Bumped version to v0.1.8

Fix pkgver in PKGBUILDS

.github/workflows/manual.yml

@@ -0,0 +1,95 @@
+### Manual build Box64 with Github Action
+name: Custom build Box64
+
+on:
+  workflow_dispatch:
+    inputs:
+      platform:
+        description: 'Target platform name'
+        required: true
+        default: 'GENERIC_ARM'
+        type: choice
+        options:
+        - X64
+        - GENERIC_ARM
+        - PHYTIUM
+        - RK3288
+        - RK3326
+        - RK3399
+        - RPI4ARM64
+        - SD845
+        - TEGRAX1
+      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 Box64 Repository"
+        uses: actions/checkout@v2
+
+      - name: "Environment preparation"
+        run: |
+          sudo apt-get update
+          if [[ ${{ github.event.inputs.platform }} != 'X64' && ${{ github.event.inputs.platform }} != 'GENERIC_ARM' ]]; then
+            echo "BOX64_PLATFORM_MARCRO=-D${{ github.event.inputs.platform }}=1" >> $GITHUB_ENV
+            echo "BOX64_COMPILER=aarch64-linux-gnu-gcc" >> $GITHUB_ENV
+            sudo apt-get -y install git gcc-aarch64-linux-gnu cmake make python3
+          else
+            if [[ ${{ github.event.inputs.platform }} == 'X64' ]]; then
+              echo "BOX64_PLATFORM_MARCRO=-DLD80BITS=1 -DNOALIGN=1" >> $GITHUB_ENV
+              echo "BOX64_COMPILER=gcc" >> $GITHUB_ENV
+              sudo apt-get -y install git cmake make python3
+            else
+              echo BOX64_PLATFORM_MARCRO="-DARM_DYNAREC=ON" >> $GITHUB_ENV
+              echo "BOX64_COMPILER=aarch64-linux-gnu-gcc" >> $GITHUB_ENV
+              sudo apt-get -y install git gcc-aarch64-linux-gnu cmake make python3
+            fi
+          fi
+
+      - name: "Display Build info"
+        run: |
+          echo "CMake Platform Macro: ${{ env.BOX64_PLATFORM_MARCRO }}"
+          echo "CMake C Compiler: ${{ env.BOX64_COMPILER }}"
+          echo "Build type: ${{ github.event.inputs.build_type }}"
+          echo "Dynarec Enabled: ${{ github.event.inputs.dynarec }}"
+          echo "Trace Enabled: ${{ github.event.inputs.trace }}"
+
+      - name: "Build Box64"
+        run: |
+          mkdir build
+          cd build
+          cmake .. -DCMAKE_C_COMPILER=${{ env.BOX64_COMPILER }} ${{ env.BOX64_PLATFORM_MARCRO }}\
+                   -DCMAKE_BUILD_TYPE=${{ 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$(nproc) VERBOSE=1
+
+      - name: "Upload Artifact"
+        uses: actions/upload-artifact@v2
+        with:
+          name: box64-${{ github.event.inputs.platform }}-${{ github.event.inputs.build_type }}
+          path: build/box64
+
+
+

.github/workflows/release.yml

@@ -1,7 +1,7 @@
 ### Auto build Box64 and release its binary with Github Action
 name: Build and Release Box64
 
-on: 
+on:
   workflow_dispatch:
   release:
   push:
@@ -10,16 +10,16 @@ on:
       - "**/*.h"
       - "**/*.S"
       - "**/*.py"
-      - "CMakeList.txt"
+      - "CMakeLists.txt"
       - "**/*.yml"
-  pull_request: 
+  pull_request:
     types: [assigned, opened, synchronize, reopened]
     paths:
       - "**/*.c"
       - "**/*.h"
       - "**/*.S"
       - "**/*.py"
-      - "CMakeList.txt"
+      - "CMakeLists.txt"
       - "**/*.yml"
 
 jobs:
@@ -27,7 +27,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        platform: [X64, OTHER_ARM, RPI4ARM64, RK3326, RK3399, RK3288, TEGRAX1, PHYTIUM, SD845]
+        platform: [X64, OTHER_ARM, RISCV, RPI4ARM64, RK3326, RK3399, RK3288, TEGRAX1, PHYTIUM, SD845]
         type: [Release, Trace]
 
     runs-on: ubuntu-latest
@@ -38,7 +38,7 @@ jobs:
       - name: "Environment preparation"
         run: |
           sudo apt-get update
-          if [[ ${{ matrix.platform }} != 'X64' && ${{ matrix.platform }} != 'OTHER_ARM' ]]; then
+          if [[ ${{ matrix.platform }} != 'X64' && ${{ matrix.platform }} != 'OTHER_ARM' && ${{ matrix.platform }} != 'RISCV' ]]; then
             echo "BOX64_PLATFORM_MARCRO=-D${{ matrix.platform }}=1" >> $GITHUB_ENV
             echo "BOX64_COMPILER=aarch64-linux-gnu-gcc" >> $GITHUB_ENV
             sudo apt-get -y install git gcc-aarch64-linux-gnu cmake make python3
@@ -47,7 +47,10 @@ jobs:
               echo "BOX64_PLATFORM_MARCRO=-DLD80BITS=1 -DNOALIGN=1" >> $GITHUB_ENV
               echo "BOX64_COMPILER=gcc" >> $GITHUB_ENV
               sudo apt-get -y install git cmake make python3
-              ls /bin/*gcc*
+            elif [[ ${{ matrix.platform }} == 'RISCV' ]]; then
+              echo BOX64_PLATFORM_MARCRO="-DRV64=ON" >> $GITHUB_ENV
+              echo "BOX64_COMPILER=riscv64-linux-gnu-gcc" >> $GITHUB_ENV
+              sudo apt-get -y install git gcc-riscv64-linux-gnu cmake make python3
             else
               echo BOX64_PLATFORM_MARCRO="-DARM_DYNAREC=ON" >> $GITHUB_ENV
               echo "BOX64_COMPILER=aarch64-linux-gnu-gcc" >> $GITHUB_ENV
@@ -61,23 +64,33 @@ jobs:
             echo BOX64_BUILD_TYPE=RelWithDebInfo >> $GITHUB_ENV
             echo BOX64_HAVE_TRACE=1 >> $GITHUB_ENV
           fi
-            
+
       - name: "Display Build info"
         run: |
           echo "CMake Platform Macro: ${{ env.BOX64_PLATFORM_MARCRO }}"
           echo "CMake C Compiler: ${{ env.BOX64_COMPILER }}"
           echo "Build type: ${{ env.BOX64_BUILD_TYPE }}"
           echo "Trace Enabled: ${{ env.BOX64_HAVE_TRACE }}"
-          
+
       - name: "Build Box64"
         run: |
           mkdir build
           cd build
-          cmake .. -DCMAKE_C_COMPILER=${{ env.BOX64_COMPILER }} ${{ env.BOX64_PLATFORM_MARCRO }} -DCMAKE_BUILD_TYPE=${{ env.BOX64_BUILD_TYPE }} -DHAVE_TRACE=${{ env.BOX64_HAVE_TRACE }} -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON
+          cmake .. -DCMAKE_C_COMPILER=${{ env.BOX64_COMPILER }}\
+                   ${{ env.BOX64_PLATFORM_MARCRO }}\
+                   -DCMAKE_BUILD_TYPE=${{ env.BOX64_BUILD_TYPE }}\
+                   -DHAVE_TRACE=${{ env.BOX64_HAVE_TRACE }}\
+                   -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON
           make -j$(nproc) VERBOSE=1
-      
+
+      - name: "Test Box64"
+        if: ${{ matrix.platform == 'X64' }}
+        run: |
+          cd build
+          ctest -j$(nproc)
+
       - name: "Upload Artifact"
         uses: actions/upload-artifact@v2
         with:
-          name: box86-${{ matrix.platform }}-${{ matrix.type }}
+          name: box64-${{ matrix.platform }}-${{ matrix.type }}
           path: build/box64

CMakeLists.txt

@@ -8,10 +8,15 @@ SET(CMAKE_CXX_FLAGS_DEBUG  "-O0 -g")
 option(RPI4ARM64 "Set to ON if targeting an RaspberryPI4 device with multiarch arm64 and armhf" ${RPI4ARM64})
 option(RK3326 "Set to ON if targeting an Rockchip RK3326 based device" ${RK3326})
 option(RK3399 "Set to ON if targeting an Rockchip RK3399 based device" ${RK3399})
+option(ODROIDN2 "Set to ON if targeting an Odroid-N2 device" ${ODROIDN2})
 option(TEGRAX1 "Set to ON if targeting an Tegra X1 based device" ${TEGRAX1})
 option(PHYTIUM "Set to ON if targeting an Phytium (D2000 or FT2000/4) based device" ${PHYTIUM})
 option(SD845 "Set to ON if targeting a Snapragon 845 based device" ${SD845})
+option(M1 "Set to ON if targeting a AppleM1 running on Asahi computer" ${M1})
 option(LARCH64 "Set to ON if targeting an Loongarch64 based device" ${LARCH64})
+option(RV64 "Set to ON if targeting an RISCV RV64G based device" ${RV64})
+option(PPC64LE "Set to ON if targeting an PowerPC 64 LE based device" ${PPC64LE})
+option(LX2160A "Set to ON if targeting an LX2160A based device" ${LX2160A})
 option(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 Box64" OFF)
@@ -19,16 +24,35 @@ option(NOGIT "Set to ON if not building from a git clone repo (like when buildin
 if(LARCH64)
     set(LD80BITS OFF CACHE BOOL "")
     set(NOALIGN OFF CACHE BOOL "")
-    set(ARM_DYNAREC OFF CACHE BOOL "")  
+    set(ARM_DYNAREC OFF CACHE BOOL "")
 endif()
-if(RK3399 OR RPI4ARM64 OR RK3326 OR TEGRAX1 OR PHYTIUM OR SD845)
+if(RV64)
+    set(LD80BITS OFF CACHE BOOL "")
+    set(NOALIGN OFF CACHE BOOL "")
+    set(ARM_DYNAREC OFF CACHE BOOL "")  
+    set(PAGE16K OFF CACHE BOOL "")
+endif()
+if(PPC64LE)
+    set(LD80BITS OFF CACHE BOOL "")
+    set(NOALIGN OFF CACHE BOOL "")
+    set(ARM_DYNAREC OFF CACHE BOOL "")
+    set(PAGE16K OFF CACHE BOOL "")
+endif()
+if(RK3399 OR ODROIDN2 OR RPI4ARM64 OR RK3326 OR TEGRAX1 OR PHYTIUM OR SD845 OR LX2160A OR M1)
     set(LD80BITS OFF CACHE BOOL "")
     set(NOALIGN OFF CACHE BOOL "")
     set(ARM_DYNAREC ON CACHE BOOL "")
 endif()
+if(RK3399 OR ODROIDN2 OR RPI4ARM64 OR RK3326 OR TEGRAX1 OR PHYTIUM OR SD845 OR LX2160A)
+    set(PAGE16K OFF CACHE BOOL "")
+endif()
+if(M1 OR LARCH64)
+    set(PAGE16K ON CACHE BOOL "")
+endif()
 option(LD80BITS "Set to ON if host device have 80bits long double (i.e. i386)" ${LD80BITS})
 option(NOALIGN "Set to ON if host device doesn't need re-align (i.e. i386)" ${NOALIGN})
 option(ARM_DYNAREC "Set to ON to use ARM Dynamic Recompilation" ${ARM_DYNAREC})
+option(PAGE16K "Set to ON if host device have PageSize of 16K (instead of 4K)" ${PAGE16K})
 
 if(${CMAKE_VERSION} VERSION_LESS "3.12.2")
     find_package(PythonInterp 3)
@@ -70,6 +94,10 @@ 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")
+elseif(ODROIDN2)
+    add_definitions(-DODROIDN2)
+    add_definitions(-march=armv8-a+crc+simd+crypto -mcpu=cortex-a73.cortex-a53+crypto)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crc+simd+crypto -mcpu=cortex-a73.cortex-a53+crypto")
 elseif(TEGRAX1)
     add_definitions(-DTEGRAX1)
     add_definitions(-pipe -march=armv8-a+crc+simd+crypto -mcpu=cortex-a57+crypto)
@@ -82,14 +110,32 @@ elseif(SD845)
     add_definitions(-DSD845)
     add_definitions(-pipe -march=armv8.2-a+simd+crypto -mtune=cortex-a75.cortex-a55)
     set(CMAKE_ASM_FLAGS  "-pipe -march=armv8.2-a+simd+crypto -mtune=cortex-a75.cortex-a55")
+elseif(M1)
+    add_definitions(-DM1)
+    add_definitions(-pipe -march=armv8.5-a+simd+crypto)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8.5-a+simd+crypto")
 elseif(LARCH64)
-    add_definitions(-DLARCH64)
+    add_definitions(-DLA464)
     add_definitions(-pipe -march=loongarch64)
     set(CMAKE_ASM_FLAGS  "-pipe -march=loongarch64")
+elseif(RV64)
+    add_definitions(-DRV64)
+    add_definitions(-pipe -march=rv64g)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=rv64g")
+elseif(PPC64LE)
+    add_definitions(-DPPC64LE)
+elseif(LX2160A)
+    add_definitions(-DLX2160A)
+    add_definitions(-pipe -march=armv8-a+crypto+crc -mcpu=cortex-a72+crypto)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crypto+crc -mcpu=cortex-a72+crypto")
 elseif(ARM_DYNAREC)
     set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crc+simd+crypto")
 endif()
 
+if(PAGE16K)
+    add_definitions(-DPAGE16K)
+endif()
+
 if(NOGIT)
     add_definitions(-DNOGIT)
 endif()
@@ -119,6 +165,10 @@ if(ARM_DYNAREC)
     add_definitions(-DDYNAREC)
     add_definitions(-DARM64)
     enable_language(ASM)
+    include_directories("${BOX64_ROOT}/src/dynarec/arm64")
+    set(DYNAREC ON)
+else()
+    set(DYNAREC OFF)
 endif()
 
 set(BOX64_ROOT ${CMAKE_SOURCE_DIR})
@@ -152,7 +202,9 @@ set(ELFLOADER_SRC
     "${BOX64_ROOT}/src/box64context.c"
     "${BOX64_ROOT}/src/build_info.c"
     "${BOX64_ROOT}/src/custommem.c"
+    "${BOX64_ROOT}/src/steam.c"
     "${BOX64_ROOT}/src/dynarec/dynarec.c"
+    "${BOX64_ROOT}/src/elfs/elfdwarf_private.c"
     "${BOX64_ROOT}/src/elfs/elfloader.c"
     "${BOX64_ROOT}/src/elfs/elfparser.c"
     "${BOX64_ROOT}/src/elfs/elfload_dump.c"
@@ -194,6 +246,7 @@ set(ELFLOADER_SRC
     "${BOX64_ROOT}/src/librarian/globalsymbols.c"
     "${BOX64_ROOT}/src/librarian/symbols.c"
     "${BOX64_ROOT}/src/libtools/auxval.c"
+    "${BOX64_ROOT}/src/libtools/obstack.c"
     "${BOX64_ROOT}/src/libtools/myalign.c"
     "${BOX64_ROOT}/src/libtools/sdl1rwops.c"
     "${BOX64_ROOT}/src/libtools/sdl2rwops.c"
@@ -224,19 +277,28 @@ set(WRAPPEDS
     "${BOX64_ROOT}/src/wrapped/wrappeddbus.c"
     "${BOX64_ROOT}/src/wrapped/wrappeddbusglib1.c"
     "${BOX64_ROOT}/src/wrapped/wrappedexpat.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedfaudio.c"
     "${BOX64_ROOT}/src/wrapped/wrappedflac.c"
     "${BOX64_ROOT}/src/wrapped/wrappedfontconfig.c"
     "${BOX64_ROOT}/src/wrapped/wrappedfreetype.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgbm.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgconf2.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgdkpixbuf2.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgdkx112.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgio2.c"
     "${BOX64_ROOT}/src/wrapped/wrappedglib2.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgmodule2.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgmp.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgnutls.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgobject2.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgssapi.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgssapikrb5.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstapp.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstaudio.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstbase.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstreamer.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgsttag.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstvideo.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgthread2.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgtkx112.c"
     "${BOX64_ROOT}/src/wrapped/wrappedkrb5.c"
@@ -246,6 +308,7 @@ set(WRAPPEDS
     "${BOX64_ROOT}/src/wrapped/wrappedlibcrypt.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibcups.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibdl.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibdrm.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibform.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibformw.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibfuse.c"
@@ -259,6 +322,7 @@ set(WRAPPEDS
     "${BOX64_ROOT}/src/wrapped/wrappedlibncursesw6.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibogg.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibpanel.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibpci.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibpcre.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibpthread.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibresolv.c"
@@ -268,12 +332,15 @@ set(WRAPPEDS
     "${BOX64_ROOT}/src/wrapped/wrappedlibssl.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibtinfo.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibtinfo6.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibusb1.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibuuid.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibvorbis.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibx11.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibx11xcb.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibxau.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibxcb.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibxcbdri2.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibxcbdri3.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibxcbglx.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibxcbkeysyms.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibxcbimage.c"
@@ -343,6 +410,7 @@ set(WRAPPEDS
     "${BOX64_ROOT}/src/wrapped/wrappedxslt.c"
     "${BOX64_ROOT}/src/wrapped/wrappedldapr.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlber.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedvulkan.c"
 )
 
 #set(WRAPPEDS_HEAD "${BOX64_ROOT}/src/wrapped/wrappedd3dadapter9_gen.h")
@@ -376,77 +444,87 @@ add_custom_command(
 add_custom_target(WRAPPERS DEPENDS "${BOX64_ROOT}/src/wrapped/generated/functions_list.txt")
 #add_custom_target(PRINTER DEPENDS "${BOX64_ROOT}/src/dynarec/last_run.txt")
 
-if(ARM_DYNAREC)
+if(DYNAREC)
     set(DYNAREC_SRC
         "${BOX64_ROOT}/src/dynarec/dynablock.c"
+        "${BOX64_ROOT}/src/dynarec/dynarec_native.c"
+    )
+endif()
 
-        "${BOX64_ROOT}/src/dynarec/dynarec_arm64.c"
-        "${BOX64_ROOT}/src/dynarec/dynarec_arm64_functions.c"
-        "${BOX64_ROOT}/src/dynarec/arm64_printer.c"
+if(ARM_DYNAREC)
+    set(DYNAREC_SRC
+        ${DYNAREC_SRC}
 
-        "${BOX64_ROOT}/src/dynarec/arm64_prolog.S"
-        "${BOX64_ROOT}/src/dynarec/arm64_epilog.S"
-        "${BOX64_ROOT}/src/dynarec/arm64_next.S"
-        "${BOX64_ROOT}/src/dynarec/arm64_lock.S"
+        "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_functions.c"
+        "${BOX64_ROOT}/src/dynarec/arm64/arm64_printer.c"
+
+        "${BOX64_ROOT}/src/dynarec/arm64/arm64_prolog.S"
+        "${BOX64_ROOT}/src/dynarec/arm64/arm64_epilog.S"
+        "${BOX64_ROOT}/src/dynarec/arm64/arm64_next.S"
+        "${BOX64_ROOT}/src/dynarec/arm64/arm64_lock.S"
     )
 
     set(DYNAREC_PASS
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_helper.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_emit_tests.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_emit_math.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_emit_logic.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_emit_shift.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_pass.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_00.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_0f.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_64.c"
-    #"${BOX64_ROOT}/src/dynarec/dynarec_arm64_65.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_66.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_67.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_d8.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_d9.c"
-    #"${BOX64_ROOT}/src/dynarec/dynarec_arm64_da.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_db.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_dc.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_dd.c"
-    #"${BOX64_ROOT}/src/dynarec/dynarec_arm64_de.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_df.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_f0.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_660f.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_6664.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_f20f.c"
-    "${BOX64_ROOT}/src/dynarec/dynarec_arm64_f30f.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_helper.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_emit_tests.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_emit_math.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_emit_logic.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_emit_shift.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_00.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_0f.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_64.c"
+    #"${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_65.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_66.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_67.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_d8.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_d9.c"
+    #"${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_da.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_db.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_dc.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_dd.c"
+    #"${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_de.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_df.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_f0.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_660f.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_6664.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_f20f.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_f30f.c"
+    )
+endif()
+
+if(DYNAREC)
+    set(DYNAREC_PASS 
+        "${BOX64_ROOT}/src/wrapped/generated/wrapper.h" 
+        ${DYNAREC_PASS}
+        "${BOX64_ROOT}/src/dynarec/dynarec_native_pass.c"
     )
 
-    set(DYNAREC_PASS "${BOX64_ROOT}/src/wrapped/generated/wrapper.h" ${DYNAREC_PASS})
+    add_library(dynarec_native OBJECT ${DYNAREC_SRC})
 
-    add_library(dynarec_arm OBJECT ${DYNAREC_SRC})
-
-    add_library(arm64_pass0 OBJECT ${DYNAREC_PASS})
-    set_target_properties(arm64_pass0 PROPERTIES COMPILE_FLAGS "-DSTEP=0")
-    add_library(arm64_pass1 OBJECT ${DYNAREC_PASS})
-    set_target_properties(arm64_pass1 PROPERTIES COMPILE_FLAGS "-DSTEP=1")
-    add_library(arm64_pass2 OBJECT ${DYNAREC_PASS})
-    set_target_properties(arm64_pass2 PROPERTIES COMPILE_FLAGS "-DSTEP=2")
-    add_library(arm64_pass3 OBJECT ${DYNAREC_PASS})
-    set_target_properties(arm64_pass3 PROPERTIES COMPILE_FLAGS "-DSTEP=3")
-    add_dependencies(arm64_pass0 WRAPPERS)
-    add_dependencies(arm64_pass1 WRAPPERS)
-    add_dependencies(arm64_pass2 WRAPPERS)
-    add_dependencies(arm64_pass3 WRAPPERS)
+    add_library(native_pass0 OBJECT ${DYNAREC_PASS})
+    set_target_properties(native_pass0 PROPERTIES COMPILE_FLAGS "-DSTEP=0")
+    add_library(native_pass1 OBJECT ${DYNAREC_PASS})
+    set_target_properties(native_pass1 PROPERTIES COMPILE_FLAGS "-DSTEP=1")
+    add_library(native_pass2 OBJECT ${DYNAREC_PASS})
+    set_target_properties(native_pass2 PROPERTIES COMPILE_FLAGS "-DSTEP=2")
+    add_library(native_pass3 OBJECT ${DYNAREC_PASS})
+    set_target_properties(native_pass3 PROPERTIES COMPILE_FLAGS "-DSTEP=3")
+    add_dependencies(native_pass0 WRAPPERS)
+    add_dependencies(native_pass1 WRAPPERS)
+    add_dependencies(native_pass2 WRAPPERS)
+    add_dependencies(native_pass3 WRAPPERS)
 
     add_library(dynarec STATIC
-        $<TARGET_OBJECTS:dynarec_arm>
-        $<TARGET_OBJECTS:arm64_pass0>
-        $<TARGET_OBJECTS:arm64_pass1>
-        $<TARGET_OBJECTS:arm64_pass2>
-        $<TARGET_OBJECTS:arm64_pass3>
+        $<TARGET_OBJECTS:dynarec_native>
+        $<TARGET_OBJECTS:native_pass0>
+        $<TARGET_OBJECTS:native_pass1>
+        $<TARGET_OBJECTS:native_pass2>
+        $<TARGET_OBJECTS:native_pass3>
     )
-
 endif()
 
 # creates git_head.h
-if(ARM_DYNAREC)
+if(DYNAREC)
     add_custom_command(
         OUTPUT "${BOX64_ROOT}/src/git_head.h"
         COMMAND bash -c "echo \\\#define GITREV \\\"$(git rev-parse --short HEAD)\\\">\"${BOX64_ROOT}/src/git_head.h\""
@@ -464,8 +542,9 @@ add_executable(${BOX64} ${ELFLOADER_SRC} ${WRAPPEDS} "${BOX64_ROOT}/src/git_head
 set_target_properties(${BOX64} PROPERTIES ENABLE_EXPORTS ON)
 add_dependencies(${BOX64} WRAPPERS)
 #add_dependencies(${BOX64} PRINTER)
-target_link_libraries(${BOX64} c m dl rt pthread)
-if(ARM_DYNAREC)
+#target_link_libraries(${BOX64} c m dl rt pthread resolv)
+set_target_properties(${BOX64} PROPERTIES LINK_FLAGS "-Wl,--no-as-needed -lc -lm -ldl -lrt -lpthread -lresolv -Wl,--as-needed -pthread")
+if(DYNAREC)
     target_link_libraries(${BOX64} dynarec)
 endif()
 
@@ -503,6 +582,7 @@ if(NOT _x86 AND NOT _x86_64)
   install(FILES ${CMAKE_BINARY_DIR}/system/box64.conf DESTINATION /etc/binfmt.d/)
   install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libstdc++.so.6 DESTINATION /usr/lib/x86_64-linux-gnu/)
   install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libgcc_s.so.1 DESTINATION /usr/lib/x86_64-linux-gnu/)
+  install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libpng12.so.0 DESTINATION /usr/lib/x86_64-linux-gnu/)
 endif()
 
 if(NOT TARGET uninstall)
@@ -545,101 +625,108 @@ endif()
 set(CPACK_DEBIAN_FILE_NAME "${BOX64}-${BOX64_MAJOR}.${BOX64_MINOR}.${BOX64_REVISION}_Linux-${CPACK_DEBIAN_PACKAGE_ARCHITECTURE}.deb")
 INCLUDE(CPack)
 
-add_test(test01 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test01 -D TEST_OUTPUT=tmpfile.txt 
+add_test(test01 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test01 -D TEST_OUTPUT=tmpfile01.txt
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref01.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-add_test(test02 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test02 -D TEST_OUTPUT=tmpfile.txt 
+add_test(test02 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test02 -D TEST_OUTPUT=tmpfile02.txt
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref02.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-add_test(test03 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test03 -D TEST_OUTPUT=tmpfile.txt 
+add_test(test03 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test03 -D TEST_OUTPUT=tmpfile03.txt
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref03.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-add_test(test04 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test04 -D TEST_ARGS2=yeah -D TEST_OUTPUT=tmpfile.txt 
+add_test(test04 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test04 -D TEST_ARGS2=yeah -D TEST_OUTPUT=tmpfile04.txt
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref04.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-add_test(test05 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test05 -D TEST_ARGS2=7 -D TEST_OUTPUT=tmpfile.txt 
+add_test(test05 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test05 -D TEST_ARGS2=7 -D TEST_OUTPUT=tmpfile05.txt
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref05.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-add_test(test06 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test06 -D TEST_OUTPUT=tmpfile.txt 
+add_test(test06 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test06 -D TEST_OUTPUT=tmpfile06.txt
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref06.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-add_test(test07 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test07 -D TEST_OUTPUT=tmpfile.txt 
+add_test(test07 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test07 -D TEST_OUTPUT=tmpfile07.txt
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref07.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-add_test(test08 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test08 -D TEST_OUTPUT=tmpfile.txt 
+add_test(test08 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test08 -D TEST_OUTPUT=tmpfile08.txt
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref08.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-add_test(test09 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test09 -D TEST_OUTPUT=tmpfile.txt 
+add_test(test09 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test09 -D TEST_OUTPUT=tmpfile09.txt
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref09.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-add_test(test10 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test10 -D TEST_OUTPUT=tmpfile.txt 
+add_test(test10 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test10 -D TEST_OUTPUT=tmpfile10.txt
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref10.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-add_test(test11 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test11 -D TEST_OUTPUT=tmpfile.txt 
+add_test(test11 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test11 -D TEST_OUTPUT=tmpfile11.txt
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref11.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-add_test(test12 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test12 -D TEST_OUTPUT=tmpfile.txt 
+add_test(test12 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test12 -D TEST_OUTPUT=tmpfile12.txt
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref12.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-add_test(test13 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test13 -D TEST_OUTPUT=tmpfile.txt 
+add_test(test13 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test13 -D TEST_OUTPUT=tmpfile13.txt
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref13.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-#add_test(test14 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-#    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test14 -D TEST_OUTPUT=tmpfile.txt 
+#add_test(test14 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+#    -D TEST_ARGS=${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 )
 
-add_test(test15 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test15 -D TEST_OUTPUT=tmpfile.txt 
+add_test(test15 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test15 -D TEST_OUTPUT=tmpfile15.txt
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref15.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-add_test(test16 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test16 -D TEST_OUTPUT=tmpfile.txt 
+add_test(test16 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${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(test17 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test17 -D TEST_OUTPUT=tmpfile.txt 
+add_test(sse ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${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 )
 
-add_test(aes ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64} 
-    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test18 -D TEST_OUTPUT=tmpfile.txt 
+set_tests_properties(sse PROPERTIES ENVIRONMENT "BOX64_DYNAREC_FASTNAN=0")
+    
+add_test(aes ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${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(backtrace ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test19 -D TEST_OUTPUT=tmpfile19.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref19.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
 file(GLOB extension_tests "${CMAKE_SOURCE_DIR}/tests/extensions/*.c")
 foreach(file ${extension_tests})
     get_filename_component(testname "${file}" NAME_WE)
     add_test(NAME "${testname}" COMMAND ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
-        -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/extensions/${testname} -D TEST_OUTPUT=tmpfile.txt
+        -D TEST_ARGS=${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()

debian/changelog

@@ -0,0 +1,29 @@
+box64 (0.1.8) unstable; urgency=low
+
+  * Added an option to get generated -NAN on SSE2 operation
+  * Many small fixes and improvement to get SteamPlay (i.e. proton) working (along with box86)
+  * Added some workaround for program calling uname to mimic x86_64 system
+  * Added some granularity to mmap, 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
+  *   - make adding new architecture easier, by sharing as much infrastructure as possible
+  * Improved elfloader
+  * Improved some tests
+  * Improved a bit the "Hotpage" handling and messages
+  * Added more dynarec opcodes
+  * A few fixes in some Dynarec opcodes
+  * A bit more optimisation in some Dynarec opcodes
+  * Don't quit on unsupported syscall, return -1 / ENOSYS error
+  * Added some more Syscall
+  * Wrapped Vulkan library. Vulkan and dxvk now supported!
+  * Support building on PPC64LE (interpretor only)
+  * Added support for Risc-V (interpretor only)
+
+ -- 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: box64
+Section: utils
+Priority: optional
+Maintainer: Ryan Fortner <[email protected]>
+Build-Depends: cmake, debhelper (>=11~), python3
+Standards-Version: 4.1.4
+Homepage: https://github.com/ptitSeb/box64
+
+Package: box64
+Architecture: any
+Multi-Arch: foreign
+Depends: ${misc:Depends}, ${shlibs:Depends}
+Description: Linux Userspace x86_64 Emulator with a twist, targeted at ARM Linux devices 
+ Box86 lets you run x86_64 Linux programs (such as games) on non-x86 Linux systems, like ARM (host system needs to be 64bit 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 @@
+box64: 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.6 

debian/source/format

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

debian/upstream/metadata

@@ -0,0 +1,5 @@
+Bug-Database: https://github.com/ptitSeb/box64/issues
+Bug-Submit: https://github.com/ptitSeb/box64/issues/new
+Repository: https://github.com/ptitSeb/box64.git
+Repository-Browse: https://github.com/ptitSeb/box64
+Security-Contact: [email protected]

debian/watch

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

docs/CHANGELOG.md

@@ -1,3 +1,31 @@
+v0.1.8
+======
+* Added an option to get generated -NAN on SSE2 operation
+* Many small fixes and improvement to get SteamPlay (i.e. proton) working (along with box86)
+* Added some workaround for program calling uname to mimic x86_64 system
+* Added some granularity to mmap, 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
+    * make adding new architecture easier, by sharing as much infrastructure as possible
+* Improved elfloader
+* Improved some tests
+* Improved a bit the "Hotpage" handling and messages
+* Added more dynarec opcodes
+* A few fixes in some Dynarec opcodes
+* A bit more optimisation in some Dynarec opcodes
+* Don't quit on unsupported syscall, return -1 / ENOSYS error
+* Added some more Syscall
+* Wrapped Vulkan library. Vulkan and dxvk now supported!
+* Support building on PPC64LE (interpretor only)
+* Added support for Risc-V (interpretor only)
+
 v0.1.6
 ======
 * Introduce "HotPage", to temporarily disable Dynarec on a page were writing is also occuring (can help speed up C# code)

docs/COMPILE.md

@@ -2,7 +2,7 @@ Compiling
 ----
 #### for RK3399
 
-Using a 64bits OS:
+Using a 64bit OS:
 ```
 git clone https://github.com/ptitSeb/box64
 cd box64
@@ -17,7 +17,7 @@ sudo systemctl restart systemd-binfmt
 
 #### for PI4
 
-Warning, you need a 64bits OS:
+Warning, you need a 64bit OS:
 
 ```
 git clone https://github.com/ptitSeb/box64
@@ -33,7 +33,7 @@ sudo systemctl restart systemd-binfmt
 
 #### for TEGRA X1
 
-Using a 64bits OS:
+Using a 64bit OS:
 
 ```
 git clone https://github.com/ptitSeb/box64
@@ -49,7 +49,7 @@ sudo systemctl restart systemd-binfmt
 
 #### for Snapdragon 845
 
-Using a 64bits OS:
+Using a 64bit OS:
 
 ```
 git clone https://github.com/ptitSeb/box64
@@ -67,7 +67,7 @@ sudo systemctl restart systemd-binfmt
 ----
 #### for Phytium
 
-Using a 64bits OS:
+Using a 64bit OS:
 ```
 git clone https://github.com/ptitSeb/box64
 cd box64
@@ -80,9 +80,30 @@ If it's the first install, you also need:
 sudo systemctl restart systemd-binfmt
 ```
 
+#### for M1
+
+Only test on Asahi for now, using the default "16K page" kernel
+
+```
+git clone https://github.com/ptitSeb/box64
+cd box64
+mkdir build; cd build; cmake .. -DM1=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
+make -j4
+sudo make install
+```
+If it's the first install, you also need:
+```
+sudo systemctl restart systemd-binfmt
+```
+
+
+#### for Other ARM64 Linux platforms
+
+ `mkdir build; cd build; cmake .. -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j$(nproc)`
+
 #### for LoongArch
 
-Using a 64bits OS:
+Using a 64bit OS:
 
 ```
 git clone https://github.com/ptitSeb/box64
@@ -96,15 +117,59 @@ If it's the first install, you also need:
 sudo systemctl restart systemd-binfmt
 ```
 
-#### for Other ARM64 Linux platforms
+#### for RISCV
 
- `mkdir build; cd build; cmake .. -DARM_DYNAREC=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j$(nproc)`
+Using a 64bit OS:
+
+```
+git clone https://github.com/ptitSeb/box64
+cd box64
+mkdir build; cd build; cmake .. -DRV64=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
+make -j4
+sudo make install
+```
+If it's the first install, you also need:
+```
+sudo systemctl restart systemd-binfmt
+```
+
+#### for PowerPC 64 LE
+
+Using a 64bit OS:
+
+```
+git clone https://github.com/ptitSeb/box64
+cd box64
+mkdir build; cd build; cmake .. -DPPC64LE=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
+make -j4
+sudo make install
+```
+If it's the first install, you also need:
+```
+sudo systemctl restart systemd-binfmt
+```
+
+#### for LX2160A
+
+Using a 64bit OS:
+
+```
+git clone https://github.com/ptitSeb/box64
+cd box64
+mkdir build; cd build; cmake .. -DLX2160A=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
+make -j$(nproc)
+sudo make install
+```
+If it's the first install, you also need:
+```
+sudo systemctl restart systemd-binfmt
+```
 
 #### for x86_64 Linux
 
  `mkdir build; cd build; cmake .. -DLD80BITS=1 -DNOALIGN=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo; make -j$(nproc)`
 
-If you encounter some linking errors, try activating `NOLOADADDR` (`cmake -DNOLOADADDR=ON; make -j$(nproc)`).
+If you encounter some linking errors, try using `NOLOADADDR=ON` (`cmake -DNOLOADADDR=ON; make -j$(nproc)`).
 
 ### use ccmake
 
@@ -112,30 +177,39 @@ Alternatively, you can **use the curses-bases ccmake (or any other gui frontend
 
 ### 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)
 
-*To 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 ( ***the interpretor will be slightly 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 to have 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`.
+Dynarec is only available on ARM (for the meantime), Activate it by using `-DARM_DYNAREC=1`.
+###### *Note: VFPv3 and NEON are required for Dynarec.*
 
-*Not building from a git clone*
+#### Not building from a git clone
 
-If you are not building from a git clone (for example, downloading a release source zipped from github), you need to activate `-DNOGIT=1` from cmake to be able to build (normal process include git sha1 of HEAD in the version that box64 print).
+If you are not building from a git clone (for example, downloading a release source code zip from github), you need to use `-DNOGIT=1` from cmake to be able to build (box64 uses git SHA1 to show last commit in version number).
 
 ----
 
 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 box64.
+
+They can be launched using the `ctest` command.
+
+The tests are very basic and only tests some functionality for now.
+
+----
+
+Debian Packaging
+----
+Box64 can also be packaged into a .deb file with `DEB_BUILD_OPTIONS=nostrip dpkg-buildpackage -us -uc -nc`.

docs/README.md

@@ -1,20 +1,20 @@
-# box64
-
 ![Official logo](img/Box64Logo.png "Official Logo")
+
 Linux Userspace x86_64 Emulator with a twist
 
+[View changelog](https://github.com/ptitSeb/box64/blob/main/docs/CHANGELOG.md) | [中文](https://github.com/ptitSeb/box64/blob/main/docs/README_CN.md) | [Report an error](https://github.com/ptitSeb/box64/issues/new)
+
+![build](https://app.travis-ci.com/ptitSeb/box64.svg?branch=main) ![stars](https://img.shields.io/github/stars/ptitSeb/box64) ![forks](https://img.shields.io/github/forks/ptitSeb/box64) ![contributors](https://img.shields.io/github/contributors/ptitSeb/box64) ![prs](https://img.shields.io/github/issues-pr/ptitSeb/box64) ![issues](https://img.shields.io/github/issues/ptitSeb/box64)
+
 ----
-![box64 build status](https://app.travis-ci.com/ptitSeb/box64.svg?branch=main)
 
-[中文](README_CN.md)
+Box64 lets you run x86_64 Linux programs (such as games) on non-x86_64 Linux systems, like ARM (host system needs to be 64-bit little-endian).
 
-Box64 lets you run x86_64 Linux programs (such as games) on non-x86_64 Linux systems, like ARM (host system needs to be 64bit little-endian).
+You can find many Box64 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.
 
-You can find many more Box64 video on Youtube in the [MicroLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ), [Pi Labs](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g) or [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA) Channels
+Since Box64 uses the native versions of some "system" libraries, like libc, libm, SDL, and OpenGL, it's easy to integrate and use with most applications, and performance can be surprisingly high in many cases. Take a look at thoses bench analysis for an example [here](https://box86.org/index.php/2021/06/game-performances/).
 
-Because Box64 uses the native versions of some "system" libraries, like libc, libm, SDL, and OpenGL, it's easy to integrate and use with most applications, and performance can be surprisingly high in many cases. Take a look at thoses bench analysis for an example [here](https://box86.org/index.php/2021/06/game-performances/).
-
-Box64 integrates a DynaRec (dynamic recompiler) for the ARM64 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/).
+Box64 integrates with DynaRec (dynamic recompiler) for the ARM64 platform, providing a speed boost between 5 to 10 times faster than using only the interpreter. Some high level information on how DynaRec works 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/).
 
 Some x64 internal opcodes use parts of "Realmode X86 Emulator Library", see [x64primop.c](../src/emu/x64primop.c) for copyright details
 
@@ -29,7 +29,7 @@ Usage
 
 There are a few environment variables to control the behaviour of Box64.
 
-See [here](USAGE.md) for all environment variables and what they do.
+See [here](USAGE.md) for all the environment variables and what they do.
 
 Note: Box64'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.
 
@@ -37,24 +37,24 @@ Note: Box64's Dynarec uses a mechanism with Memory Protection and a SegFault sig
 
 Compiling/Installation
 ----
-> Compilation instructions can be found [here](COMPILE.md)  
+> Compilation instructions can be found [here](COMPILE.md).
 
 ----
 
-Version history
+Version history/Change log
 ----
 
-The change log is available [here](CHANGELOG.md)
+The change log is available [here](CHANGELOG.md).
 
 ----
 
 Notes about 32-bit platforms
 ----
 
-Because Box64 works by directly translating function calls from x86_64 to host system, the host system (the one Box64 is running on) needs to have 64-bits libraries. Box64 doesn't include any 64-bits <-> 32-bits translation. 
+Because Box64 works by directly translating function calls from x86_64 to host system, the host system (the one Box64 is running on) needs to have 64-bit libraries. Box64 doesn't include any 64-bit <-> 32-bit translation. 
 
-So understand that box64 will only runs 64bits linux binaries. For 32bits binaries, you need box86 (with all the multiarch or proot trickery it imply on 64bits OS).
-Note that many installer (mojo setup based) will fall back to "x86" when detecting ARM64 OS, and so will try to use box86 for the setup, even if an x86_64 version exist. You can hack your way around with a fake "uname" that return "x86_64" when argument is "-m"
+So understand that box64 will only runs 64-bit linux binaries. For 32-bit binaries, you need box86 (with all the multiarch or proot trickery it imply on 64-bit OS).
+Note that many installer (mojo setup based) will fall back to "x86" when detecting ARM64 OS, and so will try to use box86 for the setup, even if an x86_64 version exist. You can hack your way around with a fake "uname" that return "x86_64" when the argument is "-m"
 
 ----
 
@@ -62,53 +62,53 @@ Notes about Unity game emulation
 ----
 
 Running Unity games is a hit or miss for now. Unity uses Mono (which uses signals that are not always emulated in a 100% accurate way).
-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.
-So, not all Unity games work and can require a high OpenGL profile, but many actualy works 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 start then quit before showing anything.
+You should also note that some Unity3D games require OpenGL 3+ which can be tricky to provide on ARM SBC (single-board computers).
+So, not all Unity games work and can require a high OpenGL profile, but many actualy work now.
+Hint: on Pi4, use `MESA_GL_VERSION_OVERRIDE=3.2` and with Panfrost use `PAN_MESA_DEBUG=gl3` to use higher profile if the game starts then quits before showing anything.
 
 ----
 
 Notes about GTK programs
 ----
 
-GTK libraries are not wrapped yet on box64 (contrary to box86). Many games use gtk (setup too), like Unity game for the Screen Selection dialog. While Unity game can run without that dialog (using default setup), many just can't. For now, you'll need o put the set of gtk x86_64 libraries in the game folder (or some other folder known by box64) to get gtk support. GTK Wrapping is planned on box64.
+GTK libraries are now wrapped on box64, bu only gtk2, not gtk3 (contrary to box86).
 
 ----
 
 Notes about Steam
 ----
 
-Not that Steam is an hybrid 32bits / 64bits. You NEED box86 to run steam, as the client app is a 32bits binaries. It also use a 64bits local server binaries, but like most stuff using libcef/chromium. it's not working correctly on box64 for now.
-So: no Steam for now on box64.
+Note that Steam is a hybrid 32-bit / 64-bit. You NEED box86 to run Steam, as the client app is a 32-bit binary. It also uses a 64-bit local server binaries, but like most stuff using libcef/chromium, it's not working correctly on box64 for now.
+So, no Steam for now on box64.
 
 ----
 
 Notes about Wine
 ----
 
-Wine 64bits is supported on box64. Be aware that 64bits wine also include 32bits componant, to be able to run 32bits windws programs. The 32bits will need box86 and will not run without it. On a system where both box64 and box86 are present and working, a wine 64bits setup can run both 32bits and 64bits windows program (just use `wine` and `wine64` respectivelly)
+Wine64 is supported on box64. Proton should also work. Be aware that 64-bit Wine also includes 32-bit components, to be able to run 32-bit Windows programs. The 32-bit apps will need box86 and will not run without it. On a system where both box64 and box86 are present and working, a wine 64-bit setup can run both 32-bit and 64-bit Windows programs (just use `wine` and `wine64` respectively)
 
 ----
 
 Notes about Vulkan
 ----
 
-Box64 Does not wrap Vulkan libraries yet. It's planned, but for now, vulkan program will not run on box64.
+Box64 wraps Vulkan libraries, but note that it as only been tested with a RX550 card, some extensions may be missing depending on your graphics card.
 
 ----
 
 Final word
 ----
 
-I want to thank everyone who has contributed to box64 developpement.
+I want to thank everyone who has contributed to box64 development.
 There are many ways to contribute: code contribution, financial, hardware and advertisement!
 So, in no particular order, I want to thank:
  * For their major code contribution: rajdakin, mogery
  * For their major financial contribution: FlyingFathead, stormchaser3000
  * For hardware contribution and LoongArch migration: [xiaoji](https://www.linuxgame.cn/), Deepin Beijing Develop Team
- * For their hardware contribution: [Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/)
+ * For their hardware contribution: [Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/), [StarFive](https://rvspace.org/)
  * For their continous advertisement of box64 project: salva ([microLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ)), [PILab](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g)/[TwisterOS](https://twisteros.com/) team, [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA), [NicoD](https://www.youtube.com/channel/UCpv7NFr0-9AB5xoklh3Snhg), ekianjo ([Boilingsteam](https://boilingsteam.com/))
 
-And I also thank the many other people who participated even once in the project.
+And I also thank the many other people who participated even once in this project.
 
 (If you use Box64 in your project, please don't forget to mention it!)

docs/README_CN.md

@@ -77,7 +77,7 @@ Box64的工作原理是直接将函数调用从x86_64转换到主机系统，主
 关于GTK程序的注意事项
 ----
 
-GTK库尚未封装在Box64内（与Box86相反）。许多游戏（包括设置）都使用gtk，例如Unity游戏的屏幕选择对话框。有些Unity游戏可以在没有该对话框的情况下运行（使用默认设置），还有一些不能。您需要将一套gtk x86_64库放在游戏目录（或其他一些Box64识别的目录）中以获得gtk支持。Box64有计划封装GTK原生库。 
+GTK 庫現在包裝在 box64 上，但只有 gtk2，而不是 gtk3（與 box86 相反）。
 
 ----
 
@@ -98,7 +98,7 @@ box64支持64位Wine，但请注意，64位的Wine还包含32位的组件，以
 关于Vulkan注意事项
 ----
 
-Box64尚未封装原生Vulkan库，有计划这么做，但目前，vulkan程序不能在box64上运行。
+Box64 包裝 Vulkan 庫。但請注意，它僅使用 RX550 卡進行了測試，根據您的圖形卡，可能會缺少某些擴展。
 
 ----
 
@@ -110,7 +110,7 @@ Box64尚未封装原生Vulkan库，有计划这么做，但目前，vulkan程序
 以下感谢不分先后： 
  * 代码贡献：rajdakin, mogery
  * 财务帮助：FlyingFathead, stormchaser3000
- * 硬件捐赠：[Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/)
+ * 硬件捐赠：[Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/), [StarFive](https://rvspace.org/)
  * 硬件捐赠和LoongArch迁移：[xiaoji](https://www.linuxgame.cn)
  * 为本项目持续的宣传：salva ([microLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ)), [PILab](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g)/[TwisterOS](https://twisteros.com/) team, [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA), [NicoD](https://www.youtube.com/channel/UCpv7NFr0-9AB5xoklh3Snhg), ekianjo ([Boilingsteam](https://boilingsteam.com/))
 

docs/USAGE.md

@@ -78,6 +78,11 @@ Disable handling of SigILL (to ease debugging mainly).
  * 0 : Let x86 program set sighandler for Illegal Instruction
  * 1 : Disables the handling of SigILL 
 
+#### BOX64_SHOWSEGV
+Show Segfault signal even if a signal handler is present
+ * 0 : Don"t force show the SIGSEGV analysis (Default.)
+ * 1 : Show SIGSEGV detail, even if a signal handler is present
+
 #### BOX64_X11THREADS
 Call XInitThreads when loading X11. (This is mostly for old Loki games with the Loki_Compat library.)
  * 0 : Don't force call XInitThreads. (Default.)
@@ -126,6 +131,16 @@ Enable/Disable simulation of Strong Memory model
 * 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
 
+#### BOX64_DYNAREC_X87DOUBLE
+Force the use of Double for x87 emulation
+* 0 : Try to use float when possible for x87 emulation (faster)
+* 1 : Only use Double for x87 emulation (slower, may be needed for some specific games, like Crysis)
+
+#### BOX64_DYNAREC_FASTNAN
+Enable/Disable generation of -NAN
+* 0 : Generate -NAN like on x86 (Default.)
+* 1 : Don't do anything special with NAN, to go as fast as possible (was default before this option exsted)
+
 #### BOX64_LIBGL
  * libXXXX set the name for libGL (defaults to libGL.so.1).
  * /PATH/TO/libGLXXX : Sets the name and path for libGL
@@ -145,13 +160,32 @@ Allow Box64 to continue even if a library is missing.
  * 0 : Box64 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.
 
+#### BOX64_PREFER_WRAPPED
+Box64 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
+
+#### BOX64_PREFER_EMULATED
+Box64 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)
+
 #### BOX64_NOPULSE
 Disables the load of pulseaudio libraries.
  * 0 : Load pulseaudio libraries if found. (Default.)
  * 1 : Disables the load of pulse audio libraries (libpulse and libpulse-simple), both the native library and the x86 library
 
-#### BOX64_JITGDB
+#### BOX64_NOGTK
+Disables the loading of wrapped GTK libraries.
+ * 0 : Load wrapped GTK libraries if found. (Default.)
+ * 1 : Disables loading wrapped GTK libraries.
 
