  \
         a = fpu_get_scratch_double(dyn);            \
         if(parity) {                                \
-            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3); \
+            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, 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;
@@ -64,12 +68,12 @@ uintptr_t dynarecF20F(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);
-                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 {
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
-                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(v0, x2, x3);
@@ -80,13 +84,13 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             INST_NAME("MOVSD Ex, Gx");
             nextop = F8;
             gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
             if((nextop&0xC0)==0xC0) {
-                d0 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                d0 = sse_get_reg(dyn, ninst, x1, nextop&7, 1);
                 VMOVD(d0, v0);
             } else {
                 VMOVfrV_D(x2, x3, v0);
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-4, 0, 0, NULL);
                 STR_IMM9(x2, ed, fixedaddress+0);
                 STR_IMM9(x3, ed, fixedaddress+4);
             }
@@ -96,12 +100,12 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             nextop = F8;
             gd = (nextop&0x38)>>3;
             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);
             } else {
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
-                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(v0, x2, x3);
@@ -113,7 +117,7 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             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);
@@ -126,9 +130,9 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             GETGD;
             s0 = fpu_get_scratch_single(dyn);
             if((nextop&0xC0)==0xC0) {
-                d0 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                d0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                 d0 = fpu_get_scratch_double(dyn);
                 VLDR_64(d0, ed, fixedaddress);
             }
@@ -142,9 +146,9 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             GETGD;
             s0 = fpu_get_scratch_single(dyn);
             if((nextop&0xC0)==0xC0) {
-                d0 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                d0 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3, 0, NULL);
                 d0 = fpu_get_scratch_double(dyn);
                 VLDR_64(d0, ed, fixedaddress);
             }
@@ -157,55 +161,105 @@ 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();
@@ -215,16 +269,28 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             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();
@@ -235,11 +301,11 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             INST_NAME("PSHUFLW Gx, Ex, Ib");
             nextop = F8;
             gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
             if((nextop&0xC0)==0xC0) {
-                v1 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 v1 = fpu_get_scratch_quad(dyn);
                 VLD1Q_64(v1, ed);
             }
@@ -270,11 +336,11 @@ uintptr_t dynarecF20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             INST_NAME("HADDPS Gx, Ex");
             nextop = F8;
             gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
             if((nextop&0xC0)==0xC0) {
-                v1 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 v1 = fpu_get_scratch_quad(dyn);
                 VLD1Q_64(v1, ed);
             }
@@ -286,44 +352,66 @@ 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;
-            if((u8&7)==6){
-                VCMP_F64(d0, v0);
-            } else {
-                VCMP_F64(v0, d0);
-            }
+            VCMP_F64(v0, d0);
             VMRS_APSR();
             MOVW(x2, 0);
             switch(u8&7) {
                 case 0: MVN_COND_REG_LSL_IMM5(cEQ, x2, x2, 0); break;   // Equal
-                case 1: MVN_COND_REG_LSL_IMM5(cMI, x2, x2, 0); break;   // Less than
-                case 2: MVN_COND_REG_LSL_IMM5(cLE, x2, x2, 0); break;   // Less or 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 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(cLT, x2, x2, 0); break;   // Greater or unordered, test inverted, N!=V so unordereded or less than (inverted)
+                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 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);
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 VMOVQ(v0, v1);
             } else {
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-12, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-12, 0, 0, NULL);
                 LDR_IMM9(x2, ed, fixedaddress+0);
                 LDR_IMM9(x3, ed, fixedaddress+4);
                 VMOVtoV_D(v0, x2, x3);

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 {                                      \
@@ -38,7 +38,7 @@
         a = b*2;                                    \
         parity = getedparity(dyn, ninst, addr, nextop, 2);  \
         if(parity) {                                \
-            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 0); \
+            addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 0, 0, NULL); \
             VLDR_32(a, ed, fixedaddress);           \
         } else {                                    \
             GETED;                                  \
@@ -46,9 +46,12 @@
         }                                           \
     }
 
-#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)
@@ -78,8 +81,8 @@ 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 {
@@ -98,19 +101,26 @@ 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);
+                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 {
@@ -128,9 +138,15 @@ 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);
+                }
             }
             break;
         case 0x12:
@@ -138,9 +154,9 @@ 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) {
-                q1 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                q1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 q1 = fpu_get_scratch_quad(dyn);
                 VLD1Q_64(q1, ed);
             }
@@ -154,9 +170,9 @@ 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) {
-                q1 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                q1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 q1 = fpu_get_scratch_quad(dyn);
                 VLD1Q_64(q1, ed);
             }
@@ -181,7 +197,7 @@ 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;
@@ -203,7 +219,7 @@ 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 {
@@ -214,10 +230,10 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             } else {
                 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);
                     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);
                 }
@@ -232,7 +248,7 @@ 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 {
@@ -243,10 +259,10 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             } else {
                 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);
                     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);
                 }
@@ -259,7 +275,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
         case 0x51:
             INST_NAME("SQRTSS Gx, Ex");
             nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
             GETEX(s1, v1);
             if(v0==v1)
                 d0 = s1/2;
@@ -279,7 +295,7 @@ 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;
@@ -307,7 +323,7 @@ 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;
@@ -330,7 +346,7 @@ 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;
@@ -350,7 +366,7 @@ 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;
@@ -370,7 +386,7 @@ 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;
@@ -378,9 +394,9 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             INST_NAME("CVTTPS2DQ Gx, Ex");
             nextop = F8;
             if((nextop&0xC0)==0xC0) {
-                v1 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 v1 = fpu_get_scratch_quad(dyn);
                 VLD1Q_64(v1, ed);
             }
@@ -391,7 +407,7 @@ 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;
@@ -411,7 +427,7 @@ 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;
@@ -434,7 +450,7 @@ 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;
@@ -454,7 +470,7 @@ 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;
@@ -480,12 +496,12 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             nextop = F8;
             gd = (nextop&0x38)>>3;
             if((nextop&0xC0)==0xC0) {
-                v1 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 VMOVQ(v0, v1);
             } else {
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-12, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 4095-12, 0, 0, NULL);
                 LDR_IMM9(x2, ed, fixedaddress+0);
                 LDR_IMM9(x3, ed, fixedaddress+4);
                 VMOVtoV_D(v0, x2, x3);
@@ -498,11 +514,11 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             INST_NAME("PSHUFHW Gx, Ex, Ib");
             nextop = F8;
             gd = (nextop&0x38)>>3;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
             if((nextop&0xC0)==0xC0) {
-                v1 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 v1 = fpu_get_scratch_quad(dyn);
                 VLD1Q_64(v1, ed);
             }
@@ -534,12 +550,12 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             nextop = F8;
             gd = (nextop&0x38)>>3;
             if((nextop&0xC0)==0xC0) {
-                v1 = sse_get_reg(dyn, ninst, x1, nextop&7);
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 VMOVD(v0, v1);
             } else {
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
-                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, 0, NULL);
                 VLD1_64(v0, ed);
             }
             VEOR(v0+1, v0+1, v0+1);
@@ -548,13 +564,12 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
             INST_NAME("MOVDQU Ex,Gx");
             nextop = F8;
             gd = (nextop&0x38)>>3;
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
             if((nextop&0xC0)==0xC0) {
-                v0 = sse_get_reg(dyn, ninst, x1, gd);
                 v1 = sse_get_reg_empty(dyn, ninst, x1, nextop&7);
                 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);
@@ -566,11 +581,11 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
 
         case 0xBC:
             INST_NAME("TZCNT Gd, Ed");
-            SETFLAGS(X_CF|X_ZF, SF_SET);
+            SETFLAGS(X_CF|X_ZF, SF_SUBSET);
+            SET_DFNONE(x3);
             nextop = F8;
             GETED;
             GETGD;
-            SET_DFNONE(x3);
             TSTS_REG_LSL_IMM5(ed, ed, 0);
             MOVW_COND(cEQ, x3, 1);
             MOVW_COND(cNE, x3, 0);
@@ -586,7 +601,7 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
         case 0xC2:
             INST_NAME("CMPSS Gx, Ex");
             nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
             GETEX(s1, v1);
             if(v0==v1)
                 d0 = s1/2;
@@ -599,31 +614,49 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 }
             s0 = d0*2;
             u8 = F8;
-            if((u8&7)==6){
-                VCMP_F32(s1, s0);
-            } else {
-                VCMP_F32(s0, s1);
-            }
+            VCMP_F32(s0, s1);
             VMRS_APSR();
             MOVW(x2, 0);
             switch(u8&7) {
                 case 0: MVN_COND_REG_LSL_IMM5(cEQ, x2, x2, 0); break;   // Equal
-                case 1: MVN_COND_REG_LSL_IMM5(cMI, x2, x2, 0); break;   // Less than
-                case 2: MVN_COND_REG_LSL_IMM5(cLE, x2, x2, 0); break;   // Less or 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 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(cLT, x2, x2, 0); break;   // Greater or unordered, test inverted, N!=V so unordereded or less than (inverted)
+                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);
+                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);
+                v1 = sse_get_reg(dyn, ninst, x1, nextop&7, 0);
                 if(v1<16)
                     v0 = v1;
                 else {
@@ -634,10 +667,10 @@ uintptr_t dynarecF30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nins
                 v0 = fpu_get_scratch_double(dyn);
                 parity = getedparity(dyn, ninst, addr, nextop, 3);
                 if(parity) {
-                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 1023, 3);
+                    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);
+                    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);

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"
@@ -25,10 +26,11 @@
 #include "dynarec_arm.h"
 #include "dynarec_arm_private.h"
 #include "dynarec_arm_functions.h"
+#include "bridge.h"
 
 void arm_fstp(x86emu_t* emu, void* p)
 {
-    if(ST0.q!=STld(0).ref)
+    if(ST0.q!=STld(0).uref)
         D2LD(&ST0.d, p);
     else
         memcpy(p, &STld(0).ld, 10);
@@ -50,6 +52,7 @@ 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)
 {
@@ -64,20 +67,12 @@ void arm_fxtract(x86emu_t* emu)
 }
 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)
 {
@@ -85,17 +80,9 @@ 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);
@@ -106,8 +93,8 @@ void arm_fild64(x86emu_t* emu, int64_t* ed)
     int64_t tmp;
     memcpy(&tmp, ed, sizeof(tmp));
     ST0.d = tmp;
-    STll(0).ll = tmp;
-    STll(0).ref = ST0.q;
+    STll(0).sq = tmp;
+    STll(0).sref = ST0.sq;
 }
 
 void arm_fbstp(x86emu_t* emu, uint8_t* ed)
@@ -118,8 +105,8 @@ void arm_fbstp(x86emu_t* emu, uint8_t* ed)
 void arm_fistp64(x86emu_t* emu, int64_t* ed)
 {
     // used of memcpy to avoid aligments issues
-    if(STll(0).ref==ST(0).q) {
-        memcpy(ed, &STll(0).ll, sizeof(int64_t));
+    if(STll(0).sref==ST(0).sq) {
+        memcpy(ed, &STll(0).sq, sizeof(int64_t));
     } else {
         int64_t tmp;
         if(isgreater(ST0.d, (double)(int64_t)0x7fffffffffffffffLL) || isless(ST0.d, (double)(int64_t)0x8000000000000000LL) || !isfinite(ST0.d))
@@ -130,6 +117,43 @@ void arm_fistp64(x86emu_t* emu, int64_t* ed)
     }
 }
 
+int64_t arm_fist64_0(double d)
+{
+    int64_t tmp;
+    if(isgreater(d, (double)(int64_t)0x7fffffffffffffffLL) || isless(d, (double)(int64_t)0x8000000000000000LL) || !isfinite(d))
+        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)(int64_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)(int64_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)(int64_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
@@ -140,8 +164,8 @@ void arm_fistt64(x86emu_t* emu, int64_t* ed)
 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.q;
+    LD2D(&STld(0).ld, &ST(0).d);
+    STld(0).uref = ST0.q;
 }
 
 void arm_ud(x86emu_t* emu)
@@ -156,7 +180,7 @@ void arm_fsave(x86emu_t* emu, uint8_t* ed)
     uint8_t* p = ed;
     p += 28;
     for (int i=0; i<8; ++i) {
-        LD2D(p, &ST(i).d);
+        D2LD(&ST(i).d, p);
         p+=10;
     }
 }
@@ -167,7 +191,7 @@ void arm_frstor(x86emu_t* emu, uint8_t* ed)
     uint8_t* p = ed;
     p += 28;
     for (int i=0; i<8; ++i) {
-        D2LD(&ST(i).d, p);
+        LD2D(p, &ST(i).d);
         p+=10;
     }
 
@@ -176,58 +200,62 @@ void arm_frstor(x86emu_t* emu, uint8_t* ed)
 void arm_fprem1(x86emu_t* emu)
 {
     // simplified version
-    int32_t tmp32s = round(ST0.d / ST1.d);
-    ST0.d -= ST1.d*tmp32s;
+    int64_t tmp64s = lrint(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);
+    emu->sw.f.F87_C1 = (tmp64s&1);
+    emu->sw.f.F87_C3 = ((tmp64s>>1)&1);
+    emu->sw.f.F87_C0 = ((tmp64s>>2)&1);
 }
 
 
 // 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)
 {
-    if(dyn->fpu_scratch>4) {
-        if(dyn->fpu_extra_qscratch) {
+    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->fpu_extra_qscratch = fpu_get_reg_quad(dyn);
-        return dyn->fpu_extra_qscratch;
+            dyn->n.fpu_extra_qscratch = fpu_get_reg_quad(dyn, NEON_CACHE_SCR, 0);
+        return dyn->n.fpu_extra_qscratch;
     }
-    int i = (dyn->fpu_scratch+3)&(~3);
-    dyn->fpu_scratch = i+4;
+    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;
-    if(dyn->fpu_extra_qscratch) {
-        fpu_free_reg_quad(dyn, dyn->fpu_extra_qscratch);
-        dyn->fpu_extra_qscratch = 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
@@ -235,28 +263,370 @@ 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->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++)
@@ -336,7 +706,7 @@ int isNativeCall(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t* calladdress, int
         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 = *(uint32_t*)a1; 
+        addr = (uintptr_t)getAlternate(*(void**)a1); 
     }
     if(addr<0x10000)    // too low, that is suspicious
         return 0;
@@ -352,3 +722,77 @@ int isNativeCall(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t* calladdress, int
 #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;
+}
+
+// is inst clean for a son branch?
+int isInstClean(dynarec_arm_t* dyn, int ninst)
+{
+    // check flags cache
+    if(dyn->insts[ninst].f_entry.dfnone || dyn->insts[ninst].f_entry.pending)
+        return 0;
+    if(dyn->insts[ninst].x86.state_flags)
+        return 0;
+    // check neoncache
+    neoncache_t* n = &dyn->insts[ninst].n;
+    if(n->news || n->stack || n->stack_next)
+        return 0;
+    for(int i=0; i<24; ++i)
+        if(n->neoncache[i].v)
+            return 0;
+    return 1;
+}
+
+static int flagsCacheNeedsTransform(dynarec_arm_t* dyn, int ninst) {
+    int jmp = dyn->insts[ninst].x86.jmp_insts;
+    if(jmp<0)
+        return 0;
+    if(dyn->insts[ninst].f_exit.dfnone)  // flags are fully known, nothing we can do more
+        return 0;
+/*    if((dyn->f.pending!=SF_SET)
+    && (dyn->f.pending!=SF_SET_PENDING)) {
+        if(dyn->f.pending!=SF_PENDING) {*/
+    switch (dyn->insts[jmp].f_entry.pending) {
+        case SF_UNKNOWN: return 0;
+        case SF_SET: 
+            if(dyn->insts[ninst].f_exit.pending!=SF_SET && dyn->insts[ninst].f_exit.pending!=SF_SET_PENDING) 
+                return 1; 
+            else 
+                return 0;
+        case SF_SET_PENDING:
+            if(dyn->insts[ninst].f_exit.pending!=SF_SET 
+            && dyn->insts[ninst].f_exit.pending!=SF_SET_PENDING
+            && dyn->insts[ninst].f_exit.pending!=SF_PENDING) 
+                return 1; 
+            else 
+                return 0;
+        case SF_PENDING:
+            if(dyn->insts[ninst].f_exit.pending!=SF_SET 
+            && dyn->insts[ninst].f_exit.pending!=SF_SET_PENDING
+            && dyn->insts[ninst].f_exit.pending!=SF_PENDING)
+                return 1;
+            else
+                return (dyn->insts[jmp].f_entry.dfnone  == dyn->insts[ninst].f_exit.dfnone)?0:1;
+    }
+    if(dyn->insts[jmp].f_entry.dfnone && !dyn->insts[ninst].f_exit.dfnone)
+        return 1;
+    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;
+}

src/dynarec/dynarec_arm_functions.h

@@ -8,20 +8,14 @@ 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);
@@ -29,8 +23,15 @@ void arm_fsave(x86emu_t* emu, uint8_t* ed);
 void arm_frstor(x86emu_t* emu, uint8_t* ed);
 void arm_fprem1(x86emu_t* emu);
 
+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);
 
+#define FPUFIRST    8
+
 // Get an FPU single scratch reg
 int fpu_get_scratch_single(dynarec_arm_t* dyn);
 // Get an FPU double scratch reg
@@ -39,17 +40,44 @@ 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);
+
+// is inst clean for a son branch?
+int isInstClean(dynarec_arm_t* dyn, int ninst);
+
+// predecessor access
+int isPred(dynarec_arm_t* dyn, int ninst, int pred);
+int getNominalPred(dynarec_arm_t* dyn, int ninst);
+
 // 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...
@@ -58,4 +86,6 @@ 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);
+
 #endif //__DYNAREC_ARM_FUNCTIONS_H__

src/dynarec/dynarec_arm_helper.h

@@ -22,9 +22,13 @@
 #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)
+
+//LOCK_* define
+#define LOCK_LOCK   (int*)1
 
