Bumped version to v0.2.6

* Update CMakeLists.txt

* Update COMPILE.md

* Update COMPILE.md

.clang-format

@@ -0,0 +1,14 @@
+---
+BasedOnStyle: WebKit
+AlignTrailingComments: true
+AllowShortBlocksOnASingleLine: true
+AllowShortCaseLabelsOnASingleLine: true
+AllowShortFunctionsOnASingleLine: true
+AllowShortIfStatementsOnASingleLine: true
+ColumnLimit: 0
+IndentWidth: 4
+SortIncludes: false
+MaxEmptyLinesToKeep: 2
+IndentCaseLabels: true
+AlignConsecutiveMacros: true
+---

.github/workflows/manual.yml

@@ -17,7 +17,10 @@ on:
         - RK3399
         - RK3588
         - RPI4ARM64
+        - RPI5ARM64
         - SD845
+        - SD888
+        - ADLINK
         - TEGRAX1
         - ARM64
       build_type:

.github/workflows/release.yml

@@ -27,7 +27,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        platform: [X64, OTHER_ARM, RISCV, RPI4ARM64, RK3326, RK3399, RK3588, PHYTIUM, SD845, ARM64]
+        platform: [X64, OTHER_ARM, RISCV, RPI4ARM64, RPI5ARM64, RK3326, RK3399, RK3588, PHYTIUM, SD845, SD888, ADLINK, ARM64, ANDROID]
         type: [Release, Trace]
         os: [ubuntu-latest]
         include:
@@ -47,9 +47,17 @@ jobs:
         run: |
           sudo apt-get update
           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
+            sudo apt-get -y install git cmake make python3
+            if [[ ${{ matrix.platform }} == 'ANDROID' ]]; then
+              wget https://dl.google.com/android/repository/android-ndk-r23b-linux.zip
+              unzip android-ndk-r23b-linux.zip
+              echo "BOX64_COMPILER=$PWD/android-ndk-r23b/toolchains/llvm/prebuilt/linux-x86_64/bin/aarch64-linux-android31-clang" >> $GITHUB_ENV
+              echo "BOX64_PLATFORM_MARCRO=-DANDROID=1 -DARM_DYNAREC=1 -DNOLOADADDR=1" >> $GITHUB_ENV
+            else
+              sudo apt-get -y install git gcc-aarch64-linux-gnu
+              echo "BOX64_PLATFORM_MARCRO=-D${{ matrix.platform }}=1" >> $GITHUB_ENV
+              echo "BOX64_COMPILER=aarch64-linux-gnu-gcc" >> $GITHUB_ENV
+            fi
           else
             if [[ ${{ matrix.platform }} == 'X64' ]]; then
               echo "BOX64_PLATFORM_MARCRO=-DLD80BITS=1 -DNOALIGN=1" >> $GITHUB_ENV
@@ -88,22 +96,32 @@ jobs:
                    ${{ env.BOX64_PLATFORM_MARCRO }}\
                    -DCMAKE_BUILD_TYPE=${{ env.BOX64_BUILD_TYPE }}\
                    -DHAVE_TRACE=${{ env.BOX64_HAVE_TRACE }}\
-                   -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON
+                   -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON\
+                   -DCI=${{ matrix.platform != 'ANDROID' }}
           make -j$(nproc) VERBOSE=1
 
       - name: "Test Box64"
         if: ${{ matrix.platform != 'TEGRAX1' }}
         run: |
-          if [[ ${{ matrix.platform }} != 'X64' ]]; then
-            sudo apt-get -y install qemu-user
+          if [[ ${{ matrix.platform }} != 'X64' && ${{ matrix.platform }} != 'ANDROID' ]]; then
+            mkdir qemu8.1.0
+            wget -O- -q https://archive.archlinux.org/packages/q/qemu-user-static/qemu-user-static-8.1.0-2-x86_64.pkg.tar.zst | tar -I zstd -C qemu8.1.0 -xf -
+            sudo cp qemu8.1.0/usr/bin/* /usr/bin/
           fi
 
           cd build
           if [[ ${{ matrix.platform }} == 'RISCV' ]]; then
-            QEMU_LD_PREFIX=/usr/riscv64-linux-gnu/ ctest -j$(nproc) --rerun-failed --output-on-failure
-          elif [[ ${{ matrix.platform }} != 'X64' ]]; then
-            QEMU_LD_PREFIX=/usr/aarch64-linux-gnu/ ctest -j$(nproc) --rerun-failed --output-on-failure
+            INTERPRETER=qemu-riscv64-static QEMU_LD_PREFIX=/usr/riscv64-linux-gnu/ ctest -j$(nproc) --rerun-failed --output-on-failure
+            INTERPRETER=qemu-riscv64-static QEMU_LD_PREFIX=/usr/riscv64-linux-gnu/ BOX64_DYNAREC=0 ctest -j$(nproc) --rerun-failed --output-on-failure
+            INTERPRETER=qemu-riscv64-static QEMU_LD_PREFIX=/usr/riscv64-linux-gnu/ QEMU_CPU=rv64,zba=true,zbb=true,zbc=true,zbs=true ctest -j$(nproc) --rerun-failed --output-on-failure
+            INTERPRETER=qemu-riscv64-static QEMU_LD_PREFIX=/usr/riscv64-linux-gnu/ QEMU_CPU=rv64,xtheadba=true,xtheadba=true,xtheadbb=true,xtheadbs=true,xtheadcondmov=true,xtheadmemidx=true,xtheadmempair=true,xtheadfmemidx=true,xtheadmac=true,xtheadfmv=true ctest -j$(nproc) --rerun-failed --output-on-failure
+          elif [[ ${{ matrix.platform }} != 'X64' && ${{ matrix.platform }} != 'ANDROID' ]]; then # AArch64
+            INTERPRETER=qemu-aarch64-static QEMU_LD_PREFIX=/usr/aarch64-linux-gnu/ BOX64_DYNAREC=0 ctest -j$(nproc) --rerun-failed --output-on-failure
+            INTERPRETER=qemu-aarch64-static QEMU_LD_PREFIX=/usr/aarch64-linux-gnu/ ctest -j$(nproc) --rerun-failed --output-on-failure
+          elif [[ ${{ matrix.platform }} == 'ANDROID' ]]; then
+            true
           else
+            BOX64_DYNAREC=0 ctest -j$(nproc) --rerun-failed --output-on-failure
             ctest -j$(nproc) --rerun-failed --output-on-failure
           fi
 

CMakeLists.txt

@@ -7,6 +7,8 @@ SET(CMAKE_CXX_FLAGS_DEBUG  "-O0 -g")
 
 option(RPI3ARM64 "Set to ON if targeting an RaspberryPI3 device with multiarch arm64 and armhf" ${RPI3ARM64})
 option(RPI4ARM64 "Set to ON if targeting an RaspberryPI4 device with multiarch arm64 and armhf" ${RPI4ARM64})
+option(RPI5ARM64 "Set to ON if targeting an RaspberryPi5 device with multiarch arm64 and armhf" ${RPI5ARM64})
+option(RPI5ARM64PS16K "Set to ON if targeting an RaspberryPi5 device with multiarch arm64, armhf, and 16k pages" ${RPI5ARM64PS16K})
 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(RK3588 "Set to ON if targeting an Rockchip RK3588(S) based device" ${RK3588})
@@ -15,19 +17,30 @@ 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(SD888 "Set to ON if targeting a Snapragon 888 based device" ${SD888})
+option(SD8G2 "Set to ON if targeting a Snapragon 8 Gen 2 based device" ${SD8G2})
+option(ADLINK "Set to ON if targeting an ADLink AmpereAltra based device" ${ADLINK})
 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 RISC-V RV64GC 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(ARM64 "Set to ON if targeting a generic ARM64 based device" ${ARM64})
+option(ANDROID "Set to ON if targeting an Android device" ${ANDROID})
+option(TERMUX "Set to ON if targeting an Android device with Termux" ${TERMUX})
 option(USE_CCACHE "Set to ON to use ccache if present in the system" ${USE_CCACHE})
 option(HAVE_TRACE "Set to ON to have Trace ability (needs ZydisInfo library)" ${HAVE_TRACE})
+option(SAVE_MEM "Set to ON to build dynarec with some slower memory saving optimisations" ${SAVE_MEM})
 option(NOLOADADDR "Set to ON to avoid fixing the load address of Box64" OFF)
 option(NOGIT "Set to ON if not building from a git clone repo (like when building from a zip download from github)" ${NOGIT})
 option(BAD_SIGNAL "Set to ON to activate the workaround for incoherent si_info on SIGSEGV" ${BAD_SIGNAL})
 option(SW64 "Set ON if targeting an SW64 based device" ${SW64})
+option(CI "Set to ON if running in CI" ${CI})
+option(WITH_MOLD "Set to ON to use with mold" ${WITH_MOLD})
 
+if(TERMUX)
+    set(TERMUX_PATH "/data/data/com.termux/files")
+    set(ANDROID ON CACHE BOOL "")
+endif()
 if(LARCH64)
     set(LD80BITS OFF CACHE BOOL "")
     set(NOALIGN OFF CACHE BOOL "")
@@ -48,19 +61,25 @@ if(PPC64LE)
     set(RV64_DYNAREC OFF CACHE BOOL "")
     set(PAGE16K OFF CACHE BOOL "")
 endif()
-if(RK3399 OR RK3588 OR ODROIDN2 OR RPI3ARM64 OR RPI4ARM64 OR RK3326 OR TEGRAX1 OR PHYTIUM OR SD845 OR SD888 OR LX2160A OR M1 OR ARM64)
+if(RPI5ARM64PS16K)
+    set(RPI5ARM64 ON CACHE BOOL "")
+endif()
+if(RK3399 OR RK3588 OR ODROIDN2 OR RPI3ARM64 OR RPI4ARM64 OR RPI5ARM64 OR RK3326 OR TEGRAX1 OR PHYTIUM OR SD845 OR SD888 OR SD8G2 OR LX2160A OR M1 OR ARM64 OR ADLINK)
     set(LD80BITS OFF CACHE BOOL "")
     set(NOALIGN OFF CACHE BOOL "")
     set(ARM_DYNAREC ON CACHE BOOL "")
     set(RV64_DYNAREC OFF CACHE BOOL "")
 endif()
-if(RK3399 OR RK3588 OR ODROIDN2 OR RPI3ARM64 OR RPI4ARM64 OR RK3326 OR TEGRAX1 OR PHYTIUM OR SD845 OR SD888 OR LX2160A)
+if(RK3399 OR RK3588 OR ODROIDN2 OR RPI3ARM64 OR RPI4ARM64 OR RK3326 OR TEGRAX1 OR PHYTIUM OR SD845 OR SD888 OR SD8G2 OR LX2160A OR ADLINK)
     set(PAGE16K OFF CACHE BOOL "")
 endif()
+if(RK3399 OR ODROIDN2 OR RPI3ARM64 OR RPI4ARM64 OR RPI5ARM64)
+    set(SAVE_MEM ON CACHE BOOL "")
+endif()
 if(RK3588)
     set(BAD_SIGNAL ON CACHE BOOL "")
 endif()
-if(M1 OR LARCH64)
+if(M1 OR LARCH64 OR RPI5ARM64PS16K)
     set(PAGE16K ON CACHE BOOL "")
 endif()
 if(SW64)
@@ -68,6 +87,10 @@ if(SW64)
     set(NOALIGN OFF CACHE BOOL "")
     set(PAGE8K ON CACHE BOOL "")
 endif()
+if(ANDROID)
+    set(NOLOADADDR ON CACHE BOOL "")
+    set(BAD_SIGNAL ON CACHE BOOL "")
+endif()
 
 option(LD80BITS "Set to ON if host device have 80bits long double (i.e. i386)" ${LD80BITS})
 option(NOALIGN "Set to ON if host device doesn't need re-align (i.e. i386)" ${NOALIGN})
@@ -117,6 +140,11 @@ elseif(RPI4ARM64)
     add_definitions(-DRPI4ARM64)
     add_definitions(-pipe -march=armv8-a+crc -mtune=cortex-a72)
     set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crc -mtune=cortex-a72")
+elseif(RPI5ARM64)
+    add_definitions(-DRPI)
+    add_definitions(-DRPI5ARM64)
+    add_definitions(-pipe -march=armv8.2-a+crc+crypto+fp16+rcpc+dotprod -mtune=cortex-a76)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8.2-a+crc+crypto+fp16+rcpc+dotprod -mtune=cortex-a76")
 elseif(RK3326)
     add_definitions(-DRK3326)
     add_definitions(-pipe -march=armv8-a+crc+simd+crypto -mcpu=cortex-a35+crypto)
@@ -148,8 +176,16 @@ elseif(SD845)
     set(CMAKE_ASM_FLAGS  "-pipe -march=armv8.2-a+simd+crypto -mtune=cortex-a75.cortex-a55")
 elseif(SD888)
     add_definitions(-DSD888)
-    add_definitions(-pipe -march=armv8.4-a+crypto)
-    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8.4-a+crypto")
+    add_definitions(-pipe -march=armv8.4-a+simd+crypto)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8.4-a+simd+crypto")
+elseif(SD8G2)
+    add_definitions(-DSD8G2)
+    add_definitions(-pipe -march=armv9-a+i8mm+sm4+sha3+rcpc+crypto+nosve+nosve2)
+    set(CMAKE_ASM_FLAGS  "-pipe -march=armv9-a+i8mm+sm4+sha3+rcpc+crypto+nosve+nosve2")
+elseif(ADLINK)
+    add_definitions(-DADLINK)
+    add_definitions(-pipe -mcpu=neoverse-n1 -fuse-ld=gold -fuse-linker-plugin)
+    set(CMAKE_ASM_FLAGS  "-pipe -mcpu=neoverse-n1")
 elseif(M1)
     add_definitions(-DM1)
     add_definitions(-pipe -march=armv8.5-a+simd+crypto)
@@ -174,13 +210,22 @@ elseif(ARM64)
     #add_definitions(-pipe -march=native)
     set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+simd")
 elseif(ARM_DYNAREC)
-    set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crc+simd+crypto")
+    if(NOT ANDROID)
+        set(CMAKE_ASM_FLAGS  "-pipe -march=armv8-a+crc+simd+crypto")
+    endif()
+endif()
+if(ANDROID)
+    add_definitions(-DANDROID)
+endif()
+if(TERMUX)
+    add_definitions(-DTERMUX)
 endif()
-
 if(BAD_SIGNAL)
     add_definitions(-DBAD_SIGNAL)
 endif()
-
+if(SAVE_MEM)
+    add_definitions(-DSAVE_MEM)
+endif()
 if(PAGE8K)
     add_definitions(-DPAGE8K)
 endif()
@@ -268,15 +313,16 @@ set(ELFLOADER_SRC
     "${BOX64_ROOT}/src/mallochook.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"
+    "${BOX64_ROOT}/src/emu/x64compstrings.c"
     "${BOX64_ROOT}/src/emu/x64emu.c"
     "${BOX64_ROOT}/src/emu/x64int3.c"
     "${BOX64_ROOT}/src/emu/x87emu_private.c"
     "${BOX64_ROOT}/src/emu/x64primop.c"
     "${BOX64_ROOT}/src/emu/x64run_private.c"
+    "${BOX64_ROOT}/src/emu/x64shaext.c"
     "${BOX64_ROOT}/src/emu/x64syscall.c"
     "${BOX64_ROOT}/src/emu/x86syscall.c"
     "${BOX64_ROOT}/src/emu/x64tls.c"
@@ -287,7 +333,6 @@ 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"
     "${BOX64_ROOT}/src/libtools/sdl2rwops.c"
@@ -305,6 +350,12 @@ set(ELFLOADER_SRC
     "${BOX64_ROOT}/src/tools/wine_tools.c"
     "${BOX64_ROOT}/src/wrapped/generated/wrapper.c"
 )
+if(NOT ANDROID)
+    list(APPEND ELFLOADER_SRC
+        "${BOX64_ROOT}/src/libtools/obstack.c"
+        "${BOX64_ROOT}/src/elfs/elfdwarf_private.c"
+    )
+endif()
 
 set(INTERPRETER
     "${BOX64_ROOT}/src/emu/x64run.c"
@@ -312,6 +363,8 @@ set(INTERPRETER
     "${BOX64_ROOT}/src/emu/x64run64.c"
     "${BOX64_ROOT}/src/emu/x64run66.c"
     "${BOX64_ROOT}/src/emu/x64run660f.c"
+    "${BOX64_ROOT}/src/emu/x64run66f20f.c"
+    "${BOX64_ROOT}/src/emu/x64run66f30f.c"
     "${BOX64_ROOT}/src/emu/x64run6664.c"
     "${BOX64_ROOT}/src/emu/x64run66d9.c"
     "${BOX64_ROOT}/src/emu/x64run66dd.c"
@@ -342,6 +395,7 @@ set(WRAPPEDS
     "${BOX64_ROOT}/src/wrapped/wrappedatkbridge.c"
     "${BOX64_ROOT}/src/wrapped/wrappedatomic.c"
     "${BOX64_ROOT}/src/wrapped/wrappedatspi.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedayatanaappindicator3.c"
     "${BOX64_ROOT}/src/wrapped/wrappedbz2.c"
     "${BOX64_ROOT}/src/wrapped/wrappedcap.c"
     "${BOX64_ROOT}/src/wrapped/wrappedcairo.c"
@@ -352,6 +406,8 @@ set(WRAPPEDS
     "${BOX64_ROOT}/src/wrapped/wrappedcurl.c"
     "${BOX64_ROOT}/src/wrapped/wrappeddbus.c"
     "${BOX64_ROOT}/src/wrapped/wrappeddbusglib1.c"
+    "${BOX64_ROOT}/src/wrapped/wrappeddbusmenuglib.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedevent21.c"
     "${BOX64_ROOT}/src/wrapped/wrappedexpat.c"
     "${BOX64_ROOT}/src/wrapped/wrappedfaudio.c"
     "${BOX64_ROOT}/src/wrapped/wrappedflac.c"
@@ -373,19 +429,31 @@ set(WRAPPEDS
     "${BOX64_ROOT}/src/wrapped/wrappedgomp.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgssapi.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgssapikrb5.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstallocators.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgstapp.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgstaudio.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgstbase.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstcheck.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstcontroller.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstfft.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgstgl.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstnet.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstpbutils.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgstreamer.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstriff.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstrtp.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstrtsp.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedgstsdp.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgsttag.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgstvideo.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgthread2.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgtk3.c"
     "${BOX64_ROOT}/src/wrapped/wrappedgtkx112.c"
     "${BOX64_ROOT}/src/wrapped/wrappedkrb5.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlcms2.c"
     "${BOX64_ROOT}/src/wrapped/wrappedldlinux.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibasound.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibbsd.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibc.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibcmusl.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibcrypt.c"
@@ -395,6 +463,7 @@ set(WRAPPEDS
     "${BOX64_ROOT}/src/wrapped/wrappedlibegl.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibform.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibformw.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedlibformw6.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibfuse.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibgl.c"
     "${BOX64_ROOT}/src/wrapped/wrappedlibglu.c"
@@ -477,6 +546,7 @@ set(WRAPPEDS
     "${BOX64_ROOT}/src/wrapped/wrappednss3.c"
     "${BOX64_ROOT}/src/wrapped/wrappednssutil3.c"
     "${BOX64_ROOT}/src/wrapped/wrappedopenal.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedopencl.c"
     "${BOX64_ROOT}/src/wrapped/wrappedpango.c"
     "${BOX64_ROOT}/src/wrapped/wrappedpangocairo.c"
     "${BOX64_ROOT}/src/wrapped/wrappedpangoft2.c"
@@ -525,11 +595,25 @@ set(WRAPPEDS
     "${BOX64_ROOT}/src/wrapped/wrappedvulkan.c"
     "${BOX64_ROOT}/src/wrapped/wrappedxshmfence.c"
     "${BOX64_ROOT}/src/wrapped/wrappedd3dadapter9.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedicuuc66.c"
+    "${BOX64_ROOT}/src/wrapped/wrappedicui18n66.c"
     "${BOX64_ROOT}/src/wrapped/wrappedicuuc67.c"
     "${BOX64_ROOT}/src/wrapped/wrappedicui18n67.c"
     "${BOX64_ROOT}/src/wrapped/wrappedicuuc72.c"
     "${BOX64_ROOT}/src/wrapped/wrappedicui18n72.c"
 )
+if(ANDROID)
+    list(APPEND WRAPPEDS
+        "${BOX64_ROOT}/src/wrapped/wrappedandroidshmem.c"
+    )
+endif()
+if(TERMUX)
+    list(APPEND WRAPPEDS
+        "${BOX64_ROOT}/src/wrapped/wrappediconv.c"
+        "${BOX64_ROOT}/src/wrapped/wrappedtermuxexec.c"
+        "${BOX64_ROOT}/src/wrapped/wrappedandroidsupport.c"
+    )
+endif()
 
 # If BOX64_ROOT contains a ".c", the build breaks...
 string(REPLACE ".c" "_private.h" MODROOT ${BOX64_ROOT})
@@ -544,16 +628,18 @@ endforeach()
 
 set(WRAPPER "${BOX64_ROOT}/src/wrapped/generated/wrapper.c" "${BOX64_ROOT}/src/wrapped/generated/wrapper.h")
 
-add_custom_command(
-    OUTPUT "${BOX64_ROOT}/src/wrapped/generated/functions_list.txt"
-    COMMAND "${PYTHON_EXECUTABLE}" "${BOX64_ROOT}/rebuild_wrappers.py"
-    "${BOX64_ROOT}"
-    "PANDORA" "HAVE_LD80BITS" "NOALIGN" "HAVE_TRACE" "--"
-    ${WRAPPEDS_HEAD}
-    MAIN_DEPENDENCY "${BOX64_ROOT}/rebuild_wrappers.py"
-    DEPENDS ${WRAPPEDS} ${WRAPPEDS_HEAD}
-    BYPRODUCTS ${WRAPPER}
-)
+if(NOT CI)
+    add_custom_command(
+        OUTPUT "${BOX64_ROOT}/src/wrapped/generated/functions_list.txt"
+        COMMAND "${PYTHON_EXECUTABLE}" "${BOX64_ROOT}/rebuild_wrappers.py"
+        "${BOX64_ROOT}"
+        "PANDORA" "HAVE_LD80BITS" "NOALIGN" "HAVE_TRACE" "ANDROID" "TERMUX" "--"
+        ${WRAPPEDS_HEAD}
+        MAIN_DEPENDENCY "${BOX64_ROOT}/rebuild_wrappers.py"
+        DEPENDS ${WRAPPEDS} ${WRAPPEDS_HEAD}
+        BYPRODUCTS ${WRAPPER}
+    )
+endif ()
 
 #add_custom_command(
 #    OUTPUT "${BOX64_ROOT}/src/dynarec/last_run.txt"
@@ -601,9 +687,11 @@ if(ARM_DYNAREC)
     #"${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"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_67_32.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_6764_32.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_da.c"
     "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_db.c"
     "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_dc.c"
     "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_dd.c"
@@ -611,6 +699,8 @@ if(ARM_DYNAREC)
     "${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"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_66f20f.c"
+    "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_66f30f.c"
     "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_6664.c"
     "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_66f0.c"
     "${BOX64_ROOT}/src/dynarec/arm64/dynarec_arm64_f20f.c"
@@ -649,9 +739,10 @@ if(RV64_DYNAREC)
     #"${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_65.c"
     "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_66.c"
     "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_67.c"
+    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_67_32.c"
     "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_d8.c"
     "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_d9.c"
-    #"${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_da.c"
+    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_da.c"
     "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_db.c"
     "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_dc.c"
     "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_dd.c"
@@ -659,6 +750,8 @@ if(RV64_DYNAREC)
     "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_df.c"
     "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_f0.c"
     "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_660f.c"
+    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_66f20f.c"
+    "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_66f30f.c"
     "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_6664.c"
     "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_66f0.c"
     "${BOX64_ROOT}/src/dynarec/rv64/dynarec_rv64_f20f.c"
@@ -725,7 +818,15 @@ add_dependencies(${BOX64} WRAPPERS)
 if(STATICBUILD)
     set_target_properties(${BOX64} PROPERTIES LINK_FLAGS "-static -Wl,--no-as-needed -Wl,--whole-archive -Wl,--allow-multiple-definition -lm -ldl -lrt -lpthread -lresolv -lc -Wl,-defsym,_DYNAMIC=0 -pthread")
 else()
-    set_target_properties(${BOX64} PROPERTIES LINK_FLAGS "-Wl,--no-as-needed -lc -lm -ldl -lrt -lpthread -lresolv -Wl,--as-needed -pthread")
+    if(ANDROID)
+        if(TERMUX)
+        target_link_libraries(${BOX64} c m dl android-sysv-semaphore)
+        else()
+        target_link_libraries(${BOX64} c m dl)
+        endif()
+    else()
+	    set_target_properties(${BOX64} PROPERTIES LINK_FLAGS "-Wl,--no-as-needed -lc -lm -ldl -lrt -lpthread -lresolv -Wl,--as-needed -pthread")
+    endif()
 endif()
 if(DYNAREC)
     target_link_libraries(${BOX64} dynarec)
@@ -734,7 +835,7 @@ target_link_libraries(${BOX64} interpreter)
 
 if(${CMAKE_VERSION} VERSION_LESS "3.13")
     if(NOT NOLOADADDR)
-        if(CMAKE_C_COMPILER_ID STREQUAL "Clang")
+        if(CMAKE_C_COMPILER_ID STREQUAL "Clang" OR WITH_MOLD)
             set_target_properties(${BOX64} PROPERTIES LINK_FLAGS "-Wl,--image-base=${BOX64_ELF_ADDRESS}")
         else()
             set_target_properties(${BOX64} PROPERTIES LINK_FLAGS "-Wl,-Ttext-segment,${BOX64_ELF_ADDRESS}")
@@ -743,7 +844,7 @@ if(${CMAKE_VERSION} VERSION_LESS "3.13")
 else()
     # If symbols are missing, try this: target_link_options(${BOX64} PUBLIC -rdynamic)
     if(NOT NOLOADADDR)
-        if(CMAKE_C_COMPILER_ID STREQUAL "Clang")
+        if(CMAKE_C_COMPILER_ID STREQUAL "Clang" OR WITH_MOLD)
             target_link_options(${BOX64} PUBLIC LINKER:--image-base=${BOX64_ELF_ADDRESS})
         else()
             target_link_options(${BOX64} PUBLIC LINKER:-Ttext-segment,${BOX64_ELF_ADDRESS})
@@ -761,18 +862,42 @@ if(_x86_64 OR _aarch64)
 endif()
 
 if(NOT _x86 AND NOT _x86_64)
-  install(TARGETS ${BOX64}
+  if(NOT TERMUX)
+   install(TARGETS ${BOX64}
     RUNTIME DESTINATION bin)
+  else()
+   install(TARGETS ${BOX64}
+    RUNTIME DESTINATION ${TERMUX_PATH}/usr/bin)
+  endif()
   if(NOT NO_CONF_INSTALL)
     configure_file(system/box64.conf.cmake system/box64.conf)
-    install(FILES ${CMAKE_BINARY_DIR}/system/box64.conf DESTINATION /etc/binfmt.d/)
-    install(FILES ${CMAKE_SOURCE_DIR}/system/box64.box64rc DESTINATION /etc/)
+	  if(NOT TERMUX)
+    		install(FILES ${CMAKE_BINARY_DIR}/system/box64.conf DESTINATION /etc/binfmt.d/)
+    		install(FILES ${CMAKE_SOURCE_DIR}/system/box64.box64rc DESTINATION /etc/)
+	  else()
+		#install(FILES ${CMAKE_BINARY_DIR}/system/box64.conf DESTINATION ${TERMUX_PATH}/usr/etc/binfmt.d/)
+		install(FILES ${CMAKE_SOURCE_DIR}/system/box64.box64rc DESTINATION ${TERMUX_PATH}/usr/etc/)
+	  endif()
+
   endif()
   if(NOT NO_LIB_INSTALL)
-    install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libstdc++.so.5 DESTINATION /usr/lib/x86_64-linux-gnu/)
-    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/)
+	if(NOT TERMUX)
+    		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libstdc++.so.5 DESTINATION /usr/lib/x86_64-linux-gnu/)
+    		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/)
+    		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libcrypto.so.1.1 DESTINATION /usr/lib/x86_64-linux-gnu/)
+    		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libssl.so.1.1 DESTINATION /usr/lib/x86_64-linux-gnu/)
+    		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libunwind.so.8 DESTINATION /usr/lib/x86_64-linux-gnu/)
+	else()
+		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libstdc++.so.5 DESTINATION ${TERMUX_PATH}/usr/lib/x86_64-linux-gnu/)
+		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libstdc++.so.6 DESTINATION ${TERMUX_PATH}/usr/lib/x86_64-linux-gnu/)
+		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libgcc_s.so.1 DESTINATION ${TERMUX_PATH}/usr/lib/x86_64-linux-gnu/)
+		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libpng12.so.0 DESTINATION ${TERMUX_PATH}/usr/lib/x86_64-linux-gnu/)
+		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libcrypto.so.1.1 DESTINATION ${TERMUX_PATH}/usr/lib/x86_64-linux-gnu/)
+		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libssl.so.1.1 DESTINATION ${TERMUX_PATH}/usr/lib/x86_64-linux-gnu/)
+		install(FILES ${CMAKE_SOURCE_DIR}/x64lib/libunwind.so.8 DESTINATION ${TERMUX_PATH}/usr/lib/x86_64-linux-gnu/)
+	endif()
    endif()
 endif()
 
@@ -790,7 +915,11 @@ set(CPACK_GENERATOR "DEB")
 set(CPACK_DEBIAN_PACKAGE_MAINTAINER "ptitSeb")
 set(CPACK_PACKAGE_CONTACT "[email protected]")
 set(CPACK_PACKAGE_DESCRIPTION "Box64 - Linux Userspace x86_64 Emulator with a twist")
-set(CPACK_DEBIAN_PACKAGE_DEPENDS "libc6")
+if(NOT TERMUX)
+	set(CPACK_DEBIAN_PACKAGE_DEPENDS "libc6")
+else()
+	set(CPACK_DEBIAN_PACKAGE_DEPENDS "libandroid-sysv-semaphore")
+endif()
 set(CPACK_PACKAGE_HOMEPAGE_URL, "https://box86.org")
 file(STRINGS "${BOX64_ROOT}/src/box64version.h" TEMP_STRING REGEX "BOX64_MAJOR\\s*")
 string(REGEX REPLACE "BOX64_MAJOR" "" TEMP_STRING ${TEMP_STRING})
@@ -804,9 +933,15 @@ string(REGEX MATCH "[0-9]" BOX64_REVISION ${TEMP_STRING})
 set(CPACK_PACKAGE_VERSION_MAJOR ${BOX64_MAJOR})
 set(CPACK_PACKAGE_VERSION_MINOR ${BOX64_MINOR})
 set(CPACK_PACKAGE_VERSION_PATCH ${BOX64_REVISION})
+if(NOT TERMUX)
 set(CPACK_DEBIAN_PACKAGE_CONTROL_EXTRA "${CMAKE_CURRENT_SOURCE_DIR}/postinst")
+endif()
 if(_aarch64)
-    set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "arm64")
+	if(TERMUX)
+		set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "aarch64")
+	else()
+    		set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "arm64")
+	endif()
 elseif(__riscv64)
     set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE "riscv64")
 elseif(_x86_64)
@@ -815,7 +950,14 @@ else()
     #probably wrong...
     execute_process(COMMAND "dpkg --print-architecture" OUTPUT_VARIABLE CPACK_DEBIAN_PACKAGE_ARCHITECTURE)
 endif()
-set(CPACK_DEBIAN_FILE_NAME "${BOX64}-${BOX64_MAJOR}.${BOX64_MINOR}.${BOX64_REVISION}_Linux-${CPACK_DEBIAN_PACKAGE_ARCHITECTURE}.deb")
+if(TERMUX)
+set(CPACK_PACKAGING_INSTALL_PREFIX "${TERMUX_PATH}/usr")
+endif()
+if(NOT TERMUX)
+set(CPACK_DEBIAN_FILE_NAME "${BOX64}-${BOX64_MAJOR}.${BOX64_MINOR}.${BOX64_REVISION}_${CMAKE_SYSTEM_NAME}-${CPACK_DEBIAN_PACKAGE_ARCHITECTURE}.deb")
+else()
+set(CPACK_DEBIAN_FILE_NAME "${BOX64}-${BOX64_MAJOR}.${BOX64_MINOR}.${BOX64_REVISION}_Termux-${CPACK_DEBIAN_PACKAGE_ARCHITECTURE}.deb")
+endif()
 INCLUDE(CPack)
 
 add_test(bootSyscall ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
@@ -922,19 +1064,56 @@ add_test(irelative_reloc ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref20.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-    add_test(longjumpInSignals ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+add_test(longjumpInSignals ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
     -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test21 -D TEST_OUTPUT=tmpfile21.txt
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref21.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-    add_test(x87 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+add_test(x87 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
     -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test22 -D TEST_OUTPUT=tmpfile22.txt
     -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref22.txt
     -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
 
-    set_tests_properties(x87 PROPERTIES ENVIRONMENT "BOX64_DYNAREC_FASTROUND=0")
+set_tests_properties(x87 PROPERTIES ENVIRONMENT "BOX64_DYNAREC_FASTROUND=0")
 
-    file(GLOB extension_tests "${CMAKE_SOURCE_DIR}/tests/extensions/*.c")
+add_test(pshufb ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test23 -D TEST_OUTPUT=tmpfile23.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref23.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(bswap ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test24 -D TEST_OUTPUT=tmpfile24.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref24.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(x87cache ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test25 -D TEST_OUTPUT=tmpfile25.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref25.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(feround ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test26 -D TEST_OUTPUT=tmpfile26.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref26.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+set_tests_properties(feround PROPERTIES ENVIRONMENT "BOX64_SYNC_ROUNDING=1")
+
+add_test(sse4_2 ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test27 -D TEST_OUTPUT=tmpfile27.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref27.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(shaext ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test28 -D TEST_OUTPUT=tmpfile28.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref28.txt
+    -P ${CMAKE_SOURCE_DIR}/runTest.cmake )
+
+add_test(lock ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
+    -D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test29 -D TEST_OUTPUT=tmpfile29.txt
+    -D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref29.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}

README.md

@@ -37,8 +37,8 @@ Note: Box64's Dynarec uses a mechanism with Memory Protection and a SegFault sig
 
 Compiling/Installation
 ----
-> Compilation instructions can be found [here](docs/COMPILE.md)\
-> Instructions for installing Wine for Box64 can be found [here](docs/X64WINE.md)
+> Compilation instructions can be found [here](https://github.com/ptitSeb/box64/blob/main/docs/COMPILE.md) \
+> Instructions for installing Wine for Box64 can be found [here](https://github.com/ptitSeb/box64/blob/main/docs/X64WINE.md)
 
 ----
 
@@ -54,7 +54,7 @@ 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-bit libraries. Box64 doesn't include any 64-bit <-> 32-bit translation. 
 
-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).
+So understand that box64 will only run 64-bit linux binaries. For 32-bit binaries, you need box86 (with all the multiarch or proot trickery it implies on a 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"
 
 ----
@@ -62,16 +62,16 @@ Note that many installer (mojo setup based) will fall back to "x86" when detecti
 Notes about Box64 configuration
 ----
 
-Box64 now have configurations files. There are 2 files loaded. `/etc/box4.box64rc` and `~/.box64rc`. Both files have the same syntax, and is basicaly an ini files. Section in square brakets define the process name, and the rest is the env. var. to set. Looke at [Usage](USAGE.md) for detail on what parameters can be put. Box64 comes with a default file that should be installed for better stability. The file in in `system/box64.box64rc` and should be installed to `/etc/box64.box64rc` If, for some reasons, you don't want to install that file here, at least copy it to `~/.box64rc` or some game may not function correctly.
-Note that the priority is: `~/.bashrc` > `/etc/box64.box64rc` > command line
-So, your settings in `~/.bashrc` may override the setting from your command line...
+Box64 now have configurations files. There are 2 files loaded. `/etc/box4.box64rc` and `~/.box64rc`. Both files have the same syntax, and is basicaly an ini files. Section in square brakets define the process name, and the rest is the env. var. to set. Looke at [Usage](USAGE.md) for detail on what parameters can be put. Box64 comes with a default file that should be installed for better stability. The file in in `system/box64.box64rc` and should be installed to `/etc/box64.box64rc` If, for some reason, you don't want to install that file here, at least copy it to `~/.box64rc` or some game may not function correctly.
+Note that the priority is: `~/.box64rc` > `/etc/box64.box64rc` > command line
+So, your settings in `~/.box64rc` may override the setting from your command line...
 
 ----
 
 Notes about Unity game emulation
 ----
 
-Running Unity games should just works, but you should also note that many Unity3D games require OpenGL 3+ which can be tricky to provide on ARM SBC (single-board computers). Also many newer Unity3D (like KSP) games use the BC7 compressed textures, wich is not supported on many ARM integrated GPU.
+Running Unity games should just work, but you should also note that many Unity3D games require OpenGL 3+ which can be tricky to provide on ARM SBC (single-board computers). Also many newer Unity3D (like KSP) games use the BC7 compressed textures, which is not supported on many ARM integrated GPU.
 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.
 
 ----
@@ -86,7 +86,7 @@ GTK libraries are now wrapped on box64, both gtk2 and gtk3.
 Notes about Steam
 ----
 
-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, and that steamwebhelper process is now mendatory, even on the "small mode". And that process will eat lots of memory. So machine with less the 6Gb of RAM will need a swapfile tp use Steam.
+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, and that steamwebhelper process is now mandatory, even on the "small mode". And that process will eat lots of memory. So machine with less the 6Gb of RAM will need a swapfile to use Steam.
 
 ----
 
@@ -94,14 +94,14 @@ Notes about Wine
 ----
 
 Wine64 is supported on box64. Proton too. 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).
-Note that the new 32bits PE in 64bits process that the Wine time in currently implementing in Wine 7.+ is now supported, but the support is quite young so there might be some residual issues.
+Note that the new 32-bit PE in 64-bit processes that the Wine team is currently implementing in Wine 7.+ is now supported, but the support is quite young so there might be some residual issues.
 
 ----
 
 Notes about Vulkan
 ----
 
-Box64 wraps Vulkan libraries, but note that it as mostly been tested with a AMD RX550 card and on the Freedreno driver, so some extensions may be missing depending on your graphics card.
+Box64 wraps Vulkan libraries, but note that it as mostly been tested with an AMD RX550 card and on the Freedreno driver, so some extensions may be missing depending on your graphics card.
 
 ----
 
@@ -112,11 +112,11 @@ 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, dennis1248, sll00, [libre-computer-project](https://libre.computer/)
+ * For their major financial contribution: FlyingFathead, stormchaser3000, dennis1248, sll00, [libre-computer-project](https://libre.computer/), [CubeCoders Limited](http://cubecoders.com/)
  * For hardware contribution and LoongArch migration: [xiaoji](https://www.linuxgame.cn/), Deepin Beijing Develop Team
  * For their major code contribution on the RV64 Dynarec: ksco, xctan
- * 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/))
+ * For their hardware contribution: [ADLINK](https://www.adlinktech.com/Products/Computer_on_Modules/COM-HPC-Server-Carrier-and-Starter-Kit/Ampere_Altra_Developer_Platform?lang=en) with [Ampere](https://amperecomputing.com/home/edge), [Radxa](https://rockpi.org/), [StarFive](https://rvspace.org/), [Pine64](https://www.pine64.org/), [AYN](https://www.ayntec.com/)
+ * For their continuous advertisements for the 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 this project.
 