+#### BOX64_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.)
+
+#### BOX64_JITGDB
  * 0 : Just print the Segfault message on segfault (default)
  * 1 : Launch `gdb` when a segfault, bus error or illegal instruction signal is trapped, attached to the offending process and go in an endless loop, waiting.
  When in gdb, you need to find the correct thread yourself (the one with `my_box64signalhandler` in is stack)

pkgbuilds/PKGBUILD-arm64

@@ -1,7 +1,7 @@
-# Maintainer: Jai-JAP <parjailu@gmail.com>
+# Maintainer: Jai-JAP <jai.jap.318@gmail.com>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box64-arm64-git
-pkgver=1116.9243ea4
+pkgver=r1495.e6e9fae
 pkgrel=1
 pkgdesc="Linux Userspace x86_64 Emulator with a twist, targeted at ARM64 Linux devices."
 arch=('aarch64')

pkgbuilds/PKGBUILD-arm64-DEBUG

@@ -1,7 +1,7 @@
-# Maintainer: Jai-JAP <parjailu@gmail.com>
+# Maintainer: Jai-JAP <jai.jap.318@gmail.com>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box64-arm64-git
-pkgver=1116.9243ea4
+pkgver=r1495.e6e9fae
 pkgrel=1
 pkgdesc="Linux Userspace x86_64 Emulator with a twist, targeted at ARM64 Linux devices."
 arch=('aarch64')

pkgbuilds/PKGBUILD-lx2160a

@@ -0,0 +1,44 @@
+# Maintainer: Jai-JAP <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box64-lx2160a-git
+pkgver=r1495.e6e9fae
+pkgrel=1
+pkgdesc="Linux Userspace x86_64 Emulator with a twist, targeted at ARM64 Linux devices."
+arch=('aarch64')
+url="https://github.com/ptitSeb/box64"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-lx2160a-git}-git" "${pkgname%-lx2160a-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-lx2160a-git}-git" "${pkgname%-lx2160a-git}")
+source=('git+https://github.com/ptitSeb/box64')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "$srcdir/${pkgname%-lx2160a-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-lx2160a-git}"
+	if [[ ! -d ./build ]]; then
+		mkdir build && cd build
+		cmake .. -DLX2160A=1 -DCMAKE_BUILD_TYPE=RelWithDebInfo
+	fi
+}
+
+build() {
+	cd "$srcdir/${pkgname%-lx2160a-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "$srcdir/${pkgname%-lx2160a-git}/build"
+	make DESTDIR="${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
+}

pkgbuilds/PKGBUILD-lx2160a-DEBUG

@@ -0,0 +1,44 @@
+# Maintainer: Jai-JAP <[email protected]>
+# Author: Sebastien Chevalier <[email protected]>
+pkgname=box64-lx2160a-git
+pkgver=r1495.e6e9fae
+pkgrel=1
+pkgdesc="Linux Userspace x86_64 Emulator with a twist, targeted at ARM64 Linux devices."
+arch=('aarch64')
+url="https://github.com/ptitSeb/box64"
+license=('MIT')
+optdepends=('gl4es: OpenGL 2 for GLES 2 devices')
+makedepends=('git' 'cmake' 'make')
+provides=("${pkgname%-git}" "${pkgname%-lx2160a-git}-git" "${pkgname%-lx2160a-git}")
+conflicts=("${pkgname%-git}" "${pkgname%-lx2160a-git}-git" "${pkgname%-lx2160a-git}")
+source=('git+https://github.com/ptitSeb/box64')
+md5sums=('SKIP')
+
+pkgver() {
+	cd "$srcdir/${pkgname%-lx2160a-git}"
+	printf "r%s.%s" "$(git rev-list --count HEAD)" "$(git rev-parse --short HEAD)"
+}
+
+prepare() {
+	cd "${srcdir}/${pkgname%-lx2160a-git}"
+	if [[ ! -d ./build ]]; then
+		mkdir build && cd build
+		cmake .. -DLX2160A=1 -DCMAKE_BUILD_TYPE=Debug
+	fi
+}
+
+build() {
+	cd "$srcdir/${pkgname%-lx2160a-git}/build"
+	make -j$(nproc)
+}
+
+package() {
+	cd "$srcdir/${pkgname%-lx2160a-git}/build"
+	make DESTDIR="${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
+}

pkgbuilds/PKGBUILD-rk3399

@@ -1,7 +1,7 @@
-# Maintainer: Jai-JAP <parjailu@gmail.com>
+# Maintainer: Jai-JAP <jai.jap.318@gmail.com>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-rk3399-git
-pkgver=1116.9243ea4
+pkgver=r1495.e6e9fae
 pkgrel=1
 pkgdesc="Linux Userspace x86_64 Emulator with a twist, targeted at ARM64 Linux devices."
 arch=('aarch64')
@@ -41,4 +41,4 @@ package() {
 	mv usr/local/bin/ usr/bin/
 	# cleanup when dir is empty
 	rmdir usr/local || exit 0
-}
+}

pkgbuilds/PKGBUILD-rk3399-DEBUG

@@ -1,7 +1,7 @@
-# Maintainer: Jai-JAP <parjailu@gmail.com>
+# Maintainer: Jai-JAP <jai.jap.318@gmail.com>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box86-rk3399-git
-pkgver=1116.9243ea4
+pkgver=r1495.e6e9fae
 pkgrel=1
 pkgdesc="Linux Userspace x86_64 Emulator with a twist, targeted at ARM64 Linux devices."
 arch=('aarch64')

pkgbuilds/PKGBUILD-rpi4

@@ -1,7 +1,7 @@
-# Maintainer: Jai-JAP <parjailu@gmail.com>
+# Maintainer: Jai-JAP <jai.jap.318@gmail.com>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box64-rpi-git
-pkgver=1116.9243ea4
+pkgver=r1495.e6e9fae
 pkgrel=1
 pkgdesc="Linux Userspace x86_64 Emulator with a twist, targeted at ARM64 Linux devices."
 arch=('aarch64')

pkgbuilds/PKGBUILD-rpi4-DEBUG

@@ -1,7 +1,7 @@
-# Maintainer: Jai-JAP <parjailu@gmail.com>
+# Maintainer: Jai-JAP <jai.jap.318@gmail.com>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box64-rpi-git
-pkgver=1116.9243ea4
+pkgver=r1495.e6e9fae
 pkgrel=1
 pkgdesc="Linux Userspace x86_64 Emulator with a twist, targeted at ARM64 Linux devices."
 arch=('aarch64')

pkgbuilds/PKGBUILD-tegra

@@ -1,7 +1,7 @@
-# Maintainer: Jai-JAP <parjailu@gmail.com>
+# Maintainer: Jai-JAP <jai.jap.318@gmail.com>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box64-tegra-git
-pkgver=1116.9243ea4
+pkgver=r1495.e6e9fae
 pkgrel=1
 pkgdesc="Linux Userspace x86_64 Emulator with a twist, targeted at ARM64 Linux devices."
 arch=('aarch64')

pkgbuilds/PKGBUILD-tegra-DEBUG

@@ -1,7 +1,7 @@
-# Maintainer: Jai-JAP <parjailu@gmail.com>
+# Maintainer: Jai-JAP <jai.jap.318@gmail.com>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box64-tegra-git
-pkgver=1116.9243ea4
+pkgver=r1495.e6e9fae
 pkgrel=1
 pkgdesc="Linux Userspace x86_64 Emulator with a twist, targeted at ARM64 Linux devices."
 arch=('aarch64')

pkgbuilds/PKGBUILD-x86_64

@@ -1,7 +1,7 @@
-# Maintainer: Jai-JAP <parjailu@gmail.com>
+# Maintainer: Jai-JAP <jai.jap.318@gmail.com>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box64-x64-git
-pkgver=1116.9243ea4
+pkgver=r1495.e6e9fae
 pkgrel=1
 pkgdesc="Linux Userspace x86_64 Emulator with a twist, targeted at ARM64 Linux devices."
 arch=('aarch64')

pkgbuilds/PKGBUILD-x86_64-DEBUG

@@ -1,7 +1,7 @@
-# Maintainer: Jai-JAP <parjailu@gmail.com>
+# Maintainer: Jai-JAP <jai.jap.318@gmail.com>
 # Author: Sebastien Chevalier <[email protected]>
 pkgname=box64-x64-git
-pkgver=1116.9243ea4
+pkgver=r1495.e6e9fae
 pkgrel=1
 pkgdesc="Linux Userspace x86_64 Emulator with a twist, targeted at ARM64 Linux devices."
 arch=('aarch64')

pkgbuilds/README.md

@@ -12,13 +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:
+### 3. Any advantages with this solution?
+This might give you some advantages like:
 - 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 `box64` 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 box64 to .deb or .rpm installer files.
+### Note: You can use [makedeb](https://github.com/makedeb/makedeb) (or) [pkgbuild2spec](https://github.com/prozum/pkgbuild2spec) to compile box64 to .deb or .rpm installer files.

runTest.cmake

@@ -32,6 +32,9 @@ execute_process(
 
 # if the return value is !=0 bail out
 if( TEST_RESULT )
+  get_filename_component(TESTNAME "${TEST_ARGS}" NAME)
+  file(RENAME "${TEST_OUTPUT}" "${CMAKE_BINARY_DIR}/${TESTNAME}.out")
+  file(WRITE  "${CMAKE_BINARY_DIR}/${TESTNAME}.err" ${TEST_ERROR})
   message( FATAL_ERROR "Failed: Test program ${TEST_PROGRAM} exited != 0.\n${TEST_ERROR}" )
 endif( TEST_RESULT )
 
@@ -43,8 +46,16 @@ execute_process(
 
 # again, if return value is !=0 scream and shout
 if( TEST_RESULT )
+  get_filename_component(TESTNAME "${TEST_ARGS}" NAME)
+  file(RENAME "${TEST_OUTPUT}" "${CMAKE_BINARY_DIR}/${TESTNAME}.out")
+  file(WRITE  "${CMAKE_BINARY_DIR}/${TESTNAME}.err" ${TEST_ERROR})
   message( FATAL_ERROR "Failed: The output of ${TEST_PROGRAM} did not match ${TEST_REFERENCE}")
 endif( TEST_RESULT )
 
+# remove the temporary files if they exist
+if( EXISTS ${TEST_OUTPUT} )
+  file(REMOVE "${TEST_OUTPUT}")
+endif()
+
 # everything went fine...
 message( "Passed: The output of ${TEST_PROGRAM} matches ${TEST_REFERENCE}" )

src/box64context.c

@@ -5,6 +5,7 @@
 #include <dlfcn.h>
 #include <signal.h>
 #include <sys/mman.h>
+#include <pthread.h>
 
 #include "box64context.h"
 #include "debug.h"
@@ -259,6 +260,9 @@ void FreeBox64Context(box64context_t** context)
     freeGLProcWrapper(ctx);
     freeALProcWrapper(ctx);
 
+    if(ctx->stack_clone)
+        free(ctx->stack_clone);
+
 
     void* ptr;
     if ((ptr = pthread_getspecific(ctx->tlskey)) != NULL) {

src/box64version.h

@@ -3,6 +3,6 @@
 
 #define BOX64_MAJOR 0
 #define BOX64_MINOR 1
-#define BOX64_REVISION 6
+#define BOX64_REVISION 8
 
 #endif //__BOX64_VERSION_H_

src/custommem.c

@@ -4,6 +4,7 @@
 #include <string.h>
 #include <dlfcn.h>
 #include <signal.h>
+#include <pthread.h>
 
 #include "box64context.h"
 #include "elfloader.h"
@@ -24,7 +25,7 @@
 #include "threads.h"
 #ifdef DYNAREC
 #include "dynablock.h"
-#include "dynarec/arm64_lock.h"
+#include "dynarec/native_lock.h"
 
 //#define USE_MMAP
 
@@ -41,13 +42,25 @@ static uintptr_t*          box64_jmptbldefault1[1<<JMPTABL_SHIFT];
 static uintptr_t           box64_jmptbldefault0[1<<JMPTABL_SHIFT];
 #endif
 static pthread_mutex_t     mutex_prot;
+#if defined(PAGE16K)
+#define MEMPROT_SHIFT 14
+#define MEMPROT_SHIFT2 (16+14)
+#else
 #define MEMPROT_SHIFT 12
 #define MEMPROT_SHIFT2 (16+12)
+#endif
 #define MEMPROT_SIZE (1<<16)
 static uint8_t *volatile memprot[1<<20];    // x86_64 mem is 48bits, page is 12bits, so memory is tracked as [20][16][page protection]
 static uint8_t memprot_default[MEMPROT_SIZE];
 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;
     size_t              maxfree;
@@ -579,17 +592,17 @@ void addDBFromAddressRange(uintptr_t addr, size_t size)
         int idx1 = (i    )&((1<<DYNAMAP_SHIFT)-1);
         if(!dynmap123[idx3]) {
             dynablocklist_t*** p = (dynablocklist_t***)calloc(1<<DYNAMAP_SHIFT, sizeof(dynablocklist_t**));
-            if(arm64_lock_storeifnull(&dynmap123[idx3], p)!=p)
+            if(native_lock_storeifnull(&dynmap123[idx3], p)!=p)
                 free(p);
         }
         if(!dynmap123[idx3][idx2]) {
             dynablocklist_t** p = (dynablocklist_t**)calloc(1<<DYNAMAP_SHIFT, sizeof(dynablocklist_t*));
-            if(arm64_lock_storeifnull(&dynmap123[idx3][idx2], p)!=p)
+            if(native_lock_storeifnull(&dynmap123[idx3][idx2], p)!=p)
                 free(p);
         }
         if(!dynmap123[idx3][idx2][idx1]) {
             dynablocklist_t* p = NewDynablockList(i<<DYNAMAP_SHIFT, 1<<DYNAMAP_SHIFT, 0);
-            if(arm64_lock_storeifnull(&dynmap123[idx3][idx2][idx1], p)!=p)
+            if(native_lock_storeifnull(&dynmap123[idx3][idx2][idx1], p)!=p)
                 FreeDynablockList(&p);
         }
     }
@@ -618,13 +631,13 @@ void cleanDBFromAddressRange(uintptr_t addr, size_t size, int destroy)
                 if(destroy) {
                     if(FreeRangeDynablock(dblist, addr, size) && 0) {    // dblist is empty, check if we can delete more...
                         // disabling this for now. It seems to cause random crash in Terraria
-                        if(!arm64_lock_storeifref(&dynmap123[idx3][idx2][idx1], NULL, dblist)) {
+                        if(!native_lock_storeifref(&dynmap123[idx3][idx2][idx1], NULL, dblist)) {
                             dynablocklist_t** p = dynmap123[idx3][idx2];
                             if(dynmapempty((void**)p)) {
-                                if(!arm64_lock_storeifref(&dynmap123[idx3][idx2], NULL, p)) {
+                                if(!native_lock_storeifref(&dynmap123[idx3][idx2], NULL, p)) {
                                     dynablocklist_t*** p2 = dynmap123[idx3];
                                     if(dynmapempty((void**)p2)) {
-                                        if(!arm64_lock_storeifref(&dynmap123[idx3], NULL, p2)) {
+                                        if(!native_lock_storeifref(&dynmap123[idx3], NULL, p2)) {
                                             free(p2);
                                         }
                                     }
@@ -643,6 +656,12 @@ void cleanDBFromAddressRange(uintptr_t addr, size_t size, int destroy)
 
 #ifdef ARM64
 void arm64_next(void);
+#define native_next arm64_next
+#elif defined(LA464)
+void la464_next(void);
+#define native_next la464_next
+#else
+#error Unsupported architecture
 #endif
 
 void addJumpTableIfDefault64(void* addr, void* jmp)
@@ -656,25 +675,25 @@ void addJumpTableIfDefault64(void* addr, void* jmp)
         uintptr_t*** tbl = (uintptr_t***)malloc((1<<JMPTABL_SHIFT)*sizeof(uintptr_t**));
         for(int i=0; i<(1<<JMPTABL_SHIFT); ++i)
             tbl[i] = box64_jmptbldefault1;
-        if(arm64_lock_storeifref(&box64_jmptbl3[idx3], tbl, box64_jmptbldefault2)!=tbl)
+        if(native_lock_storeifref(&box64_jmptbl3[idx3], tbl, box64_jmptbldefault2)!=tbl)
             free(tbl);
     }
     if(box64_jmptbl3[idx3][idx2] == box64_jmptbldefault1) {
         uintptr_t** tbl = (uintptr_t**)malloc((1<<JMPTABL_SHIFT)*sizeof(uintptr_t*));
         for(int i=0; i<(1<<JMPTABL_SHIFT); ++i)
             tbl[i] = box64_jmptbldefault0;
-        if(arm64_lock_storeifref(&box64_jmptbl3[idx3][idx2], tbl, box64_jmptbldefault1)!=tbl)
+        if(native_lock_storeifref(&box64_jmptbl3[idx3][idx2], tbl, box64_jmptbldefault1)!=tbl)
             free(tbl);
     }
     if(box64_jmptbl3[idx3][idx2][idx1] == box64_jmptbldefault0) {
         uintptr_t* tbl = (uintptr_t*)malloc((1<<JMPTABL_SHIFT)*sizeof(uintptr_t));
         for(int i=0; i<(1<<JMPTABL_SHIFT); ++i)
-            tbl[i] = (uintptr_t)arm64_next;
-        if(arm64_lock_storeifref(&box64_jmptbl3[idx3][idx2][idx1], tbl, box64_jmptbldefault0)!=tbl)
+            tbl[i] = (uintptr_t)native_next;
+        if(native_lock_storeifref(&box64_jmptbl3[idx3][idx2][idx1], tbl, box64_jmptbldefault0)!=tbl)
             free(tbl);
     }
 
-    arm64_lock_storeifref(&box64_jmptbl3[idx3][idx2][idx1][idx0], jmp, arm64_next);
+    native_lock_storeifref(&box64_jmptbl3[idx3][idx2][idx1][idx0], jmp, native_next);
 }
 void setJumpTableDefault64(void* addr)
 {
@@ -689,9 +708,9 @@ void setJumpTableDefault64(void* addr)
         return;
     if(box64_jmptbl3[idx3][idx2][idx1] == box64_jmptbldefault0)
         return;
-    if(box64_jmptbl3[idx3][idx2][idx1][idx0]==(uintptr_t)arm64_next)
+    if(box64_jmptbl3[idx3][idx2][idx1][idx0]==(uintptr_t)native_next)
         return;
-    box64_jmptbl3[idx3][idx2][idx1][idx0] = (uintptr_t)arm64_next;
+    box64_jmptbl3[idx3][idx2][idx1][idx0] = (uintptr_t)native_next;
 }
 int isJumpTableDefault64(void* addr)
 {
@@ -706,9 +725,9 @@ int isJumpTableDefault64(void* addr)
         return 1;
     if(box64_jmptbl3[idx3][idx2][idx1] == box64_jmptbldefault0)
         return 1;
-    if(box64_jmptbl3[idx3][idx2][idx1][idx0]==(uintptr_t)arm64_next)
+    if(box64_jmptbl3[idx3][idx2][idx1][idx0]==(uintptr_t)native_next)
         return 1;
-    return (box64_jmptbl3[idx3][idx2][idx1][idx0]==(uintptr_t)arm64_next)?1:0;
+    return (box64_jmptbl3[idx3][idx2][idx1][idx0]==(uintptr_t)native_next)?1:0;
 }
 uintptr_t getJumpTable64()
 {
@@ -726,21 +745,21 @@ uintptr_t getJumpTableAddress64(uintptr_t addr)
         uintptr_t*** tbl = (uintptr_t***)malloc((1<<JMPTABL_SHIFT)*sizeof(uintptr_t**));
         for(int i=0; i<(1<<JMPTABL_SHIFT); ++i)
             tbl[i] = box64_jmptbldefault1;
-        if(arm64_lock_storeifref(&box64_jmptbl3[idx3], tbl, box64_jmptbldefault2)!=tbl)
+        if(native_lock_storeifref(&box64_jmptbl3[idx3], tbl, box64_jmptbldefault2)!=tbl)
             free(tbl);
     }
     if(box64_jmptbl3[idx3][idx2] == box64_jmptbldefault1) {
         uintptr_t** tbl = (uintptr_t**)malloc((1<<JMPTABL_SHIFT)*sizeof(uintptr_t*));
         for(int i=0; i<(1<<JMPTABL_SHIFT); ++i)
             tbl[i] = box64_jmptbldefault0;
-        if(arm64_lock_storeifref(&box64_jmptbl3[idx3][idx2], tbl, box64_jmptbldefault1)!=tbl)
+        if(native_lock_storeifref(&box64_jmptbl3[idx3][idx2], tbl, box64_jmptbldefault1)!=tbl)
             free(tbl);
     }
     if(box64_jmptbl3[idx3][idx2][idx1] == box64_jmptbldefault0) {
         uintptr_t* tbl = (uintptr_t*)malloc((1<<JMPTABL_SHIFT)*sizeof(uintptr_t));
         for(int i=0; i<(1<<JMPTABL_SHIFT); ++i)
-            tbl[i] = (uintptr_t)arm64_next;
-        if(arm64_lock_storeifref(&box64_jmptbl3[idx3][idx2][idx1], tbl, box64_jmptbldefault0)!=tbl)
+            tbl[i] = (uintptr_t)native_next;
+        if(native_lock_storeifref(&box64_jmptbl3[idx3][idx2][idx1], tbl, box64_jmptbldefault0)!=tbl)
             free(tbl);
     }
 
@@ -753,8 +772,8 @@ void protectDB(uintptr_t addr, uintptr_t size)
     dynarec_log(LOG_DEBUG, "protectDB %p -> %p\n", (void*)addr, (void*)(addr+size-1));
     uintptr_t idx = (addr>>MEMPROT_SHIFT);
     uintptr_t end = ((addr+size-1LL)>>MEMPROT_SHIFT);
-    if(end>=(1LL<<(20+16)))
-        end = (1LL<<(20+16))-1;
+    if(end>=(1LL<<(48-MEMPROT_SHIFT)))
+        end = (1LL<<(48-MEMPROT_SHIFT))-1;
     if(end<idx) // memory addresses higher than 48bits are not tracked
         return;
     pthread_mutex_lock(&mutex_prot);
@@ -762,7 +781,7 @@ void protectDB(uintptr_t addr, uintptr_t size)
     for (uintptr_t i=(idx>>16); i<=(end>>16); ++i)
         if(memprot[i]==memprot_default) {
             uint8_t* newblock = calloc(1<<16, sizeof(uint8_t));
-            /*if (arm64_lock_storeifref(&memprot[i], newblock, memprot_default) != newblock) {
+            /*if (native_lock_storeifref(&memprot[i], newblock, memprot_default) != newblock) {
                 free(newblock);
             }*/
             memprot[i] = newblock;
@@ -787,15 +806,15 @@ void unprotectDB(uintptr_t addr, size_t size)
     dynarec_log(LOG_DEBUG, "unprotectDB %p -> %p\n", (void*)addr, (void*)(addr+size-1));
     uintptr_t idx = (addr>>MEMPROT_SHIFT);
     uintptr_t end = ((addr+size-1)>>MEMPROT_SHIFT);
-    if(end>=(1LL<<(20+16)))
-        end = (1LL<<(20+16))-1;
+    if(end>=(1LL<<(48-MEMPROT_SHIFT)))
+        end = (1LL<<(48-MEMPROT_SHIFT))-1;
     if(end<idx) // memory addresses higher than 48bits are not tracked
         return;
     pthread_mutex_lock(&mutex_prot);
     for (uintptr_t i=(idx>>16); i<=(end>>16); ++i)
         if(memprot[i]==memprot_default) {
             uint8_t* newblock = calloc(1<<16, sizeof(uint8_t));
-            /*if (arm64_lock_storeifref(&memprot[i], newblock, memprot_default) != newblock) {
+            /*if (native_lock_storeifref(&memprot[i], newblock, memprot_default) != newblock) {
                 free(newblock);
             }*/
             memprot[i] = newblock;
@@ -818,8 +837,8 @@ int isprotectedDB(uintptr_t addr, size_t size)
     dynarec_log(LOG_DEBUG, "isprotectedDB %p -> %p => ", (void*)addr, (void*)(addr+size-1));
     uintptr_t idx = (addr>>MEMPROT_SHIFT);
     uintptr_t end = ((addr+size-1LL)>>MEMPROT_SHIFT);
-    if(end>=(1LL<<(20+16)))
-        end = (1LL<<(20+16))-1LL;
+    if(end>=(1LL<<(48-MEMPROT_SHIFT)))
+        end = (1LL<<(48-MEMPROT_SHIFT))-1LL;
     if(end<idx) { // memory addresses higher than 48bits are not tracked
         dynarec_log(LOG_DEBUG, "00\n");
         return 0;
@@ -837,13 +856,104 @@ int isprotectedDB(uintptr_t addr, size_t size)
 
 #endif
 
+void printMapMem()
+{
+    mapmem_t* m = mapmem;
+    while(m) {
+        printf_log(LOG_INFO, " %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*)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;
+        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;
+            }
+            free(tmp);
+        } else if(begin>m->begin && end<m->end) { // the zone is totaly inside the block => split it!
+            mapmem_t* newm = (mapmem_t*)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, size_t size, uint32_t prot)
 {
     dynarec_log(LOG_DEBUG, "updateProtection %p:%p 0x%x\n", (void*)addr, (void*)(addr+size-1), prot);
+    addMapMem(addr, addr+size-1);
     uintptr_t idx = (addr>>MEMPROT_SHIFT);
     uintptr_t end = ((addr+size-1)>>MEMPROT_SHIFT);
-    if(end>=(1LL<<(20+16)))
-        end = (1LL<<(20+16))-1;
+    if(end>=(1LL<<(48-MEMPROT_SHIFT)))
+        end = (1LL<<(48-MEMPROT_SHIFT))-1;
     if(end<idx) // memory addresses higher than 48bits are not tracked
         return;
     pthread_mutex_lock(&mutex_prot);
@@ -851,7 +961,7 @@ void updateProtection(uintptr_t addr, size_t size, uint32_t prot)
         if(memprot[i]==memprot_default) {
             uint8_t* newblock = calloc(1<<16, sizeof(uint8_t));
 #if 0 //def ARM64   //disabled for now, not usefull with the mutex
-            if (arm64_lock_storeifref(&memprot[i], newblock, memprot_default) != newblock) {
+            if (native_lock_storeifref(&memprot[i], newblock, memprot_default) != newblock) {
                 free(newblock);
             }
 #else
@@ -869,10 +979,11 @@ void updateProtection(uintptr_t addr, size_t size, uint32_t prot)
 
 void setProtection(uintptr_t addr, size_t size, uint32_t prot)
 {
+    addMapMem(addr, addr+size-1);
     uintptr_t idx = (addr>>MEMPROT_SHIFT);
     uintptr_t end = ((addr+size-1)>>MEMPROT_SHIFT);
-    if(end>=(1LL<<(20+16)))
-        end = (1LL<<(20+16))-1;
+    if(end>=(1LL<<(48-MEMPROT_SHIFT)))
+        end = (1LL<<(48-MEMPROT_SHIFT))-1;
     if(end<idx) // memory addresses higher than 48bits are not tracked
         return;
     pthread_mutex_lock(&mutex_prot);
@@ -880,7 +991,7 @@ void setProtection(uintptr_t addr, size_t size, uint32_t prot)
         if(memprot[i]==memprot_default) {
             uint8_t* newblock = calloc(MEMPROT_SIZE, sizeof(uint8_t));
 #if 0 //def ARM64   //disabled for now, not usefull with the mutex
-            if (arm64_lock_storeifref(&memprot[i], newblock, memprot_default) != newblock) {
+            if (native_lock_storeifref(&memprot[i], newblock, memprot_default) != newblock) {
                 free(newblock);
             }
 #else
@@ -895,10 +1006,11 @@ void setProtection(uintptr_t addr, size_t size, uint32_t prot)
 void allocProtection(uintptr_t addr, size_t size, uint32_t prot)
 {
     dynarec_log(LOG_DEBUG, "allocProtection %p:%p 0x%x\n", (void*)addr, (void*)(addr+size-1), prot);
+    addMapMem(addr, addr+size-1);
     uintptr_t idx = (addr>>MEMPROT_SHIFT);
     uintptr_t end = ((addr+size-1LL)>>MEMPROT_SHIFT);
-    if(end>=(1LL<<(20+16)))
-        end = (1LL<<(20+16))-1;
+    if(end>=(1LL<<(48-MEMPROT_SHIFT)))
+        end = (1LL<<(48-MEMPROT_SHIFT))-1;
     if(end<idx) // memory addresses higher than 48bits are not tracked
         return;
     pthread_mutex_lock(&mutex_prot);
@@ -906,7 +1018,7 @@ void allocProtection(uintptr_t addr, size_t size, uint32_t prot)
         if(memprot[i]==memprot_default) {
             uint8_t* newblock = calloc(1<<16, sizeof(uint8_t));
 #if 0 //def ARM64   //disabled for now, not usefull with the mutex
-            if (arm64_lock_storeifref(&memprot[i], newblock, memprot_default) != newblock) {
+            if (native_lock_storeifref(&memprot[i], newblock, memprot_default) != newblock) {
                 free(newblock);
             }
 #else
@@ -928,6 +1040,8 @@ void allocProtection(uintptr_t addr, size_t size, uint32_t prot)
 
 void loadProtectionFromMap()
 {
+    if(box64_mapclean)
+        return;
     char buf[500];
     FILE *f = fopen("/proc/self/maps", "r");
     if(!f)
@@ -939,10 +1053,11 @@ void loadProtectionFromMap()
         uintptr_t s, e;
         if(sscanf(buf, "%lx-%lx %c%c%c", &s, &e, &r, &w, &x)==5) {
             int prot = ((r=='r')?PROT_READ:0)|((w=='w')?PROT_WRITE:0)|((x=='x')?PROT_EXEC:0);
-            allocProtection(s, e-s, prot|PROT_ALLOC);
+            allocProtection(s, e-s, prot);
         }
     }
     fclose(f);
+    box64_mapclean = 1;
 }
 
 static int blockempty(uint8_t* mem)
@@ -957,10 +1072,11 @@ static int blockempty(uint8_t* mem)
 void freeProtection(uintptr_t addr, size_t size)
 {
     dynarec_log(LOG_DEBUG, "freeProtection %p:%p\n", (void*)addr, (void*)(addr+size-1));
+    removeMapMem(addr, addr+size-1);
     uintptr_t idx = (addr>>MEMPROT_SHIFT);
     uintptr_t end = ((addr+size-1LL)>>MEMPROT_SHIFT);
-    if(end>=(1LL<<(20+16)))
-        end = (1LL<<(20+16))-1;
+    if(end>=(1LL<<(48-MEMPROT_SHIFT)))
+        end = (1LL<<(48-MEMPROT_SHIFT))-1;
     if(end<idx) // memory addresses higher than 48bits are not tracked
         return;
     pthread_mutex_lock(&mutex_prot);
@@ -973,9 +1089,9 @@ void freeProtection(uintptr_t addr, size_t size)
             memset(block+start, 0, finish-start+1);
 #if 0 //def ARM64   //disabled for now, not usefull with the mutex
             if (blockempty(block)) {
-                block = (void*)arm64_lock_xchg(&memprot[key], (uintptr_t)memprot_default);
+                block = (void*)native_lock_xchg(&memprot[key], (uintptr_t)memprot_default);
                 if(!blockempty(block)) {
-                    block = (void*)arm64_lock_xchg(&memprot[key], (uintptr_t)block);
+                    block = (void*)native_lock_xchg(&memprot[key], (uintptr_t)block);
                     for (int i = 0; i < 0x10000; ++i) {
                         memprot[key][i] |= block[i];
                     }
@@ -1003,107 +1119,54 @@ uint32_t getProtection(uintptr_t addr)
     return ret;
 }
 
-#define LOWEST (void*)0x20000
-static uintptr_t nextFree(uintptr_t addr)
-{
-    if(addr>=(1LL<<48))
-        return 0;
-    do {
-        uintptr_t idx = (addr>>MEMPROT_SHIFT);
-        if(memprot[idx>>16]==memprot_default) {
-            return addr;
-        }
-        for (uintptr_t i=(idx&0xffff); i<MEMPROT_SIZE; ++i)
-            if(!memprot[idx>>16][i]) {
-                return ((idx>>16)<<(16+12))+(i<<MEMPROT_SHIFT);
-            }
-        addr += (1LL<<(16+12));
-        addr &= ~((1LL<<(16+12)-1LL));
-    } while(1);
-}
-static uintptr_t maxFree(uintptr_t addr, uintptr_t sz)
-{
-    if(addr>=(1LL<<48))
-        return 0;
-    uintptr_t mfree = 0;
-    do {
-        uintptr_t idx = (addr>>MEMPROT_SHIFT);
-        if(memprot[idx>>16]==memprot_default) {
-            mfree+=(1LL<<(16+12));
-            if(mfree>sz) {
-                return addr;
-            }
-        } else {
-            for (uintptr_t i=(idx&0xffffLL); i<MEMPROT_SIZE; ++i)
-                if(!memprot[idx>>16][i]) {
-                    mfree+=(1<<MEMPROT_SHIFT);
-                } else
-                    return mfree;
-        }
-        addr += (1LL<<(12+16));
-        addr &= ~((1LL<<(12+16))-1LL);
-    } while(1);
-}
+#define LOWEST (void*)0x10000
 void* find32bitBlock(size_t size)
 {
     return findBlockNearHint(LOWEST, size);
 }
 void* find47bitBlock(size_t size)
 {
-    // slow iterative search... Would need something better one day
-    uintptr_t addr = 0x100000000LL;
-    do {
-        addr = nextFree(addr);
-        uintptr_t sz = maxFree(addr, size);
-        if(sz>=size) {
-            return (void*)addr;
-        }
-        addr += sz;
-    } while(addr<0x800000000000LL);
-    // search in 32bits as a backup
-    addr = (uintptr_t)LOWEST;
-    do {
-        addr = nextFree(addr);
-        uintptr_t sz = maxFree(addr, size);
-        if(sz>=size) {
-            return (void*)addr;
-        }
-        addr += sz;
-    } while(addr<0x100000000LL);
-    printf_log(LOG_NONE, "Warning: cannot find a 0x%zx block in 47bits address space\n", size);
-    return NULL;
+    void* ret = find47bitBlockNearHint((void*)0x100000000LL, size);
+    if(!ret)
+        ret = find32bitBlock(size);
+    return ret;
 }
 void* find47bitBlockNearHint(void* hint, size_t size)
 {
-    // slow iterative search... Would need something better one day
-    uintptr_t addr = (uintptr_t)hint;
-    do {
-        addr = nextFree(addr);
-        uintptr_t sz = maxFree(addr, size);
-        if(sz>=size) {
+    mapmem_t* m = mapmem;
+    uintptr_t h = (uintptr_t)hint;
+    if(hint<LOWEST) hint = LOWEST;
+    while(m && m->end<0x800000000000LL) {
+        // granularity 0x10000
+        uintptr_t addr = (m->end+1+0xffff)&~0xffff;
+        uintptr_t end = (m->next)?(m->next->begin-1):0xffffffffffffffffLL;
+        // 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;
-        }
-        addr += sz;
-    } while(addr<0x800000000000LL);
-    printf_log(LOG_NONE, "Warning: cannot find a 0x%zx block in 32bits address space\n", size);
+        m = m->next;
+    }
     return NULL;
 }
 void* findBlockNearHint(void* hint, size_t size)
 {
-    // slow iterative search... Would need something better one day
-    uintptr_t addr = (uintptr_t)hint;
-    do {
-        addr = nextFree(addr);
-        uintptr_t sz = maxFree(addr, size);
-        if(sz>=size) {
+    mapmem_t* m = mapmem;
+    uintptr_t h = (uintptr_t)hint;
+    if(hint<LOWEST) hint = LOWEST;
+    while(m && m->end<0x100000000LL) {
+        // granularity 0x10000
+        uintptr_t addr = (m->end+1+0xffff)&~0xffff;
+        uintptr_t end = (m->next)?(m->next->begin-1):0xffffffffffffffffLL;
+        // 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;
-        }
-        addr += sz;
-    } while(addr<0x100000000LL);
-    printf_log(LOG_NONE, "Warning: cannot find a 0x%zx block in 32bits address space\n", size);
-    return NULL;
+        m = m->next;
+    }
+    return hint;
 }
-#undef LOWEST
 
 int unlockCustommemMutex()
 {
@@ -1159,19 +1222,25 @@ void init_custommem_helper(box64context_t* ctx)
         memprot[i] = memprot_default;
     init_mutexes();
 #ifdef DYNAREC
-#ifdef ARM64
     if(box64_dynarec)
         for(int i=0; i<(1<<JMPTABL_SHIFT); ++i) {
-            box64_jmptbldefault0[i] = (uintptr_t)arm64_next;
+            box64_jmptbldefault0[i] = (uintptr_t)native_next;
             box64_jmptbldefault1[i] = box64_jmptbldefault0;
             box64_jmptbldefault2[i] = box64_jmptbldefault1;
             box64_jmptbl3[i] = box64_jmptbldefault2;
         }
-#else
-#error Unsupported architecture!
-#endif
 #endif
     pthread_atfork(NULL, NULL, atfork_child_custommem);
+    // init mapmem list
+    mapmem = (mapmem_t*)calloc(1, sizeof(mapmem_t));
+    mapmem->begin = 0x0;
+    mapmem->end = (uintptr_t)LOWEST - 1;
+    loadProtectionFromMap();
+    // check if PageSize is correctly defined
+    if(box64_pagesize != (1<<MEMPROT_SHIFT)) {
+        printf_log(LOG_NONE, "Error: PageSize configuation is wrong: configured with %d, but got %d\n", 1<<MEMPROT_SHIFT, box64_pagesize);
+        exit(-1);   // abort or let it continue?
+    }
 }
 
 void fini_custommem_helper(box64context_t *ctx)
@@ -1249,4 +1318,9 @@ void fini_custommem_helper(box64context_t *ctx)
     free(p_blocks);
     pthread_mutex_destroy(&mutex_prot);
     pthread_mutex_destroy(&mutex_blocks);
+    while(mapmem) {
+        mapmem_t *tmp = mapmem;
+        mapmem = mapmem->next;
+        free(tmp);
+    }
 }

src/dynarec/arm64/arm64_emitter.h

@@ -194,6 +194,10 @@
 #define SBCSw_REG(Rd, Rn, Rm)       EMIT(ADDSUBC_gen(0, 1, 1, Rm, Rn, Rd))
 #define SBCSxw_REG(Rd, Rn, Rm)      EMIT(ADDSUBC_gen(rex.w, 1, 1, Rm, Rn, Rd))
 
+// ADR
+#define ADR_gen(immlo, immhi, Rd)   ((immlo)<<29 | 0b10000<<24 | (immhi)<<5 | (Rd))
+#define ADR_S20(Rd, imm)            EMIT(ADR_gen((imm)&3, ((imm)>>2)&0x7ffff, (Rd))
+
 // LDR
 #define LDR_gen(size, op1, imm9, op2, Rn, Rt)    ((size)<<30 | 0b111<<27 | (op1)<<24 | 0b01<<22 | (imm9)<<12 | (op2)<<10 | (Rn)<<5 | (Rt))
 #define LDRx_S9_postindex(Rt, Rn, imm9)   EMIT(LDR_gen(0b11, 0b00, (imm9)&0x1ff, 0b01, Rn, Rt))
@@ -1315,6 +1319,15 @@
 #define VUZP1Q_64(Rt, Rn, Rm)       EMIT(UZP_gen(1, 0b11, Rm, 0, Rn, Rt))
 #define VUZP2Q_64(Rt, Rn, Rm)       EMIT(UZP_gen(1, 0b11, Rm, 1, Rn, Rt))
 
+#define DUP_gen(Q, imm5, Rn, Rd)    ((Q)<<30 | 0b01110000<<21 | (imm5)<<16 | 1<<10 | (Rn)<<5 | (Rd))
+#define VDUP_8(Vd, Vn, idx)         EMIT(DUP_gen(0, ((idx)<<1|1), Vn, Vd))
+#define VDUPQ_8(Vd, Vn, idx)        EMIT(DUP_gen(1, ((idx)<<1|1), Vn, Vd))
+#define VDUP_16(Vd, Vn, idx)        EMIT(DUP_gen(0, ((idx)<<2|0b10), Vn, Vd))
+#define VDUPQ_16(Vd, Vn, idx)       EMIT(DUP_gen(1, ((idx)<<2|0b10), Vn, Vd))
+#define VDUP_32(Vd, Vn, idx)        EMIT(DUP_gen(0, ((idx)<<3|0b100), Vn, Vd))
+#define VDUPQ_32(Vd, Vn, idx)       EMIT(DUP_gen(1, ((idx)<<3|0b100), Vn, Vd))
+#define VDUPQ_64(Vd, Vn, idx)       EMIT(DUP_gen(1, ((idx)<<4|0b1000), Vn, Vd))
+
 // TBL
 #define TBL_gen(Q, Rm, len, op, Rn, Rd) ((Q)<<30 | 0b001110<<24 | (Rm)<<16 | (len)<<13 | (op)<<12 | (Rn)<<5 | (Rd))
 //Use Rm[] to pick from Rn element and store in Rd. Out-of-range element gets 0
@@ -1454,18 +1467,35 @@
 // Vector Float CMP
 // EQual
 #define FCMP_vector(Q, U, E, sz, Rm, ac, Rn, Rd)    ((Q)<<30 | (U)<<29 | 0b01110<<24 | (E)<<23 | (sz)<<22 | 1<<21 | (Rm)<<16 | 0b1110<<12 | (ac)<<11 | 1<<10 | (Rn)<<5 | (Rd))
-#define FCMEQQD(Rd, Rn, Rm)          EMIT(FCMP_vector(1, 0, 0, 1, Rm, 0, Rn, Rd))
-#define FCMEQQS(Rd, Rn, Rm)          EMIT(FCMP_vector(1, 0, 0, 0, Rm, 0, Rn, Rd))
+#define VFCMEQQD(Rd, Rn, Rm)         EMIT(FCMP_vector(1, 0, 0, 1, Rm, 0, Rn, Rd))
+#define VFCMEQQS(Rd, Rn, Rm)         EMIT(FCMP_vector(1, 0, 0, 0, Rm, 0, Rn, Rd))
 // Greater or Equal
-#define FCMGEQD(Rd, Rn, Rm)          EMIT(FCMP_vector(1, 1, 0, 1, Rm, 0, Rn, Rd))
-#define FCMGEQS(Rd, Rn, Rm)          EMIT(FCMP_vector(1, 1, 0, 0, Rm, 0, Rn, Rd))
-#define FCMGEQD_ABS(Rd, Rn, Rm)      EMIT(FCMP_vector(1, 1, 0, 1, Rm, 1, Rn, Rd))
-#define FCMGEQS_ABS(Rd, Rn, Rm)      EMIT(FCMP_vector(1, 1, 0, 0, Rm, 1, Rn, Rd))
+#define VFCMGEQD(Rd, Rn, Rm)         EMIT(FCMP_vector(1, 1, 0, 1, Rm, 0, Rn, Rd))
+#define VFCMGEQS(Rd, Rn, Rm)         EMIT(FCMP_vector(1, 1, 0, 0, Rm, 0, Rn, Rd))
+#define VFCMGEQD_ABS(Rd, Rn, Rm)     EMIT(FCMP_vector(1, 1, 0, 1, Rm, 1, Rn, Rd))
+#define VFCMGEQS_ABS(Rd, Rn, Rm)     EMIT(FCMP_vector(1, 1, 0, 0, Rm, 1, Rn, Rd))
 // Greater Than
-#define FCMGTQD(Rd, Rn, Rm)          EMIT(FCMP_vector(1, 1, 1, 1, Rm, 0, Rn, Rd))
-#define FCMGTQS(Rd, Rn, Rm)          EMIT(FCMP_vector(1, 1, 1, 0, Rm, 0, Rn, Rd))
-#define FCMGTQD_ABS(Rd, Rn, Rm)      EMIT(FCMP_vector(1, 1, 1, 1, Rm, 1, Rn, Rd))
-#define FCMGTQS_ABS(Rd, Rn, Rm)      EMIT(FCMP_vector(1, 1, 1, 0, Rm, 1, Rn, Rd))
+#define VFCMGTQD(Rd, Rn, Rm)         EMIT(FCMP_vector(1, 1, 1, 1, Rm, 0, Rn, Rd))
+#define VFCMGTQS(Rd, Rn, Rm)         EMIT(FCMP_vector(1, 1, 1, 0, Rm, 0, Rn, Rd))
+#define VFCMGTQD_ABS(Rd, Rn, Rm)     EMIT(FCMP_vector(1, 1, 1, 1, Rm, 1, Rn, Rd))
+#define VFCMGTQS_ABS(Rd, Rn, Rm)     EMIT(FCMP_vector(1, 1, 1, 0, Rm, 1, Rn, Rd))
+
+// Scalar Float CMP to 0
+#define FCMP_0_scalar(U, sz, op, Rn, Rd) (0b01<<30 | (U)<<29| 0b11110<<24 | 1<<23 | (sz)<<22 | 0b10000<<17 | 0b011<<14 | (op)<<12 | 0b10<<10 | (Rn)<<5 | (Rd))
+// Less or equal to 0
+#define FCMLES_0(Rd, Rn)             EMIT(FCMP_0_scalar(1, 0, 0b01, (Rn), (Rd)))
+#define FCMLED_0(Rd, Rn)             EMIT(FCMP_0_scalar(1, 1, 0b01, (Rn), (Rd)))
+// Greater than 0
+#define FCMGTS_0(Rd, Rn)             EMIT(FCMP_0_scalar(0, 0, 0b00, (Rn), (Rd)))
+#define FCMGTD_0(Rd, Rn)             EMIT(FCMP_0_scalar(0, 1, 0b00, (Rn), (Rd)))
+// Less than 0
+#define FCMLTS_0(Rd, Rn)             EMIT(FCMP_0_scalar(0, 0, 0b10, (Rn), (Rd)))
+#define FCMLTD_0(Rd, Rn)             EMIT(FCMP_0_scalar(0, 1, 0b10, (Rn), (Rd)))
+
+// Scalar Float CMP
+#define FCMP_op_scalar(U, E, sz, Rm, ac, Rn, Rd)    (0b01<<30 | (U)<<29 | 0b11110<<24 | (E)<<23 | (sz)<<22 | 1<<21 | (Rm)<<16 | 0b1110<<12 | (ac)<<11 | 1<<10 | (Rn)<<5 | (Rd))
+#define FCMEQS(Rd, Rn, Rm)          EMIT(FCMP_op_scalar(0, 0, 0, (Rm), 0, (Rn), (Rd)))
+#define FCMEQD(Rd, Rn, Rm)          EMIT(FCMP_op_scalar(0, 0, 1, (Rm), 0, (Rn), (Rd)))
 
 // UMULL / SMULL
 #define MULL_vector(Q, U, size, Rm, Rn, Rd) ((Q)<<30 | (U)<<29 | 0b01110<<24 | (size)<<22 | 1<<21 | (Rm)<<16 | 0b1100<<12 |(Rn)<<5 |(Rd))

src/dynarec/arm64/arm64_printer.c

@@ -236,6 +236,15 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
             snprintf(buff, sizeof(buff), "STR %s, [%s, %s, %s %d]", (size==2)?Wt[Rt]:Xt[Rt], XtSp[Rn], ((option&1)==0)?Wt[Rm]:Xt[Rm], extend[option], amount);
         return buff;
     }
+    if(isMask(opcode, "00111000001mmmmmoooS10nnnnnttttt", &a)) {
+        const char* extend[] = {"?0", "?1", "UXTW", "LSL", "?4", "?5", "SXTW", "SXTX"};
+        int amount = a.S;
+        if(option==3 && !amount)
+            snprintf(buff, sizeof(buff), "STRB %s, [%s, %s]", Wt[Rt], XtSp[Rn], ((option&1)==0)?Wt[Rm]:Xt[Rm]);
+        else
+            snprintf(buff, sizeof(buff), "STRB %s, [%s, %s, %s %d]", Wt[Rt], XtSp[Rn], ((option&1)==0)?Wt[Rm]:Xt[Rm], extend[option], amount);
+        return buff;
+    }
     if(isMask(opcode, "0x111000010iiiiiiiii01nnnnnttttt", &a)) {
         int size = a.x;
         int offset = signExtend(imm, 9);
@@ -930,6 +939,12 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
         return buff;
     }
 
+    // ADR
+    if(isMask(opcode, "0ss10000iiiiiiiiiiiiiiiiiiiddddd", &a)) {
+      snprintf(buff, sizeof(buff), "ADR, %s, %ld", Xt[Rd], signExtend((imm)<<2|(imms), 20));
+      return buff;
+    }
+
     // LDR / STR
     if(isMask(opcode, "ss111101cciiiiiiiiiiiinnnnnttttt", &a)) {
         char s = '?';
@@ -1036,7 +1051,7 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
     //FMIN/FMAX
     if(isMask(opcode, "00011110ff1mmmmm01oo10nnnnnddddd", &a)) {
         char s = (sf==0)?'S':((sf==1)?'D':'?');
-        snprintf(buff, sizeof(buff), "F%s %c%d, %c%d, %c%d", (option==3)?"MINNM":((option==2)?"MINNMP":((!option)?"MAXNM":"MAXNMP")), s, Rd, s, Rn, s, Rm);
+        snprintf(buff, sizeof(buff), "F%s %c%d, %c%d, %c%d", (option==3)?"MINNM":((option==2)?"MAXNM":((!option)?"MAX":"MIN")), s, Rd, s, Rn, s, Rm);
         return buff;
     }
     if(isMask(opcode, "0Q001110of1mmmmm110001nnnnnddddd", &a)) {
@@ -1311,6 +1326,21 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
             snprintf(buff, sizeof(buff), "%cSHLL%s V%d.%s, V%d.%s, #%d", a.U?'U':'S', a.Q?"2":"", Rd, Va, Rn, Vd, sh);
         return buff;
     }
+
+    // DUP
+    if(isMask(opcode, "0Q001110000iiiii000001nnnnnddddd", &a)) {
+        const char* Y[] = {"8B", "16B", "4H", "8H", "2S", "4S", "2D", "??"};
+        const char* Z[] = {"B", "H", "S", "D"};
+        int sz = 3;
+        if((imm&0b0001)==0b0001) sz=0;
+        else if((imm&0b0011)==0b0010) sz=1;
+        else if((imm&0b0111)==0b0100) sz=2;
+        int sh=imm - (1<<sz);
+        const char* Vd = Y[(sz<<1)|a.Q];
+        const char* Vn = Z[sz];
+        snprintf(buff, sizeof(buff), "DUP V%d.%s, V%d.%s[%d]", Rd, Vd, Rn, Vn, sh);
+        return buff;
+    }
     
     // AES
     if(isMask(opcode, "0100111000101000010f10nnnnnddddd", &a)) {
@@ -1321,8 +1351,12 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
         snprintf(buff, sizeof(buff), "AES%sMC V%d.16B, V%d.16B", sf?"I":"", Rd, Rn);
         return buff;
     }
+    // DMB ISH
+    if(isMask(opcode, "11010101000000110011nnnn10111111", &a)) {
+        snprintf(buff, sizeof(buff), "DMB %s", (Rn==0b1011)?"ISH":"???");
+    }
 
 
     snprintf(buff, sizeof(buff), "%08X ???", __builtin_bswap32(opcode));
     return buff;
-}
+}

src/dynarec/arm64/dynarec_arm64_00.c

@@ -17,10 +17,10 @@
 #include "bridge.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
+#include "dynarec_native.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_functions.h"
 #include "dynarec_arm64_helper.h"
 
@@ -372,6 +372,9 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             i64 = F32S;
             emit_sub32c(dyn, ninst, rex, xRAX, i64, x3, x4, x5);
             break;
+        case 0x2E:
+            INST_NAME("CS:");
+            break;
 
         case 0x30:
             INST_NAME("XOR Eb, Gb");
@@ -633,21 +636,30 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             }
             break;
 
-        #define GO(GETFLAGS, NO, YES, F)    \
-            READFLAGS(F);                   \
-            i8 = F8S;   \
-            BARRIER(2); \
-            JUMP(addr+i8);\
-            GETFLAGS;   \
-            if(dyn->insts[ninst].x64.jmp_insts==-1) {   \
-                /* out of the block */                  \
-                i32 = dyn->insts[ninst+1].address-(dyn->arm_size); \
-                Bcond(NO, i32);     \
-                jump_to_next(dyn, addr+i8, 0, ninst); \
-            } else {    \
-                /* inside the block */  \
-                i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->arm_size);    \
-                Bcond(YES, i32);    \
+        #define GO(GETFLAGS, NO, YES, F)                                \
+            READFLAGS(F);                                               \
+            i8 = F8S;                                                   \
+            BARRIER(BARRIER_MAYBE);                                     \
+            JUMP(addr+i8, 1);                                           \
+            GETFLAGS;                                                   \
+            if(dyn->insts[ninst].x64.jmp_insts==-1 ||                   \
+                CHECK_CACHE()) {                                        \
+                /* out of the block */                                  \
+                i32 = dyn->insts[ninst].epilog-(dyn->native_size);      \
+                Bcond(NO, i32);                                         \
+                if(dyn->insts[ninst].x64.jmp_insts==-1) {               \
+                    if(!dyn->insts[ninst].x64.barrier)                  \
+                        fpu_purgecache(dyn, ninst, 1, x1, x2, x3);      \
+                    jump_to_next(dyn, addr+i8, 0, ninst);               \
+                } else {                                                \
+                    fpuCacheTransform(dyn, ninst, x1, x2, x3);          \
+                    i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->native_size);\
+                    B(i32);                                             \
+                }                                                       \
+            } else {                                                    \
+                /* inside the block */                                  \
+                i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->native_size);    \
+                Bcond(YES, i32);                                        \
             }
 
         GOCOND(0x70, "J", "ib");
@@ -926,6 +938,10 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             } else {                    // mem <= reg
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0xfff<<(2+rex.w), (1<<(2+rex.w))-1, rex, 0, 0);
                 STRxw_U12(gd, ed, fixedaddress);
+                if(box64_dynarec_strongmem && 
+                 (dyn->insts[ninst].x64.barrier || box64_dynarec_strongmem>1 || (dyn->insts[ninst+1].x64.barrier || dyn->insts[ninst+1].x64.jmp))) {
+                    DMB_ISH();
+                }
             }
             break;
         case 0x8A:
@@ -1007,7 +1023,19 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 }
             }
             break;
-
+        case 0x8E:
+            INST_NAME("MOV Seg,Ew");
+            nextop = F8;
+            if((nextop&0xC0)==0xC0) {
+                ed = xRAX+(nextop&7)+(rex.b<<3);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0xfff<<2, 0, rex, 0, 0);
+                LDRH_U12(x1, ed, fixedaddress);
+                ed = x1;
+            }
+            STRw_U12(ed, xEmu, offsetof(x64emu_t, segs[(nextop&0x38)>>3]));
+            STRw_U12(wZR, xEmu, offsetof(x64emu_t, segs_serial[(nextop&0x38)>>3]));
+            break;
         case 0x8F:
             INST_NAME("POP Ed");
             nextop = F8;
@@ -1461,7 +1489,7 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("ROL Ed, Ib");
-                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
                     GETED(1);
                     u8 = (F8)&(rex.w?0x3f:0x1f);
                     emit_rol32c(dyn, ninst, rex, ed, u8, x3, x4);
@@ -1469,7 +1497,7 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     break;
                 case 1:
                     INST_NAME("ROR Ed, Ib");
-                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
                     GETED(1);
                     u8 = (F8)&(rex.w?0x3f:0x1f);
                     emit_ror32c(dyn, ninst, rex, ed, u8, x3, x4);
@@ -1532,7 +1560,7 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             INST_NAME("RETN");
             //SETFLAGS(X_ALL, SF_SET);    // Hack, set all flags (to an unknown state...)
             READFLAGS(X_PEND);  // lets play safe here too
-            BARRIER(2);
+            BARRIER(BARRIER_FLOAT);
             i32 = F16;
             retn_to_epilog(dyn, ninst, i32);
             *need_epilog = 0;
@@ -1542,7 +1570,7 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             INST_NAME("RET");
             // SETFLAGS(X_ALL, SF_SET);    // Hack, set all flags (to an unknown state...)
             READFLAGS(X_PEND);  // so instead, force the defered flags, so it's not too slow, and flags are not lost
-            BARRIER(2);
+            BARRIER(BARRIER_FLOAT);
             ret_to_epilog(dyn, ninst);
             *need_epilog = 0;
             *ok = 0;
@@ -1595,7 +1623,7 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             SETFLAGS(X_ALL, SF_SET);    // Hack, set all flags (to an unknown state...)
             if(PK(0)=='S' && PK(1)=='C') {
                 addr+=2;
-                BARRIER(2);
+                BARRIER(BARRIER_FLOAT);
                 INST_NAME("Special Box64 instruction");
                 if((PK64(0)==0))
                 {
@@ -1653,7 +1681,7 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0xCF:
             INST_NAME("IRET");
             SETFLAGS(X_ALL, SF_SET);    // Not a hack, EFLAGS are restored
-            BARRIER(2);
+            BARRIER(BARRIER_FLOAT);
             iret_to_epilog(dyn, ninst, rex.w);
             *need_epilog = 0;
             *ok = 0;
@@ -1766,14 +1794,14 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("ROL Ed, 1");
-                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
                     GETED(0);
                     emit_rol32c(dyn, ninst, rex, ed, 1, x3, x4);
                     WBACK;
                     break;
                 case 1:
                     INST_NAME("ROR Ed, 1");
-                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
                     GETED(0);
                     emit_ror32c(dyn, ninst, rex, ed, 1, x3, x4);
                     WBACK;
@@ -1978,17 +2006,26 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             addr = dynarec64_DF(dyn, addr, ip, ninst, rex, rep, ok, need_epilog);
             break;
         #define GO(Z)                                                   \
-            BARRIER(2);                                                 \
-            JUMP(addr+i8);                                              \
-            if(dyn->insts[ninst].x64.jmp_insts==-1) {               \
-                /* out of the block */                              \
-                i32 = dyn->insts[ninst+1].address-(dyn->arm_size);  \
-                if(Z) {CBNZx(xRCX, i32);} else {CBZx(xRCX, i32);};  \
-                jump_to_next(dyn, addr+i8, 0, ninst);               \
-            } else {                                                \
-                /* inside the block */                              \
-                i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->arm_size);    \
-                if(Z) {CBZx(xRCX, i32);} else {CBNZx(xRCX, i32);};  \
+            BARRIER(BARRIER_MAYBE);                                     \
+            JUMP(addr+i8, 1);                                           \
+            if(dyn->insts[ninst].x64.jmp_insts==-1 ||                   \
+                CHECK_CACHE()) {                                        \
+                /* out of the block */                                  \
+                i32 = dyn->insts[ninst].epilog-(dyn->native_size);      \
+                if(Z) {CBNZx(xRCX, i32);} else {CBZx(xRCX, i32);};      \
+                if(dyn->insts[ninst].x64.jmp_insts==-1) {               \
+                    if(!dyn->insts[ninst].x64.barrier)                  \
+                        fpu_purgecache(dyn, ninst, 1, x1, x2, x3);      \
+                    jump_to_next(dyn, addr+i8, 0, ninst);               \
+                } else {                                                \
+                    fpuCacheTransform(dyn, ninst, x1, x2, x3);          \
+                    i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->native_size);    \
+                    Bcond(c__, i32);                                    \
+                }                                                       \
+            } else {                                                    \
+                /* inside the block */                                  \
+                i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->native_size);    \
+                if(Z) {CBZx(xRCX, i32);} else {CBNZx(xRCX, i32);};      \
             }
         case 0xE0:
             INST_NAME("LOOPNZ");
@@ -2038,8 +2075,8 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             switch(tmp) {
                 case 3:
                     SETFLAGS(X_ALL, SF_SET);    // Hack to set flags to "dont'care" state
-                    BARRIER(1);
-                    BARRIER_NEXT(1);
+                    BARRIER(BARRIER_FULL);
+                    //BARRIER_NEXT(BARRIER_FULL);
                     TABLE64(x2, addr);
                     PUSH1(x2);
                     MESSAGE(LOG_DUMP, "Native Call to %s (retn=%d)\n", GetNativeName(GetNativeFnc(dyn->insts[ninst].natcall-1)), dyn->insts[ninst].retn);
@@ -2073,18 +2110,16 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     }
                     break;
                 default:
-                    if(ninst && dyn->insts && dyn->insts[ninst-1].x64.set_flags) {
+                    if(ninst && dyn->insts[ninst-1].x64.set_flags) {
                         READFLAGS(X_PEND);  // that's suspicious
                     } else {
                         SETFLAGS(X_ALL, SF_SET);    // Hack to set flags to "dont'care" state
                     }
                     // regular call
-                    BARRIER(1);
-                    BARRIER_NEXT(1);
-                    if(!dyn->insts || ninst==dyn->size-1) {
-                        *need_epilog = 0;
-                        *ok = 0;
-                    }
+                    BARRIER(BARRIER_FULL);
+                    //BARRIER_NEXT(1);
+                    *need_epilog = 0;
+                    *ok = 0;
                     TABLE64(x2, addr);
                     PUSH1(x2);
                     if(addr+i32==0) {   // self modifying code maybe? so use indirect address fetching
@@ -2098,7 +2133,7 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             break;
         case 0xE9:
         case 0xEB:
-            BARRIER(1);
+            BARRIER(BARRIER_MAYBE);
             if(opcode==0xE9) {
                 INST_NAME("JMP Id");
                 i32 = F32S;
@@ -2106,13 +2141,15 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 INST_NAME("JMP Ib");
                 i32 = F8S;
             }
-            JUMP(addr+i32);
-            PASS2IF(dyn->insts[ninst].x64.jmp_insts==-1, 1) {
+            JUMP(addr+i32, 0);
+            if(dyn->insts[ninst].x64.jmp_insts==-1) {
                 // out of the block
+                fpu_purgecache(dyn, ninst, 1, x1, x2, x3);
                 jump_to_next(dyn, addr+i32, 0, ninst);
             } else {
                 // inside the block
-                tmp = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->arm_size);
+                fpuCacheTransform(dyn, ninst, x1, x2, x3);
+                tmp = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->native_size);
                 if(tmp==4) {
                     NOP;
                 } else {
@@ -2272,7 +2309,7 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         MOVw_REG(xRAX, x2);
                         MOVw_REG(xRDX, x4);
                     } else {
-                        if(ninst && dyn->insts 
+                        if(ninst 
                            && dyn->insts[ninst-1].x64.addr 
                            && *(uint8_t*)(dyn->insts[ninst-1].x64.addr)==0x31 
                            && *(uint8_t*)(dyn->insts[ninst-1].x64.addr+1)==0xD2) {
@@ -2319,9 +2356,15 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                             MOVx_REG(xRAX, x2);
                         } else {
                             GETEDH(x1, 0);  // get edd changed addr, so cannot be called 2 times for same op...
-                            CBZxw_MARK(xRDX);
+                            //Need to see if RDX==0 and RAX not signed
+                            // or RDX==-1 and RAX signed
+                            CBNZx_MARK2(xRDX);
+                            TBZ_MARK(xRAX, 31);
+                            MARK2;
                             MVNx_REG(x2, xRDX);
-                            CBZxw_MARK(x2);
+                            CBNZx_MARK3(x2);
+                            TBNZ_MARK(xRAX, 31);
+                            MARK3;
                             if(ed!=x1) {MOVx_REG(x1, ed);}
                             CALL((void*)idiv64, -1);
                             B_NEXT_nocond;
@@ -2350,14 +2393,14 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("INC Eb");
-                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
+                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
                     GETEB(x1, 0);
                     emit_inc8(dyn, ninst, x1, x2, x4);
                     EBBACK;
                     break;
                 case 1:
                     INST_NAME("DEC Eb");
-                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
+                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
                     GETEB(x1, 0);
                     emit_dec8(dyn, ninst, x1, x2, x4);
                     EBBACK;
@@ -2371,14 +2414,14 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             switch((nextop>>3)&7) {
                 case 0: // INC Ed
                     INST_NAME("INC Ed");
-                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
+                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
                     GETED(0);
                     emit_inc32(dyn, ninst, rex, ed, x3, x4);
                     WBACK;
                     break;
                 case 1: //DEC Ed
                     INST_NAME("DEC Ed");
-                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
+                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
                     GETED(0);
                     emit_dec32(dyn, ninst, rex, ed, x3, x4);
                     WBACK;
@@ -2393,19 +2436,19 @@ uintptr_t dynarec64_00(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         SETFLAGS(X_ALL, SF_SET);    //Hack to put flag in "don't care" state
                     }
                     GETEDx(0);
-                    BARRIER(1);
-                    BARRIER_NEXT(1);
+                    BARRIER(BARRIER_FLOAT);
+                    //BARRIER_NEXT(BARRIER_FULL);
                     if(!dyn->insts || ninst==dyn->size-1) {
                         *need_epilog = 0;
                         *ok = 0;
                     }
-                    GETIP(addr);
+                    GETIP_(addr);
                     PUSH1(xRIP);
                     jump_to_next(dyn, 0, ed, ninst);
                     break;
                 case 4: // JMP Ed
                     INST_NAME("JMP Ed");
-                    BARRIER(1);
+                    BARRIER(BARRIER_FULL);
                     GETEDx(0);
                     jump_to_next(dyn, 0, ed, ninst);
                     *need_epilog = 0;

src/dynarec/arm64/dynarec_arm64_0f.c

@@ -15,46 +15,46 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
+#include "dynarec_native.h"
 #include "my_cpuid.h"
 #include "emu/x87emu_private.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_functions.h"
 #include "dynarec_arm64_helper.h"
 