 // GETGD    get x86 register in gd
 #define GETGD   gd = xEAX+((nextop&0x38)>>3)
@@ -33,7 +37,7 @@
                     ed = xEAX+(nextop&7);   \
                     wback = 0;              \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 4095, 0, 0, NULL); \
                     LDR_IMM9(x1, wback, fixedaddress); \
                     ed = x1;                \
                 }
@@ -42,7 +46,7 @@
                     ed = xEAX+(nextop&7);   \
                     wback = 0;              \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, (hint==x2)?x1:x2, &fixedaddress, 4095, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, (hint==x2)?x1:x2, &fixedaddress, 4095, 0, 0, NULL); \
                     LDR_IMM9(hint, wback, fixedaddress); \
                     ed = hint;              \
                 }
@@ -52,7 +56,7 @@
                     MOV_REG(ret, ed);       \
                     wback = 0;              \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, hint, &fixedaddress, 4095, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, hint, &fixedaddress, 4095, 0, 0, NULL); \
                     ed = ret;               \
                     LDR_IMM9(ed, wback, fixedaddress); \
                 }
@@ -65,7 +69,7 @@
                     ed = xEAX+(nextop&7);   \
                     wback = 0;              \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, 0, NULL); \
                     LDR_REG_LSL_IMM5(x1, wback, O, 0);  \
                     ed = x1;                 \
                 }
@@ -74,8 +78,9 @@
                     ed = xEAX+(nextop&7);               \
                     wback = 0;                          \
                 } else {                                \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0); \
-                    ADD_REG_LSL_IMM5(wback, wback, O, 0);\
+                    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;                            \
                 }
@@ -93,7 +98,7 @@
                     ed = i;                 \
                     wb1 = 0;                \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, w, &fixedaddress, 255, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, w, &fixedaddress, 255, 0, 0, NULL); \
                     LDRH_IMM8(i, wback, fixedaddress); \
                     ed = i;                 \
                     wb1 = 1;                \
@@ -105,7 +110,7 @@
                     ed = i;                 \
                     wb1 = 0;                \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL); \
                     LDRH_IMM8(i, wback, fixedaddress); \
                     ed = i;                 \
                     wb1 = 1;                \
@@ -117,7 +122,7 @@
                     ed = i;                 \
                     wb1 = 0;                \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL); \
                     LDRSH_IMM8(i, wback, fixedaddress);\
                     ed = i;                 \
                     wb1 = 1;                \
@@ -137,7 +142,7 @@
                     wb1 = 0;                \
                     ed = i;                 \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 4095, 0, 0, NULL); \
                     LDRB_IMM9(i, wback, fixedaddress); \
                     wb1 = 1;                \
                     ed = i;                 \
@@ -151,9 +156,10 @@
                     wb1 = 0;                \
                     ed = i;                 \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0, 0); \
-                    ADD_REG_LSL_IMM5(wback, wback, i, 0);   \
-                    LDRB_IMM9(i, wback, fixedaddress);      \
+                    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;                 \
                 }
@@ -166,7 +172,7 @@
                     wb1 = 0;                \
                     ed = i;                 \
                 } else {                    \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 255, 0, 0, NULL); \
                     LDRSB_IMM8(i, wback, fixedaddress);\
                     wb1 = 1;                \
                     ed = i;                 \
@@ -204,6 +210,18 @@
 // 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);
 
 #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))
@@ -232,7 +250,7 @@
     Bcond(cond, j32)
 // Branch to next instruction if cond (use j32)
 #define B_NEXT(cond)     \
-    j32 = (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->arm_size+8)):0; \
+    j32 = dyn->insts[ninst].epilog-(dyn->arm_size+8); \
     Bcond(cond, j32)
 // Branch to MARKSEG if cond (use j32)
 #define B_MARKSEG(cond)    \
@@ -243,11 +261,11 @@
     j32 = GETMARKLOCK-(dyn->arm_size+8);   \
     Bcond(cond, j32)
 
-#define IFX(A)  if(dyn->insts && (dyn->insts[ninst].x86.need_flags&(A)))
-#define IFX_PENDOR0  if(dyn->insts && (dyn->insts[ninst].x86.need_flags&(X_PEND) || !dyn->insts[ninst].x86.need_flags))
-#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 IFXN(A, B)  if(dyn->insts && (dyn->insts[ninst].x86.need_flags&(A) && !(dyn->insts[ninst].x86.need_flags&(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)    \
@@ -271,26 +289,26 @@
         /* greater than leave 0 */                                          \
     }                                                                       \
     SET_DFNONE(s1);                                                         \
-    IFX(X_OF|X_PEND) {                                                      \
-        BFC(xFlags, F_OF, 1);                                               \
-    }                                                                       \
-    IFX(X_AF|X_PEND) {                                                      \
-        BFC(xFlags, F_AF, 1);                                               \
-    }                                                                       \
-    IFX(X_SF|X_PEND) {                                                      \
-        BFC(xFlags, F_SF, 1);                                               \
+    IFX(X_OF|X_AF|X_SF|X_PEND) {                                            \
+        /* clear 0b100010010000 */                                          \
+        BIC_IMM8(xFlags, xFlags, 0b10001001, 14);                           \
     }                                                                       \
 
 
-#define SET_DFNONE(S)    if(!dyn->dfnone) {MOVW(S, d_none); STR_IMM9(S, xEmu, offsetof(x86emu_t, df)); dyn->dfnone=1;}
-#define SET_DF(S, N)     if(N) {MOVW(S, N); STR_IMM9(S, xEmu, offsetof(x86emu_t, df)); dyn->dfnone=0;} else SET_DFNONE(S)
-#define SET_NODF()          dyn->dfnone = 0
-#define SET_DFOK()          dyn->dfnone = 1
+#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(((A)!=X_PEND) && dyn->state_flags!=SF_SET) {     \
-        if(dyn->state_flags!=SF_PENDING) {              \
+    if((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);               \
             j32 = (GETMARKF)-(dyn->arm_size+8);         \
@@ -298,18 +316,29 @@
         }                                               \
         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) 
@@ -317,17 +346,14 @@
 #ifndef BARRIER_NEXT
 #define BARRIER_NEXT(A)
 #endif
-#define UFLAG_OP1(A) if(dyn->insts && dyn->insts[ninst].x86.need_flags) {STR_IMM9(A, xEmu, offsetof(x86emu_t, op1));}
-#define UFLAG_OP2(A) if(dyn->insts && dyn->insts[ninst].x86.need_flags) {STR_IMM9(A, xEmu, offsetof(x86emu_t, op2));}
-#define UFLAG_OP12(A1, A2) if(dyn->insts && dyn->insts[ninst].x86.need_flags) {STR_IMM9(A1, xEmu, offsetof(x86emu_t, op1));STR_IMM9(A2, 0, offsetof(x86emu_t, op2));}
-#define UFLAG_RES(A) if(dyn->insts && dyn->insts[ninst].x86.need_flags) {STR_IMM9(A, xEmu, offsetof(x86emu_t, res));}
-#define UFLAG_DF(r, A) if(dyn->insts && dyn->insts[ninst].x86.need_flags) {SET_DF(r, A)}
-#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
 
 #if STEP < 2
@@ -365,6 +391,7 @@ void* arm_next(x86emu_t* emu, 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)
 
@@ -376,6 +403,9 @@ void* arm_next(x86emu_t* emu, uintptr_t addr);
 #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 emit_cmp8       STEPNAME(emit_cmp8)
@@ -444,23 +474,31 @@ void* arm_next(x86emu_t* emu, uintptr_t addr);
 #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_restoreround STEPNAME(x87_restoreround)
+#define x87_swapreg     STEPNAME(x87_swapreg)
 #define sse_setround    STEPNAME(sse_setround)
 #define mmx_get_reg     STEPNAME(mmx_get_reg)
 #define mmx_get_reg_empty STEPNAME(mmx_get_reg_empty)
@@ -482,8 +520,10 @@ void* arm_next(x86emu_t* emu, uintptr_t addr);
 // 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);
@@ -496,10 +536,11 @@ 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, 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);
-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);
@@ -566,9 +607,13 @@ 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);
 
@@ -576,15 +621,21 @@ void emit_pf(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4);
 // 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 s1);
+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, 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)
@@ -595,9 +646,43 @@ void x87_reget_st(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st);
 int x87_setround(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
 // Restore round flag
 void x87_restoreround(dynarec_arm_t* dyn, int ninst, int s1);
+// 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);
 
+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 s2, int s3, int a);
@@ -606,17 +691,17 @@ int mmx_get_reg_empty(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int
 
 //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);
-void x87_purgecache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
-void mmx_purgecache(dynarec_arm_t* dyn, int ninst, int s1);
+void fpu_reset(dynarec_arm_t* dyn);
+// purge the FPU cache (needs 3 scratch registers) next=1 if for a conditionnal branch jumping out of block (no tracking updated)
+void fpu_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1, int s2, int s3);
+void x87_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1, int s2, int s3);
+void mmx_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1);
 #ifdef HAVE_TRACE
 void fpu_reflectcache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
 #endif
@@ -644,6 +729,7 @@ 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);
 

src/dynarec/dynarec_arm_pass.c

@@ -3,6 +3,7 @@
 #include <stddef.h>
 #include <pthread.h>
 #include <errno.h>
+#include <string.h>
 
 #include "debug.h"
 #include "box86context.h"
@@ -29,23 +30,99 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
 {
     int ok = 1;
     int ninst = 0;
+    int j32;
     uintptr_t ip = addr;
     uintptr_t init_addr = addr;
+    MAYUSE(init_addr);
     int need_epilog = 1;
     dyn->sons_size = 0;
     // Clean up (because there are multiple passes)
-    dyn->state_flags = 0;
-    dyn->dfnone = 0;
-    fpu_reset(dyn, ninst);
+    dyn->f.pending = 0;
+    dyn->f.dfnone = 0;
+    fpu_reset(dyn);
+    int reset_n = -1;
     // ok, go now
     INIT;
     while(ok) {
         ip = addr;
-        if(dyn->insts && (dyn->insts[ninst].x86.barrier==1)) {
-            NEW_BARRIER_INST;
+        if (reset_n!=-1) {
+            if(reset_n==-2) {
+                MESSAGE(LOG_DEBUG, "Reset Caches to zero\n");
+                dyn->f.dfnone = 0;
+                dyn->f.pending = 0;
+                fpu_reset(dyn);
+            } else {
+                MESSAGE(LOG_DEBUG, "Reset Caches with %d\n",reset_n);
+                #if STEP > 1
+                // for STEP 2 & 3, just need to refrest with current, and undo the changes (push & swap)
+                dyn->n = dyn->insts[ninst].n;
+                neoncacheUnwind(&dyn->n);
+                #ifdef HAVE_TRACE
+                if(box86_dynarec_dump)
+                    if(memcmp(&dyn->n, &dyn->insts[reset_n].n, sizeof(neon_cache_t))) {
+                        MESSAGE(LOG_DEBUG, "Warning, difference in neoncache: reset=");
+                        for(int i=0; i<24; ++i)
+                            if(dyn->insts[reset_n].n.neoncache[i].v)
+                                MESSAGE(LOG_DEBUG, " %02d:%s", i, getCacheName(dyn->insts[reset_n].n.neoncache[i].t, dyn->insts[reset_n].n.neoncache[i].n));
+                        if(dyn->insts[reset_n].n.combined1 || dyn->insts[reset_n].n.combined2)
+                            MESSAGE(LOG_DEBUG, " %s:%02d/%02d", dyn->insts[reset_n].n.swapped?"SWP":"CMB", dyn->insts[reset_n].n.combined1, dyn->insts[reset_n].n.combined2);
+                        if(dyn->insts[reset_n].n.stack_push || dyn->insts[reset_n].n.stack_pop)
+                            MESSAGE(LOG_DEBUG, " (%d:%d)", dyn->insts[reset_n].n.stack_push, -dyn->insts[reset_n].n.stack_pop);
+                        MESSAGE(LOG_DEBUG, " ==> ");
+                        for(int i=0; i<24; ++i)
+                            if(dyn->insts[ninst].n.neoncache[i].v)
+                                MESSAGE(LOG_DEBUG, " %02d:%s", i, getCacheName(dyn->insts[ninst].n.neoncache[i].t, dyn->insts[ninst].n.neoncache[i].n));
+                        if(dyn->insts[ninst].n.combined1 || dyn->insts[ninst].n.combined2)
+                            MESSAGE(LOG_DEBUG, " %s:%02d/%02d", dyn->insts[ninst].n.swapped?"SWP":"CMB", dyn->insts[ninst].n.combined1, dyn->insts[ninst].n.combined2);
+                        if(dyn->insts[ninst].n.stack_push || dyn->insts[ninst].n.stack_pop)
+                            MESSAGE(LOG_DEBUG, " (%d:%d)", dyn->insts[ninst].n.stack_push, -dyn->insts[ninst].n.stack_pop);
+                        MESSAGE(LOG_DEBUG, " -> ");
+                        for(int i=0; i<24; ++i)
+                            if(dyn->n.neoncache[i].v)
+                                MESSAGE(LOG_DEBUG, " %02d:%s", i, getCacheName(dyn->n.neoncache[i].t, dyn->n.neoncache[i].n));
+                        if(dyn->n.combined1 || dyn->n.combined2)
+                            MESSAGE(LOG_DEBUG, " %s:%02d/%02d", dyn->n.swapped?"SWP":"CMB", dyn->n.combined1, dyn->n.combined2);
+                        if(dyn->n.stack_push || dyn->n.stack_pop)
+                            MESSAGE(LOG_DEBUG, " (%d:%d)", dyn->n.stack_push, -dyn->n.stack_pop);
+                        MESSAGE(LOG_DEBUG, "\n");
+                    }
+                #endif //HAVE_TRACE
+                #else
+                dyn->n = dyn->insts[reset_n].n;
+                #endif
+                dyn->f = dyn->insts[reset_n].f_exit;
+                if(dyn->insts[ninst].x86.barrier&BARRIER_FLOAT) {
+                    MESSAGE(LOG_DEBUG, "Apply Barrier Float\n");
+                    fpu_reset(dyn);
+                }
+                if(dyn->insts[ninst].x86.barrier&BARRIER_FLAGS) {
+                    MESSAGE(LOG_DEBUG, "Apply Barrier Flags\n");
+                    dyn->f.dfnone = 0;
+                    dyn->f.pending = 0;
+                }
+            }
+            reset_n = -1;
         }
+        // propagate ST stack state, especial stack pop that are defered
+        if(dyn->n.stack_pop) {
+            for(int j=0; j<24; ++j)
+                if((dyn->n.neoncache[j].t == NEON_CACHE_ST_D || dyn->n.neoncache[j].t == NEON_CACHE_ST_F)) {
+                    if(dyn->n.neoncache[j].n<dyn->n.stack_pop)
+                        dyn->n.neoncache[j].v = 0;
+                    else
+                        dyn->n.neoncache[j].n-=dyn->n.stack_pop;
+                }
+            dyn->n.stack_pop = 0;
+        }
+        dyn->n.stack = dyn->n.stack_next;
+        dyn->n.news = 0;
+        dyn->n.stack_push = 0;
+        dyn->n.swapped = 0;
         NEW_INST;
         fpu_reset_scratch(dyn);
+        if((dyn->insts[ninst].x86.need_before&~X_PEND) && !dyn->insts[ninst].pred_sz) {
+            READFLAGS(dyn->insts[ninst].x86.need_before&~X_PEND);
+        }
 #ifdef HAVE_TRACE
         if(my_context->dec && box86_dynarec_trace) {
         if((trace_end == 0) 
@@ -66,48 +143,89 @@ uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
 
         INST_EPILOG;
 
-        if(dyn->insts && dyn->insts[ninst+1].x86.barrier) {
-            fpu_purgecache(dyn, ninst, x1, x2, x3);
-            if(dyn->insts[ninst+1].x86.barrier!=2) {
-                dyn->state_flags = 0;
-                dyn->dfnone = 0;
+        int next = ninst+1;
+        #if STEP > 0
+        if(!dyn->insts[ninst].x86.has_next && dyn->insts[ninst].x86.jmp && dyn->insts[ninst].x86.jmp_insts!=-1)
+            next = dyn->insts[ninst].x86.jmp_insts;
+        #endif
+        if(dyn->insts[ninst].x86.has_next && dyn->insts[next].x86.barrier) {
+            if(dyn->insts[next].x86.barrier&BARRIER_FLOAT)
+                fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+            if(dyn->insts[next].x86.barrier&BARRIER_FLAGS) {
+                dyn->f.pending = 0;
+                dyn->f.dfnone = 0;
             }
         }
-        if(!ok && !need_epilog && dyn->insts && (addr < (dyn->start+dyn->isize))) {
+        #if STEP != 0
+        if(!ok && !need_epilog && (addr < (dyn->start+dyn->isize))) {
             ok = 1;
+            // we use the 1st predecessor here
+            int ii = ninst+1;
+            while(ii<dyn->size && !dyn->insts[ii].pred_sz) 
+                ++ii;
+            if((dyn->insts[ii].x86.barrier&BARRIER_FULL)==BARRIER_FULL)
+                reset_n = -2;    // hack to say Barrier!
+            else {
+                reset_n = getNominalPred(dyn, ii);  // may get -1 if no predecessor are availble
+                if(reset_n==-1) {
+                    reset_n = -2;
+                    MESSAGE(LOG_DEBUG, "Warning, Reset Caches mark not found\n");
+                }
+            }
         }
-        if(!ok && !need_epilog && !dyn->insts && getProtection(addr+3))
+        #else
+        #ifndef PROT_READ
+        #define PROT_READ 1
+        #endif
+        if(!ok && !need_epilog && box86_dynarec_bigblock && (getProtection(addr+3)&PROT_READ))
             if(*(uint32_t*)addr!=0) {   // check if need to continue (but is next 4 bytes are 0, stop)
                 uintptr_t next = get_closest_next(dyn, addr);
                 if(next && (
                     (((next-addr)<15) && is_nops(dyn, addr, next-addr)) 
-                    ||(((next-addr)<30) && is_instructions(dyn, addr, next-addr)) ))
+                    && (((box86_dynarec_bigblock<2) && isJumpTableDefault((void*)next)) || (box86_dynarec_bigblock>1))
+                    /*||(((next-addr)<30) && is_instructions(dyn, addr, next-addr))*/ ))
                 {
-                    dynarec_log(LOG_DEBUG, "Extend block %p, %p -> %p (ninst=%d)\n", dyn, (void*)addr, (void*)next, ninst);
                     ok = 1;
+                    // need to find back that instruction to copy the caches, as previous version cannot be used anymore
+                    reset_n = -2;
+                    for(int ii=0; ii<ninst; ++ii)
+                        if(dyn->insts[ii].x86.jmp == next) {
+                            reset_n = ii;
+                            ii=ninst;
+                        }
+                    if(box86_dynarec_dump) dynarec_log(LOG_NONE, "Extend block %p, %p -> %p (ninst=%d, jump from %d)\n", dyn, (void*)addr, (void*)next, ninst, reset_n);
                 } else if(next && (next-addr)<30) {
-                    dynarec_log(LOG_DEBUG, "Cannot extend block %p -> %p (%02X %02X %02X %02X %02X %02X %02X %02x)\n", (void*)addr, (void*)next, PK(0), PK(1), PK(2), PK(3), PK(4), PK(5), PK(6), PK(7));
+                    if(box86_dynarec_dump) dynarec_log(LOG_NONE, "Cannot extend block %p -> %p (%02X %02X %02X %02X %02X %02X %02X %02x)\n", (void*)addr, (void*)next, PK(0), PK(1), PK(2), PK(3), PK(4), PK(5), PK(6), PK(7));
                 }
             }
+        #endif
         if(ok<0)  {ok = 0; need_epilog=1;}
         ++ninst;
         #if STEP == 0
-        if(ok && !isJumpTableDefault((void*)addr))
+        if((ok && !isJumpTableDefault((void*)addr) && (box86_dynarec_bigblock<2)) ||
+            (ok && dyn->size >= 2000))
         #else
-        if(ok && dyn->insts && (ninst==dyn->size))
+        if(ok && (ninst==dyn->size))
         #endif
         {
-            #if STEP == 3
-            dynarec_log(LOG_DEBUG, "Stopping block %p (%d / %d)\n",(void*)init_addr, ninst, dyn->size); 
+            int j32;
+            MAYUSE(j32);
+            MESSAGE(LOG_DEBUG, "Stopping block %p (%d / %d)\n",(void*)init_addr, ninst, dyn->size); 
+            --ninst;
+            if(!dyn->insts[ninst].x86.barrier) {
+                BARRIER(BARRIER_FLOAT);
+            }
+            #if STEP == 0
+            dyn->insts[ninst].x86.need_after |= X_PEND;
             #endif
-            BARRIER(2);
-            fpu_purgecache(dyn, ninst, x1, x2, x3);
+            ++ninst;
+            fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
             jump_to_next(dyn, addr, 0, ninst);
             ok=0; need_epilog=0;
         }
     }
     if(need_epilog) {
-        fpu_purgecache(dyn, ninst, x1, x2, x3);
+        fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
         jump_to_epilog(dyn, ip, 0, ninst);  // no linker here, it's an unknow instruction
     }
     FINI;