README_CN.md

@@ -14,7 +14,7 @@ Box64 可以在非 x86_64 Linux 系统（比如 ARM64）上运行 x86_64 Linux
 
 由于 Box64 使用一些“系统”库的原生版本，如 libc、libm、SDL 和 OpenGL 等，因此很容易与大多数应用程序集成和使用，并且在许多情况下性能会相当不错。可以在[这里](https://box86.org/index.php/2021/06/game-performances/)查看一些性能测试的样例。
 
-Box64 集成了适用于 ARM64 平台的 DynaRec（动态重编译器），速度可以比纯解释模式快 5 到 10 倍。可以在[这里](https://box86.org/2021/07/inner-workings-a-high%E2%80%91level-view-of-box86-and-a-low%E2%80%91level-view-of-the-dynarec/)找到有关 DynaRec 工作原理的一些信息。
+Box64 集成了适用于 ARM64 和 RV64 平台的 DynaRec（动态重编译器），速度可以比纯解释模式快 5 到 10 倍。可以在[这里](https://box86.org/2021/07/inner-workings-a-high%E2%80%91level-view-of-box86-and-a-low%E2%80%91level-view-of-the-dynarec/)找到有关 DynaRec 工作原理的一些信息。
 
 一些 x64 内部操作码使用 “Realmode X86 Emulator Library” 的部分内容，有关版权详细信息，请参见 [x64primop.c](../src/emu/x64primop.c)。
 
@@ -54,7 +54,7 @@ LOGO 由 @grayduck 制作，感谢！
 
 因为 Box64 的工作原理是直接将函数调用从 x86_64 转换为主机系统，所以主机系统（运行 Box64 的系统）需要有 64 位库。Box64 不包含任何 64 位 <-> 32 位的转换。
 
-所以 box64 只能运行 64 位的 Linux 二进制。对于 32 位二进制则需要使用 box86 来运行（它在 64 位操作系统上使用了 multiarch 和 proot 等技巧来实现运行）。请注意，许多（基于 mojo 的）安装程序在检测到 ARM64 操作系统时将回退到 “x86”，因此即使存在 x86_64 版本，也会尝试使用 box86。这时你可以使用一个假的 `uname`，当它的参数为 `-m` 时返回 `x86_64`。
+所以 box64 只能运行 64 位的 Linux 二进制。对于 32 位二进制则需要使用 box86 来运行（它在 64 位操作系统上使用了 multiarch 和 proot 等技巧来实现运行）。请注意，许多（基于 mojo 的）安装程序在检测到 ARM64 操作系统时将回退到 “x86”，因此即使存在 x86_64 版本，也会尝试使用 box86。这时你可以使用一个假的 `uname` ，并使它在运行参数为 `-m` 时返回 `x86_64`。
 
 ----
 
@@ -77,7 +77,7 @@ box64 封装了 GTK，包括 gtk2 和 gtk3。
 关于 Steam 的注意事项
 ----
 
-请注意，Steam 是 32/64 位的混合体，所以还需要 box86 才能运行，因为客户端应用程序是 32 位的。它还使用 64 位本地服务器，但与大多数使用 libcef/chromium 的东西一样，它目前无法在 box64 上正常工作。所以，现在 box64 上暂时不能运行 Steam。
+请注意，Steam 是 32/64 位混合的应用，所以你需要 box86 才能运行，因为客户端应用程序是 32 位的。它还使用 64 位本地服务器，它的 steamwebhelper 无法被关闭（即使是在最小模式）而且会吃掉大量的内存。对于内存小于 6 GB 的机型，你将会需要创建 swapfile 来运行 Steam。
 
 ----
 
@@ -105,7 +105,7 @@ Box64 封装了 Vulkan 库，但请注意，它仅在 RX550 显卡上进行过
  * 代码贡献：rajdakin, mogery
  * 财务帮助：FlyingFathead, stormchaser3000, dennis1248, sll00, [libre-computer-project](https://libre.computer/)
  * 硬件捐赠和 LoongArch 迁移：[xiaoji](https://www.linuxgame.cn/), Deepin Beijing Develop Team
- * 硬件捐赠：[Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/), [StarFive](https://rvspace.org/)
+ * 硬件捐赠：[ADLink](https://www.adlinktech.com/Products/Computer_on_Modules/COM-HPC-Server-Carrier-and-Starter-Kit/Ampere_Altra_Developer_Platform?lang=en), [Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/), [StarFive](https://rvspace.org/)
  * 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/))
 
 我还要感谢很多人，即使他们只参与了一次本项目。

README_UK.md

@@ -111,7 +111,7 @@ Box64 використовує бібліотеки Vulkan, але зауваж
 * За великий фінансовий внесок: FlyingFathead, stormchaser3000, dennis1248, sll00, [libre-computer-project](https://libre.computer/)
 * За апаратне забезпечення та міграцію LoongArch: [xiaoji](https://www.linuxgame.cn/), команда розробників Deepin Beijing
 * За основний внесок у код RV64 Dynarec: ksco, xctan
-* За внесок у апаратне забезпечення: [Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/), [StarFive](https://rvspace.org/)
+* За внесок у апаратне забезпечення: [ADLink](https://www.adlinktech.com/Products/Computer_on_Modules/COM-HPC-Server-Carrier-and-Starter-Kit/Ampere_Altra_Developer_Platform?lang=en), [Radxa](https://rockpi.org/), [Pine64](https://www.pine64.org/), [StarFive](https://rvspace.org/)
 * За безперервну рекламу Box64: salva ([microLinux](https://www.youtube.com/channel/UCwFQAEj1lp3out4n7BeBatQ)), [PILab](https://www.youtube.com/channel/UCgfQjdc5RceRlTGfuthBs7g)/Команда [TwisterOS](https://twisteros.com/), [The Byteman](https://www.youtube.com/channel/UCEr8lpIJ3B5Ctc5BvcOHSnA), [NicoD](https://www.youtube.com/channel/UCpv7NFr0-9AB5xoklh3Snhg), ekianjo ([Boilingsteam](https://boilingsteam.com/))
 
 І я також дякую багатьом іншим людям, які хоч раз брали участь у цьому проекті.

debian/changelog

@@ -1,3 +1,60 @@
+box64 (0.2.6) unstable; urgency=low
+
+ * X64Libs: Updated libstdc++ & libgcc_s
+ * X64Libs: Use x86_64 vrsion of libunwind (it's needed by wine)
+ * Dynarec: More opcodes on ARM64 & RV64
+ * Dynarec: Fixed some flags propagation issues inside blocks
+ * Dynarec: Improved Strong Memory Model emulation, introducing a 3rd level
+ * Dynarec: ARM64: reworked shift and bit opcodes flags handling
+ * Dynarec: RV64: Many fixes, making Wow64 apps running
+ * Dynarec: ARM64: Added support for SSE4.2 (with CRC32 if supported by CPU)
+ * Dynarec: ARM64: Added support for SHA (with hardware support if CPU allows it)
+ * Dynarec: ARM64: The CALLRET optimisation has been improved and is now default
+ * CPU: Remoworked how INT, priviliged opcodes and teh Trace flags works
+ * CPU: Added full support for SSE4.2
+ * CPU: Added full support for SHA extension
+ * ElfLoader: rework how elf memory is handled
+ * Emulation: Added some more syscalls
+ * Emulation: Better handling of Signal (also handling SIGMAX correctly)
+ * Emulation: Better support for POKEUSER/PEEKUSER (helps Windows Unity games)
+ * Wrapping: More libs and some fixes in function wrapping (gtk3, vulkan...)
+ * Wrapping: Removed faked libunwind wrapping
+ * Hardware: Added more profile, with support ADLink Ampere Altra, Qualcomm 8Gen3, and Pi5 machines
+ * OS: Added profile to build for Android and Termux.
+ * OS: New option to use mold for a high speed link
+ * => More hadware support, more functions wrapped for an improve compatibity
+ * => More CPU extension, with ARM64 hardware support when possible, and improved CALLRET optimisation for an improved speed of emulation
+ * => Some work also to reduce the memory footprint
+
+-- Sebastien Chevalier <[email protected]>  Thu, 15 Dec 2023 17:30:33 -0100
+
+box64 (0.2.4) unstable; urgency=low
+
+  * Added Dynarec for RISC-V
+  * Added partial 32 bits support (WoW64 works)
+  * Changed cpuid to display box64
+  * Fixed steam, again
+  * Improved hacked version for overridden malloc/free functions
+  * More wrapped libs
+  * More wrapped functions and syscalls
+  * Reduced memory footprint
+  * Added support for the BIND_NOW flags in ELF files
+  * Added RTLD_NOLOAD support
+  * Fixed some typos and general cleanup, again
+  * Added a cosimulation mode (to compare the behaviour of the dynarec with the interpreter)
+  * Added FASTNAN/FASTROUND handling for more opcodes
+  * Better float/double/long double handling
+  * Better dlopen/dlclose handling
+  * More games in the rc file
+  * Added BOX64_DYNAREC_MISSING to show only missing opcode, independantly of the log level
+  * Added BOX64_RESERVE_HIGH to reserve addresses above 47bits
+  * Improved multitreading support
+  * Ukrainian translation of the README
+  * => Dynarec now support RISC-V, large speedup on RISC-V 64, were games are now playable (like Stardew Valley on a Vision Five 2)
+  * => Improved compatibility with more fixes in elf handling, more wrapped libs and functiond, more fixed opcodes and refined profile
+
+-- Sebastien Chevalier <[email protected]>  Thu, 10 Mar 2023 10:20:33 -0500
+
 box64 (0.2.2) unstable; urgency=low
 
  * Added a hacked version for overridden malloc/free functions, enabling libcef (and derivative) to work.

docs/CHANGELOG.md

@@ -1,3 +1,31 @@
+v0.2.6
+======
+* X64Libs: Updated libstdc++ & libgcc_s
+* X64Libs: Use x86_64 vrsion of libunwind (it's needed by wine)
+* Dynarec: More opcodes on ARM64 & RV64
+* Dynarec: Fixed some flags propagation issues inside blocks
+* Dynarec: Improved Strong Memory Model emulation, introducing a 3rd level
+* Dynarec: ARM64: reworked shift and bit opcodes flags handling
+* Dynarec: RV64: Many fixes, making Wow64 apps running
+* Dynarec: ARM64: Added support for SSE4.2 (with CRC32 if supported by CPU)
+* Dynarec: ARM64: Added support for SHA (with hardware support if CPU allows it)
+* Dynarec: ARM64: The CALLRET optimisation has been improved and is now default
+* CPU: Remoworked how INT, priviliged opcodes and teh Trace flags works
+* CPU: Added full support for SSE4.2
+* CPU: Added full support for SHA extension
+* ElfLoader: rework how elf memory is handled
+* Emulation: Added some more syscalls
+* Emulation: Better handling of Signal (also handling SIGMAX correctly)
+* Emulation: Better support for POKEUSER/PEEKUSER (helps Windows Unity games)
+* Wrapping: More libs and some fixes in function wrapping (gtk3, vulkan...)
+* Wrapping: Removed faked libunwind wrapping
+* Hardware: Added more profile, with support ADLink Ampere Altra, Qualcomm 8Gen3, and Pi5 machines
+* OS: Added profile to build for Android and Termux.
+* OS: New option to use mold for a high speed link
+=> More hadware support, more functions wrapped for an improve compatibity
+=> More CPU extension, with ARM64 hardware support when possible, and improved CALLRET optimisation for an improved speed of emulation
+=> Some work also to reduce the memory footprint
+
 v0.2.4
 ======
 * Added Dynarec for RISC-V

docs/COMPILE.md

@@ -27,9 +27,9 @@ If it's the first install, you also need:
 sudo systemctl restart systemd-binfmt
 ```
 - You can use `make -j1`, `make -j2` to prevent running out of memory
-- You can also add `-DBAD_SIGNAL=ON` to the cmake command if you are on Linux Kernel mixed with Android, like on RK3588 or maybe Termux
+- You can also add `-DBAD_SIGNAL=ON` to the cmake command if you are on Linux Kernel mixed with Android, like on RK3588.
 
-#### For instance, if you want to build box64 for Generic ARM64 Linux platforms, it would look like this:
+#### For Instance, if you want to build box64 for Generic ARM64 Linux platforms, it would look like this:
 ```
 git clone https://github.com/ptitSeb/box64
 cd box64
@@ -68,7 +68,6 @@ Still, this can be faster if your build is attended.
 
 ```
 -D RPI3ARM64=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
-make 
 ```
 
 #### for Raspberry Pi 4
@@ -79,6 +78,13 @@ Warning, you need a 64bit OS:
 -D RPI4ARM64=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
 ```
 
+#### for Raspberry Pi 5
+
+If your Raspberry Pi OS use 16K pages use `RPI5ARM64PS16K` instead of `RPI5ARM64`:
+```
+-D RPI5ARM64=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
+```
+
 #### for TEGRA X1
 
 Using a 64bit OS:
@@ -96,7 +102,7 @@ Using a 64bit OS:
 -D ODROIDN2=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
 ```
 
-#### for Snapdragon 845
+#### for Snapdragon
 
 Using a 64bit OS:
 
@@ -104,6 +110,14 @@ Using a 64bit OS:
 -D SD845=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
 ```
 
+or
+
+```
+-D SD888=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
+```
+
+Depending how recent your Snapdragon is
+
 #### for Phytium
 
 Using a 64bit OS:
@@ -111,11 +125,11 @@ Using a 64bit OS:
 -D PHYTIUM=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
 ```
 
-#### for a generic ARM64 machine
+#### for ADLink machines
 
 Using a 64bit OS:
 ```
--D ARM64=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
+-D ADLINK=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
 ```
 
 #### for M1
@@ -158,6 +172,26 @@ Using a 64bit OS:
 -D LX2160A=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
 ```
 
+#### for Termux
+
+You must have ARM64 Device for build box64.
+
+##### in CHRoot/PRoot
+
+```
+-D ARM64=1 -DCMAKE_C_COMPILER=gcc -DBAD_SIGNAL=ON -DCMAKE_BUILD_TYPE=RelWithDebInfo
+```
+
+##### in Termux (Native)
+
+NOTE: BUILDING BOX64 IN TERMUX NATIVE IS EXPERIMENTAL AND DOESN'T GONNA RUN LINUX BINARIES IN NATIVE TERMUX BOX64!!!
+
+You also needed have `libandroid-sysv-semaphore` and `libandroid-spawn` libraries:
+
+```
+-D TERMUX=1 -D -DCMAKE_C_COMPILER=clang CMAKE_BUILD_TYPE=RelWithDebInfo
+```
+
 #### for x86_64 Linux
 ```
 -D LD80BITS=1 -D NOALIGN=1 -D CMAKE_BUILD_TYPE=RelWithDebInfo
@@ -166,7 +200,7 @@ If you encounter some linking errors, try using `NOLOADADDR=ON` (`cmake -D NOLOA
 
 ----
 
-### use ccmake
+### Use ccmake
 
 Alternatively, you can **use the curses-bases ccmake (or any other gui frontend for cmake)** to select wich platform to use interactively.
 
@@ -187,12 +221,19 @@ To have a trace enabled build (***the interpreter will be slightly slower***), a
 #### To have ARM Dynarec
 
 Dynarec is only available on ARM (for the meantime), Activate it by using `-DARM_DYNAREC=1`.
-###### *Note: VFPv3 and NEON are required for Dynarec.*
+
+#### To have a build using less memory
+
+You can use `-DSAVE_MEM` to have a build that will try to save some memory. It's, for now, only increasing the jumptable from 4 level to 5 levels. The added granularity avoid wasting space, but the 1 level more to the jumptable means there is on read from memory more when jumping between blocks.
 
 #### Not building from a git clone
 
 If you are not building from a git clone (for example, downloading a release source code zip from github), you need to use `-DNOGIT=1` from cmake to be able to build (box64 uses git SHA1 to show last commit in version number).
 
+#### Use faster linker
+
+You need to add `-DWITH_MOLD=1` if GNU ld is extremely slow. Then run `mold -run make -j4` to build (make sure [Mold](https://github.com/rui314/mold) is installed).
+
 ----
 
 Testing

docs/USAGE.md

@@ -3,7 +3,10 @@ Usage
 ----
 
 There are many environment variables to control Box64 behaviour. 
-Env. var with * can also be put inside box64rc files
+Env. var with * can also be put inside box64rc files. 
+Box64 look for 2 places for rcfile: `/etc/box64.box64rc` and `~/.box64rc`
+The second takes precedence to the first, on an APP level 
+(that means if an [MYAPP] my appears in both file, only the settings in `~/.box64rc` will be applied)
 
 #### BOX64_LOG *
 Controls the Verbosity level of the logs
@@ -49,7 +52,7 @@ Only on build with trace enabled. Trace allow the logging of all instruction exe
  * 0xXXXXXXX-0xYYYYYYY : Trace only between the 2 addresses.
 
 #### BOX64_TRACE_INIT *
-Use BOX64_TRACE_INIT instead of BOX_TRACE to start trace before the initialisation of libraries and the running program
+Use BOX64_TRACE_INIT instead of BOX64_TRACE to start trace before the initialisation of libraries and the running program
  * 0 : No trace. (Default.)
  * 1 : Trace enabled. The trace start with the initialisation of all depending libraries is done.
 
@@ -153,8 +156,9 @@ Define Box64's Dynarec max allowed forward value when building Block.
 #### BOX64_DYNAREC_STRONGMEM *
 Enable/Disable simulation of Strong Memory model
 * 0 : Don't try anything special (Default.)
-* 1 : Enable some Memory Barrier when reading from memory (on some MOV opcode) to simulate Strong Memory Model while trying to limit performance impact (Default when libmonobdwgc-2.0.so is loaded)
-* 2 : Enable some Memory Barrier when reading from memory (on some MOV opcode) to simulate Strong Memory Model
+* 1 : Enable some Memory Barrier when writting to memory (on some MOV opcode) to simulate Strong Memory Model while trying to limit performance impact (Default when libmonobdwgc-2.0.so is loaded)
+* 2 : All 1. plus a memory barrier on every write to memory using MOV
+* 3 : All 2. plus Memory Barrier when reading from memory and on some SSE/SSE2 opcodes too
 
 #### BOX64_DYNAREC_X87DOUBLE *
 Force the use of Double for x87 emulation
@@ -179,8 +183,8 @@ Handling of flags on CALL/RET opcodes
 
 #### BOX64_DYNAREC_CALLRET *
 Optimisation of CALL/RET opcodes (not compatible with jit/dynarec/smc)
-* 0 : Don't optimize CALL/RET, use Jump Table for boths (Default)
-* 1 : Try to optimized CALL/RET, skipping the use of the JumpTable when possible (will crash if blacks are invalidate, so probably incompatible with JIT/Dynarec)
+* 0 : Don't optimize CALL/RET, use Jump Table for boths
+* 1 : Try to optimized CALL/RET, skipping the use of the JumpTable when possible (Default)
 
 #### BOX64_DYNAREC_HOTPAGE *
 Handling of HotPage (Page being both executed and written)
@@ -192,6 +196,11 @@ Will use a faster handling of HotPage (Page being both executed and written)
 * 0 : use regular hotpage (Default)
 * 1 : Use faster hotpage, taking the risk of running obsolete JIT code (might be faster, but more prone to crash)
 
+#### BOX64_DYNAREC_ALIGNED_ATOMICS *
+Generated code for aligned atomics only
+* 0 : The code generated can handle unaligned atomics (Default)
+* 1 : Generated code only for aligned atomics (faster and less code generated, but will SEGBUS if LOCK prefix is unsed on unaligned data)
+
 #### BOX64_DYNAREC_BLEEDING_EDGE *
 Detect MonoBleedingEdge and apply conservative settings
 * 0 : Don't detect MonoBleedingEdge
@@ -222,11 +231,21 @@ Handling of x87 80bits long double
 * 0 : Try to handle 80bits long double as precise as possible (Default)
 * 1 : Handle them as double
 
+#### BOX64_SYNC_ROUNDING *
+Box64 will sync rounding mode with fesetround/fegetround.
+* 0 : Disable rounding mode syncing. (Default.)
+* 1 : Enable rounding mode syncing.
+
 #### BOX64_LIBCEF *
 Detect libcef and apply malloc_hack settings
 * 0 : Don't detect libcef
 * 1 : Detect libcef, and apply MALLOC_HACK=2 if detected (Default)
 
+#### BOX64_SDL2_JGUID *
+Need a workaround for SDL_GetJoystickGUIDInfo function for wrapped SDL2
+* 0 : Don't use any workaround
+* 1 : Use a workaround for program that use the private SDL_GetJoystickGUIDInfo function with 1 missing argument
+
 #### BOX64_LIBGL *
  * libXXXX set the name for libGL (defaults to libGL.so.1).
  * /PATH/TO/libGLXXX : Sets the name and path for libGL
@@ -282,12 +301,17 @@ 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_FUTEX_WAITV *
+Use of the new fuext_waitc syscall
+ * 0 : Do not try to use it, return unsupported (Default for BAD_SIGNAL build)
+ * 1 : let program use the syscall if the host system support it (Default for other build)
+
 #### BOX64_BASH *
 Define x86_64 bash to launch script
  * yyyy
  Will use yyyy as x86_64 bash to launch script. yyyy needs to be a full path to a valid x86_64 version of bash
 
-#### BOX64_ENV
+#### BOX64_ENV *
  * XXX=yyyy
  will add XXX=yyyy env. var.
 
@@ -295,6 +319,10 @@ Define x86_64 bash to launch script
  * XXX=yyyy
  will add XXX=yyyy env. var. and continue with BOX86_ENV2 ... until var doesn't exist
 
+#### BOX64_RESERVE_HIGH
+* 0 : Don't try to pe-reserve high memory (beyond 47bits) (Default)
+* 1 : Try to reserve (without allocating it) memory beyond 47bits (seems unstable)
+
 #### 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.
@@ -305,6 +333,7 @@ Define x86_64 bash to launch script
  * 2 : Launch `gdbserver` when a segfault, bus error or illegal instruction signal is trapped, attached to the offending process, and go in an endless loop, waiting.
  Use `gdb /PATH/TO/box64` and then `target remote 127.0.0.1:1234` to connect to the gdbserver (or use actual IP if not on the machine). After that, the procedure is the same as with ` BOX64_JITGDB=1`.
  This mode can be usefullwhen programs redirect all console output to a file (like Unity3D Games)
+ * 3 : Launch `lldb` when a segfault, bus error or illegal instruction signal is trapped, attached to the offending process and go in an endless loop, waiting.
 
 #### BOX64_NORCFILES
 If the env var exist, no rc files (like /etc/box64.box64rc and ~/.box64rc) will be loaded
@@ -318,10 +347,10 @@ Those variables are only valid inside a rcfile:
  * 0 : Nothing special
  * 1 : Added "--no-sandbox" to command line arguments (usefull for chrome based programs)
 
+#### BOX64_INPROCESSGPU
+ * 0 : Nothing special
+ * 1 : Added "--in-process-gpu" to command line arguments (usefull for chrome based programs)
+
 #### BOX64_EXIT
  * 0 : Nothing special
  * 1 : Just exit, don't try to run the program
-
-#### BOX64_RESERVE_HIGH
-* 0 : Don't try to pe-reserve high memory (beyond 47bits) (Default)
-* 1 : Try to reserve (without allocating it) memory beyond 47bits (seems unstable)

docs/box64.pod

@@ -55,6 +55,11 @@ F<~/.bashrc> may override the setting from your command line. Example:
 When set to 1, add C<--no-sandbox> to command line arguments (useful for chrome
 based programs).
 
+=item B<BOX64_INPROCESSGPU=0|1>
+
+When set to 1, add C<--in-process-gpu> to command line arguments (useful for chrome
+based programs).
+
 =item B<BOX64_EXIT>=I<0|1>
 
 When set to 1, just exit, don't try to run the program.
@@ -146,7 +151,7 @@ Only on build with trace enabled. Trace allow the logging of all instruction exe
 
 =item B<BOX64_TRACE_INIT>=I<0|1>
 
-Use B<BOX64_TRACE_INIT> instead of B<BOX_TRACE> to start trace before the
+Use B<BOX64_TRACE_INIT> instead of B<BOX64_TRACE> to start trace before the
 initialization of libraries and the running program
 
     * 0 : No trace. (Default.)
@@ -371,6 +376,13 @@ Handling of x87 80bits long double
     * 0 : Try to handle 80bits long double as precise as possible (Default)
     * 1 : Handle them as double
 
+=item B<BOX64_SYNC_ROUNDING>=I<0|1>
+
+Box64 will sync rounding mode with fesetround/fegetround.
+
+    * 0 : Disable rounding mode syncing. (Default.)
+    * 1 : Enable rounding mode syncing.
+
 =item B<BOX64_LIBCEF>=I<0|1>
 
 Detect libcef and apply malloc_hack settings
@@ -451,6 +463,13 @@ 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.)
 
+=item B<BOX64_FUTEX_WAITV>=I<0|1>
+
+Use of the new fuext_waitc syscall
+
+     * 0 : Do not try to use it, return unsupported (default for BAD_SIGNAL build)
+    * 1 : let program use the syscall if the host system support it (Default for other build)
+
 =item B<BOX64_BASH>=I<yyyy>
 
 Define x86_64 bash to launch script. Will use yyyy as x86_64 bash to launch

rebuild_wrappers.py

@@ -402,6 +402,7 @@ def readFiles(files: Iterable[Filename]) -> Tuple[JumbledGlobals, JumbledRedirec
 							 or match("libGL", "libEGL") \
 							 or match("libc", "tcmallocminimal") \
 							 or match("libc", "tbbmallocproxy") \
+							 or match("libc", "androidshmem") \
 							 or match("tcmallocminimal", "tbbmallocproxy"):
 								continue
 							

runTest.cmake

@@ -11,6 +11,9 @@ endif( NOT TEST_OUTPUT )
 if( NOT TEST_REFERENCE )
   message( FATAL_ERROR "Require TEST_REFERENCE to be defined" )
 endif( NOT TEST_REFERENCE )
+if(NOT DEFINED ENV{INTERPRETER})
+    set(ENV{INTERPRETER} "")
+endif()
 
 set(ENV{BOX64_LOG} 0)
 set(ENV{BOX64_NOBANNER} 1)
@@ -24,7 +27,7 @@ endif( EXISTS ${CMAKE_SOURCE_DIR}/x64lib )
 
 # run the test program, capture the stdout/stderr and the result var
 execute_process(
-  COMMAND ${TEST_PROGRAM} ${TEST_ARGS} ${TEST_ARGS2}
+  COMMAND $ENV{INTERPRETER} ${TEST_PROGRAM} ${TEST_ARGS} ${TEST_ARGS2}
   OUTPUT_FILE ${TEST_OUTPUT}
   ERROR_VARIABLE TEST_ERROR
   RESULT_VARIABLE TEST_RESULT

src/box64context.c

@@ -68,6 +68,8 @@ void free_tlsdatasize(void* p)
     tlsdatasize_t *data = (tlsdatasize_t*)p;
     box_free(data->ptr);
     box_free(p);
+    if(my_context)
+        pthread_setspecific(my_context->tlskey, NULL);
 }
 
 void x64Syscall(x64emu_t *emu);
@@ -278,6 +280,7 @@ void FreeBox64Context(box64context_t** context)
 
     box64context_t* ctx = *context;   // local copy to do the cleanning
 
+    //clean_current_emuthread();    // cleaning main thread seems a bad idea
     if(ctx->local_maplib)
         FreeLibrarian(&ctx->local_maplib, NULL);
     if(ctx->maplib)
@@ -342,7 +345,6 @@ void FreeBox64Context(box64context_t** context)
     void* ptr;
     if ((ptr = pthread_getspecific(ctx->tlskey)) != NULL) {
         free_tlsdatasize(ptr);
-        pthread_setspecific(ctx->tlskey, NULL);
     }
     pthread_key_delete(ctx->tlskey);
 

src/box64version.h

@@ -3,6 +3,6 @@
 
 #define BOX64_MAJOR 0
 #define BOX64_MINOR 2
-#define BOX64_REVISION 4
+#define BOX64_REVISION 6
 
 #endif //__BOX64_VERSION_H_

src/dynarec/arm64/arm64_emitter.h

@@ -92,24 +92,24 @@
 
 // MOVZ
 #define MOVZ_gen(sf, hw, imm16, Rd)         ((sf)<<31 | 0b10<<29 | 0b100101<<23 | (hw)<<21 | (imm16)<<5 | (Rd))
-#define MOVZx(Rd, imm16)                    EMIT(MOVZ_gen(1, 0, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVZx_LSL(Rd, imm16, shift)         EMIT(MOVZ_gen(1, (shift)/16, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVZw(Rd, imm16)                    EMIT(MOVZ_gen(0, 0, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVZw_LSL(Rd, imm16, shift)         EMIT(MOVZ_gen(0, (shift)/16, ((uint16_t)imm16)&0xffff, Rd))
+#define MOVZx(Rd, imm16)                    EMIT(MOVZ_gen(1, 0, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVZx_LSL(Rd, imm16, shift)         EMIT(MOVZ_gen(1, (shift)/16, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVZw(Rd, imm16)                    EMIT(MOVZ_gen(0, 0, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVZw_LSL(Rd, imm16, shift)         EMIT(MOVZ_gen(0, (shift)/16, ((uint16_t)(imm16))&0xffff, Rd))
 
 // MOVN
 #define MOVN_gen(sf, hw, imm16, Rd)         ((sf)<<31 | 0b00<<29 | 0b100101<<23 | (hw)<<21 | (imm16)<<5 | (Rd))
-#define MOVNx(Rd, imm16)                    EMIT(MOVN_gen(1, 0, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVNx_LSL(Rd, imm16, shift)         EMIT(MOVN_gen(1, (shift)/16, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVNw(Rd, imm16)                    EMIT(MOVN_gen(0, 0, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVNw_LSL(Rd, imm16, shift)         EMIT(MOVN_gen(0, (shift)/16, ((uint16_t)imm16)&0xffff, Rd))
+#define MOVNx(Rd, imm16)                    EMIT(MOVN_gen(1, 0, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVNx_LSL(Rd, imm16, shift)         EMIT(MOVN_gen(1, (shift)/16, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVNw(Rd, imm16)                    EMIT(MOVN_gen(0, 0, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVNw_LSL(Rd, imm16, shift)         EMIT(MOVN_gen(0, (shift)/16, ((uint16_t)(imm16))&0xffff, Rd))
 
 // MOVK
 #define MOVK_gen(sf, hw, imm16, Rd)         ((sf)<<31 | 0b11<<29 | 0b100101<<23 | (hw)<<21 | (imm16)<<5 | (Rd))
-#define MOVKx(Rd, imm16)                    EMIT(MOVK_gen(1, 0, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVKx_LSL(Rd, imm16, shift)         EMIT(MOVK_gen(1, (shift)/16, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVKw(Rd, imm16)                    EMIT(MOVK_gen(0, 0, ((uint16_t)imm16)&0xffff, Rd))
-#define MOVKw_LSL(Rd, imm16, shift)         EMIT(MOVK_gen(0, (shift)/16, ((uint16_t)imm16)&0xffff, Rd))
+#define MOVKx(Rd, imm16)                    EMIT(MOVK_gen(1, 0, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVKx_LSL(Rd, imm16, shift)         EMIT(MOVK_gen(1, (shift)/16, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVKw(Rd, imm16)                    EMIT(MOVK_gen(0, 0, ((uint16_t)(imm16))&0xffff, Rd))
+#define MOVKw_LSL(Rd, imm16, shift)         EMIT(MOVK_gen(0, (shift)/16, ((uint16_t)(imm16))&0xffff, Rd))
 
 // This macro will give a -Wsign-compare warning, probably bug #38341
 #define MOV32w(Rd, imm32) \
@@ -200,6 +200,12 @@
 #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))
 
+// CCMP compare if cond is true, set nzcv if false
+#define CCMP_reg(sf, Rm, cond, Rn, nzcv)    ((sf)<<31 | 1<<30 | 1<<29 | 0b11010010<<21 | (Rm)<<16 | (cond)<<12 | (Rn)<<5 | (nzcv))
+#define CCMPw(Wn, Wm, nzcv, cond)   EMIT(CCMP_reg(0, Wm, cond, Wn, nzcv))
+#define CCMPx(Xn, Xm, nzcv, cond)   EMIT(CCMP_reg(1, Xm, cond, Xn, nzcv))
+#define CCMPxw(Xn, Xm, nzcv, cond)  EMIT(CCMP_reg(rex.w, Xm, cond, Xn, nzcv))
+
 // 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)))
@@ -240,6 +246,7 @@
 #define LDRxw_REG(Rt, Rn, Rm)           EMIT(LDR_REG_gen(0b10+rex.w, Rm, 0b011, 0, Rn, Rt))
 #define LDRz_REG(Rt, Rn, Rm)            EMIT(LDR_REG_gen(rex.is32bits?0b10:0b11, Rm, 0b011, 0, Rn, Rt))
 #define LDRB_REG(Rt, Rn, Rm)            EMIT(LDR_REG_gen(0b00, Rm, 0b011, 0, Rn, Rt))
+#define LDRB_REG_UXTW(Rt, Rn, Rm)       EMIT(LDR_REG_gen(0b00, Rm, 0b010, 0, Rn, Rt))
 #define LDRH_REG(Rt, Rn, Rm)            EMIT(LDR_REG_gen(0b01, Rm, 0b011, 0, Rn, Rt))
 
 #define LDRS_U12_gen(size, op1, opc, imm12, Rn, Rt)    ((size)<<30 | 0b111<<27 | (op1)<<24 | (opc)<<22 | (imm12)<<10 | (Rn)<<5 | (Rt))
@@ -427,6 +434,13 @@
 // Data Memory Barrier
 #define DMB_gen(CRm)                    (0b1101010100<<22 | 0b011<<16 | 0b0011<<12 | (CRm)<<8 | 1<<7 | 0b01<<5 | 0b11111)
 #define DMB_ISH()                       EMIT(DMB_gen(0b1011))
+#define DMB_SY()                        EMIT(DMB_gen(0b1111))
+
+// Data Synchronization Barrier
+#define DSB_gen(CRm)                    (0b1101010100<<22 | 0b011<<16 | 0b0011<<12 | (CRm)<<8 | 1<<7 | 0b00<<5 | 0b11111)
+#define DSB_ISH()                       EMIT(DSB_gen(0b1011))
+#define DSB_ISHST()                     EMIT(DSB_gen(0b1010))
+#define DSB_SY()                        EMIT(DSB_gen(0b1111))
 
 // Break
 #define BRK_gen(imm16)                  (0b11010100<<24 | 0b001<<21 | (((imm16)&0xffff)<<5))
@@ -593,12 +607,12 @@
 #define BFMx(Rd, Rn, immr, imms)        EMIT(BFM_gen(1, 0b01, 1, immr, imms, Rn, Rd))
 #define BFMw(Rd, Rn, immr, imms)        EMIT(BFM_gen(0, 0b01, 0, immr, imms, Rn, Rd))
 #define BFMxw(Rd, Rn, immr, imms)       EMIT(BFM_gen(rex.w, 0b01, rex.w, immr, imms, Rn, Rd))
-#define BFIx(Rd, Rn, lsb, width)        BFMx(Rd, Rn, ((-lsb)%64)&0x3f, (width)-1)
-#define BFIw(Rd, Rn, lsb, width)        BFMw(Rd, Rn, ((-lsb)%32)&0x1f, (width)-1)
+#define BFIx(Rd, Rn, lsb, width)        BFMx(Rd, Rn, ((-(lsb))%64)&0x3f, (width)-1)
+#define BFIw(Rd, Rn, lsb, width)        BFMw(Rd, Rn, ((-(lsb))%32)&0x1f, (width)-1)
 #define BFIxw(Rd, Rn, lsb, width)       if(rex.w) {BFIx(Rd, Rn, lsb, width);} else {BFIw(Rd, Rn, lsb, width);}
-#define BFCx(Rd, lsb, width)            BFMx(Rd, xZR, ((-lsb)%64)&0x3f, (width)-1)
-#define BFCw(Rd, lsb, width)            BFMw(Rd, xZR, ((-lsb)%32)&0x1f, (width)-1)
-#define BFCxw(Rd, lsb, width)           BFMxw(Rd, xZR, rex.w?(((-lsb)%64)&0x3f):(((-lsb)%32)&0x1f), (width)-1)
+#define BFCx(Rd, lsb, width)            BFMx(Rd, xZR, ((-(lsb))%64)&0x3f, (width)-1)
+#define BFCw(Rd, lsb, width)            BFMw(Rd, xZR, ((-(lsb))%32)&0x1f, (width)-1)
+#define BFCxw(Rd, lsb, width)           BFMxw(Rd, xZR, rex.w?(((-(lsb))%64)&0x3f):(((-(lsb))%32)&0x1f), (width)-1)
 // Insert lsb:width part of Rn into low part of Rd (leaving rest of Rd untouched)
 #define BFXILx(Rd, Rn, lsb, width)      EMIT(BFM_gen(1, 0b01, 1, (lsb), (lsb)+(width)-1, Rn, Rd))
 // Insert lsb:width part of Rn into low part of Rd (leaving rest of Rd untouched)
@@ -637,9 +651,9 @@
 #define SBFMx(Rd, Rn, immr, imms)       EMIT(SBFM_gen(1, 1, immr, imms, Rn, Rd))
 #define SBFMw(Rd, Rn, immr, imms)       EMIT(SBFM_gen(0, 0, immr, imms, Rn, Rd))
 #define SBFMxw(Rd, Rn, immr, imms)      EMIT(SBFM_gen(rex.w, rex.w, immr, imms, Rn, Rd))
-#define SBFXx(Rd, Rn, lsb, width)       SBFMx(Rd, Rn, lsb, lsb+width-1)
-#define SBFXw(Rd, Rn, lsb, width)       SBFMw(Rd, Rn, lsb, lsb+width-1)
-#define SBFXxw(Rd, Rn, lsb, width)      SBFMxw(Rd, Rn, lsb, lsb+width-1)
+#define SBFXx(Rd, Rn, lsb, width)       SBFMx(Rd, Rn, lsb, (lsb)+(width)-1)
+#define SBFXw(Rd, Rn, lsb, width)       SBFMw(Rd, Rn, lsb, (lsb)+(width)-1)
+#define SBFXxw(Rd, Rn, lsb, width)      SBFMxw(Rd, Rn, lsb, (lsb)+(width)-1)
 #define SXTBx(Rd, Rn)                   SBFMx(Rd, Rn, 0, 7)
 #define SXTBw(Rd, Rn)                   SBFMw(Rd, Rn, 0, 7)
 #define SXTHx(Rd, Rn)                   SBFMx(Rd, Rn, 0, 15)
@@ -783,13 +797,13 @@
 // imm16 must be 4-aligned
 #define VLDR128_U12(Qt, Rn, imm16)          EMIT(VMEM_gen(0b00, 0b11, ((uint32_t)((imm16)>>4))&0xfff, Rn, Qt))
 // (imm14) must be 3-aligned
-#define VSTR32_U12(Dt, Rn, imm14)           EMIT(VMEM_gen(0b10, 0b00, ((uint32_t)(imm14>>2))&0xfff, Rn, Dt))
+#define VSTR32_U12(Dt, Rn, imm14)           EMIT(VMEM_gen(0b10, 0b00, ((uint32_t)((imm14)>>2))&0xfff, Rn, Dt))
 // (imm15) must be 3-aligned
-#define VSTR64_U12(Dt, Rn, imm15)           EMIT(VMEM_gen(0b11, 0b00, ((uint32_t)(imm15>>3))&0xfff, Rn, Dt))
+#define VSTR64_U12(Dt, Rn, imm15)           EMIT(VMEM_gen(0b11, 0b00, ((uint32_t)((imm15)>>3))&0xfff, Rn, Dt))
 // imm16 must be 4-aligned
 #define VSTR128_U12(Qt, Rn, imm16)          EMIT(VMEM_gen(0b00, 0b10, ((uint32_t)((imm16)>>4))&0xfff, Rn, Qt))
 // (imm13) must be 1-aligned
-#define VSTR16_U12(Ht, Rn, imm13)           EMIT(VMEM_gen(0b01, 0b00, ((uint32_t)(imm13>>1))&0xfff, Rn, Ht))
+#define VSTR16_U12(Ht, Rn, imm13)           EMIT(VMEM_gen(0b01, 0b00, ((uint32_t)((imm13)>>1))&0xfff, Rn, Ht))
 