 #define GETG                            \
     gd = ((nextop&0x38)>>3)+(rex.r<<3)  \
 
-#define GETGX(a)                        \
+#define GETGX(a, w)                     \
     gd = ((nextop&0x38)>>3)+(rex.r<<3); \
-    a = sse_get_reg(dyn, ninst, x1, gd)
+    a = sse_get_reg(dyn, ninst, x1, gd, w)
 
 #define GETGX_empty(a)                          \
     gd = ((nextop&0x38)>>3)+(rex.r<<3);         \
     a = sse_get_reg_empty(dyn, ninst, x1, gd)
 
-#define GETEX(a, D)                                             \
-    if(MODREG) {                                                \
-        a = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3));   \
-    } else {                                                    \
+#define GETEX(a, w, D)                                              \
+    if(MODREG) {                                                    \
+        a = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), w);  \
+    } else {                                                        \
         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<4, 15, rex, 0, D); \
-        a = fpu_get_scratch(dyn);                               \
-        VLDR128_U12(a, ed, fixedaddress);                       \
+        a = fpu_get_scratch(dyn);                                   \
+        VLDR128_U12(a, ed, fixedaddress);                           \
     }
 
 #define GETGM(a)                        \
     gd = ((nextop&0x38)>>3);            \
-    a = mmx_get_reg(dyn, ninst, x1, gd)
+    a = mmx_get_reg(dyn, ninst, x1, x2, x3, gd)
 
-#define GETEM(a, D)                                 \
-    if(MODREG) {                                    \
-        a = mmx_get_reg(dyn, ninst, x1, (nextop&7));\
-    } else {                                        \
+#define GETEM(a, D)                                             \
+    if(MODREG) {                                                \
+        a = mmx_get_reg(dyn, ninst, x1, x2, x3, (nextop&7));    \
+    } else {                                                    \
         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<3, 7, rex, 0, D); \
-        a = fpu_get_scratch(dyn);                   \
-        VLDR64_U12(a, ed, fixedaddress);            \
+        a = fpu_get_scratch(dyn);                               \
+        VLDR64_U12(a, ed, fixedaddress);                        \
     }
 
 #define PUTEM(a)                            \
@@ -151,7 +151,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             GETG;
             if(MODREG) {
                 ed = (nextop&7)+(rex.b<<3);
-                v1 = sse_get_reg(dyn, ninst, x1, ed);
+                v1 = sse_get_reg(dyn, ninst, x1, ed, 0);
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 VMOVQ(v0, v1);
             } else {
@@ -164,7 +164,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             INST_NAME("MOVUPS Ex,Gx");
             nextop = F8;
             GETG;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
             if(MODREG) {
                 ed = (nextop&7)+(rex.b<<3);
                 v1 = sse_get_reg_empty(dyn, ninst, x1, ed);
@@ -178,12 +178,12 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             nextop = F8;
             if(MODREG) {
                 INST_NAME("MOVHLPS Gx,Ex");
-                GETGX(v0);
-                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3));
+                GETGX(v0, 1);
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);
                 VMOVeD(v0, 0, v1, 1);
             } else {
                 INST_NAME("MOVLPS Gx,Ex");
-                GETGX(v0);
+                GETGX(v0, 1);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, rex, 0, 0);
                 VLD1_64(v0, 0, ed);
             }
@@ -191,9 +191,9 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0x13:
             nextop = F8;
             INST_NAME("MOVLPS Ex,Gx");
-            GETGX(v0);
+            GETGX(v0, 0);
             if(MODREG) {
-                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3));
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 1);
                 VMOVeD(v1, 0, v0, 0);
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, rex, 0, 0);
@@ -203,27 +203,27 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0x14:
             INST_NAME("UNPCKLPS Gx, Ex");
             nextop = F8;
-            GETEX(q0, 0);
-            GETGX(v0);
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
             VZIP1Q_32(v0, v0, q0);
             break;
         case 0x15:
             INST_NAME("UNPCKHPS Gx, Ex");
             nextop = F8;
-            GETEX(q0, 0);
-            GETGX(v0);
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
             VZIP2Q_32(v0, v0, q0);
             break;
         case 0x16:
             nextop = F8;
             if(MODREG) {
                 INST_NAME("MOVLHPS Gx,Ex");
-                GETGX(v0);
-                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3));
+                GETGX(v0, 1);
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);
                 VMOVeD(v0, 1, v1, 0);
             } else {
                 INST_NAME("MOVHPS Gx,Ex");
-                GETGX(v0);
+                GETGX(v0, 1);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, rex, 0, 0);
                 VLD1_64(v0, 1, ed);
             }
@@ -231,9 +231,9 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0x17:
             nextop = F8;
             INST_NAME("MOVHPS Ex,Gx");
-            GETGX(v0);
+            GETGX(v0, 0);
             if(MODREG) {
-                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3));
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 1);
                 VMOVeD(v1, 0, v0, 1);
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, rex, 0, 0);
@@ -284,7 +284,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             GETG;
             if(MODREG) {
                 ed = (nextop&7)+(rex.b<<3);
-                v1 = sse_get_reg(dyn, ninst, x1, ed);
+                v1 = sse_get_reg(dyn, ninst, x1, ed, 0);
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 VMOVQ(v0, v1);
             } else {
@@ -297,7 +297,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             INST_NAME("MOVAPS Ex,Gx");
             nextop = F8;
             GETG;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
             if(MODREG) {
                 ed = (nextop&7)+(rex.b<<3);
                 v1 = sse_get_reg_empty(dyn, ninst, x1, ed);
@@ -312,7 +312,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             INST_NAME("MOVNTPS Ex,Gx");
             nextop = F8;
             GETG;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
             if(MODREG) {
                 ed = (nextop&7)+(rex.b<<3);
                 v1 = sse_get_reg_empty(dyn, ninst, x1, ed);
@@ -329,9 +329,9 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             if(opcode==0x2F) {INST_NAME("COMISS Gx, Ex");} else {INST_NAME("UCOMISS Gx, Ex");}
             SETFLAGS(X_ALL, SF_SET);
             nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 0);
             if(MODREG) {
-                s0 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3));
+                s0 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3), 0);
             } else {
                 s0 = fpu_get_scratch(dyn);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<2, 3, rex, 0, 0);
@@ -391,6 +391,30 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             }
             break;
 
+        case 0x3A:  // these are some more SSSE3 opcodes
+            opcode = F8;
+            switch(opcode) {
+                case 0x0F:
+                    INST_NAME("PALIGNR Gm, Em, Ib");
+                    nextop = F8;
+                    GETGM(q0);
+                    GETEM(q1, 1);
+                    u8 = F8;
+                    if(u8>15) {
+                        VEOR(q0, q0, q0);    
+                    } else if(u8>7) {
+                        d0 = fpu_get_scratch(dyn);
+                        VEOR(d0, d0, d0);
+                        VEXT_8(q0, q0, d0, u8-8);
+                    } else {
+                        VEXT_8(q0, q1, q0, u8);
+                    }
+                    break;
+                default:
+                    DEFAULT;
+            }
+            break;
+
         #define GO(GETFLAGS, NO, YES, F)            \
             READFLAGS(F);                           \
             GETFLAGS;                               \
@@ -415,7 +439,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             MOV32w(gd, 0);
             if((nextop&0xC0)==0xC0) {
                 // EX is an xmm reg
-                GETEX(q0, 0);
+                GETEX(q0, 0, 0);
                 VMOVQDto(x1, q0, 0);
                 LSRx(x1, x1, 31);
                 BFIx(gd, x1, 0, 1);
@@ -444,14 +468,14 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0x51:
             INST_NAME("SQRTPS Gx, Ex");
             nextop = F8;
-            GETEX(q0, 0);
+            GETEX(q0, 0, 0);
             GETGX_empty(v0);
             VFSQRTQS(v0, q0);
             break;
         case 0x52:
             INST_NAME("RSQRTPS Gx, Ex");
             nextop = F8;
-            GETEX(q0, 0);
+            GETEX(q0, 0, 0);
             GETGX_empty(q1);
             v0 = fpu_get_scratch(dyn);
             // more precise
@@ -467,7 +491,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0x53:
             INST_NAME("RCPPS Gx, Ex");
             nextop = F8;
-            GETEX(q0, 0);
+            GETEX(q0, 0, 0);
             GETGX_empty(q1);
             if(q0 == q1)
                 v1 = fpu_get_scratch(dyn);
@@ -481,22 +505,22 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0x54:
             INST_NAME("ANDPS Gx, Ex");
             nextop = F8;
-            GETEX(q0, 0);
-            GETGX(v0);
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
             VANDQ(v0, v0, q0);
             break;
         case 0x55:
             INST_NAME("ANDNPS Gx, Ex");
             nextop = F8;
-            GETEX(q0, 0);
-            GETGX(v0);
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
             VBICQ(v0, q0, v0);
             break;
         case 0x56:
             INST_NAME("ORPS Gx, Ex");
             nextop = F8;
-            GETEX(q0, 0);
-            GETGX(v0);
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
             VORRQ(v0, v0, q0);
             break;
         case 0x57:
@@ -508,65 +532,65 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 q0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 VEORQ(q0, q0, q0);
             } else {
-                q0 = sse_get_reg(dyn, ninst, x1, gd);
-                GETEX(q1, 0);
+                q0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+                GETEX(q1, 0, 0);
                 VEORQ(q0, q0, q1);
             }
             break;
         case 0x58:
             INST_NAME("ADDPS Gx, Ex");
             nextop = F8;
-            GETEX(q0, 0);
-            GETGX(v0);
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
             VFADDQS(v0, v0, q0);
             break;
         case 0x59:
             INST_NAME("MULPS Gx, Ex");
             nextop = F8;
-            GETEX(q0, 0);
-            GETGX(v0);
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
             VFMULQS(v0, v0, q0);
             break;
         case 0x5A:
             INST_NAME("CVTPS2PD Gx, Ex");
             nextop = F8;
-            GETEX(q0, 0);
-            GETGX(q1);
+            GETEX(q0, 0, 0);
+            GETGX(q1, 1);
             FCVTL(q1, q0);
             break;
         case 0x5B:
             INST_NAME("CVTDQ2PS Gx, Ex");
             nextop = F8;
-            GETEX(q0, 0);
+            GETEX(q0, 0, 0);
             GETGX_empty(q1);
             SCVTQFS(q1, q0);
             break;
         case 0x5C:
             INST_NAME("SUBPS Gx, Ex");
             nextop = F8;
-            GETEX(q0, 0);
-            GETGX(v0);
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
             VFSUBQS(v0, v0, q0);
             break;
         case 0x5D:
             INST_NAME("MINPS Gx, Ex");
             nextop = F8;
-            GETGX(v0);
-            GETEX(v1, 0);
+            GETGX(v0, 1);
+            GETEX(v1, 0, 0);
             VFMINQS(v0, v0, v1);
             break;
         case 0x5E:
             INST_NAME("DIVPS Gx, Ex");
             nextop = F8;
-            GETEX(q0, 0);
-            GETGX(v0);
+            GETEX(q0, 0, 0);
+            GETGX(v0, 1);
             VFDIVQS(v0, v0, q0);
             break;
         case 0x5F:
             INST_NAME("MAXPS Gx, Ex");
             nextop = F8;
-            GETGX(v0);
-            GETEX(v1, 0);
+            GETGX(v0, 1);
+            GETEX(v1, 0, 0);
             VFMAXQS(v0, v0, v1);
             break;
         case 0x60:
@@ -601,21 +625,21 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             SQXTN_8(d0, q0);
             break;
         case 0x64:
-            INST_NAME("PCMPGTB Gx,Ex");
+            INST_NAME("PCMPGTB Gm,Em");
             nextop = F8;
             GETGM(v0);
             GETEM(v1, 0);
             VCMGT_8(v0, v0, v1);
             break;
         case 0x65:
-            INST_NAME("PCMPGTW Gx,Ex");
+            INST_NAME("PCMPGTW Gm,Em");
             nextop = F8;
             GETGM(v0);
             GETEM(v1, 0);
             VCMGT_16(v0, v0, v1);
             break;
         case 0x66:
-            INST_NAME("PCMPGTD Gx,Ex");
+            INST_NAME("PCMPGTD Gm,Em");
             nextop = F8;
             GETGM(v0);
             GETEM(v1, 0);
@@ -628,7 +652,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             q0 = fpu_get_scratch(dyn);
             VMOVeD(q0, 0, v0, 0);
             if(MODREG) {
-                v1 = mmx_get_reg(dyn, ninst, x1, (nextop&7));
+                v1 = mmx_get_reg(dyn, ninst, x1, x2, x3, (nextop&7));
                 VMOVeD(q0, 1, v1, 0);
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, rex, 0, 0);
@@ -678,7 +702,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             INST_NAME("MOVD Gm, Ed");
             nextop = F8;
             gd = (nextop&0x38)>>3;
-            v0 = mmx_get_reg_empty(dyn, ninst, x3, gd);
+            v0 = mmx_get_reg_empty(dyn, ninst, x1, x2, x3, gd);
             if(MODREG) {
                 ed = xRAX + (nextop&7) + (rex.b<<3);
                 if(rex.w) {
@@ -687,7 +711,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     FMOVSw(v0, ed);
                 }
             } else {
-                v0 = mmx_get_reg_empty(dyn, ninst, x1, gd);
+                v0 = mmx_get_reg_empty(dyn, ninst, x1, x2, x3, gd);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<(2+rex.w), (1<<(2+rex.w))-1, rex, 0, 0);
                 if(rex.w) {
                     VLDR64_U12(v0, ed, fixedaddress);
@@ -701,11 +725,11 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             nextop = F8;
             GETG;
             if(MODREG) {
-                v1 = mmx_get_reg(dyn, ninst, x1, nextop&7); // no rex.b on MMX
-                v0 = mmx_get_reg_empty(dyn, ninst, x1, gd);
+                v1 = mmx_get_reg(dyn, ninst, x1, x2, x3, nextop&7); // no rex.b on MMX
+                v0 = mmx_get_reg_empty(dyn, ninst, x1, x2, x3, gd);
                 VMOVeD(v0, 0, v1, 0);
             } else {
-                v0 = mmx_get_reg_empty(dyn, ninst, x1, gd);
+                v0 = mmx_get_reg_empty(dyn, ninst, x1, x2, x3, gd);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<3, 7, rex, 0, 0);
                 VLDR64_U12(v0, ed, fixedaddress);
             }
@@ -716,8 +740,8 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             gd = (nextop&0x38)>>3;
             if(MODREG) {
                 u8 = F8;
-                v1 = mmx_get_reg(dyn, ninst, x1, (nextop&7));
-                v0 = mmx_get_reg_empty(dyn, ninst, x1, gd);
+                v1 = mmx_get_reg(dyn, ninst, x1, x2, x3, (nextop&7));
+                v0 = mmx_get_reg_empty(dyn, ninst, x1, x2, x3, gd);
                 if(u8==0x4E) {
                     if(v0==v1) {
                         VEXT_8(v0, v0, v0, 4); // Swap Up/Lower 32bits parts
@@ -773,7 +797,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     VTBL1_8(v0, v1, d0);
                 }
             } else {
-                v0 = mmx_get_reg_empty(dyn, ninst, x1, gd);
+                v0 = mmx_get_reg_empty(dyn, ninst, x1, x2, x3, gd);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, rex, 0, 1);
                 u8 = F8;
                 if (u8) {
@@ -947,7 +971,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0x77:
             INST_NAME("EMMS");
             // empty MMX, FPU now usable
-            mmx_purgecache(dyn, ninst, x1);
+            mmx_purgecache(dyn, ninst, 0, x1);
             /*emu->top = 0;
             emu->fpu_stack = 0;*/ //TODO: Check if something is needed here?
             break;
@@ -978,7 +1002,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             nextop = F8;
             GETGM(v0);
             if(MODREG) {
-                v1 = mmx_get_reg_empty(dyn, ninst, x1, nextop&7);
+                v1 = mmx_get_reg_empty(dyn, ninst, x1, x2, x3, nextop&7);
                 VMOV(v1, v0);
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<3, 7, rex, 0, 0);
@@ -987,21 +1011,30 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             break;
 
         #define GO(GETFLAGS, NO, YES, F)   \
-            READFLAGS(F);   \
-            i32_ = F32S;    \
-            BARRIER(2);     \
-            JUMP(addr+i32_);\
-            GETFLAGS;   \
-            if(dyn->insts[ninst].x64.jmp_insts==-1) {   \
-                /* out of the block */                  \
-                i32 = dyn->insts[ninst+1].address-(dyn->arm_size); \
-                Bcond(NO, i32);     \
-                jump_to_next(dyn, addr+i32_, 0, ninst); \
-            } else {    \
-                /* inside the block */  \
-                i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->arm_size);    \
-                Bcond(YES, i32);    \
-            }   \
+            READFLAGS(F);                                               \
+            i32_ = F32S;                                                \
+            BARRIER(BARRIER_MAYBE);                                     \
+            JUMP(addr+i32_, 1);                                         \
+            GETFLAGS;                                                   \
+            if(dyn->insts[ninst].x64.jmp_insts==-1 ||                   \
+                CHECK_CACHE()) {                                        \
+                /* out of the block */                                  \
+                i32 = dyn->insts[ninst].epilog-(dyn->native_size);      \
+                Bcond(NO, i32);                                         \
+                if(dyn->insts[ninst].x64.jmp_insts==-1) {               \
+                    if(!dyn->insts[ninst].x64.barrier)                  \
+                        fpu_purgecache(dyn, ninst, 1, x1, x2, x3);      \
+                    jump_to_next(dyn, addr+i32_, 0, ninst);             \
+                } else {                                                \
+                    fpuCacheTransform(dyn, ninst, x1, x2, x3);          \
+                    i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->native_size);    \
+                    B(i32);                                             \
+                }                                                       \
+            } else {                                                    \
+                /* inside the block */                                  \
+                i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->native_size);    \
+                Bcond(YES, i32);                                        \
+            }
 
         GOCOND(0x80, "J", "Id");
         #undef GO
@@ -1155,7 +1188,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     case 0:
                         INST_NAME("FXSAVE Ed");
                         MESSAGE(LOG_DUMP, "Need Optimization\n");
-                        fpu_purgecache(dyn, ninst, x1, x2, x3);
+                        fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
                         if(MODREG) {
                             DEFAULT;
                         } else {
@@ -1167,7 +1200,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     case 1:
                         INST_NAME("FXRSTOR Ed");
                         MESSAGE(LOG_DUMP, "Need Optimization\n");
-                        fpu_purgecache(dyn, ninst, x1, x2, x3);
+                        fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
                         if(MODREG) {
                             DEFAULT;
                         } else {
@@ -1511,29 +1544,29 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0xC2:
             INST_NAME("CMPPS Gx, Ex, Ib");
             nextop = F8;
-            GETGX(v0);
-            GETEX(v1, 1);
+            GETGX(v0, 1);
+            GETEX(v1, 0, 1);
             u8 = F8;
             switch(u8&7) {
                 // the inversion of the params in the comparison is there to handle NaN the same way SSE does
-                case 0: FCMEQQS(v0, v0, v1); break;   // Equal
-                case 1: FCMGTQS(v0, v1, v0); break;   // Less than
-                case 2: FCMGEQS(v0, v1, v0); break;   // Less or equal
-                case 3: FCMEQQS(v0, v0, v0); 
+                case 0: VFCMEQQS(v0, v0, v1); break;   // Equal
+                case 1: VFCMGTQS(v0, v1, v0); break;   // Less than
+                case 2: VFCMGEQS(v0, v1, v0); break;   // Less or equal
+                case 3: VFCMEQQS(v0, v0, v0); 
                         if(v0!=v1) {
                             q0 = fpu_get_scratch(dyn); 
-                            FCMEQQS(q0, v1, v1); 
+                            VFCMEQQS(q0, v1, v1); 
                             VANDQ(v0, v0, q0);
                         }
                         VMVNQ(v0, v0); 
                         break;   // NaN (NaN is not equal to himself)
-                case 4: FCMEQQS(v0, v0, v1); VMVNQ(v0, v0); break;   // Not Equal (or unordered on ARM, not on X86...)
-                case 5: FCMGTQS(v0, v1, v0); VMVNQ(v0, v0); break;   // Greater or equal or unordered
-                case 6: FCMGEQS(v0, v1, v0); VMVNQ(v0, v0); break;   // Greater or unordered
-                case 7: FCMEQQS(v0, v0, v0); 
+                case 4: VFCMEQQS(v0, v0, v1); VMVNQ(v0, v0); break;   // Not Equal (or unordered on ARM, not on X86...)
+                case 5: VFCMGTQS(v0, v1, v0); VMVNQ(v0, v0); break;   // Greater or equal or unordered
+                case 6: VFCMGEQS(v0, v1, v0); VMVNQ(v0, v0); break;   // Greater or unordered
+                case 7: VFCMEQQS(v0, v0, v0); 
                         if(v0!=v1) {
                             q0 = fpu_get_scratch(dyn); 
-                            FCMEQQS(q0, v1, v1); 
+                            VFCMEQQS(q0, v1, v1); 
                             VANDQ(v0, v0, q0);
                         }
                         break;   // not NaN
@@ -1581,7 +1614,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0xC6:
             INST_NAME("SHUFPS Gx, Ex, Ib");
             nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
             if(!MODREG)
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, rex, 0, 1);
             u8 = F8;
@@ -1592,7 +1625,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             }
             // second two from Ex
             if(MODREG) {
-                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3));
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);
                 for(int i=2; i<4; ++i) {
                     VMOVeS(d0, i, v1, (u8>>(i*2)&3));
                 }
@@ -1744,7 +1777,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             if((nextop&0xC0)==0xC0) {
                 DEFAULT;
             } else {
-                v0 = mmx_get_reg(dyn, ninst, x1, gd);
+                v0 = mmx_get_reg(dyn, ninst, x1, x2, x3, gd);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<3, 7, rex, 0, 0);
                 VSTR64_U12(v0, ed, fixedaddress);
             }
@@ -1792,10 +1825,10 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             gd = ((nextop&0x38)>>3);
             if(MODREG && ((nextop&7))==gd) {
                 // special case for PXOR Gx, Gx
-                q0 = mmx_get_reg_empty(dyn, ninst, x1, gd);
+                q0 = mmx_get_reg_empty(dyn, ninst, x1, x2, x3, gd);
                 VEOR(q0, q0, q0);
             } else {
-                q0 = mmx_get_reg(dyn, ninst, x1, gd);
+                q0 = mmx_get_reg(dyn, ninst, x1, x2, x3, gd);
                 GETEM(q1, 0);
                 VEOR(q0, q0, q1);
             }

src/dynarec/arm64/dynarec_arm64_64.c

@@ -15,10 +15,10 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
+#include "dynarec_native.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_helper.h"
 #include "dynarec_arm64_functions.h"
 
@@ -31,10 +31,10 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
     uint8_t opcode = F8;
     uint8_t nextop;
     uint8_t u8;
-    uint8_t gd, ed, eb1, eb2;
+    uint8_t gd, ed, eb1, eb2, gb1, gb2;
     uint8_t wback, wb1, wb2, wb;
     int64_t i64, j64;
-    int v0;
+    int v0, v1;
     int q0;
     int d0;
     int64_t fixedaddress;
@@ -42,10 +42,13 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
     MAYUSE(eb2);
     MAYUSE(wb1);
     MAYUSE(wb2);
+    MAYUSE(gb1);
+    MAYUSE(gb2);
     MAYUSE(j64);
     MAYUSE(d0);
     MAYUSE(q0);
     MAYUSE(v0);
+    MAYUSE(v1);
 
     while((opcode==0xF2) || (opcode==0xF3)) {
         rep = opcode-0xF1;
@@ -82,8 +85,8 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                             GETG;
                             if(MODREG) {
                                 ed = (nextop&7)+ (rex.b<<3);
-                                v0 = sse_get_reg(dyn, ninst, x1, gd);
-                                d0 = sse_get_reg(dyn, ninst, x1, ed);
+                                v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+                                d0 = sse_get_reg(dyn, ninst, x1, ed, 0);
                                 VMOVeD(v0, 0, d0, 0);
                             } else {
                                 grab_segdata(dyn, addr, ninst, x4, seg);
@@ -98,8 +101,8 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                             nextop = F8;
                             GETG;
                             if(MODREG) {
-                                v0 = sse_get_reg(dyn, ninst, x1, gd);
-                                q0 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3));
+                                v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+                                q0 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3), 0);
                                 VMOVeS(v0, 0, q0, 0);
                             } else {
                                 grab_segdata(dyn, addr, ninst, x4, seg);
@@ -115,14 +118,30 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     break;
                 case 0x11:
                     switch(rep) {
+                        case 0:
+                            INST_NAME("MOVUPS Ex,Gx");
+                            nextop = F8;
+                            GETG;
+                            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
+                            if(MODREG) {
+                                ed = (nextop&7)+(rex.b<<3);
+                                v1 = sse_get_reg_empty(dyn, ninst, x1, ed);
+                                VMOVQ(v1, v0);
+                            } else {
+                                grab_segdata(dyn, addr, ninst, x4, seg);
+                                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<4, 15, rex, 0, 0);
+                                ADDx_REG(x4, x4, ed);
+                                VSTR128_U12(v0, x4, fixedaddress);
+                            }
+                            break;
                         case 1:
                             INST_NAME("MOVSD Ex, Gx");
                             nextop = F8;
                             GETG;
-                            v0 = sse_get_reg(dyn, ninst, x1, gd);
+                            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
                             if(MODREG) {
                                 ed = (nextop&7)+ (rex.b<<3);
-                                d0 = sse_get_reg(dyn, ninst, x1, ed);
+                                d0 = sse_get_reg(dyn, ninst, x1, ed, 1);
                                 VMOVeD(d0, 0, v0, 0);
                             } else {
                                 grab_segdata(dyn, addr, ninst, x4, seg);
@@ -135,9 +154,9 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                             INST_NAME("MOVSS Ex, Gx");
                             nextop = F8;
                             GETG;
-                            v0 = sse_get_reg(dyn, ninst, x1, gd);
+                            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
                             if(MODREG) {
-                                q0 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3));
+                                q0 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3), 1);
                                 VMOVeS(q0, 0, v0, 0);
                             } else {
                                 grab_segdata(dyn, addr, ninst, x4, seg);
@@ -151,6 +170,29 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     }
                     break;
 
+                case 0x6F:
+                    switch(rep) {
+                        case 2:
+                            INST_NAME("MOVDQU Gx,Ex");// no alignment constraint on NEON here, so same as MOVDQA
+                            nextop = F8;
+                            GETG;
+                            v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                            if(MODREG) {
+                                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);
+                                VMOVQ(v0, v1);
+                            } else {
+                                grab_segdata(dyn, addr, ninst, x4, seg);
+                                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<4, 15, rex, 0, 0);
+                                ADDx_REG(x4, x4, ed);
+                                VLDR128_U12(v0, ed, fixedaddress);
+                            }
+                            break;
+                        default:
+                            DEFAULT;
+                    }
+                    break;
+
+
                 case 0xAF:
                     INST_NAME("IMUL Gd, Ed");
                     SETFLAGS(X_ALL, SF_PENDING);
@@ -184,6 +226,27 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     }
                     break;
 
+                case 0xB6:
+                    INST_NAME("MOVZX Gd, Eb");
+                    nextop = F8;
+                    grab_segdata(dyn, addr, ninst, x4, seg);
+                    GETGD;
+                    if(MODREG) {
+                        if(rex.rex) {
+                            eb1 = xRAX+(nextop&7)+(rex.b<<3);
+                            eb2 = 0;                \
+                        } else {
+                            ed = (nextop&7);
+                            eb1 = xRAX+(ed&3);  // Ax, Cx, Dx or Bx
+                            eb2 = (ed&4)>>2;    // L or H
+                        }
+                        UBFXxw(gd, eb1, eb2*8, 8);
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, rex, 0, 0);
+                        LDRB_REG(gd, ed, x4);
+                    }
+                    break;
+
                 default:
                     DEFAULT;
             }
@@ -209,16 +272,49 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             emit_cmp32(dyn, ninst, rex, ed, gd, x3, x4, x5);
             break;
 
+        case 0x3B:
+            INST_NAME("CMP Gd, Seg:Ed");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            GETGD;
+            GETEDO(x4, 0);
+            emit_cmp32(dyn, ninst, rex, gd, ed, x3, x4, x5);
+            break;
+
+        case 0x63:
+            INST_NAME("MOVSXD Gd, Ed");
+            nextop = F8;
+            GETGD;
+            if(rex.w) {
+                if(MODREG) {   // reg <= reg
+                    SXTWx(gd, xRAX+(nextop&7)+(rex.b<<3));
+                } else {                    // mem <= reg
+                    grab_segdata(dyn, addr, ninst, x4, seg);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, rex, 0, 0);
+                    LDRSW_REG(gd, ed, x4);
+                }
+            } else {
+                if(MODREG) {   // reg <= reg
+                    MOVw_REG(gd, xRAX+(nextop&7)+(rex.b<<3));
+                } else {                    // mem <= reg
+                    grab_segdata(dyn, addr, ninst, x4, seg);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0, 0, rex, 0, 0);
+                    LDRw_REG(gd, ed, x4);
+                }
+            }
+            break;
+
         case 0x66:
             addr = dynarec64_6664(dyn, addr, ip, ninst, rex, rep, ok, need_epilog);
             break;
 
         case 0x80:
             nextop = F8;
-            grab_segdata(dyn, addr, ninst, x1, seg);
             switch((nextop>>3)&7) {
                 case 0: //ADD
                     INST_NAME("ADD Eb, Ib");
+                    grab_segdata(dyn, addr, ninst, x1, seg);
                     SETFLAGS(X_ALL, SF_SET_PENDING);
                     GETEBO(x1, 1);
                     u8 = F8;
@@ -227,6 +323,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     break;
                 case 1: //OR
                     INST_NAME("OR Eb, Ib");
+                    grab_segdata(dyn, addr, ninst, x1, seg);
                     SETFLAGS(X_ALL, SF_SET_PENDING);
                     GETEBO(x1, 1);
                     u8 = F8;
@@ -235,6 +332,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     break;
                 case 2: //ADC
                     INST_NAME("ADC Eb, Ib");
+                    grab_segdata(dyn, addr, ninst, x1, seg);
                     READFLAGS(X_CF);
                     SETFLAGS(X_ALL, SF_SET_PENDING);
                     GETEBO(x1, 1);
@@ -244,6 +342,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     break;
                 case 3: //SBB
                     INST_NAME("SBB Eb, Ib");
+                    grab_segdata(dyn, addr, ninst, x1, seg);
                     READFLAGS(X_CF);
                     SETFLAGS(X_ALL, SF_SET_PENDING);
                     GETEBO(x1, 1);
@@ -253,6 +352,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     break;
                 case 4: //AND
                     INST_NAME("AND Eb, Ib");
+                    grab_segdata(dyn, addr, ninst, x1, seg);
                     SETFLAGS(X_ALL, SF_SET_PENDING);
                     GETEBO(x1, 1);
                     u8 = F8;
@@ -261,6 +361,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     break;
                 case 5: //SUB
                     INST_NAME("SUB Eb, Ib");
+                    grab_segdata(dyn, addr, ninst, x1, seg);
                     SETFLAGS(X_ALL, SF_SET_PENDING);
                     GETEBO(x1, 1);
                     u8 = F8;
@@ -269,6 +370,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     break;
                 case 6: //XOR
                     INST_NAME("XOR Eb, Ib");
+                    grab_segdata(dyn, addr, ninst, x1, seg);
                     SETFLAGS(X_ALL, SF_SET_PENDING);
                     GETEBO(x1, 1);
                     u8 = F8;
@@ -277,6 +379,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     break;
                 case 7: //CMP
                     INST_NAME("CMP Eb, Ib");
+                    grab_segdata(dyn, addr, ninst, x1, seg);
                     SETFLAGS(X_ALL, SF_SET_PENDING);
                     GETEBO(x1, 1);
                     u8 = F8;
@@ -369,7 +472,40 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     break;
             }
             break;
-            
+        case 0x8A:
+            INST_NAME("MOV Gb, Eb");
+            nextop = F8;
+            if(rex.rex) {
+                gb1 = gd = xRAX+((nextop&0x38)>>3)+(rex.r<<3);
+                gb2=0;
+            } else {
+                gd = (nextop&0x38)>>3;
+                gb1 = xRAX+(gd&3);
+                gb2 = ((gd&4)>>2);
+            }
+            if(MODREG) {
+                if(rex.rex) {
+                    wback = xRAX+(nextop&7)+(rex.b<<3);
+                    wb2 = 0;
+                } else {
+                    wback = (nextop&7);
+                    wb2 = (wback>>2);
+                    wback = xRAX+(wback&3);
+                }
+                if(wb2) {
+                    UBFXw(x4, wback, wb2*8, 8);
+                    ed = x4;
+                } else {
+                    ed = wback;
+                }
+            } else {
+                grab_segdata(dyn, addr, ninst, x4, seg);
+                addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0, 0, rex, 0, 0);
+                LDRB_REG(x4, wback, x4);
+                ed = x4;
+            }
+            BFIx(gb1, ed, gb2*8, 8);
+            break;
         case 0x89:
             INST_NAME("MOV Seg:Ed, Gd");
             grab_segdata(dyn, addr, ninst, x4, seg);
@@ -441,14 +577,14 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("ROL Ed, 1");
-                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
                     GETEDO(x6, 0);
                     emit_rol32c(dyn, ninst, rex, ed, 1, x3, x4);
                     WBACKO(x6);
                     break;
                 case 1:
                     INST_NAME("ROR Ed, 1");
-                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
                     GETEDO(x6, 0);
                     emit_ror32c(dyn, ninst, rex, ed, 1, x3, x4);
                     WBACKO(x6);
@@ -717,7 +853,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         MOVw_REG(xRAX, x2);
                         MOVw_REG(xRDX, x4);
                     } else {
-                        if(ninst && dyn->insts 
+                        if(ninst
                            && dyn->insts[ninst-1].x64.addr 
                            && *(uint8_t*)(dyn->insts[ninst-1].x64.addr)==0x31 
                            && *(uint8_t*)(dyn->insts[ninst-1].x64.addr+1)==0xD2) {
@@ -787,14 +923,14 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             switch((nextop>>3)&7) {
                 case 0: // INC Ed
                     INST_NAME("INC Ed");
-                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
+                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
                     GETEDO(x6, 0);
                     emit_inc32(dyn, ninst, rex, ed, x3, x4);
                     WBACKO(x6);
                     break;
                 case 1: //DEC Ed
                     INST_NAME("DEC Ed");
-                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
+                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
                     GETEDO(x6, 0);
                     emit_dec32(dyn, ninst, rex, ed, x3, x4);
                     WBACKO(x6);
@@ -809,13 +945,12 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         SETFLAGS(X_ALL, SF_SET);    //Hack to put flag in "don't care" state
                     }
                     GETEDOx(x6, 0);
-                    BARRIER(1);
-                    BARRIER_NEXT(1);
+                    BARRIER(BARRIER_FLOAT);
                     if(!dyn->insts || ninst==dyn->size-1) {
                         *need_epilog = 0;
                         *ok = 0;
                     }
-                    GETIP(addr);
+                    GETIP_(addr);
                     PUSH1(xRIP);
                     jump_to_next(dyn, 0, ed, ninst);
                     break;

src/dynarec/arm64/dynarec_arm64_66.c

@@ -15,10 +15,10 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
+#include "dynarec_native.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_helper.h"
 #include "dynarec_arm64_functions.h"
 
@@ -328,6 +328,25 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             GWBACK;
             break;
 
+        case 0x70:
+        case 0x71:
+        case 0x72:
+        case 0x73:
+        case 0x74:
+        case 0x75:
+        case 0x76:
+        case 0x77:
+        case 0x78:
+        case 0x79:
+        case 0x7a:
+        case 0x7b:
+        case 0x7c:
+        case 0x7d:
+        case 0x7e:
+        case 0x7f:
+            // just use regular conditional jump
+            return dynarec64_00(dyn, addr-1, ip, ninst, rex, rep, ok, need_epilog);
+
         case 0x81:
         case 0x83:
             nextop = F8;
@@ -848,14 +867,14 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             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, 0);
                     emit_inc16(dyn, ninst, x1, x2, x4);
                     EWBACK;
                     break;
                 case 1:
                     INST_NAME("DEC Ew");
-                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
+                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
                     GETEW(x1, 0);
                     emit_dec16(dyn, ninst, x1, x2, x4);
                     EWBACK;

src/dynarec/arm64/dynarec_arm64_6664.c

@@ -15,10 +15,10 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
+#include "dynarec_native.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_helper.h"
 #include "dynarec_arm64_functions.h"
 
@@ -56,7 +56,7 @@ uintptr_t dynarec64_6664(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 INST_NAME("MOVQ Ex, Gx");
                 nextop = F8;
                 GETG;
-                v0 = sse_get_reg(dyn, ninst, x1, gd);
+                v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
                 if(MODREG) {
                     v1 = sse_get_reg_empty(dyn, ninst, x1, (nextop&7) + (rex.b<<3));
                     FMOVD(v1, v0);

src/dynarec/arm64/dynarec_arm64_67.c

@@ -15,20 +15,20 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
+#include "dynarec_native.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_helper.h"
 #include "dynarec_arm64_functions.h"
 
-#define GETGX(a)                        \
+#define GETGX(a, w)                     \
     gd = ((nextop&0x38)>>3)+(rex.r<<3); \
-    a = sse_get_reg(dyn, ninst, x1, gd)
+    a = sse_get_reg(dyn, ninst, x1, gd, w)
 
 #define GETGM(a)                        \
     gd = ((nextop&0x38)>>3);            \
-    a = mmx_get_reg(dyn, ninst, x1, gd)
+    a = mmx_get_reg(dyn, ninst, x1, x2, x3, gd)
 
 #define GETGm   gd = ((nextop&0x38)>>3)
 
@@ -78,9 +78,9 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         if(opcode==0x2F) {INST_NAME("COMISS Gx, Ex");} else {INST_NAME("UCOMISS Gx, Ex");}
                         SETFLAGS(X_ALL, SF_SET);
                         nextop = F8;
-                        GETGX(v0);
+                        GETGX(v0, 0);
                         if(MODREG) {
-                            s0 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3));
+                            s0 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3), 0);
                         } else {
                             s0 = fpu_get_scratch(dyn);
                             addr = geted32(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<2, 3, rex, 0, 0);
@@ -96,11 +96,11 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     nextop = F8;
                     GETGm;
                     if(MODREG) {
-                        v1 = mmx_get_reg(dyn, ninst, x1, nextop&7); // no rex.b on MMX
-                        v0 = mmx_get_reg_empty(dyn, ninst, x1, gd);
+                        v1 = mmx_get_reg(dyn, ninst, x1, x2, x3, nextop&7); // no rex.b on MMX
+                        v0 = mmx_get_reg_empty(dyn, ninst, x1, x2, x3, gd);
                         VMOVeD(v0, 0, v1, 0);
                     } else {
-                        v0 = mmx_get_reg_empty(dyn, ninst, x1, gd);
+                        v0 = mmx_get_reg_empty(dyn, ninst, x1, x2, x3, gd);
                         addr = geted32(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<3, 7, rex, 0, 0);
                         VLDR64_U12(v0, ed, fixedaddress);
                     }
@@ -111,7 +111,7 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     nextop = F8;
                     GETGM(v0);
                     if(MODREG) {
-                        v1 = mmx_get_reg_empty(dyn, ninst, x1, nextop&7);
+                        v1 = mmx_get_reg_empty(dyn, ninst, x1, x2, x3, nextop&7);
                         VMOV(v1, v0);
                     } else {
                         addr = geted32(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<3, 7, rex, 0, 0);
@@ -156,7 +156,7 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("ROL Ed, Ib");
-                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
                     GETED32(1);
                     u8 = (F8)&(rex.w?0x3f:0x1f);
                     emit_rol32c(dyn, ninst, rex, ed, u8, x3, x4);
@@ -164,7 +164,7 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     break;
                 case 1:
                     INST_NAME("ROR Ed, Ib");
-                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
                     GETED32(1);
                     u8 = (F8)&(rex.w?0x3f:0x1f);
                     emit_ror32c(dyn, ninst, rex, ed, u8, x3, x4);
@@ -224,17 +224,26 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             }
             break;
             
-        #define GO(NO, YES)   \
-            BARRIER(2); \
-            JUMP(addr+i8);\
-            if(dyn->insts[ninst].x64.jmp_insts==-1) {   \
-                /* out of the block */                  \
-                i32 = dyn->insts[ninst+1].address-(dyn->arm_size); \
-                Bcond(NO, i32);     \
-                jump_to_next(dyn, addr+i8, 0, ninst); \
+        #define GO(NO, YES)                                             \
+            BARRIER(BARRIER_MAYBE);                                     \
+            JUMP(addr+i8, 1);                                           \
+            if(dyn->insts[ninst].x64.jmp_insts==-1 ||                   \
+                CHECK_CACHE()) {                                        \
+                /* out of the block */                                  \
+                i32 = dyn->insts[ninst].epilog-(dyn->native_size);      \
+                Bcond(NO, i32);                                         \
+                if(dyn->insts[ninst].x64.jmp_insts==-1) {               \
+                    if(!dyn->insts[ninst].x64.barrier)                  \
+                        fpu_purgecache(dyn, ninst, 1, x1, x2, x3);      \
+                    jump_to_next(dyn, addr+i8, 0, ninst);               \
+                } else {                                                \
+                    fpuCacheTransform(dyn, ninst, x1, x2, x3);          \
+                    i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->native_size);\
+                    B(i32);                                             \
+                }                                                       \
             } else {    \
                 /* inside the block */  \
-                i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->arm_size);    \
+                i32 = dyn->insts[dyn->insts[ninst].x64.jmp_insts].address-(dyn->native_size);    \
                 Bcond(YES, i32);    \
             }
         case 0xE0:

src/dynarec/arm64/dynarec_arm64_d8.c

@@ -15,11 +15,11 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
 #include "emu/x87emu_private.h"
+#include "dynarec_native.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_helper.h"
 #include "dynarec_arm64_functions.h"
 
@@ -48,9 +48,13 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FADDD(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)) {
+                FADDS(v1, v1, v2);
+            } else {
+                FADDD(v1, v1, v2);
+            }
             break;
         case 0xC8:
         case 0xC9:
@@ -61,9 +65,13 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FMULD(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)) {
+                FMULS(v1, v1, v2);
+            } else {
+                FMULD(v1, v1, v2);
+            }
             break;
         case 0xD0:
         case 0xD1:
@@ -74,9 +82,13 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FCMPD(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)) {
+                FCMPS(v1, v2);
+            } else {
+                FCMPD(v1, v2);
+            }
             FCOM(x1, x2, x3);
             break;
         case 0xD8:
@@ -88,11 +100,15 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FCMPD(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)) {
+                FCMPS(v1, v2);
+            } else {
+                FCMPD(v1, v2);
+            }
             FCOM(x1, x2, x3);
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
             break;
         case 0xE0:
         case 0xE1:
@@ -103,9 +119,13 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FSUBD(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)) {
+                FSUBS(v1, v1, v2);
+            } else {
+                FSUBD(v1, v1, v2);
+            }
             break;
         case 0xE8:
         case 0xE9:
@@ -116,9 +136,13 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FSUBD(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)) {
+                FSUBS(v1, v2, v1);
+            } else {
+                FSUBD(v1, v2, v1);
+            }
             break;
         case 0xF0:
         case 0xF1:
@@ -129,9 +153,13 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FDIVD(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)) {
+                FDIVS(v1, v1, v2);
+            } else {
+                FDIVD(v1, v1, v2);
+            }
             break;
         case 0xF8:
         case 0xF9:
@@ -142,87 +170,123 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FDIVD(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)) {
+                FDIVS(v1, v2, v1);
+            } else {
+                FDIVD(v1, v2, v1);
+            }
             break;
       
         default:
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("FADD ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
                     s0 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0xfff<<2, 3, rex, 0, 0);
                     VLDR32_U12(s0, ed, fixedaddress);
-                    FCVT_D_S(s0, s0);
-                    FADDD(v1, v1, s0);
+                    if(ST_IS_F(0)) {
+                        FADDS(v1, v1, s0);
+                    } else {
+                        FCVT_D_S(s0, s0);
+                        FADDD(v1, v1, s0);
+                    }
                     break;
                 case 1:
                     INST_NAME("FMUL ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
                     s0 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0xfff<<2, 3, rex, 0, 0);
                     VLDR32_U12(s0, ed, fixedaddress);
-                    FCVT_D_S(s0, s0);
-                    FMULD(v1, v1, s0);
+                    if(ST_IS_F(0)) {
+                        FMULS(v1, v1, s0);
+                    } else {
+                        FCVT_D_S(s0, s0);
+                        FMULD(v1, v1, s0);
+                    }
                     break;
                 case 2:
                     INST_NAME("FCOM ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
                     s0 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0xfff<<2, 3, rex, 0, 0);
                     VLDR32_U12(s0, ed, fixedaddress);
-                    FCVT_D_S(s0, s0);
-                    FCMPD(v1, s0);
+                    if(ST_IS_F(0)) {
+                        FCMPS(v1, s0);
+                    } else {
+                        FCVT_D_S(s0, s0);
+                        FCMPD(v1, s0);
+                    }
                     FCOM(x1, x2, x3);
                     break;
                 case 3:
                     INST_NAME("FCOMP ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
                     s0 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0xfff<<2, 3, rex, 0, 0);
                     VLDR32_U12(s0, ed, fixedaddress);
-                    FCVT_D_S(s0, s0);
-                    FCMPD(v1, s0);
+                    if(ST_IS_F(0)) {
+                        FCMPS(v1, s0);
+                    } else {
+                        FCVT_D_S(s0, s0);
+                        FCMPD(v1, s0);
+                    }
                     FCOM(x1, x2, x3);
-                    x87_do_pop(dyn, ninst);
+                    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(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0xfff<<2, 3, rex, 0, 0);
                     VLDR32_U12(s0, ed, fixedaddress);
-                    FCVT_D_S(s0, s0);
-                    FSUBD(v1, v1, s0);
+                    if(ST_IS_F(0)) {
+                        FSUBS(v1, v1, s0);
+                    } else {
+                        FCVT_D_S(s0, s0);
+                        FSUBD(v1, v1, s0);
+                    }
                     break;
                 case 5:
                     INST_NAME("FSUBR ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
                     s0 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0xfff<<2, 3, rex, 0, 0);
                     VLDR32_U12(s0, ed, fixedaddress);
-                    FCVT_D_S(s0, s0);
-                    FSUBD(v1, s0, v1);
+                    if(ST_IS_F(0)) {
+                        FSUBS(v1, s0, v1);
+                    } else {
+                        FCVT_D_S(s0, s0);
+                        FSUBD(v1, s0, v1);
+                    }
                     break;
                 case 6:
                     INST_NAME("FDIV ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
                     s0 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0xfff<<2, 3, rex, 0, 0);
                     VLDR32_U12(s0, ed, fixedaddress);
-                    FCVT_D_S(s0, s0);
-                    FDIVD(v1, v1, s0);
+                    if(ST_IS_F(0)) {
+                        FDIVS(v1, v1, s0);
+                    } else {
+                        FCVT_D_S(s0, s0);
+                        FDIVD(v1, v1, s0);
+                    }
                     break;
                 case 7:
                     INST_NAME("FDIVR ST0, float[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
                     s0 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0xfff<<2, 3, rex, 0, 0);
                     VLDR32_U12(s0, ed, fixedaddress);
-                    FCVT_D_S(s0, s0);
-                    FDIVD(v1, s0, v1);
+                    if(ST_IS_F(0)) {
+                        FDIVS(v1, s0, v1);
+                    } else {
+                        FCVT_D_S(s0, s0);
+                        FDIVD(v1, s0, v1);
+                    }
                     break;
                 default:
                     DEFAULT;

src/dynarec/arm64/dynarec_arm64_d9.c

@@ -15,11 +15,11 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
 #include "emu/x87emu_private.h"
+#include "dynarec_native.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_helper.h"
 #include "dynarec_arm64_functions.h"
 
@@ -50,12 +50,18 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0xC6:
         case 0xC7:
             INST_NAME("FLD STx");
-            v1 = x87_get_st(dyn, ninst, x1, x2, nextop&7);
-            v2 = x87_do_push(dyn, ninst);
-            FMOVD(v2, v1);
+            v2 = x87_do_push(dyn, ninst, x1, X87_ST(nextop&7));
+            v1 = x87_get_st(dyn, ninst, x1, x2, (nextop&7)+1, X87_COMBINE(0, (nextop&7)+1));
+            if(ST_IS_F(0)) {
+                FMOVS(v2, v1);
+            } else {
+                FMOVD(v2, v1);
+            }
             break;
 
         case 0xC8:
+            INST_NAME("FXCH ST0");
+            break;
         case 0xC9:
         case 0xCA:
         case 0xCB:
@@ -65,11 +71,8 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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:
@@ -78,19 +81,31 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
 
         case 0xE0:
             INST_NAME("FCHS");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            FNEGD(v1, v1);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+            if(ST_IS_F(0)) {
+                FNEGS(v1, v1);
+            } else {
+                FNEGD(v1, v1);
+            }
             break;
         case 0xE1:
             INST_NAME("FABS");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            FABSD(v1, v1);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+            if(ST_IS_F(0)) {
+                FABSS(v1, v1);
+            } else {
+                FABSD(v1, v1);
+            }
             break;
 
         case 0xE4:
             INST_NAME("FTST");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            FCMPD_0(v1);
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+            if(ST_IS_F(0)) {
+                FCMPS_0(v1);
+            } else {
+                FCMPD_0(v1);
+            }
             FCOM(x1, x2, x3);   // same flags...
             break;
         case 0xE5:
@@ -102,64 +117,44 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
 
         case 0xE8:
             INST_NAME("FLD1");
-            v1 = x87_do_push(dyn, ninst);
-            FTABLE64(v1, 1.0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                FMOVS_8(v1, 0b01110000);
+            } else {
+                FMOVD_8(v1, 0b01110000);
+            }
             break;
         case 0xE9:
             INST_NAME("FLDL2T");
-            v1 = x87_do_push(dyn, ninst);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
             FTABLE64(v1, L2T);
             break;
         case 0xEA:     
             INST_NAME("FLDL2E");
-            v1 = x87_do_push(dyn, ninst);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
             FTABLE64(v1, L2E);
             break;
         case 0xEB:
             INST_NAME("FLDPI");
-            v1 = x87_do_push(dyn, ninst);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
             FTABLE64(v1, PI);
             break;
         case 0xEC:
             INST_NAME("FLDLG2");
-            v1 = x87_do_push(dyn, ninst);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
             FTABLE64(v1, LG2);
             break;
         case 0xED:
             INST_NAME("FLDLN2");
-            v1 = x87_do_push(dyn, ninst);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
             FTABLE64(v1, LN2);
             break;
         case 0xEE:
             INST_NAME("FLDZ");
-            v1 = x87_do_push(dyn, ninst);
-            FTABLE64(v1, 0.0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            VEOR(v1, v1, v1);
             break;
 
-        case 0xFA:
-            INST_NAME("FSQRT");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-            FSQRTD(v1, v1);
-            break;
-
-        case 0xFC:
-            INST_NAME("FRNDINT");
-            MESSAGE(LOG_DUMP, "Need Optimization\n");
-            // use C helper for now, nothing staightforward is available
-            x87_forget(dyn, ninst, x1, x2, 0);
-            CALL(arm_frndint, -1);
-            /*
-            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");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
@@ -172,15 +167,19 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             x87_forget(dyn, ninst, x1, x2, 0);
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(arm_fyl2x, -1);
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
             break;
         case 0xF2:
-            INST_NAME("FTAN");
+            INST_NAME("FPTAN");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
             x87_forget(dyn, ninst, x1, x2, 0);
             CALL(arm_ftan, -1);
-            v1 = x87_do_push(dyn, ninst);
-            FTABLE64(v1, 1.0);
+            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            if(ST_IS_F(0)) {
+                FMOVS_8(v1, 0b01110000);
+            } else {
+                FMOVD_8(v1, 0b01110000);
+            }
             break;
         case 0xF3:
             INST_NAME("FPATAN");
@@ -188,7 +187,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             x87_forget(dyn, ninst, x1, x2, 0);
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(arm_fpatan, -1);
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
             break;
         case 0xF4:
             INST_NAME("FXTRACT");
@@ -206,7 +205,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             break;
         case 0xF6:
             INST_NAME("FDECSTP");
-            fpu_purgecache(dyn, ninst, x1, x2, x3);
+            fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
             LDRw_U12(x2, xEmu, offsetof(x64emu_t, top));
             SUBw_U12(x2, x2, 1);
             ANDw_mask(x2, x2, 0, 2);    //mask=7
@@ -214,7 +213,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             break;
         case 0xF7:
             INST_NAME("FINCSTP");
-            fpu_purgecache(dyn, ninst, x1, x2, x3);
+            fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
             LDRw_U12(x2, xEmu, offsetof(x64emu_t, top));
             ADDw_U12(x2, x2, 1);
             ANDw_mask(x2, x2, 0, 2);    //mask=7
@@ -233,7 +232,16 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             x87_forget(dyn, ninst, x1, x2, 0);
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(arm_fyl2xp1, -1);
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
+            break;
+        case 0xFA:
+            INST_NAME("FSQRT");
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+            if(ST_IS_F(0)) {
+                FSQRTS(v1, v1);
+            } else {
+                FSQRTD(v1, v1);
+            }
             break;
         case 0xFB:
             INST_NAME("FSINCOS");
@@ -242,6 +250,24 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(arm_fsincos, -1);
             break;
+        case 0xFC:
+            INST_NAME("FRNDINT");
+            MESSAGE(LOG_DUMP, "Need Optimization\n");
+            // use C helper for now, nothing staightforward is available
+            x87_forget(dyn, ninst, x1, x2, 0);
+            CALL(arm_frndint, -1);
+            /*
+            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 0xFD:
             INST_NAME("FSCALE");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
@@ -288,33 +314,46 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("FLD ST0, float[ED]");
-                    v1 = x87_do_push(dyn, ninst);
-                    s0 = fpu_get_scratch(dyn);
+                    v1 = x87_do_push(dyn, ninst, x1, box64_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+                    if(ST_IS_F(0))
+                        s0 = v1;
+                    else
+                        s0 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0xfff<<2, 3, rex, 0, 0);
                     VLDR32_U12(s0, ed, fixedaddress);
-                    FCVT_D_S(v1, s0);
+                    if(!ST_IS_F(0)) {
+                        FCVT_D_S(v1, s0);
+                    }
                     break;
                 case 2:
                     INST_NAME("FST float[ED], ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-                    s0 = fpu_get_scratch(dyn);
-                    FCVT_S_D(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(dyn);
+                        FCVT_S_D(s0, v1);
+                    }
                     addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0xfff<<2, 3, rex, 0, 0);
                     VSTR32_U12(s0, ed, fixedaddress);
                     break;
                 case 3:
                     INST_NAME("FSTP float[ED], ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
-                    s0 = fpu_get_scratch(dyn);
-                    FCVT_S_D(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(dyn);
+                        FCVT_S_D(s0, v1);
+                    }
                     addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0xfff<<2, 3, rex, 0, 0);
                     VSTR32_U12(s0, ed, fixedaddress);
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                     break;
                 case 4:
                     INST_NAME("FLDENV Ed");
                     MESSAGE(LOG_DUMP, "Need Optimization\n");
-                    fpu_purgecache(dyn, ninst, x1, x2, x3); // maybe only x87, not SSE?
+                    fpu_purgecache(dyn, ninst, 0, x1, x2, x3); // maybe only x87, not SSE?
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, rex, 0, 0);
                     if(ed!=x1) {
                         MOVx_REG(x1, ed);
@@ -332,7 +371,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 case 6:
                     INST_NAME("FNSTENV Ed");
                     MESSAGE(LOG_DUMP, "Need Optimization\n");
-                    fpu_purgecache(dyn, ninst, x1, x2, x3); // maybe only x87, not SSE?
+                    fpu_purgecache(dyn, ninst, 0, x1, x2, x3); // maybe only x87, not SSE?
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, rex, 0, 0);
                     if(ed!=x1) {
                         MOVx_REG(x1, ed);

src/dynarec/arm64/dynarec_arm64_db.c

@@ -15,11 +15,11 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
 #include "emu/x87emu_private.h"
+#include "dynarec_native.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_helper.h"
 #include "dynarec_arm64_functions.h"
 