src/dynarec/dynarec_arm_pass0.h

@@ -1,20 +1,47 @@
 
 #define INIT    uintptr_t sav_addr=addr
-#define FINI    dyn->isize = addr-sav_addr
+#define FINI    \
+    dyn->isize = addr-sav_addr;\
+    dyn->insts[ninst].x86.addr = addr;\
+    if(ninst) dyn->insts[ninst-1].x86.size = dyn->insts[ninst].x86.addr - dyn->insts[ninst-1].x86.addr
 #define MESSAGE(A, ...)  
-#define SETFLAGS(A, B)  
-#define READFLAGS(A)    
+#define MAYSETFLAGS()   dyn->insts[ninst].x86.may_set = 1
+#define READFLAGS(A)    \
+        dyn->insts[ninst].x86.use_flags = A; dyn->f.dfnone = 1;\
+        dyn->f.pending=SF_SET
+#define SETFLAGS(A,B)   \
+        dyn->insts[ninst].x86.set_flags = A;    \
+        dyn->insts[ninst].x86.state_flags = B;  \
+        dyn->f.pending=(B)&SF_SET_PENDING;      \
+        dyn->f.dfnone=((B)&SF_SET)?1:0;
 #define EMIT(A)     
-#define JUMP(A)         add_next(dyn, (uintptr_t)A)
-#define NEW_INST        ++dyn->size
-#define INST_EPILOG         
+#define JUMP(A, C)      if((A)>addr) add_next(dyn, (uintptr_t)(A)); dyn->insts[ninst].x86.jmp = A; dyn->insts[ninst].x86.jmp_cond = C
+#define BARRIER(A)      if(A!=BARRIER_MAYBE) {fpu_purgecache(dyn, ninst, 0, x1, x2, x3); dyn->insts[ninst].x86.barrier = A;} else dyn->insts[ninst].barrier_maybe = 1
+#define BARRIER_NEXT(A) dyn->insts[ninst+1].x86.barrier = A
+#define NEW_INST \
+        ++dyn->size;                            \
+        if(dyn->size+3>=dyn->cap) {             \
+                dyn->insts = (instruction_arm_t*)box_realloc(dyn->insts, sizeof(instruction_arm_t)*dyn->cap*2); \
+                memset(&dyn->insts[dyn->cap], 0, sizeof(instruction_arm_t)*dyn->cap);   \
+                dyn->cap *= 2;                  \
+        }                                       \
+        dyn->insts[ninst].x86.addr = ip;        \
+        dyn->n.combined1 = dyn->n.combined2 = 0;\
+        dyn->n.swapped = 0; dyn->n.barrier = 0; \
+        dyn->insts[ninst].f_entry = dyn->f;     \
+        if(ninst) {dyn->insts[ninst-1].x86.size = dyn->insts[ninst].x86.addr - dyn->insts[ninst-1].x86.addr;}
+
+#define INST_EPILOG                             \
+        dyn->insts[ninst].f_exit = dyn->f;      \
+        dyn->insts[ninst].n = dyn->n;           \
+        dyn->insts[ninst].x86.has_next = ok;
 #define INST_NAME(name) 
 #define DEFAULT                         \
         --dyn->size;                    \
         *ok = -1;                       \
-        if(box86_dynarec_log>=LOG_INFO) {\
-        dynarec_log(LOG_NONE, "%p: Dynarec stopped because of Opcode %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X", \
-        (void*)ip, PKip(0),             \
+        if(box86_dynarec_log>=LOG_INFO || box86_dynarec_dump) {\
+        dynarec_log(LOG_NONE, "%p: Dynarec stopped @ %s (%s:%u) because of Opcode %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X", \
+        (void*)ip, __FUNCTION__, __FILE__, __LINE__, PKip(0), \
         PKip(1), PKip(2), PKip(3),      \
         PKip(4), PKip(5), PKip(6),      \
         PKip(7), PKip(8), PKip(9),      \

src/dynarec/dynarec_arm_pass1.h

@@ -1,17 +1,15 @@
 #define INIT    
-#define FINI     \
-    dyn->insts[ninst].x86.addr = addr; \
-    if(ninst) dyn->insts[ninst-1].x86.size = dyn->insts[ninst].x86.addr - dyn->insts[ninst-1].x86.addr;
+#define FINI
 #define MESSAGE(A, ...)  
 #define EMIT(A)     
-#define READFLAGS(A)    dyn->insts[ninst].x86.use_flags = A
-#define SETFLAGS(A,B)   {dyn->insts[ninst].x86.set_flags = A; dyn->insts[ninst].x86.state_flags = B;}
-#define JUMP(A)         dyn->insts[ninst].x86.jmp = A
-#define BARRIER(A)      dyn->insts[ninst].x86.barrier = A
-#define BARRIER_NEXT(A) if(ninst<dyn->size) dyn->insts[ninst+1].x86.barrier = A
 
-#define NEW_INST \
-    dyn->insts[ninst].x86.addr = ip; \
-    if(ninst) dyn->insts[ninst-1].x86.size = dyn->insts[ninst].x86.addr - dyn->insts[ninst-1].x86.addr;
-#define INST_EPILOG 
+#define NEW_INST                                \
+        dyn->insts[ninst].f_entry = dyn->f;     \
+        dyn->n.combined1 = dyn->n.combined2 = 0;\
+        dyn->n.swapped = 0; dyn->n.barrier = 0
+
+#define INST_EPILOG                             \
+        dyn->insts[ninst].n = dyn->n;           \
+        dyn->insts[ninst].f_exit = dyn->f
+
 #define INST_NAME(name)  

src/dynarec/dynarec_arm_pass2.h

@@ -3,7 +3,11 @@
 
 #define MESSAGE(A, ...)  
 #define EMIT(A)     dyn->insts[ninst].size+=4; dyn->arm_size+=4
-#define NEW_INST    if(ninst) {dyn->insts[ninst].address = (dyn->insts[ninst-1].address+dyn->insts[ninst-1].size);}
+#define NEW_INST                                                                                        \
+        if(ninst) {                                                                                     \
+                dyn->insts[ninst].address = (dyn->insts[ninst-1].address+dyn->insts[ninst-1].size);     \
+                if(ninst && isInstClean(dyn, ninst))                                                    \
+                        ++dyn->sons_size;                                                               \
+        }
 #define INST_EPILOG dyn->insts[ninst].epilog = dyn->arm_size; 
 #define INST_NAME(name) 
-#define NEW_BARRIER_INST    if(ninst) ++dyn->sons_size

src/dynarec/dynarec_arm_pass3.h

@@ -7,28 +7,58 @@
     dyn->insts[ninst].size2 += 4
 
 #define MESSAGE(A, ...)  if(box86_dynarec_dump) dynarec_log(LOG_NONE, __VA_ARGS__)
-#define NEW_INST        
+#define NEW_INST        \
+    if(ninst && isInstClean(dyn, ninst)) {                      \
+        dyn->sons_x86[dyn->sons_size] = (uintptr_t)ip;          \
+        dyn->sons_arm[dyn->sons_size] = dyn->block;             \
+        MESSAGE(LOG_DUMP, "----> potential Son here %p/%p\n", (void*)ip, dyn->block);  \
+        ++dyn->sons_size;                                       \
+    }
+
 #define INST_EPILOG     
 #define INST_NAME(name) \
     if(box86_dynarec_dump) {\
         printf_x86_instruction(my_context->dec, &dyn->insts[ninst].x86, name); \
-        dynarec_log(LOG_NONE, "%s%p: %d emited opcodes, state=%d/%d, set=%X, use=%X, need=%X%s\n", \
+        dynarec_log(LOG_NONE, "%s%p: %d emited opcodes, inst=%d, barrier=%d state=%d/%d(%d), %s=%X/%X, use=%X, need=%X/%X", \
             (box86_dynarec_dump>1)?"\e[32m":"", \
             (void*)(dyn->arm_start+dyn->insts[ninst].address),  \
             dyn->insts[ninst].size/4,           \
+            ninst,                              \
+            dyn->insts[ninst].x86.barrier,      \
             dyn->insts[ninst].x86.state_flags,  \
-            dyn->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_flags,   \
-            (box86_dynarec_dump>1)?"\e[m":"");  \
-    }
-
-#define NEW_BARRIER_INST                            \
-    if(ninst) {                                     \
-    dyn->sons_x86[dyn->sons_size] = (uintptr_t)ip;  \
-    dyn->sons_arm[dyn->sons_size] = dyn->block;     \
-    MESSAGE(LOG_DUMP, "----> potential Son here\n");\
-    ++dyn->sons_size;                               \
+            dyn->insts[ninst].x86.need_before,  \
+            dyn->insts[ninst].x86.need_after);  \
+        if(dyn->insts[ninst].pred_sz) {         \
+            dynarec_log(LOG_NONE, ", pred=");     \
+            for(int ii=0; ii<dyn->insts[ninst].pred_sz; ++ii)\
+                dynarec_log(LOG_NONE, "%s%d", ii?"/":"", dyn->insts[ninst].pred[ii]);\
+        }                                       \
+        if(dyn->insts[ninst].x86.jmp && dyn->insts[ninst].x86.jmp_insts>=0)\
+            dynarec_log(LOG_NONE, ", jmp=%d", dyn->insts[ninst].x86.jmp_insts);\
+        if(dyn->insts[ninst].x86.jmp && dyn->insts[ninst].x86.jmp_insts==-1)\
+            dynarec_log(LOG_NONE, ", jmp=out");\
+        for(int ii=0; ii<24; ++ii) {            \
+            switch(dyn->insts[ninst].n.neoncache[ii].t) {    \
+                case NEON_CACHE_ST_D: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;            \
+                case NEON_CACHE_ST_F: dynarec_log(LOG_NONE, " S%d:%s", (ii+8)*2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;        \
+                case NEON_CACHE_MM: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;              \
+                case NEON_CACHE_XMMW: dynarec_log(LOG_NONE, " Q%d:%s", (ii+8)/2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); ++ii; break;  \
+                case NEON_CACHE_XMMR: dynarec_log(LOG_NONE, " Q%d:%s", (ii+8)/2, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); ++ii; break;  \
+                case NEON_CACHE_SCR: dynarec_log(LOG_NONE, " D%d:%s", ii+8, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;             \
+                case NEON_CACHE_NONE:           \
+                default:    break;              \
+            }                                   \
+        }                                       \
+        if(dyn->n.stack || dyn->insts[ninst].n.stack_next || dyn->insts[ninst].n.x87stack)     \
+            dynarec_log(LOG_NONE, " X87:%d/%d(+%d/-%d)%d", dyn->n.stack, dyn->insts[ninst].n.stack_next, dyn->insts[ninst].n.stack_push, dyn->insts[ninst].n.stack_pop, dyn->insts[ninst].n.x87stack); \
+        if(dyn->insts[ninst].n.combined1 || dyn->insts[ninst].n.combined2)                     \
+            dynarec_log(LOG_NONE, " %s:%d/%d", dyn->insts[ninst].n.swapped?"SWP":"CMB", dyn->insts[ninst].n.combined1, dyn->insts[ninst].n.combined2);   \
+        dynarec_log(LOG_NONE, "%s\n", (box86_dynarec_dump>1)?"\e[m":"");                       \
     }
 

src/dynarec/dynarec_arm_private.h

@@ -4,6 +4,60 @@
 #include "dynarec_private.h"
 
 typedef struct x86emu_s x86emu_t;
+typedef struct dynablock_s dynablock_t;
+
+#define BARRIER_MAYBE   8
+
+#define NEON_CACHE_NONE 0
+#define NEON_CACHE_ST_D 1
+#define NEON_CACHE_ST_F 2
+#define NEON_CACHE_MM   3
+#define NEON_CACHE_XMMW 4
+#define NEON_CACHE_XMMR 5
+#define NEON_CACHE_SCR  6
+typedef union neon_cache_s {
+    int8_t           v;
+    struct {
+        uint8_t t:4;   // reg type
+        uint8_t n:4;   // reg number
+    };
+} neon_cache_t;
+typedef union sse_cache_s {
+    int8_t     v;
+    struct {
+        uint8_t reg;
+        uint8_t write;
+    };
+} sse_cache_t;
+typedef struct neoncache_s {
+    // Neon cache
+    neon_cache_t        neoncache[24];
+    int8_t              stack;
+    int8_t              stack_next;
+    int8_t              stack_pop;
+    int8_t              stack_push;
+    uint8_t             combined1;
+    uint8_t             combined2;
+    uint8_t             swapped;    // the combined reg were swapped
+    uint8_t             barrier;    // is there a barrier at instruction epilog?
+    uint32_t            news;       // bitmask, wich neoncache are new for this opcode
+    // fpu cache
+    int8_t              x87cache[8];// cache status for the 8 x87 register behind the fpu stack
+    int8_t              x87reg[8];  // reg used for x87cache entry
+    int8_t              mmxcache[8];// cache status for the 8 MMX registers
+    sse_cache_t         ssecache[8];// cache status for the 8 SSE(2) registers
+    int8_t              fpuused[24];// all 8..31 double reg from fpu, used by x87, sse and mmx
+    int8_t              x87stack;   // cache stack counter
+    int8_t              mmxcount;   // number of mmx register used (not both mmx and x87 at the same time)
+    int8_t              fpu_scratch;// scratch counter
+    int8_t              fpu_extra_qscratch; // some opcode need an extra quad scratch register
+    int8_t              fpu_reg;    // x87/sse/mmx reg counter
+} neoncache_t;
+
+typedef struct flagcache_s {
+    int                 pending;    // is there a pending flags here, or to check?
+    int                 dfnone;     // if defered flags is already set to df_none
+} flagcache_t;
 
 typedef struct instruction_arm_s {
     instruction_x86_t   x86;
@@ -11,6 +65,8 @@ typedef struct instruction_arm_s {
     uintptr_t           epilog;     // epilog of current instruction (can be start of next, of barrier stuff)
     int                 size;       // size of the arm emited instruction
     int                 size2;      // size of the arm emited instrucion after pass2
+    int                 pred_sz;    // size of predecessor list
+    int                 *pred;      // predecessor array
     uintptr_t           mark, mark2, mark3;
     uintptr_t           markf;
     uintptr_t           markseg;
@@ -18,8 +74,13 @@ typedef struct instruction_arm_s {
     int                 pass2choice;// value for choices that are fixed on pass2 for pass3
     uintptr_t           natcall;
     int                 retn;
+    int                 barrier_maybe;
+    flagcache_t         f_exit;     // flags status at end of intruction
+    neoncache_t         n;          // neoncache at end of intruction (but before poping)
+    flagcache_t         f_entry;    // flags status before the instruction begin
 } instruction_arm_t;
 
+
 typedef struct dynarec_arm_s {
     instruction_arm_t   *insts;
     int32_t             size;
@@ -29,25 +90,16 @@ typedef struct dynarec_arm_s {
     void*               block;      // memory pointer where next instruction is emited
     uintptr_t           arm_start;  // start of the arm code
     int                 arm_size;   // size of emitted arm code
-    int                 state_flags;// actual state for on-demand flags
-    int                 x87cache[8];// cache status for the 8 x87 register behind the fpu stack
-    int                 x87reg[8];  // reg used for x87cache entry
-    int                 mmxcache[8];// cache status for the 8 MMX registers
-    int                 ssecache[8];// cache status for the 8 SSE(2) registers
-    int                 fpuused[24];// all 8..31 double reg from fpu, used by x87, sse and mmx
-    int                 x87stack;   // cache stack counter
-    int                 x87count;   // number of x87 register used
-    int                 mmxcount;   // number of mmx register used (not both mmx and x87 at the same time)
-    int                 fpu_scratch;// scratch counter
-    int                 fpu_extra_qscratch; // some opcode need an extra quad scratch register
-    int                 fpu_reg;    // x87/sse/mmx reg counter
-    int                 dfnone;     // if defered flags is already set to df_none
+    flagcache_t         f;
+    neoncache_t         n;          // cache for the 8..31 double reg from fpu, plus x87 stack delta
     uintptr_t*          next;       // variable array of "next" jump address
     int                 next_sz;
     int                 next_cap;
     uintptr_t*          sons_x86;   // the x86 address of potential dynablock sons
     void**              sons_arm;   // the arm address of potential dynablock sons
     int                 sons_size;  // number of potential dynablock sons
+    int*                predecessor;// single array of all predecessor
+    dynablock_t*        dynablock;
 } dynarec_arm_t;
 
 void add_next(dynarec_arm_t *dyn, uintptr_t addr);

src/dynarec/dynarec_private.h

@@ -1,6 +1,11 @@
 #ifndef __DYNAREC_PRIVATE_H_
 #define __DYNAREC_PRIVATE_H_
 
+#define BARRIER_NONE    0
+#define BARRIER_FLAGS   1
+#define BARRIER_FLOAT   2
+#define BARRIER_FULL    3
+
 // all flags for the use_flags field
 #define X_CF    (1<<0)
 #define X_PF    (1<<1)
@@ -9,25 +14,32 @@
 #define X_SF    (1<<4)
 #define X_OF    (1<<5)
 #define X_ALL   ((1<<6)-1)
-#define X_PEND  (0x1000)
+#define X_PEND  (0x80)
 