 #define VMEMUR_vector(size, opc, imm9, Rn, Rt)  ((size)<<30 | 0b111<<27 | 1<<26 | (opc)<<22 | (imm9)<<12 | (Rn)<<5 | (Rt))
 // signed offset, no alignement!
@@ -951,6 +965,15 @@
 #define VSRI_16(Vd, Vn, shift)              EMIT(VSRI_vector(0, 0b0010 | ((shift)>>3)&1, (shift)&7, Vn, Vd))
 #define VSRI_32(Vd, Vn, shift)              EMIT(VSRI_vector(0, 0b0100 | (((shift)>>3)&3), (shift)&7, Vn, Vd))
 
+// Reverse elements in 64-bit doublewords (vector). This instruction reverses the order of 8-bit, 16-bit, or 32-bit elements in each doubleword
+#define VREVx_vector(Q, U, size, o0, Rn, Rd)    ((Q)<<30 | (U)<<29 | 0b01110<<24 | (size)<<22 | 0b10000<<17 | (o0)<<12 | 0b10<<10 | (Rn)<<5 | (Rd))
+#define VREV64_32(Vd, Vn)                   EMIT(VREVx_vector(0, 0, 0b10, 0, Vn, Vd))
+#define VREV64_16(Vd, Vn)                   EMIT(VREVx_vector(0, 0, 0b01, 0, Vn, Vd))
+#define VREV64_8(Vd, Vn)                    EMIT(VREVx_vector(0, 0, 0b00, 0, Vn, Vd))
+#define VREV64Q_32(Vd, Vn)                  EMIT(VREVx_vector(1, 0, 0b10, 0, Vn, Vd))
+#define VREV64Q_16(Vd, Vn)                  EMIT(VREVx_vector(1, 0, 0b01, 0, Vn, Vd))
+#define VREV64Q_8(Vd, Vn)                   EMIT(VREVx_vector(1, 0, 0b00, 0, Vn, Vd))
+
 // Integer MATH
 #define ADDSUB_vector(Q, U, size, Rm, Rn, Rd)   ((Q)<<30 | (U)<<29 | 0b01110<<24 | (size)<<22 | 1<<21 | (Rm)<<16 | 0b10000<<11 | 1<<10 | (Rn)<<5 | (Rd))
 #define VADDQ_8(Vd, Vn, Vm)                 EMIT(ADDSUB_vector(1, 0, 0b00, Vm, Vn, Vd))
@@ -1094,6 +1117,12 @@
 #define FADDP_vector(Q, sz, Rm, Rn, Rd) ((Q)<<30 | 1<<29 | 0b01110<<24 | (sz)<<22 | 1<<21 | (Rm)<<16 | 0b11010<<11 | 1<<10 | (Rn)<<5 | (Rd))
 #define VFADDPQS(Vd, Vn, Vm)        EMIT(FADDP_vector(1, 0, Vm, Vn, Vd))
 #define VFADDPQD(Vd, Vn, Vm)        EMIT(FADDP_vector(1, 1, Vm, Vn, Vd))
+#define VFADDPS(Vd, Vn, Vm)         EMIT(FADDP_vector(0, 0, Vm, Vn, Vd))
+#define VFADDPD(Vd, Vn, Vm)         EMIT(FADDP_vector(0, 1, Vm, Vn, Vd))
+
+#define FADDP_scalar(sz, Rn, Rd)        (0b01<<30 | 1<<29 | 0b11110<<24 | (sz)<<22 | 0b11000<<17 | 0b01101<<12 | 0b10<<10 | (Rn)<<5 | (Rd))
+#define FADDPS(Sd, Sn)              EMIT(FADDP_scalar(0, Sn, Sd))
+#define FADDPD(Dd, Dn)              EMIT(FADDP_scalar(1, Dn, Dd))
 
 // NEG / ABS
 #define FNEGABS_scalar(type, opc, Rn, Rd)  (0b11110<<24 | (type)<<22 | 1<<21 | (opc)<<15 | 0b10000<<10 | (Rn)<<5 | (Rd))
@@ -1360,22 +1389,12 @@
 #define VFRINTRDQ(Vd,Vn, mode)      EMIT(FRINT_vector(1, 0, (mode)&1, 1, ((mode)>>1)&1, Vn, Vd))
 // round with mode, mode is 0 = TieEven, 1=+inf, 2=-inf, 3=zero
 #define VFRINTRSQ(Vd,Vn, mode)      EMIT(FRINT_vector(1, 0, (mode)&1, 0, ((mode)>>1)&1, Vn, Vd))
+#define VFRINTRS(Vd, Vn, mode)      EMIT(FRINT_vector(0, 0, (mode)&1, 0, ((mode)>>1)&1, Vn, Vd))
 
 #define FRINTI_scalar(type, Rn, Rd)  (0b11110<<24 | (type)<<22 | 1<<21 | 0b001<<18 | 0b111<<15 | 0b10000<<10 | (Rn)<<5 | (Rd))
 #define FRINTIS(Sd, Sn)             EMIT(FRINTI_scalar(0b00, Sn, Sd))
 #define FRINTID(Dd, Dn)             EMIT(FRINTI_scalar(0b01, Dn, Dd))
 
-
-#define FRINTxx_scalar(type, op, Rn, Rd)  (0b11110<<24 | (type)<<22 | 1<<21 | 0b0100<<17 | (op)<<15 | 0b10000<<10 | (Rn)<<5 | (Rd))
-#define FRINT32ZS(Sd, Sn)           EMIT(FRINTxx_scalar(0b00, 0b00, Sn, Sd))
-#define FRINT32ZD(Dd, Dn)           EMIT(FRINTxx_scalar(0b01, 0b00, Dn, Dd))
-#define FRINT32XS(Sd, Sn)           EMIT(FRINTxx_scalar(0b00, 0b01, Sn, Sd))
-#define FRINT32XD(Dd, Dn)           EMIT(FRINTxx_scalar(0b01, 0b01, Dn, Dd))
-#define FRINT64ZS(Sd, Sn)           EMIT(FRINTxx_scalar(0b00, 0b10, Sn, Sd))
-#define FRINT64ZD(Dd, Dn)           EMIT(FRINTxx_scalar(0b01, 0b10, Dn, Dd))
-#define FRINT64XS(Sd, Sn)           EMIT(FRINTxx_scalar(0b00, 0b11, Sn, Sd))
-#define FRINT64XD(Dd, Dn)           EMIT(FRINTxx_scalar(0b01, 0b11, Dn, Dd))
-
 #define FRINT_scalar(type, rmode, Rn, Rd)   (0b11110<<24 | (type)<<22 | 1<<21 | 0b001<<18 | (rmode)<<15 | 0b10000<<10 | (Rn)<<5 | (Rd))
 // round toward 0 (truncate)
 #define FRINTZS(Sd, Sn)             EMIT(FRINT_scalar(0b00, 0b011, Sn, Sd))
@@ -1490,7 +1509,6 @@
 #define VTRNQ1_32(Vd, Vn, Vm)       EMIT(TRN_gen(1, 0b10, Vm, 0, Vn, Vd))
 #define VTRNQ1_16(Vd, Vn, Vm)       EMIT(TRN_gen(1, 0b01, Vm, 0, Vn, Vd))
 #define VTRNQ1_8(Vd, Vn, Vm)        EMIT(TRN_gen(1, 0b00, Vm, 0, Vn, Vd))
-#define VSWP(Vd, Vn)                VTRNQ1_64(Vd, Vn, Vn)
 #define VTRNQ2_64(Vd, Vn, Vm)       EMIT(TRN_gen(1, 0b11, Vm, 1, Vn, Vd))
 #define VTRNQ2_32(Vd, Vn, Vm)       EMIT(TRN_gen(1, 0b10, Vm, 1, Vn, Vd))
 #define VTRNQ2_16(Vd, Vn, Vm)       EMIT(TRN_gen(1, 0b01, Vm, 1, Vn, Vd))
@@ -1599,6 +1617,15 @@
 #define CMEQQ_0_16(Rd, Rn)          EMIT(CMEQ_0_vector(1, 0b01, Rn, Rd))
 #define CMEQQ_0_32(Rd, Rn)          EMIT(CMEQ_0_vector(1, 0b10, Rn, Rd))
 #define CMEQQ_0_64(Rd, Rn)          EMIT(CMEQ_0_vector(1, 0b11, Rn, Rd))
+// Greater Than 0
+#define CMCond_0_vector(Q, U, size, op, Rn, Rd) ((Q)<<30 | (U)<<29 | 0b01110<<24 | (size)<<22 | 0b10000<<17 | 0b0100<<13 | (op)<<12 | 0b10<<10 | (Rn)<<5 | (Rd))
+#define CMGT_0_8(Rd, Rn)            EMIT(CMCond_0_vector(0, 0, 0b00, 0, Rn, Rd))
+#define CMGT_0_16(Rd, Rn)           EMIT(CMCond_0_vector(0, 0, 0b01, 0, Rn, Rd))
+#define CMGT_0_32(Rd, Rn)           EMIT(CMCond_0_vector(0, 0, 0b10, 0, Rn, Rd))
+#define CMGTQ_0_8(Rd, Rn)           EMIT(CMCond_0_vector(1, 0, 0b00, 0, Rn, Rd))
+#define CMGTQ_0_16(Rd, Rn)          EMIT(CMCond_0_vector(1, 0, 0b01, 0, Rn, Rd))
+#define CMGTQ_0_32(Rd, Rn)          EMIT(CMCond_0_vector(1, 0, 0b10, 0, Rn, Rd))
+#define CMGTQ_0_64(Rd, Rn)          EMIT(CMCond_0_vector(1, 0, 0b11, 0, Rn, Rd))
 
 // Vector Float CMP
 // EQual
@@ -1757,9 +1784,11 @@
 // MOV Immediate
 #define MOVI_vector(Q, op, abc, cmode, defgh, Rd)   ((Q)<<30 | (op)<<29 | 0b0111100000<<19 | (abc)<<16 | (cmode)<<12 | 1<<10 | (defgh)<<5 | (Rd))
 #define MOVIQ_8(Rd, imm8)           EMIT(MOVI_vector(1, 0, (((imm8)>>5)&0b111), 0b1110, ((imm8)&0b11111), Rd))
+#define MOVIQ_16(Rd, imm8, lsl8)    EMIT(MOVI_vector(1, 0, (((imm8)>>5)&0b111), 0b1000|((lsl8)?0b10:0), ((imm8)&0b11111), Rd))
 #define MOVI_8(Rd, imm8)            EMIT(MOVI_vector(0, 0, (((imm8)>>5)&0b111), 0b1110, ((imm8)&0b11111), Rd))
-#define MOVI_16(Rd, imm8)           EMIT(MOVI_vector(0, 0, (((imm8)>>5)&0b111), 0b1000, ((imm8)&0b11111), Rd))
+#define MOVI_16(Rd, imm8, lsl8)     EMIT(MOVI_vector(0, 0, (((imm8)>>5)&0b111), 0b1000|((lsl8)?0b10:0), ((imm8)&0b11111), Rd))
 #define MOVI_32(Rd, imm8)           EMIT(MOVI_vector(0, 0, (((imm8)>>5)&0b111), 0b0000, ((imm8)&0b11111), Rd))
+#define MOVI_64(Rd, imm8)           EMIT(MOVI_vector(0, 1, (((imm8)>>5)&0b111), 0b1110, ((imm8)&0b11111), Rd))
 
 // SHLL and eXtend Long
 #define SHLL_vector(Q, U, immh, immb, Rn, Rd)  ((Q)<<30 | (U)<<29 | 0b011110<<23 | (immh)<<19 | (immb)<<16 | 0b10100<<11 | 1<<10 | (Rn)<<5 | (Rd))
@@ -1919,4 +1948,272 @@
 #define PMULL_128(Rd, Rn, Rm)   EMIT(PMULL_gen(0, 0b11, Rm, Rn, Rd))
 #define PMULL2_128(Rd, Rn, Rm)  EMIT(PMULL_gen(1, 0b11, Rm, Rn, Rd))
 
+// ATOMIC extension
+#define ATOMIC_gen(size, A, R, Rs, opc, Rn, Rt) ((size)<<30 | 0b111<<27 | (A)<<23 | (R)<<22 | 1<<21 | (Rs)<<16 | (opc)<<12 | (Rn)<<5 | (Rt))
+// Atomic ADD
+#define LDADDxw(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b000, Rn, Rt))
+#define LDADDAxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 1, 0, Rs, 0b000, Rn, Rt))
+#define LDADDALxw(Rs, Rt, Rn)           EMIT(ATOMIC_gen(0b10+rex.w, 1, 1, Rs, 0b000, Rn, Rt))
+#define LDADDLxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b000, Rn, Rt))
+#define STADDxw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b000, Rn, 0b11111))
+#define STADDLxw(Rs, Rn)                EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b000, Rn, 0b11111))
+#define LDADDB(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b000, Rn, Rt))
+#define LDADDAB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 1, 0, Rs, 0b000, Rn, Rt))
+#define LDADDALB(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b00, 1, 1, Rs, 0b000, Rn, Rt))
+#define LDADDLB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b000, Rn, Rt))
+#define STADDB(Rs, Rn)                  EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b000, Rn, 0b11111))
+#define STADDLB(Rs, Rn)                 EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b000, Rn, 0b11111))
+#define LDADDH(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b000, Rn, Rt))
+#define LDADDAH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 1, 0, Rs, 0b000, Rn, Rt))
+#define LDADDALH(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b01, 1, 1, Rs, 0b000, Rn, Rt))
+#define LDADDLH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b000, Rn, Rt))
+#define STADDH(Rs, Rn)                  EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b000, Rn, 0b11111))
+#define STADDLH(Rs, Rn)                 EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b000, Rn, 0b11111))
+// Atomic AND with complement (i.e. BIC)
+#define LDCLRxw(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b001, Rn, Rt))
+#define LDCLRAxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 1, 0, Rs, 0b001, Rn, Rt))
+#define LDCLRALxw(Rs, Rt, Rn)           EMIT(ATOMIC_gen(0b10+rex.w, 1, 1, Rs, 0b001, Rn, Rt))
+#define LDCLRLxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b001, Rn, Rt))
+#define STCLRxw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b001, Rn, 0b11111))
+#define STCLRLxw(Rs, Rn)                EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b001, Rn, 0b11111))
+#define LDCLRB(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b001, Rn, Rt))
+#define LDCLRAB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 1, 0, Rs, 0b001, Rn, Rt))
+#define LDCLRALB(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b00, 1, 1, Rs, 0b001, Rn, Rt))
+#define LDCLRLB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b001, Rn, Rt))
+#define STCLRB(Rs, Rn)                  EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b001, Rn, 0b11111))
+#define STCLRLB(Rs, Rn)                 EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b001, Rn, 0b11111))
+#define LDCLRH(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b001, Rn, Rt))
+#define LDCLRAH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 1, 0, Rs, 0b001, Rn, Rt))
+#define LDCLRALH(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b01, 1, 1, Rs, 0b001, Rn, Rt))
+#define LDCLRLH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b001, Rn, Rt))
+#define STCLRH(Rs, Rn)                  EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b001, Rn, 0b11111))
+#define STCLRLH(Rs, Rn)                 EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b001, Rn, 0b11111))
+// Atomic EOR
+#define LDEORxw(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b010, Rn, Rt))
+#define LDEORAxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 1, 0, Rs, 0b010, Rn, Rt))
+#define LDEORALxw(Rs, Rt, Rn)           EMIT(ATOMIC_gen(0b10+rex.w, 1, 1, Rs, 0b010, Rn, Rt))
+#define LDEORLxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b010, Rn, Rt))
+#define STEORxw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b010, Rn, 0b11111))
+#define STEORLxw(Rs, Rn)                EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b010, Rn, 0b11111))
+#define LDEORB(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b010, Rn, Rt))
+#define LDEORAB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 1, 0, Rs, 0b010, Rn, Rt))
+#define LDEORALB(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b00, 1, 1, Rs, 0b010, Rn, Rt))
+#define LDEORLB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b010, Rn, Rt))
+#define STEORB(Rs, Rn)                  EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b010, Rn, 0b11111))
+#define STEORLB(Rs, Rn)                 EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b010, Rn, 0b11111))
+#define LDEORH(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b010, Rn, Rt))
+#define LDEORAH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 1, 0, Rs, 0b010, Rn, Rt))
+#define LDEORALH(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b01, 1, 1, Rs, 0b010, Rn, Rt))
+#define LDEORLH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b010, Rn, Rt))
+#define STEORH(Rs, Rn)                  EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b010, Rn, 0b11111))
+#define STEORLH(Rs, Rn)                 EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b010, Rn, 0b11111))
+// Atomic OR
+#define LDSETxw(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b011, Rn, Rt))
+#define LDSETAxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 1, 0, Rs, 0b011, Rn, Rt))
+#define LDSETALxw(Rs, Rt, Rn)           EMIT(ATOMIC_gen(0b10+rex.w, 1, 1, Rs, 0b011, Rn, Rt))
+#define LDSETLxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b011, Rn, Rt))
+#define STSETxw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b011, Rn, 0b11111))
+#define STSETLxw(Rs, Rn)                EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b011, Rn, 0b11111))
+#define LDSETB(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b011, Rn, Rt))
+#define LDSETAB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 1, 0, Rs, 0b011, Rn, Rt))
+#define LDSETALB(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b00, 1, 1, Rs, 0b011, Rn, Rt))
+#define LDSETLB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b011, Rn, Rt))
+#define STSETB(Rs, Rn)                  EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b011, Rn, 0b11111))
+#define STSETLB(Rs, Rn)                 EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b011, Rn, 0b11111))
+#define LDSETH(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b011, Rn, Rt))
+#define LDSETAH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 1, 0, Rs, 0b011, Rn, Rt))
+#define LDSETALH(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b01, 1, 1, Rs, 0b011, Rn, Rt))
+#define LDSETLH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b011, Rn, Rt))
+#define STSETH(Rs, Rn)                  EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b011, Rn, 0b11111))
+#define STSETLH(Rs, Rn)                 EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b011, Rn, 0b11111))
+// Atomic Signel Max
+#define LDSMAXxw(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b100, Rn, Rt))
+#define LDSMAXAxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 1, 0, Rs, 0b100, Rn, Rt))
+#define LDSMAXALxw(Rs, Rt, Rn)           EMIT(ATOMIC_gen(0b10+rex.w, 1, 1, Rs, 0b100, Rn, Rt))
+#define LDSMAXLxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b100, Rn, Rt))
+#define STSMAXxw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b100, Rn, 0b11111))
+#define STSMAXLxw(Rs, Rn)                EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b100, Rn, 0b11111))
+#define LDSMAXB(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b100, Rn, Rt))
+#define LDSMAXAB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 1, 0, Rs, 0b100, Rn, Rt))
+#define LDSMAXALB(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b00, 1, 1, Rs, 0b100, Rn, Rt))
+#define LDSMAXLB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b100, Rn, Rt))
+#define STSMAXB(Rs, Rn)                  EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b100, Rn, 0b11111))
+#define STSMAXLB(Rs, Rn)                 EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b100, Rn, 0b11111))
+#define LDSMAXH(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b100, Rn, Rt))
+#define LDSMAXAH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 1, 0, Rs, 0b100, Rn, Rt))
+#define LDSMAXALH(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b01, 1, 1, Rs, 0b100, Rn, Rt))
+#define LDSMAXLH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b100, Rn, Rt))
+#define STSMAXH(Rs, Rn)                  EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b100, Rn, 0b11111))
+#define STSMAXLH(Rs, Rn)                 EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b100, Rn, 0b11111))
+// Atomic Signed Min
+#define LDSMINxw(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b101, Rn, Rt))
+#define LDSMINAxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 1, 0, Rs, 0b101, Rn, Rt))
+#define LDSMINALxw(Rs, Rt, Rn)           EMIT(ATOMIC_gen(0b10+rex.w, 1, 1, Rs, 0b101, Rn, Rt))
+#define LDSMINLxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b101, Rn, Rt))
+#define STSMINxw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b101, Rn, 0b11111))
+#define STSMINLxw(Rs, Rn)                EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b101, Rn, 0b11111))
+#define LDSMINB(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b101, Rn, Rt))
+#define LDSMINAB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 1, 0, Rs, 0b101, Rn, Rt))
+#define LDSMINALB(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b00, 1, 1, Rs, 0b101, Rn, Rt))
+#define LDSMINLB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b101, Rn, Rt))
+#define STSMINB(Rs, Rn)                  EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b101, Rn, 0b11111))
+#define STSMINLB(Rs, Rn)                 EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b101, Rn, 0b11111))
+#define LDSMINH(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b101, Rn, Rt))
+#define LDSMINAH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 1, 0, Rs, 0b101, Rn, Rt))
+#define LDSMINALH(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b01, 1, 1, Rs, 0b101, Rn, Rt))
+#define LDSMINLH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b101, Rn, Rt))
+#define STSMINH(Rs, Rn)                  EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b101, Rn, 0b11111))
+#define STSMINLH(Rs, Rn)                 EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b101, Rn, 0b11111))
+// Atomic Unsigned Max
+#define LDUMAXxw(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b110, Rn, Rt))
+#define LDUMAXAxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 1, 0, Rs, 0b110, Rn, Rt))
+#define LDUMAXALxw(Rs, Rt, Rn)           EMIT(ATOMIC_gen(0b10+rex.w, 1, 1, Rs, 0b110, Rn, Rt))
+#define LDUMAXLxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b110, Rn, Rt))
+#define STUMAXxw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b110, Rn, 0b11111))
+#define STUMAXLxw(Rs, Rn)                EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b110, Rn, 0b11111))
+#define LDUMAXB(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b110, Rn, Rt))
+#define LDUMAXAB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 1, 0, Rs, 0b110, Rn, Rt))
+#define LDUMAXALB(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b00, 1, 1, Rs, 0b110, Rn, Rt))
+#define LDUMAXLB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b110, Rn, Rt))
+#define STUMAXB(Rs, Rn)                  EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b110, Rn, 0b11111))
+#define STUMAXLB(Rs, Rn)                 EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b110, Rn, 0b11111))
+#define LDUMAXH(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b110, Rn, Rt))
+#define LDUMAXAH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 1, 0, Rs, 0b110, Rn, Rt))
+#define LDUMAXALH(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b01, 1, 1, Rs, 0b110, Rn, Rt))
+#define LDUMAXLH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b110, Rn, Rt))
+#define STUMAXH(Rs, Rn)                  EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b110, Rn, 0b11111))
+#define STUMAXLH(Rs, Rn)                 EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b110, Rn, 0b11111))
+// Atomic Unsigned Min
+#define LDUMINxw(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b111, Rn, Rt))
+#define LDUMINAxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 1, 0, Rs, 0b111, Rn, Rt))
+#define LDUMINALxw(Rs, Rt, Rn)           EMIT(ATOMIC_gen(0b10+rex.w, 1, 1, Rs, 0b111, Rn, Rt))
+#define LDUMINLxw(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b111, Rn, Rt))
+#define STUMINxw(Rs, Rn)                 EMIT(ATOMIC_gen(0b10+rex.w, 0, 0, Rs, 0b111, Rn, 0b11111))
+#define STUMINLxw(Rs, Rn)                EMIT(ATOMIC_gen(0b10+rex.w, 0, 1, Rs, 0b111, Rn, 0b11111))
+#define LDUMINB(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b111, Rn, Rt))
+#define LDUMINAB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 1, 0, Rs, 0b111, Rn, Rt))
+#define LDUMINALB(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b00, 1, 1, Rs, 0b111, Rn, Rt))
+#define LDUMINLB(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b111, Rn, Rt))
+#define STUMINB(Rs, Rn)                  EMIT(ATOMIC_gen(0b00, 0, 0, Rs, 0b111, Rn, 0b11111))
+#define STUMINLB(Rs, Rn)                 EMIT(ATOMIC_gen(0b00, 0, 1, Rs, 0b111, Rn, 0b11111))
+#define LDUMINH(Rs, Rt, Rn)              EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b111, Rn, Rt))
+#define LDUMINAH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 1, 0, Rs, 0b111, Rn, Rt))
+#define LDUMINALH(Rs, Rt, Rn)            EMIT(ATOMIC_gen(0b01, 1, 1, Rs, 0b111, Rn, Rt))
+#define LDUMINLH(Rs, Rt, Rn)             EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b111, Rn, Rt))
+#define STUMINH(Rs, Rn)                  EMIT(ATOMIC_gen(0b01, 0, 0, Rs, 0b111, Rn, 0b11111))
+#define STUMINLH(Rs, Rn)                 EMIT(ATOMIC_gen(0b01, 0, 1, Rs, 0b111, Rn, 0b11111))
+
+// SWAPxx(Xs, Xt, Xn) : [Xn] => Xt, Xs => [Xn]
+#define SWAP_gen(size, A, R, Rs, Rn, Rt)    ((size)<<30 | 0b111<<27 | (A)<<23 | (R)<<22 | 1<<21 | (Rs)<<16 | 1<<15 | (Rn)<<5 | (Rt))
+#define SWPxw(Rs, Rt, Rn)               EMIT(SWAP_gen(0b10+rex.w, 0, 0, Rs, Rn, Rt))
+#define SWPAxw(Rs, Rt, Rn)              EMIT(SWAP_gen(0b10+rex.w, 1, 0, Rs, Rn, Rt))
+#define SWPALxw(Rs, Rt, Rn)             EMIT(SWAP_gen(0b10+rex.w, 1, 1, Rs, Rn, Rt))
+#define SWPLxw(Rs, Rt, Rn)              EMIT(SWAP_gen(0b10+rex.w, 0, 1, Rs, Rn, Rt))
+#define SWPB(Rs, Rt, Rn)                EMIT(SWAP_gen(0b00, 0, 0, Rs, Rn, Rt))
+#define SWPAB(Rs, Rt, Rn)               EMIT(SWAP_gen(0b00, 1, 0, Rs, Rn, Rt))
+#define SWPALB(Rs, Rt, Rn)              EMIT(SWAP_gen(0b00, 1, 1, Rs, Rn, Rt))
+#define SWPLB(Rs, Rt, Rn)               EMIT(SWAP_gen(0b00, 0, 1, Rs, Rn, Rt))
+#define SWPH(Rs, Rt, Rn)                EMIT(SWAP_gen(0b01, 0, 0, Rs, Rn, Rt))
+#define SWPAH(Rs, Rt, Rn)               EMIT(SWAP_gen(0b01, 1, 0, Rs, Rn, Rt))
+#define SWPALH(Rs, Rt, Rn)              EMIT(SWAP_gen(0b01, 1, 1, Rs, Rn, Rt))
+#define SWPLH(Rs, Rt, Rn)               EMIT(SWAP_gen(0b01, 0, 1, Rs, Rn, Rt))
+
+#define CAS_gen(size, L, Rs, O0, Rn, Rt)    ((size)<<30 | 0b001000<<24 | 1<<23 | (L)<<22 | 1<<21 | (Rs)<<16 | (O0)<<15 | 0b11111<<10 | (Rn)<<5 | (Rt))
+// Compare and Swap compare Rs with [Rn], write Rt is same, return old [Rn] in Rs
+#define CASxw(Rs, Rt, Rn)               EMIT(CAS_gen(0b10+rex.w, 0, Rs, 0, Rn, Rt))
+#define CASAxw(Rs, Rt, Rn)              EMIT(CAS_gen(0b10+rex.w, 1, Rs, 0, Rn, Rt))
+#define CASALxw(Rs, Rt, Rn)             EMIT(CAS_gen(0b10+rex.w, 1, Rs, 1, Rn, Rt))
+#define CASLxw(Rs, Rt, Rn)              EMIT(CAS_gen(0b10+rex.w, 0, Rs, 1, Rn, Rt))
+#define CASB(Rs, Rt, Rn)                EMIT(CAS_gen(0b00, 0, Rs, 0, Rn, Rt))
+#define CASAB(Rs, Rt, Rn)               EMIT(CAS_gen(0b00, 1, Rs, 0, Rn, Rt))
+#define CASALB(Rs, Rt, Rn)              EMIT(CAS_gen(0b00, 1, Rs, 1, Rn, Rt))
+#define CASLB(Rs, Rt, Rn)               EMIT(CAS_gen(0b00, 0, Rs, 1, Rn, Rt))
+#define CASH(Rs, Rt, Rn)                EMIT(CAS_gen(0b01, 0, Rs, 0, Rn, Rt))
+#define CASAH(Rs, Rt, Rn)               EMIT(CAS_gen(0b01, 1, Rs, 0, Rn, Rt))
+#define CASALH(Rs, Rt, Rn)              EMIT(CAS_gen(0b01, 1, Rs, 1, Rn, Rt))
+#define CASLH(Rs, Rt, Rn)               EMIT(CAS_gen(0b01, 0, Rs, 1, Rn, Rt))
+
+#define CASP_gen(size, L, Rs, O0, Rn, Rt)   ((size)<<30 | 0b001000<<24 | 0<<23 | (L)<<22 | 1<<21 | (Rs)<<16 | (O0)<<15 | 0b11111<<10 | (Rn)<<5 | (Rt))
+// Compare and Swap with pair, Rs, Rs+1 and Rt,Rt+1 with [Rt]
+#define CASPxw(Rs, Rt, Rn)              EMIT(CASP_gen(0b00+rex.w, 0, Rs, 0, Rn, Rt))
+#define CASPAxw(Rs, Rt, Rn)             EMIT(CASP_gen(0b00+rex.w, 1, Rs, 0, Rn, Rt))
+#define CASPALxw(Rs, Rt, Rn)            EMIT(CASP_gen(0b00+rex.w, 1, Rs, 1, Rn, Rt))
+#define CASPLxw(Rs, Rt, Rn)             EMIT(CASP_gen(0b00+rex.w, 0, Rs, 1, Rn, Rt))
+
+// FLAGM extension
+// Invert Carry Flag
+#define CFINV()             EMIT(0b1101010100<<22 | 0b0100<<12 | 0b000<<5 | 0b11111)
+
+#define RMIF_gen(imm6, Rn, mask)        (0b10111010000<<21 | (imm6)<<15 | 0b00001<<10 | (Rn)<<5 | (mask))
+// Rotate right reg and use as NZCV
+#define RMIF(Xn, shift, mask)           EMIT(RMIF_gen(shift, Xn, mask))
+
+#define SETF_gen(sz, Rn)                (0b00111010000<<21 | (sz)<<14 | 0b0010<<10 | (Rn)<<5 | 0b1101)
+// Set NZVc with 8bit value of reg: N=bit7, Z=[0..7]==0, V=bit8 eor bit7, C unchanged
+#define SETF8(Wn)                       EMIT(SETF_gen(0, Wn))
+// Set NZVc with 16bit value of reg: N=bit15, Z=[0..15]==0, V=bit16 eor bit15, C unchanged
+#define SETF16(Wn)                      EMIT(SETF_gen(1, Wn))
+
+// FLAGM2 extension
+// NZCV -> N=0 Z=C|V C=C&!V V=0
+#define AXFLAG()            EMIT(0b1101010100<<22 | 0b0100<<12 | 0b010<<5 | 0b11111)
+// NZCV -> N=!C&!Z Z=Z&C C=C|Z V=!C&Z
+#define XAFLAG()            EMIT(0b1101010100<<22 | 0b0100<<12 | 0b001<<5 | 0b11111)
+
+// FRINTTS extension
+#define FRINTxx_scalar(type, op, Rn, Rd)  (0b11110<<24 | (type)<<22 | 1<<21 | 0b0100<<17 | (op)<<15 | 0b10000<<10 | (Rn)<<5 | (Rd))
+#define FRINT32ZS(Sd, Sn)           EMIT(FRINTxx_scalar(0b00, 0b00, Sn, Sd))
+#define FRINT32ZD(Dd, Dn)           EMIT(FRINTxx_scalar(0b01, 0b00, Dn, Dd))
+#define FRINT32XS(Sd, Sn)           EMIT(FRINTxx_scalar(0b00, 0b01, Sn, Sd))
+#define FRINT32XD(Dd, Dn)           EMIT(FRINTxx_scalar(0b01, 0b01, Dn, Dd))
+#define FRINT64ZS(Sd, Sn)           EMIT(FRINTxx_scalar(0b00, 0b10, Sn, Sd))
+#define FRINT64ZD(Dd, Dn)           EMIT(FRINTxx_scalar(0b01, 0b10, Dn, Dd))
+#define FRINT64XS(Sd, Sn)           EMIT(FRINTxx_scalar(0b00, 0b11, Sn, Sd))
+#define FRINT64XD(Dd, Dn)           EMIT(FRINTxx_scalar(0b01, 0b11, Dn, Dd))
+
+// CRC32 extension
+#define CRC32C_gen(sf, Rm, sz, Rn, Rd)  ((sf)<<31 | 0b11010110<<21 | (Rm)<<16 | 0b010<<13 | 1<<12 | (sz)<<10 | (Rn)<<5 | (Rd))
+#define CRC32CB(Wd, Wn, Wm)         EMIT(CRC32C_gen(0, Wm, 0b00, Wn, Wd))
+#define CRC32CH(Wd, Wn, Wm)         EMIT(CRC32C_gen(0, Wm, 0b01, Wn, Wd))
+#define CRC32CW(Wd, Wn, Wm)         EMIT(CRC32C_gen(0, Wm, 0b10, Wn, Wd))
+#define CRC32CX(Wd, Wn, Xm)         EMIT(CRC32C_gen(1, Xm, 0b11, Wn, Wd))
+#define CRC32Cxw(Wd, Wn, Rm)        EMIT(CRC32C_gen(rex.w, Rm, 0b10|rex.w, Wn, Wd))
+
+// SHA1 extension
+#define SHA1H_gen(Rn, Rd)       (0b01011110<<24 | 0b10100<<17 | 0b10<<10 | (Rn)<<5 | (Rd))
+// SHA1 fixed rotate (ROL 30 of 32bits value)
+#define SHA1H(Sd, Sn)           EMIT(SHA1H_gen(Sn, Sd))
+
+#define SHA1SU1_gen(Rn, Rd)     (0b01011110<<24 | 0b10100<<17 | 0b00001<<12 | 0b10<<10 | (Rn)<<5 | (Rd))
+// SHA1 schedule update 1
+#define SHA1SU1(Vd, Vn)         EMIT(SHA1SU1_gen(Vn, Vd))
+
+#define SHA1C_gen(Rm, Rn, Rd)   (0b01011110<<24 | (Rm)<<16 | (Rn)<<5 | (Rd))
+// SHA1 hash update (choose)
+#define SHA1C(Qd, Sn, Vm)       EMIT(SHA1C_gen(Vm, Sn, Qd))
+
+#define SHA1M_gen(Rm, Rn, Rd)   (0b01011110<<24 | (Rm)<<16 | 0b010<<12 | (Rn)<<5 | (Rd))
+// SHA1 hash update (majority)
+#define SHA1M(Qd, Sn, Vm)       EMIT(SHA1M_gen(Vm, Sn, Qd))
+
+#define SHA1P_gen(Rm, Rn, Rd)   (0b01011110<<24 | (Rm)<<16 | 0b001<<12 | (Rn)<<5 | (Rd))
+// SHA1 hash update (parity)
+#define SHA1P(Qd, Sn, Vm)       EMIT(SHA1P_gen(Vm, Sn, Qd))
+
+#define SHA256SU0_gen(Rn,Rd)    (0b01011110<<24 | 0b10100<<17 | 0b00010<<12 | 0b10<<10 | (Rn)<<5 | (Rd))
+//SHA256 schedule update 0
+#define SHA256SU0(Vd, Vn)       EMIT(SHA256SU0_gen(Vn, Vd))
+
+#define SHA256SU1_gen(Rm, Rn, Rd)   (0b01011110<<24 | (Rm)<<16 | 0b110<<12 | (Rn)<<5 | (Rd))
+//SHA256 schedule update 1
+#define SHA256SU1(Vd, Vn, Vm)       EMIT(SHA256SU1_gen(Vm, Vn, Vd))
+
+#define SHA256H_gen(Rm, Rn, Rd) (0b01011110<<24 | (Rm)<<16 | 0b100<<12 | (Rn)<<5 | (Rd))
+//SHA256 hash update (part 1)
+#define SHA256H(Vd, Vn, Vm)     EMIT(SHA256H_gen(Vm, Vn, Vd))
+
+#define SHA256H2_gen(Rm, Rn, Rd)    (0b01011110<<24 | (Rm)<<16 | 0b101<<12 | (Rn)<<5 | (Rd))
+//SHA256 hash update (part 2)
+#define SHA256H2(Vd, Vn, Vm)        EMIT(SHA256H2_gen(Vm, Vn, Vd))
+
 #endif  //__ARM64_EMITTER_H__

src/dynarec/arm64/arm64_lock.S

@@ -29,6 +29,10 @@
 .global arm64_lock_incif0
 .global arm64_lock_decifnot0
 .global arm64_lock_store
+.global arm64_lock_store_dd
+.global arm64_lock_get_b
+.global arm64_lock_get_d
+.global arm64_lock_get_dd
 
 arm64_lock_read_b:
     dmb     ish
@@ -254,3 +258,20 @@ arm64_lock_store:
     str     w1, [x0]
     dmb     ish
     ret
+
+arm64_lock_store_dd:
+    str     x1, [x0]
+    dmb     ish
+    ret
+
+arm64_lock_get_b:
+    ldaxrb  w0, [x0]
+    ret
+
+arm64_lock_get_d:
+    ldaxr   w0, [x0]
+    ret
+
+arm64_lock_get_dd:
+    ldaxr   x0, [x0]
+    ret

src/dynarec/arm64/arm64_lock.h

@@ -69,4 +69,16 @@ extern int arm64_lock_decifnot0(void*p);
 // atomic store (with memory barrier)
 extern void arm64_lock_store(void*p, uint32_t v);
 