@@ -53,10 +53,14 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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));
             TSTw_mask(xFlags, 0, 0);    //mask=1<<F_CF
-            FCSELD(v1, v2, v1, cEQ);    // F_CF==0
+            if(ST_IS_F(0)) {
+                FCSELS(v1, v2, v1, cEQ);
+            } else {
+                FCSELD(v1, v2, v1, cEQ);    // F_CF==0
+            }
             break;
         case 0xC8:
         case 0xC9:
@@ -68,10 +72,14 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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));
             TSTw_mask(xFlags, 0b011010, 0); //mask=1<<F_ZF
-            FCSELD(v1, v2, v1, cEQ);        // F_ZF==0
+            if(ST_IS_F(0)) {
+                FCSELS(v1, v2, v1, cEQ);
+            } else {
+                FCSELD(v1, v2, v1, cEQ);        // F_ZF==0
+            }
             break;
         case 0xD0:
         case 0xD1:
@@ -83,11 +91,15 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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));
             MOV32w(x1, (1<<F_CF)|(1<<F_ZF));
             TSTw_REG(xFlags, x1);
-            FCSELD(v1, v2, v1, cEQ);   // F_CF==0 & F_ZF==0
+            if(ST_IS_F(0)) {
+                FCSELS(v1, v2, v1, cEQ);
+            } else {
+                FCSELD(v1, v2, v1, cEQ);   // F_CF==0 & F_ZF==0
+            }
             break;
         case 0xD8:
         case 0xD9:
@@ -99,10 +111,14 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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));
             TSTw_mask(xFlags, 0b011110, 0); //mask=1<<F_PF
-            FCSELD(v1, v2, v1, cEQ);        // F_PF==0
+            if(ST_IS_F(0)) {
+                FCSELS(v1, v2, v1, cEQ);
+            } else {
+                FCSELD(v1, v2, v1, cEQ);        // F_PF==0
+            }
             break;
         case 0xE1:
             INST_NAME("FDISI8087_NOP"); // so.. NOP?
@@ -118,7 +134,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0xE3:
             INST_NAME("FNINIT");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
-            x87_purgecache(dyn, ninst, x1, x2, x3);
+            x87_purgecache(dyn, ninst, 0, x1, x2, x3);
             CALL(reset_fpu, -1);
             break;
         case 0xE8:
@@ -131,9 +147,13 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FCMPD(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)) {
+                FCMPS(v1, v2);
+            } else {
+                FCMPD(v1, v2);
+            }
             FCOMI(x1, x2);
             break;
         case 0xF0:  
@@ -146,9 +166,13 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FCMPD(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)) {
+                FCMPS(v1, v2);
+            } else {
+                FCMPD(v1, v2);
+            }
             FCOMI(x1, x2);
             break;
 
@@ -164,7 +188,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("FILD ST0, Ed");
-                    v1 = x87_do_push(dyn, ninst);
+                    v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
                     s0 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, 0xfff<<2, 3, rex, 0, 0);
                     VLDR32_U12(s0, ed, fixedaddress);
@@ -173,7 +197,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     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(MODREG) {
                         ed = xRAX+(nextop&7)+(rex.b<<3);
                         wback = 0;
@@ -200,11 +224,11 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     STRw_U12(x5, wback, fixedaddress);
                     MARK3;
                     #endif
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                     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, x4); // x1 have the modified RPSCR reg
                     if(MODREG) {
                         ed = xRAX+(nextop&7)+(rex.b<<3);
@@ -236,7 +260,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     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, x4); // x1 have the modified RPSCR reg
                     if(MODREG) {
                         ed = xRAX+(nextop&7)+(rex.b<<3);
@@ -265,7 +289,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     MARK3;
                     #endif
                     x87_restoreround(dyn, ninst, u8);
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                     break;
                 case 5:
                     INST_NAME("FLD tbyte");
@@ -297,7 +321,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         MOVx_REG(x1, ed);
                     }
                     CALL(arm_fstp, -1);
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                     break;
                 default:
                     DEFAULT;

src/dynarec/arm64/dynarec_arm64_dc.c

@@ -15,11 +15,11 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
 #include "emu/x87emu_private.h"
+#include "dynarec_native.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_helper.h"
 #include "dynarec_arm64_functions.h"
 
@@ -46,9 +46,13 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FADDD(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)) {
+                FADDS(v1, v1, v2);
+            } else {
+                FADDD(v1, v1, v2);
+            }
             break;
         case 0xC8:
         case 0xC9:
@@ -59,9 +63,13 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FMULD(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)) {
+                FMULS(v1, v1, v2);
+            } else {
+                FMULD(v1, v1, v2);
+            }
             break;
         case 0xD0:
         case 0xD1:
@@ -72,9 +80,13 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FCMPD(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)) {
+                FCMPS(v1, v2);
+            } else {
+                FCMPD(v1, v2);
+            }
             FCOM(x1, x2, x3);
             break;
         case 0xD8:
@@ -86,11 +98,15 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FCMPD(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)) {
+                FCMPS(v1, v2);
+            } else {
+                FCMPD(v1, v2);
+            }
             FCOM(x1, x2, x3);
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
             break;
         case 0xE0:
         case 0xE1:
@@ -101,9 +117,13 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FSUBD(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)) {
+                FSUBS(v1, v2, v1);
+            } else {
+                FSUBD(v1, v2, v1);
+            }
             break;
         case 0xE8:
         case 0xE9:
@@ -114,9 +134,13 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FSUBD(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)) {
+                FSUBS(v1, v1, v2);
+            } else {
+                FSUBD(v1, v1, v2);
+            }
             break;
         case 0xF0:
         case 0xF1:
@@ -127,9 +151,13 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FDIVD(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)) {
+                FDIVS(v1, v2, v1);
+            } else {
+                FDIVD(v1, v2, v1);
+            }
             break;       
         case 0xF8:
         case 0xF9:
@@ -140,15 +168,19 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FDIVD(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)) {
+                FDIVS(v1, v1, v2);
+            } else {
+                FDIVD(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);
                     v2 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0xfff<<3, 3, rex, 0, 0);
                     VLDR64_U12(v2, wback, fixedaddress);
@@ -156,7 +188,7 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     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);
                     v2 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0xfff<<3, 3, rex, 0, 0);
                     VLDR64_U12(v2, wback, fixedaddress);
@@ -164,7 +196,7 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     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);
                     v2 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0xfff<<3, 3, rex, 0, 0);
                     VLDR64_U12(v2, wback, fixedaddress);
@@ -173,17 +205,17 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     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);
                     v2 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0xfff<<3, 3, rex, 0, 0);
                     VLDR64_U12(v2, wback, fixedaddress);
                     FCMPD(v1, v2);
                     FCOM(x1, x2, x3);
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                     break;
                 case 4:
                     INST_NAME("FSUB ST0, double[ED]");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                     v2 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0xfff<<3, 3, rex, 0, 0);
                     VLDR64_U12(v2, wback, fixedaddress);
@@ -191,7 +223,7 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     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);
                     v2 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0xfff<<3, 3, rex, 0, 0);
                     VLDR64_U12(v2, wback, fixedaddress);
@@ -199,7 +231,7 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     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);
                     v2 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0xfff<<3, 3, rex, 0, 0);
                     VLDR64_U12(v2, wback, fixedaddress);
@@ -207,7 +239,7 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     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);
                     v2 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0xfff<<3, 3, rex, 0, 0);
                     VLDR64_U12(v2, wback, fixedaddress);

src/dynarec/arm64/dynarec_arm64_dd.c

@@ -15,11 +15,11 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
 #include "emu/x87emu_private.h"
+#include "dynarec_native.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_helper.h"
 #include "dynarec_arm64_functions.h"
 
@@ -49,8 +49,8 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0xC7:
             INST_NAME("FFREE STx");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
-            x87_purgecache(dyn, ninst, x1, x2, x3);
-            MOV32w(x1, nextop-0xC0);
+            x87_purgecache(dyn, ninst, 0, x1, x2, x3);
+            MOV32w(x1, nextop&7);
             CALL(fpu_do_free, -1);
             break;
         case 0xD0:
@@ -62,13 +62,17 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FMOVD(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)) {
+                FMOVS(v2, v1);
+            } else {
+                FMOVD(v2, v1);
+            }
             break;
         case 0xD8:
             INST_NAME("FSTP ST0, ST0");
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
             break;
         case 0xD9:
         case 0xDA:
@@ -78,10 +82,9 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FMOVD(v2, v1);
-            x87_do_pop(dyn, ninst);
+            // copy the cache value for st0 to stx
+            x87_swapreg(dyn, ninst, x1, x2, 0, nextop&7);
+            x87_do_pop(dyn, ninst, x3);
             break;
 
         case 0xE0:
@@ -93,9 +96,13 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FCMPD(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)) {
+                FCMPS(v1, v2);
+            } else {
+                FCMPD(v1, v2);
+            }
             FCOM(x1, x2, x3);
             break;
         case 0xE8:
@@ -107,11 +114,15 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FCMPD(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)) {
+                FCMPS(v1, v2);
+            } else {
+                FCMPD(v1, v2);
+            }
             FCOM(x1, x2, x3);
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
             break;
 
         case 0xC8:
@@ -145,37 +156,37 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("FLD double");
-                    v1 = x87_do_push(dyn, ninst);
+                    v1 = x87_do_push(dyn, ninst, x3, NEON_CACHE_ST_D);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<3, 7, rex, 0, 0);
                     VLDR64_U12(v1, ed, fixedaddress);
                     break;
                 case 1:
                     INST_NAME("FISTTP i64, ST0");
-                    v1 = x87_do_push(dyn, ninst);
+                    v1 = x87_do_push(dyn, ninst, x3, NEON_CACHE_ST_D);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<3, 7, rex, 0, 0);
                     s0 = fpu_get_scratch(dyn);
                     FRINT64ZD(s0, v1);
                     FCVTZSxD(x2, s0);
                     STRx_U12(x2, ed, fixedaddress);
-                    x87_do_pop(dyn, ninst);
+                    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);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<3, 7, rex, 0, 0);
                     VSTR64_U12(v1, ed, fixedaddress);
                     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);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<3, 7, rex, 0, 0);
                     VSTR64_U12(v1, ed, fixedaddress);
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                     break;
                 case 4: 
                     INST_NAME("FRSTOR m108byte");
                     MESSAGE(LOG_DUMP, "Need Optimization\n");
-                    fpu_purgecache(dyn, ninst, x1, x2, x3);
+                    fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, rex, 0, 0);
                     if(ed!=x1) {MOVx_REG(x1, ed);}
                     CALL(arm_frstor, -1);
@@ -183,18 +194,27 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 case 6: 
                     INST_NAME("FSAVE m108byte");
                     MESSAGE(LOG_DUMP, "Need Optimization\n");
-                    fpu_purgecache(dyn, ninst, x1, x2, x3);
+                    fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, rex, 0, 0);
                     if(ed!=x1) {MOVx_REG(x1, ed);}
                     CALL(arm_fsave, -1);
                     break;
                 case 7:
                     INST_NAME("FNSTSW m2byte");
-                    fpu_purgecache(dyn, ninst, x1, x2, x3);
+                    fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x4, &fixedaddress, 0xfff<<1, 1, rex, 0, 0);
-                    LDRw_U12(x1, xEmu, offsetof(x64emu_t, top));
+                    LDRw_U12(x2, xEmu, offsetof(x64emu_t, top));
                     LDRH_U12(x3, xEmu, offsetof(x64emu_t, sw));
-                    BFIw(x3, x1, 11, 3); // inject TOP at bit 11 (3 bits)
+                    if(dyn->n.x87stack) {
+                        // update top
+                        if(dyn->n.x87stack>0) {
+                            SUBw_U12(x2, x2, dyn->n.x87stack);
+                        } else {
+                            ADDw_U12(x2, x2, -dyn->n.x87stack);
+                        }
+                        ANDw_mask(x2, x2, 0, 2);
+                    }
+                    BFIw(x3, x2, 11, 3); // inject TOP at bit 11 (3 bits)
                     STRH_U12(x3, ed, fixedaddress);   // store whole sw flags
                     break;
                 default:

src/dynarec/arm64/dynarec_arm64_df.c

@@ -15,11 +15,11 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
 #include "emu/x87emu_private.h"
+#include "dynarec_native.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_helper.h"
 #include "dynarec_arm64_functions.h"
 
@@ -51,7 +51,7 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0xC7:
             INST_NAME("FFREEP STx");
             // not handling Tag...
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
             break;
 
         case 0xE0:
@@ -71,11 +71,16 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FCMPD(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))
+            {
+                FCMPS(v1, v2);
+            } else {
+                FCMPD(v1, v2);
+            }
             FCOMI(x1, x2);
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
             break;
         case 0xF0:
         case 0xF1:
@@ -87,11 +92,16 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         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);
-            FCMPD(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))
+            {
+                FCMPS(v1, v2);
+            } else {
+                FCMPD(v1, v2);
+            }
             FCOMI(x1, x2);
-            x87_do_pop(dyn, ninst);
+            x87_do_pop(dyn, ninst, x3);
             break;
 
         case 0xC8:
@@ -140,19 +150,23 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("FILD ST0, Ew");
-                    v1 = x87_do_push(dyn, ninst);
+                    v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0xfff<<1, 1, rex, 0, 0);
                     LDRSHw_U12(x1, wback, fixedaddress);
-                    SCVTFDw(v1, x1);
+                    if(ST_IS_F(0)) {
+                        SCVTFSw(v1, x1);
+                    } else {
+                        SCVTFDw(v1, x1);
+                    }
                     break;
                 case 1:
                     INST_NAME("FISTTP Ew, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_F);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0xfff<<1, 1, rex, 0, 0);
                     ed = x1;
                     s0 = fpu_get_scratch(dyn);
                     #if 0
-                    // this version needs ARM v8.5, //TODO: add detection of this extensio to use it
+                    // this version needs ARM v8.5, //TODO: add detection of this extension to use it
                     FRINT32ZD(s0, v1);
                     // no saturation instruction on Arm, so using NEON
                     VFCVTZSd(s0, s0);
@@ -163,8 +177,12 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     MRS_fpsr(x5);
                     BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
                     MSR_fpsr(x5);
-                    VFCVTZSd(s0, v1);
-                    SQXTN_S_D(s0, s0);
+                    if(ST_IS_F(0)) {
+                        VFCVTZSs(s0, v1);
+                    } else {
+                        VFCVTZSd(s0, v1);
+                        SQXTN_S_D(s0, s0);
+                    }
                     SQXTN_H_S(s0, s0);
                     VSTR16_U12(s0, wback, fixedaddress);
                     MRS_fpsr(x5);   // get back FPSR to check the IOC bit
@@ -173,11 +191,11 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     STRH_U12(x5, wback, fixedaddress);
                     MARK3;
                     #endif
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                     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, x4);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0xfff<<1, 1, rex, 0, 0);
                     ed = x1;
@@ -193,9 +211,14 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     MRS_fpsr(x5);
                     BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
                     MSR_fpsr(x5);
-                    FRINTXD(s0, v1);
-                    VFCVTZSd(s0, s0);
-                    SQXTN_S_D(s0, s0);
+                    if(ST_IS_F(0)) {
+                        FRINTXS(s0, v1);
+                        VFCVTZSs(s0, s0);
+                    } else {
+                        FRINTXD(s0, v1);
+                        VFCVTZSd(s0, s0);
+                        SQXTN_S_D(s0, s0);
+                    }
                     SQXTN_H_S(s0, s0);
                     VSTR16_U12(s0, wback, fixedaddress);
                     MRS_fpsr(x5);   // get back FPSR to check the IOC bit
@@ -208,7 +231,7 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     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, x4);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0xfff<<1, 1, rex, 0, 0);
                     ed = x1;
@@ -224,9 +247,14 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     MRS_fpsr(x5);
                     BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
                     MSR_fpsr(x5);
-                    FRINTXD(s0, v1);
-                    VFCVTZSd(s0, s0);
-                    SQXTN_S_D(s0, s0);
+                    if(ST_IS_F(0)) {
+                        FRINTXS(s0, v1);
+                        VFCVTZSs(s0, s0);
+                    } else {
+                        FRINTXD(s0, v1);
+                        VFCVTZSd(s0, s0);
+                        SQXTN_S_D(s0, s0);
+                    }
                     SQXTN_H_S(s0, s0);
                     VSTR16_U12(s0, wback, fixedaddress);
                     MRS_fpsr(x5);   // get back FPSR to check the IOC bit
@@ -235,7 +263,7 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     STRH_U12(x5, wback, fixedaddress);
                     MARK3;
                     #endif
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                     x87_restoreround(dyn, ninst, u8);
                     break;
                 case 4:
@@ -247,7 +275,7 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     break;
                 case 5:
                     INST_NAME("FILD ST0, i64");
-                    v1 = x87_do_push(dyn, ninst);
+                    v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0xfff<<3, 7, rex, 0, 0);
                     LDRx_U12(x1, wback, fixedaddress);
                     SCVTFDx(v1, x1);
@@ -258,11 +286,11 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0, 0, rex, 0, 0);
                     if(ed!=x1) {MOVx_REG(x1, ed);}
                     CALL(fpu_fbst, -1);
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                     break;
                 case 7:
                     INST_NAME("FISTP i64, 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, x4);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0xfff<<3, 7, rex, 0, 0);
                     ed = x1;
@@ -285,7 +313,7 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     MARK3;
                     #endif
                     x87_restoreround(dyn, ninst, u8);
-                    x87_do_pop(dyn, ninst);
+                    x87_do_pop(dyn, ninst, x3);
                     break;
                 default:
                     DEFAULT;

src/dynarec/arm64/dynarec_arm64_emit_logic.c

@@ -15,11 +15,11 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
+#include "dynarec_native.h"
 #include "../tools/bridge_private.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_functions.h"
 #include "dynarec_arm64_helper.h"
 

src/dynarec/arm64/dynarec_arm64_emit_math.c

@@ -15,11 +15,11 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
+#include "dynarec_native.h"
 #include "../tools/bridge_private.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_functions.h"
 #include "dynarec_arm64_helper.h"
 

src/dynarec/arm64/dynarec_arm64_emit_shift.c

@@ -15,11 +15,11 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
+#include "dynarec_native.h"
 #include "../tools/bridge_private.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_functions.h"
 #include "dynarec_arm64_helper.h"
 

src/dynarec/arm64/dynarec_arm64_emit_tests.c

@@ -15,11 +15,11 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
+#include "dynarec_native.h"
 #include "../tools/bridge_private.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_functions.h"
 #include "dynarec_arm64_helper.h"
 

src/dynarec/arm64/dynarec_arm64_f0.c

@@ -15,10 +15,10 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
+#include "dynarec_native.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_helper.h"
 #include "dynarec_arm64_functions.h"
 
@@ -124,6 +124,50 @@ uintptr_t dynarec64_F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             nextop = F8;
             switch(nextop) {
 
+                case 0xB0:
+                    INST_NAME("LOCK CMPXCHG Eb, Gb");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    nextop = F8;
+                    GETGB(x1);
+                    UBFXx(x6, xRAX, 0, 8);
+                    DMB_ISH();
+                    if(MODREG) {
+                        if(rex.rex) {
+                            wback = xRAX+(nextop&7)+(rex.b<<3);
+                            wb2 = 0;
+                        } else { 
+                            wback = (nextop&7);
+                            wb2 = (wback>>2)*8;
+                            wback = xRAX+(wback&3);
+                        } 
+                        UBFXx(x2, wback, wb2, 8);
+                        wb1 = 0;
+                        ed = x2;
+                        UFLAG_IF {emit_cmp8(dyn, ninst, x6, ed, x3, x4, x5);}
+                        CMPSxw_REG(x6, x2);
+                        B_MARK2(cNE);
+                        BFIx(wback, x2, wb2, 8);
+                        MOVxw_REG(ed, gd);
+                        MARK2;
+                        BFIx(xRAX, x2, 0, 8);
+                        B_NEXT_nocond;
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0, 0, rex, 0, 0);
+                        // Aligned version
+                        MARKLOCK;
+                        LDAXRB(x2, wback);
+                        CMPSxw_REG(x6, x2);
+                        B_MARK(cNE);
+                        // EAX == Ed
+                        STLXRB(x4, gd, wback);
+                        CBNZx_MARKLOCK(x4);
+                        // done
+                        MARK;
+                        UFLAG_IF {emit_cmp32(dyn, ninst, rex, x6, x2, x3, x4, x5);}
+                        BFIx(xRAX, x2, 0, 8);    // upper par of RAX will be erase on 32bits, no mater what
+                    }
+                    DMB_ISH();
+                    break;
                 case 0xB1:
                     INST_NAME("LOCK CMPXCHG Ed, Gd");
                     SETFLAGS(X_ALL, SF_SET_PENDING);
@@ -266,6 +310,28 @@ uintptr_t dynarec64_F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0x66:
             opcode = F8;
             switch(opcode) {
+                case 0x09:
+                    INST_NAME("LOCK OR Ew, Gw");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    nextop = F8;
+                    GETGW(x5);
+                    DMB_ISH();
+                    if(MODREG) {
+                        ed = xRAX+(nextop&7)+(rex.b<<3);
+                        UXTHw(x6, ed);
+                        emit_or16(dyn, ninst, x6, x5, x3, x4);
+                        BFIx(ed, x6, 0, 16);
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, rex, 0, 0);
+                        MARKLOCK;
+                        LDAXRH(x1, wback);
+                        emit_or16(dyn, ninst, x1, x5, x3, x4);
+                        STLXRH(x3, x1, wback);
+                        CBNZx_MARKLOCK(x3);
+                    }
+                    DMB_ISH();
+                    break;
+
                 case 0x81:
                 case 0x83:
                     nextop = F8;
@@ -653,20 +719,20 @@ uintptr_t dynarec64_F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     } else {
                         addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, rex, 0, (opcode==0x81)?4:1);
                         if(opcode==0x81) i64 = F32S; else i64 = F8S;
-                        MOV64xw(x5, i64);
-                        TSTx_mask(wback, 1, 0, 0);    // mask=1
+                        TSTx_mask(wback, 1, 0, 1+rex.w);    // mask=3 or 7
                         B_MARK(cNE);
                         MARKLOCK;
                         LDAXRxw(x1, wback);
-                        emit_add32(dyn, ninst, rex, x1, x5, x3, x4);
+                        emit_add32c(dyn, ninst, rex, x1, i64, x3, x4, x5);
                         STLXRxw(x3, x1, wback);
                         CBNZx_MARKLOCK(x3);
+                        DMB_ISH();
                         B_NEXT_nocond;
                         MARK;   // unaligned! also, not enough 
                         LDRxw_U12(x1, wback, 0);
                         LDAXRB(x4, wback);
                         BFIxw(x1, x4, 0, 8); // re-inject
-                        emit_add32(dyn, ninst, rex, x1, x5, x3, x4);
+                        emit_add32c(dyn, ninst, rex, x1, i64, x3, x4, x5);
                         STLXRB(x3, x1, wback);
                         CBNZx_MARK(x3);
                         STRxw_U12(x1, wback, 0);    // put the whole value
@@ -761,20 +827,20 @@ uintptr_t dynarec64_F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     } else {
                         addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, 0, 0, rex, 0, (opcode==0x81)?4:1);
                         if(opcode==0x81) i64 = F32S; else i64 = F8S;
-                        MOV64xw(x5, i64);
-                        TSTx_mask(wback, 1, 0, 0);    // mask=1
+                        TSTx_mask(wback, 1, 0, 1+rex.w);    // mask=3 or 7
                         B_MARK(cNE);
                         MARKLOCK;
                         LDAXRxw(x1, wback);
-                        emit_sub32(dyn, ninst, rex, x1, x5, x3, x4);
+                        emit_sub32c(dyn, ninst, rex, x1, i64, x3, x4, x5);
                         STLXRxw(x3, x1, wback);
                         CBNZx_MARKLOCK(x3);
+                        DMB_ISH();
                         B_NEXT_nocond;
                         MARK;   // unaligned! also, not enough 
                         LDRxw_U12(x1, wback, 0);
                         LDAXRB(x4, wback);
                         BFIxw(x1, x4, 0, 8); // re-inject
-                        emit_sub32(dyn, ninst, rex, x1, x5, x3, x4);
+                        emit_sub32c(dyn, ninst, rex, x1, i64, x3, x4, x5);
                         STLXRB(x3, x1, wback);
                         CBNZx_MARK(x3);
                         STRxw_U12(x1, wback, 0);    // put the whole value
@@ -851,7 +917,7 @@ uintptr_t dynarec64_F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             {
                 case 0: // INC Ed
                     INST_NAME("LOCK INC Ed");
-                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
+                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
                     DMB_ISH();
                     if(MODREG) {
                         ed = xRAX+(nextop&7)+(rex.b<<3);
@@ -879,7 +945,7 @@ uintptr_t dynarec64_F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     break;
                 case 1: //DEC Ed
                     INST_NAME("LOCK DEC Ed");
-                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET);
+                    SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
                     DMB_ISH();
                     if(MODREG) {
                         ed = xRAX+(nextop&7)+(rex.b<<3);

src/dynarec/arm64/dynarec_arm64_f20f.c

@@ -15,17 +15,17 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
+#include "dynarec_native.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_functions.h"
 #include "dynarec_arm64_helper.h"
 
 // Get Ex as a double, not a quad (warning, x2 get used)
-#define GETEX(a, D)                                                                                 \
+#define GETEX(a, w, D)                                                                              \
     if(MODREG) {                                                                                    \
-        a = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3));                                     \
+        a = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), w);                                  \
     } else {                                                                                        \
         a = fpu_get_scratch(dyn);                                                                   \
         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<3, 7, rex, 0, D);     \
@@ -34,12 +34,16 @@
 
 #define GETG        gd = ((nextop&0x38)>>3)+(rex.r<<3)
 
-#define GETGX(a)    gd = ((nextop&0x38)>>3)+(rex.r<<3); \
-                    a = sse_get_reg(dyn, ninst, x1, gd)
+#define GETGX(a, w) gd = ((nextop&0x38)>>3)+(rex.r<<3); \
+                    a = sse_get_reg(dyn, ninst, x1, gd, w)
 
 #define GETGX_empty(a)  gd = ((nextop&0x38)>>3)+(rex.r<<3); \
                         a = sse_get_reg_empty(dyn, ninst, x1, gd)
 
+#define GETGM(a)                        \
+    gd = ((nextop&0x38)>>3);            \
+    a = mmx_get_reg(dyn, ninst, x1, x2, x3, gd)
+
 uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
 {
     (void)ip; (void)need_epilog;
@@ -73,8 +77,8 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             GETG;
             if(MODREG) {
                 ed = (nextop&7)+ (rex.b<<3);
-                v0 = sse_get_reg(dyn, ninst, x1, gd);
-                d0 = sse_get_reg(dyn, ninst, x1, ed);
+                v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+                d0 = sse_get_reg(dyn, ninst, x1, ed, 0);
                 VMOVeD(v0, 0, d0, 0);
             } else {
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
@@ -86,10 +90,10 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             INST_NAME("MOVSD Ex, Gx");
             nextop = F8;
             GETG;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 0);
             if(MODREG) {
                 ed = (nextop&7)+ (rex.b<<3);
-                d0 = sse_get_reg(dyn, ninst, x1, ed);
+                d0 = sse_get_reg(dyn, ninst, x1, ed, 1);
                 VMOVeD(d0, 0, v0, 0);
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<3, 7, rex, 0, 0);
@@ -101,7 +105,7 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             nextop = F8;
             GETG;
             if(MODREG) {
-                d0 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3));
+                d0 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 VMOVeD(v0, 0, d0, 0);
             } else {
@@ -115,7 +119,7 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x2A:
             INST_NAME("CVTSI2SD Gx, Ed");
             nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
             GETED(0);
             d1 = fpu_get_scratch(dyn);
             if(rex.w) {
@@ -130,23 +134,23 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             INST_NAME("CVTTSD2SI Gd, Ex");
             nextop = F8;
             GETGD;
-            GETEX(q0, 0);
+            GETEX(q0, 0, 0);
             FCVTZSxwD(gd, q0);
             break;
         case 0x2D:
             INST_NAME("CVTSD2SI Gd, Ex");
             nextop = F8;
             GETGD;
-            GETEX(q0, 0);
+            GETEX(q0, 0, 0);
             #ifdef PRECISE_CVT
             LDRH_U12(x1, xEmu, offsetof(x64emu_t, mxcsr));
             UBFXx(x1, x1, 13, 2);   // extract round requested
             LSLx_REG(x1, x1, 3);
-            ADDx_U12(x1, x1, 8);    // add the actual add+jump opcodes
             // Construct a "switch case", with each case 2 instructions, so 8 bytes
-            BL(+4); // Branch with Link to next, so LR gets next PC address
+            ADR(xLR, GETMARK);
             ADDx_REG(xLR, xLR, x1);
-            B(xLR); // could use RET, but it's not really one
+            B(xLR);
+            MARK;
             FCVTNSxwD(gd, q0);  // 0: Nearest (even)
             B_NEXT_nocond;
             FCVTMSxwD(gd, q0);  // 1: Toward -inf
@@ -163,36 +167,70 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x51:
             INST_NAME("SQRTSD Gx, Ex");
             nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
             d1 = fpu_get_scratch(dyn);
-            GETEX(d0, 0);
+            GETEX(d0, 0, 0);
+            if(!box64_dynarec_fastnan) {
+                v1 = fpu_get_scratch(dyn);
+                FCMLTD_0(v1, d0);
+                SHL_64(v1, v1, 63);
+            }
             FSQRTD(d1, d0);
+            if(!box64_dynarec_fastnan) {
+                VORR(d1, d1, v1);
+            }
             VMOVeD(v0, 0, d1, 0);
             break;
 
         case 0x58:
             INST_NAME("ADDSD Gx, Ex");
             nextop = F8;
-            GETGX(v0);
-            d1 = fpu_get_scratch(dyn);
-            GETEX(d0, 0);
-            FADDD(d1, v0, d0);  // the high part of the vector is erased...
-            VMOVeD(v0, 0, d1, 0);
+            GETGX(d1, 1);
+            v1 = fpu_get_scratch(dyn);
+            GETEX(d0, 0, 0);
+            if(!box64_dynarec_fastnan) {
+                v0 = fpu_get_scratch(dyn);
+                q0 = fpu_get_scratch(dyn);
+                // check if any input value was NAN
+                FMAXD(v0, d0, d1);    // propagate NAN
+                FCMEQD(v0, v0, v0);    // 0 if NAN, 1 if not NAN
+            }
+            FADDD(v1, d1, d0);  // the high part of the vector is erased...
+            if(!box64_dynarec_fastnan) {
+                FCMEQD(q0, v1, v1);    // 0 => out is NAN
+                VBIC(q0, v0, q0);      // forget it in any input was a NAN already
+                SHL_64(q0, q0, 63);     // only keep the sign bit
+                VORR(v1, v1, q0);      // NAN -> -NAN
+            }
+            VMOVeD(d1, 0, v1, 0);
             break;
         case 0x59:
             INST_NAME("MULSD Gx, Ex");
             nextop = F8;
-            GETGX(v0);
-            d1 = fpu_get_scratch(dyn);
-            GETEX(d0, 0);
-            FMULD(d1, v0, d0);
-            VMOVeD(v0, 0, d1, 0);
+            GETGX(d1, 1);
+            v1 = fpu_get_scratch(dyn);
+            GETEX(d0, 0, 0);
+            if(!box64_dynarec_fastnan) {
+                v0 = fpu_get_scratch(dyn);
+                q0 = fpu_get_scratch(dyn);
+                // check if any input value was NAN
+                FMAXD(v0, d0, d1);    // propagate NAN
+                FCMEQD(v0, v0, v0);    // 0 if NAN, 1 if not NAN
+            }
+            FMULD(v1, d1, d0);
+            if(!box64_dynarec_fastnan) {
+                FCMEQD(q0, v1, v1);    // 0 => out is NAN
+                VBIC(q0, v0, q0);      // forget it in any input was a NAN already
+                SHL_64(q0, q0, 63);     // only keep the sign bit
+                VORR(v1, v1, q0);      // NAN -> -NAN
+            }
+            VMOVeD(d1, 0, v1, 0);
             break;
         case 0x5A:
             INST_NAME("CVTSD2SS Gx, Ex");
             nextop = F8;
-            GETGX(v0);
-            GETEX(d0, 0);
+            GETGX(v0, 1);
+            GETEX(d0, 0, 0);
             d1 = fpu_get_scratch(dyn);
             FCVT_S_D(d1, d0);
             VMOVeS(v0, 0, d1, 0);
@@ -201,18 +239,30 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x5C:
             INST_NAME("SUBSD Gx, Ex");
             nextop = F8;
-            GETGX(v0);
-            d1 = fpu_get_scratch(dyn);
-            GETEX(d0, 0);
-            FSUBD(d1, v0, d0);
-            VMOVeD(v0, 0, d1, 0);
+            GETGX(d1, 1);
+            v1 = fpu_get_scratch(dyn);
+            GETEX(d0, 0, 0);
+            if(!box64_dynarec_fastnan) {
+                v0 = fpu_get_scratch(dyn);
+                q0 = fpu_get_scratch(dyn);
+                // check if any input value was NAN
+                FMAXD(v0, d0, d1);    // propagate NAN
+                FCMEQD(v0, v0, v0);    // 0 if NAN, 1 if not NAN
+            }
+            FSUBD(v1, d1, d0);
+            if(!box64_dynarec_fastnan) {
+                FCMEQD(q0, v1, v1);    // 0 => out is NAN
+                VBIC(q0, v0, q0);      // forget it in any input was a NAN already
+                SHL_64(q0, q0, 63);     // only keep the sign bit
+                VORR(v1, v1, q0);      // NAN -> -NAN
+            }
+            VMOVeD(d1, 0, v1, 0);
             break;
         case 0x5D:
             INST_NAME("MINSD Gx, Ex");
             nextop = F8;
-            GETG;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
-            GETEX(v1, 0);
+            GETGX(v0, 1);
+            GETEX(v1, 0, 0);
             // MINSD: if any input is NaN, or Ex[0]<Gx[0], copy Ex[0] -> Gx[0]
             #if 0
             d0 = fpu_get_scratch(dyn);
@@ -227,18 +277,30 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x5E:
             INST_NAME("DIVSD Gx, Ex");
             nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
             d1 = fpu_get_scratch(dyn);
-            GETEX(d0, 0);
-            FDIVD(d1, v0, d0);
+            GETEX(v1, 0, 0);
+            if(!box64_dynarec_fastnan) {
+                d0 = fpu_get_scratch(dyn);
+                q0 = fpu_get_scratch(dyn);
+                // check if any input value was NAN
+                FMAXD(d0, v0, v1);      // propagate NAN
+                FCMEQD(d0, d0, d0);     // 0 if NAN, 1 if not NAN
+            }
+            FDIVD(d1, v0, v1);
+            if(!box64_dynarec_fastnan) {
+                FCMEQD(q0, d1, d1);     // 0 => out is NAN
+                VBIC(q0, d0, q0);       // forget it in any input was a NAN already
+                SHL_64(q0, q0, 63);     // only keep the sign bit
+                VORR(d1, d1, q0);       // NAN -> -NAN
+            }
             VMOVeD(v0, 0, d1, 0);
             break;
         case 0x5F:
             INST_NAME("MAXSD Gx, Ex");
             nextop = F8;
-            GETG;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
-            GETEX(v1, 0);
+            GETGX(v0, 1);
+            GETEX(v1, 0, 0);
             // MAXSD: if any input is NaN, or Ex[0]>Gx[0], copy Ex[0] -> Gx[0]
             #if 0
             d0 = fpu_get_scratch(dyn);
@@ -254,8 +316,8 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x70:
             INST_NAME("PSHUFLW Gx, Ex, Ib");
             nextop = F8;
-            GETEX(v1, 1);
-            GETGX(v0);
+            GETEX(v1, 0, 1);
+            GETGX(v0, 1);
 
             u8 = F8;
             // only low part need to be suffled. VTBL only handle 8bits value, so the 16bits suffles need to be changed in 8bits
@@ -277,9 +339,9 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x7C:
             INST_NAME("HADDPS Gx, Ex");
             nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
             if(MODREG) {
-                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3));
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<4, 15, rex, 0, 0);
                 v1 = fpu_get_scratch(dyn);
@@ -291,8 +353,8 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0xC2:
             INST_NAME("CMPSD Gx, Ex, Ib");
             nextop = F8;
-            GETGX(v0);
-            GETEX(v1, 1);
+            GETGX(v0, 1);
+            GETEX(v1, 0, 1);
             u8 = F8;
             FCMPD(v0, v1);
             switch(u8&7) {
@@ -311,8 +373,8 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0xD0:
             INST_NAME("ADDSUBPS Gx, Ex");
             nextop = F8;
-            GETGX(v0);
-            GETEX(v1, 0);
+            GETGX(v0, 1);
+            GETEX(v1, 0, 0);
             q0 = fpu_get_scratch(dyn);
             static float addsubps[4] = {-1.f, 1.f, -1.f, 1.f};
             MAYUSE(addsubps);
@@ -321,13 +383,21 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             VFMLAQS(v0, v1, q0);
             break;
 
+        case 0xD6:
+            INST_NAME("MOVDQ2Q Gm, Ex");
+            nextop = F8;
+            GETGM(v0);
+            GETEX(v1, 0, 0);
+            VMOV(v0, v1);
+            break;
+
         case 0xE6:
             INST_NAME("CVTPD2DQ Gx, Ex");
             nextop = F8;
-            GETEX(v1, 0);
+            GETEX(v1, 0, 0);
             GETGX_empty(v0);
             u8 = sse_setround(dyn, ninst, x1, x2, x3);
-            FRINTXD(v0, v1);
+            VFRINTIDQ(v0, v1);
             x87_restoreround(dyn, ninst, u8);
             VFCVTNSQD(v0, v0);  // convert double -> int64
             SQXTN_32(v0, v0);   // convert int64 -> int32 with saturation in lower part, RaZ high part
@@ -338,7 +408,7 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             nextop = F8;
             GETG;
             if(MODREG) {
-                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3));
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
                 VMOVQ(v0, v1);
             } else {

src/dynarec/arm64/dynarec_arm64_f30f.c

@@ -15,17 +15,17 @@
 #include "callback.h"
 #include "emu/x64run_private.h"
 #include "x64trace.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
+#include "dynarec_native.h"
 
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
 #include "dynarec_arm64_functions.h"
 #include "dynarec_arm64_helper.h"
 
 // Get Ex as a single, not a quad (warning, x2 get used)
-#define GETEX(a, D)                                                                                 \
+#define GETEX(a, w, D)                                                                              \
     if(MODREG) {                                                                                    \
-        a = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3));                                     \
+        a = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), w);                                  \
     } else {                                                                                        \
         a = fpu_get_scratch(dyn);                                                                   \
         addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<2, 3, rex, 0, D);     \
@@ -34,8 +34,8 @@
 
 #define GETG        gd = ((nextop&0x38)>>3)+(rex.r<<3)
 
-#define GETGX(a)    gd = ((nextop&0x38)>>3)+(rex.r<<3); \
-                    a = sse_get_reg(dyn, ninst, x1, gd)
+#define GETGX(a, w) gd = ((nextop&0x38)>>3)+(rex.r<<3); \
+                    a = sse_get_reg(dyn, ninst, x1, gd, w)
 
 #define GETGX_empty(a)  gd = ((nextop&0x38)>>3)+(rex.r<<3);         \
                         a = sse_get_reg_empty(dyn, ninst, x1, gd)
@@ -70,8 +70,8 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             nextop = F8;
             GETG;
             if(MODREG) {
-                v0 = sse_get_reg(dyn, ninst, x1, gd);
-                q0 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3));
+                v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+                q0 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3), 0);
                 VMOVeS(v0, 0, q0, 0);
             } else {
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
@@ -83,9 +83,9 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             INST_NAME("MOVSS Ex, Gx");
             nextop = F8;
             GETG;
-            v0 = sse_get_reg(dyn, ninst, x1, gd);
+            v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
             if(MODREG) {
-                q0 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3));
+                q0 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3), 0);
                 VMOVeS(q0, 0, v0, 0);
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<2, 3, rex, 0, 0);
@@ -96,7 +96,7 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             INST_NAME("MOVSLDUP Gx, Ex");
             nextop = F8;
             if(MODREG) {
-                q1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3));
+                q1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<4, 15, rex, 0, 0);
                 q1 = fpu_get_scratch(dyn);
@@ -110,7 +110,7 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             INST_NAME("MOVSHDUP Gx, Ex");
             nextop = F8;
             if(MODREG) {
-                q1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3));
+                q1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<4, 15, rex, 0, 0);
                 q1 = fpu_get_scratch(dyn);
@@ -129,7 +129,7 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x2A:
             INST_NAME("CVTSI2SS Gx, Ed");
             nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
             GETED(0);
             d1 = fpu_get_scratch(dyn);
             if(rex.w) {
@@ -144,23 +144,22 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             INST_NAME("CVTTSS2SI Gd, Ex");
             nextop = F8;
             GETGD;
-            GETEX(d0, 0);
+            GETEX(d0, 0, 0);
             FCVTZSxwS(gd, d0);
             break;
         case 0x2D:
             INST_NAME("CVTSS2SI Gd, Ex");
             nextop = F8;
             GETGD;
-            GETEX(q0, 0);
+            GETEX(q0, 0, 0);
             #ifdef PRECISE_CVT
             LDRH_U12(x1, xEmu, offsetof(x64emu_t, mxcsr));
             UBFXx(x1, x1, 13, 2);   // extract round requested
             LSLx_REG(x1, x1, 3);
-            ADDx_U12(x1, x1, 8);    // add the actual add+jump opcodes
             // Construct a "switch case", with each case 2 instructions, so 8 bytes
-            BL(+4); // Branch with Link to next, so LR gets next PC address
+            ADR(xLR, GETMARK);
             ADDx_REG(xLR, xLR, x1);
-            B(xLR); // could use RET, but it's not really one
+            B(xLR);
             FCVTNSxwS(gd, q0);  // 0: Nearest (even)
             B_NEXT_nocond;
             FCVTMSxwS(gd, q0);  // 1: Toward -inf
@@ -175,17 +174,17 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x51:
             INST_NAME("SQRTSS Gx, Ex");
             nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
             d1 = fpu_get_scratch(dyn);
-            GETEX(d0, 0);
+            GETEX(d0, 0, 0);
             FSQRTS(d1, d0);
             VMOVeS(v0, 0, d1, 0);
             break;
         case 0x52:
             INST_NAME("RSQRTSS Gx, Ex");
             nextop = F8;
-            GETEX(v1, 0);
-            GETGX_empty(v0);
+            GETGX(v0, 1);
+            GETEX(v1, 0, 0);
             d0 = fpu_get_scratch(dyn);
             d1 = fpu_get_scratch(dyn);
             // so here: F32: Imm8 = abcd efgh that gives => aBbbbbbc defgh000 00000000 00000000
@@ -201,8 +200,8 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x53:
             INST_NAME("RCPSS Gx, Ex");
             nextop = F8;
-            GETGX(v0);
-            GETEX(v1, 0);
+            GETGX(v0, 1);
+            GETEX(v1, 0, 0);
             d0 = fpu_get_scratch(dyn);
             FMOVS_8(d0, 0b01110000);    //1.0f
             FDIVS(d0, d0, v1);
@@ -212,26 +211,26 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x58:
             INST_NAME("ADDSS Gx, Ex");
             nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
             d1 = fpu_get_scratch(dyn);
-            GETEX(d0, 0);
+            GETEX(d0, 0, 0);
             FADDS(d1, v0, d0);  // the high part of the vector is erased...
             VMOVeS(v0, 0, d1, 0);
             break;
         case 0x59:
             INST_NAME("MULSS Gx, Ex");
             nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
             d1 = fpu_get_scratch(dyn);
-            GETEX(d0, 0);
+            GETEX(d0, 0, 0);
             FMULS(d1, v0, d0);
             VMOVeS(v0, 0, d1, 0);
             break;
         case 0x5A:
             INST_NAME("CVTSS2SD Gx, Ex");
             nextop = F8;
-            GETGX(v0);
-            GETEX(v1, 0);
+            GETGX(v0, 1);
+            GETEX(v1, 0, 0);
             d0 = fpu_get_scratch(dyn);
             FCVT_D_S(d0, v1);
             VMOVeD(v0, 0, d0, 0);
@@ -239,7 +238,7 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x5B:
             INST_NAME("CVTPS2DQ Gx, Ex");
             nextop = F8;
-            GETEX(d0, 0);
+            GETEX(d0, 0, 0);
             GETGX_empty(v0);
             VFCVTZSQS(v0, d0);
             break;
@@ -247,17 +246,17 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x5C:
             INST_NAME("SUBSS Gx, Ex");
             nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
             d1 = fpu_get_scratch(dyn);
-            GETEX(d0, 0);
+            GETEX(d0, 0, 0);
             FSUBS(d1, v0, d0);
             VMOVeS(v0, 0, d1, 0);
             break;
         case 0x5D:
             INST_NAME("MINSS Gx, Ex");
             nextop = F8;
-            GETGX(v0);
-            GETEX(v1, 0);
+            GETGX(v0, 1);
+            GETEX(v1, 0, 0);
             // MINSS: if any input is NaN, or Ex[0]<Gx[0], copy Ex[0] -> Gx[0]
             #if 0
             d0 = fpu_get_scratch(dyn);
@@ -272,17 +271,17 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x5E:
             INST_NAME("DIVSS Gx, Ex");
             nextop = F8;
-            GETGX(v0);
+            GETGX(v0, 1);
             d1 = fpu_get_scratch(dyn);
-            GETEX(d0, 0);
+            GETEX(d0, 0, 0);
             FDIVS(d1, v0, d0);
             VMOVeS(v0, 0, d1, 0);
             break;
         case 0x5F:
             INST_NAME("MAXSS Gx, Ex");
             nextop = F8;
-            GETGX(v0);
-            GETEX(v1, 0);
+            GETGX(v0, 1);
+            GETEX(v1, 0, 0);
             // MAXSS: if any input is NaN, or Ex[0]>Gx[0], copy Ex[0] -> Gx[0]
             #if 0
             d0 = fpu_get_scratch(dyn);
@@ -299,11 +298,12 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             INST_NAME("MOVDQU Gx,Ex");// no alignment constraint on NEON here, so same as MOVDQA
             nextop = F8;
             GETG;
-            v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
             if(MODREG) {
-                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3));
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);
+                GETGX_empty(v0);
                 VMOVQ(v0, v1);
             } else {
+                GETGX_empty(v0);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<4, 15, rex, 0, 0);
                 VLDR128_U12(v0, ed, fixedaddress);
             }
@@ -311,9 +311,8 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x70:
             INST_NAME("PSHUFHW Gx, Ex, Ib");
             nextop = F8;
-            GETEX(v1, 1);
-            GETGX(v0);
-
+            GETEX(v1, 0, 1);
+            GETGX(v0, 1);
             u8 = F8;
             // only high part need to be suffled. VTBL only handle 8bits value, so the 16bits suffles need to be changed in 8bits
             u64 = 0;
@@ -334,13 +333,12 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x7E:
             INST_NAME("MOVQ Gx, Ex");
             nextop = F8;
-            GETG;
             if(MODREG) {
-                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3));
-                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3), 0);
+                GETGX_empty(v0);
                 FMOVD(v0, v1);
             } else {
-                v0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                GETGX_empty(v0);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<3, 7, rex, 0, 0);
                 VLDR64_U12(v0, ed, fixedaddress);
             }
@@ -348,13 +346,11 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x7F:
             INST_NAME("MOVDQU Ex,Gx");
             nextop = F8;
-            GETG;
+            GETGX(v0, 0);
             if(MODREG) {
-                v0 = sse_get_reg(dyn, ninst, x1, gd);
                 v1 = sse_get_reg_empty(dyn, ninst, x1, (nextop&7) + (rex.b<<3));
                 VMOVQ(v1, v0);
             } else {
-                v0 = sse_get_reg(dyn, ninst, x1, gd);
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<4, 15, rex, 0, 0);
                 VSTR128_U12(v0, ed, fixedaddress);
             }
@@ -393,8 +389,8 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0xC2:
             INST_NAME("CMPSS Gx, Ex, Ib");
             nextop = F8;
-            GETGX(v0);
-            GETEX(v1, 1);
+            GETGX(v0, 1);
+            GETEX(v1, 0, 1);
             u8 = F8;
             FCMPS(v0, v1);
             switch(u8&7) {
@@ -415,7 +411,7 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             nextop = F8;
             GETGX_empty(v0);
             if(MODREG) {
-                v1 = mmx_get_reg(dyn, ninst, x1, (nextop&7));
+                v1 = mmx_get_reg(dyn, ninst, x1, x2, x3, (nextop&7));
                 VEORQ(v0, v0, v0);  // usefull?
                 VMOV(v0, v1);
             } else {
@@ -427,7 +423,14 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0xE6:
             INST_NAME("CVTDQ2PD Gx, Ex");
             nextop = F8;
-            GETEX(v1, 0);
+            // cannot use GETEX because we want 64bits not 32bits
+            if(MODREG) {
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);
+            } else {
+                v1 = fpu_get_scratch(dyn);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, 0xfff<<3, 7, rex, 0, 0);
+                VLDR64_U12(v1, ed, fixedaddress);
+            }
             GETGX_empty(v0);
             d0 = fpu_get_scratch(dyn);
             SXTL_32(v0, v1);