 // all state flags
 #define SF_UNKNOWN  0
 #define SF_SET      1
 #define SF_PENDING  2
-#define SF_MAYSET   3
-#define SF_SUBSET   4
+#define SF_SET_PENDING (SF_SET|SF_PENDING)
+#define SF_SUB      4
+#define SF_SUBSET   (SF_SUB|SF_SET)
+#define SF_SUBSET_PENDING   (SF_SUBSET|SF_PENDING)
 
 typedef struct instruction_x86_s {
     uintptr_t   addr;   //address of the instruction
     int32_t     size;   // size of the instruction
-    int         barrier; // next instruction is a jump point, so no optim allowed
     uintptr_t   jmp;    // offset to jump to, even if conditionnal (0 if not), no relative offset here
     int         jmp_insts;  // instuction to jump to (-1 if out of the block)
-    uint32_t    use_flags;  // 0 or combination of X_?F
-    uint32_t    set_flags;  // 0 or combination of X_?F
-    uint32_t    need_flags; // calculated
-    int         state_flags; // One of SF_XXX state
+    uint8_t     jmp_cond;   // 1 of conditionnal jump
+    uint8_t     has_next;   // does this opcode can continue to the next?
+    uint8_t     barrier;    // next instruction is a jump point, so no optim allowed
+    uint8_t     state_flags;// One of SF_XXX state
+    uint8_t     use_flags;  // 0 or combination of X_?F
+    uint8_t     set_flags;  // 0 or combination of X_?F
+    uint8_t     may_set;    // 1 if the flags may not be set
+    uint8_t     gen_flags;  // calculated
+    uint8_t     need_before; // calculated
+    uint8_t     need_after;    // calculated
 } instruction_x86_t;
 
 void printf_x86_instruction(zydis_dec_t* dec, instruction_x86_t* inst, const char* name);

src/dynarec/last_run.txt

@@ -1,3 +1,4 @@
+ARM_ 1 1 1 1 0 0 0 1 0 0 0 0 (0) (0) (0) 1 (0) (0) (0) (0) (0) (0) @<1> (0) 0 0 0 0 (0) (0) (0) (0) SETEND \%cparam ? 'B' : 'L'\E
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 0 op 0 N Q M 0 Vm VH\(op ? "SUB" : "ADD")\<c>.<dt> <Qd>, <Qn>, <Qm>
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 0 0 0 N Q M 1 Vm VQADD<c>.<dt> <Qd>, <Qn>, <Qm>
 NEON 1 1 1 1 0 0 1 U 0 D size Vn Vd 0 0 0 1 N Q M 0 Vm VRHADD<c>.<dt> <Qd>, <Qn>, <Qm>
@@ -82,7 +83,7 @@ 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>
@@ -141,99 +142,128 @@ NEON 1 1 1 1 0 0 1 0 1 D size Vn Vd 1 0 1 1 N 1 M 0 Vm @ size=0b01=0b10 unsigned
 NEON 1 1 1 1 0 0 1 Q 1 D size Vn Vd 1 1 0 0 N 1 M 0 Vm @ size=0b01=0b10 unsigned int decodedImm !@ decodedImm = m >> %; m %@ 3@ &= 0x7@ 4@ &= 0xF@ 5@ = 0x21@ @ VQDMULH<c>.<dt> <Qd>, <Qn>, <Dm[x]>
 NEON 1 1 1 1 0 0 1 Q 1 D size Vn Vd 1 1 0 1 N 1 M 0 Vm @ size=0b01=0b10 unsigned int decodedImm !@ decodedImm = m >> %; m %@ 3@ &= 0x7@ 4@ &= 0xF@ 5@ = 0x21@ @ VQRDMULH<c>.<dt> <Qd>, <Qn>, <Dm[x]>
 INVALIDATE 1 1 1 1 0 0 1 <1> 1 <16> 0 <1> 0 <4>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 0 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 0 1 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 0 size @<2> Rm @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 0 1 size @<2> Rm @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 0 1 1 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 0 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 0 0 Rn Vd 1 0 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\], <Rm>
-INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 0 0 <8> 1 1 <10>
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 0 @<4> 1 1 1 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 0 @<4> 1 1 0 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]!
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 0 @<4> Rm @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]!
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]!
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> Rm @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]!
-NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\, \vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 0 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\, \vecname[0x26 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 0 1 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x23 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 0 size @<2> Rm @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 0 1 size @<2> Rm @ param=0b00 char align[4] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 63 + param)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\, \vecname[0x24 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\-\vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 0 1 1 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 1 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 1 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 0 1 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x22 + d]\}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 1 0 size @<2> 1 1 0 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 1 0 size @<2> 1 1 1 1 @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 0 D 1 0 Rn Vd 1 0 1 0 size @<2> Rm @ param=0b00 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 32 << param)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\, \vecname[0x21 + d]\}, [<Rn>\align\], <Rm>
-INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <8> 1 1 <10>
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 0 @<4> 1 1 1 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]
+INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 0 0 <8> 1 1 <10>
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 0 @<4> 1 1 0 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 0 @<4> 1 1 1 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 0 @<4> Rm @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VST1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 0 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 0 @<4> 1 1 0 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]!
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 0 @<4> 1 1 1 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 0 @<4> Rm @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VST3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 0 0 Rn Vd size 1 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VST4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
+INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <8> 1 1 <2> 1 1 <6>
+INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <8> 1 1 <2> 0 0 <1> 1 <4>
+INVALIDATE 1 1 1 1 0 1 0 0 1 <1> 1 0 <8> 1 1 1 0 <3> 1 <4>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 0 @<1> 1 1 0 1 @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 0 @<1> 1 1 1 1 @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 0 @<1> Rm @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 1 @<1> 1 1 0 1 @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 1 @<1> 1 1 1 1 @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + d]\[]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 0 size 1 @<1> Rm @ param=1 char align[4] !@ %@ sprintf(align, ":%d", 8 << size)@ align[0] = '\0'@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + d]\[]}, [<Rn>\align\], <Rm>
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 0 @<4> 1 1 0 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 0 @<4> 1 1 1 1 @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 0 @<4> Rm @ set uint8_t temp = (2 << size) - 1@ set uint8_t temp2 = param & temp@ temp2=0=param>>1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 8 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ @ VLD1<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\]}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> 1 1 0 1 @ param2_1=1 char align[4] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> 1 1 1 1 @ param2_1=1 char align[4] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 0 1 size @<1> @<1> Rm @ param2_1=1 char align[4] !@ %@ sprintf(align, ":%d", 16 << size)@ align[0] = '\0'@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[]}, [<Rn>\align\], <Rm>
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\]
 NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0=1 char align[5] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", 16 << size)@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD2<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]!
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> Rm @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>], <Rm>
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]!
-NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 0 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 0 size @<1> @<1> 1 1 0 1 VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[]}, [<Rn>]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 0 size @<1> @<1> 1 1 1 1 VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[]}, [<Rn>]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 0 size @<1> @<1> Rm VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[]}, [<Rn>], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 0 @<4> 1 1 0 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 0 @<4> 1 1 1 1 @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 0 @<4> Rm @ set uint8_t temp = param & (1 << size)@ temp=0 uint8_t spacing spacing,0,1@ set uint8_t x = param >> (size + 1)@ @ VLD3<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\]}, [<Rn>], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 1 size @<1> @<1> 1 1 0 1 @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 32 << size)@ align[0] = '\0'@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[], \vecname[0x23 + param1_1+param1_1+param1_1 + d]\[]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 1 size @<1> @<1> 1 1 1 1 @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 32 << size)@ align[0] = '\0'@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[], \vecname[0x23 + param1_1+param1_1+param1_1 + d]\[]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd 1 1 1 1 size @<1> @<1> Rm @ param2_1=1 char align[5] !@ %@ sprintf(align, ":%d", 32 << size)@ align[0] = '\0'@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[], \vecname[0x21 + param1_1 + d]\[], \vecname[0x22 + param1_1+param1_1 + d]\[], \vecname[0x23 + param1_1+param1_1+param1_1 + d]\[]}, [<Rn>\align\], <Rm>
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 1 @<4> 1 1 0 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]!
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 1 @<4> 1 1 1 1 @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\]
+NEON 1 1 1 1 0 1 0 0 1 D 1 0 Rn Vd size 1 1 @<4> Rm @ size=0b00=0b01 uint8_t temp temp,1,1,3@ set uint8_t temp2 = param & temp@ temp2=0b0=0b1=0b10 char align[6] !@ %@ align[0] = '\0'@ sprintf(align, ":%d", (size ? 64 : 32))@ sprintf(align, ":128")@ sprintf(align, "???")@ set uint8_t x = param >> (size + 1)@ set int spacing = size ? ((param >> size) & 1) : 0@ @ VLD4<c>.\%d8 &l&l size\ {\vecname[0x20 + d]\[\%dx\], \vecname[0x21 + d + spacing]\[\%dx\], \vecname[0x22 + d + 2*spacing]\[\%dx\], \vecname[0x23 + d + 3*spacing]\[\%dx\]}, [<Rn>\align\], <Rm>
 INVALIDATE 1 1 1 1 0 1 0 0 <3> 0 <20>
+ARM_ 1 1 1 1 0 1 0 0 @<1> 0 0 1 Rn @<8> @<4> @<4> 'NOP'
+ARM_ 1 1 1 1 0 1 0 0 U 1 0 1 Rn (1) (1) (1) (1) imm12 PLI [<Rn>, #+/-<imm12>]
+ARM_ 1 1 1 1 0 1 0 1 U @<1> 0 1 Rn (1) (1) (1) (1) imm12 PLD\param ? "W" : ""\ [<Rn>, #+/-<imm12>]
+ARM_ 1 1 1 1 0 1 0 1 U (1) 0 1 1 1 1 1 (1) (1) (1) (1) imm12 PLD [PC, #+/-<imm>]
+ARM_ 1 1 1 1 0 1 0 1 0 1 1 1 (1) (1) (1) (1) (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 1 (1) (1) (1) (1) CLREX
+ARM_ 1 1 1 1 0 1 0 1 0 1 1 1 (1) (1) (1) (1) (1) (1) (1) (1) (0) (0) (0) (0) 0 1 0 0 @<4> @ param=0b1111=0b1110=0b1011=0b1010=0b0111=0b0110=0b0011=0b0010 const char* option option,"SY","ST","ISH","ISHST","NSH","NSHST","OSH","OSHST","!!!"@ @ DSB \option\
+ARM_ 1 1 1 1 0 1 0 1 0 1 1 1 (1) (1) (1) (1) (1) (1) (1) (1) (0) (0) (0) (0) 0 1 0 1 @<4> @ param=0b1111=0b1110=0b1011=0b1010=0b0111=0b0110=0b0011=0b0010 const char* option option,"SY","ST","ISH","ISHST","NSH","NSHST","OSH","OSHST","!!!"@ @ DMB \option\
+ARM_ 1 1 1 1 0 1 0 1 0 1 1 1 (1) (1) (1) (1) (1) (1) (1) (1) (0) (0) (0) (0) 0 1 1 0 @<4> @ param=0b1111 const char* option option,"SY","!!!"@ @ ISB \option\
+ARM_ 1 1 1 1 0 1 1 0 @<1> 0 0 1 Rn @<8> @<3> 0 @<4> 'NOP'
+ARMs 1 1 1 1 0 1 1 0 U 1 0 1 Rn (1) (1) (1) (1) imm5 type 0 Rm PLI [<Rn>, +/-<Rm>{, <shift>}]
+ARMs 1 1 1 1 0 1 1 1 U @<1> 0 1 Rn (1) (1) (1) (1) imm5 type 0 Rm PLD\param ? "W" : ""\ [<Rn>, +/-<Rm>{, <shift>}]
+INVALIDATE 1 1 1 1 0 1 1 1 1 1 1 1 <4> <8> 1 1 1 1 <4>
 INVALIDATE 1 1 1 1 <28>
 ARMS cond 0 0 0 0 0 0 0 S Rn Rd imm5 type 0 Rm AND{S}<c> <Rd>, <Rn>, <Rm>{, <shift>}
 ARM_ cond 0 0 0 0 0 0 0 S Rd (0) (0) (0) (0) Rm 1 0 0 1 Rn MUL{S}<c> <Rd>, <Rn>, <Rm>
@@ -283,6 +313,8 @@ 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>
@@ -330,6 +362,17 @@ ARM_ cond 0 0 1 0 1 1 1 S Rn Rd imm12 RSC{S}<c> <Rd>, <Rn>, #<const>
 ARM_ cond 0 0 1 1 0 0 0 0 imm4 Rd imm12 MOVW<c> <Rd>, #<imm16>
 ARM_ cond 0 0 1 1 0 0 0 1 Rn (0) (0) (0) (0) imm12 TST<c> <Rn>, #<const>
 ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 0 0 0 0 0 NOP<c>
+ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 0 0 0 0 1 YIELD<c>
+ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 0 0 0 1 0 WFE<c>
+ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 0 0 0 1 1 WFI<c>
+ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 0 0 1 0 0 SEV<c>
+ARM_ cond/=1110 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 0 0 0 1 0 1 0 0 CSDB<c>
+ARM_ cond 0 0 1 1 0 0 1 0 0 0 0 0 (1) (1) (1) (1) (0) (0) (0) (0) 1 1 1 1 imm4 DBG<c> #<imm>
+INVALIDATE <4> 0 0 1 1 0 0 1 0 0 0 0 0 <8> <8>
+ARM_ cond 0 0 1 1 0 0 1 0 @<2> 0 0 (1) (1) (1) (1) imm12 @ param=0b01=0b10 const char* spec_reg spec_reg,"APSR_g","APSR_nzcvq","APSR_nzcvqg"@ @ MSR<c> \spec_reg\, #<imm>
+INVALIDATE <4> 0 0 1 1 0 0 1 0 <2> 0 1 <8> <8>
+INVALIDATE <4> 0 0 1 1 0 0 1 0 <2> 1 <1> <8> <8>
+INVALIDATE <4> 0 0 1 1 0 1 1 0 <4> <8> <8>
 ARM_ cond 0 0 1 1 0 0 1 1 Rn (0) (0) (0) (0) imm12 TEQ<c> <Rn>, #<const>
 ARM_ cond 0 0 1 1 0 1 0 0 imm4 Rd imm12 MOVT<c> <Rd>, #<imm16>
 ARM_ cond 0 0 1 1 0 1 0 1 Rn (0) (0) (0) (0) imm12 CMP<c> <Rn>, #<const>
@@ -388,6 +431,8 @@ ARMs cond 0 1 1 0 U 1 0 0 Rn Rt imm5 type 0 Rm STRB<c> <Rt>, [<Rn>], +/-<Rm>{, <
 ARMs cond 0 1 1 0 U 1 0 1 Rn Rt imm5 type 0 Rm LDRB<c> <Rt>, [<Rn>], +/-<Rm>{, <shift>}
 ARMs cond 0 1 1 0 U 1 1 0 Rn Rt imm5 type 0 Rm STRBT<c> <Rt>, [<Rn>], +/-<Rm>{, <shift>}
 ARMs cond 0 1 1 0 U 1 1 1 Rn Rt imm5 type 0 Rm LDRBT<c> <Rt>, [<Rn>], +/-<Rm>{, <shift>}
+ARM_ cond 0 1 1 1 0 0 0 1 Rd (1) (1) (1) (1) Rm 0 0 0 1 Rn SDIV<c> <Rd>, <Rn>, <Rm>
+ARM_ cond 0 1 1 1 0 0 1 1 Rd (1) (1) (1) (1) Rm 0 0 0 1 Rn UDIV<c> <Rd>, <Rn>, <Rm>
 ARMs cond 0 1 1 1 U 0 W 0 Rn Rt imm5 type 0 Rm STR<c> <Rt>, [<Rn>, +/-<Rm>{, <shift>}]{!}
 ARMs cond 0 1 1 1 U 0 W 1 Rn Rt imm5 type 0 Rm LDR<c> <Rt>, [<Rn>, +/-<Rm>{, <shift>}]{!}
 ARMs cond 0 1 1 1 U 1 W 0 Rn Rt imm5 type 0 Rm STRB<c> <Rt>, [<Rn>, +/-<Rm>{, <shift>}]{!}
@@ -397,6 +442,7 @@ ARM_ cond 0 1 1 1 1 0 0 0 Rd Ra Rm 0 0 0 1 Rn USADA8<c> <Rd>, <Rn>, <Rm>, <Ra>
 ARM_ cond 0 1 1 1 1 1 0 msb Rd lsb 0 0 1 1 1 1 1 BFC<c> <Rd>, #<lsb>, #<width>
 ARM_ cond 0 1 1 1 1 1 0 msb Rd lsb 0 0 1 Rn BFI<c> <Rd>, <Rn>, #<lsb>, #<width>
 ARM_ cond 0 1 1 1 1 1 1 widthm1 Rd lsb 1 0 1 Rn UBFX<c> <Rd>, <Rn>, #<lsb>, #<width>
+ARM_ cond/=1110 0 1 1 1 1 1 1 1 imm12 1 1 1 1 imm4 UDF<c> #<imm16>
 ARM_ cond 1 0 0 0 0 0 W 0 Rn/=1101 register_list STMED<c> <Rn>{!}, <registers>
 ARM_ cond 1 0 0 0 0 0 W 0 Rn register_list STMDA<c> <Rn>{!}, <registers>
 ARM_ cond 1 0 0 0 0 0 W 1 Rn/=1101 register_list LDMED<c> <Rn>{!}, <registers>
@@ -427,14 +473,14 @@ INVALIDATE <4> 1 1 0 1 1 <1> 1 <9> 1 0 1 <9>
 INVALIDATE <4> 1 1 0 1 1 <1> 1 <9> 1 0 1 <9>
 VFPU cond 1 1 0 1 0 D 1 0 1 1 0 1 Vd 1 0 1 1 imm8/01 @ set --imm8@ @ FSTMX<c> <LIst>
 VFPU cond 1 1 0 1 0 D 1 0 1 1 0 1 Vd 1 0 1 sz imm8 VPUSH<c> <list>
-VFPU cond 1 1 0 1 U D 0 0 Rn Vd 1 0 1 sz imm8 VSTR<c> <Dd>, [<Rn>{, #+/-<imm>}]
+VFPU cond 1 1 0 1 U D 0 0 Rn Vd 1 0 1 sz imm8 @ set imm8*=4@ @ VSTR<c> <Dd>, [<Rn>{, #+/-<imm>}]
 VFPU cond 1 1 0 @<2> D W 0 Rn Vd 1 0 1 1 imm8/01 @ set --imm8@ @ FSTM\(param == 0b01 ? "IA" : "DB")\X<c> <Rn>{!}, <LIst>
 VFPU cond 1 1 0 @<2> D W 0 Rn Vd 1 0 1 sz imm8 VSTM\(param == 0b01 ? "IA" : "DB")\<c> <Rn>{!}, <list>
 VFPU cond 1 1 0 0 1 D 1 1 1 1 0 1 Vd 1 0 1 1 imm8/01 @ set --imm8@ @ FLDMX<c> <LIst>
 VFPU cond 1 1 0 0 1 D 1 1 1 1 0 1 Vd 1 0 1 sz imm8 VPOP<c> <list>
 VFPU cond 1 1 0 1 0 D 0 1 1 1 1 1 Vd 1 0 1 sz 0 0 0 0 0 0 0 0 VLDR<c> <Dd>, [PC, #-0]
 VFPU cond 1 1 0 1 U D 0 1 1 1 1 1 Vd 1 0 1 sz imm8 VLDR<c> <Dd>, <label>
-VFPU cond 1 1 0 1 U D 0 1 Rn Vd 1 0 1 sz imm8 VLDR<c> <Dd>, [<Rn>{, #+/-<imm>}]
+VFPU cond 1 1 0 1 U D 0 1 Rn Vd 1 0 1 sz imm8 @ set imm8*=4@ @ VLDR<c> <Dd>, [<Rn>{, #+/-<imm>}]
 VFPU cond 1 1 0 @<2> D W 1 Rn Vd 1 0 1 1 imm8/01 @ set --imm8@ @ FLDM\(param == 0b01 ? "IA" : "DB")\X<c> <Rn>{!}, <LIst>
 VFPU cond 1 1 0 @<2> D W 1 Rn Vd 1 0 1 sz imm8 VLDM\(param == 0b01 ? "IA" : "DB")\<c> <Rn>{!}, <list>
 INVALIDATE <4> 1 1 0 <25>
@@ -446,6 +492,7 @@ 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 << 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>
@@ -472,6 +519,6 @@ VFPU cond 1 1 1 0 1 D 1 1 0 1 1 1 Vd 1 0 1 0 1 1 M 0 Vm VCVT<c>.F64.F32 <Dd>, <S
 VFPU cond 1 1 1 0 1 D 1 1 0 1 1 1 Vd 1 0 1 1 1 1 M 0 Vm VCVT<c>.F32.F64 <Sd>, <Dm>
 VFPU cond 1 1 1 0 1 D 1 1 1 0 0 0 Vd 1 0 1 sz op 1 M 0 Vm VCVT<c>.F\(size ? "64" : "32")\.\(op ? "S" : "U")\32 <Dd>, <Sm>
 VFPU cond 1 1 1 0 1 D 1 1 1 0 1 U Vd 1 0 1 sz op 1 i 0 imm4 @ set uint8_t td = (u << 2) + op + 1@ td=0b001=0b010=0b101 const char* Td Td,"S16","U16","S32","U32"@ set imm5 = (16 << op) - (imm5 >> 4) - ((imm5 & 0xF) << 1)@ @ VCVT<c>.F\(size ? "64" : "32")\.\Td\ <Dd>, <Dd>, #<imm>
-VFPU cond 1 1 1 0 1 D 1 1 1 1 0 U Vd 1 0 1 sz op 1 M 0 Vm VCVT\(op ? "" : "R")\<c>.\(u ? "U" : "S")\32.F\(size ? "64" : "32")\ <Sd>, <Dm>
+VFPU cond 1 1 1 0 1 D 1 1 1 1 U 1 Vd 1 0 1 sz op 1 M 0 Vm VCVT\(op ? "" : "R")\<c>.\(u ? "U" : "S")\32.F\(size ? "64" : "32")\ <Sd>, <Dm>
 VFPU cond 1 1 1 0 1 D 1 1 1 1 1 U Vd 1 0 1 sz op 1 i 0 imm4 @ set uint8_t td = (u << 2) + op + 1@ td=0b001=0b010=0b101 const char* Td Td,"S16","U16","S32","U32"@ set imm5 = (16 << op) - (imm5 >> 4) - ((imm5 & 0xF) << 1)@ @ VCVT<c>.\Td\.F\(size ? "64" : "32")\ <Dd>, <Dd>, #<imm>
 INVALIDATE <32>

src/elfs/elfload_dump.c

@@ -200,13 +200,14 @@ const char* DumpRelType(int t)
     return buff;
 }
 
-const char* DumpSym(elfheader_t *h, Elf32_Sym* sym)
+const char* DumpSym(elfheader_t *h, Elf32_Sym* sym, int version)
 {
     static char buff[4096];
+    const char* vername = (version==-1)?"(none)":((version==0)?"*local*":((version==1)?"*global*":GetSymbolVersion(h, version)));
     memset(buff, 0, sizeof(buff));
-    sprintf(buff, "\"%s\", value=%p, size=%d, info/other=%d/%d index=%d", 
+    sprintf(buff, "\"%s\", value=%p, size=%d, info/other=%d/%d index=%d (ver=%d/%s)", 
         h->DynStr+sym->st_name, (void*)sym->st_value, sym->st_size,
-        sym->st_info, sym->st_other, sym->st_shndx);
+        sym->st_info, sym->st_other, sym->st_shndx, version, vername);
     return buff;
 }
 