+// atomic store (with memory barrier)
+extern void arm64_lock_store_dd(void*p, uint64_t v);
+
+// atomic get (with memory barrier)
+extern uint8_t arm64_lock_get_b(void*p);
+
+// atomic get (with memory barrier)
+extern uint32_t arm64_lock_get_d(void*p);
+
+// atomic get (with memory barrier)
+extern void* arm64_lock_get_dd(void*p);
+
 #endif  //__ARM64_LOCK__H__

src/dynarec/arm64/arm64_printer.c

@@ -16,7 +16,7 @@ static const char* conds[] = {"cEQ", "cNE", "cCS", "cCC", "cMI", "cPL", "cVS", "
 #define abs(A) (((A)<0)?(-(A)):(A))
 
 typedef struct arm64_print_s {
-    int N, S, U, L, Q;
+    int N, S, U, L, Q, A, R;
     int t, n, m, d, t2, a;
     int f, c, o, h, p;
     int i, r, s;
@@ -60,6 +60,8 @@ int isMask(uint32_t opcode, const char* mask, arm64_print_t *a)
             case 'N': a->N = (a->N<<1) | v; break;
             case 'S': a->S = (a->S<<1) | v; break;
             case 'U': a->U = (a->U<<1) | v; break;
+            case 'A': a->A = (a->A<<1) | v; break;
+            case 'R': a->R = (a->R<<1) | v; break;
             case 'L': a->L = (a->L<<1) | v; break;
             case 'Q': a->Q = (a->Q<<1) | v; break;
             case 't': a->t = (a->t<<1) | v; break;
@@ -182,22 +184,22 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
     }
     if(isMask(opcode, "01111000010iiiiiiiii00nnnnnttttt", &a)) {
         int offset = signExtend(imm, 9);
-        snprintf(buff, sizeof(buff), "LDRUH %s, [%s, %s0x%x]", Wt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
+        snprintf(buff, sizeof(buff), "LDURH %s, [%s, %s0x%x]", Wt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
         return buff;
     }
     if(isMask(opcode, "001110001x0iiiiiiiii00nnnnnttttt", &a)) {
         int offset = signExtend(imm, 9);
-        snprintf(buff, sizeof(buff), "LDRUSB %s, [%s, %s0x%x]", a.x?Xt[Rt]:Wt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
+        snprintf(buff, sizeof(buff), "LDURSB %s, [%s, %s0x%x]", a.x?Xt[Rt]:Wt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
         return buff;
     }
     if(isMask(opcode, "011110001x0iiiiiiiii00nnnnnttttt", &a)) {
         int offset = signExtend(imm, 9);
-        snprintf(buff, sizeof(buff), "LDRUSH %s, [%s, %s0x%x]", a.x?Xt[Rt]:Wt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
+        snprintf(buff, sizeof(buff), "LDURSH %s, [%s, %s0x%x]", a.x?Xt[Rt]:Wt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
         return buff;
     }
     if(isMask(opcode, "10111000100iiiiiiiii00nnnnnttttt", &a)) {
         int offset = signExtend(imm, 9);
-        snprintf(buff, sizeof(buff), "LDRUSW %s, [%s, %s0x%x]", Xt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
+        snprintf(buff, sizeof(buff), "LDURSW %s, [%s, %s0x%x]", Xt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
         return buff;
     }
     if(isMask(opcode, "1x111000000iiiiiiiii00nnnnnttttt", &a)) {
@@ -213,7 +215,7 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
     }
     if(isMask(opcode, "01111000000iiiiiiiii00nnnnnttttt", &a)) {
         int offset = signExtend(imm, 9);
-        snprintf(buff, sizeof(buff), "STRUH %s, [%s, %s0x%x]", Wt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
+        snprintf(buff, sizeof(buff), "STURH %s, [%s, %s0x%x]", Wt[Rt], XtSp[Rn], (offset<0)?"-":"", abs(offset));
         return buff;
     }
     if(isMask(opcode, "1x11100101iiiiiiiiiiiinnnnnttttt", &a)) {
@@ -795,6 +797,11 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
         snprintf(buff, sizeof(buff), "CSEL %s, %s, %s, %s", sf?Xt[Rd]:Wt[Rd], sf?Xt[Rn]:Wt[Rn], sf?Xt[Rm]:Wt[Rm], conds[cond]);    
         return buff;
     }
+    if(isMask(opcode, "f1111010010mmmmmcccc00nnnnn0iiii", &a)) {
+        snprintf(buff, sizeof(buff), "CCMP %s, %s, %s 0x%x", sf?Xt[Rn]:Wt[Rn], sf?Xt[Rm]:Wt[Rm], conds[cond], imm);
+        return buff;
+    }
+
     // MISC Bits
     if(isMask(opcode, "f10110101100000000010onnnnnddddd", &a)) {
         snprintf(buff, sizeof(buff), "CL%c %s, %s", option?'S':'Z', sf?Xt[Rd]:Wt[Rd], sf?Xt[Rn]:Wt[Rn]);
@@ -950,6 +957,22 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
         snprintf(buff, sizeof(buff), "VCMEQ V%d.%s, V%d.%s, V%d.%s", Rd, Vd, Rn, Vd, Rm, Vd);
         return buff;
     }
+    // CMP zero
+    if(isMask(opcode, "0QU01110ff100000100o10nnnnnddddd", &a)) {
+        const char* Y[] = {"8B", "16B", "4H", "8H", "2S", "4S", "??", "2D"};
+        const char* Z[] = {"GT", "GE", "EQ", "LE"};
+        const char* Vd = Y[((sf)<<1) | a.Q];
+        const char* Cond = Z[(a.o << 1 | a.U)];
+        snprintf(buff, sizeof(buff), "VCM%s V%d.%s, V%d.%s, #0", Cond, Rd, Vd, Rn, Vd);
+        return buff;
+    }
+    // CMPLT zero
+    if(isMask(opcode, "0Q001110ff100000101010nnnnnddddd", &a)) {
+        const char* Y[] = {"8B", "16B", "4H", "8H", "2S", "4S", "??", "2D"};
+        const char* Vd = Y[((sf)<<1) | a.Q];
+        snprintf(buff, sizeof(buff), "VCMLT V%d.%s, V%d.%s, #0", Rd, Vd, Rn, Vd);
+        return buff;
+    }
     // MIN/MAX
     if(isMask(opcode, "0QU01110ff1mmmmm0110o1nnnnnddddd", &a)) {
         const char* Y[] = {"8B", "16B", "4H", "8H", "2S", "4S", "??", "2D"};
@@ -965,11 +988,21 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
         snprintf(buff, sizeof(buff), "MOVI V%d.%s, #0x%x", Rd, Vd, imm);
         return buff;
     }
-    // MOV immediate (not)shifted 16bits & 32bits
+    // MOV immediate notshifted 16bits & 32bits
     if(isMask(opcode, "0Q00111100000iiif00001iiiiiddddd", &a)) {
         const char* Y[] = {"2S", "4S", "4H", "8H"};
         const char* Vd = Y[(sf<<1)| a.Q];
-        snprintf(buff, sizeof(buff), "MOVI V%d.%s, #0x%x", Rd, Vd, imm);
+        int sh = 0;
+
+        snprintf(buff, sizeof(buff), "MOVI V%d.%s, #0x%x", Rd, Vd, imm<<sh);
+        return buff;
+    }
+    // MOV immediate shifted 16bits
+    if(isMask(opcode, "0Q00111100000iii101001iiiiiddddd", &a)) {
+        const char* Y[] = {"4H", "8H"};
+        const char* Vd = Y[a.Q];
+
+        snprintf(buff, sizeof(buff), "MOVI V%d.%s, #0x%x", Rd, Vd, imm<<8);
         return buff;
     }
 
@@ -1034,7 +1067,30 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
         if(!offset)
             snprintf(buff, sizeof(buff), "%s %c%d, [%s]", op?"STR":"LDR", s, Rt, XtSp[Rn]);
         else
-            snprintf(buff, sizeof(buff), "%s %c%d, [%s, %d]", op?"STR":"LDR", s, Rt, XtSp[Rn], offset);
+            snprintf(buff, sizeof(buff), "%s %c%d, [%s, 0x%x]", op?"STR":"LDR", s, Rt, XtSp[Rn], offset);
+        return buff;
+    }
+    if(isMask(opcode, "ss111100cc1mmmmmoooS10nnnnnttttt", &a)) {
+        char s = '?';
+        int size=imms;
+        int op=0;
+        if(size==0 && opc==1) {s='B';}
+        else if(size==1 && opc==1) {s='H';}
+        else if(size==2 && opc==1) {s='S';}
+        else if(size==3 && opc==1) {s='D';}
+        else if(size==0 && opc==3) {s='Q'; size = 4;}
+        else if(size==0 && opc==0) {s='B'; op=1;}
+        else if(size==1 && opc==0) {s='H'; op=1;}
+        else if(size==2 && opc==0) {s='S'; op=1;}
+        else if(size==3 && opc==0) {s='D'; op=1;}
+        else if(size==0 && opc==2) {s='Q'; op=1; size = 4;}
+
+        const char* extend[] = {"?0", "?1", "UXTW", "LSL", "?4", "?5", "SXTW", "SXTX"};
+        int amount = size*a.S;
+        if(option==3 && !amount)
+            snprintf(buff, sizeof(buff), "%s %c%d, [%s, %s]", op?"STR":"LDR", s, Rt, XtSp[Rn], ((option&1)==0)?Wt[Rm]:Xt[Rm]);
+        else
+            snprintf(buff, sizeof(buff), "%s %c%d, [%s, %s, %s %d]", op?"STR":"LDR", s, Rt, XtSp[Rn], ((option&1)==0)?Wt[Rm]:Xt[Rm], extend[option], amount);
         return buff;
     }
 
@@ -1094,6 +1150,20 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
         return buff;
     }
 
+    // FADDP
+    if(isMask(opcode, "0Q1011100f1mmmmm110101nnnnnddddd", &a)) {
+        char s = a.Q?'V':'D';
+        char d = sf?'D':'S';
+        int n = (a.Q && !sf)?4:2;
+        snprintf(buff, sizeof(buff), "VFADDP %c%d.%d%c, %c%d.%d%c, %c%d.%d%c", s, Rd, n, d, s, Rn, n, d, s, Rm, n, d);
+        return buff;
+    }
+    if(isMask(opcode, "011111100f110000110110nnnnnddddd", &a)) {
+        char s = (sf==0)?'S':((sf==1)?'D':'?');
+        snprintf(buff, sizeof(buff), "FADDP %c%d, %c%d, %c%d", s, Rd, s, Rn, s, Rm);
+        return buff;
+    }
+
     // SQRT
     if(isMask(opcode, "0Q1011101f100001111110nnnnnddddd", &a)) {
         char s = a.Q?'V':'D';
@@ -1302,6 +1372,14 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
         snprintf(buff, sizeof(buff), "FCVT%sS %s, %c%d", roundings[a.c], sf?Xt[Rd]:Wt[Rd], s, Rn);
         return buff;
     }
+    if(isMask(opcode, "0QU01110of100001101o10nnnnnddddd", &a)) {
+        const char* Y[] = {"2S", "4S", "??", "2D"};
+        const char* Z[] = {"S", "S", "??", "D"};
+        const char* Vd = Y[(sf<<1) | a.Q];
+        const char* roundings[] = {"N", "M", "P", "Z"};
+        snprintf(buff, sizeof(buff), "VFCVT%s%s%s%s V%d.%s, V%d.%s", roundings[option], a.U?"U":"S", a.Q?"Q":"", Z[(sf<<1)|a.Q], Rd, Vd, Rn, Vd);
+        return buff;
+    }
 
     // FMOV
     if(isMask(opcode, "00011110pp100000010000nnnnnddddd", &a)) {
@@ -1548,7 +1626,139 @@ const char* arm64_print(uint32_t opcode, uintptr_t addr)
         snprintf(buff, sizeof(buff), "DMB %s", (Rn==0b1011)?"ISH":"???");
         return buff;
     }
+    // DSB ISH/ISHST
+    if(isMask(opcode, "11010101000000110011nnnn10011111", &a)) {
+        snprintf(buff, sizeof(buff), "DSB %s", (Rn==0b1011)?"ISH":((Rn==0b1010)?"ISHST":"???"));
+        return buff;
+    }
 
+    // CASxw
+    if(isMask(opcode, "1f0010001L1ssssso11111nnnnnttttt", &a)) {
+        snprintf(buff, sizeof(buff), "CAS%s%s %s, %s, [%s]", a.o?"A":"", a.L?"L":"", sf?Xt[Rs]:Wt[Rs], sf?Xt[Rt]:Wt[Rt], XtSp[Rn]);
+        return buff;
+    }
+    // CAS B/H
+    if(isMask(opcode, "0f0010001L1ssssso11111nnnnnttttt", &a)) {
+        snprintf(buff, sizeof(buff), "CAS%s%s%s %s, %s, [%s]", a.o?"A":"", a.L?"L":"", sf?"H":"B", Xt[Rs], Xt[Rt], XtSp[Rn]);
+        return buff;
+    }
+    // CASPxw
+    if(isMask(opcode, "0f0010000L1ssssso11111nnnnnttttt", &a)) {
+        snprintf(buff, sizeof(buff), "CASP%s%s %s,%s, %s,%s, [%s]", a.o?"A":"", a.L?"L":"", sf?Xt[Rs]:Wt[Rs], sf?Xt[Rs+1]:Wt[Rs+1], sf?Xt[Rt]:Wt[Rt], sf?Xt[Rt+1]:Wt[Rt+1], XtSp[Rn]);
+        return buff;
+    }
+    // SWPxw
+    if(isMask(opcode, "1f111000AR1sssss100000nnnnnttttt", &a)) {
+        snprintf(buff, sizeof(buff), "SWP%s%s %s, %s, [%s]", a.A?"A":"", a.R?"L":"", sf?Xt[Rs]:Wt[Rs], sf?Xt[Rt]:Wt[Rt], XtSp[Rn]);
+        return buff;
+    }
+    // SWPxw
+    if(isMask(opcode, "0f111000AR1sssss100000nnnnnttttt", &a)) {
+        snprintf(buff, sizeof(buff), "CAS%s%s%s %s, %s, [%s]", a.A?"A":"", a.R?"L":"", sf?"H":"B", Xt[Rs], Xt[Rt], XtSp[Rn]);
+        return buff;
+    }
+    // LDXXXxw
+    if(isMask(opcode, "1f111000AR1sssss0ooo00nnnnnttttt", &a)) {
+        const char* ops[] = { "ADD", "CLR", "EOR", "SET", "SMAX, SMIN", "UMAX", "UMIN" };
+        if((Rt == 0b11111) && !a.A) {
+            snprintf(buff, sizeof(buff), "ST%s%s %s, [%s]", ops[a.o], a.R?"L":"", sf?Xt[Rs]:Wt[Rs], XtSp[Rn]);
+        } else {
+            snprintf(buff, sizeof(buff), "LD%s%s%s %s, %s, [%s]", ops[a.o], a.A?"A":"", a.R?"L":"", sf?Xt[Rs]:Wt[Rs], sf?Xt[Rt]:Wt[Rt], XtSp[Rn]);
+        }
+        return buff;
+    }
+    // LDXXX B/H
+    if(isMask(opcode, "0f111000AR1sssss0ooo00nnnnnttttt", &a)) {
+        const char* ops[] = { "ADD", "CLR", "EOR", "SET", "SMAX, SMIN", "UMAX", "UMIN" };
+        if((Rt == 0b11111) && !a.A) {
+            snprintf(buff, sizeof(buff), "ST%s%s%s %s, [%s]", ops[a.o], a.R?"L":"", sf?"H":"B", Xt[Rs], XtSp[Rn]);
+        } else {
+            snprintf(buff, sizeof(buff), "LD%s%s%s%s %s, %s, [%s]", ops[a.o], a.A?"A":"", a.R?"L":"", sf?"H":"B", Xt[Rs], Xt[Rt], XtSp[Rn]);
+        }
+        return buff;
+    }
+    // AXFLAG
+    if(isMask(opcode, "11010101000000000100000001011111", &a)) {
+        snprintf(buff, sizeof(buff), "AXFLAG");
+        return buff;
+    }
+    // XAFLAG
+    if(isMask(opcode, "11010101000000000100000000111111", &a)) {
+        snprintf(buff, sizeof(buff), "XAFLAG");
+        return buff;
+    }
+    // CFINV
+    if(isMask(opcode, "11010101000000000100000000011111", &a)) {
+        snprintf(buff, sizeof(buff), "CFINV");
+        return buff;
+    }
+    // RMIF
+    if(isMask(opcode, "10111010000iiiiii00001nnnnn0oooo", &a)) {
+        snprintf(buff, sizeof(buff), "RMIF %s, #%d, #0x%x", Xt[Rn], imm, opc);
+        return buff;
+    }
+    // SETF
+    if(isMask(opcode, "00111010000000000f0010nnnnn01101", &a)) {
+        snprintf(buff, sizeof(buff), "SETF%d %s", 8<<sf, Xt[Rn]);
+        return buff;
+    }
+    // REV64
+    if(isMask(opcode, "0Q001110ff100000000010nnnnnddddd", &a)) {
+        const char* T[] = {"8B", "16B", "4H", "8H", "2S", "4S", "??", "??"};
+        int sz = sf;
+        const char* Vn = T[(sz<<1)|a.Q];
+        const char* Vd = T[(sz<<1)|a.Q];
+        snprintf(buff, sizeof(buff), "REV64 V%d.%s, V%d.%s",Rd, Vd, Rn, Vn);
+        return buff;
+    }
+    //TRNx
+    if(isMask(opcode, "0Q001110ff0mmmmm0o1010nnnnnddddd", &a)) {
+        const char* T[] = {"8B", "16B", "4H", "8H", "2S", "4S", "??", "2D"};
+        int sz = sf;
+        const char* Vn = T[(sz<<1)|a.Q];
+        const char* Vm = T[(sz<<1)|a.Q];
+        const char* Vd = T[(sz<<1)|a.Q];
+        snprintf(buff, sizeof(buff), "TRN%d V%d.%s, V%d.%s, V%d.%s", a.o+1, Rd, Vd, Rn, Vn, Rm, Vm);
+        return buff;
+    }
+    //SHA1 stuffs
+    if(isMask(opcode, "0101111000101000000010nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA1H S%d, S%d", Rd, Rn);
+        return buff;
+    }
+    if(isMask(opcode, "0101111000101000000110nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA1SU1 V%d.4S, V%d.4S", Rd, Rn);
+        return buff;
+    }
+    if(isMask(opcode, "01011110000mmmmm000000nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA1C Q%d, S%d, V%d.4S", Rd, Rn, Rm);
+        return buff;
+    }
+    if(isMask(opcode, "01011110000mmmmm001000nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA1M Q%d, S%d, V%d.4S", Rd, Rn, Rm);
+        return buff;
+    }
+    if(isMask(opcode, "01011110000mmmmm000100nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA1P Q%d, S%d, V%d.4S", Rd, Rn, Rm);
+        return buff;
+    }
+    //SHA256 stuffs
+    if(isMask(opcode, "0101111000101000001010nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA256SU0 V%d.4S, V%d.4S", Rd, Rn);
+        return buff;
+    }
+    if(isMask(opcode, "01011110000mmmmm011000nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA256SU1 V%d.4S, V%d.4S, V%d.4S", Rd, Rn, Rm);
+        return buff;
+    }
+    if(isMask(opcode, "01011110000mmmmm010000nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA256H Q%d, Q%d, V%d.4S", Rd, Rn, Rm);
+        return buff;
+    }
+    if(isMask(opcode, "01011110000mmmmm010100nnnnnddddd", &a)) {
+        snprintf(buff, sizeof(buff), "SHA256H2 Q%d, Q%d, V%d.4S", Rd, Rn, Rm);
+        return buff;
+    }
 
     snprintf(buff, sizeof(buff), "%08X ???", __builtin_bswap32(opcode));
     return buff;

src/dynarec/arm64/dynarec_arm64_0f.c

@@ -17,6 +17,7 @@
 #include "dynarec_native.h"
 #include "my_cpuid.h"
 #include "emu/x87emu_private.h"
+#include "emu/x64shaext.h"
 
 #include "arm64_printer.h"
 #include "dynarec_arm64_private.h"
@@ -302,7 +303,17 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 SMWRITE2();
             }
             break;
-
+        case 0x2A:
+            INST_NAME("CVTPI2PS Gx,Em");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEM(q1, 0);
+            d0 = fpu_get_scratch(dyn);
+            u8 = sse_setround(dyn, ninst, x1, x2, x3);
+            SCVTFS(d0, q1);
+            x87_restoreround(dyn, ninst, u8);
+            VMOVeD(v0, 0, d0, 0);
+            break;
         case 0x2B:
             INST_NAME("MOVNTPS Ex,Gx");
             nextop = F8;
@@ -317,7 +328,69 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 VST128(v0, ed, fixedaddress);
             }
             break;
-
+        case 0x2C:
+            INST_NAME("CVTTPS2PI Gm,Ex");
+            nextop = F8;
+            GETGM(q0);
+            GETEX(v1, 0, 0);
+            if (box64_dynarec_fastround) {
+                VFCVTZSS(q0, v1);
+            } else {
+                MRS_fpsr(x5);
+                BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
+                MSR_fpsr(x5);
+                ORRw_mask(x2, xZR, 1, 0);    //0x80000000
+                d0 = fpu_get_scratch(dyn);
+                for (int i=0; i<2; ++i) {
+                    BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
+                    if (i) {
+                        VMOVeS(d0, 0, v1, i);
+                        FRINTZS(d0, d0);
+                    } else {
+                        FRINTZS(d0, v1);
+                    }
+                    FCVTZSwS(x1, d0);
+                    MRS_fpsr(x5);   // get back FPSR to check the IOC bit
+                    TBZ(x5, FPSR_IOC, 4+4);
+                    MOVw_REG(x1, x2);
+                    VMOVQSfrom(q0, i, x1);
+                }
+            }
+            break;
+        case 0x2D:
+            INST_NAME("CVTPS2PI Gm, Ex");
+            nextop = F8;
+            GETGM(q0);
+            GETEX(v1, 0, 0);
+            if (box64_dynarec_fastround) {
+                u8 = sse_setround(dyn, ninst, x1, x2, x3);
+                VFRINTIS(q0, v1);
+                x87_restoreround(dyn, ninst, u8);
+                VFCVTZSS(q0, q0);
+            } else {
+                u8 = sse_setround(dyn, ninst, x1, x2, x3);
+                MRS_fpsr(x5);
+                BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
+                MSR_fpsr(x5);
+                ORRw_mask(x2, xZR, 1, 0);    //0x80000000
+                d0 = fpu_get_scratch(dyn);
+                for (int i=0; i<2; ++i) {
+                    BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
+                    if (i) {
+                        VMOVeS(d0, 0, v1, i);
+                        FRINTIS(d0, d0);
+                    } else {
+                        FRINTIS(d0, v1);
+                    }
+                    FCVTZSwS(x1, d0);
+                    MRS_fpsr(x5);   // get back FPSR to check the IOC bit
+                    TBZ(x5, FPSR_IOC, 4+4);
+                    MOVw_REG(x1, x2);
+                    VMOVQSfrom(q0, i, x1);
+                }
+                x87_restoreround(dyn, ninst, u8);
+            }
+            break;
         case 0x2E:
             // no special check...
         case 0x2F:
@@ -349,11 +422,34 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     GETGM(q0);
                     GETEM(q1, 0);
                     d0 = fpu_get_scratch(dyn);
-                    MOVI_8(d0, 0b10001111);
+                    MOVI_8(d0, 0b10000111);
                     VAND(d0, d0, q1);  // mask the index
                     VTBL1_8(q0, q0, d0);
                     break;
-
+                case 0x01:
+                    INST_NAME("PHADDW Gm, Em");
+                    nextop = F8;
+                    GETGM(q0);
+                    GETEM(q1, 0);
+                    VADDP_16(q0, q0, q1);
+                    break;
+                case 0x2:
+                    INST_NAME("PHADDD Gm, Em");
+                    nextop = F8;
+                    GETGM(q0);
+                    GETEM(q1, 0);
+                    VADDP_32(q0, q0, q1);
+                    break;
+                case 0x03:
+                    INST_NAME("PHADDSW Gm, Em");
+                    nextop = F8;
+                    GETGM(q0);
+                    GETEM(q1, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    VUZP1_16(v0, q0, q1);
+                    VUZP2_16(q0, q0, q1);
+                    SQADD_16(q0, q0, v0);
+                    break;
                 case 0x04:
                     INST_NAME("PMADDUBSW Gm,Em");
                     nextop = F8;
@@ -367,7 +463,75 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     SADDLPQ_16(v1, v0);
                     SQXTN_16(q0, v1);
                     break;
-
+                case 0x05:
+                    INST_NAME("PHSUBW Gm,Em");
+                    nextop = F8;
+                    GETGM(q0);
+                    GETEM(q1, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    VUZP1_16(v0, q0, q1);
+                    VUZP2_16(q0, q0, q1);
+                    VSUB_16(q0, v0, q0);
+                    break;
+                case 0x06:
+                    INST_NAME("PHSUBD Gm,Em");
+                    nextop = F8;
+                    GETGM(q0);
+                    GETEM(q1, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    VUZP1_32(v0, q0, q1);
+                    VUZP2_32(q0, q0, q1);
+                    VSUB_32(q0, v0, q0);
+                    break;
+                case 0x07:
+                    INST_NAME("PHSUBSW Gm,Em");
+                    nextop = F8;
+                    GETGM(q0);
+                    GETEM(q1, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    VUZP1_16(v0, q0, q1);
+                    VUZP2_16(q0, q0, q1);
+                    SQSUB_16(q0, v0, q0);
+                    break;
+                case 0x08:
+                    INST_NAME("PSIGNB Gm,Em");
+                    nextop = F8;
+                    GETGM(q0);
+                    GETEM(q1, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    v1 = fpu_get_scratch(dyn);
+                    CMGT_0_8(v0, q1);
+                    VAND(v0, v0, q0);
+                    CMLT_0_8(v1, q1);
+                    VMUL_8(q0, q0, v1);
+                    VORR(q0, q0, v0);
+                    break;
+                case 0x09:
+                    INST_NAME("PSIGNW Gm,Em");
+                    nextop = F8;
+                    GETGM(q0);
+                    GETEM(q1, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    v1 = fpu_get_scratch(dyn);
+                    CMGT_0_16(v0, q1);
+                    VAND(v0, v0, q0);
+                    CMLT_0_16(v1, q1);
+                    VMUL_16(q0, q0, v1);
+                    VORR(q0, q0, v0);
+                    break;
+                case 0x0A:
+                    INST_NAME("PSIGND Gm,Em");
+                    nextop = F8;
+                    GETGM(q0);
+                    GETEM(q1, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    v1 = fpu_get_scratch(dyn);
+                    CMGT_0_32(v0, q1);
+                    VAND(v0, v0, q0);
+                    CMLT_0_32(v1, q1);
+                    VMUL_32(q0, q0, v1);
+                    VORR(q0, q0, v0);
+                    break;
                 case 0x0B:
                     INST_NAME("PMULHRSW Gm,Em");
                     nextop = F8;
@@ -375,6 +539,199 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     GETEM(q1, 0);
                     SQRDMULH_16(q0, q0, q1);
                     break;
+                case 0x1C:
+                    INST_NAME("PABSB Gm,Em");
+                    nextop = F8;
+                    GETGM(q0);
+                    GETEM(q1, 0);
+                    ABS_8(q0, q1);
+                    break;
+                case 0x1D:
+                    INST_NAME("PABSW Gm,Em");
+                    nextop = F8;
+                    GETGM(q0);
+                    GETEM(q1, 0);
+                    ABS_16(q0, q1);
+                    break;
+                case 0x1E:
+                    INST_NAME("PABSD Gm,Em");
+                    nextop = F8;
+                    GETGM(q0);
+                    GETEM(q1, 0);
+                    ABS_32(q0, q1);
+                    break;
+                
+                case 0xC8:
+                    INST_NAME("SHA1NEXTE Gx, Ex");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    GETEX(q1, 0, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    VEORQ(v0, v0, v0);
+                    if(arm64_sha1) {
+                        v1 = fpu_get_scratch(dyn);
+                        VMOVeS(v1, 0, q0, 3);
+                        SHA1H(v1, v1);
+                        VMOVeS(v0, 3, v1, 0);
+                    } else {
+                        VMOVSto(x1, q0, 3);
+                        RORw(x1, x1, 2);    // i.e. ROL 30
+                        VMOVQSfrom(v0, 3, x1);
+                    }
+                    VADDQ_32(q0, v0, q1);
+                    break;
+                case 0xC9:
+                    INST_NAME("SHA1MSG1 Gx, Ex");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    GETEX(q1, 0, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    VEXTQ_8(v0, q1, q0, 8);
+                    VEORQ(q0, q0, v0);
+                    break;
+                case 0xCA:
+                    INST_NAME("SHA1MSG2 Gx, Ex");
+                    nextop = F8;
+                    if(arm64_sha1) {
+                        GETGX(q0, 1);
+                        GETEX(q1, 0, 0);
+                        VEXTQ_8(q0, q0, q0, 8);
+                        VREV64Q_32(q0, q0);
+                        if(MODREG) {
+                            if(q0==q1)
+                                v0 = q0;
+                            else {
+                                v0 = fpu_get_scratch(dyn);
+                                VEXTQ_8(v0, q1, q1, 8);
+                                VREV64Q_32(v0, v0);
+                            }
+                        } else {
+                            v0 = q1;
+                            VEXTQ_8(v0, q1, q1, 8);
+                            VREV64Q_32(v0, v0);
+                        }
+                        SHA1SU1(q0, v0);
+                        VEXTQ_8(q0, q0, q0, 8);
+                        VREV64Q_32(q0, q0);
+                    } else {
+                        if(MODREG) {
+                            ed = (nextop&7)+(rex.b<<3);
+                            sse_reflect_reg(dyn, ninst, ed);
+                            ADDx_U12(x2, xEmu, offsetof(x64emu_t, xmm[ed]));
+                        } else {
+                            SMREAD();
+                            addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
+                            if(wback!=x2) {
+                                MOVx_REG(x2, wback);
+                            }
+                        }
+                        GETG;
+                        sse_forget_reg(dyn, ninst, gd);
+                        ADDx_U12(x1, xEmu, offsetof(x64emu_t, xmm[gd]));
+                        CALL(sha1msg2, -1);
+                    }
+                    break;
+                case 0xCB:
+                    INST_NAME("SHA256RNDS2 Gx, Ex (, XMM0)");
+                    nextop = F8;
+                    if(arm64_sha2) {
+                        GETGX(q0, 1);
+                        GETEX(q1, 0, 0);
+                        d0 = sse_get_reg(dyn, ninst, x1, 0, 0);
+                        v0 = fpu_get_scratch(dyn);
+                        d1 = fpu_get_scratch(dyn);
+                        if(MODREG) {
+                            v1 = fpu_get_scratch(dyn);
+                        } else
+                            v1 = q1;
+                        VREV64Q_32(q0, q0);
+                        VREV64Q_32(v1, q1);
+                        VZIP1Q_64(v0, v1, q0);
+                        VZIP2Q_64(v1, v1, q0);
+                        SHA256H(v1, v0, d0);
+                        VREV64Q_32(d1, q1);
+                        VZIP2Q_64(d1, d1, q0);
+                        SHA256H2(v0, d1, d0);
+                        VZIP2Q_64(q0, v0, v1);
+                        VREV64Q_32(q0, q0);
+                    } else {
+                        if(MODREG) {
+                            ed = (nextop&7)+(rex.b<<3);
+                            sse_reflect_reg(dyn, ninst, ed);
+                            ADDx_U12(x2, xEmu, offsetof(x64emu_t, xmm[ed]));
+                        } else {
+                            SMREAD();
+                            addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
+                            if(wback!=x2) {
+                                MOVx_REG(x2, wback);
+                            }
+                        }
+                        GETG;
+                        sse_forget_reg(dyn, ninst, gd);
+                        ADDx_U12(x1, xEmu, offsetof(x64emu_t, xmm[gd]));
+                        sse_reflect_reg(dyn, ninst, 0);
+                        CALL(sha256rnds2, -1);
+                    }
+                    break;
+                case 0xCC:
+                    INST_NAME("SHA256MSG1 Gx, Ex");
+                    nextop = F8;
+                    if(arm64_sha2) {
+                        GETGX(q0, 1);
+                        GETEX(q1, 0, 0);
+                        SHA256SU0(q0, q1);
+                    } else {
+                        if(MODREG) {
+                            ed = (nextop&7)+(rex.b<<3);
+                            sse_reflect_reg(dyn, ninst, ed);
+                            ADDx_U12(x2, xEmu, offsetof(x64emu_t, xmm[ed]));
+                        } else {
+                            SMREAD();
+                            addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
+                            if(wback!=x2) {
+                                MOVx_REG(x2, wback);
+                            }
+                        }
+                        GETG;
+                        sse_forget_reg(dyn, ninst, gd);
+                        ADDx_U12(x1, xEmu, offsetof(x64emu_t, xmm[gd]));
+                        CALL(sha256msg1, -1);
+                    }
+                    break;
+                case 0xCD:
+                    INST_NAME("SHA256MSG2 Gx, Ex");
+                    nextop = F8;
+                    if(arm64_sha2) {
+                        GETGX(q0, 1);
+                        GETEX(q1, 0, 0);
+                        v0 = fpu_get_scratch(dyn);
+                        v1 = fpu_get_scratch(dyn);
+                        d0 = fpu_get_scratch(dyn);
+                        VEORQ(v1, v1, v1);
+                        VMOVQ(v0, q0);
+                        SHA256SU1(v0, v1, q1);  // low v0 are ok and also need to be feed again SHA256SU1 to get the high part
+                        VTRNQ1_64(d0, v0, v0);  // duplicate low to hi
+                        VEXTQ_8(q0, q0, q0, 8); // invert low/high now
+                        SHA256SU1(q0, v1, d0);  // low is destination high
+                        VEXTQ_8(q0, d0, q0, 8);
+                    } else {
+                        if(MODREG) {
+                            ed = (nextop&7)+(rex.b<<3);
+                            sse_reflect_reg(dyn, ninst, ed);
+                            ADDx_U12(x2, xEmu, offsetof(x64emu_t, xmm[ed]));
+                        } else {
+                            SMREAD();
+                            addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
+                            if(wback!=x2) {
+                                MOVx_REG(x2, wback);
+                            }
+                        }
+                        GETG;
+                        sse_forget_reg(dyn, ninst, gd);
+                        ADDx_U12(x1, xEmu, offsetof(x64emu_t, xmm[gd]));
+                        CALL(sha256msg2, -1);
+                    }
+                    break;
 
                 case 0xF0:
                     INST_NAME("MOVBE Gd, Ed");
@@ -427,6 +784,81 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         VEXT_8(q0, q1, q0, u8);
                     }
                     break;
+
+                case 0xCC:
+                    INST_NAME("SHA1RNDS4 Gx, Ex, Ib");
+                    nextop = F8;
+                    if(arm64_sha1) {
+                        GETGX(q0, 1);
+                        GETEX(q1, 0, 1);
+                        u8 = F8&3;
+                        d0 = fpu_get_scratch(dyn);
+                        d1 = fpu_get_scratch(dyn);
+                        v0 = fpu_get_scratch(dyn);
+                        VEXTQ_8(v0, q0, q0, 8);
+                        VREV64Q_32(v0, v0);
+                        VEORQ(d1, d1, d1);
+                        if(MODREG) {
+                            if(q0==q1)
+                                v1 = v0;
+                            else
+                                v1 = fpu_get_scratch(dyn);
+                        } else 
+                            v1 = q1;
+                        if(v1!=v0) {
+                            VEXTQ_8(v1, q1, q1, 8);
+                            VREV64Q_32(v1, v1);
+                        }
+                        switch(u8) {
+                            case 0:
+                                MOV32w(x1, 0x5A827999);
+                                VDUPQS(d0, x1);
+                                VADDQ_32(v1, v1, d0);
+                                SHA1C(v0, d1, v1);
+                                break;
+                            case 1:
+                                MOV32w(x1, 0x6ED9EBA1);
+                                VDUPQS(d0, x1);
+                                VADDQ_32(v1, v1, d0);
+                                SHA1P(v0, d1, v1);
+                                break;
+                            case 2:
+                                MOV32w(x1, 0X8F1BBCDC);
+                                VDUPQS(d0, x1);
+                                VADDQ_32(v1, v1, d0);
+                                SHA1M(v0, d1, v1);
+                                break;
+                            case 3:
+                                MOV32w(x1, 0xCA62C1D6);
+                                VDUPQS(d0, x1);
+                                VADDQ_32(v1, v1, d0);
+                                SHA1P(v0, d1, v1);
+                                break;
+                        }
+                        VREV64Q_32(v0, v0);
+                        VEXTQ_8(q0, v0, v0, 8);
+                        break;
+                    } else {
+                        if(MODREG) {
+                            ed = (nextop&7)+(rex.b<<3);
+                            sse_reflect_reg(dyn, ninst, ed);
+                            ADDx_U12(x2, xEmu, offsetof(x64emu_t, xmm[ed]));
+                        } else {
+                            SMREAD();
+                            addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 1);
+                            if(wback!=x2) {
+                                MOVx_REG(x2, wback);
+                            }
+                        }
+                        u8 = F8;
+                        GETG;
+                        sse_forget_reg(dyn, ninst, gd);
+                        ADDx_U12(x1, xEmu, offsetof(x64emu_t, xmm[gd]));
+                        MOV32w(x3, u8);
+                        CALL(sha1rnds4, -1);
+                    }
+                    break;
+
                 default:
                     DEFAULT;
             }
@@ -787,33 +1219,17 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         VMOVeS(v0, 1, v1, 0);
                     }
                 } else if(u8==0x00) {
-                    // dumplicate lower 16bits to all spot
-                    if(v0!=v1) {
-                        VMOVeH(v0, 0, v1, 0);
-                    }
-                    VMOVeH(v0, 1, v1, 0);
-                    VMOVeS(v0, 1, v1, 0);
+                    // duplicate lower 16bits to all spot
+                    VDUP_16(v0, v1, 0);
                 } else if(u8==0x55) {
-                    // dumplicate 16bits slot 1 to all spot
-                    if(v0!=v1) {
-                        VMOVeH(v0, 1, v1, 1);
-                    }
-                    VMOVeH(v0, 0, v1, 1);
-                    VMOVeS(v0, 1, v1, 0);
+                    // duplicate 16bits slot 1 to all spot
+                    VDUP_16(v0, v1, 1);
                 } else if(u8==0xAA) {
-                    // dumplicate 16bits slot 2 to all spot
-                    if(v0!=v1) {
-                        VMOVeH(v0, 2, v1, 2);
-                    }
-                    VMOVeH(v0, 3, v1, 2);
-                    VMOVeS(v0, 0, v1, 1);
+                    // duplicate 16bits slot 2 to all spot
+                    VDUP_16(v0, v1, 2);
                 } else if(u8==0xFF) {
-                    // dumplicate 16bits slot 3 to all spot
-                    if(v0!=v1) {
-                        VMOVeH(v0, 3, v1, 3);
-                    }
-                    VMOVeH(v0, 2, v1, 3);
-                    VMOVeS(v0, 0, v1, 1);
+                    // duplicate 16bits slot 3 to all spot
+                    VDUP_16(v0, v1, 3);
                 } else if(v0!=v1) {
                     VMOVeH(v0, 0, v1, (u8>>(0*2))&3);
                     VMOVeH(v0, 1, v1, (u8>>(1*2))&3);
@@ -1133,13 +1549,27 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0xA5:
             nextop = F8;
             INST_NAME("SHLD Ed, Gd, CL");
-            MESSAGE(LOG_DUMP, "Need Optimization\n");
-            UXTBw(x3, xRCX);
-            SETFLAGS(X_ALL, SF_SET);
-            GETEDW(x4, x1, 0);
+            SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+            if(box64_dynarec_safeflags>1)
+                MAYSETFLAGS();
             GETGD;
-            MOVxw_REG(x2, gd);
-            CALL_(rex.w?((void*)shld64):((void*)shld32), ed, x4);
+            GETED(0);
+            if(!rex.w && !rex.is32bits && MODREG) {MOVw_REG(ed, ed);}
+            UFLAG_IF {
+                if(rex.w) {
+                    ANDSx_mask(x3, xRCX, 1, 0, 0b00101);  //mask=0x000000000000003f
+                } else {
+                    ANDSw_mask(x3, xRCX, 0, 0b00100);  //mask=0x00000001f
+                }
+                B_NEXT(cEQ);
+            } else {
+                if(rex.w) {
+                    ANDx_mask(x3, xRCX, 1, 0, 0b00101);  //mask=0x000000000000003f
+                } else {
+                    ANDw_mask(x3, xRCX, 0, 0b00100);  //mask=0x00000001f
+                }
+            }
+            emit_shld32(dyn, ninst, rex, ed, gd, x3, x5, x4);
             WBACK;
             break;
 