src/dynarec/arm64/dynarec_arm64_functions.c

@@ -0,0 +1,858 @@
+#define _GNU_SOURCE
+#include <stdio.h>
+#include <stdlib.h>
+#include <pthread.h>
+#include <errno.h>
+#include <string.h>
+#include <math.h>
+#include <signal.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "tools/bridge_private.h"
+#include "x64run.h"
+#include "x64emu.h"
+#include "box64stack.h"
+#include "callback.h"
+#include "emu/x64run_private.h"
+#include "emu/x87emu_private.h"
+#include "x64trace.h"
+#include "signals.h"
+#include "dynarec_native.h"
+#include "dynarec_arm64_private.h"
+#include "dynarec_arm64_functions.h"
+#include "custommem.h"
+#include "bridge.h"
+
+void arm_fstp(x64emu_t* emu, void* p)
+{
+    if(ST0.q!=STld(0).uref)
+        D2LD(&ST0.d, p);
+    else
+        memcpy(p, &STld(0).ld, 10);
+}
+
+void arm_print_armreg(x64emu_t* emu, uintptr_t reg, uintptr_t n)
+{
+    (void)emu;
+    dynarec_log(LOG_DEBUG, "R%lu=0x%lx (%lu)\n", n, reg, reg);
+}
+
+void arm_f2xm1(x64emu_t* emu)
+{
+    ST0.d = exp2(ST0.d) - 1.0;
+}
+void arm_fyl2x(x64emu_t* emu)
+{
+    ST(1).d = log2(ST0.d)*ST(1).d;
+}
+void arm_ftan(x64emu_t* emu)
+{
+    ST0.d = tan(ST0.d);
+    emu->sw.f.F87_C2 = 0;
+}
+void arm_fpatan(x64emu_t* emu)
+{
+    ST1.d = atan2(ST1.d, ST0.d);
+}
+void arm_fxtract(x64emu_t* emu)
+{
+    int32_t tmp32s = (ST1.q&0x7ff0000000000000LL)>>52;
+    tmp32s -= 1023;
+    ST1.d /= exp2(tmp32s);
+    ST0.d = tmp32s;
+}
+void arm_fprem(x64emu_t* emu)
+{
+    int32_t tmp32s = ST0.d / ST1.d;
+    ST0.d -= ST1.d * tmp32s;
+    emu->sw.f.F87_C2 = 0;
+    emu->sw.f.F87_C0 = (tmp32s&1);
+    emu->sw.f.F87_C3 = ((tmp32s>>1)&1);
+    emu->sw.f.F87_C1 = ((tmp32s>>2)&1);
+}
+void arm_fyl2xp1(x64emu_t* emu)
+{
+    ST(1).d = log2(ST0.d + 1.0)*ST(1).d;
+}
+void arm_fsincos(x64emu_t* emu)
+{
+    sincos(ST1.d, &ST1.d, &ST0.d);
+    emu->sw.f.F87_C2 = 0;
+}
+void arm_frndint(x64emu_t* emu)
+{
+    ST0.d = fpu_round(emu, ST0.d);
+}
+void arm_fscale(x64emu_t* emu)
+{
+    if(ST0.d!=0.0)
+        ST0.d *= exp2(trunc(ST1.d));
+}
+void arm_fsin(x64emu_t* emu)
+{
+    ST0.d = sin(ST0.d);
+    emu->sw.f.F87_C2 = 0;
+}
+void arm_fcos(x64emu_t* emu)
+{
+    ST0.d = cos(ST0.d);
+    emu->sw.f.F87_C2 = 0;
+}
+
+void arm_fbld(x64emu_t* emu, uint8_t* ed)
+{
+    fpu_fbld(emu, ed);
+}
+
+void arm_fild64(x64emu_t* emu, int64_t* ed)
+{
+    int64_t tmp;
+    memcpy(&tmp, ed, sizeof(tmp));
+    ST0.d = tmp;
+    STll(0).sq = tmp;
+    STll(0).sref = ST0.sq;
+}
+
+void arm_fbstp(x64emu_t* emu, uint8_t* ed)
+{
+    fpu_fbst(emu, ed);
+}
+
+void arm_fistp64(x64emu_t* emu, int64_t* ed)
+{
+    // used of memcpy to avoid aligments issues
+    if(STll(0).sref==ST(0).sq) {
+        memcpy(ed, &STll(0).sq, sizeof(int64_t));
+    } else {
+        int64_t tmp;
+        if(isgreater(ST0.d, (double)(int64_t)0x7fffffffffffffffLL) || isless(ST0.d, (double)(int64_t)0x8000000000000000LL) || !isfinite(ST0.d))
+            tmp = 0x8000000000000000LL;
+        else
+            tmp = fpu_round(emu, ST0.d);
+        memcpy(ed, &tmp, sizeof(tmp));
+    }
+}
+
+void arm_fistt64(x64emu_t* emu, int64_t* ed)
+{
+    // used of memcpy to avoid aligments issues
+    int64_t tmp = ST0.d;
+    memcpy(ed, &tmp, sizeof(tmp));
+}
+
+void arm_fld(x64emu_t* emu, uint8_t* ed)
+{
+    memcpy(&STld(0).ld, ed, 10);
+    LD2D(&STld(0), &ST(0).d);
+    STld(0).uref = ST0.q;
+}
+
+void arm_ud(x64emu_t* emu)
+{
+    emit_signal(emu, SIGILL, (void*)R_RIP, 0);
+}
+
+void arm_fsave(x64emu_t* emu, uint8_t* ed)
+{
+    fpu_savenv(emu, (char*)ed, 0);
+
+    uint8_t* p = ed;
+    p += 28;
+    for (int i=0; i<8; ++i) {
+        LD2D(p, &ST(i).d);
+        p+=10;
+    }
+}
+void arm_frstor(x64emu_t* emu, uint8_t* ed)
+{
+    fpu_loadenv(emu, (char*)ed, 0);
+
+    uint8_t* p = ed;
+    p += 28;
+    for (int i=0; i<8; ++i) {
+        D2LD(&ST(i).d, p);
+        p+=10;
+    }
+
+}
+
+void arm_fprem1(x64emu_t* emu)
+{
+    // simplified version
+    int32_t tmp32s = round(ST0.d / ST1.d);
+    ST0.d -= ST1.d*tmp32s;
+    emu->sw.f.F87_C2 = 0;
+    emu->sw.f.F87_C0 = (tmp32s&1);
+    emu->sw.f.F87_C3 = ((tmp32s>>1)&1);
+    emu->sw.f.F87_C1 = ((tmp32s>>2)&1);
+}
+
+static uint8_t ff_mult(uint8_t a, uint8_t b)
+{
+	int retval = 0;
+
+	for(int i = 0; i < 8; i++) {
+		if((b & 1) == 1) 
+			retval ^= a;
+		
+		if((a & 0x80)) {
+			a <<= 1;
+			a  ^= 0x1b;
+		} else {
+			a <<= 1;
+		}
+		
+		b >>= 1;
+	}
+	
+	return retval;
+}
+
+void arm_aesimc(x64emu_t* emu, int xmm)
+{
+    sse_regs_t eax1 = emu->xmm[xmm];
+
+    for(int j=0; j<4; ++j) {
+        emu->xmm[xmm].ub[0+j*4] = ff_mult(0x0E, eax1.ub[0+j*4]) ^ ff_mult(0x0B, eax1.ub[1+j*4]) ^ ff_mult(0x0D, eax1.ub[2+j*4]) ^ ff_mult(0x09, eax1.ub[3+j*4]);
+        emu->xmm[xmm].ub[1+j*4] = ff_mult(0x09, eax1.ub[0+j*4]) ^ ff_mult(0x0E, eax1.ub[1+j*4]) ^ ff_mult(0x0B, eax1.ub[2+j*4]) ^ ff_mult(0x0D, eax1.ub[3+j*4]);
+        emu->xmm[xmm].ub[2+j*4] = ff_mult(0x0D, eax1.ub[0+j*4]) ^ ff_mult(0x09, eax1.ub[1+j*4]) ^ ff_mult(0x0E, eax1.ub[2+j*4]) ^ ff_mult(0x0B, eax1.ub[3+j*4]);
+        emu->xmm[xmm].ub[3+j*4] = ff_mult(0x0B, eax1.ub[0+j*4]) ^ ff_mult(0x0D, eax1.ub[1+j*4]) ^ ff_mult(0x09, eax1.ub[2+j*4]) ^ ff_mult(0x0E, eax1.ub[3+j*4]);
+    }
+}
+void arm_aesmc(x64emu_t* emu, int xmm)
+{
+    sse_regs_t eax1 = emu->xmm[xmm];
+
+    for(int j=0; j<4; ++j) {
+        emu->xmm[xmm].ub[0+j*4] = ff_mult(0x02, eax1.ub[0+j*4]) ^ ff_mult(0x03, eax1.ub[1+j*4]) ^               eax1.ub[2+j*4]  ^               eax1.ub[3+j*4] ;
+        emu->xmm[xmm].ub[1+j*4] =               eax1.ub[0+j*4]  ^ ff_mult(0x02, eax1.ub[1+j*4]) ^ ff_mult(0x03, eax1.ub[2+j*4]) ^               eax1.ub[3+j*4] ;
+        emu->xmm[xmm].ub[2+j*4] =               eax1.ub[0+j*4]  ^               eax1.ub[1+j*4]  ^ ff_mult(0x02, eax1.ub[2+j*4]) ^ ff_mult(0x03, eax1.ub[3+j*4]);
+        emu->xmm[xmm].ub[3+j*4] = ff_mult(0x03, eax1.ub[0+j*4]) ^               eax1.ub[1+j*4]  ^               eax1.ub[2+j*4]  ^ ff_mult(0x02, eax1.ub[3+j*4]);
+    }
+}
+void arm_aesdlast(x64emu_t* emu, int xmm)
+{
+                            //   A0 B1 C2 D3 E4 F5 G6 H7 I8 J9 Ka Lb Mc Nd Oe Pf
+                            //   A  N  K  H  E  B  O  L  I  F  C  P  M  J  G  D
+    const uint8_t invshiftrows[] = {0,13,10, 7, 4, 1,14,11, 8, 5, 2,15,12, 9, 6, 3};
+    const uint8_t invsubbytes[256] = {
+        0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38, 0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb,
+        0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87, 0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb,
+        0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d, 0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e,
+        0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2, 0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25,
+        0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16, 0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92,
+        0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda, 0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84,
+        0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a, 0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06,
+        0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02, 0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b,
+        0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea, 0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73,
+        0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85, 0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e,
+        0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89, 0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b,
+        0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20, 0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4,
+        0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31, 0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f,
+        0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d, 0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef,
+        0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0, 0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61,
+        0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26, 0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d,
+    };
+
+    sse_regs_t eax1;
+    for(int i=0; i<16; ++i)
+        eax1.ub[i] = emu->xmm[xmm].ub[invshiftrows[i]];
+    //STATE ← InvSubBytes( STATE );
+    for(int i=0; i<16; ++i)
+        emu->xmm[xmm].ub[i] = invsubbytes[eax1.ub[i]];
+
+}
+void arm_aeselast(x64emu_t* emu, int xmm)
+{
+                            //   A0 B1 C2 D3 E4 F5 G6 H7 I8 J9 Ka Lb Mc Nd Oe Pf
+                            //   A  F  K  P  E  J  O  D  I  N  C  H  M  B  G  L
+    const uint8_t shiftrows[] = {0, 5,10,15, 4, 9,14, 3, 8,13, 2, 7,12, 1, 6,11};
+    const uint8_t subbytes[256] = {
+        0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5, 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76,
+        0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0, 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0,
+        0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc, 0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15,
+        0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a, 0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75,
+        0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0, 0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84,
+        0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b, 0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf,
+        0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85, 0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8,
+        0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5, 0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2,
+        0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17, 0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73,
+        0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88, 0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb,
+        0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c, 0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79,
+        0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9, 0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08,
+        0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6, 0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a,
+        0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e, 0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e,
+        0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94, 0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf,
+        0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68, 0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16,
+    };
+
+    sse_regs_t eax1;
+    for(int i=0; i<16; ++i)
+        eax1.ub[i] = emu->xmm[xmm].ub[shiftrows[i]];
+    //STATE ← SubBytes( STATE );
+    for(int i=0; i<16; ++i)
+        emu->xmm[xmm].ub[i] = subbytes[eax1.ub[i]];
+}
+void arm_aesd(x64emu_t* emu, int xmm)
+{
+    arm_aesdlast(emu, xmm);
+    arm_aesimc(emu, xmm);
+}
+void arm_aese(x64emu_t* emu, int xmm)
+{
+    arm_aeselast(emu, xmm);
+    arm_aesmc(emu, xmm);
+}
+
+
+#define XMM0    0
+#define XMM8    16
+#define X870    8
+#define EMM0    8
+#define SCRATCH0    24
+
+// Get a FPU scratch reg
+int fpu_get_scratch(dynarec_arm_t* dyn)
+{
+    return SCRATCH0 + dyn->n.fpu_scratch++;  // return an Sx
+}
+// Reset scratch regs counter
+void fpu_reset_scratch(dynarec_arm_t* dyn)
+{
+    dyn->n.fpu_scratch = 0;
+}
+// Get a x87 double reg
+int fpu_get_reg_x87(dynarec_arm_t* dyn, int t, int n)
+{
+    int i=X870;
+    while (dyn->n.fpuused[i]) ++i;
+    dyn->n.fpuused[i] = 1;
+    dyn->n.neoncache[i].n = n;
+    dyn->n.neoncache[i].t = t;
+    dyn->n.news |= (1<<i);
+    return i; // return a Dx
+}
+// Free a FPU double reg
+void fpu_free_reg(dynarec_arm_t* dyn, int reg)
+{
+    // TODO: check upper limit?
+    dyn->n.fpuused[reg] = 0;
+    if(dyn->n.neoncache[reg].t!=NEON_CACHE_ST_F && dyn->n.neoncache[reg].t!=NEON_CACHE_ST_D)
+        dyn->n.neoncache[reg].v = 0;
+}
+// Get an MMX double reg
+int fpu_get_reg_emm(dynarec_arm_t* dyn, int emm)
+{
+    dyn->n.fpuused[EMM0 + emm] = 1;
+    dyn->n.neoncache[EMM0 + emm].t = NEON_CACHE_MM;
+    dyn->n.neoncache[EMM0 + emm].n = emm;
+    return EMM0 + emm;
+}
+// Get an XMM quad reg
+int fpu_get_reg_xmm(dynarec_arm_t* dyn, int t, int xmm)
+{
+    int i;
+    if(xmm>7) {
+        i = XMM8 + xmm - 8;
+    } else {
+        i = XMM0 + xmm;
+    }
+    dyn->n.fpuused[i] = 1;
+    dyn->n.neoncache[i].t = t;
+    dyn->n.neoncache[i].n = xmm;
+    return i;
+}
+// Reset fpu regs counter
+void fpu_reset_reg(dynarec_arm_t* dyn)
+{
+    dyn->n.fpu_reg = 0;
+    for (int i=0; i<24; ++i) {
+        dyn->n.fpuused[i]=0;
+        dyn->n.neoncache[i].v = 0;
+    }
+
+}
+
+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(box64_dynarec_dump) dynarec_log(LOG_NONE, "neoncache_promote_double_combined, ninst=%d combined%c %d i=%d (stack:%d/%d)\n", ninst, (a == dyn->insts[ninst].n.combined2)?'2':'1', a ,i, dyn->insts[ninst].n.stack_push, -dyn->insts[ninst].n.stack_pop);
+        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(box64_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(box64_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(box64_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(box64_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(box64_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(box64_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].x64.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(box64_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(box64_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(box64_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];
+}
+
+int isCacheEmpty(dynarec_arm_t* dyn, int ninst) {
+    if(dyn->insts[ninst].n.stack_next) {
+        return 0;
+    }
+    for(int i=0; i<24; ++i)
+        if(dyn->insts[ninst].n.neoncache[i].v) {       // there is something at ninst for i
+            if(!(
+            (dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_F || dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_D)
+            && dyn->insts[ninst].n.neoncache[i].n<dyn->insts[ninst].n.stack_pop))
+                return 0;
+        }
+    return 1;
+
+}
+
+int fpuCacheNeedsTransform(dynarec_arm_t* dyn, int ninst) {
+    int i2 = dyn->insts[ninst].x64.jmp_insts;
+    if(i2<0)
+        return 1;
+    if((dyn->insts[i2].x64.barrier&BARRIER_FLOAT))
+        // if the barrier as already been apply, no transform needed
+        return ((dyn->insts[ninst].x64.barrier&BARRIER_FLOAT))?0:(isCacheEmpty(dyn, ninst)?0:1);
+    int ret = 0;
+    if(!i2) { // just purge
+        if(dyn->insts[ninst].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 x64 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->x87reg[i] = 0;
+        cache->ssecache[i*2].v = -1;
+        cache->ssecache[i*2+1].v = -1;
+    }
+    int x87reg = 0;
+    for(int i=0; i<24; ++i) {
+        if(cache->neoncache[i].v) {
+            cache->fpuused[i] = 1;
+            switch (cache->neoncache[i].t) {
+                case NEON_CACHE_MM:
+                    cache->mmxcache[cache->neoncache[i].n] = i;
+                    ++cache->mmxcount;
+                    ++cache->fpu_reg;
+                    break;
+                case NEON_CACHE_XMMR:
+                case NEON_CACHE_XMMW:
+                    cache->ssecache[cache->neoncache[i].n].reg = i;
+                    cache->ssecache[cache->neoncache[i].n].write = (cache->neoncache[i].t==NEON_CACHE_XMMW)?1:0;
+                    ++cache->fpu_reg;
+                    break;
+                case NEON_CACHE_ST_F:
+                case NEON_CACHE_ST_D:
+                    cache->x87cache[x87reg] = cache->neoncache[i].n;
+                    cache->x87reg[x87reg] = i;
+                    ++x87reg;
+                    ++cache->fpu_reg;
+                    break;
+                case NEON_CACHE_SCR:
+                    cache->fpuused[i] = 0;
+                    cache->neoncache[i].v = 0;
+                    break;
+            }
+        } else {
+            cache->fpuused[i] = 0;
+        }
+    }
+}
+
+#define F8      *(uint8_t*)(addr++)
+#define F32     *(uint32_t*)(addr+=4, addr-4)
+#define F32S64  (uint64_t)(int64_t)*(int32_t*)(addr+=4, addr-4)
+// Get if ED will have the correct parity. Not emiting anything. Parity is 2 for DWORD or 3 for QWORD
+int getedparity(dynarec_arm_t* dyn, int ninst, uintptr_t addr, uint8_t nextop, int parity, int delta)
+{
+    (void)dyn; (void)ninst;
+
+    uint32_t tested = (1<<parity)-1;
+    if((nextop&0xC0)==0xC0)
+        return 0;   // direct register, no parity...
+    if(!(nextop&0xC0)) {
+        if((nextop&7)==4) {
+            uint8_t sib = F8;
+            int sib_reg = (sib>>3)&7;
+            if((sib&0x7)==5) {
+                uint64_t tmp = F32S64;
+                if (sib_reg!=4) {
+                    // if XXXXXX+reg<<N then check parity of XXXXX and N should be enough
+                    return ((tmp&tested)==0 && (sib>>6)>=parity)?1:0;
+                } else {
+                    // just a constant...
+                    return (tmp&tested)?0:1;
+                }
+            } else {
+                if(sib_reg==4 && parity<3)
+                    return 0;   // simple [reg]
+                // don't try [reg1 + reg2<<N], unless reg1 is ESP
+                return ((sib&0x7)==4 && (sib>>6)>=parity)?1:0;
+            }
+        } else if((nextop&7)==5) {
+            uint64_t tmp = F32S64;
+            tmp+=addr+delta;
+            return (tmp&tested)?0:1;
+        } else {
+            return 0;
+        }
+    } else {
+        return 0; //Form [reg1 + reg2<<N + XXXXXX]
+    }
+}
+
+// Do the GETED, but don't emit anything...
+uintptr_t fakeed(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop) 
+{
+    (void)dyn; (void)addr; (void)ninst;
+
+    if((nextop&0xC0)==0xC0)
+        return addr;
+    if(!(nextop&0xC0)) {
+        if((nextop&7)==4) {
+            uint8_t sib = F8;
+            if((sib&0x7)==5) {
+                addr+=4;
+            }
+        } else if((nextop&7)==5) {
+            addr+=4;
+        }
+    } else {
+        if((nextop&7)==4) {
+            ++addr;
+        }
+        if(nextop&0x80) {
+            addr+=4;
+        } else {
+            ++addr;
+        }
+    }
+    return addr;
+}
+#undef F8
+#undef F32
+
+int isNativeCall(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t* calladdress, int* retn)
+{
+    (void)dyn;
+
+#define PK(a)       *(uint8_t*)(addr+a)
+#define PK32(a)     *(int32_t*)(addr+a)
+
+    if(!addr || !getProtection(addr))
+        return 0;
+    if(PK(0)==0xff && PK(1)==0x25) {            // "absolute" jump, maybe the GOT (well, RIP relative in fact)
+        uintptr_t a1 = addr + 6 + (PK32(2));    // need to add a check to see if the address is from the GOT !
+        addr = (uintptr_t)getAlternate(*(void**)a1);
+    }
+    if(!addr || !getProtection(addr))
+        return 0;
+    onebridge_t *b = (onebridge_t*)(addr);
+    if(b->CC==0xCC && b->S=='S' && b->C=='C' && b->w!=(wrapper_t)0 && b->f!=(uintptr_t)PltResolver) {
+        // found !
+        if(retn) *retn = (b->C3==0xC2)?b->N:0;
+        if(calladdress) *calladdress = addr+1;
+        return 1;
+    }
+    return 0;
+#undef PK32
+#undef PK
+}
+
+const char* getCacheName(int t, int n)
+{
+    static char buff[20];
+    switch(t) {
+        case NEON_CACHE_ST_D: sprintf(buff, "ST%d", n); break;
+        case NEON_CACHE_ST_F: sprintf(buff, "st%d", n); break;
+        case NEON_CACHE_MM: sprintf(buff, "MM%d", n); break;
+        case NEON_CACHE_XMMW: sprintf(buff, "XMM%d", n); break;
+        case NEON_CACHE_XMMR: sprintf(buff, "xmm%d", n); break;
+        case NEON_CACHE_SCR: sprintf(buff, "Scratch"); break;
+        case NEON_CACHE_NONE: buff[0]='\0'; break;
+    }
+    return buff;
+}
+
+// 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].x64.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;
+}

src/dynarec/arm64/dynarec_arm64_functions.h

@@ -43,16 +43,43 @@ int fpu_get_scratch(dynarec_arm_t* dyn);
 // Reset scratch regs counter
 void fpu_reset_scratch(dynarec_arm_t* dyn);
 // Get an x87 double reg
-int fpu_get_reg_x87(dynarec_arm_t* dyn);
+int fpu_get_reg_x87(dynarec_arm_t* dyn, int t, int n);
 // Get an MMX double reg
 int fpu_get_reg_emm(dynarec_arm_t* dyn, int emm);
 // Get an XMM quad reg
-int fpu_get_reg_xmm(dynarec_arm_t* dyn, int xmm);
+int fpu_get_reg_xmm(dynarec_arm_t* dyn, int t, int xmm);
 // Free a FPU/MMX/XMM reg
 void fpu_free_reg(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 fpuCacheNeedsTransform(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, int delta);
 // Do the GETED, but don't emit anything...
@@ -61,4 +88,6 @@ 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/arm64/dynarec_arm64_helper.h

@@ -244,7 +244,8 @@
                     ed = i;                     \
                 } else {                        \
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, 0, 0, rex, 0, D); \
-                    ADDx_REG(wback, wback, i);  \
+                    ADDx_REG(x3, wback, i);     \
+                    if(wback!=x3) wback = x3;   \
                     LDRB_U12(i, wback, fixedaddress);      \
                     wb1 = 1;                    \
                     ed = i;                     \
@@ -312,122 +313,138 @@
 // R0 will not be pushed/popd if ret is -2. Flags are not save/restored
 #define CALL_S(F, ret) call_c(dyn, ninst, F, x7, ret, 0, 0)
 
-#define MARK    dyn->insts[ninst].mark = dyn->arm_size
+#define MARK    dyn->insts[ninst].mark = dyn->native_size
 #define GETMARK dyn->insts[ninst].mark
-#define MARK2   dyn->insts[ninst].mark2 = dyn->arm_size
+#define MARK2   dyn->insts[ninst].mark2 = dyn->native_size
 #define GETMARK2 dyn->insts[ninst].mark2
-#define MARK3   dyn->insts[ninst].mark3 = dyn->arm_size
+#define MARK3   dyn->insts[ninst].mark3 = dyn->native_size
 #define GETMARK3 dyn->insts[ninst].mark3
-#define MARKF   dyn->insts[ninst].markf = dyn->arm_size
+#define MARKF   dyn->insts[ninst].markf = dyn->native_size
 #define GETMARKF dyn->insts[ninst].markf
-#define MARKSEG dyn->insts[ninst].markseg = dyn->arm_size
+#define MARKSEG dyn->insts[ninst].markseg = dyn->native_size
 #define GETMARKSEG dyn->insts[ninst].markseg
-#define MARKLOCK dyn->insts[ninst].marklock = dyn->arm_size
+#define MARKLOCK dyn->insts[ninst].marklock = dyn->native_size
 #define GETMARKLOCK dyn->insts[ninst].marklock
 
 // Branch to MARK if cond (use j64)
 #define B_MARK(cond)                \
-    j64 = GETMARK-(dyn->arm_size);  \
+    j64 = GETMARK-(dyn->native_size);  \
     Bcond(cond, j64)
 // Branch to MARK unconditionnal (use j64)
 #define B_MARK_nocond               \
-    j64 = GETMARK-(dyn->arm_size);  \
+    j64 = GETMARK-(dyn->native_size);  \
     B(j64)
 // Branch to MARK if reg is 0 (use j64)
 #define CBZxw_MARK(reg)             \
-    j64 = GETMARK-(dyn->arm_size);  \
+    j64 = GETMARK-(dyn->native_size);  \
     CBZxw(reg, j64)
 // Branch to MARK if reg is not 0 (use j64)
 #define CBNZx_MARK(reg)             \
-    j64 = GETMARK-(dyn->arm_size);  \
+    j64 = GETMARK-(dyn->native_size);  \
     CBNZx(reg, j64)
 // Branch to MARK if reg is not 0 (use j64)
 #define CBNZw_MARK(reg)             \
-    j64 = GETMARK-(dyn->arm_size);  \
+    j64 = GETMARK-(dyn->native_size);  \
     CBNZw(reg, j64)
+// Test bit N of A and branch to MARK if not set
+#define TBZ_MARK(A, N)              \
+    j64 = GETMARK-(dyn->native_size);  \
+    TBZ(A, N, j64)
+// Test bit N of A and branch to MARK if set
+#define TBNZ_MARK(A, N)             \
+    j64 = GETMARK-(dyn->native_size);  \
+    TBNZ(A, N, j64)
 // Branch to MARK2 if cond (use j64)
 #define B_MARK2(cond)               \
-    j64 = GETMARK2-(dyn->arm_size); \
+    j64 = GETMARK2-(dyn->native_size); \
     Bcond(cond, j64)
 // Branch to MARK2 unconditionnal (use j64)
 #define B_MARK2_nocond              \
-    j64 = GETMARK2-(dyn->arm_size); \
+    j64 = GETMARK2-(dyn->native_size); \
     B(j64)
 // Branch to MARK2 if reg is not 0 (use j64)
 #define CBNZx_MARK2(reg)            \
-    j64 = GETMARK2-(dyn->arm_size); \
+    j64 = GETMARK2-(dyn->native_size); \
     CBNZx(reg, j64)
 // Test bit N of A and branch to MARK2 if set
 #define TBNZ_MARK2(A, N)            \
-    j64 = GETMARK2-(dyn->arm_size); \
+    j64 = GETMARK2-(dyn->native_size); \
     TBNZ(A, N, j64)
 // Branch to MARK3 if cond (use j64)
 #define B_MARK3(cond)               \
-    j64 = GETMARK3-(dyn->arm_size); \
+    j64 = GETMARK3-(dyn->native_size); \
     Bcond(cond, j64)
+// Test bit N of A and branch to MARK3 if not set
+#define TBZ_MARK2(A, N)            \
+    j64 = GETMARK2-(dyn->native_size); \
+    TBZ(A, N, j64)
 // Branch to MARK3 unconditionnal (use j64)
 #define B_MARK3_nocond              \
-    j64 = GETMARK3-(dyn->arm_size); \
+    j64 = GETMARK3-(dyn->native_size); \
     B(j64)
 // Branch to MARK3 if reg is not 0 (use j64)
 #define CBNZx_MARK3(reg)            \
-    j64 = GETMARK3-(dyn->arm_size); \
+    j64 = GETMARK3-(dyn->native_size); \
     CBNZx(reg, j64)
-// Test bit N of A and branch to MARK3 if set
-#define TBNZ_MARK3(A, N)            \
-    j64 = GETMARK3-(dyn->arm_size); \
-    TBNZ(A, N, j64)
+// Branch to MARK3 if reg is 0 (use j64)
+#define CBZx_MARK3(reg)             \
+    j64 = GETMARK3-(dyn->native_size); \
+    CBZx(reg, j64)
 // Test bit N of A and branch to MARK3 if not set
 #define TBZ_MARK3(A, N)             \
-    j64 = GETMARK3-(dyn->arm_size); \
+    j64 = GETMARK3-(dyn->native_size); \
     TBZ(A, N, j64)
+// Test bit N of A and branch to MARK3 if set
+#define TBNZ_MARK3(A, N)            \
+    j64 = GETMARK3-(dyn->native_size); \
+    TBNZ(A, N, j64)
 // Branch to next instruction if cond (use j64)
 #define B_NEXT(cond)     \
-    j64 = (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->arm_size)):0; \
+    j64 = (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->native_size)):0; \
     Bcond(cond, j64)
 // Branch to next instruction unconditionnal (use j64)
 #define B_NEXT_nocond                                               \
-    j64 = (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->arm_size)):0;\
+    j64 = (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->native_size)):0;\
     B(j64)
 // Branch to next instruction if reg is 0 (use j64)
 #define CBZw_NEXT(reg)    \
-    j64 =  (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->arm_size)):0; \
+    j64 =  (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->native_size)):0; \
     CBZw(reg, j64)
 // Branch to next instruction if reg is 0 (use j64)
 #define CBZx_NEXT(reg)    \
-    j64 =  (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->arm_size)):0; \
+    j64 =  (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->native_size)):0; \
     CBZx(reg, j64)
 // Branch to next instruction if reg is not 0 (use j64)
 #define CBNZx_NEXT(reg)   \
-    j64 =  (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->arm_size)):0; \
+    j64 =  (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->native_size)):0; \
     CBNZx(reg, j64)
 // Test bit N of A and branch to next instruction if not set
 #define TBZ_NEXT(A, N)              \
-    j64 = (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->arm_size)):0; \
+    j64 = (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->native_size)):0; \
     TBZ(A, N, j64)
 // Test bit N of A and branch to next instruction if set
 #define TBNZ_NEXT(A, N)             \
-    j64 = (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->arm_size)):0; \
+    j64 = (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->native_size)):0; \
     TBNZ(A, N, j64)
 // Branch to MARKSEG if cond (use j64)
 #define B_MARKSEG(cond)    \
-    j64 = GETMARKSEG-(dyn->arm_size);   \
+    j64 = GETMARKSEG-(dyn->native_size);   \
     Bcond(cond, j64)
 // Branch to MARKSEG if reg is 0 (use j64)
 #define CBZw_MARKSEG(reg)    \
-    j64 = GETMARKSEG-(dyn->arm_size);   \
+    j64 = GETMARKSEG-(dyn->native_size);   \
     CBZw(reg, j64)
 // Branch to MARKSEG if reg is not 0 (use j64)
 #define CBNZw_MARKSEG(reg)              \
-    j64 = GETMARKSEG-(dyn->arm_size);   \
+    j64 = GETMARKSEG-(dyn->native_size);   \
     CBNZw(reg, j64)
 // Branch to MARKLOCK if cond (use j64)
 #define B_MARKLOCK(cond)    \
-    j64 = GETMARKLOCK-(dyn->arm_size);   \
+    j64 = GETMARKLOCK-(dyn->native_size);   \
     Bcond(cond, j64)
 // Branch to MARKLOCK if reg is not 0 (use j64)
 #define CBNZx_MARKLOCK(reg)             \
-    j64 = GETMARKLOCK-(dyn->arm_size);  \
+    j64 = GETMARKLOCK-(dyn->native_size);  \
     CBNZx(reg, j64)
 
 #define IFX(A)  if((dyn->insts[ninst].x64.need_flags&(A)))
@@ -553,35 +570,47 @@
     LDP_REGS(R12, R13);     \
     LDP_REGS(R14, R15)
 
-#define SET_DFNONE(S)    if(!dyn->dfnone) {MOVZw(S, d_none); STRw_U12(S, xEmu, offsetof(x64emu_t, df)); dyn->dfnone=1;}
-#define SET_DF(S, N)     if((N)!=d_none) {MOVZw(S, (N)); STRw_U12(S, xEmu, offsetof(x64emu_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) {MOVZw(S, d_none); STRw_U12(S, xEmu, offsetof(x64emu_t, df)); dyn->f.dfnone=1;}
+#define SET_DF(S, N)     if((N)!=d_none) {MOVZw(S, (N)); STRw_U12(S, xEmu, offsetof(x64emu_t, df)); dyn->f.dfnone=0;} else SET_DFNONE(S)
+#define SET_NODF()          dyn->f.dfnone = 0
+#define SET_DFOK()          dyn->f.dfnone = 1
 
 #ifndef READFLAGS
 #define READFLAGS(A) \
-    if(((A)!=X_PEND) && dyn->state_flags!=SF_SET && dyn->state_flags!=SF_SET_PENDING) { \
-        if(dyn->state_flags!=SF_PENDING) {              \
+    if(((A)!=X_PEND && dyn->f.pending!=SF_SET)          \
+    && (dyn->f.pending!=SF_SET_PENDING)) {              \
+        if(dyn->f.pending!=SF_PENDING) {                \
             LDRw_U12(x3, xEmu, offsetof(x64emu_t, df)); \
-            j64 = (GETMARKF)-(dyn->arm_size);           \
+            j64 = (GETMARKF)-(dyn->native_size);        \
             CBZw(x3, j64);                              \
         }                                               \
         CALL_(UpdateFlags, -1, 0);                      \
         MARKF;                                          \
-        dyn->state_flags = SF_SET;                      \
+        dyn->f.pending = SF_SET;                        \
         SET_DFOK();                                     \
     }
 #endif
+// SF_MAYSET doesn't change the flags status cache
+// it also doesn't consume any needed flags
 #ifndef SETFLAGS
-#define SETFLAGS(A, B)  \
-    if(dyn->state_flags!=SF_SET && B==SF_SUBSET && (dyn->insts[ninst].x64.need_flags&(~((A)/*|X_PEND*/)))) \
-        READFLAGS(dyn->insts[ninst].x64.need_flags&(~(A)|X_PEND));  \
-    dyn->state_flags = (B==SF_SUBSET)?SF_SET:                       \
-        ((B==SF_SET_PENDING && !(dyn->insts[ninst].x64.need_flags&X_PEND)?SF_SET:B))
-
+#define SETFLAGS(A, B)                                                                          \
+    if(dyn->f.pending!=SF_SET                                                                   \
+    && (B==SF_SUBSET || B==SF_SUBSET_PENDING)                                                   \
+    && (dyn->insts[ninst].x64.need_flags&(~((A)|((B==SF_SUBSET_PENDING)?X_PEND:0)))))           \
+        READFLAGS(dyn->insts[ninst].x64.need_flags&(~(A)|X_PEND));                              \
+    if(dyn->insts[ninst].x64.need_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].x64.need_flags&X_PEND)?SF_SET_PENDING:SF_SET;   \
+            break;                                                                              \
+        case SF_MAYSET: break;                                                                  \
+    } else dyn->f.pending = SF_SET
 #endif
 #ifndef JUMP
-#define JUMP(A) 
+#define JUMP(A, C) 
 #endif
 #ifndef BARRIER
 #define BARRIER(A) 
@@ -638,6 +667,7 @@
             TABLE64(xRIP, (A));                         \
     }
 #endif
+#define CLEARIP()   dyn->last_ip=0
 
 #if STEP < 2
 #define PASS2IF(A, B) if(A)
@@ -658,7 +688,7 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 #define STEPNAME(N) STEPNAME2(N, STEP)
 #endif
 
-#define arm_pass        STEPNAME(arm_pass)
+#define native_pass        STEPNAME(native_pass)
 
 #define dynarec64_00       STEPNAME(dynarec64_00)
 #define dynarec64_0F       STEPNAME(dynarec64_0F)
@@ -766,12 +796,16 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 #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_swapreg     STEPNAME(x87_swapreg)
 #define x87_setround    STEPNAME(x87_setround)
 #define x87_restoreround STEPNAME(x87_restoreround)
 #define sse_setround    STEPNAME(sse_setround)
@@ -792,6 +826,8 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 #define fpu_reflectcache STEPNAME(fpu_reflectcache)
 #endif
 
+#define fpuCacheTransform       STEPNAME(fpuCacheTransform)
+
 /* setup r2 to address pointed by */
 uintptr_t geted(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop, uint8_t* ed, uint8_t hint, int64_t* fixaddress, int absmax, uint32_t mask, rex_t rex, int s, int delta);
 
@@ -887,21 +923,29 @@ 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 x87_do_push(dynarec_arm_t* dyn, int ninst, int s1, int t);
 // fpu push. Do not allocate a cache register. Needs a scratch register to do x87stack synch (or 0 to not do it)
 void x87_do_push_empty(dynarec_arm_t* dyn, int ninst, int s1);
 // fpu pop. All previous returned Dd should be considered invalid
-void x87_do_pop(dynarec_arm_t* dyn, int ninst);
+void x87_do_pop(dynarec_arm_t* dyn, int ninst, int s1);
+// get cache index for a x87 reg, return -1 if cache doesn't exist
+int x87_get_current_cache(dynarec_arm_t* dyn, int ninst, int st, int t);
 // get cache index for a x87 reg, create the entry if needed
-int x87_get_cache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a);
+int x87_get_cache(dynarec_arm_t* dyn, int ninst, int populate, int s1, int s2, int a, int t);
+// get neoncache index for a x87 reg
+int x87_get_neoncache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a);
 // get vfpu register for a x87 reg, create the entry if needed
-int x87_get_st(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a);
+int x87_get_st(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a, int t);
+// get vfpu register for a x87 reg, create the entry if needed. Do not fetch the Stx if not already in cache
+int x87_get_st_empty(dynarec_arm_t* dyn, int ninst, int s1, int s2, int a, int t);
 // refresh a value from the cache ->emu (nothing done if value is not cached)
 void x87_refresh(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st);
 // refresh a value from the cache ->emu and then forget the cache (nothing done if value is not cached)
 void x87_forget(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st);
 // refresh the cache value from emu
 void x87_reget_st(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st);
+// 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 cw flags, return reg to restore flags
 int x87_setround(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
 // Restore round flag
@@ -909,15 +953,47 @@ void x87_restoreround(dynarec_arm_t* dyn, int ninst, int s1);
 // Set rounding according to mxcsr flags, return reg to restore flags
 int sse_setround(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
 
+void fpuCacheTransform(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
+
+#if STEP < 2
+#define CHECK_CACHE()   0
+#else
+#define CHECK_CACHE()   fpuCacheNeedsTransform(dyn, ninst)
+#endif
+
+#define neoncache_st_coherency STEPNAME(neoncache_st_coherency)
+int neoncache_st_coherency(dynarec_arm_t* dyn, int ninst, int a, int b);
+
+#if STEP == 0
+#define ST_IS_F(A)          0
+#define X87_COMBINE(A, B)   NEON_CACHE_ST_D
+#define X87_ST0             NEON_CACHE_ST_D
+#define X87_ST(A)           NEON_CACHE_ST_D
+#elif STEP == 1
+#define ST_IS_F(A) (neoncache_get_current_st(dyn, ninst, A)==NEON_CACHE_ST_F)
+#define X87_COMBINE(A, B) neoncache_combine_st(dyn, ninst, A, B)
+#define X87_ST0     neoncache_get_current_st(dyn, ninst, 0)
+#define X87_ST(A)   neoncache_get_current_st(dyn, ninst, A)
+#else
+#define ST_IS_F(A) (neoncache_get_st(dyn, ninst, A)==NEON_CACHE_ST_F)
+#if STEP == 3
+#define X87_COMBINE(A, B) neoncache_st_coherency(dyn, ninst, A, B)
+#else
+#define X87_COMBINE(A, B) neoncache_get_st(dyn, ninst, A)
+#endif
+#define X87_ST0     neoncache_get_st(dyn, ninst, 0)
+#define X87_ST(A)   neoncache_get_st(dyn, ninst, A)
+#endif
+
 //MMX helpers
 // get neon register for a MMX reg, create the entry if needed
-int mmx_get_reg(dynarec_arm_t* dyn, int ninst, int s1, int a);
+int mmx_get_reg(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int a);
 // get neon register for a MMX reg, but don't try to synch it if it needed to be created
-int mmx_get_reg_empty(dynarec_arm_t* dyn, int ninst, int s1, int a);
+int mmx_get_reg_empty(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int a);
 
 //SSE/SSE2 helpers
 // get neon register for a SSE reg, create the entry if needed
-int sse_get_reg(dynarec_arm_t* dyn, int ninst, int s1, int a);
+int sse_get_reg(dynarec_arm_t* dyn, int ninst, int s1, int a, int forwrite);
 // get neon register for a SSE reg, but don't try to synch it if it needed to be created
 int sse_get_reg_empty(dynarec_arm_t* dyn, int ninst, int s1, int a);
 // forget neon register for a SSE reg, create the entry if needed
@@ -927,13 +1003,13 @@ void sse_purge07cache(dynarec_arm_t* dyn, int ninst, int s1);
 
 // common coproc helpers
 // reset the cache
-void fpu_reset(dynarec_arm_t* dyn, int ninst);
+void fpu_reset(dynarec_arm_t* dyn);
 // purge the FPU cache (needs 3 scratch registers)
-void fpu_purgecache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
+void fpu_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1, int s2, int s3);
 // purge MMX cache
-void mmx_purgecache(dynarec_arm_t* dyn, int ninst, int s1);
+void mmx_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1);
 // purge x87 cache
-void x87_purgecache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
+void x87_purgecache(dynarec_arm_t* dyn, int ninst, int next, int s1, int s2, int s3);
 #ifdef HAVE_TRACE
 void fpu_reflectcache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3);
 #endif

src/dynarec/arm64/dynarec_arm64_pass0.h

@@ -0,0 +1,54 @@
+
+#define INIT    uintptr_t sav_addr=addr
+#define FINI    \
+    dyn->isize = addr-sav_addr;         \
+    dyn->insts[ninst].x64.addr = addr;  \
+    if(ninst) dyn->insts[ninst-1].x64.size = dyn->insts[ninst].x64.addr - dyn->insts[ninst-1].x64.addr
+
+#define MESSAGE(A, ...)  
+#define READFLAGS(A)    \
+        dyn->insts[ninst].x64.use_flags = A; dyn->f.dfnone = 1;\
+        dyn->f.pending=SF_SET
+#define SETFLAGS(A,B)   \
+        dyn->insts[ninst].x64.set_flags = A;            \
+        if(B!=SF_MAYSET) {                              \
+                dyn->insts[ninst].x64.state_flags = B;  \
+                dyn->f.pending=(B)&SF_SET_PENDING;      \
+                dyn->f.dfnone=((B)&SF_SET)?1:0;         \
+        }
+#define EMIT(A)     
+#define JUMP(A, C)         add_next(dyn, (uintptr_t)A); dyn->insts[ninst].x64.jmp = A; dyn->insts[ninst].x64.jmp_cond = C
+#define BARRIER(A)      if(A!=BARRIER_MAYBE) {fpu_purgecache(dyn, ninst, 0, x1, x2, x3); dyn->insts[ninst].x64.barrier = A;} else dyn->insts[ninst].barrier_maybe = 1
+#define BARRIER_NEXT(A) dyn->insts[ninst+1].x64.barrier = A
+#define NEW_INST \
+        ++dyn->size;                            \
+        if(dyn->size+3>=dyn->cap) {             \
+                dyn->insts = (instruction_native_t*)realloc(dyn->insts, sizeof(instruction_native_t)*dyn->cap*2);     \
+                memset(&dyn->insts[dyn->cap], 0, sizeof(instruction_native_t)*dyn->cap);   \
+                dyn->cap *= 2;                  \
+        }                                       \
+        dyn->insts[ninst].x64.addr = ip;        \
+        dyn->n.combined1 = dyn->n.combined2 = 0;\
+        dyn->n.swapped = 0; dyn->n.barrier = 0; \
+        dyn->insts[ninst].f_entry = dyn->f;     \
+        if(ninst) {dyn->insts[ninst-1].x64.size = dyn->insts[ninst].x64.addr - dyn->insts[ninst-1].x64.addr;}
+
+#define INST_EPILOG                             \
+        dyn->insts[ninst].f_exit = dyn->f;      \
+        dyn->insts[ninst].n = dyn->n;           \
+        dyn->insts[ninst].x64.has_next = (ok>0)?1:0;
+#define INST_NAME(name) 
+#define DEFAULT                         \
+        --dyn->size;                    \
+        *ok = -1;                       \
+        if(box64_dynarec_log>=LOG_INFO) {\
+        dynarec_log(LOG_NONE, "%p: Dynarec stopped because of Opcode %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X", \
+        (void*)ip, PKip(0),             \
+        PKip(1), PKip(2), PKip(3),      \
+        PKip(4), PKip(5), PKip(6),      \
+        PKip(7), PKip(8), PKip(9),      \
+        PKip(10),PKip(11),PKip(12),     \
+        PKip(13),PKip(14));             \
+        printFunctionAddr(ip, " => ");  \
+        dynarec_log(LOG_NONE, "\n");    \
+        }

src/dynarec/arm64/dynarec_arm64_pass1.h

@@ -0,0 +1,14 @@
+#define INIT    
+#define FINI
+#define MESSAGE(A, ...)  
+#define EMIT(A)     
+#define NEW_INST                                \
+        dyn->insts[ninst].f_entry = dyn->f;     \
+        dyn->n.combined1 = dyn->n.combined2 = 0;\
+        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/arm64/dynarec_arm64_pass2.h

@@ -0,0 +1,17 @@
+#define INIT        dyn->native_size = 0
+#define FINI        if(ninst) {dyn->insts[ninst].address = (dyn->insts[ninst-1].address+dyn->insts[ninst-1].size);}
+
+#define MESSAGE(A, ...)  
+#define EMIT(A)     dyn->insts[ninst].size+=4; dyn->native_size+=4
+#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->last_ip = 0;                                                               \
+                        ++dyn->sons_size;                                                               \
+                }                                                                                       \
+        }
+#define INST_EPILOG dyn->insts[ninst].epilog = dyn->native_size; 
+#define INST_NAME(name) 
+#define TABLE64(A, V)   {Table64(dyn, (V)); EMIT(0);}
+#define FTABLE64(A, V)  {mmx87_regs_t v = {.d = V}; Table64(dyn, v.q); EMIT(0);}

src/dynarec/arm64/dynarec_arm64_pass3.h

@@ -0,0 +1,65 @@
+#define INIT    
+#define FINI
+#define EMIT(A)     \
+    if(box64_dynarec_dump) {dynarec_log(LOG_NONE, "\t%08x\t%s\n", (uint32_t)(A), arm64_print(A, (uintptr_t)dyn->block));} \
+    *(uint32_t*)(dyn->block) = (uint32_t)(A);       \
+    dyn->block += 4; dyn->native_size += 4;         \
+    dyn->insts[ninst].size2 += 4
+
+#define MESSAGE(A, ...)  if(box64_dynarec_dump) dynarec_log(LOG_NONE, __VA_ARGS__)
+#define NEW_INST        \
+    if(ninst && isInstClean(dyn, ninst)) {                      \
+        dyn->last_ip = 0;                                       \
+        dyn->sons_x64[dyn->sons_size] = (uintptr_t)ip;          \
+        dyn->sons_native[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(box64_dynarec_dump) {\
+        printf_x64_instruction(my_context->dec, &dyn->insts[ninst].x64, name); \
+        dynarec_log(LOG_NONE, "%s%p: %d emited opcodes, inst=%d, barrier=%d state=%d/%d(%d), set=%X, use=%X, need=%X", \
+            (box64_dynarec_dump>1)?"\e[32m":"", \
+            (void*)(dyn->native_start+dyn->insts[ninst].address),  \
+            dyn->insts[ninst].size/4,           \
+            ninst,                              \
+            dyn->insts[ninst].x64.barrier,      \
+            dyn->insts[ninst].x64.state_flags,  \
+            dyn->f.pending,                     \
+            dyn->f.dfnone,                      \
+            dyn->insts[ninst].x64.set_flags,    \
+            dyn->insts[ninst].x64.use_flags,    \
+            dyn->insts[ninst].x64.need_flags);  \
+        if(dyn->insts[ninst].pred_sz) {         \
+            dynarec_log(LOG_NONE, ", pred=");   \
+            for(int ii=0; ii<dyn->insts[ninst].pred_sz; ++ii)\
+                dynarec_log(LOG_NONE, "%s%d", ii?"/":"", dyn->insts[ninst].pred[ii]);\
+        }                                       \
+        if(dyn->insts[ninst].x64.jmp && dyn->insts[ninst].x64.jmp_insts>=0)\
+            dynarec_log(LOG_NONE, ", jmp=%d", dyn->insts[ninst].x64.jmp_insts);\
+        if(dyn->insts[ninst].x64.jmp && dyn->insts[ninst].x64.jmp_insts==-1)\
+            dynarec_log(LOG_NONE, ", jmp=out"); \
+        if(dyn->last_ip)                        \
+            dynarec_log(LOG_NONE, ", last_ip=%p", (void*)dyn->last_ip);\
+        for(int ii=0; ii<24; ++ii) {            \
+            switch(dyn->insts[ninst].n.neoncache[ii].t) {    \
+                case NEON_CACHE_ST_D: dynarec_log(LOG_NONE, " D%d:%s", ii, 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, 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, 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, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;  \
+                case NEON_CACHE_XMMR: dynarec_log(LOG_NONE, " Q%d:%s", ii, getCacheName(dyn->insts[ninst].n.neoncache[ii].t, dyn->insts[ninst].n.neoncache[ii].n)); break;  \
+                case NEON_CACHE_SCR: dynarec_log(LOG_NONE, " D%d:%s", ii, 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", (box64_dynarec_dump>1)?"\e[m":"");                       \
+    }
+
+#define TABLE64(A, V)   {int val64offset = Table64(dyn, (V)); MESSAGE(LOG_DUMP, "  Table64: 0x%lx\n", (V)); LDRx_literal(A, val64offset);}
+#define FTABLE64(A, V)  {mmx87_regs_t v = {.d = V}; int val64offset = Table64(dyn, v.q); MESSAGE(LOG_DUMP, "  FTable64: %g\n", v.d); VLDR64_literal(A, val64offset);}

src/dynarec/arm64/dynarec_arm64_private.h

@@ -0,0 +1,124 @@
+#ifndef __DYNAREC_ARM_PRIVATE_H_
+#define __DYNAREC_ARM_PRIVATE_H_
+
+#include "../dynarec_private.h"
+
+typedef struct x64emu_s x64emu_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 {
+        unsigned int t:4;   // reg type
+        unsigned int n:4;   // reg number
+    };
+} neon_cache_t;
+typedef union sse_cache_s {
+    int8_t      v;
+    struct {
+        unsigned int reg:7;
+        unsigned int write:1;
+    };
+} 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[16];   // cache status for the 16 SSE(2) registers
+    int8_t              fpuused[24];    // all 0..24 double reg from fpu, used by x87, sse and mmx
+    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_arm64_s {
+    instruction_x64_t   x64;
+    uintptr_t           address;    // (start) address of the arm emited instruction
+    uintptr_t           epilog;     // epilog of current instruction (can be start of next, or 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;
+    uintptr_t           marklock;
+    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_arm64_t;
+
+typedef struct dynarec_arm_s {
+    instruction_arm64_t *insts;
+    int32_t             size;
+    int32_t             cap;
+    uintptr_t           start;      // start of the block
+    uint32_t            isize;      // size in byte of x64 instructions included
+    void*               block;      // memory pointer where next instruction is emited
+    uintptr_t           native_start;  // start of the arm code
+    size_t              native_size;   // size of emitted arm code
+    uintptr_t           last_ip;    // last set IP in RIP (or NULL if unclean state) TODO: move to a cache something
+    uint64_t            *table64;   // table of 64bits value
+    int                 table64size;// size of table (will be appended at end of executable code)
+    int                 table64cap;
+    uintptr_t           tablestart;
+    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_x64;   // the x64 address of potential dynablock sons
+    void**              sons_native;   // 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);
+uintptr_t get_closest_next(dynarec_arm_t *dyn, uintptr_t addr);
+int is_nops(dynarec_arm_t *dyn, uintptr_t addr, int n);
+int is_instructions(dynarec_arm_t *dyn, uintptr_t addr, int n);
+
+int Table64(dynarec_arm_t *dyn, uint64_t val);  // add a value to etable64 (if needed) and gives back the imm19 to use in LDR_literal
+
+#define GO_TRACE() \
+    GETIP_(ip);             \
+    MOVx_REG(x1, xRIP);     \
+    STORE_XEMU_CALL(xRIP);  \
+    MOV32w(x2, 1);          \
+    CALL(PrintTrace, -1);   \
+    LOAD_XEMU_CALL(xRIP)
+
+#endif //__DYNAREC_ARM_PRIVATE_H_

src/dynarec/dynablock.c

@@ -19,12 +19,10 @@
 #include "dynablock_private.h"
 #include "dynarec_private.h"
 #include "elfloader.h"
-#ifdef ARM64
-#include "dynarec_arm64.h"
-#include "arm64_lock.h"
-#else
-#error Unsupported architecture!
-#endif
+
+#include "dynarec_native.h"
+#include "native_lock.h"
+
 #include "custommem.h"
 #include "khash.h"
 
@@ -74,13 +72,13 @@ void FreeDynablock(dynablock_t* db, int need_lock)
         if(db->parent->direct) {
             uintptr_t addr = (uintptr_t)db->x64_addr;
             if(addr>=startdb && addr<enddb)
-                arm64_lock_xchg(&db->parent->direct[addr-startdb], 0);   // secured write
+                native_lock_xchg(&db->parent->direct[addr-startdb], 0);   // secured write
         }
         // remove jumptable
         setJumpTableDefault64(db->x64_addr);
         // remove and free the sons
         for (int i=0; i<db->sons_size; ++i) {
-            dynablock_t *son = (dynablock_t*)arm64_lock_xchg(&db->sons[i], 0);
+            dynablock_t *son = (dynablock_t*)native_lock_xchg(&db->sons[i], 0);
             FreeDynablock(son, 0);
         }
         // only the father free the DynarecMap
@@ -189,7 +187,7 @@ int FreeRangeDynablock(dynablocklist_t* dynablocks, uintptr_t addr, uintptr_t si
             end = enddb;
         if(end>startdb && start<enddb)
             for(uintptr_t i = start; i<end; ++i) {
-                db = (dynablock_t*)arm64_lock_xchg(&dynablocks->direct[i-startdb], 0);
+                db = (dynablock_t*)native_lock_xchg(&dynablocks->direct[i-startdb], 0);
                 if(db) {
                     if(db->father)
                         db = db->father;
@@ -274,7 +272,7 @@ dynablock_t *AddNewDynablock(dynablocklist_t* dynablocks, uintptr_t addr, int* c
     pthread_mutex_lock(&my_context->mutex_dyndump);
     if(!dynablocks->direct) {
         dynablock_t** p = (dynablock_t**)calloc(dynablocks->textsz, sizeof(dynablock_t*));
-        if(arm64_lock_storeifnull(&dynablocks->direct, p)!=p)
+        if(native_lock_storeifnull(&dynablocks->direct, p)!=p)
             free(p);    // someone already create the direct array, too late...
     }
 
@@ -283,7 +281,7 @@ dynablock_t *AddNewDynablock(dynablocklist_t* dynablocks, uintptr_t addr, int* c
 
     block = (dynablock_t*)calloc(1, sizeof(dynablock_t));
     block->parent = dynablocks; 
-    dynablock_t* tmp = (dynablock_t*)arm64_lock_storeifnull(&dynablocks->direct[addr-dynablocks->text], block);
+    dynablock_t* tmp = (dynablock_t*)native_lock_storeifnull(&dynablocks->direct[addr-dynablocks->text], block);
     if(tmp !=  block) {
         // a block appeard!
         pthread_mutex_unlock(&my_context->mutex_dyndump);
@@ -353,7 +351,7 @@ static dynablock_t* internalDBGetBlock(x64emu_t* emu, uintptr_t addr, uintptr_t
         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*)arm64_lock_storeifref(&dynablocks->direct[addr-dynablocks->text], 0, block);
+        void* old = (void*)native_lock_storeifref(&dynablocks->direct[addr-dynablocks->text], 0, block);
         if(old!=block && old) {// put it back in place, strange things are happening here!
             dynarec_log(LOG_INFO, "Warning, a wild block appeared at %p: %p\n", (void*)addr, old);
             // doing nothing else, the block has not be writen
@@ -407,6 +405,7 @@ int IsInHotPage(uintptr_t addr) {
                 hotpage_size[i] = 0;
                 dynarec_log(LOG_DEBUG, "End of Hotpage %p\n", (void*)hotpage[i]);
             }
+            __sync_synchronize();
             return 1;
         }
     }
@@ -439,17 +438,20 @@ void AddHotPage(uintptr_t addr) {
             if(!hotpage_count[i])
                 ++hotpages;
             hotpage_count[i] = HOTPAGE_STEP;
+            __sync_synchronize();
             return;
         }
         if(addr==hotpage[i]+0x1000*(hotpage_size[i]+1)) {
             ++hotpage_size[i];
             hotpage_count[i] = HOTPAGE_STEP;
+            __sync_synchronize();
             return;
         }
         if(addr+0x1000==hotpage[i]) {
             ++hotpage_size[i];
             hotpage[i] = addr;
             hotpage_count[i] = HOTPAGE_STEP;
+            __sync_synchronize();
             return;
         }
     }
@@ -462,23 +464,30 @@ void AddHotPage(uintptr_t addr) {
             minidx = i;
         }
     if(hotpage_count[minidx]) {
-        dynarec_log(LOG_NONE, "Warning, not enough Hotpage, replacing %p(%p/%d) with %p\n", (void*)hotpage[minidx], (void*)(0x1000*(hotpage_size[minidx]+1)), hotpage_count[minidx], (void*)addr);
+        static int cnt = 0;
+        if(cnt<50) {
+            dynarec_log(LOG_NONE, "Warning, not enough Hotpage, replacing %p(%p/%d) with %p\n", (void*)hotpage[minidx], (void*)(0x1000*(hotpage_size[minidx]+1)), hotpage_count[minidx], (void*)addr);
+            ++cnt;
+            if(cnt==50)   // stop spamming console with message...
+                dynarec_log(LOG_NONE, "    will stop warning about not enough Hotpage now\n");
+        }
         hotpage_size[minidx] = 0;
     } else
         ++hotpages;
     hotpage[minidx] = addr;
     hotpage_count[minidx] = HOTPAGE_STEP;
+    __sync_synchronize();
 }
 
 dynablock_t* DBGetBlock(x64emu_t* emu, uintptr_t addr, int create, dynablock_t** 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))) {
+    dynablock_t *father = (db && db->father)?db->father:db;
+    if(father && father->done && db->block && father->need_test) {
         if(pthread_mutex_trylock(&my_context->mutex_dyndump)) {
             dynarec_log(LOG_DEBUG, "mutex_dyndump not available when trying to validate block %p from %p:%p (hash:%X) with %d son(s) for %p\n", db, db->x64_addr, db->x64_addr+db->x64_size-1, db->hash, db->sons_size, (void*)addr);
             return NULL;
         }
-        dynablock_t *father = db->father?db->father:db;
         if(AreaInHotPage((uintptr_t)father->x64_addr, (uintptr_t)father->x64_addr + father->x64_size - 1)) {
             dynarec_log(LOG_DEBUG, "Not running block %p from %p:%p with %d son(s) for %p because it's in a hotpage\n", father, father->x64_addr, father->x64_addr+father->x64_size-1, father->sons_size, (void*)addr);
             pthread_mutex_unlock(&my_context->mutex_dyndump);
@@ -518,10 +527,10 @@ dynablock_t* DBAlternateBlock(x64emu_t* emu, uintptr_t addr, uintptr_t filladdr)
     dynarec_log(LOG_DEBUG, "Creating AlternateBlock at %p for %p\n", (void*)addr, (void*)filladdr);
     int create = 1;
     dynablock_t *db = internalDBGetBlock(emu, addr, filladdr, create, NULL, 1);
-    if(db && db->done && db->block && (db->need_test || (db->father && db->father->need_test))) {
+    dynablock_t *father = (db && db->father)?db->father:db;
+    if(father && father->done && db->block && 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->x64_addr, father->x64_size);
         if(hash!=father->hash) {
             father->done = 0;   // invalidating the block

src/dynarec/dynarec.c

@@ -26,7 +26,15 @@
 #ifdef ARM64
 void arm64_prolog(x64emu_t* emu, void* addr) EXPORTDYN;
 void arm64_epilog() EXPORTDYN;
-void arm64_epilog_fast() EXPORTDYN;
+#define native_prolog       arm64_prolog
+#define native_epilog       arm64_epilog
+#elif defined(LA464)
+void la464_prolog(x64emu_t* emu, void* addr) EXPORTDYN;
+void la464_epilog() EXPORTDYN;
+#define native_prolog       la464_prolog
+#define native_epilog       la464_epilog
+#else
+#error Unsupported architecture
 #endif
 #endif
 