@@ -221,116 +222,118 @@ const char* IdxSymName(elfheader_t *h, int sym)
 
 void DumpMainHeader(Elf32_Ehdr *header, elfheader_t *h)
 {
-    if(box86_log>=LOG_DUMP) {
-        printf_log(LOG_DUMP, "ELF Dump main header\n");
-        printf_log(LOG_DUMP, "  Entry point = %p\n", (void*)header->e_entry);
-        printf_log(LOG_DUMP, "  Program Header table offset = %p\n", (void*)header->e_phoff);
-        printf_log(LOG_DUMP, "  Section Header table offset = %p\n", (void*)header->e_shoff);
-        printf_log(LOG_DUMP, "  Flags = 0x%X\n", header->e_flags);
-        printf_log(LOG_DUMP, "  ELF Header size = %d\n", header->e_ehsize);
-        printf_log(LOG_DUMP, "  Program Header Entry num/size = %d(%d)/%d\n", h->numPHEntries, header->e_phnum, header->e_phentsize);
-        printf_log(LOG_DUMP, "  Section Header Entry num/size = %d(%d)/%d\n", h->numSHEntries, header->e_shnum, header->e_shentsize);
-        printf_log(LOG_DUMP, "  Section Header index num = %d(%d)\n", h->SHIdx, header->e_shstrndx);
-        printf_log(LOG_DUMP, "ELF Dump ==========\n");
+    if(box86_dump) {
+        printf_dump(LOG_NEVER, "ELF Dump main header\n");
+        printf_dump(LOG_NEVER, "  Entry point = %p\n", (void*)header->e_entry);
+        printf_dump(LOG_NEVER, "  Program Header table offset = %p\n", (void*)header->e_phoff);
+        printf_dump(LOG_NEVER, "  Section Header table offset = %p\n", (void*)header->e_shoff);
+        printf_dump(LOG_NEVER, "  Flags = 0x%X\n", header->e_flags);
+        printf_dump(LOG_NEVER, "  ELF Header size = %d\n", header->e_ehsize);
+        printf_dump(LOG_NEVER, "  Program Header Entry num/size = %d(%d)/%d\n", h->numPHEntries, header->e_phnum, header->e_phentsize);
+        printf_dump(LOG_NEVER, "  Section Header Entry num/size = %d(%d)/%d\n", h->numSHEntries, header->e_shnum, header->e_shentsize);
+        printf_dump(LOG_NEVER, "  Section Header index num = %d(%d)\n", h->SHIdx, header->e_shstrndx);
+        printf_dump(LOG_NEVER, "ELF Dump ==========\n");
 
-        printf_log(LOG_DUMP, "ELF Dump PEntries (%d)\n", h->numSHEntries);
+        printf_dump(LOG_NEVER, "ELF Dump PEntries (%d)\n", h->numSHEntries);
         for (int i=0; i<h->numPHEntries; ++i)
-            printf_log(LOG_DUMP, "  PHEntry %04d : %s\n", i, DumpPHEntry(h->PHEntries+i));
-        printf_log(LOG_DUMP, "ELF Dump PEntries ====\n");
+            printf_dump(LOG_NEVER, "  PHEntry %04d : %s\n", i, DumpPHEntry(h->PHEntries+i));
+        printf_dump(LOG_NEVER, "ELF Dump PEntries ====\n");
 
-        printf_log(LOG_DUMP, "ELF Dump Sections (%d)\n", h->numSHEntries);
+        printf_dump(LOG_NEVER, "ELF Dump Sections (%d)\n", h->numSHEntries);
         for (int i=0; i<h->numSHEntries; ++i)
-            printf_log(LOG_DUMP, "  Section %04d : %s\n", i, DumpSection(h->SHEntries+i, h->SHStrTab));
-        printf_log(LOG_DUMP, "ELF Dump Sections ====\n");
+            printf_dump(LOG_NEVER, "  Section %04d : %s\n", i, DumpSection(h->SHEntries+i, h->SHStrTab));
+        printf_dump(LOG_NEVER, "ELF Dump Sections ====\n");
     }
 }
 
 void DumpSymTab(elfheader_t *h)
 {
-    if(box86_log>=LOG_DUMP && h->SymTab) {
+    if(box86_dump && h->SymTab) {
         const char* name = ElfName(h);
-        printf_log(LOG_DUMP, "ELF Dump SymTab(%d)=\n", h->numSymTab);
+        printf_dump(LOG_NEVER, "ELF Dump SymTab(%d)=\n", h->numSymTab);
         for (int i=0; i<h->numSymTab; ++i)
-            printf_log(LOG_DUMP, "  %s:SymTab[%d] = \"%s\", value=%p, size=%d, info/other=%d/%d index=%d\n", name, 
+            printf_dump(LOG_NEVER, "  %s:SymTab[%d] = \"%s\", value=%p, size=%d, info/other=%d/%d index=%d\n", name, 
                 i, h->StrTab+h->SymTab[i].st_name, (void*)h->SymTab[i].st_value, h->SymTab[i].st_size,
                 h->SymTab[i].st_info, h->SymTab[i].st_other, h->SymTab[i].st_shndx);
-        printf_log(LOG_DUMP, "ELF Dump SymTab=====\n");
+        printf_dump(LOG_NEVER, "ELF Dump SymTab=====\n");
     }
 }
 
 void DumpDynamicSections(elfheader_t *h)
 {
-    if(box86_log>=LOG_DUMP && h->Dynamic) {
-        printf_log(LOG_DUMP, "ELF Dump Dynamic(%d)=\n", h->numDynamic);
+    if(box86_dump && h->Dynamic) {
+        printf_dump(LOG_NEVER, "ELF Dump Dynamic(%d)=\n", h->numDynamic);
         for (int i=0; i<h->numDynamic; ++i)
-            printf_log(LOG_DUMP, "  Dynamic %04d : %s\n", i, DumpDynamic(h->Dynamic+i));
-        printf_log(LOG_DUMP, "ELF Dump Dynamic=====\n");
+            printf_dump(LOG_NEVER, "  Dynamic %04d : %s\n", i, DumpDynamic(h->Dynamic+i));
+        printf_dump(LOG_NEVER, "ELF Dump Dynamic=====\n");
     }
 }
 
 void DumpDynSym(elfheader_t *h)
 {
-    if(box86_log>=LOG_DUMP && h->DynSym) {
+    if(box86_dump && h->DynSym) {
         const char* name = ElfName(h);
-        printf_log(LOG_DUMP, "ELF Dump DynSym(%d)=\n", h->numDynSym);
-        for (int i=0; i<h->numDynSym; ++i)
-            printf_log(LOG_DUMP, "  %s:DynSym[%d] = %s\n", name, i, DumpSym(h, h->DynSym+i));
-        printf_log(LOG_DUMP, "ELF Dump DynSym=====\n");
+        printf_dump(LOG_NEVER, "ELF Dump DynSym(%d)=\n", h->numDynSym);
+        for (int i=0; i<h->numDynSym; ++i) {
+            int version = h->VerSym?((Elf32_Half*)((uintptr_t)h->VerSym+h->delta))[i]:-1;
+            printf_dump(LOG_NEVER, "  %s:DynSym[%d] = %s\n", name, i, DumpSym(h, h->DynSym+i, version));
+        }
+        printf_dump(LOG_NEVER, "ELF Dump DynSym=====\n");
     }
 }
 
 void DumpDynamicNeeded(elfheader_t *h)
 {
-    if(box86_log>=LOG_DUMP && h->DynStrTab) {
-        printf_log(LOG_DUMP, "ELF Dump DT_NEEDED=====\n");
+    if(box86_dump && h->DynStrTab) {
+        printf_dump(LOG_NEVER, "ELF Dump DT_NEEDED=====\n");
         for (int i=0; i<h->numDynamic; ++i)
             if(h->Dynamic[i].d_tag==DT_NEEDED) {
-                printf_log(LOG_DUMP, "  Needed : %s\n", h->DynStrTab+h->Dynamic[i].d_un.d_val + h->delta);
+                printf_dump(LOG_NEVER, "  Needed : %s\n", h->DynStrTab+h->Dynamic[i].d_un.d_val + h->delta);
             }
-        printf_log(LOG_DUMP, "ELF Dump DT_NEEDED=====\n");
+        printf_dump(LOG_NEVER, "ELF Dump DT_NEEDED=====\n");
     }
 }
 
 void DumpDynamicRPath(elfheader_t *h)
 {
-    if(box86_log>=LOG_DUMP && h->DynStrTab) {
-        printf_log(LOG_DUMP, "ELF Dump DT_RPATH/DT_RUNPATH=====\n");
+    if(box86_dump && h->DynStrTab) {
+        printf_dump(LOG_NEVER, "ELF Dump DT_RPATH/DT_RUNPATH=====\n");
         for (int i=0; i<h->numDynamic; ++i) {
             if(h->Dynamic[i].d_tag==DT_RPATH) {
-                printf_log(LOG_DUMP, "   RPATH : %s\n", h->DynStrTab+h->Dynamic[i].d_un.d_val + h->delta);
+                printf_dump(LOG_NEVER, "   RPATH : %s\n", h->DynStrTab+h->Dynamic[i].d_un.d_val + h->delta);
             }
             if(h->Dynamic[i].d_tag==DT_RUNPATH) {
-                printf_log(LOG_DUMP, " RUNPATH : %s\n", h->DynStrTab+h->Dynamic[i].d_un.d_val + h->delta);
+                printf_dump(LOG_NEVER, " RUNPATH : %s\n", h->DynStrTab+h->Dynamic[i].d_un.d_val + h->delta);
             }
         }
-        printf_log(LOG_DUMP, "=====ELF Dump DT_RPATH/DT_RUNPATH\n");
+        printf_dump(LOG_NEVER, "=====ELF Dump DT_RPATH/DT_RUNPATH\n");
     }
 }
 
 void DumpRelTable(elfheader_t *h, int cnt, Elf32_Rel *rel, const char* name)
 {
-    if(box86_log>=LOG_DUMP) {
+    if(box86_dump) {
         const char* elfname = ElfName(h);
-        printf_log(LOG_DUMP, "ELF Dump %s Table(%d) @%p\n", name, cnt, rel);
+        printf_dump(LOG_NEVER, "ELF Dump %s Table(%d) @%p\n", name, cnt, rel);
         for (int i = 0; i<cnt; ++i)
-            printf_log(LOG_DUMP, "  %s:Rel[%d] = %p (0x%X: %s, sym=0x%0X/%s)\n", elfname,
+            printf_dump(LOG_NEVER, "  %s:Rel[%d] = %p (0x%X: %s, sym=0x%0X/%s)\n", elfname,
                 i, (void*)rel[i].r_offset, rel[i].r_info, DumpRelType(ELF32_R_TYPE(rel[i].r_info)), 
                 ELF32_R_SYM(rel[i].r_info), IdxSymName(h, ELF32_R_SYM(rel[i].r_info)));
-        printf_log(LOG_DUMP, "ELF Dump Rel Table=====\n");
+        printf_dump(LOG_NEVER, "ELF Dump Rel Table=====\n");
     }
 }
 
 void DumpRelATable(elfheader_t *h, int cnt, Elf32_Rela *rela, const char* name)
 {
-    if(box86_log>=LOG_DUMP && h->rela) {
+    if(box86_dump && h->rela) {
         const char* elfname = ElfName(h);
-        printf_log(LOG_DUMP, "ELF Dump %s Table(%d) @%p\n", name, cnt, rela);
+        printf_dump(LOG_NEVER, "ELF Dump %s Table(%d) @%p\n", name, cnt, rela);
         for (int i = 0; i<cnt; ++i)
-            printf_log(LOG_DUMP, "  %s:RelA[%d] = %p (0x%X: %s, sym=0x%X/%s) Addend=%d\n", elfname,
+            printf_dump(LOG_NEVER, "  %s:RelA[%d] = %p (0x%X: %s, sym=0x%X/%s) Addend=%d\n", elfname,
                 i, (void*)rela[i].r_offset, rela[i].r_info, DumpRelType(ELF32_R_TYPE(rela[i].r_info)), 
                 ELF32_R_SYM(rela[i].r_info), IdxSymName(h, ELF32_R_SYM(rela[i].r_info)), 
                 rela[i].r_addend);
-        printf_log(LOG_DUMP, "ELF Dump RelA Table=====\n");
+        printf_dump(LOG_NEVER, "ELF Dump RelA Table=====\n");
     }
 }
 
@@ -341,24 +344,24 @@ void DumpBinary(char* p, int sz)
     unsigned char* d = (unsigned char*)p;
     int delta = ((uintptr_t)p)&0xf;
     for (int i = 0; sz; ++i) {
-        printf_log(LOG_DUMP, "%p ", (void*)(((uintptr_t)d)&~0xf));
+        printf_dump(LOG_NEVER, "%p ", (void*)(((uintptr_t)d)&~0xf));
         int n = 16 - delta;
         if (n>sz) n = sz;
         int fill = 16-sz;
         for (int j = 0; j<delta; ++j)
-            printf_log(LOG_DUMP, "   ");
+            printf_dump(LOG_NEVER, "   ");
         for (int j = 0; j<n; ++j)
-            printf_log(LOG_DUMP, "%02X ", d[j]);
+            printf_dump(LOG_NEVER, "%02X ", d[j]);
         for (int j = 0; j<fill; ++j)
-            printf_log(LOG_DUMP, "   ");
-        printf_log(LOG_DUMP, " | ");
+            printf_dump(LOG_NEVER, "   ");
+        printf_dump(LOG_NEVER, " | ");
         for (int j = 0; j<delta; ++j)
-            printf_log(LOG_DUMP, " ");
+            printf_dump(LOG_NEVER, " ");
         for (int j = 0; j<n; ++j)
-            printf_log(LOG_DUMP, "%c", (d[j]<32 || d[j]>127)?'.':d[j]);
+            printf_dump(LOG_NEVER, "%c", (d[j]<32 || d[j]>127)?'.':d[j]);
         for (int j = 0; j<fill; ++j)
-            printf_log(LOG_DUMP, " ");
-        printf_log(LOG_DUMP, "\n");
+            printf_dump(LOG_NEVER, " ");
+        printf_dump(LOG_NEVER, "\n");
         d+=n;
         sz-=n;
         delta=0;

src/elfs/elfloader_private.h

@@ -7,8 +7,10 @@ typedef struct dynablocklist_s dynablocklist_t;
 
 typedef struct library_s library_t;
 typedef struct needed_libs_s needed_libs_t;
+typedef struct kh_mapsymbols_s kh_mapsymbols_t;
 
-#include <pthread.h>
+#include <elf.h>
+#include "elfloader.h"
 
 struct elfheader_s {
     char*       name;
@@ -33,6 +35,8 @@ struct elfheader_s {
     Elf32_Half* VerSym;
     Elf32_Verneed*  VerNeed;
     int         szVerNeed;
+    Elf32_Verdef*   VerDef;
+    int         szVerDef;
     int         e_type;
 
     intptr_t    delta;  // should be 0
@@ -64,14 +68,19 @@ struct elfheader_s {
     uintptr_t   plt_end;
     uintptr_t   text;
     int         textsz;
+    uintptr_t   bss;
+    int         bsssz;
 
     uintptr_t   paddr;
     uintptr_t   vaddr;
     int         align;
     uint32_t    memsz;
+    uint32_t    reserve;
     uint32_t    stacksz;
     int         stackalign;
+    uintptr_t   tlsaddr;
     uint32_t    tlssize;
+    uint32_t    tlsfilesize;
     int         tlsalign;
 
     int32_t     tlsbase;    // the base of the tlsdata in the global tlsdata (always negative)
@@ -87,6 +96,10 @@ struct elfheader_s {
 
     library_t   *lib;
     needed_libs_t *neededlibs;
+
+    kh_mapsymbols_t   *mapsymbols;
+    kh_mapsymbols_t   *weaksymbols;
+    kh_mapsymbols_t   *localsymbols;
 };
 