@@ -1166,17 +1596,13 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             } else {
                 ANDw_mask(x2, gd, 0, 0b00100);  //mask=0x00000001f
             }
-            LSRxw_REG(x4, ed, x2);
-            if(rex.w) {
-                ANDSx_mask(x4, x4, 1, 0, 0);  //mask=1
-            } else {
-                ANDSw_mask(x4, x4, 0, 0);  //mask=1
+            IFX(X_CF) {
+                LSRxw_REG(x4, ed, x2);
+                BFIw(xFlags, x4, F_CF, 1);
             }
-            BFIw(xFlags, x4, F_CF, 1);
             MOV32w(x4, 1);
             LSLxw_REG(x4, x4, x2);
-            EORxw_REG(x4, ed, x4);
-            CSELxw(ed, ed, x4, cNE);
+            ORRxw_REG(ed, ed, x4);
             if(wback) {
                 STRxw_U12(ed, wback, fixedaddress);
                 SMWRITE();
@@ -1195,13 +1621,27 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0xAD:
             nextop = F8;
             INST_NAME("SHRD Ed, Gd, CL");
-            MESSAGE(LOG_DUMP, "Need Optimization\n");
-            SETFLAGS(X_ALL, SF_SET);
-            UXTBw(x3, xRCX);
-            GETEDW(x4, x1, 0);
+            SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+            if(box64_dynarec_safeflags>1)
+                MAYSETFLAGS();
             GETGD;
-            MOVxw_REG(x2, gd);
-            CALL_(rex.w?((void*)shrd64):((void*)shrd32), ed, x4);
+            GETED(0);
+            if(!rex.w && !rex.is32bits && MODREG) {MOVw_REG(ed, ed);}
+            UFLAG_IF {
+                if(rex.w) {
+                    ANDSx_mask(x3, xRCX, 1, 0, 0b00101);  //mask=0x000000000000003f
+                } else {
+                    ANDSw_mask(x3, xRCX, 0, 0b00100);  //mask=0x00000001f
+                }
+                B_NEXT(cEQ);
+            } else {
+                if(rex.w) {
+                    ANDx_mask(x3, xRCX, 1, 0, 0b00101);  //mask=0x000000000000003f
+                } else {
+                    ANDw_mask(x3, xRCX, 0, 0b00100);  //mask=0x00000001f
+                }
+            }
+            emit_shrd32(dyn, ninst, rex, ed, gd, x3, x5, x4);
             WBACK;
             break;
 
@@ -1279,7 +1719,7 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             break;
         case 0xAF:
             INST_NAME("IMUL Gd, Ed");
-            SETFLAGS(X_ALL, SF_PENDING);
+            SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGD;
             GETED(0);
@@ -1288,9 +1728,20 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 UFLAG_IF {
                     SMULH(x3, gd, ed);
                     MULx(gd, gd, ed);
-                    UFLAG_OP1(x3);
-                    UFLAG_RES(gd);
-                    UFLAG_DF(x3, d_imul64);
+                    IFX(X_PEND) {
+                        UFLAG_OP1(x3);
+                        UFLAG_RES(gd);
+                        UFLAG_DF(x4, d_imul64);
+                    } else IFX(X_CF|X_OF) {
+                        SET_DFNONE(x4);
+                    }
+                    IFX(X_CF|X_OF) {
+                        ASRx(x4, gd, 63);
+                        CMPSx_REG(x3, x4);
+                        CSETw(x3, cNE);
+                        BFIw(xFlags, x3, F_CF, 1);
+                        BFIw(xFlags, x3, F_OF, 1);
+                    }
                 } else {
                     MULxw(gd, gd, ed);
                 }
@@ -1298,10 +1749,29 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 // 32bits imul
                 UFLAG_IF {
                     SMULL(gd, gd, ed);
-                    UFLAG_RES(gd);
-                    LSRx(x3, gd, 32);
-                    UFLAG_OP1(x3);
-                    UFLAG_DF(x3, d_imul32);
+                    IFX(X_PEND) {
+                        UFLAG_RES(gd);
+                        LSRx(x3, gd, 32);
+                        UFLAG_OP1(x3);
+                        UFLAG_DF(x4, d_imul32);
+                    } else IFX(X_CF|X_OF) {
+                        SET_DFNONE(x4);
+                    }
+                    IFX(X_CF|X_OF) {
+                        ASRx(x3, gd, 31);
+                        ASRw(x4, gd, 31);
+                        CMPSw_REG(x3, x4);
+                        CSETw(x3, cNE);
+                        BFIw(xFlags, x3, F_CF, 1);
+                        BFIw(xFlags, x3, F_OF, 1);
+                    }
+                    if(box64_dynarec_test) {
+                        // to avoid noise during test
+                        BFCw(xFlags, F_AF, 1);
+                        BFCw(xFlags, F_PF, 1);
+                        BFCw(xFlags, F_ZF, 1);
+                        BFCw(xFlags, F_SF, 1);
+                    }
                     MOVw_REG(gd, gd);
                 } else {
                     MULxw(gd, gd, ed);
@@ -1332,13 +1802,10 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             } else {
                 ANDw_mask(x2, gd, 0, 0b00100);  //mask=0x00000001f
             }
-            LSRxw_REG(x4, ed, x2);
-            if(rex.w) {
-                ANDx_mask(x4, x4, 1, 0, 0);  //mask=1
-            } else {
-                ANDw_mask(x4, x4, 0, 0);  //mask=1
+            IFX(X_CF) {
+                LSRxw_REG(x4, ed, x2);
+                BFIw(xFlags, x4, F_CF, 1);
             }
-            BFIw(xFlags, x4, F_CF, 1);
             MOV32w(x4, 1);
             LSLxw_REG(x4, x4, x2);
             BICxw_REG(ed, ed, x4);
@@ -1417,15 +1884,15 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     }
                     u8 = F8;
                     u8&=(rex.w?0x3f:0x1f);
-                    BFXILxw(xFlags, ed, u8, 1);  // inject 1 bit from u8 to F_CF (i.e. pos 0)
-                    TBNZ_MARK3(xFlags, 0); // bit already set, jump to next instruction
+                    IFX(X_CF) {
+                        BFXILxw(xFlags, ed, u8, 1);  // inject 1 bit from u8 to F_CF (i.e. pos 0)
+                    }
                     MOV32w(x4, 1);
                     ORRxw_REG_LSL(ed, ed, x4, u8);
                     if(wback) {
                         STxw(ed, wback, fixedaddress);
                         SMWRITE();
                     }
-                    MARK3;
                     break;
                 case 6:
                     INST_NAME("BTR Ed, Ib");
@@ -1442,15 +1909,14 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     }
                     u8 = F8;
                     u8&=(rex.w?0x3f:0x1f);
-                    BFXILxw(xFlags, ed, u8, 1);  // inject 1 bit from u8 to F_CF (i.e. pos 0)
-                    TBZ_MARK3(xFlags, 0); // bit already clear, jump to next instruction
-                    MOV32w(x4, 1);
-                    BICxw_REG_LSL(ed, ed, x4, u8);
+                    IFX(X_CF) {
+                        BFXILxw(xFlags, ed, u8, 1);  // inject 1 bit from u8 to F_CF (i.e. pos 0)
+                    }
+                    BFCxw(ed, u8, 1);
                     if(wback) {
                         STxw(ed, wback, fixedaddress);
                         SMWRITE();
                     }
-                    MARK3;
                     break;
                 case 7:
                     INST_NAME("BTC Ed, Ib");
@@ -1467,14 +1933,15 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     }
                     u8 = F8;
                     u8&=(rex.w?0x3f:0x1f);
-                    BFXILxw(xFlags, ed, u8, 1);  // inject 1 bit from u8 to F_CF (i.e. pos 0)
+                    IFX(X_CF) {
+                        BFXILxw(xFlags, ed, u8, 1);  // inject 1 bit from u8 to F_CF (i.e. pos 0)
+                    }
                     MOV32w(x4, 1);
                     EORxw_REG_LSL(ed, ed, x4, u8);
                     if(wback) {
                         STxw(ed, wback, fixedaddress);
                         SMWRITE();
                     }
-                    MARK3;
                     break;
                 default:
                     DEFAULT;
@@ -1503,13 +1970,10 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             } else {
                 ANDw_mask(x2, gd, 0, 0b00100);  //mask=0x00000001f
             }
-            LSRxw_REG(x4, ed, x2);
-            if(rex.w) {
-                ANDx_mask(x4, x4, 1, 0, 0);  //mask=1
-            } else {
-                ANDw_mask(x4, x4, 0, 0);  //mask=1
+            IFX(X_CF) {
+                LSRxw_REG(x4, ed, x2);
+                BFIw(xFlags, x4, F_CF, 1);
             }
-            BFIw(xFlags, x4, F_CF, 1);
             MOV32w(x4, 1);
             LSLxw_REG(x4, x4, x2);
             EORxw_REG(ed, ed, x4);
@@ -1780,12 +2244,13 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             nextop = F8;
             GETGM(d0);
             GETEM(d1, 0);
-            if(MODREG)
-                q0 = fpu_get_scratch(dyn);
-            else
-                q0 = d1;
-            NEG_64(q0, d1);
-            USHL_R_64(d0, d0, q0);
+            v0 = fpu_get_scratch(dyn);
+            //MOVI_64(v0, 64);  not 64!
+            MOV32w(x1, 64);
+            VMOVQDfrom(v0, 0, x1);
+            UMIN_32(v0, v0, d1);    // limit to 0 .. +64 values (will force 32bits upper part to 0)
+            NEG_64(v0, v0);
+            USHL_R_64(d0, d0, v0);
             break;
         case 0xD4:
             INST_NAME("PADDQ Gm,Em");
@@ -1975,7 +2440,16 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 VEOR(q0, q0, q1);
             }
             break;
-
+        case 0xF1:
+            INST_NAME("PSLLW Gm,Em");
+            nextop = F8;
+            GETGM(d0);
+            GETEM(d1, 0);
+            v0 = fpu_get_scratch(dyn);
+            VMOVHto(x1, d1, 0);
+            VDUPH(v0, x1);
+            USHL_16(d0, d0, v0);
+            break;
         case 0xF2:
             INST_NAME("PSLLD Gm,Em");
             nextop = F8;
@@ -1987,11 +2461,18 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             SQXTN_32(v0, v0); // 2*q1 in 32bits now
             SSHL_32(d0, d0, v0);
             break;
-        case 0xF4:
-            INST_NAME("PMULUDQ Gx,Ex");
+        case 0xF3:
+            INST_NAME("PSLLQ Gm,Em");
             nextop = F8;
-            GETGX(v0, 1);
-            GETEX(v1, 0, 0);
+            GETGM(d0);
+            GETEM(d1, 0);
+            USHL_R_64(d0, d0, d1);
+            break;
+        case 0xF4:
+            INST_NAME("PMULUDQ Gm,Em");
+            nextop = F8;
+            GETGM(v0);
+            GETEM(v1, 0);
             VUMULL_32(v0, v0, v1);
             break;
         case 0xF5:
@@ -2016,7 +2497,20 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             UADDLVQ_16(d1, d0);
             VMOVeD(q0, 0, d1, 0);
             break;
-
+        case 0xF7:
+            INST_NAME("MASKMOVQ Gm, Em");
+            nextop = F8;
+            GETGM(q0);
+            GETEM(q1, 0);
+            d0 = fpu_get_scratch(dyn);
+            d1 = fpu_get_scratch(dyn);
+            VSSHR_8(d1, q1, 7); // d1 = byte slection mask
+            VLDR64_U12(d0, xRDI, 0);
+            VBIC(d0, d0, d1);   // d0 = clear masked byte
+            VAND(d1, q0, d1);   // d1 = masked Gm
+            VORR(d0, d0, d1);
+            VSTR64_U12(d0, xRDI, 0);
+            break;
         case 0xF8:
             INST_NAME("PSUBB Gm, Em");
             nextop = F8;
@@ -2038,14 +2532,6 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             GETEM(v1, 0);
             VSUB_32(v0, v0, v1);
             break;
-
-        case 0xFC:
-            INST_NAME("PADDB Gm, Em");
-            nextop = F8;
-            GETGM(v0);
-            GETEM(v1, 0);
-            VADD_8(v0, v0, v1);
-            break;
         case 0xFB:
             INST_NAME("PSUBQ Gm, Em");
             nextop = F8;
@@ -2053,6 +2539,13 @@ uintptr_t dynarec64_0F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             GETEM(v1, 0);
             SUB_64(v0, v0, v1);
             break;
+        case 0xFC:
+            INST_NAME("PADDB Gm, Em");
+            nextop = F8;
+            GETGM(v0);
+            GETEM(v1, 0);
+            VADD_8(v0, v0, v1);
+            break;
         case 0xFD:
             INST_NAME("PADDW Gm, Em");
             nextop = F8;

src/dynarec/arm64/dynarec_arm64_64.c

@@ -58,6 +58,16 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
 
     switch(opcode) {
 
+        case 0x01:
+            INST_NAME("ADD Seg:Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            GETGD;
+            GETEDO(x4, 0);
+            emit_add32(dyn, ninst, rex, ed, gd, x3, x5);
+            WBACKO(x4);
+            break;
         case 0x03:
             INST_NAME("ADD Gd, Seg:Ed");
             SETFLAGS(X_ALL, SF_SET_PENDING);
@@ -223,8 +233,8 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                                     UFLAG_RES(gd);
                                     LSRx(x3, gd, 32);
                                     UFLAG_OP1(x3);
-                                    UFLAG_DF(x3, d_imul32);
                                     MOVw_REG(gd, gd);
+                                    UFLAG_DF(x3, d_imul32);
                                 } else {
                                     MULxw(gd, gd, ed);
                                 }
@@ -268,6 +278,28 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             }
             break;
 
+        case 0x21:
+            INST_NAME("AND Seg:Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            GETGD;
+            GETEDO(x4, 0);
+            emit_and32(dyn, ninst, rex, ed, gd, x3, x5);
+            WBACKO(x4);
+            break;
+
+        case 0x29:
+            INST_NAME("SUB Seg:Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            GETGD;
+            GETEDO(x4, 0);
+            emit_sub32(dyn, ninst, rex, ed, gd, x3, x5);
+            WBACKO(x4);
+            break;
+
         case 0x2B:
             INST_NAME("SUB Gd, Seg:Ed");
             SETFLAGS(X_ALL, SF_SET_PENDING);
@@ -278,6 +310,17 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             emit_sub32(dyn, ninst, rex, gd, ed, x3, x4);
             break;
 
+        case 0x31:
+            INST_NAME("XOR Seg:Ed, Gd");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            GETGD;
+            GETEDO(x4, 0);
+            emit_xor32(dyn, ninst, rex, ed, gd, x3, x5);
+            WBACKO(x4);
+            break;
+
         case 0x33:
             INST_NAME("XOR Gd, Seg:Ed");
             SETFLAGS(X_ALL, SF_SET_PENDING);
@@ -337,10 +380,22 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 }
             }
             break;
-
+        case 0x64:
+            addr = dynarec64_64(dyn, addr, ip, ninst, rex, rep, _FS, ok, need_epilog);
+            break;
+        case 0x65:
+            addr = dynarec64_64(dyn, addr, ip, ninst, rex, rep, _GS, ok, need_epilog);
+            break;
         case 0x66:
             addr = dynarec64_6664(dyn, addr, ip, ninst, rex, seg, ok, need_epilog);
             break;
+        case 0x67:
+            if(rex.is32bits)
+                addr = dynarec64_6764_32(dyn, addr, ip, ninst, rex, rep, seg, ok, need_epilog);
+            else {
+                DEFAULT;
+            }
+            break;
 
         case 0x80:
             nextop = F8;
@@ -514,7 +569,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             } else {
                 gd = (nextop&0x38)>>3;
                 gb1 = xRAX+(gd&3);
-                gb2 = ((gd&4)>>2);
+                gb2 = ((gd&4)<<1);
             }
             if(MODREG) {
                 if(rex.rex) {
@@ -538,7 +593,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 LDRB_REG(x4, wback, x4);
                 ed = x4;
             }
-            BFIx(gb1, ed, gb2*8, 8);
+            BFIx(gb1, ed, gb2, 8);
             break;
         case 0x89:
             INST_NAME("MOV Seg:Ed, Gd");
@@ -584,7 +639,22 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 }
             }
             break;
-
+        case 0x8E:
+            INST_NAME("MOV Seg,Ew");
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            u8 = (nextop&0x38)>>3;
+            if((nextop&0xC0)==0xC0) {
+                ed = xRAX+(nextop&7)+(rex.b<<3);
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
+                LDRH_REG(x1, wback, x4);
+                ed = x1;
+            }
+            STRH_U12(ed, xEmu, offsetof(x64emu_t, segs[u8]));
+            STRw_U12(wZR, xEmu, offsetof(x64emu_t, segs_serial[u8]));
+            break;
         case 0x8F:
             INST_NAME("POP FS:Ed");
             grab_segdata(dyn, addr, ninst, x4, seg);
@@ -604,6 +674,9 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 }
             }
             break;
+        case 0x90:
+            INST_NAME("NOP");
+            break;
 
         case 0xA1:
             INST_NAME("MOV EAX,FS:Od");
@@ -902,7 +975,6 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 case 4:
                     INST_NAME("MUL EAX, Ed");
                     SETFLAGS(X_ALL, SF_PENDING);
-                    UFLAG_DF(x2, rex.w?d_mul64:d_mul32);
                     GETEDO(x6, 0);
                     if(rex.w) {
                         if(ed==xRDX) gd=x3; else gd=xRDX;
@@ -916,11 +988,11 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     }
                     UFLAG_RES(xRAX);
                     UFLAG_OP1(xRDX);
+                    UFLAG_DF(x2, rex.w?d_mul64:d_mul32);
                     break;
                 case 5:
                     INST_NAME("IMUL EAX, Ed");
                     SETFLAGS(X_ALL, SF_PENDING);
-                    UFLAG_DF(x2, rex.w?d_imul64:d_imul32);
                     GETEDO(x6, 0);
                     if(rex.w) {
                         if(ed==xRDX) gd=x3; else gd=xRDX;
@@ -934,6 +1006,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     }
                     UFLAG_RES(xRAX);
                     UFLAG_OP1(xRDX);
+                    UFLAG_DF(x2, rex.w?d_imul64:d_imul32);
                     break;
                 case 6:
                     INST_NAME("DIV Ed");
@@ -1048,6 +1121,7 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     GETEDOz(x6, 0);
                     if(box64_dynarec_callret && box64_dynarec_bigblock>1) {
                         BARRIER(BARRIER_FULL);
+                        BARRIER_NEXT(BARRIER_FULL);
                     } else {
                         BARRIER(BARRIER_FLOAT);
                         *need_epilog = 0;
@@ -1055,12 +1129,15 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     }
                     GETIP_(addr);
                     if(box64_dynarec_callret) {
+                        SET_HASCALLRET();
                         // Push actual return address
                         if(addr < (dyn->start+dyn->isize)) {
                             // there is a next...
                             j64 = (dyn->insts)?(dyn->insts[ninst].epilog-(dyn->native_size)):0;
                             ADR_S20(x4, j64);
+                            MESSAGE(LOG_NONE, "\tCALLRET set return to +%di\n", j64>>2);
                         } else {
+                            MESSAGE(LOG_NONE, "\tCALLRET set return to Jmptable(%p)\n", (void*)addr);
                             j64 = getJumpTableAddress64(addr);
                             TABLE64(x4, j64);
                             LDRx_U12(x4, x4, 0);

src/dynarec/arm64/dynarec_arm64_66.c

@@ -15,6 +15,7 @@
 #include "emu/x64run_private.h"
 #include "x64trace.h"
 #include "dynarec_native.h"
+#include "custommem.h"
 
 #include "arm64_printer.h"
 #include "dynarec_arm64_private.h"
@@ -42,7 +43,7 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
     MAYUSE(j64);
     MAYUSE(lock);
 
-    while((opcode==0x2E) || (opcode==0x36) || (opcode==0x66))   // ignoring CS:, SS: or multiple 0x66
+    while((opcode==0x2E) || (opcode==0x36) || (opcode==0x26) || (opcode==0x66))   // ignoring CS:, SS:, ES: or multiple 0x66
         opcode = F8;
 
     while((opcode==0xF2) || (opcode==0xF3)) {
@@ -82,6 +83,25 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             emit_add16(dyn, ninst, x1, x2, x3, x4);
             BFIx(xRAX, x1, 0, 16);
             break;
+        case 0x06:
+            if(rex.is32bits) {
+                INST_NAME("PUSH ES");
+                LDRH_U12(x1, xEmu, offsetof(x64emu_t, segs[_ES]));
+                PUSH1_32(x1);
+            } else {
+                DEFAULT;
+            }
+            break;
+        case 0x07:
+            if(rex.is32bits) {
+                INST_NAME("POP ES");
+                POP1_32(x1);
+                STRH_U12(x1, xEmu, offsetof(x64emu_t, segs[_ES]));
+                STRw_U12(xZR, xEmu, offsetof(x64emu_t, segs_serial[_ES]));
+            } else {
+                DEFAULT;
+            }
+            break;
 
         case 0x09:
             INST_NAME("OR Ew, Gw");
@@ -112,7 +132,11 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             break;
 
         case 0x0F:
-            addr = dynarec64_660F(dyn, addr, ip, ninst, rex, rep, ok, need_epilog);
+            switch(rep) {
+                case 0: addr = dynarec64_660F(dyn, addr, ip, ninst, rex, ok, need_epilog); break;
+                case 1: addr = dynarec64_66F20F(dyn, addr, ip, ninst, rex, ok, need_epilog); break;
+                case 2: addr = dynarec64_66F30F(dyn, addr, ip, ninst, rex, ok, need_epilog); break;
+            }
             break;
         case 0x11:
             INST_NAME("ADC Ew, Gw");
@@ -175,6 +199,25 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             emit_sbb16(dyn, ninst, x1, x2, x3, x4);
             BFIx(xRAX, x1, 0, 16);
             break;
+        case 0x1E:
+            if(rex.is32bits) {
+                INST_NAME("PUSH DS");
+                LDRH_U12(x1, xEmu, offsetof(x64emu_t, segs[_DS]));
+                PUSH1_32(x1);
+            } else {
+                DEFAULT;
+            }
+            break;
+        case 0x1F:
+            if(rex.is32bits) {
+                INST_NAME("POP DS");
+                POP1_32(x1);
+                STRH_U12(x1, xEmu, offsetof(x64emu_t, segs[_DS]));
+                STRw_U12(xZR, xEmu, offsetof(x64emu_t, segs_serial[_DS]));
+            } else {
+                DEFAULT;
+            }
+            break;
 
         case 0x21:
             INST_NAME("AND Ew, Gw");
@@ -330,7 +373,7 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             if(rex.is32bits) {
                 INST_NAME("PUSH reg");
                 gd = xRAX+(opcode&0x07);
-                if (gd == xRSP) {
+                if (gd==xRSP) {
                     MOVw_REG(x1, xRSP);
                     PUSH1_16(x1);
                 } else {
@@ -415,7 +458,6 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             }
             SETFLAGS(X_ALL, SF_PENDING);
             nextop = F8;
-            UFLAG_DF(x1, d_imul16);
             GETSEW(x1, (opcode==0x69)?2:1);
             if(opcode==0x69) i32 = F16S; else i32 = F8S;
             MOV32w(x2, i32);
@@ -423,6 +465,7 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             UFLAG_RES(x2);
             gd=x2;
             GWBACK;
+            UFLAG_DF(x1, d_imul16);
             break;
 
         case 0x70:
@@ -548,19 +591,36 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             } else {
                 GETGD;
                 addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                SMDMB();
-                TSTx_mask(ed, 1, 0, 0);    // mask=1
-                B_MARK(cNE);
-                MARKLOCK;
-                LDAXRH(x1, ed);
-                STLXRH(x3, gd, ed);
-                CBNZx_MARKLOCK(x3);
-                B_MARK2_nocond;
-                MARK;
-                LDRH_U12(x1, ed, 0);
-                STRH_U12(gd, ed, 0);
-                MARK2;
-                SMDMB();
+                if(!ALIGNED_ATOMICH) {
+                    TSTx_mask(ed, 1, 0, 0);    // mask=1
+                    B_MARK(cNE);
+                }
+                if(arm64_atomics) {
+                    SWPALH(gd, x1, ed);
+                    SMDMB();
+                    if(!ALIGNED_ATOMICH) {
+                        B_MARK2_nocond;
+                    }
+                } else {
+                    MARKLOCK;
+                    LDAXRH(x1, ed);
+                    STLXRH(x3, gd, ed);
+                    CBNZx_MARKLOCK(x3);
+                    SMDMB();
+                    if(!ALIGNED_ATOMICH) {
+                        B_MARK2_nocond;
+                    }
+                }
+                if(!ALIGNED_ATOMICH) {
+                    MARK;
+                    LDRH_U12(x1, ed, 0);
+                    LDAXRB(x3, ed);
+                    STLXRB(x3, gd, ed);
+                    CBNZx_MARK(x3);
+                    STRH_U12(gd, ed, 0);
+                    SMDMB();
+                    MARK2;
+                }
                 BFIx(gd, x1, 0, 16);
             }
             break;
@@ -600,7 +660,7 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             INST_NAME("MOV EW, Seg");
             nextop=F8;
             u8 = (nextop&0x38)>>3;
-            LDRw_U12(x3, xEmu, offsetof(x64emu_t, segs[u8]));
+            LDRH_U12(x3, xEmu, offsetof(x64emu_t, segs[u8]));
             if((nextop&0xC0)==0xC0) {   // reg <= seg
                 UXTHw(xRAX+(nextop&7)+(rex.b<<3), x3);
             } else {                    // mem <= seg
@@ -610,6 +670,22 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             }
             break;
 
+        case 0x8E:
+            INST_NAME("MOV Seg,Ew");
+            nextop = F8;
+            u8 = (nextop&0x38)>>3;
+            if((nextop&0xC0)==0xC0) {
+                ed = xRAX+(nextop&7)+(rex.b<<3);
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
+                LDH(x1, wback, fixedaddress);
+                ed = x1;
+            }
+            STRH_U12(ed, xEmu, offsetof(x64emu_t, segs[u8]));
+            STRw_U12(wZR, xEmu, offsetof(x64emu_t, segs_serial[u8]));
+            break;
+
             case 0x90:
             case 0x91:
             case 0x92:
@@ -634,7 +710,11 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             SXTBw(x1, xRAX);
             BFIx(xRAX, x1, 0, 16);
             break;
-
+        case 0x99:
+            INST_NAME("CWD");
+            SXTHw(x1, xRAX);
+            BFXILx(xRDX, x1, 16, 16);
+            break;
         case 0x9C:
             INST_NAME("PUSHF");
             READFLAGS(X_ALL);
@@ -665,7 +745,8 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             else
                 u64 = F64;
             MOV64z(x1, u64);
-            SMREAD();
+            if(isLockAddress(u64)) lock=1; else lock = 0;
+            SMREADLOCK(lock);
             LDRH_U12(x2, x1, 0);
             BFIx(xRAX, x2, 0, 16);
             break;
@@ -677,8 +758,9 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             else
                 u64 = F64;
             MOV64z(x1, u64);
+            if(isLockAddress(u64)) lock=1; else lock = 0;
             STRH_U12(xRAX, x1, 0);
-            SMWRITE();
+            SMWRITELOCK(lock);
             break;
 
         case 0xA5:
@@ -864,23 +946,29 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("ROL Ew, Ib");
-                    MESSAGE(LOG_DUMP, "Need Optimization\n");
-                    SETFLAGS(X_OF|X_CF, SF_SET);
-                    GETEW(x1, 1);
-                    u8 = F8;
-                    MOV32w(x2, u8);
-                    CALL_(rol16, x1, x3);
-                    EWBACK;
+                    if(geted_ib(dyn, addr, ninst, nextop)&15) {
+                        SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
+                        GETEW(x1, 1);
+                        u8 = F8;
+                        emit_rol16c(dyn, ninst, x1, u8&15, x4, x5);
+                        EWBACK;
+                    } else {
+                        FAKEED;
+                        F8;
+                    }
                     break;
                 case 1:
                     INST_NAME("ROR Ew, Ib");
-                    MESSAGE(LOG_DUMP, "Need Optimization\n");
-                    SETFLAGS(X_OF|X_CF, SF_SET);
-                    GETEW(x1, 1);
-                    u8 = F8;
-                    MOV32w(x2, u8);
-                    CALL_(ror16, x1, x3);
-                    EWBACK;
+                    if(geted_ib(dyn, addr, ninst, nextop)&15) {
+                        SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
+                        GETEW(x1, 1);
+                        u8 = F8;
+                        emit_ror16c(dyn, ninst, x1, u8&15, x4, x5);
+                        EWBACK;
+                    } else {
+                        FAKEED;
+                        F8;
+                    }
                     break;
                 case 2:
                     INST_NAME("RCL Ew, Ib");
@@ -907,42 +995,42 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 case 4:
                 case 6:
                     INST_NAME("SHL Ew, Ib");
-                    UFLAG_IF {MESSAGE(LOG_DUMP, "Need Optimization for flags\n");}
-                    SETFLAGS(X_ALL, SF_PENDING);
-                    GETEW(x1, 1);
-                    u8 = F8;
-                    UFLAG_IF {MOV32w(x2, (u8&0x1f));}
-                    UFLAG_OP12(ed, x2)
-                    LSLw_IMM(ed, ed, u8&0x1f);
-                    EWBACK;
-                    UFLAG_RES(ed);
-                    UFLAG_DF(x3, d_shl16);
+                    if(geted_ib(dyn, addr, ninst, nextop)&0x1f) {
+                        SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                        GETEW(x1, 0);
+                        u8 = (F8)&0x1f;
+                        emit_shl16c(dyn, ninst, x1, u8, x5, x4);
+                        EWBACK;
+                    } else {
+                        FAKEED;
+                        F8;
+                    }
                     break;
                 case 5:
                     INST_NAME("SHR Ed, Ib");
-                    UFLAG_IF {MESSAGE(LOG_DUMP, "Need Optimization for flags\n");}
-                    SETFLAGS(X_ALL, SF_PENDING);
-                    GETEW(x1, 1);
-                    u8 = F8;
-                    UFLAG_IF {MOV32w(x2, (u8&0x1f));}
-                    UFLAG_OP12(ed, x2)
-                    LSRw_IMM(ed, ed, u8&0x1f);
-                    EWBACK;
-                    UFLAG_RES(ed);
-                    UFLAG_DF(x3, d_shr16);
+                    if(geted_ib(dyn, addr, ninst, nextop)&0x1f) {
+                        SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                        GETEW(x1, 0);
+                        u8 = (F8)&0x1f;
+                        emit_shr16c(dyn, ninst, x1, u8, x5, x4);
+                        EWBACK;
+                    } else {
+                        FAKEED;
+                        F8;
+                    }
                     break;
                 case 7:
                     INST_NAME("SAR Ed, Ib");
-                    SETFLAGS(X_ALL, SF_PENDING);
-                    UFLAG_IF {MESSAGE(LOG_DUMP, "Need Optimization for flags\n");}
-                    GETSEW(x1, 1);
-                    u8 = F8;
-                    UFLAG_IF {MOV32w(x2, (u8&0x1f));}
-                    UFLAG_OP12(ed, x2)
-                    ASRw(ed, ed, (u8&0x1f));
-                    EWBACK;
-                    UFLAG_RES(ed);
-                    UFLAG_DF(x3, d_sar16);
+                    if(geted_ib(dyn, addr, ninst, nextop)&0x1f) {
+                        SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                        GETSEW(x1, 0);
+                        u8 = (F8)&0x1f;
+                        emit_sar16c(dyn, ninst, x1, u8, x5, x4);
+                        EWBACK;
+                    } else {
+                        FAKEED;
+                        F8;
+                    }
                     break;
             }
             break;
@@ -965,108 +1053,186 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             break;
 
         case 0xD1:
-        case 0xD3:
             nextop = F8;
             switch((nextop>>3)&7) {
                 case 0:
-                    if(opcode==0xD1) {
-                        INST_NAME("ROL Ew, 1");
-                        MOV32w(x2, 1);
-                    } else {
-                        INST_NAME("ROL Ew, CL");
-                        ANDw_mask(x2, xRCX, 0, 0b00100);
-                    }
-                    MESSAGE(LOG_DUMP, "Need Optimization\n");
-                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    INST_NAME("ROL Ew, 1");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
                     GETEW(x1, 0);
-                    CALL_(rol16, x1, x3);
+                    emit_rol16c(dyn, ninst, x1, 1, x5, x4);
                     EWBACK;
                     break;
                 case 1:
-                    if(opcode==0xD1) {
-                        INST_NAME("ROR Ew, 1");
-                        MOV32w(x2, 1);
-                    } else {
-                        INST_NAME("ROR Ew, CL");
-                        ANDw_mask(x2, xRCX, 0, 0b00100);
-                    }
-                    MESSAGE(LOG_DUMP, "Need Optimization\n");
-                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    INST_NAME("ROR Ew, 1");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET_PENDING);
                     GETEW(x1, 0);
-                    CALL_(ror16, x1, x3);
+                    emit_ror16c(dyn, ninst, x1, 1, x5, x4);
                     EWBACK;
                     break;
                 case 2:
-                    if(opcode==0xD1) {INST_NAME("RCL Ew, 1"); } else { INST_NAME("RCL Ew, CL");}
+                    INST_NAME("RCL Ew, 1");
                     MESSAGE(LOG_DUMP, "Need Optimization\n");
                     READFLAGS(X_CF);
                     SETFLAGS(X_OF|X_CF, SF_SET);
-                    if(opcode==0xD1) {MOV32w(x2, 1);} else {ANDw_mask(x2, xRCX, 0, 0b00100);}
+                    MOV32w(x2, 1);
                     GETEW(x1, 0);
                     CALL_(rcl16, x1, x3);
                     EWBACK;
                     break;
                 case 3:
-                    if(opcode==0xD1) {INST_NAME("RCR Ew, 1");} else {INST_NAME("RCR Ew, CL");}
+                    INST_NAME("RCR Ew, 1");
                     MESSAGE(LOG_DUMP, "Need Optimization\n");
                     READFLAGS(X_CF);
                     SETFLAGS(X_OF|X_CF, SF_SET);
-                    if(opcode==0xD1) {MOV32w(x2, 1);} else {ANDw_mask(x2, xRCX, 0, 0b00100);}
+                    MOV32w(x2, 1);
                     GETEW(x1, 0);
                     CALL_(rcr16, x1, x3);
                     EWBACK;
                     break;
                 case 4:
                 case 6:
-                    if(opcode==0xD1) {
-                        INST_NAME("SHL Ew, 1");
-                        MOV32w(x4, 1);
-                    } else {
-                        INST_NAME("SHL Ew, CL");
-                        ANDw_mask(x4, xRCX, 0, 0b00100);
-                    }
-                    UFLAG_IF {MESSAGE(LOG_DUMP, "Need Optimization for flags\n");}
-                    SETFLAGS(X_ALL, SF_PENDING);
+                    INST_NAME("SHL Ew, 1");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
                     GETEW(x1, 0);
-                    UFLAG_OP12(ed, x4)
-                    LSLw_REG(ed, ed, x4);
+                    emit_shl16c(dyn, ninst, x1, 1, x5, x4);
                     EWBACK;
-                    UFLAG_RES(ed);
-                    UFLAG_DF(x3, d_shl16);
                     break;
                 case 5:
-                    if(opcode==0xD1) {
-                        INST_NAME("SHR Ew, 1");
-                        MOV32w(x4, 1);
-                    } else {
-                        INST_NAME("SHR Ew, CL");
-                        ANDw_mask(x4, xRCX, 0, 0b00100);
-                    }
-                    UFLAG_IF {MESSAGE(LOG_DUMP, "Need Optimization for flags\n");}
-                    SETFLAGS(X_ALL, SF_PENDING);
+                    INST_NAME("SHR Ew, 1");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
                     GETEW(x1, 0);
-                    UFLAG_OP12(ed, x4)
-                    LSRw_REG(ed, ed, x4);
+                    emit_shr16c(dyn, ninst, x1, 1, x5, x4);
                     EWBACK;
-                    UFLAG_RES(ed);
-                    UFLAG_DF(x3, d_shr16);
                     break;
                 case 7:
-                    if(opcode==0xD1) {
-                        INST_NAME("SAR Ew, 1");
-                        MOV32w(x4, 1);
-                    } else {
-                        INST_NAME("SAR Ew, CL");
-                        ANDw_mask(x4, xRCX, 0, 0b00100);
-                    }
-                    UFLAG_IF {MESSAGE(LOG_DUMP, "Need Optimization for flags\n");}
-                    SETFLAGS(X_ALL, SF_PENDING);
+                    INST_NAME("SAR Ew, 1");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
                     GETSEW(x1, 0);
-                    UFLAG_OP12(ed, x4)
-                    ASRw_REG(ed, ed, x4);
+                    emit_sar16c(dyn, ninst, x1, 1, x5, x4);
+                    EWBACK;
+                    break;
+            }
+            break;
+
+        case 0xD3:
+            nextop = F8;
+            switch((nextop>>3)&7) {
+                case 0:
+                    INST_NAME("ROL Ew, CL");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    if(box64_dynarec_safeflags>1)
+                        MAYSETFLAGS();
+                    UFLAG_IF {
+                        TSTw_mask(xRCX, 0, 0b00100);  //mask=0x00000001f
+                        B_NEXT(cEQ);
+                    }
+                    ANDw_mask(x2, xRCX, 0, 0b00011);  //mask=0x00000000f
+                    MOV32w(x4, 16);
+                    SUBx_REG(x2, x4, x2);
+                    GETEW(x1, 0);
+                    ORRw_REG_LSL(ed, ed, ed, 16);
+                    LSRw_REG(ed, ed, x2);
+                    EWBACK;
+                    UFLAG_IF {  // calculate flags directly
+                        CMPSw_U12(x2, 15);
+                        B_MARK(cNE);
+                            ADDxw_REG_LSR(x3, ed, ed, 15);
+                            BFIw(xFlags, x3, F_OF, 1);
+                        MARK;
+                        BFIw(xFlags, ed, F_CF, 1);
+                        UFLAG_DF(x2, d_none);
+                    }
+                    break;
+                case 1:
+                    INST_NAME("ROR Ew, CL");
+                    SETFLAGS(X_OF|X_CF, SF_SUBSET);
+                    if(box64_dynarec_safeflags>1)
+                        MAYSETFLAGS();
+                    UFLAG_IF {
+                        TSTw_mask(xRCX, 0, 0b00100);  //mask=0x00000001f
+                        B_NEXT(cEQ);
+                    }
+                    ANDw_mask(x2, xRCX, 0, 0b00011);  //mask=0x00000000f
+                    GETEW(x1, 0);
+                    ORRw_REG_LSL(ed, ed, ed, 16);
+                    LSRw_REG(ed, ed, x2);
+                    EWBACK;
+                    UFLAG_IF {  // calculate flags directly
+                        CMPSw_U12(x2, 1);
+                        B_MARK(cNE);
+                            LSRxw(x2, ed, 14); // x2 = d>>14
+                            EORw_REG_LSR(x2, x2, x2, 1); // x2 = ((d>>14) ^ ((d>>14)>>1))
+                            BFIw(xFlags, x2, F_OF, 1);
+                        MARK;
+                        BFXILw(xFlags, ed, 15, 1);
+                        UFLAG_DF(x2, d_none);
+                    }
+                    break;
+                case 2:
+                    INST_NAME("RCL Ew, CL");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    ANDw_mask(x2, xRCX, 0, 0b00100);
+                    GETEW(x1, 0);
+                    CALL_(rcl16, x1, x3);
+                    EWBACK;
+                    break;
+                case 3:
+                    INST_NAME("RCR Ew, CL");
+                    MESSAGE(LOG_DUMP, "Need Optimization\n");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    ANDw_mask(x2, xRCX, 0, 0b00100);
+                    GETEW(x1, 0);
+                    CALL_(rcr16, x1, x3);
+                    EWBACK;
+                    break;
+                case 4:
+                case 6:
+                    INST_NAME("SHL Ew, CL");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                    if(box64_dynarec_safeflags>1)
+                        MAYSETFLAGS();
+                    UFLAG_IF {
+                        ANDSw_mask(x2, xRCX, 0, 0b00100);  //mask=0x00000001f
+                        B_NEXT(cEQ);
+                    } else {
+                        ANDw_mask(x2, xRCX, 0, 0b00100);  //mask=0x00000001f
+                    }
+                    GETEW(x1, 0);
+                    emit_shl16(dyn, ninst, x1, x2, x5, x4);
+                    EWBACK;
+                    break;
+                case 5:
+                    INST_NAME("SHR Ew, CL");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                    if(box64_dynarec_safeflags>1)
+                        MAYSETFLAGS();
+                    UFLAG_IF {
+                        ANDSw_mask(x2, xRCX, 0, 0b00100);  //mask=0x00000001f
+                        B_NEXT(cEQ);
+                    } else {
+                        ANDw_mask(x2, xRCX, 0, 0b00100);  //mask=0x00000001f
+                    }
+                    GETEW(x1, 0);
+                    emit_shr16(dyn, ninst, x1, x2, x5, x4);
+                    EWBACK;
+                    break;
+                case 7:
+                    INST_NAME("SAR Ew, CL");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+                    if(box64_dynarec_safeflags>1)
+                        MAYSETFLAGS();
+                    UFLAG_IF {
+                        ANDSw_mask(x2, xRCX, 0, 0b00100);  //mask=0x00000001f
+                        B_NEXT(cEQ);
+                    } else {
+                        ANDw_mask(x2, xRCX, 0, 0b00100);  //mask=0x00000001f
+                    }
+                    GETSEW(x1, 0);
+                    emit_sar16(dyn, ninst, x1, x2, x5, x4);
                     EWBACK;
-                    UFLAG_RES(ed);
-                    UFLAG_DF(x3, d_sar16);
                     break;
             }
             break;
@@ -1102,24 +1268,24 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 case 4:
                     INST_NAME("MUL AX, Ew");
                     SETFLAGS(X_ALL, SF_PENDING);
-                    UFLAG_DF(x1, d_mul16);
                     GETEW(x1, 0);
                     UXTHw(x2, xRAX);
                     MULw(x1, x2, x1);
                     UFLAG_RES(x1);
                     BFIx(xRAX, x1, 0, 16);
                     BFXILx(xRDX, x1, 16, 16);
+                    UFLAG_DF(x1, d_mul16);
                     break;
                 case 5:
                     INST_NAME("IMUL AX, Ew");
                     SETFLAGS(X_ALL, SF_PENDING);
-                    UFLAG_DF(x1, d_imul16);
                     GETSEW(x1, 0);
                     SXTHw(x2, xRAX);
                     MULw(x1, x2, x1);
                     UFLAG_RES(x1);
                     BFIx(xRAX, x1, 0, 16);
                     BFXILx(xRDX, x1, 16, 16);
+                    UFLAG_DF(x1, d_imul16);
                     break;
                 case 6:
                     INST_NAME("DIV Ew");
@@ -1134,7 +1300,7 @@ uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     break;
                 case 7:
                     INST_NAME("IDIV Ew");
-                    NOTEST(x1);
+                    SKIPTEST(x1);
                     SETFLAGS(X_ALL, SF_SET);
                     GETSEW(x1, 0);
                     UXTHw(x2, xRAX);

src/dynarec/arm64/dynarec_arm64_660f.c

@@ -20,10 +20,11 @@
 #include "dynarec_arm64_private.h"
 #include "dynarec_arm64_functions.h"
 #include "dynarec_arm64_helper.h"
+#include "emu/x64compstrings.h"
 
-uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
 {
-    (void)ip; (void)rep; (void)need_epilog;
+    (void)ip; (void)need_epilog;
 
     uint8_t opcode = F8;
     uint8_t nextop, u8;
@@ -48,6 +49,8 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
     MAYUSE(j64);
     #if STEP > 1
     static const int8_t mask_shift8[] = { -7, -6, -5, -4, -3, -2, -1, 0 };
+    static const int8_t mask_string8[] = { 7, 6, 5, 4, 3, 2, 1, 0 };
+    static const int8_t mask_string16[] = { 15, 14, 13, 12, 11, 10, 9, 8 };
     static const int8_t round_round[] = { 0, 2, 1, 3};
     #endif
 
@@ -169,7 +172,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             nextop = F8;
             FAKEED;
             break;
-        
+
         case 0x28:
             INST_NAME("MOVAPD Gx,Ex");
             nextop = F8;
@@ -201,7 +204,92 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 SMWRITE2();
             }
             break;
-
+        case 0x2A:
+            INST_NAME("CVTPI2PD Gx,Em");
+            nextop = F8;
+            GETGX(v0, 1);
+            GETEM(q1, 0);
+            SXTL_32(v0, q1);
+            SCVTQFD(v0, v0);
+            break;
+        case 0x2B:
+            INST_NAME("MOVNTPD Ex,Gx");
+            nextop = F8;
+            GETGX(v0, 0);
+            if(MODREG) {
+                ed = (nextop&7)+(rex.b<<3);
+                v1 = sse_get_reg_empty(dyn, ninst, x1, ed);
+                VMOVQ(v1, v0);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                VST128(v0, ed, fixedaddress);
+            }
+            break;
+        case 0x2C:
+            INST_NAME("CVTTPD2PI Gm,Ex");
+            nextop = F8;
+            GETGM(q0);
+            GETEX(v1, 0, 0);
+            if(box64_dynarec_fastround) {
+                VFCVTZSQD(q0, v1);
+                SQXTN_32(q0, q0);
+            } else {
+                MRS_fpsr(x5);
+                BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
+                MSR_fpsr(x5);
+                ORRw_mask(x2, xZR, 1, 0);    //0x80000000
+                d0 = fpu_get_scratch(dyn);
+                for (int i=0; i<2; ++i) {
+                    BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
+                    if (i) {
+                        VMOVeD(d0, 0, v1, i);
+                        FRINTZD(d0, d0);
+                    } else {
+                        FRINTZD(d0, v1);
+                    }
+                    FCVTZSwD(x1, d0);
+                    MRS_fpsr(x5);   // get back FPSR to check the IOC bit
+                    TBZ(x5, FPSR_IOC, 4+4);
+                    MOVw_REG(x1, x2);
+                    VMOVQSfrom(q0, i, x1);
+                }
+            }
+            break;
+        case 0x2D:
+            INST_NAME("CVTPD2PI Gm,Ex");
+            nextop = F8;
+            GETGM(q0);
+            GETEX(v1, 0, 0);
+            if(box64_dynarec_fastround) {
+                u8 = sse_setround(dyn, ninst, x1, x2, x3);
+                VFRINTIDQ(q0, v1);
+                FCVTXN(q0, q0);
+                x87_restoreround(dyn, ninst, u8);
+                VFCVTZSS(q0, q0);
+            } else {
+                u8 = sse_setround(dyn, ninst, x1, x2, x3);
+                MRS_fpsr(x5);
+                BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
+                MSR_fpsr(x5);
+                ORRw_mask(x2, xZR, 1, 0);    //0x80000000
+                d0 = fpu_get_scratch(dyn);
+                for (int i=0; i<2; ++i) {
+                    BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
+                    if (i) {
+                        VMOVeD(d0, 0, v1, i);
+                        FRINTID(d0, d0);
+                    } else {
+                        FRINTID(d0, v1);
+                    }
+                    FCVTZSwD(x1, d0);
+                    MRS_fpsr(x5);   // get back FPSR to check the IOC bit
+                    TBZ(x5, FPSR_IOC, 4+4);
+                    MOVw_REG(x1, x2);
+                    VMOVQSfrom(q0, i, x1);
+                }
+                x87_restoreround(dyn, ninst, u8);
+            }
+            break;
         case 0x2E:
             // no special check...
         case 0x2F:
@@ -241,7 +329,16 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     GETEX(q1, 0, 0);
                     VADDPQ_32(q0, q0, q1);
                     break;
-
+                case 0x03:
+                    INST_NAME("PHADDSW Gx, Ex");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    GETEX(q1, 0, 0);
+                    v0 = fpu_get_scratch(dyn);
+                    VUZP1Q_16(v0, q0, q1);
+                    VUZP2Q_16(q0, q0, q1);
+                    SQADDQ_16(q0, q0, v0);
+                    break;
                 case 0x04:
                     INST_NAME("PMADDUBSW Gx,Ex");
                     nextop = F8;
@@ -443,14 +540,14 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 case 0x20:
                     INST_NAME("PMOVSXBW Gx, Ex");  // SSE4 opcode!
                     nextop = F8;
-                    GETEX(q1, 0, 0);
+                    GETEX64(q1, 0, 0);
                     GETGX_empty(q0);
                     SXTL_8(q0, q1);     // 8bits->16bits
                     break;
                 case 0x21:
                     INST_NAME("PMOVSXBD Gx, Ex");  // SSE4 opcode!
                     nextop = F8;
-                    GETEX(q1, 0, 0);
+                    GETEX32(q1, 0, 0);
                     GETGX_empty(q0);
                     SXTL_8(q0, q1);     // 8bits->16bits
                     SXTL_16(q0, q0);    //16bits->32bits
@@ -458,7 +555,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 case 0x22:
                     INST_NAME("PMOVSXBQ Gx, Ex");  // SSE4 opcode!
                     nextop = F8;
-                    GETEX(q1, 0, 0);
+                    GETEX16(q1, 0, 0);
                     GETGX_empty(q0);
                     SXTL_8(q0, q1);     // 8bits->16bits
                     SXTL_16(q0, q0);    //16bits->32bits
@@ -467,14 +564,14 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 case 0x23:
                     INST_NAME("PMOVSXWD Gx, Ex");  // SSE4 opcode!
                     nextop = F8;
-                    GETEX(q1, 0, 0);
+                    GETEX64(q1, 0, 0);
                     GETGX_empty(q0);
                     SXTL_16(q0, q1);     // 16bits->32bits
                     break;
                 case 0x24:
                     INST_NAME("PMOVSXWQ Gx, Ex");  // SSE4 opcode!
                     nextop = F8;
-                    GETEX(q1, 0, 0);
+                    GETEX32(q1, 0, 0);
                     GETGX_empty(q0);
                     SXTL_16(q0, q1);     // 16bits->32bits
                     SXTL_32(q0, q0);     // 32bits->64bits
@@ -482,7 +579,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 case 0x25:
                     INST_NAME("PMOVSXDQ Gx, Ex");  // SSE4 opcode!
                     nextop = F8;
-                    GETEX(q1, 0, 0);
+                    GETEX64(q1, 0, 0);
                     GETGX_empty(q0);
                     SXTL_32(q0, q1);     // 32bits->64bits
                     break;
@@ -539,14 +636,14 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 case 0x30:
                     INST_NAME("PMOVZXBW Gx, Ex");  // SSE4 opcode!
                     nextop = F8;
-                    GETEX(q1, 0, 0);
+                    GETEX64(q1, 0, 0);
                     GETGX_empty(q0);
                     UXTL_8(q0, q1);     // 8bits->16bits
                     break;
                 case 0x31:
                     INST_NAME("PMOVZXBD Gx, Ex");  // SSE4 opcode!
                     nextop = F8;
-                    GETEX(q1, 0, 0);
+                    GETEX32(q1, 0, 0);
                     GETGX_empty(q0);
                     UXTL_8(q0, q1);     // 8bits->16bits
                     UXTL_16(q0, q0);    //16bits->32bits
@@ -554,7 +651,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 case 0x32:
                     INST_NAME("PMOVZXBQ Gx, Ex");  // SSE4 opcode!
                     nextop = F8;
-                    GETEX(q1, 0, 0);
+                    GETEX16(q1, 0, 0);
                     GETGX_empty(q0);
                     UXTL_8(q0, q1);     // 8bits->16bits
                     UXTL_16(q0, q0);    //16bits->32bits
@@ -563,14 +660,14 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 case 0x33:
                     INST_NAME("PMOVZXWD Gx, Ex");  // SSE4 opcode!
                     nextop = F8;
-                    GETEX(q1, 0, 0);
+                    GETEX64(q1, 0, 0);
                     GETGX_empty(q0);
                     UXTL_16(q0, q1);     // 16bits->32bits
                     break;
                 case 0x34:
                     INST_NAME("PMOVZXWQ Gx, Ex");  // SSE4 opcode!
                     nextop = F8;
-                    GETEX(q1, 0, 0);
+                    GETEX32(q1, 0, 0);
                     GETGX_empty(q0);
                     UXTL_16(q0, q1);     // 16bits->32bits
                     UXTL_32(q0, q0);     // 32bits->64bits
@@ -578,11 +675,18 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 case 0x35:
                     INST_NAME("PMOVZXDQ Gx, Ex");  // SSE4 opcode!
                     nextop = F8;
-                    GETEX(q1, 0, 0);
+                    GETEX64(q1, 0, 0);
                     GETGX_empty(q0);
                     UXTL_32(q0, q1);     // 32bits->64bits
                     break;
 
+                case 0x37:
+                    INST_NAME("PCMPGTQ Gx, Ex"); // SSE4 opcode!
+                    nextop = F8;
+                    GETEX(q1, 0, 0);
+                    GETGX(q0, 1);
+                    VCMGTQ_64(q0, q0, q1);
+                    break;
                 case 0x38:
                     INST_NAME("PMINSB Gx, Ex");  // SSE4 opcode!
                     nextop = F8;
@@ -757,7 +861,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                         BFIx(gd, x1, 0, 16);
                     } else {
                         SMREAD();
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, (1<<2)-1, rex, NULL, 0, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<1, (1<<1)-1, rex, NULL, 0, 0);
                         LDH(x1, ed, fixedaddress);
                         REV16x(x1, x1);
                         BFIx(gd, x1, 0, 16);
@@ -773,7 +877,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                         BFIx(ed, x1, 0, 16);
                     } else {
                         SMREAD();
-                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, (1<<2)-1, rex, NULL, 0, 0);
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<1, (1<<1)-1, rex, NULL, 0, 0);
                         REV16x(x1, gd);
                         STH(x1, ed, fixedaddress);
                     }
@@ -863,7 +967,20 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                             VMOVeS(q0, i, q1, i);
                         }
                     break;
-
+                case 0x0D:
+                    INST_NAME("PBLENDPD Gx, Ex, Ib");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    GETEX(q1, 0, 1);
+                    u8 = F8&0b11;
+                    if(u8==0b01) {
+                        VMOVeD(q0, 0, q1, 0);
+                    } else if(u8==0b10) {
+                        VMOVeD(q0, 1, q1, 1);
+                    } else if(u8==0b11) {
+                        VMOVQ(q0, q1);
+                    }
+                    break;
                 case 0x0E:
                     INST_NAME("PBLENDW Gx, Ex, Ib");
                     nextop = F8;
@@ -907,7 +1024,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     GETEX(q1, 0, 1);
                     u8 = F8;
                     if(u8>31) {
-                        VEORQ(q0, q0, q0);    
+                        VEORQ(q0, q0, q0);
                     } else if(u8>15) {
                         d0 = fpu_get_scratch(dyn);
                         VEORQ(d0, d0, d0);
@@ -970,7 +1087,22 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                         SMWRITE2();
                     }
                     break;
-
+                case 0x17:
+                    INST_NAME("EXTRACTPS Ew, Gx, Ib");
+                    nextop = F8;
+                    GETGX(q0, 0);
+                    if (MODREG) {
+                        ed = xRAX+(nextop&7)+(rex.b<<3);
+                        u8 = F8&0b11;
+                        MOVx_REG(ed, xZR);
+                        VMOVSto(ed, q0, u8);
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0xfff<<2, 3, rex, NULL, 0, 1);
+                        u8 = F8&0b11;
+                        VMOVSto(x1, q0, u8);
+                        STW(x1, wback, fixedaddress);
+                    }
+                    break;
                 case 0x20:
                     INST_NAME("PINSRB Gx, ED, Ib");
                     nextop = F8;
@@ -979,7 +1111,30 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     u8 = F8;
                     VMOVQBfrom(q0, (u8&15), ed);
                     break;
-
+                case 0x21:
+                    INST_NAME("INSERTPS Gx, Ex, Ib");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    d0 = fpu_get_scratch(dyn);
+                    VMOVQ(d0, q0);
+                    if (MODREG) {
+                        q1 = sse_get_reg(dyn, ninst, x1, (nextop & 7) + (rex.b << 3), 0);
+                        u8 = F8;
+                        VMOVeS(q0, (u8>>4)&3, q1, (u8>>6)&3);
+                    } else {
+                        SMREAD();
+                        addr = geted(dyn, addr, ninst, nextop, &wback, x1, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 1);
+                        u8 = F8;
+                        LDW(x2, wback, fixedaddress);
+                        VMOVQSfrom(q0, (u8>>4)&3, x2);
+                    }
+                    uint8_t zmask = u8 & 0xf;
+                    for (uint8_t i=0; i<4; i++) {
+                        if (zmask & (1<<i)) {
+                            VMOVQSfrom(q0, i, wZR);
+                        }
+                    }
+                    break;
                 case 0x22:
                     INST_NAME("PINSRD Gx, ED, Ib");
                     nextop = F8;
@@ -993,6 +1148,28 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     }
                     break;
 
+                case 0x40:
+                    INST_NAME("DPPS Gx, Ex, Ib");
+                    nextop = F8;
+                    GETGX(q0, 1);
+                    GETEX(q1, 0, 1);
+                    u8 = F8;
+                    v0 = fpu_get_scratch(dyn);
+                    VFMULQS(v0, q0, q1);
+                    // mask some, duplicate all, mask some
+                    for(int i=0; i<4; ++i)
+                        if(!(u8&(1<<(4+i)))) {
+                            VMOVQSfrom(v0, i, xZR);
+                        }
+                    VFADDPQS(v0, v0, v0);
+                    FADDPS(v0, v0);
+                    VDUPQ_32(q0, v0, 0);
+                    for(int i=0; i<4; ++i)
+                        if(!(u8&(1<<i))) {
+                            VMOVQSfrom(q0, i, xZR);
+                        }
+                    break;
+
                 case 0x44:
                     INST_NAME("PCLMULQDQ Gx, Ex, Ib");
                     nextop = F8;
@@ -1021,7 +1198,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                         sse_forget_reg(dyn, ninst, gd);
                         MOV32w(x1, gd); // gx
                         if(MODREG) {
-                            ed = (nextop&7)+(rex.b<<3); 
+                            ed = (nextop&7)+(rex.b<<3);
                             sse_forget_reg(dyn, ninst, ed);
                             MOV32w(x2, ed);
                             MOV32w(x3, 0);  //p = NULL
@@ -1038,6 +1215,181 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     }
                     break;
 
+                case 0x60:
+                    INST_NAME("PCMPESTRM Gx, Ex, Ib");
+                    SETFLAGS(X_OF|X_CF|X_AF|X_ZF|X_SF|X_PF, SF_SET);
+                    nextop = F8;
+                    GETG;
+                    sse_forget_reg(dyn, ninst, gd);
+                    ADDx_U12(x3, xEmu, offsetof(x64emu_t, xmm[gd]));
+                    if(MODREG) {
+                        ed = (nextop&7)+(rex.b<<3);
+                        sse_reflect_reg(dyn, ninst, ed);
+                        ADDx_U12(x1, xEmu, offsetof(x64emu_t, xmm[ed]));
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 1);
+                        if(ed!=x1) {
+                            MOVx_REG(x1, ed);
+                        }
+                    }
+                    MOVx_REG(x2, xRDX);
+                    MOVx_REG(x4, xRAX);
+                    u8 = F8;
+                    MOV32w(x5, u8);
+                    CALL(sse42_compare_string_explicit_len, x1);
+                    q0 = sse_get_reg_empty(dyn, ninst, x2, gd);
+                    q1 = fpu_get_scratch(dyn);
+                    if(u8&0b1000000) {
+                        switch(u8&1) {
+                            case 0b00:
+                                VDUPQB(q0, x1); // load the low 8bits of the mask
+                                LSRw_IMM(x1, x1, 8);
+                                VDUPQB(q1, x1); // load the high 8bits of the mask
+                                VEXTQ_8(q0, q0, q1, 8); // low and hig bits mask
+                                TABLE64(x2, (uintptr_t)&mask_string8);
+                                VLDR64_U12(q1, x2, 0);     // load shift
+                                VDUPQ_64(q1, q1, 0);
+                                USHLQ_8(q0, q0, q1); // extract 1 bit
+                                MOVIQ_8(q1, 0x80);   // load mask
+                                VANDQ(q0, q0, q1);
+                                VSSHRQ_8(q0, q0, 7);    // saturate the mask
+                                break;
+                            case 0b01:
+                                VDUPQH(q0, x1); // load the 8bits of the mask
+                                TABLE64(x2, (uintptr_t)&mask_string16);
+                                VLDR64_U12(q1, x2, 0);     // load shift
+                                UXTL_8(q1, q1);     // extend mask to 16bits
+                                USHLQ_16(q0, q0, q1); // extract 1 bit
+                                MOVIQ_16(q1, 0x80, 1);   // load mask
+                                VANDQ(q0, q0, q1);
+                                VSSHRQ_16(q0, q0, 15);    // saturate the mask
+                        }
+                    } else {
+                        VEORQ(q0, q0, q0);
+                        VMOVQHfrom(q0, 0, x1);
+                    }
+                    break;
+                case 0x61:
+                    INST_NAME("PCMPESTRI Gx, Ex, Ib");
+                    SETFLAGS(X_OF|X_CF|X_AF|X_ZF|X_SF|X_PF, SF_SET);
+                    nextop = F8;
+                    GETG;
+                    sse_reflect_reg(dyn, ninst, gd);
+                    ADDx_U12(x3, xEmu, offsetof(x64emu_t, xmm[gd]));
+                    if(MODREG) {
+                        ed = (nextop&7)+(rex.b<<3);
+                        sse_reflect_reg(dyn, ninst, ed);
+                        ADDx_U12(x1, xEmu, offsetof(x64emu_t, xmm[ed]));
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 1);
+                        if(ed!=x1) {
+                            MOVx_REG(x1, ed);
+                        }
+                    }
+                    MOVx_REG(x2, xRDX);
+                    MOVx_REG(x4, xRAX);
+                    u8 = F8;
+                    MOV32w(x5, u8);
+                    CALL(sse42_compare_string_explicit_len, x1);
+                    CBNZw_MARK(x1);
+                    MOV32w(xRCX, (u8&1)?8:16);
+                    B_NEXT_nocond;
+                    MARK;
+                    if(u8&0b1000000) {
+                        CLZw(xRCX, x1);
+                        MOV32w(x2, 31);
+                        SUBw_REG(xRCX, x2, xRCX);
+                    } else {
+                        RBITxw(xRCX, x1);
+                        CLZw(xRCX, xRCX);
+                    }
+                    break;
+                case 0x62:
+                    INST_NAME("PCMPISTRM Gx, Ex, Ib");
+                    SETFLAGS(X_OF|X_CF|X_AF|X_ZF|X_SF|X_PF, SF_SET);
+                    nextop = F8;
+                    GETG;
+                    sse_forget_reg(dyn, ninst, gd);
+                    ADDx_U12(x2, xEmu, offsetof(x64emu_t, xmm[gd]));
+                    if(MODREG) {
+                        ed = (nextop&7)+(rex.b<<3);
+                        sse_reflect_reg(dyn, ninst, ed);
+                        ADDx_U12(x1, xEmu, offsetof(x64emu_t, xmm[ed]));
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 1);
+                        if(ed!=x1) {
+                            MOVx_REG(x1, ed);
+                        }
+                    }
+                    u8 = F8;
+                    MOV32w(x3, u8);
+                    CALL(sse42_compare_string_implicit_len, x1);
+                    q0 = sse_get_reg_empty(dyn, ninst, x2, gd);
+                    q1 = fpu_get_scratch(dyn);
+                    if(u8&0b1000000) {
+                        switch(u8&1) {
+                            case 0b00:
+                                VDUPQB(q0, x1); // load the low 8bits of the mask
+                                LSRw_IMM(x1, x1, 8);
+                                VDUPQB(q1, x1); // load the high 8bits of the mask
+                                VEXTQ_8(q0, q0, q1, 8); // low and hig bits mask
+                                TABLE64(x2, (uintptr_t)&mask_string8);
+                                VLDR64_U12(q1, x2, 0);     // load shift
+                                VDUPQ_64(q1, q1, 0);
+                                USHLQ_8(q0, q0, q1); // extract 1 bit
+                                MOVIQ_8(q1, 0x80);   // load mask
+                                VANDQ(q0, q0, q1);
+                                VSSHRQ_8(q0, q0, 7);    // saturate the mask
+                                break;
+                            case 0b01:
+                                VDUPQH(q0, x1); // load the 8bits of the mask
+                                TABLE64(x2, (uintptr_t)&mask_string16);
+                                VLDR64_U12(q1, x2, 0);     // load shift
+                                UXTL_8(q1, q1);     // extend mask to 16bits
+                                USHLQ_16(q0, q0, q1); // extract 1 bit
+                                MOVIQ_16(q1, 0x80, 1);   // load mask
+                                VANDQ(q0, q0, q1);
+                                VSSHRQ_16(q0, q0, 15);    // saturate the mask
+                        }
+                    } else {
+                        VEORQ(q0, q0, q0);
+                        VMOVQHfrom(q0, 0, x1);
+                    }
+                    break;
+                case 0x63:
+                    INST_NAME("PCMPISTRI Gx, Ex, Ib");
+                    SETFLAGS(X_OF|X_CF|X_AF|X_ZF|X_SF|X_PF, SF_SET);
+                    nextop = F8;
+                    GETG;
+                    sse_reflect_reg(dyn, ninst, gd);
+                    ADDx_U12(x2, xEmu, offsetof(x64emu_t, xmm[gd]));
+                    if(MODREG) {
+                        ed = (nextop&7)+(rex.b<<3);
+                        sse_reflect_reg(dyn, ninst, ed);
+                        ADDx_U12(x1, xEmu, offsetof(x64emu_t, xmm[ed]));
+                    } else {
+                        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 1);
+                        if(ed!=x1) {
+                            MOVx_REG(x1, ed);
+                        }
+                    }
+                    u8 = F8;
+                    MOV32w(x3, u8);
+                    CALL(sse42_compare_string_implicit_len, x1);
+                    CBNZw_MARK(x1);
+                    MOV32w(xRCX, (u8&1)?8:16);
+                    B_NEXT_nocond;
+                    MARK;
+                    if(u8&0b1000000) {
+                        CLZw(xRCX, x1);
+                        MOV32w(x2, 31);
+                        SUBw_REG(xRCX, x2, xRCX);
+                    } else {
+                        RBITxw(xRCX, x1);
+                        CLZw(xRCX, xRCX);
+                    }
+                    break;
+
                 case 0xDF:
                     INST_NAME("AESKEYGENASSIST Gx, Ex, Ib");  // AES-NI
                     nextop = F8;
@@ -1045,7 +1397,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     sse_forget_reg(dyn, ninst, gd);
                     MOV32w(x1, gd); // gx
                     if(MODREG) {
-                        ed = (nextop&7)+(rex.b<<3); 
+                        ed = (nextop&7)+(rex.b<<3);
                         sse_forget_reg(dyn, ninst, ed);
                         MOV32w(x2, ed);
                         MOV32w(x3, 0);  //p = NULL
@@ -1095,7 +1447,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             LSRx(x1, x1, 63);
             LSRx(gd, gd, 63);
             BFIx(gd, x1, 1, 1);
-            break;    
+            break;
         case 0x51:
             INST_NAME("SQRTPD Gx, Ex");
             nextop = F8;
@@ -1115,7 +1467,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 VFSQRTQD(q1, q0);
             }
             break;
-     
+
         case 0x54:
             INST_NAME("ANDPD Gx, Ex");
             nextop = F8;
@@ -1140,9 +1492,16 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
         case 0x57:
             INST_NAME("XORPD Gx, Ex");
             nextop = F8;
-            GETEX(q0, 0, 0);
-            GETGX(v0, 1);
-            VEORQ(v0, v0, q0);
+            GETG;
+            if(MODREG && ((nextop&7)+(rex.b<<3)==gd)) {
+                // special case for XORPD Gx, Gx
+                q0 = sse_get_reg_empty(dyn, ninst, x1, gd);
+                VEORQ(q0, q0, q0);
+            } else {
+                q0 = sse_get_reg(dyn, ninst, x1, gd, 1);
+                GETEX(q1, 0, 0);
+                VEORQ(q0, q0, q1);
+            }
             break;
         case 0x58:
             INST_NAME("ADDPD Gx, Ex");
@@ -1470,32 +1829,25 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     }
                 } else if(u8==0x00) {
                     // duplicate lower 32bits to all spot
-                    if(v0!=v1) {
-                        VMOVeS(v0, 0, v1, 0);
-                    }
-                    VMOVeS(v0, 1, v1, 0);
-                    VMOVeD(v0, 1, v0, 0);
+                    VDUPQ_32(v0, v1, 0);
                 } else if(u8==0x55) {
                     // duplicate slot 1 to all spot
-                    if(v0!=v1) {
-                        VMOVeS(v0, 1, v1, 1);
-                    }
-                    VMOVeS(v0, 0, v1, 1);
-                    VMOVeD(v0, 1, v0, 0);
+                    VDUPQ_32(v0, v1, 1);
                 } else if(u8==0xAA) {
                     // duplicate slot 2 to all spot
-                    if(v0!=v1) {
-                        VMOVeS(v0, 2, v1, 2);
-                    }
-                    VMOVeS(v0, 3, v1, 2);
-                    VMOVeD(v0, 0, v0, 1);
+                    VDUPQ_32(v0, v1, 2);
                 } else if(u8==0xFF) {
                     // duplicate slot 3 to all spot
-                    if(v0!=v1) {
-                        VMOVeS(v0, 3, v1, 3);
-                    }
-                    VMOVeS(v0, 2, v1, 3);
-                    VMOVeD(v0, 0, v0, 1);
+                    VDUPQ_32(v0, v1, 3);
+                } else if(u8==0x44) {
+                    // duplicate slot 0/1 to all spot
+                    VDUPQ_64(v0, v1, 0);
+                } else if(u8==0xEE) {
+                    // duplicate slot 2/3 to all spot
+                    VDUPQ_64(v0, v1, 1);
+                } else if(u8==0xB1) {
+                    // invert 0/1 and 2/3
+                    VREV64Q_32(v0, v1);
                 } else if(v0!=v1) {
                     VMOVeS(v0, 0, v1, (u8>>(0*2))&3);
                     VMOVeS(v0, 1, v1, (u8>>(1*2))&3);
@@ -1736,7 +2088,16 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 VORRQ(q1, q1, v1);      // NAN -> -NAN
             }
             break;
-
+        case 0x7D:
+            INST_NAME("HSUBPD Gx, Ex");  // SSE4 opcode!
+            nextop = F8;
+            GETEX(q1, 0, 0);
+            GETGX(q0, 1);
+            v0 = fpu_get_scratch(dyn);
+            VUZP1Q_64(v0, q0, q1);
+            VUZP2Q_64(q0, q0, q1);
+            VFSUBQD(q0, v0, q0);
+            break;
         case 0x7E:
             INST_NAME("MOVD Ed,Gx");
             nextop = F8;
@@ -1796,24 +2157,31 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             BFIw(xFlags, x1, F_CF, 1);
             break;
         case 0xA4:
+            INST_NAME("SHLD Ew, Gw, Ib");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETEW(x1, 1);
+            GETGW(x2);
+            u8 = F8;
+            emit_shld16c(dyn, ninst, ed, gd, u8, x4, x5);
+            EWBACK;
+            break;
         case 0xA5:
             nextop = F8;
-            if(opcode==0xA4) {
-                INST_NAME("SHLD Ew, Gw, Ib");
-            } else {
-                INST_NAME("SHLD Ew, Gw, CL");
-                UXTBw(x3, xRCX);
-            }
-            MESSAGE(LOG_DUMP, "Need Optimization\n");
-            SETFLAGS(X_ALL, SF_SET);
-            GETEWW(x4, x1, (opcode==0xA4)?1:0);
+            INST_NAME("SHLD Ew, Gw, CL");
+            SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+            if(box64_dynarec_safeflags>1)
+                MAYSETFLAGS();
             GETGW(x2);
-            if(opcode==0xA4) {
-                u8 = F8;
-                MOV32w(x3, u8);
+            GETEW(x1, 0);
+            UFLAG_IF {
+                ANDSw_mask(x4, xRCX, 0, 0b00100);  //mask=0x00000001f
+                B_NEXT(cEQ);
+            } else {
+                ANDw_mask(x4, xRCX, 0, 0b00100);  //mask=0x00000001f
             }
-            CALL_(shld16, x1, wback);
-            EWBACKW(x1);
+            emit_shld16(dyn, ninst, ed, gd, x4, x5, x6);
+            EWBACK;
             break;
 
         case 0xAB:
@@ -1834,48 +2202,55 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 ed = x4;
             }
             ANDw_mask(x2, gd, 0, 0b000011);  // mask=0x0f
-            LSRw_REG(x1, ed, x2);
-            BFIw(xFlags, x1, F_CF, 1);
-            ANDSw_mask(x1, x1, 0, 0);  //mask=1
-            B_NEXT(cNE);
+            IFX(X_CF) {
+                LSRw_REG(x1, ed, x2);
+                BFIw(xFlags, x1, F_CF, 1);
+            }
             MOV32w(x1, 1);
-            LSLxw_REG(x1, x1, x2);
+            LSLw_REG(x1, x1, x2);
             ORRx_REG(ed, ed, x1);
             if(wback) {
                 STRH_U12(ed, wback, fixedaddress);
             }
             break;
         case 0xAC:
+            nextop = F8;
+            INST_NAME("SHRD Ew, Gw, Ib");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            GETEW(x1, 1);
+            GETGW(x2);
+            u8 = F8;
+            emit_shrd16c(dyn, ninst, ed, gd, u8, x4, x5);
+            EWBACK;
+            break;
         case 0xAD:
             nextop = F8;
-            if(opcode==0xAC) {
-                INST_NAME("SHRD Ew, Gw, Ib");
-            } else {
-                INST_NAME("SHRD Ew, Gw, CL");
-                UXTBw(x3, xRCX);
-            }
-            MESSAGE(LOG_DUMP, "Need Optimization\n");
-            SETFLAGS(X_ALL, SF_SET);
-            GETEWW(x4, x1, (opcode==0xAC)?1:0);
+            INST_NAME("SHRD Ew, Gw, CL");
+            SETFLAGS(X_ALL, SF_SET_PENDING);    // some flags are left undefined
+            if(box64_dynarec_safeflags>1)
+                MAYSETFLAGS();
             GETGW(x2);
-            if(opcode==0xAC) {
-                u8 = F8;
-                MOV32w(x3, u8);
+            GETEW(x1, 0);
+            UFLAG_IF {
+                ANDSw_mask(x4, xRCX, 0, 0b00100);  //mask=0x00000001f
+                B_NEXT(cEQ);
+            } else {
+                ANDw_mask(x4, xRCX, 0, 0b00100);  //mask=0x00000001f
             }
-            CALL_(shrd16, x1, wback);
-            EWBACKW(x1);
+            emit_shrd16(dyn, ninst, ed, gd, x4, x5, x6);
+            EWBACK;
             break;
 
         case 0xAF:
             INST_NAME("IMUL Gw,Ew");
             SETFLAGS(X_ALL, SF_PENDING);
             nextop = F8;
-            UFLAG_DF(x1, d_imul16);
             GETSEW(x1, 0);
             GETSGW(x2);
             MULw(x2, x2, x1);
             UFLAG_RES(x2);
             GWBACK;
+            UFLAG_DF(x1, d_imul16);
             break;
 
         case 0xB3:
@@ -1899,8 +2274,6 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             ANDw_mask(x2, gd, 0, 0b000011);  // mask=0x0f
             LSRw_REG(x1, ed, x2);
             BFIw(xFlags, x1, F_CF, 1);
-            ANDSw_mask(x1, x1, 0, 0);  //mask=1
-            B_NEXT(cEQ);
             MOV32w(x1, 1);
             LSLxw_REG(x1, x1, x2);
             BICx_REG(ed, ed, x1);
@@ -1916,7 +2289,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             if(MODREG) {
                 if(rex.rex) {
                     eb1 = xRAX+(nextop&7)+(rex.b<<3);
-                    eb2 = 0;                \
+                    eb2 = 0;
                 } else {
                     ed = (nextop&7);
                     eb1 = xRAX+(ed&3);  // Ax, Cx, Dx or Bx
@@ -1967,12 +2340,12 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     GETEW(x1, 1);
                     u8 = F8;
                     u8&=(rex.w?0x3f:0x0f);
-                    BFXILxw(xFlags, ed, u8, 1);  // inject 1 bit from u8 to F_CF (i.e. pos 0)
-                    TBNZ_MARK3(xFlags, 0); // bit already set, jump to next instruction
+                    IFX(X_CF) {
+                        BFXILxw(xFlags, ed, u8, 1);  // inject 1 bit from u8 to F_CF (i.e. pos 0)
+                    }
                     MOV32w(x4, 1);
-                    ORRxw_REG_LSL(ed, ed, x4, u8);
+                    BFIxw(ed, x4, u8, 1);
                     EWBACK(x1);
-                    MARK3;
                     break;
                 case 6:
                     INST_NAME("BTR Ew, Ib");
@@ -1981,12 +2354,11 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     GETEW(x1, 1);
                     u8 = F8;
                     u8&=(rex.w?0x3f:0x0f);
-                    BFXILxw(xFlags, ed, u8, 1);  // inject 1 bit from u8 to F_CF (i.e. pos 0)
-                    TBZ_MARK3(xFlags, 0); // bit already clear, jump to next instruction
-                    MOV32w(x4, 1);
-                    BICxw_REG_LSL(ed, ed, x4, u8);
+                    IFX(X_CF) {
+                        BFXILxw(xFlags, ed, u8, 1);  // inject 1 bit from u8 to F_CF (i.e. pos 0)
+                    }
+                    BFCxw(ed, u8, 1);
                     EWBACK(x1);
-                    MARK3;
                     break;
                 case 7:
                     INST_NAME("BTC Ew, Ib");
@@ -1995,7 +2367,9 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     GETEW(x1, 1);
                     u8 = F8;
                     u8&=(rex.w?0x3f:0x0f);
-                    BFXILxw(xFlags, ed, u8, 1);  // inject 1 bit from u8 to F_CF (i.e. pos 0)
+                    IFX(X_CF) {
+                        BFXILxw(xFlags, ed, u8, 1);  // inject 1 bit from u8 to F_CF (i.e. pos 0)
+                    }
                     MOV32w(x4, 1);
                     EORxw_REG_LSL(ed, ed, x4, u8);
                     EWBACK(x1);
@@ -2023,9 +2397,10 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 ed = x4;
             }
             ANDw_mask(x2, gd, 0, 0b000011);  // mask=0x0f
-            LSRw_REG(x1, ed, x2);
-            BFIw(xFlags, x1, F_CF, 1);
-            ANDw_mask(x1, x1, 0, 0);  //mask=1
+            IFX(X_CF) {
+                LSRw_REG(x1, ed, x2);
+                BFIw(xFlags, x1, F_CF, 1);
+            }
             MOV32w(x1, 1);
             LSLxw_REG(x1, x1, x2);
             EORx_REG(ed, ed, x1);
@@ -2091,6 +2466,16 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             BFIx(gd, x1, 0, 16);
             break;
 