@@ -47,11 +55,11 @@ void* LinkNext(x64emu_t* emu, uintptr_t addr, void* x2, uintptr_t* x3)
     if(!block) {
         // no block, let link table as is...
         if(hasAlternate((void*)addr)) {
-            printf_log(LOG_INFO, "Jmp address has alternate: %p", (void*)addr);
+            printf_log(LOG_DEBUG, "Jmp address has alternate: %p", (void*)addr);
             addr = (uintptr_t)getAlternate((void*)addr);    // set new address
             R_RIP = addr;   // but also new RIP!
             *x3 = addr; // and the RIP in x27 register
-            printf_log(LOG_INFO, " -> %p\n", (void*)addr);
+            printf_log(LOG_DEBUG, " -> %p\n", (void*)addr);
             block = DBGetBlock(emu, addr, 1, &current);
         }
         if(!block) {
@@ -59,18 +67,17 @@ void* LinkNext(x64emu_t* emu, uintptr_t addr, void* x2, uintptr_t* x3)
             dynablock_t* db = FindDynablockFromNativeAddress(x2-4);
             dynarec_log(LOG_INFO, "Warning, jumping to a no-block address %p from %p (db=%p, x64addr=%p)\n", (void*)addr, x2-4, db, db?(void*)getX64Address(db, (uintptr_t)x2-4):NULL);
             #endif
-            //tableupdate(arm64_epilog, addr, table);
-            return arm64_epilog;
+            //tableupdate(native_epilog, addr, table);
+            return native_epilog;
         }
     }
     if(!block->done) {
         // not finished yet... leave linker
-        //tableupdate(arm_linker, addr, table);
-        return arm64_epilog;
+        return native_epilog;
     }
     if(!(jblock=block->block)) {
         // null block, but done: go to epilog, no linker here
-        return arm64_epilog;
+        return native_epilog;
     }
     //dynablock_t *father = block->father?block->father:block;
     return jblock;
@@ -129,9 +136,7 @@ void DynaCall(x64emu_t* emu, uintptr_t addr)
                 dynarec_log(LOG_DEBUG, "%04d|Calling DynaRec Block @%p (%p) of %d x64 instructions (father=%p) emu=%p\n", GetTID(), (void*)R_RIP, block->block, block->isize ,block->father, emu);
                 CHECK_FLAGS(emu);
                 // block is here, let's run it!
-                #ifdef ARM64
-                arm64_prolog(emu, block->block);
-                #endif
+                native_prolog(emu, block->block);
                 block = NULL;
             }
             if(emu->fork) {
@@ -210,9 +215,7 @@ int DynaRun(x64emu_t* emu)
             } else {
                 dynarec_log(LOG_DEBUG, "%04d|Running DynaRec Block @%p (%p) of %d x64 insts (father=%p) emu=%p\n", GetTID(), (void*)R_RIP, block->block, block->isize, block->father, emu);
                 // block is here, let's run it!
-                #ifdef ARM64
-                arm64_prolog(emu, block->block);
-                #endif
+                native_prolog(emu, block->block);
                 block = NULL;
             }
             if(emu->fork) {

src/dynarec/dynarec_arch.h

@@ -0,0 +1,25 @@
+#ifndef __DYNAREC_ARCH__H_
+#define __DYNAREC_ARCH__H_
+
+#ifdef ARM64
+#include "arm64/arm64_printer.h"
+#include "arm64/dynarec_arm64_private.h"
+#include "arm64/dynarec_arm64_functions.h"
+
+#define instruction_native_t        instruction_arm64_t
+#define dynarec_native_t            dynarec_arm_t
+
+#elif defined(LA464)
+#include "la464/la464_printer.h"
+#include "la464/dynarec_la464_private.h"
+#include "la464/dynarec_la464_functions.h"
+
+#define instruction_native_t        instruction_la464_t
+#define dynarec_native_t            dynarec_la464_t
+
+
+#else
+#error Unsupported platform
+#endif
+
+#endif //__DYNAREC_ARCH__H_

src/dynarec/dynarec_arm64_functions.c

@@ -1,466 +0,0 @@
-#define _GNU_SOURCE
-#include <stdio.h>
-#include <stdlib.h>
-#include <pthread.h>
-#include <errno.h>
-#include <string.h>
-#include <math.h>
-#include <signal.h>
-#include <sys/types.h>
-#include <unistd.h>
-
-#include "debug.h"
-#include "box64context.h"
-#include "dynarec.h"
-#include "emu/x64emu_private.h"
-#include "tools/bridge_private.h"
-#include "x64run.h"
-#include "x64emu.h"
-#include "box64stack.h"
-#include "callback.h"
-#include "emu/x64run_private.h"
-#include "emu/x87emu_private.h"
-#include "x64trace.h"
-#include "signals.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "dynarec_arm64_functions.h"
-#include "custommem.h"
-#include "bridge.h"
-
-void arm_fstp(x64emu_t* emu, void* p)
-{
-    if(ST0.q!=STld(0).uref)
-        D2LD(&ST0.d, p);
-    else
-        memcpy(p, &STld(0).ld, 10);
-}
-
-void arm_print_armreg(x64emu_t* emu, uintptr_t reg, uintptr_t n)
-{
-    (void)emu;
-    dynarec_log(LOG_DEBUG, "R%lu=0x%lx (%lu)\n", n, reg, reg);
-}
-
-void arm_f2xm1(x64emu_t* emu)
-{
-    ST0.d = exp2(ST0.d) - 1.0;
-}
-void arm_fyl2x(x64emu_t* emu)
-{
-    ST(1).d = log2(ST0.d)*ST(1).d;
-}
-void arm_ftan(x64emu_t* emu)
-{
-    ST0.d = tan(ST0.d);
-    emu->sw.f.F87_C2 = 0;
-}
-void arm_fpatan(x64emu_t* emu)
-{
-    ST1.d = atan2(ST1.d, ST0.d);
-}
-void arm_fxtract(x64emu_t* emu)
-{
-    int32_t tmp32s = (ST1.q&0x7ff0000000000000LL)>>52;
-    tmp32s -= 1023;
-    ST1.d /= exp2(tmp32s);
-    ST0.d = tmp32s;
-}
-void arm_fprem(x64emu_t* emu)
-{
-    int32_t tmp32s = ST0.d / ST1.d;
-    ST0.d -= ST1.d * tmp32s;
-    emu->sw.f.F87_C2 = 0;
-    emu->sw.f.F87_C0 = (tmp32s&1);
-    emu->sw.f.F87_C3 = ((tmp32s>>1)&1);
-    emu->sw.f.F87_C1 = ((tmp32s>>2)&1);
-}
-void arm_fyl2xp1(x64emu_t* emu)
-{
-    ST(1).d = log2(ST0.d + 1.0)*ST(1).d;
-}
-void arm_fsincos(x64emu_t* emu)
-{
-    sincos(ST1.d, &ST1.d, &ST0.d);
-    emu->sw.f.F87_C2 = 0;
-}
-void arm_frndint(x64emu_t* emu)
-{
-    ST0.d = fpu_round(emu, ST0.d);
-}
-void arm_fscale(x64emu_t* emu)
-{
-    if(ST0.d!=0.0)
-        ST0.d *= exp2(trunc(ST1.d));
-}
-void arm_fsin(x64emu_t* emu)
-{
-    ST0.d = sin(ST0.d);
-    emu->sw.f.F87_C2 = 0;
-}
-void arm_fcos(x64emu_t* emu)
-{
-    ST0.d = cos(ST0.d);
-    emu->sw.f.F87_C2 = 0;
-}
-
-void arm_fbld(x64emu_t* emu, uint8_t* ed)
-{
-    fpu_fbld(emu, ed);
-}
-
-void arm_fild64(x64emu_t* emu, int64_t* ed)
-{
-    int64_t tmp;
-    memcpy(&tmp, ed, sizeof(tmp));
-    ST0.d = tmp;
-    STll(0).sq = tmp;
-    STll(0).sref = ST0.sq;
-}
-
-void arm_fbstp(x64emu_t* emu, uint8_t* ed)
-{
-    fpu_fbst(emu, ed);
-}
-
-void arm_fistp64(x64emu_t* emu, int64_t* ed)
-{
-    // used of memcpy to avoid aligments issues
-    if(STll(0).sref==ST(0).sq) {
-        memcpy(ed, &STll(0).sq, sizeof(int64_t));
-    } else {
-        int64_t tmp;
-        if(isgreater(ST0.d, (double)(int64_t)0x7fffffffffffffffLL) || isless(ST0.d, (double)(int64_t)0x8000000000000000LL) || !isfinite(ST0.d))
-            tmp = 0x8000000000000000LL;
-        else
-            tmp = fpu_round(emu, ST0.d);
-        memcpy(ed, &tmp, sizeof(tmp));
-    }
-}
-
-void arm_fistt64(x64emu_t* emu, int64_t* ed)
-{
-    // used of memcpy to avoid aligments issues
-    int64_t tmp = ST0.d;
-    memcpy(ed, &tmp, sizeof(tmp));
-}
-
-void arm_fld(x64emu_t* emu, uint8_t* ed)
-{
-    memcpy(&STld(0).ld, ed, 10);
-    LD2D(&STld(0), &ST(0).d);
-    STld(0).uref = ST0.q;
-}
-
-void arm_ud(x64emu_t* emu)
-{
-    emit_signal(emu, SIGILL, (void*)R_RIP, 0);
-}
-
-void arm_fsave(x64emu_t* emu, uint8_t* ed)
-{
-    fpu_savenv(emu, (char*)ed, 0);
-
-    uint8_t* p = ed;
-    p += 28;
-    for (int i=0; i<8; ++i) {
-        LD2D(p, &ST(i).d);
-        p+=10;
-    }
-}
-void arm_frstor(x64emu_t* emu, uint8_t* ed)
-{
-    fpu_loadenv(emu, (char*)ed, 0);
-
-    uint8_t* p = ed;
-    p += 28;
-    for (int i=0; i<8; ++i) {
-        D2LD(&ST(i).d, p);
-        p+=10;
-    }
-
-}
-
-void arm_fprem1(x64emu_t* emu)
-{
-    // simplified version
-    int32_t tmp32s = round(ST0.d / ST1.d);
-    ST0.d -= ST1.d*tmp32s;
-    emu->sw.f.F87_C2 = 0;
-    emu->sw.f.F87_C0 = (tmp32s&1);
-    emu->sw.f.F87_C3 = ((tmp32s>>1)&1);
-    emu->sw.f.F87_C1 = ((tmp32s>>2)&1);
-}
-
-static uint8_t ff_mult(uint8_t a, uint8_t b)
-{
-	int retval = 0;
-
-	for(int i = 0; i < 8; i++) {
-		if((b & 1) == 1) 
-			retval ^= a;
-		
-		if((a & 0x80)) {
-			a <<= 1;
-			a  ^= 0x1b;
-		} else {
-			a <<= 1;
-		}
-		
-		b >>= 1;
-	}
-	
-	return retval;
-}
-
-void arm_aesimc(x64emu_t* emu, int xmm)
-{
-    sse_regs_t eax1 = emu->xmm[xmm];
-
-    for(int j=0; j<4; ++j) {
-        emu->xmm[xmm].ub[0+j*4] = ff_mult(0x0E, eax1.ub[0+j*4]) ^ ff_mult(0x0B, eax1.ub[1+j*4]) ^ ff_mult(0x0D, eax1.ub[2+j*4]) ^ ff_mult(0x09, eax1.ub[3+j*4]);
-        emu->xmm[xmm].ub[1+j*4] = ff_mult(0x09, eax1.ub[0+j*4]) ^ ff_mult(0x0E, eax1.ub[1+j*4]) ^ ff_mult(0x0B, eax1.ub[2+j*4]) ^ ff_mult(0x0D, eax1.ub[3+j*4]);
-        emu->xmm[xmm].ub[2+j*4] = ff_mult(0x0D, eax1.ub[0+j*4]) ^ ff_mult(0x09, eax1.ub[1+j*4]) ^ ff_mult(0x0E, eax1.ub[2+j*4]) ^ ff_mult(0x0B, eax1.ub[3+j*4]);
-        emu->xmm[xmm].ub[3+j*4] = ff_mult(0x0B, eax1.ub[0+j*4]) ^ ff_mult(0x0D, eax1.ub[1+j*4]) ^ ff_mult(0x09, eax1.ub[2+j*4]) ^ ff_mult(0x0E, eax1.ub[3+j*4]);
-    }
-}
-void arm_aesmc(x64emu_t* emu, int xmm)
-{
-    sse_regs_t eax1 = emu->xmm[xmm];
-
-    for(int j=0; j<4; ++j) {
-        emu->xmm[xmm].ub[0+j*4] = ff_mult(0x02, eax1.ub[0+j*4]) ^ ff_mult(0x03, eax1.ub[1+j*4]) ^               eax1.ub[2+j*4]  ^               eax1.ub[3+j*4] ;
-        emu->xmm[xmm].ub[1+j*4] =               eax1.ub[0+j*4]  ^ ff_mult(0x02, eax1.ub[1+j*4]) ^ ff_mult(0x03, eax1.ub[2+j*4]) ^               eax1.ub[3+j*4] ;
-        emu->xmm[xmm].ub[2+j*4] =               eax1.ub[0+j*4]  ^               eax1.ub[1+j*4]  ^ ff_mult(0x02, eax1.ub[2+j*4]) ^ ff_mult(0x03, eax1.ub[3+j*4]);
-        emu->xmm[xmm].ub[3+j*4] = ff_mult(0x03, eax1.ub[0+j*4]) ^               eax1.ub[1+j*4]  ^               eax1.ub[2+j*4]  ^ ff_mult(0x02, eax1.ub[3+j*4]);
-    }
-}
-void arm_aesdlast(x64emu_t* emu, int xmm)
-{
-                            //   A0 B1 C2 D3 E4 F5 G6 H7 I8 J9 Ka Lb Mc Nd Oe Pf
-                            //   A  N  K  H  E  B  O  L  I  F  C  P  M  J  G  D
-    const uint8_t invshiftrows[] = {0,13,10, 7, 4, 1,14,11, 8, 5, 2,15,12, 9, 6, 3};
-    const uint8_t invsubbytes[256] = {
-        0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38, 0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb,
-        0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87, 0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb,
-        0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d, 0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e,
-        0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2, 0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25,
-        0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16, 0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92,
-        0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda, 0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84,
-        0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a, 0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06,
-        0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02, 0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b,
-        0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea, 0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73,
-        0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85, 0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e,
-        0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89, 0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b,
-        0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20, 0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4,
-        0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31, 0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f,
-        0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d, 0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef,
-        0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0, 0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61,
-        0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26, 0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d,
-    };
-
-    sse_regs_t eax1;
-    for(int i=0; i<16; ++i)
-        eax1.ub[i] = emu->xmm[xmm].ub[invshiftrows[i]];
-    //STATE ← InvSubBytes( STATE );
-    for(int i=0; i<16; ++i)
-        emu->xmm[xmm].ub[i] = invsubbytes[eax1.ub[i]];
-
-}
-void arm_aeselast(x64emu_t* emu, int xmm)
-{
-                            //   A0 B1 C2 D3 E4 F5 G6 H7 I8 J9 Ka Lb Mc Nd Oe Pf
-                            //   A  F  K  P  E  J  O  D  I  N  C  H  M  B  G  L
-    const uint8_t shiftrows[] = {0, 5,10,15, 4, 9,14, 3, 8,13, 2, 7,12, 1, 6,11};
-    const uint8_t subbytes[256] = {
-        0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5, 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76,
-        0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0, 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0,
-        0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc, 0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15,
-        0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a, 0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75,
-        0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0, 0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84,
-        0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b, 0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf,
-        0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85, 0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8,
-        0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5, 0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2,
-        0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17, 0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73,
-        0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88, 0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb,
-        0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c, 0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79,
-        0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9, 0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08,
-        0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6, 0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a,
-        0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e, 0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e,
-        0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94, 0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf,
-        0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68, 0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16,
-    };
-
-    sse_regs_t eax1;
-    for(int i=0; i<16; ++i)
-        eax1.ub[i] = emu->xmm[xmm].ub[shiftrows[i]];
-    //STATE ← SubBytes( STATE );
-    for(int i=0; i<16; ++i)
-        emu->xmm[xmm].ub[i] = subbytes[eax1.ub[i]];
-}
-void arm_aesd(x64emu_t* emu, int xmm)
-{
-    arm_aesdlast(emu, xmm);
-    arm_aesimc(emu, xmm);
-}
-void arm_aese(x64emu_t* emu, int xmm)
-{
-    arm_aeselast(emu, xmm);
-    arm_aesmc(emu, xmm);
-}
-
-
-#define XMM0    0
-#define XMM8    16
-#define X870    8
-#define EMM0    8
-#define SCRATCH0    24
-
-// Get a FPU scratch reg
-int fpu_get_scratch(dynarec_arm_t* dyn)
-{
-    return SCRATCH0 + dyn->fpu_scratch++;  // return an Sx
-}
-// Reset scratch regs counter
-void fpu_reset_scratch(dynarec_arm_t* dyn)
-{
-    dyn->fpu_scratch = 0;
-}
-// Get a x87 double reg
-int fpu_get_reg_x87(dynarec_arm_t* dyn)
-{
-    int i=X870;
-    while (dyn->fpuused[i]) ++i;
-    dyn->fpuused[i] = 1;
-    return i; // return a Dx
-}
-// Free a FPU double reg
-void fpu_free_reg(dynarec_arm_t* dyn, int reg)
-{
-    // TODO: check upper limit?
-    dyn->fpuused[reg] = 0;
-}
-// Get an MMX double reg
-int fpu_get_reg_emm(dynarec_arm_t* dyn, int emm)
-{
-    dyn->fpuused[EMM0 + emm] = 1;
-    return EMM0 + emm;
-}
-// Get an XMM quad reg
-int fpu_get_reg_xmm(dynarec_arm_t* dyn, int xmm)
-{
-    if(xmm>7) {
-        dyn->fpuused[XMM8 + xmm - 8] = 1;
-        return XMM8 + xmm - 8;
-    } else {
-        dyn->fpuused[XMM0 + xmm] = 1;
-        return XMM0 + xmm;
-    }
-}
-// Reset fpu regs counter
-void fpu_reset_reg(dynarec_arm_t* dyn)
-{
-    dyn->fpu_reg = 0;
-    for (int i=0; i<32; ++i)
-        dyn->fpuused[i]=0;
-}
-
-#define F8      *(uint8_t*)(addr++)
-#define F32     *(uint32_t*)(addr+=4, addr-4)
-#define F32S64  (uint64_t)(int64_t)*(int32_t*)(addr+=4, addr-4)
-// Get if ED will have the correct parity. Not emiting anything. Parity is 2 for DWORD or 3 for QWORD
-int getedparity(dynarec_arm_t* dyn, int ninst, uintptr_t addr, uint8_t nextop, int parity, int delta)
-{
-    (void)dyn; (void)ninst;
-
-    uint32_t tested = (1<<parity)-1;
-    if((nextop&0xC0)==0xC0)
-        return 0;   // direct register, no parity...
-    if(!(nextop&0xC0)) {
-        if((nextop&7)==4) {
-            uint8_t sib = F8;
-            int sib_reg = (sib>>3)&7;
-            if((sib&0x7)==5) {
-                uint64_t tmp = F32S64;
-                if (sib_reg!=4) {
-                    // if XXXXXX+reg<<N then check parity of XXXXX and N should be enough
-                    return ((tmp&tested)==0 && (sib>>6)>=parity)?1:0;
-                } else {
-                    // just a constant...
-                    return (tmp&tested)?0:1;
-                }
-            } else {
-                if(sib_reg==4 && parity<3)
-                    return 0;   // simple [reg]
-                // don't try [reg1 + reg2<<N], unless reg1 is ESP
-                return ((sib&0x7)==4 && (sib>>6)>=parity)?1:0;
-            }
-        } else if((nextop&7)==5) {
-            uint64_t tmp = F32S64;
-            tmp+=addr+delta;
-            return (tmp&tested)?0:1;
-        } else {
-            return 0;
-        }
-    } else {
-        return 0; //Form [reg1 + reg2<<N + XXXXXX]
-    }
-}
-
-// Do the GETED, but don't emit anything...
-uintptr_t fakeed(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop) 
-{
-    (void)dyn; (void)addr; (void)ninst;
-
-    if((nextop&0xC0)==0xC0)
-        return addr;
-    if(!(nextop&0xC0)) {
-        if((nextop&7)==4) {
-            uint8_t sib = F8;
-            if((sib&0x7)==5) {
-                addr+=4;
-            }
-        } else if((nextop&7)==5) {
-            addr+=4;
-        }
-    } else {
-        if((nextop&7)==4) {
-            ++addr;
-        }
-        if(nextop&0x80) {
-            addr+=4;
-        } else {
-            ++addr;
-        }
-    }
-    return addr;
-}
-#undef F8
-#undef F32
-
-int isNativeCall(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t* calladdress, int* retn)
-{
-    (void)dyn;
-
-#define PK(a)       *(uint8_t*)(addr+a)
-#define PK32(a)     *(int32_t*)(addr+a)
-
-    if(!addr || !getProtection(addr))
-        return 0;
-    if(PK(0)==0xff && PK(1)==0x25) {            // "absolute" jump, maybe the GOT (well, RIP relative in fact)
-        uintptr_t a1 = addr + 6 + (PK32(2));    // need to add a check to see if the address is from the GOT !
-        addr = (uintptr_t)getAlternate(*(void**)a1);
-    }
-    if(!addr || !getProtection(addr))
-        return 0;
-    onebridge_t *b = (onebridge_t*)(addr);
-    if(b->CC==0xCC && b->S=='S' && b->C=='C' && b->w!=(wrapper_t)0 && b->f!=(uintptr_t)PltResolver) {
-        // found !
-        if(retn) *retn = (b->C3==0xC2)?b->N:0;
-        if(calladdress) *calladdress = addr+1;
-        return 1;
-    }
-    return 0;
-#undef PK32
-#undef PK
-}

src/dynarec/dynarec_arm64_pass.c

@@ -1,144 +0,0 @@
-#include <stdio.h>
-#include <stdlib.h>
-#include <stddef.h>
-#include <pthread.h>
-#include <errno.h>
-#include <string.h>
-
-#include "debug.h"
-#include "box64context.h"
-#include "dynarec.h"
-#include "emu/x64emu_private.h"
-#include "emu/x64run_private.h"
-#include "x64run.h"
-#include "x64emu.h"
-#include "box64stack.h"
-#include "emu/x64run_private.h"
-#include "x64trace.h"
-#include "dynablock.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "arm64_printer.h"
-#include "dynarec_arm64_functions.h"
-#include "dynarec_arm64_helper.h"
-#include "custommem.h"
-
-#ifndef STEP
-#error No STEP defined
-#endif
-
-uintptr_t arm_pass(dynarec_arm_t* dyn, uintptr_t addr)
-{
-    int ok = 1;
-    int ninst = 0;
-    uintptr_t ip = addr;
-    uintptr_t init_addr = addr;
-    rex_t rex;
-    int rep;    // 0 none, 1=F2 prefix, 2=F3 prefix
-    int need_epilog = 1;
-    dyn->sons_size = 0;
-    // Clean up (because there are multiple passes)
-    dyn->state_flags = 0;
-    dyn->dfnone = 0;
-    dyn->last_ip = ip;  // RIP is always set at start of block!
-    MAYUSE(init_addr);
-    fpu_reset(dyn, ninst);
-    // ok, go now
-    INIT;
-    while(ok) {
-        ip = addr;
-        if((dyn->insts[ninst].x64.barrier==1)) {
-            dyn->last_ip = 0;
-            NEW_BARRIER_INST;
-        }
-        NEW_INST;
-        fpu_reset_scratch(dyn);
-#ifdef HAVE_TRACE
-        if(my_context->dec && box64_dynarec_trace) {
-        if((trace_end == 0) 
-            || ((ip >= trace_start) && (ip < trace_end)))  {
-                MESSAGE(LOG_DUMP, "TRACE ----\n");
-                fpu_reflectcache(dyn, ninst, x1, x2, x3);
-                GETIP(ip);
-                MOVx_REG(x1, xRIP);
-                STORE_XEMU_CALL(xRIP);
-                MOV32w(x2, 1);
-                CALL(PrintTrace, -1);
-                LOAD_XEMU_CALL(xRIP);
-                MESSAGE(LOG_DUMP, "----------\n");
-            }
-        }
-#endif
-
-        rep = 0;
-        uint8_t pk = PK(0);
-        while((pk==0xF2) || (pk==0xF3)) {
-            rep = pk-0xF1;
-            ++addr;
-            pk = PK(0);
-        }
-        while(pk==0x3E) {   //Branch Taken Hint ignored
-            ++addr;
-            pk = PK(0);
-        }
-        rex.rex = 0;
-        while(pk>=0x40 && pk<=0x4f) {
-            rex.rex = pk;
-            ++addr;
-            pk = PK(0);
-        }
-
-        addr = dynarec64_00(dyn, addr, ip, ninst, rex, rep, &ok, &need_epilog);
-
-        INST_EPILOG;
-
-        if(dyn->insts[ninst+1].x64.barrier) {
-            fpu_purgecache(dyn, ninst, x1, x2, x3);
-            if(dyn->insts[ninst+1].x64.barrier!=2) {
-                dyn->state_flags = 0;
-                dyn->dfnone = 0;
-            }
-        }
-        if(!ok && !need_epilog && (addr < (dyn->start+dyn->isize))) {
-            ok = 1;
-        }
-        #if STEP == 0
-        if(!ok && !need_epilog && box64_dynarec_bigblock && getProtection(addr+3)&~PROT_CUSTOM && !IsInHotPage(addr+3))
-            if(*(uint32_t*)addr!=0) {   // check if need to continue (but is next 4 bytes are 0, stop)
-                uintptr_t next = get_closest_next(dyn, addr);
-                if(next && (
-                    (((next-addr)<15) && is_nops(dyn, addr, next-addr)) 
-                    ||(((next-addr)<30) && is_instructions(dyn, addr, next-addr)) ))
-                {
-                    dynarec_log(LOG_DEBUG, "Extend block %p, %p -> %p (ninst=%d)\n", dyn, (void*)addr, (void*)next, ninst);
-                    ok = 1;
-                } else if(next && (next-addr)<30) {
-                    dynarec_log(LOG_DEBUG, "Cannot extend block %p -> %p (%02X %02X %02X %02X %02X %02X %02X %02x)\n", (void*)addr, (void*)next, PK(0), PK(1), PK(2), PK(3), PK(4), PK(5), PK(6), PK(7));
-                }
-            }
-        #endif
-        if(ok<0)  {ok = 0; need_epilog=1;}
-        ++ninst;
-        #if STEP == 0
-        if(ok && !isJumpTableDefault64((void*)addr))
-        #else
-        if(ok && dyn->insts && (ninst==dyn->size))
-        #endif
-        {
-            #if STEP == 3
-            dynarec_log(LOG_DEBUG, "Stopping block %p (%d / %d)\n",(void*)init_addr, ninst, dyn->size); 
-            #endif
-            BARRIER(2);
-            fpu_purgecache(dyn, ninst, 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);
-        jump_to_epilog(dyn, ip, 0, ninst);  // no linker here, it's an unknow instruction
-    }
-    FINI;
-    MESSAGE(LOG_DUMP, "---- END OF BLOCK ---- (%d, %d sons)\n", dyn->size, dyn->sons_size);
-    return addr;
-}

src/dynarec/dynarec_arm64_pass0.h

@@ -1,33 +0,0 @@
-
-#define INIT    uintptr_t sav_addr=addr
-#define FINI    dyn->isize = addr-sav_addr
-#define MESSAGE(A, ...)  
-#define SETFLAGS(A, B)  
-#define READFLAGS(A)    
-#define EMIT(A)     
-#define JUMP(A)         add_next(dyn, (uintptr_t)A); dyn->insts[ninst].x64.jmp = A
-#define BARRIER(A)      dyn->insts[ninst].x64.barrier = A
-#define BARRIER_NEXT(A) if(ninst<dyn->size) dyn->insts[ninst+1].x64.barrier = A
-#define NEW_INST \
-        if(dyn->size+3>=dyn->cap) {             \
-                dyn->insts = (instruction_arm64_t*)realloc(dyn->insts, sizeof(instruction_arm64_t)*dyn->cap*2);     \
-                memset(&dyn->insts[dyn->cap], 0, sizeof(instruction_arm64_t)*dyn->cap);   \
-                dyn->cap *= 2;                  \
-        }                                       \
-        ++dyn->size
-#define INST_EPILOG         
-#define INST_NAME(name) 
-#define DEFAULT                         \
-        --dyn->size;                    \
-        *ok = -1;                       \
-        if(box64_dynarec_log>=LOG_INFO) {\
-        dynarec_log(LOG_NONE, "%p: Dynarec stopped because of Opcode %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X %02X", \
-        (void*)ip, PKip(0),             \
-        PKip(1), PKip(2), PKip(3),      \
-        PKip(4), PKip(5), PKip(6),      \
-        PKip(7), PKip(8), PKip(9),      \
-        PKip(10),PKip(11),PKip(12),     \
-        PKip(13),PKip(14));             \
-        printFunctionAddr(ip, " => ");  \
-        dynarec_log(LOG_NONE, "\n");    \
-        }

src/dynarec/dynarec_arm64_pass1.h

@@ -1,14 +0,0 @@
-#define INIT    
-#define FINI     \
-    dyn->insts[ninst].x64.addr = addr; \
-    if(ninst) dyn->insts[ninst-1].x64.size = dyn->insts[ninst].x64.addr - dyn->insts[ninst-1].x64.addr;
-#define MESSAGE(A, ...)  
-#define EMIT(A)     
-#define READFLAGS(A)    dyn->insts[ninst].x64.use_flags = A
-#define SETFLAGS(A,B)   {dyn->insts[ninst].x64.set_flags = A; dyn->insts[ninst].x64.state_flags = B;}
-
-#define NEW_INST \
-    dyn->insts[ninst].x64.addr = ip; \
-    if(ninst) dyn->insts[ninst-1].x64.size = dyn->insts[ninst].x64.addr - dyn->insts[ninst-1].x64.addr;
-#define INST_EPILOG
-#define INST_NAME(name)  

src/dynarec/dynarec_arm64_pass2.h

@@ -1,11 +0,0 @@
-#define INIT        dyn->arm_size = 0
-#define FINI        if(ninst) {dyn->insts[ninst].address = (dyn->insts[ninst-1].address+dyn->insts[ninst-1].size);}
-
-#define 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 INST_EPILOG dyn->insts[ninst].epilog = dyn->arm_size; 
-#define INST_NAME(name) 
-#define NEW_BARRIER_INST    if(ninst) ++dyn->sons_size
-#define TABLE64(A, V)   {Table64(dyn, (V)); EMIT(0);}
-#define FTABLE64(A, V)  {mmx87_regs_t v = {.d = V}; Table64(dyn, v.q); EMIT(0);}

src/dynarec/dynarec_arm64_pass3.h

@@ -1,36 +0,0 @@
-#define INIT    
-#define FINI
-#define EMIT(A)     \
-    if(box64_dynarec_dump) {dynarec_log(LOG_NONE, "\t%08x\t%s\n", (uint32_t)(A), arm64_print(A, (uintptr_t)dyn->block));} \
-    *(uint32_t*)(dyn->block) = (uint32_t)(A);       \
-    dyn->block += 4; dyn->arm_size += 4;\
-    dyn->insts[ninst].size2 += 4
-
-#define MESSAGE(A, ...)  if(box64_dynarec_dump) dynarec_log(LOG_NONE, __VA_ARGS__)
-#define NEW_INST        
-#define INST_EPILOG     
-#define INST_NAME(name) \
-    if(box64_dynarec_dump) {\
-        printf_x64_instruction(my_context->dec, &dyn->insts[ninst].x64, name); \
-        dynarec_log(LOG_NONE, "%s%p: %d emited opcodes, state=%d/%d, set=%X, use=%X, need=%X%s\n", \
-            (box64_dynarec_dump>1)?"\e[32m":"", \
-            (void*)(dyn->arm_start+dyn->insts[ninst].address),  \
-            dyn->insts[ninst].size/4,           \
-            dyn->insts[ninst].x64.state_flags,  \
-            dyn->state_flags,                   \
-            dyn->insts[ninst].x64.set_flags,    \
-            dyn->insts[ninst].x64.use_flags,    \
-            dyn->insts[ninst].x64.need_flags,   \
-            (box64_dynarec_dump>1)?"\e[m":"");  \
-    }
-
-#define NEW_BARRIER_INST                            \
-    if(ninst) {                                     \
-    dyn->sons_x64[dyn->sons_size] = (uintptr_t)ip;  \
-    dyn->sons_arm[dyn->sons_size] = dyn->block;     \
-    MESSAGE(LOG_DUMP, "----> potential Son here\n");\
-    ++dyn->sons_size;                               \
-    }
-
-#define TABLE64(A, V)   {int val64offset = Table64(dyn, (V)); MESSAGE(LOG_DUMP, "  Table64: 0x%lx\n", (V)); LDRx_literal(A, val64offset);}
-#define FTABLE64(A, V)  {mmx87_regs_t v = {.d = V}; int val64offset = Table64(dyn, v.q); MESSAGE(LOG_DUMP, "  FTable64: %g\n", v.d); VLDR64_literal(A, val64offset);}

src/dynarec/dynarec_arm64_private.h

@@ -1,64 +0,0 @@
-#ifndef __DYNAREC_ARM_PRIVATE_H_
-#define __DYNAREC_ARM_PRIVATE_H_
-
-#include "dynarec_private.h"
-
-typedef struct x64emu_s x64emu_t;
-typedef struct dynablock_s dynablock_t;
-
-typedef struct instruction_arm64_s {
-    instruction_x64_t   x64;
-    uintptr_t           address;    // (start) address of the arm emited instruction
-    uintptr_t           epilog;     // epilog of current instruction (can be start of next, of barrier stuff)
-    int                 size;       // size of the arm emited instruction
-    int                 size2;      // size of the arm emited instrucion after pass2
-    uintptr_t           mark, mark2, mark3;
-    uintptr_t           markf;
-    uintptr_t           markseg;
-    uintptr_t           marklock;
-    int                 pass2choice;// value for choices that are fixed on pass2 for pass3
-    uintptr_t           natcall;
-    int                 retn;
-} instruction_arm64_t;
-
-typedef struct dynarec_arm_s {
-    instruction_arm64_t *insts;
-    int32_t             size;
-    int32_t             cap;
-    uintptr_t           start;      // start of the block
-    uint32_t            isize;      // size in byte of x64 instructions included
-    void*               block;      // memory pointer where next instruction is emited
-    uintptr_t           arm_start;  // start of the arm code
-    size_t              arm_size;   // size of emitted arm code
-    int                 state_flags;// actual state for on-demand flags
-    uintptr_t           last_ip;    // last set IP in RIP (or NULL if unclean state)
-    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
-    int8_t              ssecache[16];// cache status for the 16 SSE(2) registers
-    int8_t              fpuused[32];// all 8..31 Q reg from fpu, used by x87, sse and mmx
-    int                 x87stack;   // cache stack counter
-    int                 fpu_scratch;// scratch counter
-    int                 fpu_reg;    // x87/sse/mmx reg counter
-    int                 dfnone;     // if defered flags is already set to df_none
-    uint64_t            *table64;   // table of 64bits value
-    int                 table64size;// size of table (will be appended at end of executable code)
-    int                 table64cap;
-    uintptr_t           tablestart;
-    uintptr_t*          next;       // variable array of "next" jump address
-    int                 next_sz;
-    int                 next_cap;
-    uintptr_t*          sons_x64;   // the x64 address of potential dynablock sons
-    void**              sons_arm;   // the arm address of potential dynablock sons
-    int                 sons_size;  // number of potential dynablock sons
-    dynablock_t*        dynablock;
-} dynarec_arm_t;
-
-void add_next(dynarec_arm_t *dyn, uintptr_t addr);
-uintptr_t get_closest_next(dynarec_arm_t *dyn, uintptr_t addr);
-int is_nops(dynarec_arm_t *dyn, uintptr_t addr, int n);
-int is_instructions(dynarec_arm_t *dyn, uintptr_t addr, int n);
-
-int Table64(dynarec_arm_t *dyn, uint64_t val);  // add a value to etable64 (if needed) and gives back the imm19 to use in LDR_literal
-
-#endif //__DYNAREC_ARM_PRIVATE_H_

src/dynarec/dynarec_helper.h

@@ -0,0 +1,12 @@
+#ifndef __DYNAREC_HELPER__H_
+#define __DYNAREC_HELPER__H_
+
+#ifdef ARM64
+#include "arm64/dynarec_arm64_helper.h"
+#elif defined(LA464)
+#include "la464/dynarec_la464_helper.h"
+#else
+#error Unsupported architecture
+#endif
+
+#endif //__DYNAREC_HELPER__H_

src/dynarec/dynarec_native.c

@@ -3,6 +3,7 @@
 #include <pthread.h>
 #include <errno.h>
 #include <string.h>
+#include <assert.h>
 
 #include "debug.h"
 #include "box64context.h"
@@ -18,11 +19,11 @@
 #include "x64trace.h"
 #include "dynablock.h"
 #include "dynablock_private.h"
-#include "dynarec_arm64.h"
-#include "dynarec_arm64_private.h"
-#include "dynarec_arm64_functions.h"
 #include "elfloader.h"
 
+#include "dynarec_native.h"
+#include "dynarec_arch.h"
+
 void printf_x64_instruction(zydis_dec_t* dec, instruction_x64_t* inst, const char* name) {
     uint8_t *ip = (uint8_t*)inst->addr;
     if(ip[0]==0xcc && ip[1]=='S' && ip[2]=='C') {
@@ -57,7 +58,7 @@ void printf_x64_instruction(zydis_dec_t* dec, instruction_x64_t* inst, const cha
     }
 }
 
-void add_next(dynarec_arm_t *dyn, uintptr_t addr) {
+void add_next(dynarec_native_t *dyn, uintptr_t addr) {
     if(!box64_dynarec_bigblock)
         return;
     for(int i=0; i<dyn->next_sz; ++i)
@@ -69,7 +70,7 @@ void add_next(dynarec_arm_t *dyn, uintptr_t addr) {
     }
     dyn->next[dyn->next_sz++] = addr;
 }
-uintptr_t get_closest_next(dynarec_arm_t *dyn, uintptr_t addr) {
+uintptr_t get_closest_next(dynarec_native_t *dyn, uintptr_t addr) {
     // get closest, but no addresses befores
     uintptr_t best = 0;
     int i = 0;
@@ -86,7 +87,7 @@ uintptr_t get_closest_next(dynarec_arm_t *dyn, uintptr_t addr) {
     return best;
 }
 #define PK(A) (*((uint8_t*)(addr+(A))))
-int is_nops(dynarec_arm_t *dyn, uintptr_t addr, int n)
+int is_nops(dynarec_native_t *dyn, uintptr_t addr, int n)
 {
     if(!n)
         return 1;
@@ -117,7 +118,7 @@ int is_nops(dynarec_arm_t *dyn, uintptr_t addr, int n)
 
 // return size of next instuciton, -1 is unknown
 // not all instrction are setup
-int next_instruction(dynarec_arm_t *dyn, uintptr_t addr)
+int next_instruction(dynarec_native_t *dyn, uintptr_t addr)
 {
     uint8_t opcode = PK(0);
     uint8_t nextop;
@@ -220,7 +221,7 @@ int next_instruction(dynarec_arm_t *dyn, uintptr_t addr)
 }
 #undef PK
 
-int is_instructions(dynarec_arm_t *dyn, uintptr_t addr, int n)
+int is_instructions(dynarec_native_t *dyn, uintptr_t addr, int n)
 {
     int i = 0;
     while(i<n) {
@@ -231,61 +232,14 @@ 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].x64.use_flags;
-    if(needed) {
-        setf &= ~needed;
-        if(!setf)   // all flags already used, no need to continue
-            return needed;
-    }
-
-    if(!needed && !dyn->insts[ninst].x64.set_flags && !dyn->insts[ninst].x64.jmp_insts) {
-        int start = ninst;
-        int end = ninst;
-        while(end<dyn->size && !dyn->insts[end].x64.use_flags && !dyn->insts[end].x64.set_flags && !dyn->insts[end].x64.jmp_insts)
-            ++end;
-        needed = needed_flags(dyn, end, setf, recurse);
-        for(int i=start; i<end; ++i)
-            dyn->insts[i].x64.need_flags = needed;
-        return needed;
-    }
-
-    if(dyn->insts[ninst].x64.set_flags && (dyn->insts[ninst].x64.state_flags!=SF_MAYSET)) {
-        if((setf & ~dyn->insts[ninst].x64.set_flags) == 0)
-            return needed;    // all done, gives all the flags needed
-        setf |= dyn->insts[ninst].x64.set_flags;    // add new flags to continue
-    }
-
-    int jinst = dyn->insts[ninst].x64.jmp_insts;
-    if(dyn->insts[ninst].x64.jmp) {
-        dyn->insts[ninst].x64.need_flags = (jinst==-1)?X_PEND:needed_flags(dyn, jinst, setf, recurse+1);
-        if(dyn->insts[ninst].x64.use_flags)  // conditionnal jump
-             dyn->insts[ninst].x64.need_flags |= needed_flags(dyn, ninst+1, setf, recurse);
-    } else
-        dyn->insts[ninst].x64.need_flags = needed_flags(dyn, ninst+1, setf, recurse);
-    if(dyn->insts[ninst].x64.state_flags==SF_MAYSET)
-        needed |= dyn->insts[ninst].x64.need_flags;
-    else
-        needed |= (dyn->insts[ninst].x64.need_flags & ~dyn->insts[ninst].x64.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 x64_size, int arm_size)
+instsize_t* addInst(instsize_t* insts, size_t* size, size_t* cap, int x64_size, int native_size)
 {
     // x64 instruction is <16 bytes
     int toadd;
-    if(x64_size>arm_size)
+    if(x64_size>native_size)
         toadd = 1 + x64_size/15;
     else
-        toadd = 1 + arm_size/15;
+        toadd = 1 + native_size/15;
     if((*size)+toadd>(*cap)) {
         *cap = (*size)+toadd;
         insts = (instsize_t*)realloc(insts, (*cap)*sizeof(instsize_t));
@@ -296,11 +250,11 @@ instsize_t* addInst(instsize_t* insts, size_t* size, size_t* cap, int x64_size,
         else
             insts[*size].x64 = x64_size;
         x64_size -= insts[*size].x64;
-        if(arm_size>15)
+        if(native_size>15)
             insts[*size].nat = 15;
         else
-            insts[*size].nat = arm_size;
-        arm_size -= insts[*size].nat;
+            insts[*size].nat = native_size;
+        native_size -= insts[*size].nat;
         ++(*size);
         --toadd;
     }
@@ -308,7 +262,7 @@ instsize_t* addInst(instsize_t* insts, size_t* size, size_t* cap, int x64_size,
 }
 
 // add a value to table64 (if needed) and gives back the imm19 to use in LDR_literal
-int Table64(dynarec_arm_t *dyn, uint64_t val)
+int Table64(dynarec_native_t *dyn, uint64_t val)
 {
     // find the value if already present
     int idx = -1;
@@ -329,17 +283,103 @@ int Table64(dynarec_arm_t *dyn, uint64_t val)
     return delta;
 }
 
+static void fillPredecessors(dynarec_native_t* dyn)
+{
+    int pred_sz = 0;
+    // 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].x64.jmp && dyn->insts[i].x64.jmp_insts!=-1) {
+            ++pred_sz;
+            ++dyn->insts[dyn->insts[i].x64.jmp_insts].pred_sz;
+        }
+    }
+    // second the "has_next"
+    for(int i=0; i<dyn->size; ++i) {
+        if(i!=dyn->size-1 && dyn->insts[i].x64.has_next && (!i || dyn->insts[i].pred_sz)) {
+            ++pred_sz;
+            ++dyn->insts[i+1].pred_sz;
+        }
+    }
+    dyn->predecessor = (int*)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[]
+    }
+    assert(p==dyn->predecessor+pred_sz);
+    // fill pred
+    for(int i=0; i<dyn->size; ++i) {
+        if(i!=dyn->size-1 && dyn->insts[i].x64.has_next && (!i || dyn->insts[i].pred_sz))
+            dyn->insts[i+1].pred[dyn->insts[i+1].pred_sz++] = i;
+        if(dyn->insts[i].x64.jmp && dyn->insts[i].x64.jmp_insts!=-1)
+            dyn->insts[dyn->insts[i].x64.jmp_insts].pred[dyn->insts[dyn->insts[i].x64.jmp_insts].pred_sz++] = i;
+    }
 
-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);
+}
+
+static void updateNeed(dynarec_native_t* dyn, int ninst, uint32_t need) {
+    uint32_t old_need = dyn->insts[ninst].x64.need_flags;
+    uint32_t new_need = old_need | need;
+    uint32_t new_use = dyn->insts[ninst].x64.use_flags;
+    uint32_t old_use = dyn->insts[ninst].x64.old_use;
+
+    if((new_need&X_PEND) && dyn->insts[ninst].x64.state_flags==SF_SUBSET) {
+        new_need &=~X_PEND;
+        new_need |= X_ALL;
+    }
+
+    uint32_t new_set = 0;
+    if(dyn->insts[ninst].x64.state_flags & SF_SET)
+        new_set = dyn->insts[ninst].x64.set_flags;
+    if(dyn->insts[ninst].x64.state_flags & SF_PENDING)
+        new_set |= X_PEND;
+    if((new_need&X_PEND) && (
+        dyn->insts[ninst].x64.state_flags==SF_SET || dyn->insts[ninst].x64.state_flags==SF_SUBSET)) {
+        new_need &=~X_PEND;
+        new_need |=X_ALL;
+    }
+    
+    dyn->insts[ninst].x64.need_flags = new_need;
+    dyn->insts[ninst].x64.old_use = new_use;
+
+    if(dyn->insts[ninst].x64.jmp_insts==-1)
+        new_need |= X_PEND;
+
+    // a Flag Barrier will change all need to "Pending", as it clear all flags optimisation
+    if(new_need && dyn->insts[ninst].x64.barrier&BARRIER_FLAGS)
+        new_need = X_PEND;
+    
+    if((new_need == old_need) && (new_use == old_use))    // no changes, bye
+        return;
+    
+    if(!(new_need && dyn->insts[ninst].x64.barrier&BARRIER_FLAGS)) {
+        new_need &=~new_set;    // clean needed flag that were suplied
+        new_need |= new_use;    // new need
+    }
+
+    if((new_need == (X_ALL|X_PEND)) && (dyn->insts[ninst].x64.state_flags & SF_SET))
+        new_need = X_ALL;
+
+    //update need to new need on predecessor
+    for(int i=0; i<dyn->insts[ninst].pred_sz; ++i)
+        updateNeed(dyn, dyn->insts[ninst].pred[i], new_need);
+}
+
+static void resetNeed(dynarec_native_t* dyn) {
+    for(int i = dyn->size; i-- > 0;) {
+        dyn->insts[i].x64.old_use = 0;
+        dyn->insts[i].x64.need_flags = dyn->insts[i].x64.default_need;
+    }
+}
 
 __thread void* current_helper = NULL;
 
 void CancelBlock64()
 {
-    dynarec_arm_t* helper = (dynarec_arm_t*)current_helper;
+    dynarec_native_t* helper = (dynarec_native_t*)current_helper;
     current_helper = NULL;
     if(!helper)
         return;
@@ -347,11 +387,16 @@ void CancelBlock64()
     free(helper->insts);
     free(helper->table64);
     free(helper->sons_x64);
-    free(helper->sons_arm);
+    free(helper->sons_native);
     if(helper->dynablock && helper->dynablock->block)
         FreeDynarecMap(helper->dynablock, (uintptr_t)helper->dynablock->block, helper->dynablock->size);
 }
 
+uintptr_t native_pass0(dynarec_native_t* dyn, uintptr_t addr);
+uintptr_t native_pass1(dynarec_native_t* dyn, uintptr_t addr);
+uintptr_t native_pass2(dynarec_native_t* dyn, uintptr_t addr);
+uintptr_t native_pass3(dynarec_native_t* dyn, uintptr_t addr);
+
 void* FillBlock64(dynablock_t* block, uintptr_t addr) {
     if(IsInHotPage(addr)) {
         dynarec_log(LOG_DEBUG, "Cancelling dynarec FillBlock on hotpage for %p\n", (void*)addr);
@@ -364,15 +409,15 @@ void* FillBlock64(dynablock_t* block, uintptr_t addr) {
     // protect the 1st page
     protectDB(addr, 1);
     // init the helper
-    dynarec_arm_t helper = {0};
+    dynarec_native_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_arm64_t*)calloc(helper.cap, sizeof(instruction_arm64_t));
-    // pass 0, addresses, x86 jump addresses, overall size of the block
-    uintptr_t end = arm_pass0(&helper, addr);
+    helper.insts = (instruction_native_t*)calloc(helper.cap, sizeof(instruction_native_t));
+    // pass 0, addresses, x64 jump addresses, overall size of the block
+    uintptr_t end = native_pass0(&helper, addr);
     // no need for next anymore
     free(helper.next);
     helper.next_sz = helper.next_cap = 0;
@@ -393,37 +438,98 @@ void* FillBlock64(dynablock_t* block, uintptr_t addr) {
         protectDB(addr, end-addr);  //end is 1byte after actual end
     // compute hash signature
     uint32_t hash = X31_hash_code((void*)addr, end-addr);
+    // Compute flag_need, without with current barriers
+    resetNeed(&helper);
+    for(int i = helper.size; i-- > 0;)
+        updateNeed(&helper, i, 0);
     // calculate barriers
     for(int i=0; i<helper.size; ++i)
         if(helper.insts[i].x64.jmp) {
             uintptr_t j = helper.insts[i].x64.jmp;
-            if(j<start || j>=end)
+            if(j<start || j>=end) {
                 helper.insts[i].x64.jmp_insts = -1;
-            else {
+                helper.insts[i].x64.use_flags |= X_PEND;
+            } else {
                 // find jump address instruction
                 int k=-1;
                 for(int i2=0; i2<helper.size && k==-1; ++i2) {
                     if(helper.insts[i2].x64.addr==j)
                         k=i2;
                 }
-                if(k!=-1)   // -1 if not found, mmm, probably wrong, exit anyway
-                    helper.insts[k].x64.barrier = 1;
+                if(k!=-1 && !helper.insts[i].barrier_maybe)
+                    helper.insts[k].x64.barrier |= BARRIER_FULL;
                 helper.insts[i].x64.jmp_insts = k;
             }
         }
-    // pass 1, flags
-    arm_pass1(&helper, addr);
-    for(int i=0; i<helper.size; ++i)
-        if(helper.insts[i].x64.set_flags && !helper.insts[i].x64.need_flags) {
-            helper.insts[i].x64.need_flags = needed_flags(&helper, i+1, helper.insts[i].x64.set_flags, 0);
-            if((helper.insts[i].x64.need_flags&X_PEND) && (helper.insts[i].x64.state_flags==SF_MAYSET))
-                helper.insts[i].x64.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].x64.jmp_insts == -1) {
+                // nothing for now
+            } else {
+                // inside block jump
+                int k = helper.insts[i].x64.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].x64.barrier|=BARRIER_FLAGS;
+                    }
+                } else {
+                    // jump back
+                    helper.insts[k].x64.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].x64.jmp
+        && ((k=helper.insts[i].x64.jmp_insts)>=0)
+        && helper.insts[k].x64.barrier&BARRIER_FLAGS) {
+            //TODO: optimize FPU barrier too
+            if((!helper.insts[k].x64.need_flags)
+             ||(helper.insts[k].x64.set_flags==X_ALL
+                  && helper.insts[k].x64.state_flags==SF_SET)
+             ||(helper.insts[k].x64.state_flags==SF_SET_PENDING)) {
+                //if(box64_dynarec_dump) dynarec_log(LOG_NONE, "Removed barrier for inst %d\n", k);
+                helper.insts[k].x64.barrier &= ~BARRIER_FLAGS; // remove flag barrier
+             }
         }
+    }*/
+    // reset need_flags and compute again, now taking barrier into account (because barrier change use_flags)
+    for(int i = helper.size; i-- > 0;) {
+        int k;
+        if(helper.insts[i].x64.jmp 
+        && ((k=helper.insts[i].x64.jmp_insts)>=0)
+        ) {
+            if(helper.insts[k].x64.barrier&BARRIER_FLAGS)
+                // jumpto barrier
+                helper.insts[i].x64.use_flags |= X_PEND;
+            if(helper.insts[i].x64.barrier&BARRIER_FLAGS && (helper.insts[k].x64.need_flags | helper.insts[k].x64.use_flags))
+                helper.insts[k].x64.barrier|=BARRIER_FLAGS;
+            else
+                helper.insts[i].x64.use_flags |= (helper.insts[k].x64.need_flags | helper.insts[k].x64.use_flags);
+        }
+        if(helper.insts[i].x64.barrier&BARRIER_FLAGS)
+            // immediate barrier
+            helper.insts[i].x64.use_flags |= X_PEND;
+    }
+    resetNeed(&helper);
+    for(int i = helper.size; i-- > 0;)
+        updateNeed(&helper, i, 0);
+
+    // pass 1, float optimisations, first pass for flags
+    native_pass1(&helper, addr);
     