 #define R_386_NONE	0

src/elfs/elfparser.c

@@ -17,16 +17,16 @@
 int LoadSH(FILE *f, Elf32_Shdr *s, void** SH, const char* name, uint32_t type)
 {
     if(type && (s->sh_type != type)) {
-        printf_log(LOG_INFO, "Section Header \"%s\" (off=%d, size=%d) has incorect type (%d != %d)\n", name, s->sh_offset, s->sh_size, s->sh_type, type);
+        printf_dump(LOG_INFO, "Section Header \"%s\" (off=%d, size=%d) has incorect type (%d != %d)\n", name, s->sh_offset, s->sh_size, s->sh_type, type);
         return -1;
     }
     if (type==SHT_SYMTAB && s->sh_size%sizeof(Elf32_Sym)) {
-        printf_log(LOG_INFO, "Section Header \"%s\" (off=%d, size=%d) has size (not multiple of %d)\n", name, s->sh_offset, s->sh_size, sizeof(Elf32_Sym));
+        printf_dump(LOG_INFO, "Section Header \"%s\" (off=%d, size=%d) has size (not multiple of %d)\n", name, s->sh_offset, s->sh_size, sizeof(Elf32_Sym));
     }
-    *SH = calloc(1, s->sh_size);
+    *SH = box_calloc(1, s->sh_size);
     fseeko64(f, s->sh_offset ,SEEK_SET);
     if(fread(*SH, s->sh_size, 1, f)!=1) {
-            printf_log(LOG_INFO, "Cannot read Section Header \"%s\" (off=%d, size=%d)\n", name, s->sh_offset, s->sh_size);
+            printf_dump(LOG_INFO, "Cannot read Section Header \"%s\" (off=%d, size=%d)\n", name, s->sh_offset, s->sh_size);
             return -1;
     }
 
@@ -46,7 +46,7 @@ int FindSection(Elf32_Shdr *s, int n, char* SHStrTab, const char* name)
 void LoadNamedSection(FILE *f, Elf32_Shdr *s, int size, char* SHStrTab, const char* name, const char* clearname, uint32_t type, void** what, int* num)
 {
     int n = FindSection(s, size, SHStrTab, name);
-    printf_log(LOG_DEBUG, "Loading %s (idx = %d)\n", clearname, n);
+    printf_dump(LOG_DEBUG, "Loading %s (idx = %d)\n", clearname, n);
     if(n)
         LoadSH(f, s+n, what, name, type);
     if(type==SHT_SYMTAB || type==SHT_DYNSYM) {
@@ -63,59 +63,59 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
     Elf32_Ehdr header;
     int level = (exec)?LOG_INFO:LOG_DEBUG;
     if(fread(&header, sizeof(Elf32_Ehdr), 1, f)!=1) {
-        printf_log(level, "Cannot read ELF Header\n");
+        printf_dump(level, "Cannot read ELF Header\n");
         return NULL;
     }
     if(memcmp(header.e_ident, ELFMAG, SELFMAG)!=0) {
-        printf_log(LOG_INFO, "Not an ELF file (sign=%c%c%c%c)\n", header.e_ident[0], header.e_ident[1], header.e_ident[2], header.e_ident[3]);
+        printf_dump(LOG_INFO, "Not an ELF file (sign=%c%c%c%c)\n", header.e_ident[0], header.e_ident[1], header.e_ident[2], header.e_ident[3]);
         return NULL;
     }
     if(header.e_ident[EI_CLASS]!=ELFCLASS32) {
         if(header.e_ident[EI_CLASS]==ELFCLASS64) {
-            printf_log(LOG_INFO, "This is a 64bits ELF! box86 can only run 32bits ELF!\n");
+            printf_dump(LOG_INFO, "This is a 64bits ELF! box86 can only run 32bits ELF!\n");
         } else {
-            printf_log(LOG_INFO, "Not a 32bits ELF (%d)\n", header.e_ident[EI_CLASS]);
+            printf_dump(LOG_INFO, "Not a 32bits ELF (%d)\n", header.e_ident[EI_CLASS]);
         }
         return NULL;
     }
     if(header.e_ident[EI_DATA]!=ELFDATA2LSB) {
-        printf_log(LOG_INFO, "Not a LittleEndian ELF (%d)\n", header.e_ident[EI_DATA]);
+        printf_dump(LOG_INFO, "Not a LittleEndian ELF (%d)\n", header.e_ident[EI_DATA]);
         return NULL;
     }
     if(header.e_ident[EI_VERSION]!=EV_CURRENT) {
-        printf_log(LOG_INFO, "Incorrect ELF version (%d)\n", header.e_ident[EI_VERSION]);
+        printf_dump(LOG_INFO, "Incorrect ELF version (%d)\n", header.e_ident[EI_VERSION]);
         return NULL;
     }
     if(header.e_ident[EI_OSABI]!=ELFOSABI_LINUX && header.e_ident[EI_OSABI]!=ELFOSABI_NONE && header.e_ident[EI_OSABI]!=ELFOSABI_SYSV) {
-        printf_log(LOG_INFO, "Not a Linux ELF (%d)\n",header.e_ident[EI_OSABI]);
+        printf_dump(LOG_INFO, "Not a Linux ELF (%d)\n",header.e_ident[EI_OSABI]);
         return NULL;
     }
 
     if(header.e_type != ET_EXEC && header.e_type != ET_DYN) {
-        printf_log(LOG_INFO, "Not an Executable (%d)\n", header.e_type);
+        printf_dump(LOG_INFO, "Not an Executable (%d)\n", header.e_type);
         return NULL;
     }
 
     if(header.e_machine != EM_386) {
-        printf_log(level, "Not an i386 ELF (%d)\n", header.e_machine);
+        printf_dump(level, "Not an i386 ELF (%d)\n", header.e_machine);
         return NULL;
     }
 
     if(header.e_entry == 0 && exec) {
-        printf_log(LOG_INFO, "No entry point in ELF\n");
+        printf_dump(LOG_INFO, "No entry point in ELF\n");
         return NULL;
     }
     if(header.e_phentsize != sizeof(Elf32_Phdr)) {
-        printf_log(LOG_INFO, "Program Header Entry size incorrect (%d != %d)\n", header.e_phentsize, sizeof(Elf32_Phdr));
+        printf_dump(LOG_INFO, "Program Header Entry size incorrect (%d != %d)\n", header.e_phentsize, sizeof(Elf32_Phdr));
         return NULL;
     }
     if(header.e_shentsize != sizeof(Elf32_Shdr) && header.e_shentsize != 0) {
-        printf_log(LOG_INFO, "Section Header Entry size incorrect (%d != %d)\n", header.e_shentsize, sizeof(Elf32_Shdr));
+        printf_dump(LOG_INFO, "Section Header Entry size incorrect (%d != %d)\n", header.e_shentsize, sizeof(Elf32_Shdr));
         return NULL;
     }
 
-    elfheader_t *h = calloc(1, sizeof(elfheader_t));
-    h->name = strdup(name);
+    elfheader_t *h = box_calloc(1, sizeof(elfheader_t));
+    h->name = box_strdup(name);
     h->entrypoint = header.e_entry;
     h->numPHEntries = header.e_phnum;
     h->numSHEntries = header.e_shnum;
@@ -124,67 +124,67 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
     if(header.e_shentsize && header.e_shnum) {
         // special cases for nums
         if(h->numSHEntries == 0) {
-            printf_log(LOG_DEBUG, "Read number of Sections in 1st Section\n");
+            printf_dump(LOG_DEBUG, "Read number of Sections in 1st Section\n");
             // read 1st section header and grab actual number from here
             fseeko64(f, header.e_shoff, SEEK_SET);
             Elf32_Shdr section;
             if(fread(&section, sizeof(Elf32_Shdr), 1, f)!=1) {
-                free(h);
-                printf_log(LOG_INFO, "Cannot read Initial Section Header\n");
+                box_free(h);
+                printf_dump(LOG_INFO, "Cannot read Initial Section Header\n");
                 return NULL;
             }
             h->numSHEntries = section.sh_size;
         }
         // now read all section headers
-        printf_log(LOG_DEBUG, "Read %d Section header\n", h->numSHEntries);
-        h->SHEntries = (Elf32_Shdr*)calloc(h->numSHEntries, sizeof(Elf32_Shdr));
+        printf_dump(LOG_DEBUG, "Read %d Section header\n", h->numSHEntries);
+        h->SHEntries = (Elf32_Shdr*)box_calloc(h->numSHEntries, sizeof(Elf32_Shdr));
         fseeko64(f, header.e_shoff ,SEEK_SET);
         if(fread(h->SHEntries, sizeof(Elf32_Shdr), h->numSHEntries, f)!=h->numSHEntries) {
                 FreeElfHeader(&h);
-                printf_log(LOG_INFO, "Cannot read all Section Header\n");
+                printf_dump(LOG_INFO, "Cannot read all Section Header\n");
                 return NULL;
         }
 
         if(h->numPHEntries == PN_XNUM) {
-            printf_log(LOG_DEBUG, "Read number of Program Header in 1st Section\n");
+            printf_dump(LOG_DEBUG, "Read number of Program Header in 1st Section\n");
             // read 1st section header and grab actual number from here
             h->numPHEntries = h->SHEntries[0].sh_info;
         }
     }
 
-    printf_log(LOG_DEBUG, "Read %d Program header\n", h->numPHEntries);
-    h->PHEntries = (Elf32_Phdr*)calloc(h->numPHEntries, sizeof(Elf32_Phdr));
+    printf_dump(LOG_DEBUG, "Read %d Program header\n", h->numPHEntries);
+    h->PHEntries = (Elf32_Phdr*)box_calloc(h->numPHEntries, sizeof(Elf32_Phdr));
     fseeko64(f, header.e_phoff ,SEEK_SET);
     if(fread(h->PHEntries, sizeof(Elf32_Phdr), h->numPHEntries, f)!=h->numPHEntries) {
             FreeElfHeader(&h);
-            printf_log(LOG_INFO, "Cannot read all Program Header\n");
+            printf_dump(LOG_INFO, "Cannot read all Program Header\n");
             return NULL;
     }
 
     if(header.e_shentsize && header.e_shnum) {
         if(h->SHIdx == SHN_XINDEX) {
-            printf_log(LOG_DEBUG, "Read number of String Table in 1st Section\n");
+            printf_dump(LOG_DEBUG, "Read number of String Table in 1st Section\n");
             h->SHIdx = h->SHEntries[0].sh_link;
         }
         if(h->SHIdx > h->numSHEntries) {
-            printf_log(LOG_INFO, "Incoherent Section String Table Index : %d / %d\n", h->SHIdx, h->numSHEntries);
+            printf_dump(LOG_INFO, "Incoherent Section String Table Index : %d / %d\n", h->SHIdx, h->numSHEntries);
             FreeElfHeader(&h);
             return NULL;
         }
         // load Section table
-        printf_log(LOG_DEBUG, "Loading Sections Table String (idx = %d)\n", h->SHIdx);
+        printf_dump(LOG_DEBUG, "Loading Sections Table String (idx = %d)\n", h->SHIdx);
         if(LoadSH(f, h->SHEntries+h->SHIdx, (void*)&h->SHStrTab, ".shstrtab", SHT_STRTAB)) {
             FreeElfHeader(&h);
             return NULL;
         }
-        if(box86_log>=LOG_DUMP) DumpMainHeader(&header, h);
+        if(box86_dump) DumpMainHeader(&header, h);
 
         LoadNamedSection(f, h->SHEntries, h->numSHEntries, h->SHStrTab, ".strtab", "SymTab Strings", SHT_STRTAB, (void**)&h->StrTab, NULL);
         LoadNamedSection(f, h->SHEntries, h->numSHEntries, h->SHStrTab, ".symtab", "SymTab", SHT_SYMTAB, (void**)&h->SymTab, &h->numSymTab);
-        if(box86_log>=LOG_DUMP && h->SymTab) DumpSymTab(h);
+        if(box86_dump && h->SymTab) DumpSymTab(h);
 
         LoadNamedSection(f, h->SHEntries, h->numSHEntries, h->SHStrTab, ".dynamic", "Dynamic", SHT_DYNAMIC, (void**)&h->Dynamic, &h->numDynamic);
-        if(box86_log>=LOG_DUMP && h->Dynamic) DumpDynamicSections(h);
+        if(box86_dump && h->Dynamic) DumpDynamicSections(h);
         // grab DT_REL & DT_RELA stuffs
         // also grab the DT_STRTAB string table
         {
@@ -212,7 +212,7 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
                     h->relaent = val;
                     break;
                 case DT_PLTGOT:
-                    h->pltgot = val;
+                    h->pltgot = ptr;
                     break;
                 case DT_PLTREL:
                     h->pltrel = val;
@@ -221,7 +221,7 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
                     h->pltsz = val;
                     break;
                 case DT_JMPREL:
-                    h->jmprel = val;
+                    h->jmprel = ptr;
                     break;
                 case DT_STRTAB:
                     h->DynStrTab = (char *)(ptr);
@@ -231,53 +231,65 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
                     break;
                 case DT_INIT: // Entry point
                     h->initentry = ptr;
-                    printf_log(LOG_DEBUG, "The DT_INIT is at address %p\n", (void*)h->initentry);
+                    printf_dump(LOG_DEBUG, "The DT_INIT is at address %p\n", (void*)h->initentry);
                     break;
                 case DT_INIT_ARRAY:
                     h->initarray = ptr;
-                    printf_log(LOG_DEBUG, "The DT_INIT_ARRAY is at address %p\n", (void*)h->initarray);
+                    printf_dump(LOG_DEBUG, "The DT_INIT_ARRAY is at address %p\n", (void*)h->initarray);
                     break;
                 case DT_INIT_ARRAYSZ:
                     h->initarray_sz = val / sizeof(Elf32_Addr);
-                    printf_log(LOG_DEBUG, "The DT_INIT_ARRAYSZ is %d\n", h->initarray_sz);
+                    printf_dump(LOG_DEBUG, "The DT_INIT_ARRAYSZ is %d\n", h->initarray_sz);
+                    break;
+                case DT_PREINIT_ARRAYSZ:
+                    if(val)
+                        printf_dump(LOG_NONE, "Warning, PreInit Array (size=%d) present and ignored!\n", val);
                     break;
                 case DT_FINI: // Exit hook
                     h->finientry = ptr;
-                    printf_log(LOG_DEBUG, "The DT_FINI is at address %p\n", (void*)h->finientry);
+                    printf_dump(LOG_DEBUG, "The DT_FINI is at address %p\n", (void*)h->finientry);
                     break;
                 case DT_FINI_ARRAY:
                     h->finiarray = ptr;
-                    printf_log(LOG_DEBUG, "The DT_FINI_ARRAY is at address %p\n", (void*)h->finiarray);
+                    printf_dump(LOG_DEBUG, "The DT_FINI_ARRAY is at address %p\n", (void*)h->finiarray);
                     break;
                 case DT_FINI_ARRAYSZ:
                     h->finiarray_sz = val / sizeof(Elf32_Addr);
-                    printf_log(LOG_DEBUG, "The DT_FINI_ARRAYSZ is %d\n", h->finiarray_sz);
+                    printf_dump(LOG_DEBUG, "The DT_FINI_ARRAYSZ is %d\n", h->finiarray_sz);
                     break;
                 case DT_VERNEEDNUM:
-                    h->szVerNeed = val / sizeof(Elf32_Verneed);
-                    printf_log(LOG_DEBUG, "The DT_VERNEEDNUM is %d\n", h->szVerNeed);
+                    h->szVerNeed = val;
+                    printf_dump(LOG_DEBUG, "The DT_VERNEEDNUM is %d\n", h->szVerNeed);
                     break;
                 case DT_VERNEED:
                     h->VerNeed = (Elf32_Verneed*)ptr;
-                    printf_log(LOG_DEBUG, "The DT_VERNEED is at address %p\n", h->VerNeed);
+                    printf_dump(LOG_DEBUG, "The DT_VERNEED is at address %p\n", h->VerNeed);
+                    break;
+                case DT_VERDEFNUM:
+                    h->szVerDef = val;
+                    printf_dump(LOG_DEBUG, "The DT_VERDEFNUM is %d\n", h->szVerDef);
+                    break;
+                case DT_VERDEF:
+                    h->VerDef = (Elf32_Verdef*)ptr;
+                    printf_dump(LOG_DEBUG, "The DT_VERDEF is at address %p\n", h->VerDef);
                     break;
                 }
             }
             if(h->rel) {
                 if(h->relent != sizeof(Elf32_Rel)) {
-                    printf_log(LOG_NONE, "Rel Table Entry size invalid (0x%x should be 0x%x)\n", h->relent, sizeof(Elf32_Rel));
+                    printf_dump(LOG_NONE, "Rel Table Entry size invalid (0x%x should be 0x%x)\n", h->relent, sizeof(Elf32_Rel));
                     FreeElfHeader(&h);
                     return NULL;
                 }
-                printf_log(LOG_DEBUG, "Rel Table @%p (0x%x/0x%x)\n", (void*)h->rel, h->relsz, h->relent);
+                printf_dump(LOG_DEBUG, "Rel Table @%p (0x%x/0x%x)\n", (void*)h->rel, h->relsz, h->relent);
             }
             if(h->rela) {
                 if(h->relaent != sizeof(Elf32_Rela)) {
-                    printf_log(LOG_NONE, "RelA Table Entry size invalid (0x%x should be 0x%x)\n", h->relaent, sizeof(Elf32_Rela));
+                    printf_dump(LOG_NONE, "RelA Table Entry size invalid (0x%x should be 0x%x)\n", h->relaent, sizeof(Elf32_Rela));
                     FreeElfHeader(&h);
                     return NULL;
                 }
-                printf_log(LOG_DEBUG, "RelA Table @%p (0x%x/0x%x)\n", (void*)h->rela, h->relasz, h->relaent);
+                printf_dump(LOG_DEBUG, "RelA Table @%p (0x%x/0x%x)\n", (void*)h->rela, h->relasz, h->relaent);
             }
             if(h->jmprel) {
                 if(h->pltrel == DT_REL) {
@@ -285,16 +297,16 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
                 } else if(h->pltrel == DT_RELA) {
                     h->pltent = sizeof(Elf32_Rela);
                 } else {
-                    printf_log(LOG_NONE, "PLT Table type is unknown (size = 0x%x, type=%d)\n", h->pltsz, h->pltrel);
+                    printf_dump(LOG_NONE, "PLT Table type is unknown (size = 0x%x, type=%d)\n", h->pltsz, h->pltrel);
                     FreeElfHeader(&h);
                     return NULL;
                 }
                 if((h->pltsz / h->pltent)*h->pltent != h->pltsz) {
-                    printf_log(LOG_NONE, "PLT Table Entry size invalid (0x%x, ent=0x%x, type=%d)\n", h->pltsz, h->pltent, h->pltrel);
+                    printf_dump(LOG_NONE, "PLT Table Entry size invalid (0x%x, ent=0x%x, type=%d)\n", h->pltsz, h->pltent, h->pltrel);
                     FreeElfHeader(&h);
                     return NULL;
                 }
-                printf_log(LOG_DEBUG, "PLT Table @%p (type=%d 0x%x/0x%0x)\n", (void*)h->jmprel, h->pltrel, h->pltsz, h->pltent);
+                printf_dump(LOG_DEBUG, "PLT Table @%p (type=%d 0x%x/0x%0x)\n", (void*)h->jmprel, h->pltrel, h->pltsz, h->pltent);
             }
             if(h->DynStrTab && h->szDynStrTab) {
                 //DumpDynamicNeeded(h); cannot dump now, it's not loaded yet
@@ -305,37 +317,43 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
         if(ii) {
             h->gotplt = h->SHEntries[ii].sh_addr;
             h->gotplt_end = h->gotplt + h->SHEntries[ii].sh_size;
-            printf_log(LOG_DEBUG, "The GOT.PLT Table is at address %p\n", (void*)h->gotplt);
+            printf_dump(LOG_DEBUG, "The GOT.PLT Table is at address %p\n", (void*)h->gotplt);
         }
         ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".got");
         if(ii) {
             h->got = h->SHEntries[ii].sh_addr;
             h->got_end = h->got + h->SHEntries[ii].sh_size;
-            printf_log(LOG_DEBUG, "The GOT Table is at address %p..%p\n", (void*)h->got, (void*)h->got_end);
+            printf_dump(LOG_DEBUG, "The GOT Table is at address %p..%p\n", (void*)h->got, (void*)h->got_end);
         }
         ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".plt");
         if(ii) {
             h->plt = h->SHEntries[ii].sh_addr;
             h->plt_end = h->plt + h->SHEntries[ii].sh_size;
-            printf_log(LOG_DEBUG, "The PLT Table is at address %p..%p\n", (void*)h->plt, (void*)h->plt_end);
+            printf_dump(LOG_DEBUG, "The PLT Table is at address %p..%p\n", (void*)h->plt, (void*)h->plt_end);
         }
         // grab version of symbols
         ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".gnu.version");
         if(ii) {
             h->VerSym = (Elf32_Half*)(h->SHEntries[ii].sh_addr);
-            printf_log(LOG_DEBUG, "The .gnu.version is at address %p\n", h->VerSym);
+            printf_dump(LOG_DEBUG, "The .gnu.version is at address %p\n", h->VerSym);
         }
-        // grab .text for main code
+        // grab .bss for main code
         ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".text");
         if(ii) {
             h->text = (uintptr_t)(h->SHEntries[ii].sh_addr);
             h->textsz = h->SHEntries[ii].sh_size;
-            printf_log(LOG_DEBUG, "The .text is at address %p, and is %d big\n", (void*)h->text, h->textsz);
+            printf_dump(LOG_DEBUG, "The .text is at address %p, and is %d big\n", (void*)h->text, h->textsz);
+        }
+        // grab .bss to find brk
+        ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".bss");
+        if(ii) {
+            h->bss = (uintptr_t)(h->SHEntries[ii].sh_addr);
+            h->bsssz = h->SHEntries[ii].sh_size;
+            printf_dump(LOG_DEBUG, "The .bss is at address %p, and is %d big\n", (void*)h->bss, h->bsssz);
         }
 