+        case 0xC1:
+            INST_NAME("XADD Gw, Ew");
+            SETFLAGS(X_ALL, SF_SET_PENDING);
+            nextop = F8;
+            GETGW(x1);
+            GETEW(x2, 0);
+            BFIx(xRAX+((nextop&0x38)>>3)+(rex.r<<3), ed, 0, 16);
+            emit_add16(dyn, ninst, ed, gd, x4, x5);
+            EWBACK;
+            break;
         case 0xC2:
             INST_NAME("CMPPD Gx, Ex, Ib");
             nextop = F8;
@@ -2102,21 +2487,21 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 case 0: VFCMEQQD(v0, v0, v1); break;   // Equal
                 case 1: VFCMGTQD(v0, v1, v0); break;   // Less than
                 case 2: VFCMGEQD(v0, v1, v0); break;   // Less or equal
-                case 3: VFCMEQQD(v0, v0, v0); 
+                case 3: VFCMEQQD(v0, v0, v0);
                         if(v0!=v1) {
-                            q0 = fpu_get_scratch(dyn); 
-                            VFCMEQQD(q0, v1, v1); 
+                            q0 = fpu_get_scratch(dyn);
+                            VFCMEQQD(q0, v1, v1);
                             VANDQ(v0, v0, q0);
                         }
-                        VMVNQ(v0, v0); 
+                        VMVNQ(v0, v0);
                         break;   // NaN (NaN is not equal to himself)
                 case 4: VFCMEQQD(v0, v0, v1); VMVNQ(v0, v0); break;   // Not Equal (or unordered on ARM, not on X86...)
                 case 5: VFCMGTQD(v0, v1, v0); VMVNQ(v0, v0); break;   // Greater or equal or unordered
                 case 6: VFCMGEQD(v0, v1, v0); VMVNQ(v0, v0); break;   // Greater or unordered
-                case 7: VFCMEQQD(v0, v0, v0); 
+                case 7: VFCMEQQD(v0, v0, v0);
                         if(v0!=v1) {
-                            q0 = fpu_get_scratch(dyn); 
-                            VFCMEQQD(q0, v1, v1); 
+                            q0 = fpu_get_scratch(dyn);
+                            VFCMEQQD(q0, v1, v1);
                             VANDQ(v0, v0, q0);
                         }
                         break;   // not NaN
@@ -2206,6 +2591,16 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             VDUPQ_16(v0, v0, 0);    // only the low 8bits will be used anyway
             USHLQ_16(q0, q0, v0);   // SHR x8
             break;
+        case 0xD0:
+            INST_NAME("ADDSUBPD Gx,Ex");
+            nextop = F8;
+            GETGX(q0, 1);
+            GETEX(q1, 0, 0);
+            v0 = fpu_get_scratch(dyn);
+            VFSUBQD(v0, q0, q1);
+            VFADDQD(q0, q0, q1);
+            VMOVeD(q0, 0, v0, 0);
+            break;
         case 0xD2:
             INST_NAME("PSRLD Gx,Ex");
             nextop = F8;
@@ -2283,7 +2678,7 @@ uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             UADDLV_8(q1, q1);   // accumalte
             VMOVBto(x1, q1, 0);
             BFIx(gd, x1, 8, 8);
-            break;        
+            break;
         case 0xD8:
             INST_NAME("PSUBUSB Gx, Ex");
             nextop = F8;

src/dynarec/arm64/dynarec_arm64_66f0.c

@@ -53,7 +53,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGW(x5);
-            SMDMB();
             if(MODREG) {
                 ed = xRAX+(nextop&7)+(rex.b<<3);
                 UXTHw(x6, ed);
@@ -61,13 +60,22 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 BFIx(ed, x6, 0, 16);
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                MARKLOCK;
-                LDAXRH(x1, wback);
-                emit_or16(dyn, ninst, x1, x5, x3, x4);
-                STLXRH(x3, x1, wback);
-                CBNZx_MARKLOCK(x3);
+                if(arm64_atomics) {
+                    UFLAG_IF {
+                        LDSETALH(x5, x1, wback);
+                        emit_or16(dyn, ninst, x1, x5, x3, x4);
+                    } else {
+                        STSETLH(x5, wback);
+                    }
+                } else {
+                    MARKLOCK;
+                    LDAXRH(x1, wback);
+                    emit_or16(dyn, ninst, x1, x5, x3, x4);
+                    STLXRH(x3, x1, wback);
+                    CBNZx_MARKLOCK(x3);
+                }
+                SMDMB();
             }
-            SMDMB();
             break;
 
         case 0x0F:
@@ -79,7 +87,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     SETFLAGS(X_ALL, SF_SET_PENDING);
                     nextop = F8;
                     GETGD;
-                    SMDMB();
                     UXTHw(x6, xRAX);
                     if(MODREG) {
                         ed = xRAX+(nextop&7)+(rex.b<<3);
@@ -90,34 +97,43 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                         BFIx(ed, gd, 0, 16);
                     } else {
                         addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                        TSTx_mask(wback, 1, 0, 0);    // mask=1
-                        B_MARK3(cNE);
+                        if(!ALIGNED_ATOMICH) {
+                            TSTx_mask(wback, 1, 0, 0);    // mask=1
+                            B_MARK3(cNE);
+                        }
                         // Aligned version
-                        MARKLOCK;
-                        LDAXRH(x1, wback);
-                        CMPSw_REG(x6, x1);
-                        B_MARK(cNE);
-                        // EAX == Ed
-                        STLXRH(x4, gd, wback);
-                        CBNZx_MARKLOCK(x4);
-                        // done
-                        B_MARK_nocond;
-                        // Unaligned version
-                        MARK3;
-                        LDRH_U12(x1, wback, 0);
-                        LDAXRB(x3, wback); // dummy read, to arm the write...
-                        CMPSw_REG(x6, x1);
-                        B_MARK(cNE);
-                        // EAX == Ed
-                        STLXRB(x4, gd, wback);
-                        CBNZx_MARK3(x4);
-                        STRH_U12(gd, wback, 0);
+                        if(arm64_atomics) {
+                            MOVw_REG(x1, x6);
+                            CASALH(x1, gd, wback);
+                        } else {
+                            MARKLOCK;
+                            LDAXRH(x1, wback);
+                            CMPSw_REG(x6, x1);
+                            B_MARK(cNE);
+                            // EAX == Ed
+                            STLXRH(x4, gd, wback);
+                            CBNZx_MARKLOCK(x4);
+                            // done
+                        }
+                        if(!ALIGNED_ATOMICH) {
+                            B_MARK_nocond;
+                            // Unaligned version
+                            MARK3;
+                            LDRH_U12(x1, wback, 0);
+                            LDAXRB(x3, wback); // dummy read, to arm the write...
+                            CMPSw_REG(x6, x1);
+                            B_MARK(cNE);
+                            // EAX == Ed
+                            STLXRB(x4, gd, wback);
+                            CBNZx_MARK3(x4);
+                            STRH_U12(gd, wback, 0);
+                        }
                     }
+                    SMDMB();
                     MARK;
                     // Common part (and fallback for EAX != Ed)
                     UFLAG_IF {emit_cmp16(dyn, ninst, x6, x1, x3, x4, x5);}
                     BFIx(xRAX, x1, 0, 16);
-                    SMDMB();
                     break;
 
                 case 0xC1:
@@ -126,7 +142,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     nextop = F8;
                     gd = xRAX+((nextop&0x38)>>3)+(rex.r<<3);
                     UXTHx(x5, gd);
-                    SMDMB();
                     if(MODREG) {
                         ed = xRAX+(nextop&7)+(rex.b<<3);
                         BFIx(gd, ed, 0, 16);
@@ -134,18 +149,22 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                         BFIx(ed, x5, 0, 16);
                     } else {
                         addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                        MARKLOCK;
-                        LDAXRH(x1, wback);
-                        ADDxw_REG(x4, x1, x5);
-                        STLXRH(x3, x4, wback);
-                        CBNZx_MARKLOCK(x3);
+                        if(arm64_atomics) {
+                            LDADDALH(x5, x1, wback);
+                        } else {
+                            MARKLOCK;
+                            LDAXRH(x1, wback);
+                            ADDxw_REG(x4, x1, x5);
+                            STLXRH(x3, x4, wback);
+                            CBNZx_MARKLOCK(x3);
+                        }
+                        SMDMB();
                         IFX(X_ALL|X_PEND) {
                             MOVxw_REG(x2, x1);
                             emit_add16(dyn, ninst, x2, x5, x3, x4);
                         }
                         BFIx(gd, x1, 0, 16);
                     }
-                    SMDMB();
                     break;
 
                 default:
@@ -158,7 +177,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             SETFLAGS(X_ALL, SF_SET_PENDING);
             nextop = F8;
             GETGW(x5);
-            SMDMB();
             if(MODREG) {
                 ed = xRAX+(nextop&7)+(rex.b<<3);
                 UXTHw(x6, ed);
@@ -171,13 +189,12 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                 emit_adc16(dyn, ninst, x1, x5, x3, x4);
                 STLXRH(x3, x1, wback);
                 CBNZx_MARKLOCK(x3);
+                SMDMB();
             }
-            SMDMB();
             break;
         case 0x81:
         case 0x83:
             nextop = F8;
-            SMDMB();
             switch((nextop>>3)&7) {
                 case 0: //ADD
                     if(opcode==0x81) {
@@ -197,22 +214,35 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                         addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                         if(opcode==0x81) i32 = F16S; else i32 = F8S;
                         MOV32w(x5, i32);
-                        TSTx_mask(wback, 1, 0, 0);    // mask=1
-                        B_MARK(cNE);
-                        MARKLOCK;
-                        LDAXRH(x1, wback);
-                        emit_add16(dyn, ninst, x1, x5, x3, x4);
-                        STLXRH(x3, x1, wback);
-                        CBNZx_MARKLOCK(x3);
-                        B_NEXT_nocond;
-                        MARK;   // unaligned! also, not enough
-                        LDRH_U12(x1, wback, 0);
-                        LDAXRB(x4, wback);
-                        BFIw(x1, x4, 0, 8); // re-inject
-                        emit_add16(dyn, ninst, x1, x5, x3, x4);
-                        STLXRB(x3, x1, wback);
-                        CBNZx_MARK(x3);
-                        STRH_U12(x1, wback, 0);    // put the whole value
+                        if(!ALIGNED_ATOMICH) {
+                            TSTx_mask(wback, 1, 0, 0);    // mask=1
+                            B_MARK(cNE);
+                        }
+                        if(arm64_atomics) {
+                            LDADDALH(x5, x1, wback);
+                            UFLAG_IF {
+                                emit_add16(dyn, ninst, x1, x5, x3, x4);    
+                            }
+                        } else {
+                            MARKLOCK;
+                            LDAXRH(x1, wback);
+                            emit_add16(dyn, ninst, x1, x5, x3, x4);
+                            STLXRH(x3, x1, wback);
+                            CBNZx_MARKLOCK(x3);
+                        }
+                        SMDMB();
+                        if(!ALIGNED_ATOMICH) {
+                            B_NEXT_nocond;
+                            MARK;   // unaligned! also, not enough
+                            LDRH_U12(x1, wback, 0);
+                            LDAXRB(x4, wback);
+                            BFIw(x1, x4, 0, 8); // re-inject
+                            emit_add16(dyn, ninst, x1, x5, x3, x4);
+                            STLXRB(x3, x1, wback);
+                            CBNZx_MARK(x3);
+                            STRH_U12(x1, wback, 0);    // put the whole value
+                            SMDMB();
+                        }
                     }
                     break;
                 case 1: //OR
@@ -229,11 +259,21 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                         addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                         if(opcode==0x81) i32 = F16S; else i32 = F8S;
                         MOV32w(x5, i32);
-                        MARKLOCK;
-                        LDAXRH(x1, wback);
-                        emit_or16(dyn, ninst, x1, x5, x3, x4);
-                        STLXRH(x3, x1, wback);
-                        CBNZx_MARKLOCK(x3);
+                        if(arm64_atomics) {
+                            UFLAG_IF {
+                                LDSETALH(x5, x1, wback);
+                                emit_or16(dyn, ninst, x1, x5, x3, x4);
+                            } else {
+                                STSETLH(x5, wback);
+                            }
+                        } else {
+                            MARKLOCK;
+                            LDAXRH(x1, wback);
+                            emit_or16(dyn, ninst, x1, x5, x3, x4);
+                            STLXRH(x3, x1, wback);
+                            CBNZx_MARKLOCK(x3);
+                        }
+                        SMDMB();
                     }
                     break;
                 case 2: //ADC
@@ -256,6 +296,7 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                         emit_adc16(dyn, ninst, x1, x5, x3, x4);
                         STLXRH(x3, x1, wback);
                         CBNZx_MARKLOCK(x3);
+                        SMDMB();
                     }
                     break;
                 case 3: //SBB
@@ -278,6 +319,7 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                         emit_sbb16(dyn, ninst, x1, x5, x3, x4);
                         STLXRH(x3, x1, wback);
                         CBNZx_MARKLOCK(x3);
+                        SMDMB();
                     }
                     break;
                 case 4: //AND
@@ -293,12 +335,24 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     } else {
                         addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                         if(opcode==0x81) i32 = F16S; else i32 = F8S;
-                        MOV32w(x5, i32);
-                        MARKLOCK;
-                        LDAXRH(x1, wback);
-                        emit_and16(dyn, ninst, x1, x5, x3, x4);
-                        STLXRH(x3, x1, wback);
-                        CBNZx_MARKLOCK(x3);
+                        if(arm64_atomics) {
+                            MOV32w(x5, ~i32);
+                            UFLAG_IF {
+                                LDCLRALH(x5, x1, wback);
+                                MVNw_REG(x5, x5);
+                                emit_and16(dyn, ninst, x1, x5, x3, x4);
+                            } else {
+                                STCLRLH(x5, wback);
+                            }
+                        } else {
+                            MOV32w(x5, i32);
+                            MARKLOCK;
+                            LDAXRH(x1, wback);
+                            emit_and16(dyn, ninst, x1, x5, x3, x4);
+                            STLXRH(x3, x1, wback);
+                            CBNZx_MARKLOCK(x3);
+                            SMDMB();
+                        }
                     }
                     break;
                 case 5: //SUB
@@ -315,22 +369,38 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                         addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                         if(opcode==0x81) i32 = F16S; else i32 = F8S;
                         MOV32w(x5, i32);
-                        TSTx_mask(wback, 1, 0, 0);    // mask=1
-                        B_MARK(cNE);
-                        MARKLOCK;
-                        LDAXRH(x1, wback);
-                        emit_sub16(dyn, ninst, x1, x5, x3, x4);
-                        STLXRH(x3, x1, wback);
-                        CBNZx_MARKLOCK(x3);
-                        B_NEXT_nocond;
-                        MARK;   // unaligned! also, not enough
-                        LDRH_U12(x1, wback, 0);
-                        LDAXRB(x4, wback);
-                        BFIw(x1, x4, 0, 8); // re-inject
-                        emit_sub16(dyn, ninst, x1, x5, x3, x4);
-                        STLXRB(x3, x1, wback);
-                        CBNZx_MARK(x3);
-                        STRH_U12(x1, wback, 0);    // put the whole value
+                        if(!ALIGNED_ATOMICH) {
+                            TSTx_mask(wback, 1, 0, 0);    // mask=1
+                            B_MARK(cNE);
+                        }
+                        if(arm64_atomics) {
+                            NEGw_REG(x4, x5);
+                            UFLAG_IF {
+                                LDADDALH(x4, x1, wback);
+                                emit_sub16(dyn, ninst, x1, x5, x3, x4);
+                            } else {
+                                STADDLH(x4, wback);
+                            }
+                        } else {
+                            MARKLOCK;
+                            LDAXRH(x1, wback);
+                            emit_sub16(dyn, ninst, x1, x5, x3, x4);
+                            STLXRH(x3, x1, wback);
+                            CBNZx_MARKLOCK(x3);
+                        }
+                        SMDMB();
+                        if(!ALIGNED_ATOMICH) {
+                            B_NEXT_nocond;
+                            MARK;   // unaligned! also, not enough
+                            LDRH_U12(x1, wback, 0);
+                            LDAXRB(x4, wback);
+                            BFIw(x1, x4, 0, 8); // re-inject
+                            emit_sub16(dyn, ninst, x1, x5, x3, x4);
+                            STLXRB(x3, x1, wback);
+                            CBNZx_MARK(x3);
+                            STRH_U12(x1, wback, 0);    // put the whole value
+                            SMDMB();
+                        }
                     }
                     break;
                 case 6: //XOR
@@ -347,11 +417,21 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                         addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, (opcode==0x81)?2:1);
                         if(opcode==0x81) i32 = F16S; else i32 = F8S;
                         MOV32w(x5, i32);
-                        MARKLOCK;
-                        LDAXRH(x1, wback);
-                        emit_xor16(dyn, ninst, x1, x5, x3, x4);
-                        STLXRH(x3, x1, wback);
-                        CBNZx_MARKLOCK(x3);
+                        if(arm64_atomics) {
+                            UFLAG_IF {
+                                LDEORALH(x5, x1, wback);
+                                emit_xor16(dyn, ninst, x1, x5, x3, x4);
+                            } else {
+                                STEORLH(x5, wback);
+                            }
+                        } else {
+                            MARKLOCK;
+                            LDAXRH(x1, wback);
+                            emit_xor16(dyn, ninst, x1, x5, x3, x4);
+                            STLXRH(x3, x1, wback);
+                            CBNZx_MARKLOCK(x3);
+                            SMDMB();
+                        }
                     }
                     break;
                 case 7: //CMP
@@ -369,7 +449,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     }
                     break;
             }
-            SMDMB();
             break;
 
             case 0xFF:
@@ -379,7 +458,6 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                     case 0: // INC Ew
                         INST_NAME("LOCK INC Ew");
                         SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
-                        SMDMB();
                         if(MODREG) {
                             ed = xRAX+(nextop&7)+(rex.b<<3);
                             UXTHw(x6, ed);
@@ -387,18 +465,27 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                             BFIx(ed, x6, 0, 16);
                         } else {
                             addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                            MARKLOCK;
-                            LDAXRH(x1, wback);
-                            emit_inc16(dyn, ninst, x1, x3, x4);
-                            STLXRH(x3, x1, wback);
-                            CBNZx_MARKLOCK(x3);
+                            if(arm64_atomics) {
+                                MOV32w(x3, 1);
+                                UFLAG_IF {
+                                    LDADDALH(x3, x1, wback);
+                                    emit_inc16(dyn, ninst, x1, x3, x4);
+                                } else {
+                                    STADDLH(x3, wback);
+                                }
+                            } else {
+                                MARKLOCK;
+                                LDAXRH(x1, wback);
+                                emit_inc16(dyn, ninst, x1, x3, x4);
+                                STLXRH(x3, x1, wback);
+                                CBNZx_MARKLOCK(x3);
+                            }
+                            SMDMB();
                         }
-                        SMDMB();
                         break;
                     case 1: //DEC Ew
                         INST_NAME("LOCK DEC Ew");
                         SETFLAGS(X_ALL&~X_CF, SF_SUBSET_PENDING);
-                        SMDMB();
                         if(MODREG) {
                             ed = xRAX+(nextop&7)+(rex.b<<3);
                             UXTHw(x6, ed);
@@ -406,13 +493,23 @@ uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
                             BFIx(ed, x6, 0, 16);
                         } else {
                             addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, NULL, 0, 0, rex, LOCK_LOCK, 0, 0);
-                            MARKLOCK;
-                            LDAXRH(x1, wback);
-                            emit_dec16(dyn, ninst, x1, x3, x4);
-                            STLXRH(x3, x1, wback);
-                            CBNZx_MARKLOCK(x3);
+                            if(arm64_atomics) {
+                                MOV32w(x3, -1);
+                                UFLAG_IF {
+                                    LDADDALH(x3, x1, wback);
+                                    emit_dec16(dyn, ninst, x1, x3, x4);
+                                } else {
+                                    STADDLH(x3, wback);
+                                }
+                            } else {
+                                MARKLOCK;
+                                LDAXRH(x1, wback);
+                                emit_dec16(dyn, ninst, x1, x3, x4);
+                                STLXRH(x3, x1, wback);
+                                CBNZx_MARKLOCK(x3);
+                            }
+                            SMDMB();
                         }
-                        SMDMB();
                         break;
                     default:
                         DEFAULT;

src/dynarec/arm64/dynarec_arm64_66f20f.c

@@ -0,0 +1,93 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.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 "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynarec_native.h"
+
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
+#include "dynarec_arm64_functions.h"
+#include "dynarec_arm64_helper.h"
+#include "emu/x64compstrings.h"
+
+uintptr_t dynarec64_66F20F(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;
+
+    uint8_t opcode = F8;
+    uint8_t nextop, u8;
+    int32_t i32;
+    uint8_t gd, ed;
+    uint8_t wback, wb1, wb2;
+    uint8_t eb1, eb2;
+    int64_t j64;
+    uint64_t tmp64u, tmp64u2;
+    int v0, v1;
+    int q0, q1;
+    int d0, d1;
+    int64_t fixedaddress;
+    int unscaled;
+
+    MAYUSE(d0);
+    MAYUSE(d1);
+    MAYUSE(q0);
+    MAYUSE(q1);
+    MAYUSE(eb1);
+    MAYUSE(eb2);
+    MAYUSE(j64);
+    #if STEP > 1
+    static const int8_t mask_shift8[] = { -7, -6, -5, -4, -3, -2, -1, 0 };
+    static const int8_t mask_string8[] = { 7, 6, 5, 4, 3, 2, 1, 0 };
+    static const int8_t mask_string16[] = { 15, 14, 13, 12, 11, 10, 9, 8 };
+    static const int8_t round_round[] = { 0, 2, 1, 3};
+    #endif
+
+    switch(opcode) {
+
+        case 0x38:  // SSE 4.x
+            opcode = F8;
+            switch(opcode) {
+
+                case 0xF1:
+                    INST_NAME("CRC32 Gd, Ew");
+                    nextop = F8;
+                    GETEW(x1, 0);
+                    GETGD;
+                    if(arm64_crc32) {
+                        CRC32CH(gd, gd, ed);
+                    } else {
+                        MOV32w(x2, 0x82f63b78);
+                        for(int j=0; j<2; ++j) {
+                            UBFXxw(x3, ed, 8*j, 8);
+                            EORw_REG(gd, gd, x3);
+                            for(int i=0; i<8; ++i) {
+                                LSRw_IMM((i&1)?gd:x4, (i&1)?x4:gd, 1);
+                                TBZ((i&1)?x4:gd, 0, 4+4);
+                                EORw_REG((i&1)?gd:x4, (i&1)?gd:x4, x2);
+                            }
+                        }
+                    }
+                    break;
+
+                default:
+                    DEFAULT;
+            }
+            break;
+
+        default:
+            DEFAULT;
+    }
+    return addr;
+}

src/dynarec/arm64/dynarec_arm64_66f30f.c

@@ -0,0 +1,64 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.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 "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynarec_native.h"
+
+#include "arm64_printer.h"
+#include "dynarec_arm64_private.h"
+#include "dynarec_arm64_functions.h"
+#include "dynarec_arm64_helper.h"
+#include "emu/x64compstrings.h"
+
+uintptr_t dynarec64_66F30F(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;
+
+    uint8_t opcode = F8;
+    uint8_t nextop, u8;
+    int32_t i32;
+    uint8_t gd, ed;
+    uint8_t wback, wb1, wb2;
+    uint8_t eb1, eb2;
+    int64_t j64;
+    uint64_t tmp64u, tmp64u2;
+    int v0, v1;
+    int q0, q1;
+    int d0, d1;
+    int64_t fixedaddress;
+    int unscaled;
+
+    MAYUSE(d0);
+    MAYUSE(d1);
+    MAYUSE(q0);
+    MAYUSE(q1);
+    MAYUSE(eb1);
+    MAYUSE(eb2);
+    MAYUSE(j64);
+    #if STEP > 1
+    static const int8_t mask_shift8[] = { -7, -6, -5, -4, -3, -2, -1, 0 };
+    static const int8_t mask_string8[] = { 7, 6, 5, 4, 3, 2, 1, 0 };
+    static const int8_t mask_string16[] = { 15, 14, 13, 12, 11, 10, 9, 8 };
+    static const int8_t round_round[] = { 0, 2, 1, 3};
+    #endif
+
+    switch(opcode) {
+
+
+        default:
+            DEFAULT;
+    }
+    return addr;
+}

src/dynarec/arm64/dynarec_arm64_67.c

@@ -276,7 +276,7 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     if(MODREG) {
                         if(rex.rex) {
                             eb1 = xRAX+(nextop&7)+(rex.b<<3);
-                            eb2 = 0;                \
+                            eb2 = 0;
                         } else {
                             ed = (nextop&7);
                             eb1 = xRAX+(ed&3);  // Ax, Cx, Dx or Bx
@@ -761,12 +761,12 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 gb2 = 0;
                 gb1 = xRAX + gd;
             } else {
-                gb2 = ((gd&4)>>2);
+                gb2 = ((gd&4)<<1);
                 gb1 = xRAX+(gd&3);
             }
             if(gb2) {
                 gd = x4;
-                UBFXw(gd, gb1, gb2*8, 8);
+                UBFXw(gd, gb1, gb2, 8);
             } else {
                 gd = gb1;   // no need to extract
             }
@@ -798,7 +798,40 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 SMWRITELOCK(lock);
             }
             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)<<1);
+            }
+            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 {
+                addr = geted32(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff, 0, rex, &lock, 0, 0);
+                SMREADLOCK(lock);
+                LDB(x4, wback, fixedaddress);
+                ed = x4;
+            }
+            BFIx(gb1, ed, gb2, 8);
+            break;
         case 0x8B:
             INST_NAME("MOV Gd, Ed");
             nextop=F8;
@@ -1008,7 +1041,6 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 case 4:
                     INST_NAME("MUL EAX, Ed");
                     SETFLAGS(X_ALL, SF_PENDING);
-                    UFLAG_DF(x2, rex.w?d_mul64:d_mul32);
                     GETED32(0);
                     if(rex.w) {
                         if(ed==xRDX) gd=x3; else gd=xRDX;
@@ -1022,11 +1054,11 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     }
                     UFLAG_RES(xRAX);
                     UFLAG_OP1(xRDX);
+                    UFLAG_DF(x2, rex.w?d_mul64:d_mul32);
                     break;
                 case 5:
                     INST_NAME("IMUL EAX, Ed");
                     SETFLAGS(X_ALL, SF_PENDING);
-                    UFLAG_DF(x2, rex.w?d_imul64:d_imul32);
                     GETED32(0);
                     if(rex.w) {
                         if(ed==xRDX) gd=x3; else gd=xRDX;
@@ -1040,6 +1072,7 @@ uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     }
                     UFLAG_RES(xRAX);
                     UFLAG_OP1(xRDX);
+                    UFLAG_DF(x2, rex.w?d_imul64:d_imul32);
                     break;
                 case 6:
                     INST_NAME("DIV Ed");

src/dynarec/arm64/dynarec_arm64_6764_32.c

@@ -0,0 +1,103 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.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 "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.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 GETGm   gd = ((nextop&0x38)>>3)
+
+uintptr_t dynarec64_6764_32(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int seg, int* ok, int* need_epilog)
+{
+    (void)ip; (void)need_epilog;
+
+    uint8_t opcode = F8;
+    uint8_t nextop;
+    uint8_t gd, ed, wback, wb, wb1, wb2, gb1, gb2, eb1, eb2;
+    int64_t fixedaddress;
+    int unscaled;
+    int8_t  i8;
+    uint8_t u8;
+    int32_t i32;
+    int64_t j64, i64;
+    int cacheupd = 0;
+    int lock;
+    int v0, v1, s0;
+    MAYUSE(i32);
+    MAYUSE(j64);
+    MAYUSE(v0);
+    MAYUSE(v1);
+    MAYUSE(s0);
+    MAYUSE(lock);
+    MAYUSE(cacheupd);
+
+    if(!rex.is32bits) {
+        // should do a different file
+        DEFAULT;
+        return addr;
+    }
+
+    rep = 0;
+    while((opcode==0xF2) || (opcode==0xF3)) {
+        rep = opcode-0xF1;
+        opcode = F8;
+    }
+
+    switch(opcode) {
+
+        case 0xA3:
+            INST_NAME("MOV Seg:[Od], EAX");
+            i32 = F16;
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            if(i32<4096 && !(i32&3)) {
+                STRw_U12(xRAX, x4, i32);
+            } else if(i32<256) {
+                STURw_I9(xRAX, x4, i32);
+            } else {
+                MOV32w(x1, i32);
+                STRw_REG(xRAX, x4, x1);
+            }
+            break;
+
+        case 0xFF:
+            nextop = F8;
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            switch((nextop>>3)&7) {
+                case 6: // Push Ed
+                    INST_NAME("PUSH FS:Ew");
+                    if((nextop&0xC0)==0xC0) {   // reg
+                        DEFAULT;
+                    } else {                    // mem <= i32
+                        SMREAD();
+                        addr = geted16(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0, 0, 0);
+                        LDRw_REG(x3, ed, x4);
+                        PUSH1_32(x3);
+                    }
+                    break;
+                default:
+                    DEFAULT;
+            }
+            break;
+
+
+        default:
+            DEFAULT;
+    }
+    return addr;
+}

src/dynarec/arm64/dynarec_arm64_67_32.c

@@ -0,0 +1,72 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.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 "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.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 GETGm   gd = ((nextop&0x38)>>3)
+
+uintptr_t dynarec64_67_32(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+{
+    (void)ip; (void)need_epilog;
+
+    uint8_t opcode = F8;
+    uint8_t nextop;
+    uint8_t gd, ed, wback, wb, wb1, wb2, gb1, gb2, eb1, eb2;
+    int64_t fixedaddress;
+    int unscaled;
+    int8_t  i8;
+    uint8_t u8;
+    int32_t i32;
+    int64_t j64, i64;
+    int cacheupd = 0;
+    int lock;
+    int v0, v1, s0;
+    MAYUSE(i32);
+    MAYUSE(j64);
+    MAYUSE(v0);
+    MAYUSE(v1);
+    MAYUSE(s0);
+    MAYUSE(lock);
+    MAYUSE(cacheupd);
+
+    if(!rex.is32bits) {
+        // should do a different file
+        DEFAULT;
+        return addr;
+    }
+
+    rep = 0;
+    while((opcode==0xF2) || (opcode==0xF3)) {
+        rep = opcode-0xF1;
+        opcode = F8;
+    }
+
+    switch(opcode) {
+        
+        case 64:
+            addr = dynarec64_6764_32(dyn, addr, ip, ninst, rex, rep, _FS, ok, need_epilog);
+            break;
+
+        default:
+            DEFAULT;
+    }
+    return addr;
+}

src/dynarec/arm64/dynarec_arm64_d8.c

@@ -108,7 +108,7 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 FCMPD(v1, v2);
             }
             FCOM(x1, x2, x3);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xE0:
         case 0xE1:
@@ -178,7 +178,7 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 FDIVD(v1, v2, v1);
             }
             break;
-      
+
         default:
             switch((nextop>>3)&7) {
                 case 0:
@@ -234,7 +234,7 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         FCMPD(v1, s0);
                     }
                     FCOM(x1, x2, x3);
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 case 4:
                     INST_NAME("FSUB ST0, float[ED]");

src/dynarec/arm64/dynarec_arm64_d9.c

@@ -51,7 +51,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0xC6:
         case 0xC7:
             INST_NAME("FLD STx");
-            v2 = x87_do_push(dyn, ninst, x1, X87_ST(nextop&7));
+            X87_PUSH_OR_FAIL(v2, 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);
@@ -84,7 +84,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
 
         case 0xD8:
             INST_NAME("FSTPNCE ST0, ST0");
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xD9:
         case 0xDA:
@@ -98,7 +98,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             x87_get_st_empty(dyn, ninst, x1, x2, nextop&7, X87_ST(nextop&7));
             x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
             x87_swapreg(dyn, ninst, x1, x2, 0, nextop&7);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xE0:
             INST_NAME("FCHS");
@@ -138,7 +138,7 @@ 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, x1, NEON_CACHE_ST_F);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_F);
             if(ST_IS_F(0)) {
                 FMOVS_8(v1, 0b01110000);
             } else {
@@ -147,82 +147,123 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             break;
         case 0xE9:
             INST_NAME("FLDL2T");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
             FTABLE64(v1, L2T);
             break;
-        case 0xEA:     
+        case 0xEA:
             INST_NAME("FLDL2E");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
             FTABLE64(v1, L2E);
             break;
         case 0xEB:
             INST_NAME("FLDPI");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
             FTABLE64(v1, PI);
             break;
         case 0xEC:
             INST_NAME("FLDLG2");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
             FTABLE64(v1, LG2);
             break;
         case 0xED:
             INST_NAME("FLDLN2");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
             FTABLE64(v1, LN2);
             break;
         case 0xEE:
             INST_NAME("FLDZ");
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_F);
             VEOR(v1, v1, v1);
             break;
 
         case 0xF0:
             INST_NAME("F2XM1");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
+            i1 = x87_stackcount(dyn, ninst, x1);
             x87_forget(dyn, ninst, x1, x2, 0);
             CALL(native_f2xm1, -1);
+            x87_unstackcount(dyn, ninst, x1, i1);
             break;
         case 0xF1:
             INST_NAME("FYL2X");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
+            i1 = x87_stackcount(dyn, ninst, x1);
             x87_forget(dyn, ninst, x1, x2, 0);
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(native_fyl2x, -1);
-            x87_do_pop(dyn, ninst, x3);
+            x87_unstackcount(dyn, ninst, x1, i1);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xF2:
             INST_NAME("FPTAN");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
+            i1 = x87_stackcount(dyn, ninst, x1);
             x87_forget(dyn, ninst, x1, x2, 0);
             CALL(native_ftan, -1);
-            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
-            if(ST_IS_F(0)) {
-                FMOVS_8(v1, 0b01110000);
+            x87_unstackcount(dyn, ninst, x1, i1);
+            if(PK(0)==0xdd && PK(1)==0xd8) {
+                MESSAGE(LOG_DUMP, "Optimized next DD D8 fstp st0, st0, not emiting 1\n");
+                u8 = F8;
+                u8 = F8;
             } else {
-                FMOVD_8(v1, 0b01110000);
+                X87_PUSH_OR_FAIL(v1, 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");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
+            i1 = x87_stackcount(dyn, ninst, x1);
             x87_forget(dyn, ninst, x1, x2, 0);
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(native_fpatan, -1);
-            x87_do_pop(dyn, ninst, x3);
+            x87_unstackcount(dyn, ninst, x1, i1);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xF4:
             INST_NAME("FXTRACT");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
-            x87_do_push_empty(dyn, ninst, 0);
+            X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, 0);
+            i1 = x87_stackcount(dyn, ninst, x1);
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(native_fxtract, -1);
+            x87_unstackcount(dyn, ninst, x1, i1);
             break;
         case 0xF5:
             INST_NAME("FPREM1");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
+            #if 0
+            i1 = x87_stackcount(dyn, ninst, x1);
             x87_forget(dyn, ninst, x1, x2, 0);
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(native_fprem1, -1);
+            x87_unstackcount(dyn, ninst, x1, i1);
+            #else
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            v2 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
+            s0 = fpu_get_scratch(dyn);
+            FDIVD(s0, v1, v2);
+            FRINTRRD(s0, s0, 0b00); // Nearest == TieToEven?
+            FCVTZSxD(x4, s0);
+            FMULD(s0, s0, v2);
+            FSUBD(v1, v1, s0);
+            LDRw_U12(x1, xEmu, offsetof(x64emu_t, sw));
+            // set C2 = 0
+            BFCw(x1, 10, 1);
+            // set C1 = Q0
+            BFIw(x1, x4, 9, 1);
+            // set C3 = Q1
+            LSRx_IMM(x4, x4, 1);
+            BFIw(x1, x4, 14, 1);
+            // Set C0 = Q2
+            LSRx(x4, x4, 1);
+            BFIw(x1, x4, 8, 1);
+            STRw_U12(x1, xEmu, offsetof(x64emu_t, sw));
+            #endif
             break;
         case 0xF6:
             INST_NAME("FDECSTP");
@@ -242,18 +283,45 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             break;
         case 0xF8:
             INST_NAME("FPREM");
+            #if 0
             MESSAGE(LOG_DUMP, "Need Optimization\n");
+            i1 = x87_stackcount(dyn, ninst, x1);
             x87_forget(dyn, ninst, x1, x2, 0);
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(native_fprem, -1);
+            x87_unstackcount(dyn, ninst, x1, i1);
+            #else
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+            v2 = x87_get_st(dyn, ninst, x1, x2, 1, NEON_CACHE_ST_D);
+            s0 = fpu_get_scratch(dyn);
+            FDIVD(s0, v1, v2);
+            FRINTZD(s0, s0);
+            FCVTZSxD(x4, s0);
+            FMULD(s0, s0, v2);
+            FSUBD(v1, v1, s0);
+            LDRw_U12(x1, xEmu, offsetof(x64emu_t, sw));
+            // set C2 = 0
+            BFCw(x1, 10, 1);
+            // set C1 = Q0
+            BFIw(x1, x4, 9, 1);
+            // set C3 = Q1
+            LSRx_IMM(x4, x4, 1);
+            BFIw(x1, x4, 14, 1);
+            // Set C0 = Q2
+            LSRx(x4, x4, 1);
+            BFIw(x1, x4, 8, 1);
+            STRw_U12(x1, xEmu, offsetof(x64emu_t, sw));
+            #endif
             break;
         case 0xF9:
             INST_NAME("FYL2XP1");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
+            i1 = x87_stackcount(dyn, ninst, x1);
             x87_forget(dyn, ninst, x1, x2, 0);
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(native_fyl2xp1, -1);
-            x87_do_pop(dyn, ninst, x3);
+            x87_unstackcount(dyn, ninst, x1, i1);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xFA:
             INST_NAME("FSQRT");
@@ -267,9 +335,11 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0xFB:
             INST_NAME("FSINCOS");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
-            x87_do_push_empty(dyn, ninst, 0);
+            X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, 0);
+            i1 = x87_stackcount(dyn, ninst, x1);
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(native_fsincos, -1);
+            x87_unstackcount(dyn, ninst, x1, i1);
             break;
         case 0xFC:
             INST_NAME("FRNDINT");
@@ -292,21 +362,27 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0xFD:
             INST_NAME("FSCALE");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
+            i1 = x87_stackcount(dyn, ninst, x1);
             x87_forget(dyn, ninst, x1, x2, 0);
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(native_fscale, -1);
+            x87_unstackcount(dyn, ninst, x1, i1);
             break;
         case 0xFE:
             INST_NAME("FSIN");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
+            i1 = x87_stackcount(dyn, ninst, x1);
             x87_forget(dyn, ninst, x1, x2, 0);
             CALL(native_fsin, -1);
+            x87_unstackcount(dyn, ninst, x1, i1);
             break;
         case 0xFF:
             INST_NAME("FCOS");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
+            i1 = x87_stackcount(dyn, ninst, x1);
             x87_forget(dyn, ninst, x1, x2, 0);
             CALL(native_fcos, -1);
+            x87_unstackcount(dyn, ninst, x1, i1);
             break;
 