     // pass 2, instruction size
-    arm_pass2(&helper, addr);
+    native_pass2(&helper, addr);
     // ok, now allocate mapped memory, with executable flag on
-    size_t sz = helper.arm_size + helper.table64size*sizeof(uint64_t);
+    size_t sz = helper.native_size + helper.table64size*sizeof(uint64_t);
     void* p = (void*)AllocDynarecMap(block, sz);
     if(p==NULL) {
         dynarec_log(LOG_INFO, "AllocDynarecMap(%p, %zu) failed, cancelling block\n", block, sz);
@@ -431,25 +537,25 @@ void* FillBlock64(dynablock_t* block, uintptr_t addr) {
         return NULL;
     }
     helper.block = p;
-    helper.arm_start = (uintptr_t)p;
-    helper.tablestart = helper.arm_start + helper.arm_size;
+    helper.native_start = (uintptr_t)p;
+    helper.tablestart = helper.native_start + helper.native_size;
     if(helper.sons_size) {
         helper.sons_x64 = (uintptr_t*)calloc(helper.sons_size, sizeof(uintptr_t));
-        helper.sons_arm = (void**)calloc(helper.sons_size, sizeof(void*));
+        helper.sons_native = (void**)calloc(helper.sons_size, sizeof(void*));
     }
-    // pass 3, emit (log emit arm opcode)
+    // pass 3, emit (log emit native opcode)
     if(box64_dynarec_dump) {
-        dynarec_log(LOG_NONE, "%s%04d|Emitting %zu bytes for %u x64 bytes", (box64_dynarec_dump>1)?"\e[01;36m":"", GetTID(), helper.arm_size, helper.isize); 
+        dynarec_log(LOG_NONE, "%s%04d|Emitting %zu bytes for %u x64 bytes", (box64_dynarec_dump>1)?"\e[01;36m":"", GetTID(), helper.native_size, helper.isize); 
         printFunctionAddr(helper.start, " => ");
         dynarec_log(LOG_NONE, "%s\n", (box64_dynarec_dump>1)?"\e[m":"");
     }
     int oldtable64size = helper.table64size;
-    size_t oldarmsize = helper.arm_size;
-    helper.arm_size = 0;
+    size_t oldnativesize = helper.native_size;
+    helper.native_size = 0;
     helper.table64size = 0; // reset table64 (but not the cap)
-    arm_pass3(&helper, addr);
-    if((oldarmsize!=helper.arm_size) || (oldtable64size<helper.table64size)) {
-        printf_log(LOG_NONE, "BOX64: Warning, size difference in block between pass2 (%zu) & pass3 (%zu)!\n", sz, helper.arm_size+helper.table64size*8);
+    native_pass3(&helper, addr);
+    if((oldnativesize!=helper.native_size) || (oldtable64size<helper.table64size)) {
+        printf_log(LOG_NONE, "BOX64: Warning, size difference in block between pass2 (%zu) & pass3 (%zu)!\n", sz, helper.native_size+helper.table64size*8);
         uint8_t *dump = (uint8_t*)helper.start;
         printf_log(LOG_NONE, "Dump of %d x64 opcodes:\n", helper.size);
         for(int i=0; i<helper.size; ++i) {
@@ -496,13 +602,14 @@ void* FillBlock64(dynablock_t* block, uintptr_t addr) {
         dynarec_log(LOG_INFO, "Warning, a block changed while beeing processed hash(%p:%ld)=%x/%x\n", block->x64_addr, block->x64_size, block->hash, hash);
         CancelBlock64();
         return NULL;
-    }    // fill sons if any
+    }
     if(!isprotectedDB(addr, end-addr)) {
         dynarec_log(LOG_INFO, "Warning, block unprotected while beeing processed %p:%ld, cancelling\n", block->x64_addr, block->x64_size);
         CancelBlock64();
         return NULL;
         //protectDB(addr, end-addr);
     }
+    // fill sons if any
     dynablock_t** sons = NULL;
     int sons_size = 0;
     if(helper.sons_size) {
@@ -511,10 +618,10 @@ void* FillBlock64(dynablock_t* block, uintptr_t addr) {
             int created = 1;
             dynablock_t *son = AddNewDynablock(block->parent, helper.sons_x64[i], &created);
             if(created) {    // avoid breaking a working block!
-                son->block = helper.sons_arm[i];
+                son->block = helper.sons_native[i];
                 son->x64_addr = (void*)helper.sons_x64[i];
                 son->x64_size = end-helper.sons_x64[i];
-                if(!son->x64_size) {printf_log(LOG_NONE, "Warning, son with null x64 size! (@%p / ARM=%p)", son->x64_addr, son->block);}
+                if(!son->x64_size) {printf_log(LOG_NONE, "Warning, son with null x64 size! (@%p / Native=%p)", son->x64_addr, son->block);}
                 son->father = block;
                 son->size = sz + son->block - block->block; // update size count, for debugging
                 //son->done = 1;
@@ -532,8 +639,8 @@ void* FillBlock64(dynablock_t* block, uintptr_t addr) {
     }
     free(helper.sons_x64);
     helper.sons_x64 = NULL;
-    free(helper.sons_arm);
-    helper.sons_arm = NULL;
+    free(helper.sons_native);
+    helper.sons_native = NULL;
     current_helper = NULL;
     //block->done = 1;
     return (void*)block;

src/dynarec/dynarec_native_pass.c

@@ -0,0 +1,247 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <pthread.h>
+#include <errno.h>
+#include <string.h>
+
+#include "debug.h"
+#include "box64context.h"
+#include "dynarec.h"
+#include "emu/x64emu_private.h"
+#include "emu/x64run_private.h"
+#include "x64run.h"
+#include "x64emu.h"
+#include "box64stack.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynablock.h"
+#include "dynarec_native.h"
+#include "custommem.h"
+
+#include "dynarec_arch.h"
+#include "dynarec_helper.h"
+
+#ifndef STEP
+#error No STEP defined
+#endif
+
+uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr)
+{
+    int ok = 1;
+    int ninst = 0;
+    uintptr_t ip = addr;
+    uintptr_t init_addr = addr;
+    rex_t rex;
+    int rep;    // 0 none, 1=F2 prefix, 2=F3 prefix
+    int need_epilog = 1;
+    dyn->sons_size = 0;
+    // Clean up (because there are multiple passes)
+    dyn->f.pending = 0;
+    dyn->f.dfnone = 0;
+    fpu_reset(dyn);
+    int reset_n = -1;
+    dyn->last_ip = (dyn->insts && dyn->insts[0].pred_sz)?0:ip;  // RIP is always set at start of block unless there is a predecessor!
+    // ok, go now
+    INIT;
+    while(ok) {
+        ip = addr;
+        if (reset_n!=-1) {
+            dyn->last_ip = 0;
+            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(box64_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].x64.barrier&BARRIER_FLOAT) {
+                    MESSAGE(LOG_DEBUG, "Apply Barrier Float\n");
+                    fpu_reset(dyn);
+                }
+                if(dyn->insts[ninst].x64.barrier&BARRIER_FLAGS) {
+                    MESSAGE(LOG_DEBUG, "Apply Barrier Flags\n");
+                    dyn->f.dfnone = 0;
+                    dyn->f.pending = 0;
+                }
+            }
+            reset_n = -1;
+        } else if(dyn->insts[ninst].pred_sz!=1)
+            dyn->last_ip = 0;   // reset IP if some jump are comming here
+        // 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);
+#ifdef HAVE_TRACE
+        if(my_context->dec && box64_dynarec_trace) {
+        if((trace_end == 0) 
+            || ((ip >= trace_start) && (ip < trace_end)))  {
+                MESSAGE(LOG_DUMP, "TRACE ----\n");
+                fpu_reflectcache(dyn, ninst, x1, x2, x3);
+                GO_TRACE();
+                MESSAGE(LOG_DUMP, "----------\n");
+            }
+        }
+#endif
+
+        rep = 0;
+        uint8_t pk = PK(0);
+        while((pk==0xF2) || (pk==0xF3)) {
+            rep = pk-0xF1;
+            ++addr;
+            pk = PK(0);
+        }
+        while(pk==0x3E) {   //Branch Taken Hint ignored
+            ++addr;
+            pk = PK(0);
+        }
+        rex.rex = 0;
+        while(pk>=0x40 && pk<=0x4f) {
+            rex.rex = pk;
+            ++addr;
+            pk = PK(0);
+        }
+
+        addr = dynarec64_00(dyn, addr, ip, ninst, rex, rep, &ok, &need_epilog);
+
+        INST_EPILOG;
+
+        int next = ninst+1;
+        #if STEP > 0
+        if(!dyn->insts[ninst].x64.has_next && dyn->insts[ninst].x64.jmp && dyn->insts[ninst].x64.jmp_insts!=-1)
+            next = dyn->insts[ninst].x64.jmp_insts;
+        #endif
+        if(dyn->insts[ninst].x64.has_next && dyn->insts[next].x64.barrier) {
+            if(dyn->insts[next].x64.barrier&BARRIER_FLOAT)
+                fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+            if(dyn->insts[next].x64.barrier&BARRIER_FLAGS) {
+                dyn->f.pending = 0;
+                dyn->f.dfnone = 0;
+                dyn->last_ip = 0;
+            }
+        }
+        #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].x64.barrier&BARRIER_FULL)==BARRIER_FULL)
+                reset_n = -2;    // hack to say Barrier!
+            else {
+                reset_n = getNominalPred(dyn, ii);  // may get -1 if no predecessor are availble
+                if(reset_n==-1) {
+                    reset_n = -2;
+                    MESSAGE(LOG_DEBUG, "Warning, Reset Caches mark not found\n");
+                }
+            }
+        }
+        #else
+        if(!ok && !need_epilog && box64_dynarec_bigblock && getProtection(addr+3)&~PROT_CUSTOM)
+            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))*/ ))
+                {
+                    ok = 1;
+                    // need to find back that instruction to copy the caches, as previous version cannot be used anymore
+                    reset_n = -2;
+                    for(int ii=0; ii<ninst; ++ii)
+                        if(dyn->insts[ii].x64.jmp == next) {
+                            reset_n = ii;
+                            ii=ninst;
+                        }
+                    if(box64_dynarec_dump) dynarec_log(LOG_NONE, "Extend block %p, %p -> %p (ninst=%d, jump from %d)\n", dyn, (void*)addr, (void*)next, ninst, reset_n);
+                } else if(next && (next-addr)<30) {
+                    if(box64_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 && !isJumpTableDefault64((void*)addr) && (box64_dynarec_bigblock<2))
+        #else
+        if(ok && (ninst==dyn->size))
+        #endif
+        {
+            int j32;
+            MAYUSE(j32);
+            MESSAGE(LOG_DEBUG, "Stopping block %p (%d / %d)\n",(void*)init_addr, ninst, dyn->size); 
+            --ninst;
+            if(!dyn->insts[ninst].x64.barrier) {
+                BARRIER(BARRIER_FLOAT);
+            }
+            #if STEP == 0
+            if(dyn->insts[ninst].x64.set_flags)
+                dyn->insts[ninst].x64.default_need |= X_PEND;
+            else
+                dyn->insts[ninst].x64.use_flags |= X_PEND;
+            #endif
+            ++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, 0, x1, x2, x3);
+        jump_to_epilog(dyn, ip, 0, ninst);  // no linker here, it's an unknow instruction
+    }
+    FINI;
+    MESSAGE(LOG_DUMP, "---- END OF BLOCK ---- (%d, %d sons)\n", dyn->size, dyn->sons_size);
+    return addr;
+}

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)
@@ -9,26 +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_SET_PENDING      3
-#define SF_MAYSET   4
-#define SF_SUBSET   5
+#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)
+#define SF_MAYSET   8
 
 typedef struct instruction_x64_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
+    uintptr_t   addr;       //address of the instruction
+    int32_t     size;       // size of the instruction
+    uintptr_t   jmp;        // offset to jump to, even if conditionnal (0 if not), no relative offset here
     int         jmp_insts;  // instuction to jump to (-1 if out of the block)
-    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     default_need;// 0 or X_PEND basically
+    uint8_t     need_flags; // calculated
+    uint8_t     old_use;    // calculated
 } instruction_x64_t;
 
 void printf_x64_instruction(zydis_dec_t* dec, instruction_x64_t* inst, const char* name);

src/dynarec/native_lock.h

@@ -0,0 +1,25 @@
+#ifndef __NATIVE_LOCK__H__
+#define __NATIVE_LOCK__H__
+
+#ifdef ARM64
+#include "arm64/arm64_lock.h"
+
+#define native_lock_read_b(A)               arm64_lock_read_b(A)
+#define native_lock_write_b(A, B)           arm64_lock_write_b(A, B)
+#define native_lock_read_h(A)               arm64_lock_read_h(A)
+#define native_lock_write_h(A, B)           arm64_lock_write_h(A, B)
+#define native_lock_read_d(A)               arm64_lock_read_d(A)
+#define native_lock_write_d(A, B)           arm64_lock_write_d(A, B)
+#define native_lock_read_dd(A)              arm64_lock_read_dd(A)
+#define native_lock_write_dd(A, B)          arm64_lock_write_dd(A, B)
+#define native_lock_read_dq(A, B, C)        arm64_lock_read_dq(A, B, C)
+#define native_lock_write_dq(A, B, C)       arm64_lock_write_dq(A, B, C)
+#define native_lock_xchg(A, B)              arm64_lock_xchg(A, B)
+#define native_lock_storeifref(A, B, C)     arm64_lock_storeifref(A, B, C)
+#define native_lock_storeifnull(A, B)       arm64_lock_storeifnull(A, B)
+
+#else
+#error Unsupported architecture
+#endif
+
+#endif //#define __NATIVE_LOCK__H__

src/elfs/elfdwarf_private.c

@@ -0,0 +1,717 @@
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+
+#include "debug.h"
+#include "elfloader_private.h"
+#include "elfdwarf_private.h"
+
+typedef struct dwarf_unwind_constr_s {
+    uint8_t reg_count;
+    uint64_t *table;
+    uint8_t *statuses;
+#define GET_STATUS(n,i) (((n).statuses[(i) >> 1] >> (((i) & 1) << 2)) & 0xF)
+#define SET_STATUS(n,i,v) (n).statuses[(i) >> 1] = ((n).statuses[(i) >> 1] & (0xF0 >> (((i) & 1) << 2))) | (v << (((i) & 1) << 2))
+#define REGSTATUS_undefined     0b0000
+#define REGSTATUS_same_val      0b0101
+#define REGSTATUS_offset        0b0010
+#define REGSTATUS_val_offset    0b0011
+#define REGSTATUS_register      0b0100
+} dwarf_unwind_constr_t;
+
+#define DW_EH_PE_uptr       0x00
+#define DW_EH_PE_uleb128    0x01
+#define DW_EH_PE_udata2     0x02
+#define DW_EH_PE_udata4     0x03
+#define DW_EH_PE_udata8     0x04
+#define DW_EH_PE_SIZE_MASK  0x07
+#define DW_EH_PE_signed     0x08
+#define DW_EH_PE_absptr     0x00
+#define DW_EH_PE_pcrel      0x10
+#define DW_EH_PE_textrel    0x20 // Unsupported
+#define DW_EH_PE_datarel    0x30 // Unsupported
+#define DW_EH_PE_indirect   0x80 // May crash
+#define DW_EH_PE_omit       0xff
+
+#define READ_U1(var, ptr) do { (var) = *( uint8_t*)(ptr); (ptr) += 1; } while (0)
+#define READ_S1(var, ptr) do { (var) = *(  int8_t*)(ptr); (ptr) += 1; } while (0)
+#define READ_U2(var, ptr) do { (var) = *(uint16_t*)(ptr); (ptr) += 2; } while (0)
+#define READ_S2(var, ptr) do { (var) = *( int16_t*)(ptr); (ptr) += 2; } while (0)
+#define READ_U4(var, ptr) do { (var) = *(uint32_t*)(ptr); (ptr) += 4; } while (0)
+#define READ_S4(var, ptr) do { (var) = *( int32_t*)(ptr); (ptr) += 4; } while (0)
+#define READ_U8(var, ptr) do { (var) = *(uint64_t*)(ptr); (ptr) += 8; } while (0)
+#define READ_S8(var, ptr) do { (var) = *( int64_t*)(ptr); (ptr) += 8; } while (0)
+#define READ_ULEB128(var, ptr) readULEB(&(var), &(ptr))
+#define READ_SLEB128(var, ptr) readSLEB((int64_t*)&(var), &(ptr))
+// Seems like DW_EH_PE_signed is actually an invalid encoding
+#define READ_ENCODED(var, ptr, enc, isptr) do { uintptr_t optr = (uintptr_t)(ptr); \
+         if ((((enc) & DW_EH_PE_SIZE_MASK) == DW_EH_PE_uleb128) &&  ((enc) & DW_EH_PE_signed)) READ_SLEB128(var, ptr); \
+    else if ((((enc) & DW_EH_PE_SIZE_MASK) == DW_EH_PE_uleb128) && !((enc) & DW_EH_PE_signed)) READ_ULEB128(var, ptr); \
+    else if ((((enc) & DW_EH_PE_SIZE_MASK) == DW_EH_PE_udata2 ) &&  ((enc) & DW_EH_PE_signed)) READ_S2     (var, ptr); \
+    else if ((((enc) & DW_EH_PE_SIZE_MASK) == DW_EH_PE_udata2 ) && !((enc) & DW_EH_PE_signed)) READ_U2     (var, ptr); \
+    else if ((((enc) & DW_EH_PE_SIZE_MASK) == DW_EH_PE_udata4 ) &&  ((enc) & DW_EH_PE_signed)) READ_S4     (var, ptr); \
+    else if ((((enc) & DW_EH_PE_SIZE_MASK) == DW_EH_PE_udata4 ) && !((enc) & DW_EH_PE_signed)) READ_U4     (var, ptr); \
+    else if ((((enc) & DW_EH_PE_SIZE_MASK) == DW_EH_PE_udata8 ) &&  ((enc) & DW_EH_PE_signed)) READ_S8     (var, ptr); \
+    else if ((((enc) & DW_EH_PE_SIZE_MASK) == DW_EH_PE_udata8 ) && !((enc) & DW_EH_PE_signed)) READ_U8     (var, ptr); \
+    else if ((((enc) & DW_EH_PE_SIZE_MASK) == DW_EH_PE_uptr   ) &&  ((enc) & DW_EH_PE_signed)) READ_S8     (var, ptr); \
+    else if ((((enc) & DW_EH_PE_SIZE_MASK) == DW_EH_PE_uptr   ) && !((enc) & DW_EH_PE_signed)) READ_U8     (var, ptr); \
+    else { printf_log(LOG_DEBUG, "Invalid encoding 0x%02X\n", (enc)); (var) = 0; } \
+    if (((enc) & DW_EH_PE_pcrel) && isptr) (var) += optr; \
+    if ((enc) & DW_EH_PE_indirect) (var) = *(uint64_t*)(var); \
+} while (0)
+#define SKIP_1(ptr) (ptr) += 1
+#define SKIP_2(ptr) (ptr) += 2
+#define SKIP_4(ptr) (ptr) += 4
+#define SKIP_8(ptr) (ptr) += 8
+#define SKIP_LEB128(ptr) do { ++(ptr); } while (((*(((unsigned char*)(ptr))-1))) & (1 << 7))
+#define SKIP_ENCODED(ptr, enc) \
+    do { if (((enc) & DW_EH_PE_SIZE_MASK) == DW_EH_PE_uleb128) SKIP_LEB128(ptr); \
+    else if (((enc) & DW_EH_PE_SIZE_MASK) == DW_EH_PE_udata2 ) SKIP_2     (ptr); \
+    else if (((enc) & DW_EH_PE_SIZE_MASK) == DW_EH_PE_udata4 ) SKIP_4     (ptr); \
+    else if (((enc) & DW_EH_PE_SIZE_MASK) == DW_EH_PE_udata8 ) SKIP_8     (ptr); \
+    else if (((enc) & DW_EH_PE_SIZE_MASK) == DW_EH_PE_uptr   ) SKIP_8     (ptr); \
+    else printf_log(LOG_DEBUG, "Invalid encoding 0x%02X\n", (enc)); \
+} while (0)
+
+// LEB128 wil only work on 64 bits numbers (unsigned long, like the Linux kernel)
+void readULEB(uint64_t *var, unsigned char **ptr) {
+    *var = 0;
+    int shift = 0;
+    while (1) {
+        unsigned char byte = **ptr; ++*ptr;
+        *var |= ((byte & 0x7F) << shift);
+        if (!(byte & (1 << 7))) break;
+        shift += 7;
+    }
+}
+void readSLEB(int64_t *var, unsigned char **ptr) {
+    uint64_t tmp = 0;
+    int shift = 0;
+    unsigned char byte;
+    while (1) {
+        byte = **ptr; ++*ptr;
+        tmp |= ((byte & 0x7F) << shift);
+        if (!(byte & (1 << 7))) break;
+        shift += 7;
+    }
+    // Sign extend
+    if (byte & (1 << 6)) {
+        *(uint64_t*)var = tmp | (0xFFFFFFFFFFFFFFFFull << (shift + 7));
+    } else {
+        *var = tmp;
+    }
+}
+
+uintptr_t get_parent_registers(dwarf_unwind_t *unwind, const elfheader_t *ehdr, uintptr_t addr, char *success) {
+    if (!ehdr) {
+        *success = 0;
+        return 0;
+    }
+    unsigned char ehfh_version = *(unsigned char*)ehdr->ehframehdr;
+    if (ehfh_version != 1) {
+        *success = 0;
+        return 0;
+    }
+
+    // printf_log(LOG_NONE, "Address: 0x%016lX\nEH frame address: 0x%016lX\n", addr, ehdr->ehframe);
+
+    // Not using the binary search table (for now)
+
+    unsigned char *cur_addr = (unsigned char*)ehdr->ehframe;
+    unsigned char *end_addr = (unsigned char*)ehdr->ehframe_end;
+
+#define AUG_EHDATA      (1 << 0)
+#define AUG_AUGDATA     (1 << 1)
+#define AUG_LSDA        (1 << 2)
+#define AUG_PARG        (1 << 3)
+#define AUG_FDEENC      (1 << 4)
+#define AUG_SIGHDLER    (1 << 5)
+    unsigned short aug_fields = 0;
+    unsigned char lsda_encoding = DW_EH_PE_omit;
+    unsigned char fde_encoding  = DW_EH_PE_omit;
+    uint64_t eh_data = 0;               // Note: these are never reset, so use aug_fields & AUG_EHDATA
+    uint64_t code_alignment_factor = 0;
+    int64_t data_alignment_factor = 0;
+    uint64_t cie_aug_data_len = 0;      // Note: these are never reset, so use aug_fields & AUG_AUGDATA
+    unsigned char *cie_aug_data = NULL; // Note: these are never reset, so use aug_fields & AUG_AUGDATA
+    unsigned char *initCIEinstr = NULL; // Never NULL if initialized
+    unsigned char *endCIEinstr = NULL;  // Never NULL if initialized
+    uint64_t return_addr_reg = 0;       // Description absent in the LSBCS 5.0, but still present
+    while (cur_addr < end_addr) {
+        // Length
+        uint64_t len; int extended = 0;
+        READ_U4(len, cur_addr);
+        if (!len) {
+            // Terminator, end parsing
+            *success = 0;
+            return 0;
+        } else if (len == 0xFFFFFFFF) {
+            // Extended Length (optional)
+            extended = 1;
+            READ_U8(len, cur_addr);
+        }
+        (void)extended;
+        // printf_log(LOG_NONE, "[???] Length          %02X\n", len);
+        // printf_log(LOG_NONE, "[???] Extended        %c\n", extended ? 'Y' : 'N');
+        unsigned char *next_addr = cur_addr + len;
+
+        // CIE ID (always 0) or CIE Pointer (never 0)
+        uint32_t cie_ptr;
+        READ_U4(cie_ptr, cur_addr);
+        if (cie_ptr == 0) {
+            // Current block is a CIE
+            // printf_log(LOG_NONE, "[CIE] ID              %02X\n", cie_ptr);
+
+            // Version (always 1 or 3)
+            uint8_t cie_version;
+            READ_U1(cie_version, cur_addr);
+            // printf_log(LOG_NONE, "[CIE] Version         %01X\n", cie_version);
+            if ((cie_version != 1) && (cie_version != 3)) { // Only 1 is in the LSBCS 5.0 but I found 3 exists as well
+                // Invalid CIE version
+                printf_log(LOG_DEBUG, "Invalid CIE version %d (should be 1 or 3), stopping parsing\n", cie_version);
+                // Failed to process CIE
+                *success = 0;
+                return 0;
+            }
+
+            // Augmentation String
+            char *aug_str = (char*)cur_addr; cur_addr += strlen(aug_str) + 1;
+            // printf_log(LOG_NONE, "[CIE] Augmentation st %s\n", aug_str);
+            aug_fields = (aug_str[0] == 'z') ? AUG_AUGDATA : 0;
+            for (char *aug_str2 = aug_str + ((aug_fields & AUG_AUGDATA) ? 1 : 0); *aug_str2; ++aug_str2) {
+                if ((aug_str2[0] == 'e') && (aug_str2[1] == 'h')) {
+                    // Use has also been left untold in the Linux Standard Base Core Specification 3.0RC1
+                    printf_log(LOG_DEBUG, "Warning: EH data detected but this was removed from the standard\n");
+                    aug_fields |= AUG_EHDATA; ++aug_str2;
+                } else if (aug_fields & AUG_AUGDATA) {
+                    if (aug_str2[0] == 'L') {
+                        aug_fields |= AUG_LSDA;
+                    } else if (aug_str2[0] == 'P') {
+                        aug_fields |= AUG_PARG;
+                        printf_log(LOG_DEBUG, "Warning: augmentation string attribute 'P' unsupported\n");
+                    } else if (aug_str2[0] == 'S') {
+                        aug_fields |= AUG_SIGHDLER;
+                        printf_log(LOG_DEBUG, "Warning: augmentation string attribute 'S' ignored\n");
+                    } else if (aug_str2[0] == 'R') {
+                        aug_fields |= AUG_FDEENC;
+                    } else {
+                        printf_log(LOG_DEBUG, "Error: invalid augmentation string %s\n", aug_str);
+                        // Failed to process augmentation string
+                        break;
+                    }
+                } else {
+                    printf_log(LOG_DEBUG, "Error: invalid augmentation string %s\n", aug_str);
+                    // Failed to process augmentation string
+                    break;
+                }
+            }
+            // printf_log(LOG_NONE, "[CIE] Augmentation fs %02x\n", aug_fields);
+
+            // Code Alignment Factor
+            READ_ULEB128(code_alignment_factor, cur_addr);
+            // printf_log(LOG_NONE, "[CIE] Code AF         %d\n", code_alignment_factor);
+            // Data Alignment Factor
+            READ_SLEB128(data_alignment_factor, cur_addr);
+            // printf_log(LOG_NONE, "[CIE] Data AF         %d\n", data_alignment_factor);
+
+            if (cie_version == 1) {
+                READ_U1(return_addr_reg, cur_addr);
+            } else {
+                READ_ULEB128(return_addr_reg, cur_addr);
+            }
+            // printf_log(LOG_NONE, "[CIE] Return addr reg %01X\n", return_addr_reg);
+
+            lsda_encoding = DW_EH_PE_omit;
+            fde_encoding  = DW_EH_PE_absptr;
+            if (aug_fields & AUG_AUGDATA) {
+                // Augmentation Data Length (optional)
+                READ_ULEB128(cie_aug_data_len, cur_addr);
+                // printf_log(LOG_NONE, "[CIE] Aug data len    %01X\n", cie_aug_data_len);
+                // Augmentation Data (optional)
+                cie_aug_data = cur_addr; cur_addr += cie_aug_data_len;
+                char *cie_aug_data2 = (char*)cie_aug_data;
+                for (char *aug_str2 = aug_str + ((aug_fields & AUG_AUGDATA) ? 1 : 0); *aug_str2; ++aug_str2) {
+                    if ((aug_str2[0] == 'e') && (aug_str2[1] == 'h')) {
+                        break; // Unknown usage
+                    } else if (aug_fields & AUG_AUGDATA) {
+                        if (aug_str2[0] == 'L') {
+                            READ_U1(lsda_encoding, cie_aug_data2);
+                            if (lsda_encoding != DW_EH_PE_omit) printf_log(LOG_DEBUG, "Warning: LSDA unsupported\n");
+                        } else if (aug_str2[0] == 'P') {
+                            unsigned char pencoding;
+                            READ_U1(pencoding, cie_aug_data2);
+                            SKIP_ENCODED(cie_aug_data2, pencoding);
+                        } else if (aug_str2[0] == 'S') {
+                        } else if (aug_str2[0] == 'R') {
+                            READ_U1(fde_encoding, cie_aug_data2);
+                            // printf_log(LOG_NONE, "[CIE] FDE encoding    %02X\n", fde_encoding);
+                            if (fde_encoding == DW_EH_PE_omit) printf_log(LOG_DEBUG, "Error: FDE encoding set to 'omit'\n");
+                        } else {
+                            break;
+                        }
+                    } else {
+                        break;
+                    }
+                }
+            }
+        
+            // Initial Instructions
+            initCIEinstr = cur_addr;
+            endCIEinstr = next_addr;
+        } else if (initCIEinstr) {
+            // Current block is a FDE
+            // printf_log(LOG_NONE, "[FDE] Ptr             %02X\n", cie_ptr);
+
+            // PC Begin
+            uintptr_t pc_begin;
+            READ_ENCODED(pc_begin, cur_addr, fde_encoding, 1);
+            // printf_log(LOG_NONE, "[FDE] PC begin        %016lX\n", pc_begin);
+            // PC Range
+            uint64_t pc_range;
+            READ_ENCODED(pc_range, cur_addr, fde_encoding, 0);
+            // printf_log(LOG_NONE, "[FDE] PC end          %016lX\n", pc_begin + pc_range);
+            
+            uint64_t aug_data_len;
+            unsigned char *aug_data;
+            if (aug_fields & AUG_AUGDATA) {
+                // Augmentation Data Length (optional)
+                READ_ULEB128(aug_data_len, cur_addr);
+                // printf_log(LOG_NONE, "[FDE] Aug data len    %01X\n", aug_data_len);
+                // Augmentation Data (optional)
+                aug_data = cur_addr; cur_addr += aug_data_len;
+                (void)aug_data;
+            }
+
+            if (lsda_encoding != DW_EH_PE_omit) {
+                SKIP_ENCODED(cur_addr, lsda_encoding);
+            }
+
+            // Call Frame Instructions
+            if ((pc_begin <= addr) && (addr < pc_begin + pc_range)) {
+                // Corresponding Frame Descriptor Entry found!
+
+#define setmax(a, b) a = (a < (b) ? (b) : a)
+                dwarf_unwind_constr_t unwind_constr;
+                unwind_constr.reg_count = return_addr_reg;
+                uint64_t maxstacksize = 0;
+                uint64_t curstacksize = 0;
+                unsigned char *cur_inst = initCIEinstr;
+                uint64_t nreg;
+#define PARSE_INST \
+        switch (inst >> 6) {                                        \
+        case 0b01:                                                  \
+            break;                                                  \
+        case 0b10:                                                  \
+            setmax(unwind_constr.reg_count, inst & 0x3F);           \
+            SKIP_LEB128(cur_inst);                                  \
+            break;                                                  \
+        case 0b11:                                                  \
+            setmax(unwind_constr.reg_count, inst & 0x3F);           \
+            break;                                                  \
+        case 0b00:                                                  \
+            switch (inst & 0x3F) {                                  \
+            case 0b000000:                                          \
+                break;                                              \
+            case 0b001010:                                          \
+                ++curstacksize;                                     \
+                setmax(maxstacksize, curstacksize);                 \
+                break;                                              \
+            case 0b001011:                                          \
+                if (!curstacksize) {                                \
+                    printf_log(LOG_DEBUG, "Negative stack size during CFI execution\n"); \
+                    *success = 0;                                   \
+                    return 0;                                       \
+                }                                                   \
+                --curstacksize;                                     \
+                break;                                              \
+            case 0b000010:                                          \
+                SKIP_1(cur_inst);                                   \
+                break;                                              \
+            case 0b000011:                                          \
+                SKIP_2(cur_inst);                                   \
+                break;                                              \
+            case 0b000100:                                          \
+                SKIP_4(cur_inst);                                   \
+                break;                                              \
+            case 0b000001:                                          \
+                SKIP_8(cur_inst);                                   \
+                break;                                              \
+            case 0b001110:                                          \
+            case 0b010011:                                          \
+                SKIP_LEB128(cur_inst);                              \
+                break;                                              \
+            case 0b000110:                                          \
+            case 0b000111:                                          \
+            case 0b001000:                                          \
+            case 0b001101:                                          \
+                READ_ULEB128(nreg, cur_inst);                       \
+                setmax(unwind_constr.reg_count, nreg);              \
+                break;                                              \
+            case 0b000101:                                          \
+            case 0b001100:                                          \
+            case 0b010001:                                          \
+            case 0b010010:                                          \
+            case 0b010100:                                          \
+            case 0b010101:                                          \
+                READ_ULEB128(nreg, cur_inst);                       \
+                setmax(unwind_constr.reg_count, nreg);              \
+                SKIP_LEB128(cur_inst);                              \
+                break;                                              \
+            case 0b001001:                                          \
+                READ_ULEB128(nreg, cur_inst);                       \
+                setmax(unwind_constr.reg_count, nreg);              \
+                READ_ULEB128(nreg, cur_inst);                       \
+                setmax(unwind_constr.reg_count, nreg);              \
+                break;                                              \
+            default:                                                \
+                /* Contains undefined, user and expression CFIs */  \
+                printf_log(LOG_DEBUG, "Unknown CFI 0x%02X\n", inst); \
+                *success = 0;                                       \
+                return 0;                                        \
+            }                                                       \
+            break;                                                  \
+        }
+                while (cur_inst < endCIEinstr) {
+                    unsigned char inst; READ_U1(inst, cur_inst);
+                    PARSE_INST
+                }
+                cur_inst = cur_addr;
+                while (cur_inst < next_addr) {
+                    unsigned char inst; READ_U1(inst, cur_inst);
+                    PARSE_INST
+                }
+#undef PARSE_INST
+
+                uint64_t cfa_reg = -1;
+                unsigned char cfa_signed = 0;
+                union { uint64_t uoff; int64_t soff; } cfa_offset = { .uoff = 0 };
+                ++unwind_constr.reg_count;
+                size_t tablelen = unwind_constr.reg_count * sizeof(uint64_t);
+                size_t statuseslen = ((unwind_constr.reg_count+1) >> 1) * sizeof(uint8_t);
+                unwind_constr.table = (uint64_t*)malloc(tablelen);
+                unwind_constr.statuses = (uint8_t*)calloc((unwind_constr.reg_count+1) >> 1, sizeof(uint8_t));
+                // ~~undefined is 0: no initialization needed~~ still initialize the first 17 to same_val
+                for (int i = 0; (i < 17) && (i <= unwind_constr.reg_count); ++i) {
+                    SET_STATUS(unwind_constr, i, REGSTATUS_same_val);
+                }
+
+                curstacksize = 0;
+                uint64_t **table_stack = (uint64_t**)malloc(maxstacksize * sizeof(uint64_t*));
+                for (uint64_t i = 0; i < maxstacksize; ++i) {
+                    table_stack[i] = (uint64_t*)malloc(tablelen);
+                }
+                cur_inst = initCIEinstr;
+                uintptr_t cur_pointed_addr = pc_begin;
+                // Missing:
+                /* DW_CFA_def_cfa_expression    0   0x0f      BLOCK                              */
+                /* DW_CFA_expression            0   0x10      ULEB128 register  BLOCK            */
+                /* DW_CFA_val_expression        0   0x16      ULEB128 BLOCK                      */
+                /* DW_CFA_lo_user               0   0x1c                                         */
+                /* DW_CFA_GNU_args_size         0   0x2e      ULEB128 argsize                    */
+                /* DW_CFA_GNU_negative_offset_extended 0 0x2f ULEB128 register  ULEB128 offset (obsoleted by DW_CFA_offset_extended_sf) */
+                /* DW_CFA_hi_user               0   0x3f                                         */
+                // Known "bug": DW_CFA_set_loc can go backwards, this is ignored
+                uint64_t tmpreg;
+                uint64_t tmpuval;
+                int64_t tmpsval;
+#define PARSE_INST \
+        switch (inst >> 6) {                                                                                    \
+        case 0b01:         /* DW_CFA_advance_loc           0x1 delta                                        */  \
+            cur_pointed_addr += (inst & 0x3F) * code_alignment_factor;                                          \
+            break;                                                                                              \
+        case 0b10:         /* DW_CFA_offset                0x2 register  ULEB128 offset                     */  \
+            READ_ULEB128(tmpuval, cur_inst);                                                                    \
+            SET_STATUS(unwind_constr, inst & 0x3F, REGSTATUS_offset);                                           \
+            unwind_constr.table[inst & 0x3F] = (int64_t)tmpuval * data_alignment_factor;                        \
+            break;                                                                                              \
+        case 0b11:         /* DW_CFA_restore               0x3 register                                     */  \
+            RESTORE_REG(inst & 0x3F)                                                                            \
+            break;                                                                                              \
+        case 0b00:                                                                                              \
+            switch (inst & 0x3F) {                                                                              \
+            case 0b000001: /* DW_CFA_set_loc               0   0x01      address                            */  \
+                READ_U8(cur_pointed_addr, cur_inst);                                                            \
+                break;                                                                                          \
+            case 0b000010: /* DW_CFA_advance_loc1          0   0x02      1-byte delta                       */  \
+                READ_U1(tmpuval, cur_inst);                                                                     \
+                cur_pointed_addr += tmpuval * code_alignment_factor;                                            \
+                break;                                                                                          \
+            case 0b000011: /* DW_CFA_advance_loc2          0   0x03      2-byte delta                       */  \
+                READ_U2(tmpuval, cur_inst);                                                                     \
+                cur_pointed_addr += tmpuval * code_alignment_factor;                                            \
+                break;                                                                                          \
+            case 0b000100: /* DW_CFA_advance_loc4          0   0x04      4-byte delta                       */  \
+                READ_U4(tmpuval, cur_inst);                                                                     \
+                cur_pointed_addr += tmpuval * code_alignment_factor;                                            \
+                break;                                                                                          \
+            \
+            case 0b001100: /* DW_CFA_def_cfa               0   0x0c      ULEB128 register  ULEB128 offset   */  \
+                READ_ULEB128(cfa_reg, cur_inst);                                                                \
+                cfa_signed = 0;                                                                                 \
+                READ_ULEB128(cfa_offset.uoff, cur_inst);                                                        \
+                break;                                                                                          \
+            case 0b010010: /* DW_CFA_def_cfa_sf            0   0x12      ULEB128 register  SLEB128 offset   */  \
+                READ_ULEB128(cfa_reg, cur_inst);                                                                \
+                cfa_signed = 1;                                                                                 \
+                READ_SLEB128(cfa_offset.soff, cur_inst);                                                        \
+                cfa_offset.soff *= data_alignment_factor;                                                       \
+                break;                                                                                          \
+            case 0b001101: /* DW_CFA_def_cfa_register      0   0x0d      ULEB128 register                   */  \
+                READ_ULEB128(cfa_reg, cur_inst);                                                                \
+                break;                                                                                          \
+            case 0b001110: /* DW_CFA_def_cfa_offset        0   0x0e      ULEB128 offset                     */  \
+                cfa_signed = 0;                                                                                 \
+                READ_ULEB128(cfa_offset.uoff, cur_inst);                                                        \
+                break;                                                                                          \
+            case 0b010011: /* DW_CFA_def_cfa_offset_sf     0   0x13      SLEB128 offset                     */  \
+                READ_SLEB128(cfa_offset.soff, cur_inst);                                                        \
+                cfa_offset.soff *= data_alignment_factor;                                                       \
+                break;                                                                                          \
+            /* DW_CFA_def_cfa_expression */ \
+            \
+            case 0b000111: /* DW_CFA_undefined             0   0x07      ULEB128 register                   */  \
+                READ_ULEB128(tmpreg, cur_inst);                                                                 \
+                SET_STATUS(unwind_constr, tmpreg, REGSTATUS_undefined);                                         \
+                break;                                                                                          \
+            case 0b001000: /* DW_CFA_same_value            0   0x08      ULEB128 register                   */  \
+                READ_ULEB128(tmpreg, cur_inst);                                                                 \
+                SET_STATUS(unwind_constr, tmpreg, REGSTATUS_same_val);                                          \
+                break;                                                                                          \
+            case 0b000101: /* DW_CFA_offset_extended       0   0x05      ULEB128 register  ULEB128 offset   */  \
+                READ_ULEB128(tmpreg, cur_inst);                                                                 \
+                READ_ULEB128(tmpuval, cur_inst);                                                                \
+                SET_STATUS(unwind_constr, tmpreg, REGSTATUS_offset);                                            \
+                unwind_constr.table[tmpreg] = (int64_t)tmpuval * data_alignment_factor;                         \
+                break;                                                                                          \
+            case 0b010001: /* DW_CFA_offset_extended_sf    0   0x11      ULEB128 register  SLEB128 offset   */  \
+                READ_ULEB128(tmpreg, cur_inst);                                                                 \
+                READ_SLEB128(tmpsval, cur_inst);                                                                \
+                SET_STATUS(unwind_constr, tmpreg, REGSTATUS_offset);                                            \
+                unwind_constr.table[tmpreg] = tmpsval * data_alignment_factor;                                  \
+                break;                                                                                          \
+            case 0b010100: /* DW_CFA_val_offset            0   0x14      ULEB128 register  ULEB128 offset   */  \
+                READ_ULEB128(tmpreg, cur_inst);                                                                 \
+                READ_ULEB128(tmpuval, cur_inst);                                                                \
+                SET_STATUS(unwind_constr, tmpreg, REGSTATUS_val_offset);                                        \
+                unwind_constr.table[tmpreg] = (int64_t)tmpuval * data_alignment_factor;                         \
+                break;                                                                                          \
+            case 0b010101: /* DW_CFA_val_offset_sf         0   0x15      ULEB128 register  SLEB128 offset   */  \
+                READ_ULEB128(tmpreg, cur_inst);                                                                 \
+                READ_SLEB128(tmpsval, cur_inst);                                                                \
+                SET_STATUS(unwind_constr, tmpreg, REGSTATUS_val_offset);                                        \
+                unwind_constr.table[tmpreg] = tmpsval * data_alignment_factor;                                  \
+                break;                                                                                          \
+            case 0b001001: /* DW_CFA_register              0   0x09      ULEB128 register  ULEB128 register */  \
+                READ_ULEB128(tmpreg, cur_inst);                                                                 \
+                READ_ULEB128(tmpuval, cur_inst);                                                                \
+                SET_STATUS(unwind_constr, tmpreg, REGSTATUS_register);                                          \
+                unwind_constr.table[tmpreg] = tmpuval;                                                          \
+                break;                                                                                          \
+            /* DW_CFA_expression */ \
+            /* DW_CFA_val_expression */ \
+            case 0b000110: /* DW_CFA_restore_extended      0   0x06      ULEB128 register                   */  \
+                READ_ULEB128(tmpreg, cur_inst);                                                                 \
+                RESTORE_REG(tmpreg)                                                                             \
+                break;                                                                                          \
+            \
+            case 0b001010: /* DW_CFA_remember_state        0   0x0a                                         */  \
+                memcpy(table_stack[curstacksize], unwind_constr.table, tablelen);                               \
+                ++curstacksize;                                                                                 \
+                break;                                                                                          \
+            case 0b001011: /* DW_CFA_restore_state         0   0x0b                                         */  \
+                --curstacksize;                                                                                 \
+                memcpy(unwind_constr.table, table_stack[curstacksize], tablelen);                               \
+                break;                                                                                          \
+            \
+            case 0b000000: /* DW_CFA_nop                   0   0                                            */  \
+                break;                                                                                          \
+            default:                                                                                            \
+                /* Contains undefined, user and expression CFIs */                                              \
+                printf_log(LOG_DEBUG, "Unknown CFI 0x%02X\n", inst);                                             \
+                FAILED                                                                                          \
+            }                                                                                                   \
+            break;                                                                                              \
+        }
+#define RESTORE_REG(reg) \
+    printf_log(LOG_DEBUG, "Trying to restore register 0x%02lX while in the intial CFIs\n", (uint64_t)reg); \
+    FAILED
+#define FAILED \
+    free(unwind_constr.statuses);                                       \
+    free(unwind_constr.table);                                          \
+    for (uint64_t i = 0; i < maxstacksize; ++i) free(table_stack[i]);   \
+    free(table_stack);                                                  \
+    *success = 0; return 0;
+                while (cur_inst < endCIEinstr) {
+                    unsigned char inst; READ_U1(inst, cur_inst);
+                    // printf_log(LOG_NONE, "Executing pre 0x%02X\n", inst);
+                    PARSE_INST
+                }
+#undef FAILED
+#undef RESTORE_REG
+                uint64_t *init_table = (uint64_t*)malloc(tablelen);
+                memcpy(init_table, unwind_constr.table, tablelen);
+                uint8_t *init_statuses = (uint8_t*)malloc(statuseslen);
+                memcpy(init_statuses, unwind_constr.statuses, statuseslen);
+                cur_inst = cur_addr;
+#define RESTORE_REG(reg) \
+    unwind_constr.table[reg] = init_table[reg]; \
+    SET_STATUS(unwind_constr, (reg), ((init_statuses[(reg) >> 1] >> (((reg) & 1) << 2)) & 0xF));
+#define FAILED \
+    free(init_statuses);                                                \
+    free(init_table);                                                   \
+    free(unwind_constr.statuses);                                       \
+    free(unwind_constr.table);                                          \
+    for (uint64_t i = 0; i < maxstacksize; ++i) free(table_stack[i]);   \
+    free(table_stack);                                                  \
+    *success = 0; return 0;
+                while ((cur_inst < next_addr) && (cur_pointed_addr <= addr)) {
+                    unsigned char inst; READ_U1(inst, cur_inst);
+                    // printf_log(LOG_NONE, "Executing post 0x%02X\n", inst);
+                    PARSE_INST
+                }
+#undef FAILED
+#undef RESTORE_REG
+#undef PARSE_INST
+                free(init_statuses);
+                free(init_table);
+                for (int i = 0; i < maxstacksize; ++i) {
+                    free(table_stack[i]);
+                }
+                free(table_stack);
+
+                dwarf_unwind_t new_unwind;
+                new_unwind.reg_count = unwind_constr.reg_count;
+                new_unwind.regs = calloc(unwind_constr.reg_count, sizeof(uint64_t));
+                uintptr_t cfa = unwind->regs[cfa_reg];
+                if (cfa_signed) cfa += cfa_offset.soff;
+                else            cfa += cfa_offset.uoff;
+                
+                // printf_log(LOG_NONE, "Done, rewriting registers (CFA at r%d(0x%016lX) + %lld (%c) -> 0x%016lX\n", cfa_reg, unwind->regs[cfa_reg], cfa_offset.soff, cfa_signed ? 's' : 'u', cfa);
+                for (uint64_t i = 0; i < unwind_constr.reg_count; ++i) {
+                    switch (GET_STATUS(unwind_constr, i)) {
+                    case REGSTATUS_undefined:
+                        // printf_log(LOG_NONE, "Register %02lX: (undefined)\n", i);
+                        break;
+                    case REGSTATUS_same_val:
+                        if (i >= unwind->reg_count) {
+                            printf_log(LOG_DEBUG, "Invalid register status (value copied from register 0x%02lX)\n", i);
+                            free(unwind_constr.statuses);
+                            free(unwind_constr.table);
+                            free(new_unwind.regs);
+                            *success = 0;
+                            return 0;
+                        }
+                        new_unwind.regs[i] = unwind->regs[i];
+                        // printf_log(LOG_NONE, "Register %02lX: copy   %016lX\n", i, new_unwind.regs[i]);
+                        break;
+                    case REGSTATUS_offset:
+                        new_unwind.regs[i] = *(uint64_t*)(cfa + (int64_t)unwind_constr.table[i]);
+                        // printf_log(LOG_NONE, "Register %02lX: offset %016lX [%016lX + %lld]\n", i, new_unwind.regs[i], cfa, (int64_t)unwind_constr.table[i]);
+                        break;
+                    case REGSTATUS_val_offset:
+                        new_unwind.regs[i] = (uint64_t)(cfa + (int64_t)unwind_constr.table[i]);
+                        // printf_log(LOG_NONE, "Register %02lX: voff   %016lX\n", i, new_unwind.regs[i]);
+                        break;
+                    case REGSTATUS_register:
+                        if (unwind_constr.table[i] >= unwind->reg_count) {
+                            printf_log(LOG_DEBUG, "Invalid register status (value copied from register 0x%02lX)\n", unwind_constr.table[i]);
+                            free(unwind_constr.statuses);
+                            free(unwind_constr.table);
+                            free(new_unwind.regs);
+                            *success = 0;
+                            return 0;
+                        }
+                        new_unwind.regs[i] = unwind->regs[unwind_constr.table[i]];
+                        // printf_log(LOG_NONE, "Register %02lX: reg    %016lX\n", i, new_unwind.regs[i]);
+                        break;
+                    }
+                }
+                *success = (GET_STATUS(unwind_constr, return_addr_reg) == REGSTATUS_undefined) ? 0 : ((aug_fields & AUG_SIGHDLER) ? 2 : 1);
+                free(unwind_constr.statuses);
+                free(unwind_constr.table);
+                
+                free(unwind->regs);
+                unwind->reg_count = new_unwind.reg_count;
+                unwind->regs = new_unwind.regs;
+
+                // Maybe?
+                unwind->regs[cfa_reg] = cfa;
+
+                // printf_log(LOG_NONE, "Returning %016lX\n", unwind->regs[return_addr_reg]);
+                return unwind->regs[return_addr_reg];
+            }
+        } else {
+            // printf_log(LOG_NONE, "[FDE] Ptr             %02X\n", cie_ptr);
+            printf_log(LOG_DEBUG, "Unexpected FDE, corresponding CIE missing\n");
+            return 0;
+        }
+
+        cur_addr = next_addr;
+    }
+    *success = 0;
+    return 0;
+}
+
+dwarf_unwind_t *init_dwarf_unwind_registers(x64emu_t *emu) {
+    dwarf_unwind_t *unwind_struct = (dwarf_unwind_t*)malloc(sizeof(dwarf_unwind_t));
+    unwind_struct->reg_count = 17;
+    unwind_struct->regs = (uint64_t*)malloc(17*sizeof(uint64_t));
+    /* x86_64-abi-0.99.pdf
+     * Register Name                    | Number | Abbreviation
+     * General Purpose Register RAX     | 0      | %rax
+     * General Purpose Register RDX     | 1      | %rdx
+     * General Purpose Register RCX     | 2      | %rcx
+     * General Purpose Register RBX     | 3      | %rbx
+     * General Purpose Register RSI     | 4      | %rsi
+     * General Purpose Register RDI     | 5      | %rdi
+     * Frame Pointer Register   RBP     | 6      | %rbp
+     * Stack Pointer Register   RSP     | 7      | %rsp
+     * Extended Integer Registers 8-15  | 8-15   | %r8-%r15
+     * Return Address RA                | 16     |
+     * Vector Registers 0-7             | 17-24  | %xmm0-%xmm7
+     * Extended Vector Registers 8-15   | 25-32  | %xmm8-%xmm15
+     * Floating Point Registers 0-7     | 33-40  | %st0-%st7
+     * MMX Registers 0-7                | 41-48  | %mm0-%mm7
+     * Flag Register                    | 49     | %rFLAGS
+     * Segment Register ES              | 50     | %es
+     * Segment Register CS              | 51     | %cs
+     * Segment Register SS              | 52     | %ss
+     * Segment Register DS              | 53     | %ds
+     * Segment Register FS              | 54     | %fs
+     * Segment Register GS              | 55     | %gs
+     * Reserved                         | 56-57  |
+     * FS Base address                  | 58     | %fs.base
+     * GS Base address                  | 59     | %gs.base
+     * Reserved                         | 60-61  |
+     * Task Register                    | 62     | %tr
+     * LDT Register                     | 63     | %ldtr
+     * 128-bit Media Control and Status | 64     | %mxcsr
+     * x87 Control Word                 | 65     | %fcw
+     * x87 Status Word                  | 66     | %fsw
+     */
+    unwind_struct->regs[ 0] = emu->regs[_RAX].q[0];
+    unwind_struct->regs[ 1] = emu->regs[_RDX].q[0];
+    unwind_struct->regs[ 2] = emu->regs[_RCX].q[0];
+    unwind_struct->regs[ 3] = emu->regs[_RBX].q[0];
+    unwind_struct->regs[ 4] = emu->regs[_RSI].q[0];
+    unwind_struct->regs[ 5] = emu->regs[_RDI].q[0];
+    unwind_struct->regs[ 6] = emu->regs[_RBP].q[0];
+    unwind_struct->regs[ 7] = emu->regs[_RSP].q[0] + 8;
+    unwind_struct->regs[ 8] = emu->regs[_R8 ].q[0];
+    unwind_struct->regs[ 9] = emu->regs[_R9 ].q[0];
+    unwind_struct->regs[10] = emu->regs[_R10].q[0];
+    unwind_struct->regs[11] = emu->regs[_R11].q[0];
+    unwind_struct->regs[12] = emu->regs[_R12].q[0];
+    unwind_struct->regs[13] = emu->regs[_R13].q[0];
+    unwind_struct->regs[14] = emu->regs[_R14].q[0];
+    unwind_struct->regs[15] = emu->regs[_R15].q[0];
+    unwind_struct->regs[16] = emu->ip.q[0];
+    return unwind_struct;
+}
+
+void free_dwarf_unwind_registers(dwarf_unwind_t **unwind_struct) {
+    free((*unwind_struct)->regs);
+    free(*unwind_struct);
+    *unwind_struct = NULL;
+}

src/elfs/elfdwarf_private.h

@@ -0,0 +1,19 @@
+#ifndef __ELFDWARF_PRIVATE_H_
+#define __ELFDWARF_PRIVATE_H_
+
+#include "emu/x64emu_private.h"
+
+typedef struct dwarf_unwind_s {
+    uint8_t reg_count;
+    uint64_t *regs;
+} dwarf_unwind_t;
+typedef struct elfheader_s elfheader_t;
+
+dwarf_unwind_t *init_dwarf_unwind_registers(x64emu_t *emu);
+void free_dwarf_unwind_registers(dwarf_unwind_t **unwind_struct);
+
+// Returns the callee's address on success or NULL on failure (may be NULL regardless).
+// If success equals 2, the frame is a signal frame.
+uintptr_t get_parent_registers(dwarf_unwind_t *emu, const elfheader_t *ehdr, uintptr_t addr, char *success);
+
+#endif // __ELFDWARF_PRIVATE_H_

src/elfs/elfloader.c

@@ -134,9 +134,9 @@ int CalcLoadAddr(elfheader_t* head)
                     head->tlssize++;
         }
     }
-    printf_log(LOG_DEBUG, "Elf Addr(v/p)=%p/%p Memsize=0x%lx (align=0x%zx)\n", (void*)head->vaddr, (void*)head->paddr, head->memsz, head->align);
-    printf_log(LOG_DEBUG, "Elf Stack Memsize=%lu (align=%zu)\n", head->stacksz, head->stackalign);
-    printf_log(LOG_DEBUG, "Elf TLS Memsize=%lu (align=%zu)\n", head->tlssize, head->tlsalign);
+    printf_log(LOG_DEBUG, "Elf Addr(v/p)=%p/%p Memsize=0x%zx (align=0x%zx)\n", (void*)head->vaddr, (void*)head->paddr, head->memsz, head->align);
+    printf_log(LOG_DEBUG, "Elf Stack Memsize=%zu (align=%zu)\n", head->stacksz, head->stackalign);
+    printf_log(LOG_DEBUG, "Elf TLS Memsize=%zu (align=%zu)\n", head->tlssize, head->tlsalign);
 