         LoadNamedSection(f, h->SHEntries, h->numSHEntries, h->SHStrTab, ".dynstr", "DynSym Strings", SHT_STRTAB, (void**)&h->DynStr, NULL);
         LoadNamedSection(f, h->SHEntries, h->numSHEntries, h->SHStrTab, ".dynsym", "DynSym", SHT_DYNSYM, (void**)&h->DynSym, &h->numDynSym);
-        if(box86_log>=LOG_DUMP && h->DynSym) DumpDynSym(h);
     }
     
     return h;
@@ -343,17 +361,41 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
 
 const char* GetSymbolVersion(elfheader_t* h, int version)
 {
-    if(!h->VerNeed || (version<2))
+    version&=0x7fff;    // remove bit15 that switch between hidden/public
+    if(version<2)
         return NULL;
-    Elf32_Verneed *ver = (Elf32_Verneed*)((uintptr_t)h->VerNeed + h->delta);
-    while(ver) {
-        Elf32_Vernaux *aux = (Elf32_Vernaux*)((uintptr_t)ver + ver->vn_aux);
-        for(int j=0; j<ver->vn_cnt; ++j) {
-            if(aux->vna_other==version)
-                return h->DynStr+aux->vna_name;
-            aux = (Elf32_Vernaux*)((uintptr_t)aux + aux->vna_next);
+    /*if(version==1)
+        return "*";*/
+    if(h->VerNeed) {
+        Elf32_Verneed *ver = (Elf32_Verneed*)((uintptr_t)h->VerNeed + h->delta);
+        while(ver) {
+            Elf32_Vernaux *aux = (Elf32_Vernaux*)((uintptr_t)ver + ver->vn_aux);
+            for(int j=0; j<ver->vn_cnt; ++j) {
+                if(aux->vna_other==version)
+                    return h->DynStr+aux->vna_name;
+                aux = (Elf32_Vernaux*)((uintptr_t)aux + aux->vna_next);
+            }
+            ver = ver->vn_next?((Elf32_Verneed*)((uintptr_t)ver + ver->vn_next)):NULL;
         }
-        ver = ver->vn_next?((Elf32_Verneed*)((uintptr_t)ver + ver->vn_next)):NULL;
+    }
+    return GetParentSymbolVersion(h, version);  // if symbol is "internal", use Def table instead
+}
+
+const char* GetParentSymbolVersion(elfheader_t* h, int index)
+{
+    if(!h->VerDef || (index<1))
+        return NULL;
+    Elf32_Verdef *def = (Elf32_Verdef*)((uintptr_t)h->VerDef + h->delta);
+    while(def) {
+        if(def->vd_ndx==index) {
+            if(def->vd_cnt<1)
+                return NULL;
+            /*if(def->vd_flags&VER_FLG_BASE)
+                return NULL;*/
+            Elf32_Verdaux *aux = (Elf32_Verdaux*)((uintptr_t)def + def->vd_aux);
+            return h->DynStr+aux->vda_name; // return Parent, so 1st aux
+        }
+        def = def->vd_next?((Elf32_Verdef*)((uintptr_t)def + def->vd_next)):NULL;
     }
     return NULL;
-}
+}

src/emu/modrm.h

@@ -111,7 +111,7 @@
     } else getecommono(A, reg32_t, O)
 #define getem(A) \
     if((nextop&0xC0)==0xC0) { \
-        A = &emu->mmx87[(nextop&7)]; \
+        A = &emu->mmx[(nextop&7)]; \
     } else getecommon(A, mmx87_regs_t)
 #define getex(A) \
     if((nextop&0xC0)==0xC0) { \
@@ -141,7 +141,7 @@
 #define EX          opex
 #define GB          emu->regs[(nextop>>3)&3].byte[(nextop>>5)&0x1]
 #define GD          emu->regs[((nextop&0x38)>>3)]
-#define GM          emu->mmx87[((nextop&0x38)>>3)]
+#define GM          emu->mmx[((nextop&0x38)>>3)]
 #define GX          emu->xmm[((nextop&0x38)>>3)]
 
 // Alias

src/emu/run0f.h

@@ -3,6 +3,18 @@
         _0f_0x00:                       /* VERx Ed */
             nextop = F8;
             switch((nextop>>3)&7) {
+                case 0:                 /* SLDT Ew */
+                    GET_EW;
+                    EW->word[0] = 0;
+                    if((nextop&0xC0)==0xC0)
+                        EW->word[1] = 0;
+                    break;
+                case 1:                 /* STR Ew */
+                    GET_EW;
+                    EW->word[0] = 0x7f; // dummy return
+                    if((nextop&0xC0)==0xC0)
+                        EW->word[1] = 0;
+                    break;
                 case 4: //VERR
                 case 5: //VERW
                     GET_EW;
@@ -18,6 +30,18 @@
         _0f_0x01:                      
             nextop = F8;
             switch((nextop>>3)&7) {
+                case 0:                 /* SGDT Ed */
+                    GET_ED;
+                    ED->word[0] = 0x7f;    // dummy return...
+                    ED->word[1] = 0x000c;
+                    ED->word[2] = 0xd000;
+                    break;
+                case 1:                 /* SIDT Ed */
+                    GET_ED;
+                    ED->word[0] = 0xfff;    // dummy return, like "disabled"
+                    ED->word[1] = 0;
+                    ED->word[2] = 0;
+                    break;
                 case 4:                 /* SMSW Ew */
                     GET_ED;
                     // dummy for now... Do I need to track CR0 state?
@@ -36,6 +60,26 @@
             if(emu->quit) goto fini;
             STEP;
             NEXT;
+        
+        _0f_0x0D:                   /* PREFETCH group */
+            nextop = F8;
+            switch((nextop>>3)&7) {
+                case 1: /* PREFETCHW */
+                    GET_ED; //nothing for now. Could maybe use some builtin prefetch instruction
+                    break;
+                default:
+                    goto _default;
+            }
+            NEXT;
+        _0f_0x0E:                      /* FEMMS */
+            emu->old_ip = R_EIP;
+            R_EIP = ip-2;
+            emit_signal(emu, SIGILL, (void*)R_EIP, 0);
+            ip = R_EIP;
+            if(emu->quit) goto fini;
+            STEP;
+            NEXT;
+
         _0f_0x10:                      /* MOVUPS Gx,Ex */
             nextop = F8;
             GET_EX;
@@ -203,7 +247,7 @@
                         if(EM->ub[i]&128)
                             GM.ub[i] = 0;
                         else
-                            GM.ub[i] = eax1.ub[EM->ub[i]&7];
+                            GM.ub[i] = eam1.ub[EM->ub[i]&7];
                     }
                     break;
                 case 0x04:  /* PMADDUBSW Gm,Em */
@@ -214,6 +258,17 @@
                         GM.sw[i] = (tmp32s>32767)?32767:((tmp32s<-32768)?-32768:tmp32s);
                     }
                     break;
+                case 0x09:  /* PSIGNW Gm, Em */
+                    nextop = F8;
+                    GET_EM;
+                    for (int i=0; i<4; ++i) {
+                        if (EM->sw[i] < 0) {
+                            GM.sw[i] = -GM.sw[i];
+                        } else if (EM->sw[i] == 0) {
+                            GM.sw[i] = 0;
+                        }
+                    }
+                    break;
                 case 0x0B:  /* PMULHRSW Gm, Em */
                     nextop = F8;
                     GET_EM;
@@ -245,6 +300,44 @@
                     }
                     break;
 
+                case 0xF0: /* MOVBE Gd, Ed*/
+                    nextop = F8;
+                    GET_ED;
+                    GD.dword[0] = __builtin_bswap32(ED->dword[0]);
+                    break;
+                case 0xF1: /* MOVBE Ed, Gd*/
+                    nextop = F8;
+                    GET_ED;
+                    ED->dword[0] = __builtin_bswap32(GD.dword[0]);
+                    break;
+
+                default:
+                    goto _default;
+            }
+            NEXT;
+
+        _0f_0x3A:
+            opcode = F8;
+            switch(opcode) {
+                case 0x0F:  /* palignr */
+                    nextop = F8;
+                    GET_EM;
+                    tmp8u = F8;
+                    if (tmp8u >= 16) {
+                        GM.q = 0;
+                    } else if (tmp8u > 8) {
+                        tmp8u -= 8;
+                        GM.q >>= tmp8u*8;
+                    } else if (tmp8u == 8) {
+                        // nothing
+                    } else if (tmp8u == 0) {
+                        GM.q = EM->q;
+                    } else {
+                        GM.q <<= (8-tmp8u)*8;
+                        GM.q |= (EM->q >> tmp8u*8);
+                    }
+                    break;
+
                 default:
                     goto _default;
             }
@@ -375,8 +468,18 @@
         _0f_0x52:                      /* RSQRTPS Gx, Ex */
             nextop = F8;
             GET_EX;
-            for(int i=0; i<4; ++i)
-                GX.f[i] = 1.0f/sqrtf(EX->f[i]);
+            for(int i=0; i<4; ++i) {
+                if(EX->f[i]==0)
+                    GX.f[i] = 1.0f/EX->f[i];
+                else if (EX->f[i]<0)
+                    GX.f[i] = NAN;
+                else if (isnan(EX->f[i]))
+                    GX.f[i] = EX->f[i];
+                else if (isinf(EX->f[i]))
+                    GX.f[i] = 0.0;
+                else
+                    GX.f[i] = 1.0f/sqrtf(EX->f[i]);
+            }
             NEXT;
         _0f_0x53:                      /* RCPPS Gx, Ex */
             nextop = F8;
@@ -722,12 +825,13 @@
             CHECK_FLAGS(emu);
             nextop = F8;
             GET_ED;
-            tmp8u = GD.byte[0];
+            tmp32s = GD.sdword[0];
+            tmp8u=tmp32s&31;
+            tmp32s >>= 5;
             if((nextop&0xC0)!=0xC0)
             {
-                ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
+                ED=(reg32_t*)(((uint32_t*)(ED))+tmp32s);
             }
-            tmp8u&=31;
             if(ED->dword[0] & (1<<tmp8u))
                 SET_FLAG(F_CF);
             else
@@ -754,12 +858,13 @@
             CHECK_FLAGS(emu);
             nextop = F8;
             GET_ED;
-            tmp8u = GD.byte[0];
+            tmp32s = GD.sdword[0];
+            tmp8u=tmp32s&31;
+            tmp32s >>= 5;
             if((nextop&0xC0)!=0xC0)
             {
-                ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
+                ED=(reg32_t*)(((uint32_t*)(ED))+tmp32s);
             }
-            tmp8u&=31;
             if(ED->dword[0] & (1<<tmp8u))
                 SET_FLAG(F_CF);
             else {
@@ -789,20 +894,10 @@
             GET_ED;
             switch((nextop>>3)&7) {
                 case 0:                 /* FXSAVE m512byte */
-                    // should save flags & all
-                    // copy MMX regs...
-                    for(int i=0; i<8; ++i)
-                        memcpy(((void*)(ED))+32+i*16, &emu->mmx87[0], sizeof(emu->mmx87[0]));
-                    // copy SSE regs
-                    memcpy(((void*)(ED))+160, &emu->xmm[0], sizeof(emu->xmm));
+                    fpu_fxsave(emu, ED);
                     break;
                 case 1:                 /* FXRSTOR m512byte */
-                    // should restore flags & all
-                    // copy back MMX regs...
-                    for(int i=0; i<8; ++i)
-                        memcpy(&emu->mmx87[i], ((void*)(ED))+32+i*16, sizeof(emu->mmx87[0]));
-                    // copy SSE regs
-                    memcpy(&emu->xmm[0], ((void*)(ED))+160/4, sizeof(emu->xmm));
+                    fpu_fxrstor(emu, ED);
                     break;
                 case 2:                 /* LDMXCSR Md */
                     emu->mxcsr = ED->dword[0];
@@ -844,12 +939,13 @@
             CHECK_FLAGS(emu);
             nextop = F8;
             GET_ED;
-            tmp8u = GD.byte[0];
+            tmp32s = GD.sdword[0];
+            tmp8u=tmp32s&31;
+            tmp32s >>= 5;
             if((nextop&0xC0)!=0xC0)
             {
-                ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
+                ED=(reg32_t*)(((uint32_t*)(ED))+tmp32s);
             }
-            tmp8u&=31;
             if(ED->dword[0] & (1<<tmp8u)) {
                 SET_FLAG(F_CF);
                 ED->dword[0] ^= (1<<tmp8u);
@@ -875,10 +971,6 @@
                     CHECK_FLAGS(emu);
                     GET_ED;
                     tmp8u = F8;
-                    if((nextop&0xC0)!=0xC0)
-                    {
-                        ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
-                    }
                     tmp8u&=31;
                     if(ED->dword[0] & (1<<tmp8u))
                         SET_FLAG(F_CF);
@@ -889,10 +981,6 @@
                     CHECK_FLAGS(emu);
                     GET_ED;
                     tmp8u = F8;
-                    if((nextop&0xC0)!=0xC0)
-                    {
-                        ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
-                    }
                     tmp8u&=31;
                     if(ED->dword[0] & (1<<tmp8u)) {
                         SET_FLAG(F_CF);
@@ -905,10 +993,6 @@
                     CHECK_FLAGS(emu);
                     GET_ED;
                     tmp8u = F8;
-                    if((nextop&0xC0)!=0xC0)
-                    {
-                        ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
-                    }
                     tmp8u&=31;
                     if(ED->dword[0] & (1<<tmp8u)) {
                         SET_FLAG(F_CF);
@@ -920,10 +1004,6 @@
                     CHECK_FLAGS(emu);
                     GET_ED;
                     tmp8u = F8;
-                    if((nextop&0xC0)!=0xC0)
-                    {
-                        ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
-                    }
                     tmp8u&=31;
                     if(ED->dword[0] & (1<<tmp8u))
                         SET_FLAG(F_CF);
@@ -940,12 +1020,13 @@
             CHECK_FLAGS(emu);
             nextop = F8;
             GET_ED;
-            tmp8u = GD.byte[0];
+            tmp32s = GD.sdword[0];
+            tmp8u=tmp32s&31;
+            tmp32s >>= 5;
             if((nextop&0xC0)!=0xC0)
             {
-                ED=(reg32_t*)(((uint32_t*)(ED))+(tmp8u>>5));
+                ED=(reg32_t*)(((uint32_t*)(ED))+tmp32s);
             }
-            tmp8u&=31;
             if(ED->dword[0] & (1<<tmp8u))
                 SET_FLAG(F_CF);
             else
@@ -1055,19 +1136,25 @@
             GX.q[1] = eax1.q[1];
             NEXT;
         _0f_0xC7:                      /* CMPXCHG8B Gq */
-            CHECK_FLAGS(emu);
             nextop = F8;
             GET_ED;
-            tmp32u = ED->dword[0];
-            tmp32u2= ED->dword[1];
-            if(R_EAX == tmp32u && R_EDX == tmp32u2) {
-                SET_FLAG(F_ZF);
-                ED->dword[0] = R_EBX;
-                ED->dword[1] = R_ECX;
-            } else {
-                CLEAR_FLAG(F_ZF);
-                R_EAX = tmp32u;
-                R_EDX = tmp32u2;
+            switch((nextop>>3)&7) {
+                case 1:
+                    CHECK_FLAGS(emu);
+                    tmp32u = ED->dword[0];
+                    tmp32u2= ED->dword[1];
+                    if(R_EAX == tmp32u && R_EDX == tmp32u2) {
+                        SET_FLAG(F_ZF);
+                        ED->dword[0] = R_EBX;
+                        ED->dword[1] = R_ECX;
+                    } else {
+                        CLEAR_FLAG(F_ZF);
+                        R_EAX = tmp32u;
+                        R_EDX = tmp32u2;
+                    }
+                    break;
+                default:
+                    goto _default;
             }
             NEXT;
         _0f_0xC8:

src/emu/run66.h

@@ -55,6 +55,9 @@
         STEP
         NEXT;
 
+    _66_0x1E:                      /* PUSH DS */
+        Push16(emu, emu->segs[_DS]);
+        NEXT;
     _66_0x1F:                      /* POP DS */
         emu->segs[_DS] = Pop16(emu);    // no check, no use....
         emu->segs_serial[_DS] = 0;
@@ -86,6 +89,10 @@
     _66_0x3D:
         cmp16(emu, R_AX, F16);
         NEXT;
+    _66_0x3E:                      /* DS: */
+        // ignored
+        opcode = F8;
+        goto *opcodes66[opcode];
     
     _66_0x40:
     _66_0x41:
@@ -482,6 +489,11 @@
         R_ECX = tmp32u;
         NEXT;
 
+    _66_0xF5:                      /* CMC */
+        CHECK_FLAGS(emu);
+        CONDITIONAL_SET_FLAG(!ACCESS_FLAG(F_CF), F_CF);
+        NEXT;
+
     _66_0xF7:                      /* GRP3 Ew(,Iw) */
         nextop = F8;
         GET_EW;

src/emu/rund9.h

@@ -8,9 +8,9 @@
         case 0xC5:
         case 0xC6:
         case 0xC7:  /* FLD STx */
-            ll = ST(nextop&7).q;
+            ll = ST(nextop&7).sq;
             fpu_do_push(emu);
-            ST0.q = ll;
+            ST0.sq = ll;
             break;
         case 0xC8:
         case 0xC9:
@@ -20,9 +20,10 @@
         case 0xCD:
         case 0xCE:
         case 0xCF:  /* FXCH STx */
-            ll = ST(nextop&7).q;
-            ST(nextop&7).q = ST0.q;
-            ST0.q = ll;
+            ll = ST(nextop&7).sq;
+            ST(nextop&7).sq = ST0.sq;
+            ST0.sq = ll;
+            emu->sw.f.F87_C1 = 0;
             break;
 
         case 0xD0:  /* FNOP */
@@ -30,9 +31,11 @@
 
         case 0xE0:  /* FCHS */
             ST0.d = -ST0.d;
+            emu->sw.f.F87_C1 = 0;
             break;
         case 0xE1:  /* FABS */
             ST0.d = fabs(ST0.d);
+            emu->sw.f.F87_C1 = 0;
             break;
         
         case 0xE4:  /* FTST */
@@ -73,25 +76,30 @@
 
         case 0xF0:  /* F2XM1 */
             ST0.d = exp2(ST0.d) - 1.0;
+            emu->sw.f.F87_C1 = 0;
             break;
         case 0xF1:  /* FYL2X */
             ST(1).d *= log2(ST0.d);
             fpu_do_pop(emu);
+            emu->sw.f.F87_C1 = 0;
             break;
         case 0xF2:  /* FPTAN */
             ST0.d = tan(ST0.d);
             fpu_do_push(emu);
             ST0.d = 1.0;
             emu->sw.f.F87_C2 = 0;
+            emu->sw.f.F87_C1 = 0;
             break;
         case 0xF3:  /* FPATAN */
             ST1.d = atan2(ST1.d, ST0.d);
             fpu_do_pop(emu);
+            emu->sw.f.F87_C1 = 0;
             break;
         case 0xF4:  /* FXTRACT */
             ST0.d = frexp(ST0.d, &tmp32s);
             fpu_do_push(emu);
             ST0.d = tmp32s;
+            // C1 set only if stack under/overflow occurs
             break;
 
         case 0xF8:  /* FPREM */
@@ -146,26 +154,37 @@
         case 0xF9:  /* FYL2XP1 */
             ST(1).d *= log2(ST0.d + 1.0);
             fpu_do_pop(emu);
+            emu->sw.f.F87_C1 = 0;
             break;
         case 0xFA:  /* FSQRT */
             ST0.d = sqrt(ST0.d);
+            emu->sw.f.F87_C1 = 0;
             break;
         case 0xFB:  /* FSINCOS */
             fpu_do_push(emu);
             sincos(ST1.d, &ST1.d, &ST0.d);
+            emu->sw.f.F87_C2 = 0;
+            emu->sw.f.F87_C1 = 0;
             break;
         case 0xFC:  /* FRNDINT */
             ST0.d = fpu_round(emu, ST0.d);
+            emu->sw.f.F87_C1 = 0;
             break;
         case 0xFD:  /* FSCALE */
             // this could probably be done by just altering the exponant part of the float...
-            ST0.d *= exp2(trunc(ST1.d));
+            if(ST0.d!=0.0)
+                ST0.d *= exp2(trunc(ST1.d));
+            emu->sw.f.F87_C1 = 0;
             break;
         case 0xFE:  /* FSIN */
             ST0.d = sin(ST0.d);
+            emu->sw.f.F87_C2 = 0;
+            emu->sw.f.F87_C1 = 0;
             break;
         case 0xFF:  /* FCOS */
             ST0.d = cos(ST0.d);
+            emu->sw.f.F87_C2 = 0;
+            emu->sw.f.F87_C1 = 0;
             break;
 
 
@@ -226,9 +245,8 @@
                 break;
             case 5:     /* FLDCW Ew */
                 GET_EW;
-                emu->cw = EW->word[0];
+                emu->cw.x16 = EW->word[0];
                 // do something with cw?
-                emu->round = (fpu_round_t)((emu->cw >> 10) & 3);
                 break;
             case 6:     /* FNSTENV m */
                 // warning, incomplete
@@ -240,7 +258,7 @@
                 break;
             case 7: /* FNSTCW Ew */
                 GET_EW;
-                EW->word[0] = emu->cw;
+                EW->word[0] = emu->cw.x16;
                 break;
             default:
                 goto _default;

src/emu/rundb.h

@@ -132,12 +132,12 @@
                 GET_ED;
                 fpu_do_push(emu);
                 memcpy(&STld(0).ld, ED, 10);
-                LD2D(&STld(0), &ST(0).d);
-                STld(0).ref = ST0.q;
+                LD2D(&STld(0).ld, &ST(0).d);
+                STld(0).uref = ST0.q;
                 break;
             case 7: /* FSTP tbyte */
                 GET_ED;
-                if(ST0.q!=STld(0).ref)
+                if(ST0.q!=STld(0).uref)
                     D2LD(&ST0.d, ED);
                 else
                     memcpy(ED, &STld(0).ld, 10);

src/emu/rundf.h

@@ -125,8 +125,8 @@
             tmp64s = *(int64_t*)ED;
             fpu_do_push(emu);
             ST0.d = tmp64s;
-            STll(0).ll = tmp64s;
-            STll(0).ref = ST0.q;
+            STll(0).sq = tmp64s;
+            STll(0).sref = ST0.sq;
             break;
         case 6: /* FBSTP tbytes, ST0 */
             GET_ED;
@@ -137,8 +137,8 @@
             GET_ED;
             if((uintptr_t)ED & 0x7) {
                 // un-aligned!
-                if(STll(0).ref==ST(0).q)
-                    memcpy(ED, &STll(0).ll, sizeof(int64_t));
+                if(STll(0).sref==ST(0).sq)
+                    memcpy(ED, &STll(0).sq, sizeof(int64_t));
                 else {
                     int64_t i64;
                     if(isgreater(ST0.d, (double)(int64_t)0x7fffffffffffffffLL) || isless(ST0.d, -(double)(int64_t)0x7fffffffffffffffLL) || !isfinite(ST0.d))
@@ -148,8 +148,8 @@
                     memcpy(ED, &i64, sizeof(int64_t));
                 }
             } else {
-                if(STll(0).ref==ST(0).q)
-                    *(int64_t*)ED = STll(0).ll;
+                if(STll(0).sref==ST(0).sq)
+                    *(int64_t*)ED = STll(0).sq;
                 else {
                     if(isgreater(ST0.d, (double)(int64_t)0x7fffffffffffffffLL) || isless(ST0.d, -(double)(int64_t)0x7fffffffffffffffLL) || !isfinite(ST0.d))
                         *(int64_t*)ED = 0x8000000000000000LL;

src/emu/runf20f.h

@@ -53,17 +53,32 @@
     case 0x51:  /* SQRTSD Gx, Ex */
         nextop = F8;
         GET_EX;
-        GX.d[0] = sqrt(EX->d[0]);
+        if(EX->d[0]<0.0 )
+            GX.d[0] = -NAN;
+        else
+            GX.d[0] = sqrt(EX->d[0]);
         break;
 
     case 0x58:  /* ADDSD Gx, Ex */
         nextop = F8;
         GET_EX;
+        #ifndef NOALIGN
+        // add generate a -NAN only if doing inf + -inf
+        if((isinf(GX.d[0]) && isinf(EX->d[0]) && (EX->q[0]&0x8000000000000000LL)!=(GX.q[0]&0x8000000000000000LL)))
+            GX.d[0] = -NAN;
+        else
+        #endif
         GX.d[0] += EX->d[0];
         break;
     case 0x59:  /* MULSD Gx, Ex */
         nextop = F8;
         GET_EX;
+        #ifndef NOALIGN
+        // mul generate a -NAN only if doing (+/-)inf * (+/-)0
+        if((isinf(GX.d[0]) && EX->d[0]==0.0) || (isinf(EX->d[0]) && GX.d[0]==0.0))
+            GX.d[0] = -NAN;
+        else
+        #endif
         GX.d[0] *= EX->d[0];
         break;
     case 0x5A:  /* CVTSD2SS Gx, Ex */
@@ -75,18 +90,31 @@
     case 0x5C:  /* SUBSD Gx, Ex */
         nextop = F8;
         GET_EX;
+        #ifndef NOALIGN
+            // sub generate a -NAN only if doing inf - inf
+            if((isinf(GX.d[0]) && isinf(EX->d[0]) && (EX->q[0]&0x8000000000000000LL)==(GX.q[0]&0x8000000000000000LL)))
+                GX.d[0] = -NAN;
+            else
+        #endif
         GX.d[0] -= EX->d[0];
         break;
     case 0x5D:  /* MINSD Gx, Ex */
         nextop = F8;
         GET_EX;
-        if (isnan(GX.d[0]) || isnan(EX->d[0]) || isless(EX->d[0], GX.d[0]))
+        if (isnan(GX.d[0]) || isnan(EX->d[0]) || (EX->d[0]<=GX.d[0]))
             GX.d[0] = EX->d[0];
         break;
     case 0x5E:  /* DIVSD Gx, Ex */
         nextop = F8;
         GET_EX;
+        #ifndef NOALIGN
+        is_nan = isnan(GX.d[0]) || isnan(EX->d[0]);
+        #endif
         GX.d[0] /= EX->d[0];
+        #ifndef NOALIGN
+        if(!is_nan && isnan(GX.d[0]))
+            GX.d[0] = -NAN;
+        #endif
         break;
     case 0x5F:  /* MAXSD Gx, Ex */
         nextop = F8;
@@ -218,8 +246,8 @@
         tmp8s = 0;
         switch(tmp8u&7) {
             case 0: tmp8s=(GX.d[0] == EX->d[0]); break;
-            case 1: tmp8s=isless(GX.d[0], EX->d[0]); break;
-            case 2: tmp8s=islessequal(GX.d[0], EX->d[0]); break;
+            case 1: tmp8s=isless(GX.d[0], EX->d[0]) && !(isnan(GX.d[0]) || isnan(EX->d[0])); break;
+            case 2: tmp8s=islessequal(GX.d[0], EX->d[0]) && !(isnan(GX.d[0]) || isnan(EX->d[0])); break;
             case 3: tmp8s=isnan(GX.d[0]) || isnan(EX->d[0]); break;
             case 4: tmp8s=(GX.d[0] != EX->d[0]); break;
             case 5: tmp8s=isnan(GX.d[0]) || isnan(EX->d[0]) || isgreaterequal(GX.d[0], EX->d[0]); break;

src/emu/runf30f.h

@@ -182,6 +182,24 @@
         }
         break;
 
+    case 0xBD:                      /* LZCNT Ed,Gd */
+        CHECK_FLAGS(emu);
+        nextop = F8;
+        GET_ED;
+        tmp32u = ED->dword[0];
+        if(tmp32u) {
+            tmp8u = 0;
+            while(!(tmp32u&(1<<(31-tmp8u)))) ++tmp8u;
+            GD.dword[0] = tmp8u;
+            CONDITIONAL_SET_FLAG(tmp8u==0, F_ZF);
+            CLEAR_FLAG(F_CF);
+        } else {
+            SET_FLAG(F_CF);
+            CLEAR_FLAG(F_ZF);
+            GD.dword[0] = 32;
+        }
+        break;
+
     case 0xC2:  /* CMPSS Gx, Ex, Ib */
         nextop = F8;
         GET_EX;
@@ -189,8 +207,8 @@
         tmp8s = 0;
         switch(tmp8u&7) {
             case 0: tmp8s=(GX.f[0] == EX->f[0]); break;
-            case 1: tmp8s=isless(GX.f[0], EX->f[0]); break;
-            case 2: tmp8s=islessequal(GX.f[0], EX->f[0]); break;
+            case 1: tmp8s=isless(GX.f[0], EX->f[0]) && !(isnan(GX.f[0]) || isnan(EX->f[0])); break;
+            case 2: tmp8s=islessequal(GX.f[0], EX->f[0]) && !(isnan(GX.f[0]) || isnan(EX->f[0])); break;
             case 3: tmp8s=isnan(GX.f[0]) || isnan(EX->f[0]); break;
             case 4: tmp8s=(GX.f[0] != EX->f[0]); break;
             case 5: tmp8s=isnan(GX.f[0]) || isnan(EX->f[0]) || isgreaterequal(GX.f[0], EX->f[0]); break;

src/emu/x86emu.c

@@ -49,13 +49,13 @@ uint32_t* GetParityTab()
 
 void PushExit(x86emu_t* emu)
 {
-    uintptr_t endMarker = AddCheckBridge(my_context->system, NULL, NULL, 0);
+    uintptr_t endMarker = AddCheckBridge(my_context->system, NULL, NULL, 0, "(ExitEmulation)");
     Push(emu, endMarker);
 }
 
 void* GetExit()
 {
-    return (void*)AddCheckBridge(my_context->system, NULL, NULL, 0);
+    return (void*)AddCheckBridge(my_context->system, NULL, NULL, 0, "(ExitEmulation)");
 }
 
 static void internalX86Setup(x86emu_t* emu, box86context_t *context, uintptr_t start, uintptr_t stack, int stacksize, int ownstack)
@@ -89,7 +89,7 @@ x86emu_t *NewX86Emu(box86context_t *context, uintptr_t start, uintptr_t stack, i
 {
     printf_log(LOG_DEBUG, "Allocate a new X86 Emu, with EIP=%p and Stack=%p/0x%X\n", (void*)start, (void*)stack, stacksize);
 
-    x86emu_t *emu = (x86emu_t*)calloc(1, sizeof(x86emu_t));
+    x86emu_t *emu = (x86emu_t*)box_calloc(1, sizeof(x86emu_t));
 
     internalX86Setup(emu, context, start, stack, stacksize, ownstack);
 
@@ -130,7 +130,7 @@ void AddCleanup(x86emu_t *emu, void *p)
 {
     if(my_context->clean_sz == my_context->clean_cap) {
         my_context->clean_cap += 4;
-        my_context->cleanups = (cleanup_t*)realloc(my_context->cleanups, sizeof(cleanup_t)*my_context->clean_cap);
+        my_context->cleanups = (cleanup_t*)box_realloc(my_context->cleanups, sizeof(cleanup_t)*my_context->clean_cap);
     }
     my_context->cleanups[my_context->clean_sz].arg = 0;
     my_context->cleanups[my_context->clean_sz].a = NULL;
@@ -141,7 +141,7 @@ void AddCleanup1Arg(x86emu_t *emu, void *p, void* a)
 {
     if(my_context->clean_sz == my_context->clean_cap) {
         my_context->clean_cap += 4;
-        my_context->cleanups = (cleanup_t*)realloc(my_context->cleanups, sizeof(cleanup_t)*my_context->clean_cap);
+        my_context->cleanups = (cleanup_t*)box_realloc(my_context->cleanups, sizeof(cleanup_t)*my_context->clean_cap);
     }
     my_context->cleanups[my_context->clean_sz].arg = 1;
     my_context->cleanups[my_context->clean_sz].a = a;
@@ -171,7 +171,7 @@ void CallAllCleanup(x86emu_t *emu)
         RunFunctionWithEmu(emu, 0, (uintptr_t)(my_context->cleanups[i].f), my_context->cleanups[i].arg, my_context->cleanups[i].a );
     }
     my_context->clean_sz = 0;
-    free(my_context->cleanups);
+    box_free(my_context->cleanups);
     my_context->cleanups = NULL;
 }
 
@@ -190,7 +190,7 @@ void FreeX86Emu(x86emu_t **emu)
 
     internalFreeX86(*emu);
 
-    free(*emu);
+    box_free(*emu);
     *emu = NULL;
 }
 
@@ -211,16 +211,15 @@ void CloneEmu(x86emu_t *newemu, const x86emu_t* emu)
     newemu->old_ip = emu->old_ip;
     memcpy(newemu->segs, emu->segs, sizeof(emu->segs));
     memset(newemu->segs_serial, 0, sizeof(newemu->segs_serial));
-	memcpy(newemu->mmx87, emu->mmx87, sizeof(emu->mmx87));
+	memcpy(newemu->x87, emu->x87, sizeof(emu->x87));
+	memcpy(newemu->mmx, emu->mmx, sizeof(emu->mmx));
     memcpy(newemu->fpu_ld, emu->fpu_ld, sizeof(emu->fpu_ld));
     memcpy(newemu->fpu_ll, emu->fpu_ll, sizeof(emu->fpu_ll));
 	memcpy(newemu->p_regs, emu->p_regs, sizeof(emu->p_regs));
 	newemu->cw = emu->cw;
-    newemu->cw_mask_all = emu->cw_mask_all;
-	memcpy(&newemu->sw, &emu->sw, sizeof(emu->sw));
+	newemu->sw = emu->sw;
 	newemu->top = emu->top;
     newemu->fpu_stack = emu->fpu_stack;
-	memcpy(&newemu->round, &emu->round, sizeof(emu->round));
     memcpy(newemu->xmm, emu->xmm, sizeof(emu->xmm));
     newemu->mxcsr = emu->mxcsr;
     newemu->quit = emu->quit;
@@ -308,7 +307,7 @@ const char* DumpCPURegs(x86emu_t* emu, uintptr_t ip)
     if(trace_emm) {
         // do emm reg is needed
         for(int i=0; i<8; ++i) {
-            sprintf(tmp, "mm%d:%016llx", i, emu->mmx87[i].q);
+            sprintf(tmp, "mm%d:%016llx", i, emu->mmx[i].q);
             strcat(buff, tmp);
             if ((i&3)==3) strcat(buff, "\n"); else strcat(buff, " ");
         }
@@ -323,13 +322,18 @@ const char* DumpCPURegs(x86emu_t* emu, uintptr_t ip)
     }
     // start with FPU regs...
     if(emu->fpu_stack) {
-        for (int i=0; i<emu->fpu_stack; i++) {
-            sprintf(tmp, "ST%d=%f", i, emu->mmx87[(emu->top+i)&7].d);
+        if(emu->fpu_stack<9)
+            for (int i=0; i<emu->fpu_stack; i++) {
+                //if(i==4) strcat(buff, "\n");
+                sprintf(tmp, "ST%d=%f", i, emu->x87[(emu->top+i)&7].d);
+                strcat(buff, tmp);
+                int c = 10-strlen(tmp);
+                if(c<1) c=1;
+                while(c--) strcat(buff, " ");
+            }
+        else {
+            sprintf(tmp, "Warning, incoherent fpu_stack=%d", emu->fpu_stack);
             strcat(buff, tmp);
-            int c = 10-strlen(tmp);
-            if(c<1) c=1;
-            while(c--) strcat(buff, " ");
-            if(i==3) strcat(buff, "\n");
         }
         strcat(buff, "\n");
     }
@@ -401,7 +405,8 @@ void EmuCall(x86emu_t* emu, uintptr_t addr)
     emu->quit = 0;  // reset Quit flags...
     emu->df = d_none;
     if(emu->quitonlongjmp && emu->longjmp) {
-        emu->longjmp = 0;   // don't change anything because of the longjmp
+        if(emu->quitonlongjmp==1)
+            emu->longjmp = 0;   // don't change anything because of the longjmp
     } else {
         R_EBX = old_ebx;
         R_EDI = old_edi;