     return 0;
 }
@@ -147,13 +147,19 @@ const char* ElfName(elfheader_t* head)
         return "(noelf)";
     return head->name;
 }
+const char* ElfPath(elfheader_t* head)
+{
+    if(!head)
+        return NULL;
+    return head->path;
+}
 int AllocElfMemory(box64context_t* context, elfheader_t* head, int mainbin)
 {
     uintptr_t offs = 0;
     if(mainbin && head->vaddr==0) {
         char* load_addr = getenv("BOX64_LOAD_ADDR");
         if(load_addr)
-            if(sscanf(load_addr, "0x%lx", &offs)!=1)
+            if(sscanf(load_addr, "0x%zx", &offs)!=1)
                 offs = 0;
     }
     if(!offs)
@@ -210,12 +216,12 @@ int AllocElfMemory(box64context_t* context, elfheader_t* head, int mainbin)
                 , MAP_PRIVATE | MAP_ANONYMOUS /*| ((wine_preloaded)?MAP_FIXED:0)*/
                 , -1, 0);
             if(p==MAP_FAILED) {
-                printf_log(LOG_NONE, "Cannot create memory map (@%p 0x%lx/0x%zx) for elf \"%s\"\n", (void*)head->multiblock_offs[i], head->multiblock_size[i], head->align, head->name);
+                printf_log(LOG_NONE, "Cannot create memory map (@%p 0x%zx/0x%zx) for elf \"%s\"\n", (void*)head->multiblock_offs[i], head->multiblock_size[i], head->align, head->name);
                 return 1;
             }
             if(head->multiblock_offs[i] &&( p!=(void*)head->multiblock_offs[i])) {
                 if((head->e_type!=ET_DYN)) {
-                    printf_log(LOG_NONE, "Error, memory map (@%p 0x%lx/0x%zx) for elf \"%s\" allocated @%p\n", (void*)head->multiblock_offs[i], head->multiblock_size[i], head->align, head->name, p);
+                    printf_log(LOG_NONE, "Error, memory map (@%p 0x%zx/0x%zx) for elf \"%s\" allocated @%p\n", (void*)head->multiblock_offs[i], head->multiblock_size[i], head->align, head->name, p);
                     return 1;
                 } else {
                     printf_log(LOG_INFO, "Allocated memory is not at hinted %p but %p (size %lx) \"%s\"\n", (void*)head->multiblock_offs[i], p, head->multiblock_size[i], head->name);
@@ -239,17 +245,17 @@ int AllocElfMemory(box64context_t* context, elfheader_t* head, int mainbin)
         // vaddr is 0, load everything has a One block
         if(!offs && box64_wine)
             offs = (uintptr_t)find47bitBlock(head->memsz); // limit to 47bits...
-        printf_log(LOG_DEBUG, "Allocating 0x%lx memory @%p for Elf \"%s\"\n", head->memsz, (void*)offs, head->name);
+        printf_log(LOG_DEBUG, "Allocating 0x%zx memory @%p for Elf \"%s\"\n", head->memsz, (void*)offs, head->name);
         void* p = mmap((void*)offs, head->memsz
             , PROT_READ | PROT_WRITE | PROT_EXEC
             , MAP_PRIVATE | MAP_ANONYMOUS /*| (((offs&&wine_preloaded)?MAP_FIXED:0))*/
             , -1, 0);
         if(p==MAP_FAILED) {
-            printf_log(LOG_NONE, "Cannot create memory map (@%p 0x%lx/0x%zx) for elf \"%s\"\n", (void*)offs, head->memsz, head->align, head->name);
+            printf_log(LOG_NONE, "Cannot create memory map (@%p 0x%zx/0x%zx) for elf \"%s\"\n", (void*)offs, head->memsz, head->align, head->name);
             return 1;
         }
         if(offs && (p!=(void*)offs) && (head->e_type!=ET_DYN)) {
-            printf_log(LOG_NONE, "Error, memory map (@%p 0x%lx/0x%zx) for elf \"%s\" allocated @%p\n", (void*)offs, head->memsz, head->align, head->name, p);
+            printf_log(LOG_NONE, "Error, memory map (@%p 0x%zx/0x%zx) for elf \"%s\" allocated @%p\n", (void*)offs, head->memsz, head->align, head->name, p);
             return 1;
         }
         setProtection((uintptr_t)p, head->memsz, PROT_READ | PROT_WRITE | PROT_EXEC);
@@ -289,7 +295,7 @@ int LoadElfMemory(FILE* f, box64context_t* context, elfheader_t* head)
         if(head->PHEntries[i].p_type == PT_LOAD) {
             Elf64_Phdr * e = &head->PHEntries[i];
             char* dest = (char*)e->p_paddr + head->delta;
-            printf_log(LOG_DEBUG, "MMap block #%zu @%p offset=%p (0x%lx/0x%lx)\n", i, dest, (void*)e->p_offset, e->p_filesz, e->p_memsz);
+            printf_log(LOG_DEBUG, "MMap block #%zu @%p offset=%p (0x%zx/0x%zx)\n", i, dest, (void*)e->p_offset, e->p_filesz, e->p_memsz);
             void* p = (void*)-1;
             if(e->p_memsz==e->p_filesz && !(e->p_align&0xfff)) {
                 printf_log(LOG_DEBUG, "MMap block #%zu @%p offset=%p (0x%zx/0x%zx, flags:0x%x)\n", i, dest, (void*)e->p_offset, e->p_filesz, e->p_memsz, e->p_flags);
@@ -319,11 +325,11 @@ int LoadElfMemory(FILE* f, box64context_t* context, elfheader_t* head)
         if(head->PHEntries[i].p_type == PT_TLS) {
             Elf64_Phdr * e = &head->PHEntries[i];
             char* dest = (char*)(context->tlsdata+context->tlssize+head->tlsbase);
-            printf_log(LOG_DEBUG, "Loading TLS block #%zu @%p (0x%lx/0x%lx)\n", i, dest, e->p_filesz, e->p_memsz);
+            printf_log(LOG_DEBUG, "Loading TLS block #%zu @%p (0x%zx/0x%zx)\n", i, dest, e->p_filesz, e->p_memsz);
             if(e->p_filesz) {
                 fseeko64(f, e->p_offset, SEEK_SET);
                 if(fread(dest, e->p_filesz, 1, f)!=1) {
-                    printf_log(LOG_NONE, "Fail to read PT_TLS part #%zu (size=%ld)\n", i, e->p_filesz);
+                    printf_log(LOG_NONE, "Fail to read PT_TLS part #%zu (size=%zd)\n", i, e->p_filesz);
                     return 1;
                 }
             }
@@ -343,7 +349,7 @@ int ReloadElfMemory(FILE* f, box64context_t* context, elfheader_t* head)
         if(head->PHEntries[i].p_type == PT_LOAD) {
             Elf64_Phdr * e = &head->PHEntries[i];
             char* dest = (char*)e->p_paddr + head->delta;
-            printf_log(LOG_DEBUG, "Re-loading block #%zu @%p (0x%lx/0x%lx)\n", i, dest, e->p_filesz, e->p_memsz);
+            printf_log(LOG_DEBUG, "Re-loading block #%zu @%p (0x%zx/0x%zx)\n", i, dest, e->p_filesz, e->p_memsz);
             int ret = fseeko64(f, e->p_offset, SEEK_SET);
             if(ret==-1) {printf_log(LOG_NONE, "Fail to (re)seek PT_LOAD part #%zu (offset=%ld, errno=%d/%s)\n", i, e->p_offset, errno, strerror(errno)); return 1;}
             #ifdef DYNAREC
@@ -354,7 +360,7 @@ int ReloadElfMemory(FILE* f, box64context_t* context, elfheader_t* head)
             if(e->p_filesz) {
                 ssize_t r = -1;
                 if((r=fread(dest, e->p_filesz, 1, f))!=1) {
-                    printf_log(LOG_NONE, "Fail to (re)read PT_LOAD part #%zu (dest=%p, size=%ld, return=%ld, feof=%d/ferror=%d/%s)\n", i, dest, e->p_filesz, r, feof(f), ferror(f), strerror(ferror(f)));
+                    printf_log(LOG_NONE, "Fail to (re)read PT_LOAD part #%zu (dest=%p, size=%zd, return=%zd, feof=%d/ferror=%d/%s)\n", i, dest, e->p_filesz, r, feof(f), ferror(f), strerror(ferror(f)));
                     return 1;
                 }
             }
@@ -371,22 +377,22 @@ int FindR64COPYRel(elfheader_t* h, const char* name, uintptr_t *offs, uint64_t**
 {
     if(!h)
         return 0;
-    Elf64_Rela * rel = (Elf64_Rela *)(h->rela + h->delta);
+    Elf64_Rela * rela = (Elf64_Rela *)(h->rela + h->delta);
     if(!h->rela)
         return 0;
     int cnt = h->relasz / h->relaent;
     for (int i=0; i<cnt; ++i) {
-        int t = ELF64_R_TYPE(rel[i].r_info);
-        Elf64_Sym *sym = &h->DynSym[ELF64_R_SYM(rel[i].r_info)];
+        int t = ELF64_R_TYPE(rela[i].r_info);
+        Elf64_Sym *sym = &h->DynSym[ELF64_R_SYM(rela[i].r_info)];
         const char* symname = SymName(h, sym);
         if(t==R_X86_64_COPY && symname && !strcmp(symname, name)) {
-            int version2 = h->VerSym?((Elf64_Half*)((uintptr_t)h->VerSym+h->delta))[ELF64_R_SYM(rel[i].r_info)]:-1;
+            int version2 = h->VerSym?((Elf64_Half*)((uintptr_t)h->VerSym+h->delta))[ELF64_R_SYM(rela[i].r_info)]:-1;
             if(version2!=-1) version2 &= 0x7fff;
             if(version && !version2) version2=-1;   // match a versionned symbol against a global "local" symbol
             const char* vername2 = GetSymbolVersion(h, version2);
             if(SameVersionnedSymbol(name, version, vername, symname, version2, vername2)) {
                 *offs = sym->st_value + h->delta;
-                *p = (uint64_t*)(rel[i].r_offset + h->delta + rel[i].r_addend);
+                *p = (uint64_t*)(rela[i].r_offset + h->delta + rela[i].r_addend);
                 return 1;
             }
         }
@@ -466,7 +472,7 @@ int RelocateElfREL(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t*
                     }
                     if (!offs) {
                         if(strcmp(symname, "__gmon_start__") && strcmp(symname, "data_start") && strcmp(symname, "__data_start"))
-                            printf_log(LOG_NONE, "Error: Global Symbol %s not found, cannot apply R_X86_64_GLOB_DAT @%p (%p) in %s\n", symname, p, *(void**)p, head->name);
+                            printf_log(LOG_NONE, "%s: Global Symbol %s not found, cannot apply R_X86_64_GLOB_DAT @%p (%p) in %s\n", (bind==STB_WEAK)?"Warning":"Error", symname, p, *(void**)p, head->name);
                     } else {
                         printf_dump(LOG_NEVER, "Apply %s R_X86_64_GLOB_DAT @%p (%p -> %p) on sym=%s (ver=%d/%s)\n", (bind==STB_LOCAL)?"Local":"Global", p, (void*)(p?(*p):0), (void*)offs, symname, version, vername?vername:"(none)");
                         *p = offs;
@@ -604,7 +610,7 @@ int RelocateElfRELA(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t
         uint64_t* globp;
         uintptr_t tmp = 0;
         intptr_t delta;
-        switch(ELF64_R_TYPE(rela[i].r_info)) {
+        switch(t) {
             case R_X86_64_NONE:
                 break;
             case R_X86_64_PC32:
@@ -639,8 +645,8 @@ int RelocateElfRELA(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t
                     // set global offs / size for the symbol
                     offs = sym->st_value + head->delta;
                     end = offs + sym->st_size;
-                    if(sym->st_size) {
-                        printf_dump(LOG_NEVER, "Apply %s R_X86_64_GLOB_DAT with R_X86_64_COPY @%p/%p (%p/%p -> %p/%p) size=%ld on sym=%s (ver=%d/%s) \n", 
+                    if(sym->st_size && offs) {
+                        printf_dump(LOG_NEVER, "Apply %s R_X86_64_GLOB_DAT with R_X86_64_COPY @%p/%p (%p/%p -> %p/%p) size=%zd on sym=%s (ver=%d/%s) \n", 
                             (bind==STB_LOCAL)?"Local":"Global", p, globp, (void*)(p?(*p):0), 
                             (void*)(globp?(*globp):0), (void*)offs, (void*)globoffs, sym->st_size, symname, version, vername?vername:"(none)");
                         //memmove((void*)globoffs, (void*)offs, sym->st_size);   // preapply to copy part from lib to main elf
@@ -659,7 +665,7 @@ int RelocateElfRELA(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t
                     }
                     if (!offs) {
                         if(strcmp(symname, "__gmon_start__") && strcmp(symname, "data_start") && strcmp(symname, "__data_start"))
-                            printf_log(LOG_NONE, "Error: Global Symbol %s not found, cannot apply R_X86_64_GLOB_DAT @%p (%p) in %s\n", symname, p, *(void**)p, head->name);
+                            printf_log(LOG_NONE, "%s: Global Symbol %s not found, cannot apply R_X86_64_GLOB_DAT @%p (%p) in %s\n", (bind==STB_WEAK)?"Warning":"Error", symname, p, *(void**)p, head->name);
                     } else {
                         printf_dump(LOG_NEVER, "Apply %s R_X86_64_GLOB_DAT @%p (%p -> %p) on sym=%s (ver=%d/%s)\n", (bind==STB_LOCAL)?"Local":"Global", p, (void*)(p?(*p):0), (void*)offs, symname, version, vername?vername:"(none)");
                         *p = offs/* + rela[i].r_addend*/;   // not addend it seems
@@ -939,7 +945,7 @@ void AddSymbols(lib_t *maplib, kh_mapsymbols_t* mapsymbols, kh_mapsymbols_t* wea
         int bind = ELF64_ST_BIND(h->SymTab[i].st_info);
         int type = ELF64_ST_TYPE(h->SymTab[i].st_info);
         int vis = h->SymTab[i].st_other&0x3;
-        uint64_t sz = h->SymTab[i].st_size;
+        size_t sz = h->SymTab[i].st_size;
         if((type==STT_OBJECT || type==STT_FUNC || type==STT_COMMON || type==STT_TLS  || type==STT_NOTYPE) 
         && (vis==STV_DEFAULT || vis==STV_PROTECTED) && (h->SymTab[i].st_shndx!=0)) {
             if(sz && strstr(symname, "@@")) {
@@ -997,7 +1003,7 @@ void AddSymbols(lib_t *maplib, kh_mapsymbols_t* mapsymbols, kh_mapsymbols_t* wea
         if((type==STT_OBJECT || type==STT_FUNC || type==STT_COMMON || type==STT_TLS  || type==STT_NOTYPE) 
         && (vis==STV_DEFAULT || vis==STV_PROTECTED) && (h->DynSym[i].st_shndx!=0 && h->DynSym[i].st_shndx<=65521)) {
             uintptr_t offs = (type==STT_TLS)?h->DynSym[i].st_value:(h->DynSym[i].st_value + h->delta);
-            uint64_t sz = h->DynSym[i].st_size;
+            size_t sz = h->DynSym[i].st_size;
             int version = h->VerSym?((Elf64_Half*)((uintptr_t)h->VerSym+h->delta))[i]:-1;
             if(version!=-1) version &= 0x7fff;
             const char* vername = GetSymbolVersion(h, version);
@@ -1018,7 +1024,7 @@ void AddSymbols(lib_t *maplib, kh_mapsymbols_t* mapsymbols, kh_mapsymbols_t* wea
                 if(bind==STB_WEAK) {
                     AddSymbol(weaksymbols, symname, offs, sz, version, vername);
                 } else {
-                    AddSymbol(mapsymbols, symname, offs, sz, version, vername);
+                    AddWeakSymbol(mapsymbols, symname, offs, sz, version?version:1, vername);
                 }
         }
     }
@@ -1075,6 +1081,20 @@ int LoadNeededLibs(elfheader_t* h, lib_t *maplib, needed_libs_t* neededlibs, lib
                 rpath = tmp;
                 free(origin);
             }
+            while(strstr(rpath, "${PLATFORM}")) {
+                char* platform = strdup("x86_64");
+                char* p = strrchr(platform, '/');
+                if(p) *p = '\0';    // remove file name to have only full path, without last '/'
+                char* tmp = (char*)calloc(1, strlen(rpath)-strlen("${PLATFORM}")+strlen(platform)+1);
+                p = strstr(rpath, "${PLATFORM}");
+                memcpy(tmp, rpath, p-rpath);
+                strcat(tmp, platform);
+                strcat(tmp, p+strlen("${PLATFORM}"));
+                if(rpath!=rpathref)
+                    free(rpath);
+                rpath = tmp;
+                free(platform);
+            }
             if(strchr(rpath, '$')) {
                 printf_log(LOG_INFO, "BOX64: Warning, RPATH with $ variable not supported yet (%s)\n", rpath);
             } else {
@@ -1117,6 +1137,8 @@ int ElfCheckIfUseTCMallocMinimal(elfheader_t* h)
             char *needed = h->DynStrTab+h->delta+h->Dynamic[i].d_un.d_val;
             if(!strcmp(needed, "libtcmalloc_minimal.so.4")) // tcmalloc needs to be the 1st lib loaded
                 return 1;
+            else if(!strcmp(needed, "libtcmalloc_minimal.so.0")) // tcmalloc needs to be the 1st lib loaded
+                return 1;
             else
                 return 0;
         }
@@ -1368,6 +1390,13 @@ void* GetTLSPointer(box64context_t* context, elfheader_t* h)
     return ptr->tlsdata+(ptr->tlssize+h->tlsbase);
 }
 
+void* GetDynamicSection(elfheader_t* h)
+{
+    if(!h)
+        return NULL;
+    return h->Dynamic;
+}
+
 #ifdef DYNAREC
 dynablocklist_t* GetDynablocksFromAddress(box64context_t *context, uintptr_t addr)
 {
@@ -1518,6 +1547,8 @@ void CreateMemorymapFile(box64context_t* context, int fd)
         st.st_ino = 0;
     }
 
+    // TODO: create heap entry?
+
     for (size_t i=0; i<h->numPHEntries; ++i) {
         if (h->PHEntries[i].p_memsz == 0) continue;
 
@@ -1530,6 +1561,11 @@ void CreateMemorymapFile(box64context_t* context, int fd)
         
         dummy = write(fd, buff, strlen(buff));
     }
+    // create stack entry
+    sprintf(buff, "%16lx-%16lx %c%c%c%c %16lx %02x:%02x %ld %s\n", 
+        (uintptr_t)context->stack, (uintptr_t)context->stack+context->stacksz,
+        'r','w','-','p', 0L, 0, 0, 0L, "[stack]");
+    dummy = write(fd, buff, strlen(buff));
 }
 
 void ElfAttachLib(elfheader_t* head, library_t* lib)
@@ -1539,6 +1575,77 @@ void ElfAttachLib(elfheader_t* head, library_t* lib)
     head->lib = lib;
 }
 
+typedef struct search_symbol_s{
+    const char* name;
+    void*       addr;
+    void*       lib;
+} search_symbol_t;
+int dl_iterate_phdr_findsymbol(struct dl_phdr_info* info, size_t size, void* data)
+{
+    search_symbol_t* s = (search_symbol_t*)data;
+
+    for(int j = 0; j<info->dlpi_phnum; ++j) {
+        if (info->dlpi_phdr[j].p_type == PT_DYNAMIC) {
+            ElfW(Sym)* sym = NULL;
+            ElfW(Word) sym_cnt = 0;
+            ElfW(Verdef)* verdef = NULL;
+            ElfW(Word) verdef_cnt = 0;
+            char *strtab = NULL;
+            ElfW(Dyn)* dyn = (ElfW(Dyn)*)(info->dlpi_addr +  info->dlpi_phdr[j].p_vaddr); //Dynamic Section
+            // grab the needed info
+            while(dyn->d_tag != DT_NULL) {
+                switch(dyn->d_tag) {
+                    case DT_STRTAB:
+                        strtab = (char *)(dyn->d_un.d_ptr);
+                        break;
+                    case DT_VERDEF:
+                        verdef = (ElfW(Verdef)*)(info->dlpi_addr +  dyn->d_un.d_ptr);
+                        break;
+                    case DT_VERDEFNUM:
+                        verdef_cnt = dyn->d_un.d_val;
+                        break;
+                }
+                ++dyn;
+            }
+            if(strtab && verdef && verdef_cnt) {
+                if((uintptr_t)strtab < (uintptr_t)info->dlpi_addr) // this test is need for linux-vdso on PI and some other OS (looks like a bug to me)
+                    strtab=(char*)((uintptr_t)strtab + info->dlpi_addr);
+                // Look fr all defined versions now
+                ElfW(Verdef)* v = verdef;
+                while(v) {
+                    ElfW(Verdaux)* vda = (ElfW(Verdaux)*)(((uintptr_t)v) + v->vd_aux);
+                    if(v->vd_version>0 && !v->vd_flags)
+                        for(int i=0; i<v->vd_cnt; ++i) {
+                            const char* vername = (strtab+vda->vda_name);
+                            if(vername && vername[0] && (s->addr = dlvsym(s->lib, s->name, vername))) {
+                                printf_log(/*LOG_DEBUG*/LOG_INFO, "Found symbol with version %s, value = %p\n", vername, s->addr);
+                                return 1;   // stop searching
+                            }
+                            vda = (ElfW(Verdaux)*)(((uintptr_t)vda) + vda->vda_next);
+                        }
+                    v = v->vd_next?(ElfW(Verdef)*)((uintptr_t)v + v->vd_next):NULL;
+                }
+            }
+        }
+    }
+    return 0;
+}
+
+void* GetNativeSymbolUnversionned(void* lib, const char* name)
+{
+    // try to find "name" in loaded elf, whithout checking for the symbol version (like dlsym, but no version check)
+    search_symbol_t s;
+    s.name = name;
+    s.addr = NULL;
+    if(lib) 
+        s.lib = lib;
+    else 
+        s.lib = my_context->box64lib;
+    printf_log(LOG_INFO, "Look for %s in loaded elfs\n", name);
+    dl_iterate_phdr(dl_iterate_phdr_findsymbol, &s);
+    return s.addr;
+}
+
 uintptr_t pltResolver = ~0LL;
 EXPORT void PltResolver(x64emu_t* emu)
 {

src/elfs/elfloader_private.h

@@ -8,7 +8,8 @@ typedef struct dynablocklist_s dynablocklist_t;
 typedef struct library_s library_t;
 typedef struct needed_libs_s needed_libs_t;
 
-#include <pthread.h>
+#include <elf.h>
+#include "elfloader.h"
 
 struct elfheader_s {
     char*       name;
@@ -66,6 +67,9 @@ struct elfheader_s {
     uintptr_t   plt_end;
     uintptr_t   text;
     size_t      textsz;
+    uintptr_t   ehframe;
+    uintptr_t   ehframe_end;
+    uintptr_t   ehframehdr;
 
     uintptr_t   paddr;
     uintptr_t   vaddr;

src/elfs/elfparser.c

@@ -246,6 +246,10 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
                     h->initarray_sz = val / sizeof(Elf64_Addr);
                     printf_log(LOG_DEBUG, "The DT_INIT_ARRAYSZ is %zu\n", h->initarray_sz);
                     break;
+                case DT_PREINIT_ARRAYSZ:
+                    if(val)
+                        printf_log(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);
@@ -345,6 +349,17 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
             h->textsz = h->SHEntries[ii].sh_size;
             printf_log(LOG_DEBUG, "The .text is at address %p, and is %zu big\n", (void*)h->text, h->textsz);
         }
+        ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".eh_frame");
+        if(ii) {
+            h->ehframe = (uintptr_t)(h->SHEntries[ii].sh_addr);
+            h->ehframe_end = h->ehframe + h->SHEntries[ii].sh_size;
+            printf_log(LOG_DEBUG, "The .eh_frame section is at address %p..%p\n", (void*)h->ehframe, (void*)h->ehframe_end);
+        }
+        ii = FindSection(h->SHEntries, h->numSHEntries, h->SHStrTab, ".eh_frame_hdr");
+        if(ii) {
+            h->ehframehdr = (uintptr_t)(h->SHEntries[ii].sh_addr);
+            printf_log(LOG_DEBUG, "The .eh_frame_hdr section is at address %p\n", (void*)h->ehframehdr);
+        }
 
         LoadNamedSection(f, h->SHEntries, h->numSHEntries, h->SHStrTab, ".dynstr", "DynSym Strings", SHT_STRTAB, (void**)&h->DynStr, NULL);
         LoadNamedSection(f, h->SHEntries, h->numSHEntries, h->SHStrTab, ".dynsym", "DynSym", SHT_DYNSYM, (void**)&h->DynSym, &h->numDynSym);

src/emu/x64emu.c

@@ -459,6 +459,8 @@ void EmuCall(x64emu_t* emu, uintptr_t addr)
     uint64_t old_rsi = R_RSI;
     uint64_t old_rbp = R_RBP;
     uint64_t old_rip = R_RIP;
+    //Push64(emu, GetRBP(emu));   // set frame pointer
+    //SetRBP(emu, GetRSP(emu));   // save RSP
     PushExit(emu);
     R_RIP = addr;
     emu->df = d_none;

src/emu/x64int3.c

@@ -97,13 +97,17 @@ void x64Int3(x64emu_t* emu)
                 s = GetNativeName((void*)addr);
                 if(addr==(uintptr_t)PltResolver) {
                     snprintf(buff, 256, "%s", " ... ");
-                } else if (!strcmp(s, "__open") || !strcmp(s, "open") || !strcmp(s, "open ")) {
+                } else if (!strcmp(s, "__open") || !strcmp(s, "open") || !strcmp(s, "open ") || !strcmp(s, "open64")) {
                     tmp = (char*)(R_RDI);
                     snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", %d (,%d))", tid, *(void**)(R_RSP), s, (tmp)?tmp:"(nil)", (int)(R_ESI), (int)(R_EDX));
                     perr = 1;
+                } else if (!strcmp(s, "shm_open")) {
+                    tmp = (char*)(R_RDI);
+                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", %d, %d)", tid, *(void**)(R_RSP), s, (tmp)?tmp:"(nil)", (int)(R_ESI), (int)(R_EDX));
+                    perr = 1;
                 } else if (!strcmp(s, "fopen") || !strcmp(s, "fopen64")) {
                     tmp = (char*)(R_RDI);
-                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", %d)", tid, *(void**)(R_RSP), s, (tmp)?tmp:"(nil)", (int)(R_ESI));
+                    snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", \"%s\")", tid, *(void**)(R_RSP), s, (tmp)?tmp:"(nil)", (char*)(R_RSI));
                     perr = 2;
                 } else if (!strcmp(s, "__openat64") || !strcmp(s, "openat64") || !strcmp(s, "__openat64_2")) {
                     tmp = (char*)(R_RSI);
@@ -127,10 +131,10 @@ void x64Int3(x64emu_t* emu)
                     tmp = (char*)(R_RDI);
                     snprintf(buff, 255, "%04d|%p: Calling %s(\"%s\", 0x%x)", tid, *(void**)(R_RSP), s, (tmp)?tmp:"(nil)", R_ESI);
                     perr = 1;
-                } else if (!strcmp(s, "lseek64") || !strcmp(s, "my_lseek64")) {
+                } else if (!strcmp(s, "lseek64")) {
                     snprintf(buff, 255, "%04d|%p: Calling %s(%d, %ld, %d)", tid, *(void**)(R_RSP), s, (int)R_EDI, (int64_t)R_RSI, (int)R_EDX);
                     perr = 1;
-                } else if (!strcmp(s, "lseek") || !strcmp(s, "my_lseek")) {
+                } else if (!strcmp(s, "lseek")) {
                     snprintf(buff, 255, "%04d|%p: Calling %s(%d, %ld, %d)", tid, *(void**)(R_RSP), s, (int)R_EDI, (int64_t)R_RSI, (int)R_EDX);
                     perr = 1;
                 } else if (strstr(s, "puts")==s) {
@@ -148,15 +152,15 @@ void x64Int3(x64emu_t* emu)
                 } else if (!strcmp(s, "poll")) {
                     struct pollfd* pfd = (struct pollfd*)(R_RDI);
                     snprintf(buff, 255, "%04d|%p: Calling %s(%p[%d/%d/%d, ...], %d, %d)", tid, *(void**)(R_RSP), s, pfd, pfd->fd, pfd->events, pfd->revents, R_ESI, R_EDX);
-                } else if (strstr(s, "my___printf_chk") || !strcmp(s, "__printf_chk")) {
+                } else if (strstr(s, "__printf_chk")) {
                     tmp = (char*)(R_RSI);
                     snprintf(buff, 255, "%04d|%p: Calling %s(%d, \"%s\" (,%p))", tid, *(void**)(R_RSP), s, R_EDI, (tmp)?tmp:"(nil)", (void*)(R_RDX));
-                } else if (strstr(s, "my___snprintf_chk") || !strcmp(s, "__snprintf_chk")) {
+                } else if (strstr(s, "__snprintf_chk")) {
                     tmp = (char*)(R_R8);
                     pu64 = (uint64_t*)R_RDI;
                     post = 3;
                     snprintf(buff, 255, "%04d|%p: Calling %s(%p, %zu, %d, %zu, \"%s\" (,%p))", tid, *(void**)(R_RSP), s, (void*)R_RDI, R_RSI, R_EDX, R_RCX, (tmp)?tmp:"(nil)", (void*)(R_R9));
-                } else if (strstr(s, "my_snprintf") || !strcmp(s, "snprintf")) {
+                } else if (!strcmp(s, "snprintf")) {
                     tmp = (char*)(R_RDX);
                     pu64 = (uint64_t*)R_RDI;
                     post = 3;

src/emu/x64primop.c

@@ -1191,9 +1191,9 @@ uint16_t sbb16(x64emu_t *emu, uint16_t d, uint16_t s)
         res = d - s - 1;
     else
         res = d - s;
-	CONDITIONAL_SET_FLAG(res & 0x8000, F_SF);
-	CONDITIONAL_SET_FLAG((res & 0xffff) == 0, F_ZF);
-	CONDITIONAL_SET_FLAG(PARITY(res & 0xff), F_PF);
+   	CONDITIONAL_SET_FLAG(res & 0x8000, F_SF);
+   	CONDITIONAL_SET_FLAG((res & 0xffff) == 0, F_ZF);
+   	CONDITIONAL_SET_FLAG(PARITY(res & 0xff), F_PF);
 
 	/* calculate the borrow chain.  See note at top */
 	bc = (res & (~d | s)) | (~d & s);
@@ -1213,9 +1213,9 @@ uint32_t sbb32(x64emu_t *emu, uint32_t d, uint32_t s)
         res = d - s - 1;
     else
         res = d - s;
-	CONDITIONAL_SET_FLAG(res & 0x80000000, F_SF);
-	CONDITIONAL_SET_FLAG(!res, F_ZF);
-	CONDITIONAL_SET_FLAG(PARITY(res & 0xff), F_PF);
+  	CONDITIONAL_SET_FLAG(res & 0x80000000, F_SF);
+  	CONDITIONAL_SET_FLAG(!res, F_ZF);
+  	CONDITIONAL_SET_FLAG(PARITY(res & 0xff), F_PF);
 
 	/* calculate the borrow chain.  See note at top */
 	bc = (res & (~d | s)) | (~d & s);
@@ -1235,9 +1235,9 @@ uint64_t sbb64(x64emu_t *emu, uint64_t d, uint64_t s)
         res = d - s - 1;
     else
         res = d - s;
-	CONDITIONAL_SET_FLAG(res & 0x8000000000000000LL, F_SF);
-	CONDITIONAL_SET_FLAG(!res, F_ZF);
-	CONDITIONAL_SET_FLAG(PARITY(res & 0xff), F_PF);
+  	CONDITIONAL_SET_FLAG(res & 0x8000000000000000LL, F_SF);
+  	CONDITIONAL_SET_FLAG(!res, F_ZF);
+  	CONDITIONAL_SET_FLAG(PARITY(res & 0xff), F_PF);
 
 	/* calculate the borrow chain.  See note at top */
 	bc = (res & (~d | s)) | (~d & s);

src/emu/x64run.c

@@ -21,7 +21,7 @@
 #include "bridge.h"
 #include "signals.h"
 #ifdef DYNAREC
-#include "../dynarec/arm64_lock.h"
+#include "../dynarec/native_lock.h"
 #endif
 
 #include "modrm.h"
@@ -163,6 +163,7 @@ x64emurun:
         GO(0x30, xor)                   /* XOR 0x30 -> 0x35 */
         #undef GO
 
+	case 0x2E:	    /* segments are ignored */
         case 0x36:          /* SS: (ignored) */
             break;
 
@@ -412,8 +413,8 @@ x64emurun:
                 EB->byte[0] = tmp8u;
             } else {
                 do {
-                    tmp8u = arm64_lock_read_b(EB);
-                } while(arm64_lock_write_b(EB, GB));
+                    tmp8u = native_lock_read_b(EB);
+                } while(native_lock_write_b(EB, GB));
                 GB = tmp8u;
             }
             // dynarec use need it's own mecanism
@@ -446,11 +447,11 @@ x64emurun:
                 }
             } else {
                 if(rex.w) {
-                    GD->q[0] = arm64_lock_xchg(ED, GD->q[0]);
+                    GD->q[0] = native_lock_xchg(ED, GD->q[0]);
                 } else {
                     do {
-                        tmp32u = arm64_lock_read_d(ED);
-                    } while(arm64_lock_write_d(ED, GD->dword[0]));
+                        tmp32u = native_lock_read_d(ED);
+                    } while(native_lock_write_d(ED, GD->dword[0]));
                     GD->q[0] = tmp32u;
                 }
             }

src/emu/x64run0f.c

@@ -23,7 +23,7 @@
 #include "signals.h"
 #ifdef DYNAREC
 #include "custommem.h"
-#include "../dynarec/arm64_lock.h"
+#include "../dynarec/native_lock.h"
 #endif
 
 #include "modrm.h"
@@ -311,6 +311,33 @@ int Run0F(x64emu_t *emu, rex_t rex)
             }
             break;
 
+        case 0x3A:
+            opcode = F8;
+            switch(opcode) {
+                case 0xF:  /* palignr */
+                    nextop = F8;
+                    GETEM(1);
+                    GETGM;
+                    tmp8u = F8;
+                    if (tmp8u >= 16) {
+                        GM->q = 0;
+                    } else if (tmp8u > 8) {
+                        tmp8u -= 8;
+                        GM->q >>= tmp8u*8;
+                    } else if (tmp8u == 8 || tmp8u == 0) {
+
+                    } else {
+                        GM->q <<= (8-tmp8u)*8;
+                        GM->q |= (EM->q >> tmp8u*8);
+                    }
+                    break;
+
+                default:
+                    return 1;
+            }
+            break;
+
+
         GOCOND(0x40
             , nextop = F8;
             GETED(0);
@@ -339,8 +366,18 @@ int Run0F(x64emu_t *emu, rex_t rex)
             nextop = F8;
             GETEX(0);
             GETGX;
-            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]);
+            }
             break;
         case 0x53:                      /* RCPPS Gx, Ex */
             nextop = F8;
@@ -1221,30 +1258,36 @@ int Run0F(x64emu_t *emu, rex_t rex)
             CHECK_FLAGS(emu);
             nextop = F8;
             GETED(0);
-            if(rex.w) {
-                tmp64u = ED->q[0];
-                tmp64u2= ED->q[1];
-                if(R_RAX == tmp64u && R_RDX == tmp64u2) {
-                    SET_FLAG(F_ZF);
-                    ED->q[0] = R_EBX;
-                    ED->q[1] = R_ECX;
-                } else {
-                    CLEAR_FLAG(F_ZF);
-                    R_RAX = tmp64u;
-                    R_RDX = tmp64u2;
-                }
-            } else {
-                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_RAX = tmp32u;
-                    R_RDX = tmp32u2;
-                }
+            switch((nextop>>3)&7) {
+                case 1:
+                    if(rex.w) {
+                        tmp64u = ED->q[0];
+                        tmp64u2= ED->q[1];
+                        if(R_RAX == tmp64u && R_RDX == tmp64u2) {
+                            SET_FLAG(F_ZF);
+                            ED->q[0] = R_EBX;
+                            ED->q[1] = R_ECX;
+                        } else {
+                            CLEAR_FLAG(F_ZF);
+                            R_RAX = tmp64u;
+                            R_RDX = tmp64u2;
+                        }
+                    } else {
+                        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_RAX = tmp32u;
+                            R_RDX = tmp32u2;
+                        }
+                    }
+                    break;
+                default:
+                    return 1;
             }
             break;
         case 0xC8:

src/emu/x64run64.c

@@ -136,6 +136,12 @@ int Run64(x64emu_t *emu, rex_t rex, int seg)
                     break;
                 case 0x11:
                     switch(rep) {
+                        case 0: /* MOVUPS FS:Ex, Gx */
+                            nextop = F8;
+                            GETEX_OFFS(0, tlsdata);
+                            GETGX;
+                            EX->u128 = GX->u128;
+                            break;
                         case 1: /* MOVSD Ex, Gx */
                             nextop = F8;
                             GETEX_OFFS(0, tlsdata);
@@ -181,6 +187,20 @@ int Run64(x64emu_t *emu, rex_t rex, int seg)
                     }
                     break;
 
+                case 0x6F:
+                    switch(rep) {
+                        case 2: /* MOVDQU Gx, Ex */
+                            nextop = F8;
+                            GETEX_OFFS(0, tlsdata);
+                            GETGX;
+                            memcpy(GX, EX, 16);    // unaligned...
+                            break;
+                            
+                        default:
+                            return 1;
+                    }
+                    break;
+
                 case 0xAF:                      /* IMUL Gd,Ed */
                     nextop = F8;
                     GETED_OFFS(0, tlsdata);
@@ -230,6 +250,19 @@ int Run64(x64emu_t *emu, rex_t rex, int seg)
             break;
 
 
+        case 0x63:                      /* MOVSXD Gd,Ed */
+            nextop = F8;
+            GETED_OFFS(0, tlsdata);
+            GETGD;
+            if(rex.w)
+                GD->sq[0] = ED->sdword[0];
+            else
+                if(MODREG)
+                    GD->q[0] = ED->dword[0];    // not really a sign extension
+                else
+                    GD->sdword[0] = ED->sdword[0];  // meh?
+            break;
+
         case 0x66:
             return Run6664(emu, rex);
 
@@ -316,7 +349,12 @@ int Run64(x64emu_t *emu, rex_t rex, int seg)
                     ED->dword[0] = GD->dword[0];
             }
             break;
-
+        case 0x8A:                      /* MOV Gb,Eb */
+            nextop = F8;
+            GETEB_OFFS(0, tlsdata);
+            GETGB;
+            GB = EB->byte[0];
+            break;
         case 0x8B:                      /* MOV Gd,Ed */
             nextop = F8;
             GETED_OFFS(0, tlsdata);

src/emu/x64run66.c

@@ -20,7 +20,7 @@
 #include "box64context.h"
 #include "bridge.h"
 #ifdef DYNAREC
-#include "../dynarec/arm64_lock.h"
+#include "../dynarec/native_lock.h"
 #endif
 
 #include "modrm.h"
@@ -160,6 +160,26 @@ int Run66(x64emu_t *emu, rex_t rex, int rep)
             GW->word[0] = imul16(emu, EW->word[0], (uint16_t)tmp16s);
         }
         break;
+    
+    case 0x70:
+    case 0x71:
+    case 0x72:
+    case 0x73:
+    case 0x74:
+    case 0x75:
+    case 0x76:
+    case 0x77:
+    case 0x78:
+    case 0x79:
+    case 0x7a:
+    case 0x7b:
+    case 0x7c:
+    case 0x7d:
+    case 0x7e:
+    case 0x7f:
+        // just ignore the 66 (and maybe other) prefix and use regular conditional jump
+        R_RIP--;
+        return 0;
 
     case 0x81:                              /* GRP3 Ew,Iw */
     case 0x83:                              /* GRP3 Ew,Ib */
@@ -451,12 +471,12 @@ int Run66(x64emu_t *emu, rex_t rex, int rep)
                 if(rex.w) {
                     while(tmp64u) {
                         --tmp64u;
-                        tmp64u = *(uint64_t*)R_RDI;
+                        tmp64u2 = *(uint64_t*)R_RDI;
                         R_RDI += tmp8s;
-                        if(R_RAX==tmp64u)
+                        if(R_RAX==tmp64u2)
                             break;
                     }
-                    if(R_RCX) cmp64(emu, R_RAX, tmp64u);
+                    if(R_RCX) cmp64(emu, R_RAX, tmp64u2);
                 } else {
                     while(tmp64u) {
                         --tmp64u;
@@ -474,12 +494,12 @@ int Run66(x64emu_t *emu, rex_t rex, int rep)
                 if(rex.w) {
                     while(tmp64u) {
                         --tmp64u;
-                        tmp64u = *(uint64_t*)R_RDI;
+                        tmp64u2 = *(uint64_t*)R_RDI;
                         R_RDI += tmp8s;
-                        if(R_RAX!=tmp64u)
+                        if(R_RAX!=tmp64u2)
                             break;
                     }
-                    if(R_RCX) cmp64(emu, R_RAX, tmp64u);
+                    if(R_RCX) cmp64(emu, R_RAX, tmp64u2);
                 } else {
                     while(tmp64u) {
                         --tmp64u;

src/emu/x64run660f.c

@@ -55,6 +55,9 @@ int Run660F(x64emu_t *emu, rex_t rex)
     uint32_t tmp32u;
     uint64_t tmp64u;
     float tmpf;
+    #ifndef NOALIGN
+    int is_nan;
+    #endif
     reg64_t *oped, *opgd;
     sse_regs_t *opex, *opgx, eax1, *opex2;
     mmx87_regs_t *opem, *opgm;
@@ -358,6 +361,7 @@ int Run660F(x64emu_t *emu, rex_t rex)
                 nextop = F8;
                 GETEX(0);
                 GETGX;
+                CHECK_FLAGS(emu);
                 CONDITIONAL_SET_FLAG(!(GX->u128&EX->u128), F_ZF);
                 CONDITIONAL_SET_FLAG(!((~GX->u128)&EX->u128), F_CF);
                 break;
@@ -367,7 +371,7 @@ int Run660F(x64emu_t *emu, rex_t rex)
                 GETEX(0);
                 GETGX;
                 for (int i=0; i<16; ++i) {
-                    GX->sb[i] = abs(EX->sb[i]);
+                    GX->ub[i] = abs(EX->sb[i]);
                 }
                 break;
             case 0x1D:  /* PABSW Gx, Ex */
@@ -375,7 +379,7 @@ int Run660F(x64emu_t *emu, rex_t rex)
                 GETEX(0);
                 GETGX;
                 for (int i=0; i<8; ++i) {
-                    GX->sw[i] = abs(EX->sw[i]);
+                    GX->uw[i] = abs(EX->sw[i]);
                 }
                 break;
             case 0x1E:  /* PABSD Gx, Ex */
@@ -383,7 +387,7 @@ int Run660F(x64emu_t *emu, rex_t rex)
                 GETEX(0);
                 GETGX;
                 for (int i=0; i<4; ++i) {
-                    GX->sd[i] = abs(EX->sd[i]);
+                    GX->ud[i] = abs(EX->sd[i]);
                 }
                 break;
 
@@ -429,6 +433,49 @@ int Run660F(x64emu_t *emu, rex_t rex)
                 for(int i=1; i>=0; --i)
                     GX->sq[i] = EX->sd[i];
                 break;
+            
+            case 0x30: /* PMOVZXBW Gx, Ex */
+                nextop = F8;
+                GETEX(0);
+                GETGX;
+                for(int i=7; i>=0; --i)
+                    GX->uw[i] = EX->ub[i];
+                break;
+            case 0x31: /* PMOVZXBD Gx, Ex */
+                nextop = F8;
+                GETEX(0);
+                GETGX;
+                for(int i=3; i>=0; --i)
+                    GX->ud[i] = EX->ub[i];
+                break;
+            case 0x32: /* PMOVZXBQ Gx, Ex */
+                nextop = F8;
+                GETEX(0);
+                GETGX;
+                for(int i=1; i>=0; --i)
+                    GX->q[i] = EX->ub[i];
+                break;
+            case 0x33: /* PMOVZXWD Gx, Ex */
+                nextop = F8;
+                GETEX(0);
+                GETGX;
+                for(int i=3; i>=0; --i)
+                    GX->ud[i] = EX->uw[i];
+                break;
+            case 0x34: /* PMOVZXWQ Gx, Ex */
+                nextop = F8;
+                GETEX(0);
+                GETGX;
+                for(int i=1; i>=0; --i)
+                    GX->q[i] = EX->uw[i];
+                break;
+            case 0x35: /* PMOVZXDQ Gx, Ex */
+                nextop = F8;
+                GETEX(0);
+                GETGX;
+                for(int i=1; i>=0; --i)
+                    GX->q[i] = EX->ud[i];
+                break;
 
             case 0x39:  /* PMINSD Gx, Ex */
                 nextop = F8;
@@ -438,8 +485,16 @@ int Run660F(x64emu_t *emu, rex_t rex)
                     if(GX->sd[i]>EX->sd[i])
                         GX->sd[i] = EX->sd[i];
                 break;
+            case 0x3A:  /* PMINUW Gx, Ex */
+                nextop = F8;
+                GETEX(0);
+                GETGX;
+                for(int i=0; i<8; ++i)
+                    if(GX->uw[i]<EX->uw[i])
+                        GX->uw[i] = EX->uw[i];
+                break;
 
-            case 0x3A:  /* PMAXSD Gx, Ex */
+            case 0x3D:  /* PMAXSD Gx, Ex */
                 nextop = F8;
                 GETEX(0);
                 GETGX;
@@ -633,6 +688,25 @@ int Run660F(x64emu_t *emu, rex_t rex)
                 }
                 break;
 
+            case 0x14:      // PEXTRB EB, GX, u8
+                nextop = F8;
+                GETEB(1);
+                GETGX;
+                tmp8u = F8;
+                EB->byte[0] = GX->ub[tmp8u&0x0f];
+                break;
+
+            case 0x16:      // PEXTRD/Q ED, GX, u8
+                nextop = F8;
+                GETED(1);
+                GETGX;
+                tmp8u = F8;
+                if(rex.w) {
+                    ED->q[0] = GX->q[tmp8u&1];
+                } else {
+                    ED->q[0] = GX->ud[tmp8u&3];
+                }
+                break;
             case 0x17:      // EXTRACTPS ED, GX, u8
                 nextop = F8;
                 GETED(1);
@@ -731,8 +805,14 @@ int Run660F(x64emu_t *emu, rex_t rex)
         nextop = F8;
         GETEX(0);
         GETGX;
-        GX->d[0] = sqrt(EX->d[0]);
-        GX->d[1] = sqrt(EX->d[1]);
+        for (int i=0; i<2; ++i) {
+            #ifndef NOALIGN
+            if(EX->d[i]<0.0)        // on x86, default nan are negative
+                GX->d[i] = -NAN;    // but input NAN are not touched (so sqrt(+nan) -> +nan)
+            else
+            #endif
+            GX->d[i] = sqrt(EX->d[i]);
+        }
         break;
 
     case 0x54:                      /* ANDPD Gx, Ex */
@@ -767,15 +847,29 @@ int Run660F(x64emu_t *emu, rex_t rex)
         nextop = F8;
         GETEX(0);
         GETGX;
-        GX->d[0] += EX->d[0];
-        GX->d[1] += EX->d[1];
+        for(int i=0; i<2; ++i) {
+            #ifndef NOALIGN
+                // add generate a -NAN only if doing inf + -inf
+                if((isinf(GX->d[i]) && isinf(EX->d[i]) && (EX->q[i]&0x8000000000000000LL)!=(GX->q[i]&0x8000000000000000LL)))
+                    GX->d[i] = -NAN;
+                else
+            #endif
+            GX->d[i] += EX->d[i];
+        }
         break;
     case 0x59:                      /* MULPD Gx, Ex */
         nextop = F8;
         GETEX(0);
         GETGX;
-        GX->d[0] *= EX->d[0];
-        GX->d[1] *= EX->d[1];
+        for(int i=0; i<2; ++i) {
+            #ifndef NOALIGN
+                // mul generate a -NAN only if doing (+/-)inf * (+/-)0
+                if((isinf(GX->d[i]) && EX->d[i]==0.0) || (isinf(EX->d[i]) && GX->d[i]==0.0))
+                    GX->d[i] = -NAN;
+                else
+            #endif
+            GX->d[i] *= EX->d[i];
+        }
         break;
     case 0x5A:                      /* CVTPD2PS Gx, Ex */
         nextop = F8;
@@ -820,8 +914,15 @@ int Run660F(x64emu_t *emu, rex_t rex)
         nextop = F8;
         GETEX(0);
         GETGX;
-        GX->d[0] -= EX->d[0];
-        GX->d[1] -= EX->d[1];
+        for(int i=0; i<2; ++i) {
+            #ifndef NOALIGN
+                // sub generate a -NAN only if doing inf - inf
+                if((isinf(GX->d[i]) && isinf(EX->d[i]) && (EX->q[i]&0x8000000000000000LL)==(GX->q[i]&0x8000000000000000LL)))
+                    GX->d[i] = -NAN;
+                else
+            #endif
+            GX->d[i] -= EX->d[i];
+        }
         break;
     case 0x5D:                      /* MINPD Gx, Ex */
         nextop = F8;
@@ -836,8 +937,16 @@ int Run660F(x64emu_t *emu, rex_t rex)
         nextop = F8;
         GETEX(0);
         GETGX;
-        GX->d[0] /= EX->d[0];
-        GX->d[1] /= EX->d[1];
+        for (int i=0; i<2; ++i) {
+            #ifndef NOALIGN
+            is_nan = isnan(GX->d[i]) || isnan(EX->d[i]);
+            #endif
+            GX->d[i] /= EX->d[i];
+            #ifndef NOALIGN
+            if(!is_nan && isnan(GX->d[i]))
+                GX->d[i] = -NAN;
+            #endif
+        }
         break;
     case 0x5F:                      /* MAXPD Gx, Ex */
         nextop = F8;
@@ -1157,11 +1266,25 @@ int Run660F(x64emu_t *emu, rex_t rex)
         nextop = F8;
         GETEX(0);
         GETGX;
+        #ifndef NOALIGN
+        is_nan = isnan(GX->d[0]) || isnan(GX->d[1]);
+        #endif
         GX->d[0] += GX->d[1];
+        #ifndef NOALIGN
+        if(!is_nan && isnan(GX->d[0]))
+            GX->d[0] = -NAN;
+        #endif
         if(EX==GX) {
             GX->d[1] = GX->d[0];
         } else {
+            #ifndef NOALIGN
+            is_nan = isnan(EX->d[0]) || isnan(EX->d[1]);
+            #endif
             GX->d[1] = EX->d[0] + EX->d[1];
+            #ifndef NOALIGN
+            if(!is_nan && isnan(GX->d[1]))
+                GX->d[1] = -NAN;
+            #endif
         }
         break;
 

src/emu/x64run66f0.c

@@ -19,7 +19,9 @@
 #include "x87emu_private.h"
 #include "box64context.h"
 #include "bridge.h"
-#include "dynarec/arm64_lock.h"
+#ifdef DYNAREC
+#include "dynarec/native_lock.h"
+#endif
 
 #include "modrm.h"
 
@@ -55,10 +57,10 @@ int Run66F0(x64emu_t *emu, rex_t rex)
                     GETGW;
 #ifdef DYNAREC
                     do {
-                        tmp16u = arm64_lock_read_h(EW);
+                        tmp16u = native_lock_read_h(EW);
                         cmp16(emu, R_AX, tmp16u);
                         if(ACCESS_FLAG(F_ZF)) {
-                            tmp32s = arm64_lock_write_h(EW, GW->word[0]);
+                            tmp32s = native_lock_write_h(EW, GW->word[0]);
                         } else {
                             R_AX = tmp16u;
                             tmp32s = 0;
@@ -81,6 +83,86 @@ int Run66F0(x64emu_t *emu, rex_t rex)
             }
             break;
 
+#ifdef DYNAREC
+        #define GO(B, OP)                                           \
+        case B+1:                                                   \
+            nextop = F8;                                            \
+            GETEW(0);                                               \
+            GETGW;                                                  \
+            if(rex.w) {                                             \
+                do {                                                \
+                    tmp64u = native_lock_read_dd(ED);                \
+                    tmp64u = OP##64(emu, tmp64u, GD->q[0]);         \
+                } while (native_lock_write_dd(ED, tmp64u));          \
+            } else {                                                \
+                do {                                                \
+                    tmp16u = native_lock_read_h(ED);                 \
+                    tmp16u = OP##16(emu, tmp16u, GW->word[0]);      \
+                } while (native_lock_write_d(ED, tmp16u));           \
+                if(MODREG)                                          \
+                    EW->word[1] = 0;                                \
+            }                                                       \
+            break;                                                  \
+        case B+3:                                                   \
+            nextop = F8;                                            \
+            GETEW(0);                                               \
+            GETGW;                                                  \
+            if(rex.w)                                               \
+                GD->q[0] = OP##64(emu, GD->q[0], ED->q[0]);         \
+            else                                                    \
+                GW->word[0] = OP##16(emu, GW->word[0], EW->word[0]);\
+            break;                                                  \
+        case B+5:                                                   \
+            if(rex.w)                                               \
+                R_RAX = OP##64(emu, R_RAX, F32S64);                 \
+            else                                                    \
+                R_AX = OP##16(emu, R_AX, F16);                      \
+            break;
+#else
+        #define GO(B, OP)                                           \
+        case B+1:                                                   \
+            nextop = F8;                                            \
+            GETEW(0);                                               \
+            GETGW;                                                  \
+            pthread_mutex_lock(&emu->context->mutex_lock);          \
+            if(rex.w)                                               \
+                ED->q[0] = OP##64(emu, ED->q[0], GD->q[0]);         \
+            else                                                    \
+                if(MODREG)                                          \
+                    ED->q[0] = OP##32(emu, ED->dword[0], GD->dword[0]);     \
+                else                                                        \
+                    EW->word[0] = OP##16(emu, EW->word[0], GW->word[0]);    \
+            pthread_mutex_unlock(&emu->context->mutex_lock);        \
+            break;                                                  \
+        case B+3:                                                   \
+            nextop = F8;                                            \
+            GETEW(0);                                               \
+            GETGW;                                                  \
+            pthread_mutex_lock(&emu->context->mutex_lock);          \
+            if(rex.w)                                               \
+                GD->q[0] = OP##64(emu, GD->q[0], ED->q[0]);         \
+            else                                                    \
+                GW->word[0] = OP##16(emu, GW->word[0], EW->word[0]);\
+            pthread_mutex_unlock(&emu->context->mutex_lock);        \
+            break;                                                  \
+        case B+5:                                                   \
+            pthread_mutex_lock(&emu->context->mutex_lock);          \
+            if(rex.w)                                               \
+                R_RAX = OP##64(emu, R_RAX, F32S64);                 \
+            else                                                    \
+                R_AX = OP##16(emu, R_AX, F16);                      \
+            pthread_mutex_unlock(&emu->context->mutex_lock);        \
+            break;
+#endif
+        GO(0x00, add)                   /* ADD 0x00 -> 0x05 */
+        GO(0x08, or)                    /*  OR 0x08 -> 0x0D */
+        GO(0x10, adc)                   /* ADC 0x10 -> 0x15 */
+        GO(0x18, sbb)                   /* SBB 0x18 -> 0x1D */
+        GO(0x20, and)                   /* AND 0x20 -> 0x25 */
+        GO(0x28, sub)                   /* SUB 0x28 -> 0x2D */
+        GO(0x30, xor)                   /* XOR 0x30 -> 0x35 */
+        #undef GO
+
         case 0x81:              /* GRP Ew,Iw */
         case 0x83:              /* GRP Ew,Ib */
             nextop = F8;
@@ -101,13 +183,13 @@ int Run66F0(x64emu_t *emu, rex_t rex)
                 }
             else
                 switch((nextop>>3)&7) {
-                    case 0: do { tmp16u2 = arm64_lock_read_h(ED); tmp16u2 = add16(emu, tmp16u2, tmp64u);} while(arm64_lock_write_h(ED, tmp16u2)); break;
-                    case 1: do { tmp16u2 = arm64_lock_read_h(ED); tmp16u2 =  or16(emu, tmp16u2, tmp64u);} while(arm64_lock_write_h(ED, tmp16u2)); break;
-                    case 2: do { tmp16u2 = arm64_lock_read_h(ED); tmp16u2 = adc16(emu, tmp16u2, tmp64u);} while(arm64_lock_write_h(ED, tmp16u2)); break;
-                    case 3: do { tmp16u2 = arm64_lock_read_h(ED); tmp16u2 = sbb16(emu, tmp16u2, tmp64u);} while(arm64_lock_write_h(ED, tmp16u2)); break;
-                    case 4: do { tmp16u2 = arm64_lock_read_h(ED); tmp16u2 = and16(emu, tmp16u2, tmp64u);} while(arm64_lock_write_h(ED, tmp16u2)); break;
-                    case 5: do { tmp16u2 = arm64_lock_read_h(ED); tmp16u2 = sub16(emu, tmp16u2, tmp64u);} while(arm64_lock_write_h(ED, tmp16u2)); break;
-                    case 6: do { tmp16u2 = arm64_lock_read_h(ED); tmp16u2 = xor16(emu, tmp16u2, tmp64u);} while(arm64_lock_write_h(ED, tmp16u2)); break;
+                    case 0: do { tmp16u2 = native_lock_read_h(ED); tmp16u2 = add16(emu, tmp16u2, tmp64u);} while(native_lock_write_h(ED, tmp16u2)); break;
+                    case 1: do { tmp16u2 = native_lock_read_h(ED); tmp16u2 =  or16(emu, tmp16u2, tmp64u);} while(native_lock_write_h(ED, tmp16u2)); break;
+                    case 2: do { tmp16u2 = native_lock_read_h(ED); tmp16u2 = adc16(emu, tmp16u2, tmp64u);} while(native_lock_write_h(ED, tmp16u2)); break;
+                    case 3: do { tmp16u2 = native_lock_read_h(ED); tmp16u2 = sbb16(emu, tmp16u2, tmp64u);} while(native_lock_write_h(ED, tmp16u2)); break;
+                    case 4: do { tmp16u2 = native_lock_read_h(ED); tmp16u2 = and16(emu, tmp16u2, tmp64u);} while(native_lock_write_h(ED, tmp16u2)); break;
+                    case 5: do { tmp16u2 = native_lock_read_h(ED); tmp16u2 = sub16(emu, tmp16u2, tmp64u);} while(native_lock_write_h(ED, tmp16u2)); break;
+                    case 6: do { tmp16u2 = native_lock_read_h(ED); tmp16u2 = xor16(emu, tmp16u2, tmp64u);} while(native_lock_write_h(ED, tmp16u2)); break;
                     case 7:                                                 cmp16(emu, ED->word[0], tmp64u); break;
                 }
 #else
@@ -126,6 +208,90 @@ int Run66F0(x64emu_t *emu, rex_t rex)
 #endif
             break;
 
+        case 0xFF:              /* GRP 5 Ed */
+            nextop = F8;
+            GETED(0);
+            switch((nextop>>3)&7) {
+                case 0:                 /* INC Ed */
+#ifdef DYNAREC
+                    if(rex.w)
+                        if(((uintptr_t)ED)&7) {
+                            // unaligned
+                            do {
+                                tmp64u = ED->q[0] & 0xffffffffffffff00LL;
+                                tmp64u |= native_lock_read_b(ED);
+                                tmp64u = inc64(emu, tmp64u);
+                            } while(native_lock_write_b(ED, tmp64u&0xff));
+                            ED->q[0] = tmp64u;
+                        }
+                        else
+                            do {
+                                tmp64u = native_lock_read_dd(ED);
+                            } while(native_lock_write_dd(ED, inc64(emu, tmp64u)));
+                    else {
+                        if((uintptr_t)ED&1) { 
+                            //meh.
+                            do {
+                                tmp16u = ED->word[0];
+                                tmp16u &=~0xff;
+                                tmp16u |= native_lock_read_b(ED);
+                                tmp16u = inc16(emu, tmp16u);
+                        } while(native_lock_write_b(ED, tmp16u&0xff));
+                            ED->word[0] = tmp16u;
+                        } else {
+                            do {
+                                tmp16u = native_lock_read_h(ED);
+                            } while(native_lock_write_h(ED, inc16(emu, tmp16u)));
+                        }
+                    }
+#else
+                    pthread_mutex_lock(&emu->context->mutex_lock);
+                    if(rex.w) {
+                        ED->q[0] = inc64(emu, ED->q[0]);
+                    } else {
+                        ED->word[0] = inc16(emu, ED->word[0]);
+                    }
+                    pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                    break;
+                case 1:                 /* DEC Ed */
+#ifdef DYNAREC
+                    if(rex.w)
+                        if(((uintptr_t)ED)&7) {
+                            // unaligned
+                            do {
+                                tmp64u = ED->q[0] & 0xffffffffffffff00LL;
+                                tmp64u |= native_lock_read_b(ED);
+                                tmp64u = dec64(emu, tmp64u);
+                            } while(native_lock_write_b(ED, tmp64u&0xff));
+                            ED->q[0] = tmp64u;
+                        }
+                        else
+                            do {
+                                tmp64u = native_lock_read_dd(ED);
+                            } while(native_lock_write_dd(ED, dec64(emu, tmp64u)));
+                    else {
+                        do {
+                            tmp16u = native_lock_read_h(ED);
+                        } while(native_lock_write_h(ED, dec16(emu, tmp16u)));
+                    }
+#else
+                    pthread_mutex_lock(&emu->context->mutex_lock);
+                    if(rex.w) {
+                        ED->q[0] = dec64(emu, ED->q[0]);
+                    } else {
+                        ED->word[0] = dec16(emu, ED->word[0]);
+                    }
+                    pthread_mutex_unlock(&emu->context->mutex_lock);
+#endif
+                    break;
+                default:
+                    printf_log(LOG_NONE, "Illegal Opcode 0xF0 0xFF 0x%02X 0x%02X\n", nextop, PK(0));
+                    emu->quit=1;
+                    emu->error |= ERR_ILLEGAL;
+                    break;
+            }
+            break;
        default:
             return 1;
     }