 
@@ -322,12 +398,12 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
         case 0xEF:
             DEFAULT;
             break;
-             
+
         default:
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("FLD ST0, float[ED]");
-                    v1 = x87_do_push(dyn, ninst, x1, box64_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
+                    X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, box64_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
                     VLD32(v1, ed, fixedaddress);
                     if(!ST_IS_F(0)) {
@@ -354,7 +430,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         FCVT_S_D(v1, v1);
                     }
                     VST32(v1, ed, fixedaddress);
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 case 4:
                     INST_NAME("FLDENV Ed");

src/dynarec/arm64/dynarec_arm64_da.c

@@ -0,0 +1,239 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.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 "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.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"
+
+
+uintptr_t dynarec64_DA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+{
+    uint8_t nextop = F8;
+    int64_t j64;
+    uint8_t ed;
+    uint8_t wback;
+    int v1, v2;
+    int d0;
+    int s0;
+    int64_t fixedaddress;
+    int unscaled;
+
+    MAYUSE(s0);
+    MAYUSE(d0);
+    MAYUSE(v2);
+    MAYUSE(v1);
+    MAYUSE(ed);
+    MAYUSE(j64);
+
+    switch(nextop) {
+        case 0xC0:
+        case 0xC1:
+        case 0xC2:
+        case 0xC3:
+        case 0xC4:
+        case 0xC5:
+        case 0xC6:
+        case 0xC7:
+            INST_NAME("FCMOVB ST0, STx");
+            READFLAGS(X_CF);
+            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
+            if(ST_IS_F(0)) {
+                FCSELS(v1, v2, v1, cNE);   // F_CF==1
+            } else {
+                FCSELD(v1, v2, v1, cNE);   // F_CF==1
+            }
+            break;
+        case 0xC8:
+        case 0xC9:
+        case 0xCA:
+        case 0xCB:
+        case 0xCC:
+        case 0xCD:
+        case 0xCE:
+        case 0xCF:
+            INST_NAME("FCMOVE ST0, STx");
+            READFLAGS(X_ZF);
+            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
+            if(ST_IS_F(0)) {
+                FCSELS(v1, v2, v1, cNE);
+            } else {
+                FCSELD(v1, v2, v1, cNE);        // F_ZF==0
+            }
+            break;
+        case 0xD0:
+        case 0xD1:
+        case 0xD2:
+        case 0xD3:
+        case 0xD4:
+        case 0xD5:
+        case 0xD6:
+        case 0xD7:
+            INST_NAME("FCMOVBE ST0, STx");
+            READFLAGS(X_CF|X_ZF);
+            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);
+            if(ST_IS_F(0)) {
+                FCSELS(v1, v2, v1, cNE);
+            } else {
+                FCSELD(v1, v2, v1, cNE);   // F_CF==0 & F_ZF==0
+            }
+            break;
+        case 0xD8:
+        case 0xD9:
+        case 0xDA:
+        case 0xDB:
+        case 0xDC:
+        case 0xDD:
+        case 0xDE:
+        case 0xDF:
+            INST_NAME("FCMOVU ST0, STx");
+            READFLAGS(X_PF);
+            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
+            if(ST_IS_F(0)) {
+                FCSELS(v1, v2, v1, cNE);
+            } else {
+                FCSELD(v1, v2, v1, cNE);        // F_PF==0
+            }
+            break;
+        case 0xE9:
+            INST_NAME("FUCOMPP ST0, ST1");
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
+            v2 = x87_get_st(dyn, ninst, x1, x2, 1, X87_COMBINE(0, nextop&7));
+            if(ST_IS_F(0)) {
+                FCMPS(v1, v2);
+            } else {
+                FCMPD(v1, v2);
+            }
+            FCOM(x1, x2, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
+            break;
+
+        case 0xE4:
+        case 0xF0:
+        case 0xF1:
+        case 0xF4:
+        case 0xF5:
+        case 0xF6:
+        case 0xF7:
+        case 0xF8:
+        case 0xF9:
+        case 0xFD:
+            DEFAULT;
+            break;
+
+        default:
+            switch((nextop>>3)&7) {
+                case 0:
+                    INST_NAME("FIADD ST0, Ed");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                    v2 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                    VLD32(v2, ed, fixedaddress);
+                    SXTL_32(v2, v2);    // i32 -> i64
+                    SCVTFDD(v2, v2);    // i64 -> double
+                    FADDD(v1, v1, v2);
+                    break;
+                case 1:
+                    INST_NAME("FIMUL ST0, Ed");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                    v2 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                    VLD32(v2, ed, fixedaddress);
+                    SXTL_32(v2, v2);    // i32 -> i64
+                    SCVTFDD(v2, v2);    // i64 -> double
+                    FMULD(v1, v1, v2);
+                    break;
+                case 2:
+                    INST_NAME("FICOM ST0, Ed");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                    v2 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                    VLD32(v2, ed, fixedaddress);
+                    SXTL_32(v2, v2);    // i32 -> i64
+                    SCVTFDD(v2, v2);    // i64 -> double
+                    FCMPD(v1, v2);
+                    FCOM(x1, x2, x3);
+                    break;
+                case 3:
+                    INST_NAME("FICOMP ST0, Ed");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                    v2 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                    VLD32(v2, ed, fixedaddress);
+                    SXTL_32(v2, v2);    // i32 -> i64
+                    SCVTFDD(v2, v2);    // i64 -> double
+                    FCMPD(v1, v2);
+                    FCOM(x1, x2, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
+                    break;
+                case 4:
+                    INST_NAME("FISUB ST0, Ed");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                    v2 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                    VLD32(v2, ed, fixedaddress);
+                    SXTL_32(v2, v2);    // i32 -> i64
+                    SCVTFDD(v2, v2);    // i64 -> double
+                    FSUBD(v1, v1, v2);
+                    break;
+                case 5:
+                    INST_NAME("FISUBR ST0, Ed");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                    v2 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                    VLD32(v2, ed, fixedaddress);
+                    SXTL_32(v2, v2);    // i32 -> i64
+                    SCVTFDD(v2, v2);    // i64 -> double
+                    FSUBD(v1, v2, v1);
+                    break;
+                case 6:
+                    INST_NAME("FIDIV ST0, Ed");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                    v2 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                    VLD32(v2, ed, fixedaddress);
+                    SXTL_32(v2, v2);    // i32 -> i64
+                    SCVTFDD(v2, v2);    // i64 -> double
+                    FDIVD(v1, v1, v2);
+                    break;
+                case 7:
+                    INST_NAME("FIDIVR ST0, Ed");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                    v2 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
+                    VLD32(v2, ed, fixedaddress);
+                    SXTL_32(v2, v2);    // i32 -> i64
+                    SCVTFDD(v2, v2);    // i64 -> double
+                    FDIVD(v1, v2, v1);
+                    break;
+            }
+    }
+    return addr;
+}
+

src/dynarec/arm64/dynarec_arm64_db.c

@@ -156,7 +156,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             }
             FCOMI(x1, x2);
             break;
-        case 0xF0:  
+        case 0xF0:
         case 0xF1:
         case 0xF2:
         case 0xF3:
@@ -188,7 +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, x1, NEON_CACHE_ST_D);
+                    X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
                     VLD32(v1, ed, fixedaddress);
                     SXTL_32(v1, v1);    // i32 -> i64
@@ -218,7 +218,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     STW(x5, wback, fixedaddress);
                     MARK3;
                     #endif
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 case 2:
                     INST_NAME("FIST Ed, ST0");
@@ -273,7 +273,7 @@ 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, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 case 5:
                     INST_NAME("FLD tbyte");
@@ -296,13 +296,13 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         STRH_U12(x6, ed, 8);
                     } else {
                         if(box64_x87_no80bits) {
-                            v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
+                            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
                             VLDR64_U12(v1, ed, fixedaddress);
                         } else {
                             if(ed!=x1) {
                                 MOVx_REG(x1, ed);
                             }
-                            x87_do_push_empty(dyn, ninst, x3);
+                            X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, x3);
                             CALL(native_fld, -1);
                         }
                     }
@@ -344,7 +344,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         CMPSw_REG(x3, x4);
                         B_MARK2(cNE);
                         // NaN and Infinite
-                        ORRw_mask(x3, x5, 0, 0b1110);    //x3 = sign | 0x7fff 
+                        ORRw_mask(x3, x5, 0, 0b1110);    //x3 = sign | 0x7fff
                         TSTx_mask(x1, 1, 0, 0b110011); //0x000fffffffffffffL
                         ORRx_mask(x5, xZR, 1, 1, 0);    //0x8000000000000000
                         ORRx_mask(x4, xZR, 1, 0b10, 0b01); //0xc000000000000000
@@ -372,7 +372,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         STRH_U12(x3, wback, 8);
                         #endif
                     }
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 default:
                     DEFAULT;

src/dynarec/arm64/dynarec_arm64_dc.c

@@ -106,7 +106,7 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 FCMPD(v1, v2);
             }
             FCOM(x1, x2, x3);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xE0:
         case 0xE1:
@@ -158,7 +158,7 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             } else {
                 FDIVD(v1, v2, v1);
             }
-            break;       
+            break;
         case 0xF8:
         case 0xF9:
         case 0xFA:
@@ -211,7 +211,7 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     VLD64(v2, wback, fixedaddress);
                     FCMPD(v1, v2);
                     FCOM(x1, x2, x3);
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 case 4:
                     INST_NAME("FSUB ST0, double[ED]");

src/dynarec/arm64/dynarec_arm64_dd.c

@@ -74,7 +74,7 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             break;
         case 0xD8:
             INST_NAME("FSTP ST0, ST0");
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xD9:
         case 0xDA:
@@ -88,7 +88,7 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             x87_get_st_empty(dyn, ninst, x1, x2, nextop&7, X87_ST(nextop&7));
             x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
             x87_swapreg(dyn, ninst, x1, x2, 0, nextop&7);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
 
         case 0xE0:
@@ -126,7 +126,7 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 FCMPD(v1, v2);
             }
             FCOM(x1, x2, x3);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
 
         case 0xC8:
@@ -160,34 +160,38 @@ 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, x3, NEON_CACHE_ST_D);
+                    X87_PUSH_OR_FAIL(v1, dyn, ninst, x3, NEON_CACHE_ST_D);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
                     VLD64(v1, ed, fixedaddress);
                     break;
                 case 1:
                     INST_NAME("FISTTP i64, ST0");
-                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_I64);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
-                    s0 = fpu_get_scratch(dyn);
-                    #if 0
-                    // those are ARM 8.5 opcode!
-                    FRINT64ZD(s0, v1);
-                    FCVTZSxD(x2, s0);
-                    STRx_U12(x2, ed, fixedaddress);
-                    #else
-                    MRS_fpsr(x5);
-                    BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
-                    MSR_fpsr(x5);
-                    FRINTRRD(s0, v1, 3);
-                    FCVTZSxD(x2, s0);
-                    STx(x2, ed, fixedaddress);
-                    MRS_fpsr(x5);   // get back FPSR to check the IOC bit
-                    TBZ_MARK3(x5, FPSR_IOC);
-                    ORRx_mask(x5, xZR, 1, 1, 0);    //0x8000000000000000
-                    STx(x5, ed, fixedaddress);
-                    MARK3;
-                    #endif
-                    x87_do_pop(dyn, ninst, x3);
+                    if(ST_IS_I64(0)) {
+                        VST64(v1, ed, fixedaddress);
+                    } else {
+                        s0 = fpu_get_scratch(dyn);
+                        #if 0
+                        // those are ARM 8.5 opcode!
+                        FRINT64ZD(s0, v1);
+                        FCVTZSxD(x2, s0);
+                        STRx_U12(x2, ed, fixedaddress);
+                        #else
+                        MRS_fpsr(x5);
+                        BFCw(x5, FPSR_IOC, 1);   // reset IOC bit
+                        MSR_fpsr(x5);
+                        FRINTRRD(s0, v1, 3);
+                        FCVTZSxD(x2, s0);
+                        STx(x2, ed, fixedaddress);
+                        MRS_fpsr(x5);   // get back FPSR to check the IOC bit
+                        TBZ_MARK3(x5, FPSR_IOC);
+                        ORRx_mask(x5, xZR, 1, 1, 0);    //0x8000000000000000
+                        STx(x5, ed, fixedaddress);
+                        MARK3;
+                        #endif
+                    }
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 case 2:
                     INST_NAME("FST double");
@@ -200,9 +204,9 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
                     VST64(v1, ed, fixedaddress);
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
-                case 4: 
+                case 4:
                     INST_NAME("FRSTOR m108byte");
                     MESSAGE(LOG_DUMP, "Need Optimization\n");
                     fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
@@ -210,7 +214,7 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     if(ed!=x1) {MOVx_REG(x1, ed);}
                     CALL(native_frstor, -1);
                     break;
-                case 6: 
+                case 6:
                     INST_NAME("FSAVE m108byte");
                     MESSAGE(LOG_DUMP, "Need Optimization\n");
                     fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
@@ -220,7 +224,7 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     break;
                 case 7:
                     INST_NAME("FNSTSW m2byte");
-                    fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
+                    //fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x4, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
                     LDRw_U12(x2, xEmu, offsetof(x64emu_t, top));
                     LDRH_U12(x3, xEmu, offsetof(x64emu_t, sw));
@@ -234,6 +238,7 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         ANDw_mask(x2, x2, 0, 2);
                     }
                     BFIw(x3, x2, 11, 3); // inject TOP at bit 11 (3 bits)
+                    STRH_U12(x3, xEmu, offsetof(x64emu_t, sw));
                     STH(x3, ed, fixedaddress);   // store whole sw flags
                     break;
                 default:

src/dynarec/arm64/dynarec_arm64_de.c

@@ -53,7 +53,7 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             } else {
                 FADDD(v1, v1, v2);
             }
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xC8:
         case 0xC9:
@@ -71,7 +71,7 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             } else {
                 FMULD(v1, v1, v2);
             }
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xD0:
         case 0xD1:
@@ -90,7 +90,7 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 FCMPD(v1, v2);
             }
             FCOM(x1, x2, x3);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xD9:
             INST_NAME("FCOMPP ST0, STx");
@@ -102,8 +102,8 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 FCMPD(v1, v2);
             }
             FCOM(x1, x2, x3);
-            x87_do_pop(dyn, ninst, x3);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xE0:
         case 0xE1:
@@ -121,7 +121,7 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             } else {
                 FSUBD(v1, v2, v1);
             }
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xE8:
         case 0xE9:
@@ -139,7 +139,7 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             } else {
                 FSUBD(v1, v1, v2);
             }
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xF0:
         case 0xF1:
@@ -157,8 +157,8 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             } else {
                 FDIVD(v1, v2, v1);
             }
-            x87_do_pop(dyn, ninst, x3);
-            break;       
+            X87_POP_OR_FAIL(dyn, ninst, x3);
+            break;
         case 0xF8:
         case 0xF9:
         case 0xFA:
@@ -175,7 +175,7 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
             } else {
                 FDIVD(v1, v1, v2);
             }
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xD8:
         case 0xDA:
@@ -233,7 +233,7 @@ uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     SCVTFDD(v2, v2);
                     FCMPD(v1, v2);
                     FCOM(x1, x2, x3);
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 case 4:
                     INST_NAME("FISUB ST0, word[ED]");

src/dynarec/arm64/dynarec_arm64_df.c

@@ -34,6 +34,7 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
     int64_t j64;
     int64_t fixedaddress;
     int unscaled;
+    int i1;
 
     MAYUSE(s0);
     MAYUSE(v2);
@@ -51,12 +52,20 @@ 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, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
 
         case 0xE0:
             INST_NAME("FNSTSW AX");
             LDRw_U12(x2, xEmu, offsetof(x64emu_t, top));
+            if(dyn->n.x87stack) {
+                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);  //mask=7
+            }
             LDRH_U12(x1, xEmu, offsetof(x64emu_t, sw));
             BFIw(x1, x2, 11, 3); // inject top
             STRH_U12(x1, xEmu, offsetof(x64emu_t, sw));
@@ -81,7 +90,7 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 FCMPD(v1, v2);
             }
             FCOMI(x1, x2);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xF0:
         case 0xF1:
@@ -102,7 +111,7 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                 FCMPD(v1, v2);
             }
             FCOMI(x1, x2);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
 
         case 0xC8:
@@ -151,7 +160,7 @@ 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, x1, NEON_CACHE_ST_F);
+                    X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_F);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, 0);
                     LDSHw(x1, wback, fixedaddress);
                     if(ST_IS_F(0)) {
@@ -196,7 +205,7 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     MARK3;
                     STH(x3, wback, fixedaddress);
                     #endif
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 case 2:
                     INST_NAME("FIST Ew, ST0");
@@ -275,19 +284,19 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     MARK3;
                     STH(x3, wback, fixedaddress);
                     #endif
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     x87_restoreround(dyn, ninst, u8);
                     break;
                 case 4:
                     INST_NAME("FBLD ST0, tbytes");
-                    x87_do_push_empty(dyn, ninst, x1);
+                    X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, x1);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                     if(ed!=x1) {MOVx_REG(x1, ed);}
                     CALL(fpu_fbld, -1);
                     break;
                 case 5:
                     INST_NAME("FILD ST0, i64");
-                    v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_I64);
+                    X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_I64);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
                     VLD64(v1, wback, fixedaddress);
                     if(!ST_IS_I64(0)) {
@@ -315,11 +324,13 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                     break;
                 case 6:
                     INST_NAME("FBSTP tbytes, ST0");
+                    i1 = x87_stackcount(dyn, ninst, x1);
                     x87_forget(dyn, ninst, x1, x2, 0);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
                     if(ed!=x1) {MOVx_REG(x1, ed);}
                     CALL(fpu_fbst, -1);
-                    x87_do_pop(dyn, ninst, x3);
+                    x87_unstackcount(dyn, ninst, x1, i1);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 case 7:
                     INST_NAME("FISTP i64, ST0");
@@ -375,7 +386,7 @@ uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
                         #endif
                         x87_restoreround(dyn, ninst, u8);
                     }
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 default:
                     DEFAULT;

src/dynarec/arm64/dynarec_arm64_emit_logic.c

@@ -230,15 +230,7 @@ void emit_or8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
         MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
         BICw(xFlags, xFlags, s3);
     }
-    IFX(X_ZF) {
-        TSTw_REG(s1, s1);
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -261,15 +253,7 @@ void emit_or8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4)
         MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
         BICw(xFlags, xFlags, s3);
     }
-    IFX(X_ZF) {
-        TSTw_REG(s1, s1);
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -292,15 +276,7 @@ void emit_xor8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
         MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
         BICw(xFlags, xFlags, s3);
     }
-    IFX(X_ZF) {
-        TSTw_REG(s1, s1);
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -323,15 +299,7 @@ void emit_xor8c(dynarec_arm_t* dyn, int ninst, int s1, int32_t c, int s3, int s4
         MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
         BICw(xFlags, xFlags, s3);
     }
-    IFX(X_ZF) {
-        TSTw_REG(s1, s1);
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -423,15 +391,7 @@ void emit_or16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
         MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
         BICw(xFlags, xFlags, s3);
     }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 15);  //mask=0xffff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 15);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 16)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -500,15 +460,7 @@ void emit_xor16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
         MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
         BICw(xFlags, xFlags, s3);
     }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 15);  //mask=0xffff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 15);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 16)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }

src/dynarec/arm64/dynarec_arm64_emit_math.c

@@ -303,15 +303,7 @@ void emit_add8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
     IFX(X_PEND) {
         STRH_U12(s1, xEmu, offsetof(x64emu_t, res));
     }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 7);   //mask=0xff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -329,7 +321,7 @@ void emit_add8c(dynarec_arm_t* dyn, int ninst, int s1, int c, int s3, int s4)
         SET_DFNONE(s3);
     }
     IFX(X_AF | X_OF) {
-        if(X_PEND) {} else {MOV32w(s4, c&0xff);}
+        IFX(X_PEND) {} else {MOV32w(s4, c&0xff);}
         ORRw_REG(s3, s1, s4);       // s3 = op1 | op2
         ANDw_REG(s4, s1, s4);       // s4 = op1 & op2
     }
@@ -355,15 +347,7 @@ void emit_add8c(dynarec_arm_t* dyn, int ninst, int s1, int c, int s3, int s4)
     IFX(X_PEND) {
         STRH_U12(s1, xEmu, offsetof(x64emu_t, res));
     }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 0b000111);    //mask=000000ff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -407,15 +391,7 @@ void emit_sub8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
             BFIw(xFlags, s4, F_OF, 1);    // OF: ((bc >> 6) ^ ((bc>>6)>>1)) & 1
         }
     }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 7);   //mask=0xff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -465,15 +441,7 @@ void emit_sub8c(dynarec_arm_t* dyn, int ninst, int s1, int c, int s3, int s4, in
             BFIw(xFlags, s4, F_OF, 1);    // OF: ((bc >> 6) ^ ((bc>>6)>>1)) & 1
         }
     }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 0b000111);    //mask=000000ff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -516,15 +484,7 @@ void emit_add16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
     IFX(X_PEND) {
         STRw_U12(s1, xEmu, offsetof(x64emu_t, res));
     }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 15);  //mask=0xffff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 15);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 16)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -634,15 +594,7 @@ void emit_sub16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
             BFIw(xFlags, s4, F_OF, 1);    // OF: ((bc >> 14) ^ ((bc>>14)>>1)) & 1
         }
     }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 15);  //mask=0xffff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 15);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 16)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -792,16 +744,7 @@ void emit_inc8(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
             BFIw(xFlags, s4, F_OF, 1);    // OF: ((bc >> 6) ^ ((bc>>6)>>1)) & 1
         }
     }
-
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 7);   //mask=0xff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -838,15 +781,7 @@ void emit_inc16(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
             BFIw(xFlags, s4, F_OF, 1);    // OF: ((bc >> 14) ^ ((bc>>14)>>1)) & 1
         }
     }
-    IFX(X_ZF) {
-        TSTw_mask(s1, 0, 0b001111); // mask=0xffff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 15);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 16)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -906,7 +841,7 @@ void emit_dec32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s3, int s4
 void emit_dec8(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
 {
     IFX(X_PEND) {
-        STRB_U12(s3, xEmu, offsetof(x64emu_t, op2));
+        STRB_U12(s1, xEmu, offsetof(x64emu_t, op1));
         SET_DF(s3, d_dec8);
     } else IFX(X_ZF|X_OF|X_AF|X_SF|X_PF) {
         SET_DFNONE(s3);
@@ -1162,15 +1097,7 @@ void emit_adc8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
         LSRw(s3, s1, 8);
         BFIw(xFlags, s3, F_CF, 1);
     }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 7);   //mask=0xff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -1224,15 +1151,7 @@ void emit_adc16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
         LSRw(s3, s1, 16);
         BFIw(xFlags, s3, F_CF, 1);
     }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 15);   //mask=0xffff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 15);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 16)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -1474,15 +1393,7 @@ void emit_sbb8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
             BFIw(xFlags, s4, F_OF, 1);    // OF: ((bc >> 6) ^ ((bc>>6)>>1)) & 1
         }
     }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 7);   //mask=0xff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -1537,15 +1448,7 @@ void emit_sbb16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4)
             BFIw(xFlags, s4, F_OF, 1);    // OF: ((bc >> 14) ^ ((bc>>14)>>1)) & 1
         }
     }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 15);  //mask=0xffff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 15);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 16)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -1732,10 +1635,9 @@ void emit_neg8(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
     IFX(X_AF|X_OF) {
         MOVw_REG(s3, s1);
     }
+    NEGw_REG(s1, s1);
     IFX(X_ZF) {
-        NEGSw_REG(s1, s1);
-    } else {
-        NEGw_REG(s1, s1);
+        ANDSw_mask(s1, s1, 0, 7);   // mask 0xff
     }
     IFX(X_PEND) {
         STRB_U12(s1, xEmu, offsetof(x64emu_t, res));

src/dynarec/arm64/dynarec_arm64_emit_tests.c

@@ -111,22 +111,11 @@ void emit_cmp16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, i
     } else {
         SET_DFNONE(s3);
     }
-    IFX(X_ZF) {
-        SUBSw_REG(s5, s1, s2);   // res = s1 - s2
-    } else {
-        SUBw_REG(s5, s1, s2);   // res = s1 - s2
-    }
+    SUBw_REG(s5, s1, s2);   // res = s1 - s2
     IFX_PENDOR0 {
         STRH_U12(s5, xEmu, offsetof(x64emu_t, res));
     }
-    IFX(X_ZF) {
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s4, s5, 15);
-        BFIw(xFlags, s4, F_SF, 1);
-    }
+    COMP_ZFSF(s5, 16)
     // bc = (res & (~d | s)) | (~d & s)
     IFX(X_CF|X_AF|X_OF) {
         MVNw_REG(s4, s1);        // s4 = ~d
@@ -170,15 +159,7 @@ void emit_cmp16_0(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
         MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
         BICw(xFlags, xFlags, s3);
     }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 15);  //mask=0xffff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s4, s1, 15);
-        BFIw(xFlags, s4, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 16)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }
@@ -194,22 +175,11 @@ void emit_cmp8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4, in
     } else {
         SET_DFNONE(s4);
     }
-    IFX(X_ZF) {
-        SUBSw_REG(s5, s1, s2);   // res = s1 - s2
-    } else {
-        SUBw_REG(s5, s1, s2);   // res = s1 - s2
-    }
+    SUBw_REG(s5, s1, s2);   // res = s1 - s2
     IFX_PENDOR0 {
         STRB_U12(s5, xEmu, offsetof(x64emu_t, res));
     }
-    IFX(X_ZF) {
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s5, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s5, 8)
     // bc = (res & (~d | s)) | (~d & s)
     IFX(X_CF|X_AF|X_OF) {
         ORNw_REG(s4, s2, s1);   // s4 = ~d | s
@@ -251,15 +221,7 @@ void emit_cmp8_0(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
         MOV32w(s3, (1<<F_CF)|(1<<F_AF)|(1<<F_OF));
         BICw(xFlags, xFlags, s3);
     }
-    IFX(X_ZF) {
-        ANDSw_mask(s1, s1, 0, 0b000111);    //mask=000000ff
-        CSETw(s3, cEQ);
-        BFIw(xFlags, s3, F_ZF, 1);
-    }
-    IFX(X_SF) {
-        LSRw(s3, s1, 7);
-        BFIw(xFlags, s3, F_SF, 1);
-    }
+    COMP_ZFSF(s1, 8)
     IFX(X_PF) {
         emit_pf(dyn, ninst, s1, s3, s4);
     }

src/dynarec/arm64/dynarec_arm64_f20f.c

@@ -28,7 +28,7 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
     uint8_t opcode = F8;
     uint8_t nextop;
     uint8_t gd, ed;
-    uint8_t wback;
+    uint8_t wback, wb1, wb2;
     uint8_t u8;
     uint64_t u64, j64;
     int v0, v1;
@@ -159,30 +159,41 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             switch(opcode) {
 
                 case 0xF0:
-                    INST_NAME("(unsupported) CRC32 Gd, Eb)");
+                    INST_NAME("CRC32 Gd, Eb");
                     nextop = F8;
-                    addr = fakeed(dyn, addr, ninst, nextop);
-                    SETFLAGS(X_ALL, SF_SET);    // Hack to set flags in "don't care" state
-                    GETIP(ip);
-                    STORE_XEMU_CALL(xRIP);
-                    CALL(native_ud, -1);
-                    LOAD_XEMU_CALL(xRIP);
-                    jump_to_epilog(dyn, 0, xRIP, ninst);
-                    *need_epilog = 0;
-                    *ok = 0;
+                    GETEB(x1, 0);
+                    GETGD;
+                    if(arm64_crc32) {
+                        CRC32CB(gd, gd, ed);
+                    } else {
+                        EORw_REG(gd, gd, ed);
+                        MOV32w(x2, 0x82f63b78);
+                        for(int i=0; i<8; ++i) {
+                            LSRw_IMM((i&1)?gd:x4, (i&1)?x4:gd, 1);
+                            TBZ((i&1)?x4:gd, 0, 4+4);
+                            EORw_REG((i&1)?gd:x4, (i&1)?gd:x4, x2);
+                        }
+                    }
                     break;
                 case 0xF1:
-                    INST_NAME("(unsupported) CRC32 Gd, Ed)");
+                    INST_NAME("CRC32 Gd, Ed");
                     nextop = F8;
-                    addr = fakeed(dyn, addr, ninst, nextop);
-                    SETFLAGS(X_ALL, SF_SET);    // Hack to set flags in "don't care" state
-                    GETIP(ip);
-                    STORE_XEMU_CALL(xRIP);
-                    CALL(native_ud, -1);
-                    LOAD_XEMU_CALL(xRIP);
-                    jump_to_epilog(dyn, 0, xRIP, ninst);
-                    *need_epilog = 0;
-                    *ok = 0;
+                    GETED(0);
+                    GETGD;
+                    if(arm64_crc32) {
+                        CRC32Cxw(gd, gd, ed);
+                    } else {
+                        MOV32w(x2, 0x82f63b78);
+                        for(int j=0; j<4*(1+rex.w); ++j) {
+                            UBFXxw(x3, ed, 8*j, 8);
+                            EORw_REG(gd, gd, x3);
+                            for(int i=0; i<8; ++i) {
+                                LSRw_IMM((i&1)?gd:x4, (i&1)?x4:gd, 1);
+                                TBZ((i&1)?x4:gd, 0, 4+4);
+                                EORw_REG((i&1)?gd:x4, (i&1)?gd:x4, x2);
+                            }
+                        }
+                    }
                     break;
 
                 default:
@@ -376,7 +387,22 @@ uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             }
             VFADDPQS(v0, v0, v1);
             break;
-            
+        case 0x7D:
+            INST_NAME("HSUBPS Gx, Ex");
+            nextop = F8;
+            GETGX(v0, 1);
+            if(MODREG) {
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), 0);
+            } else {
+                addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<4, 15, rex, NULL, 0, 0);
+                v1 = fpu_get_scratch(dyn);
+                VLD128(v1, ed, fixedaddress);
+            }
+            d0 = fpu_get_scratch(dyn);
+            VUZP1Q_32(d0, v0, v1);
+            VUZP2Q_32(v0, v0, v1);
+            VFSUBQS(v0, d0, v0);
+            break;
         case 0xC2:
             INST_NAME("CMPSD Gx, Ex, Ib");
             nextop = F8;

src/dynarec/arm64/dynarec_arm64_functions.c

@@ -30,7 +30,6 @@
 #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)
@@ -96,6 +95,15 @@ void fpu_reset_reg(dynarec_arm_t* dyn)
 
 }
 
+int neoncache_no_i64(dynarec_arm_t* dyn, int ninst, int st, int a)
+{
+    if(a==NEON_CACHE_ST_I64) {
+        neoncache_promote_double(dyn, ninst, st);
+        return NEON_CACHE_ST_D;
+    }
+    return a;
+}
+
 int neoncache_get_st(dynarec_arm_t* dyn, int ninst, int a)
 {
     if (dyn->insts[ninst].n.swapped) {
@@ -139,7 +147,7 @@ int neoncache_get_st_f(dynarec_arm_t* dyn, int ninst, int a)
          && dyn->insts[ninst].n.neoncache[i].n==a)
             return i;
     return -1;
-} 
+}
 int neoncache_get_st_f_i64(dynarec_arm_t* dyn, int ninst, int a)
 {
     /*if(a+dyn->insts[ninst].n.stack_next-st<0)
@@ -150,7 +158,7 @@ int neoncache_get_st_f_i64(dynarec_arm_t* dyn, int ninst, int a)
          && 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)
@@ -158,7 +166,7 @@ int neoncache_get_st_f_noback(dynarec_arm_t* dyn, int ninst, int a)
          && dyn->insts[ninst].n.neoncache[i].n==a)
             return i;
     return -1;
-} 
+}
 int neoncache_get_st_f_i64_noback(dynarec_arm_t* dyn, int ninst, int a)
 {
     for(int i=0; i<24; ++i)
@@ -166,7 +174,7 @@ int neoncache_get_st_f_i64_noback(dynarec_arm_t* dyn, int ninst, int a)
          && 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)
@@ -174,7 +182,7 @@ int neoncache_get_current_st_f(dynarec_arm_t* dyn, int a)
          && dyn->n.neoncache[i].n==a)
             return i;
     return -1;
-} 
+}
 int neoncache_get_current_st_f_i64(dynarec_arm_t* dyn, int a)
 {
     for(int i=0; i<24; ++i)
@@ -182,7 +190,7 @@ int neoncache_get_current_st_f_i64(dynarec_arm_t* dyn, int a)
          && 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)
@@ -190,7 +198,7 @@ static void neoncache_promote_double_combined(dynarec_arm_t* dyn, int ninst, int
     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 
+        } else
             a = dyn->insts[ninst].n.combined1;
         int i = neoncache_get_st_f_i64_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);
@@ -314,7 +322,7 @@ static int isCacheEmpty(dynarec_native_t* dyn, int ninst) {
     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_F
              || dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_D
              || dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_I64)
             && dyn->insts[ninst].n.neoncache[i].n<dyn->insts[ninst].n.stack_pop))
@@ -339,7 +347,7 @@ int fpuCacheNeedsTransform(dynarec_arm_t* dyn, int ninst) {
         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_F
                 || dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_D
                 || dyn->insts[ninst].n.neoncache[i].t==NEON_CACHE_ST_I64)
                 && dyn->insts[ninst].n.neoncache[i].n<dyn->insts[ninst].n.stack_pop))
@@ -377,7 +385,7 @@ void neoncacheUnwind(neoncache_t* cache)
 {
     if(cache->swapped) {
         // unswap
-        int a = -1; 
+        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 || cache->neoncache[j].t == NEON_CACHE_ST_I64)) {
@@ -580,4 +588,4 @@ void inst_name_pass3(dynarec_native_t* dyn, int ninst, const char* name, rex_t r
 void print_opcode(dynarec_native_t* dyn, int ninst, uint32_t opcode)
 {
     dynarec_log(LOG_NONE, "\t%08x\t%s\n", opcode, arm64_print(opcode, (uintptr_t)dyn->block));
-}
+}

src/dynarec/arm64/dynarec_arm64_functions.h

@@ -3,6 +3,8 @@
 
 #include "../dynarec_native_functions.h"
 
+#define SCRATCH0    24
+
 // Get an FPU scratch reg
 int fpu_get_scratch(dynarec_arm_t* dyn);
 // Reset scratch regs counter
@@ -35,6 +37,9 @@ int neoncache_get_current_st_f_i64(dynarec_arm_t* dyn, int a);
 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*
+// Do not allow i64 type
+int neoncache_no_i64(dynarec_arm_t* dyn, int ninst, int st, int a);
+
 
 // FPU Cache transformation (for loops) // Specific, need to be written by backend
 int fpuCacheNeedsTransform(dynarec_arm_t* dyn, int ninst);

src/dynarec/arm64/dynarec_arm64_helper.c

@@ -32,6 +32,8 @@ uintptr_t geted(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop, u
 {
     MAYUSE(dyn); MAYUSE(ninst); MAYUSE(delta);
 
+    if(l==LOCK_LOCK) { /*SMDMB();*/DMB_ISH(); }
+
     if(rex.is32bits)
         return geted_32(dyn, addr, ninst, nextop, ed, hint, fixaddress, unscaled, absmax, mask, l, s);
 
@@ -69,7 +71,7 @@ uintptr_t geted(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop, u
                     }
                 } else {
                     switch(lock) {
-                        case 1: addLockAddress(tmp); break;
+                        case 1: addLockAddress(tmp); if(fixaddress) *fixaddress=tmp; break;
                         case 2: if(isLockAddress(tmp)) *l=1; break;
                     }
                     MOV64x(ret, tmp);
@@ -106,7 +108,7 @@ uintptr_t geted(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop, u
                 ADDx_REG(ret, ret, xRIP);
             }
             switch(lock) {
-                case 1: addLockAddress(addr+delta+tmp); break;
+                case 1: addLockAddress(addr+delta+tmp); if(fixaddress) *fixaddress=addr+delta+tmp; break;
                 case 2: if(isLockAddress(addr+delta+tmp)) *l=1; break;
             }
         } else {
@@ -126,7 +128,7 @@ uintptr_t geted(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t nextop, u
             i64 = F8S;
         if(i64==0 || ((i64>=absmin) && (i64<=absmax)  && !(i64&mask)) || (unscaled && (i64>-256) && (i64<256))) {
             *fixaddress = i64;
-            if(unscaled && (i64>-256) && (i64<256))
+            if(unscaled && i64 && (i64>-256) && (i64<256))
                 *unscaled = 1;
             if((nextop&7)==4) {
                 if (sib_reg!=4) {
@@ -221,7 +223,7 @@ static uintptr_t geted_32(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t
                     }
                 } else {
                     switch(lock) {
-                        case 1: addLockAddress((int32_t)tmp); break;
+                        case 1: addLockAddress((int32_t)tmp); if(fixaddress) *fixaddress=(int32_t)tmp; break;
                         case 2: if(isLockAddress((int32_t)tmp)) *l=1; break;
                     }
                     MOV32w(ret, tmp);
@@ -237,7 +239,7 @@ static uintptr_t geted_32(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t
             uint64_t tmp = F32;
             MOV32w(ret, tmp);
             switch(lock) {
-                case 1: addLockAddress(tmp); break;
+                case 1: addLockAddress(tmp); if(fixaddress) *fixaddress=tmp; break;
                 case 2: if(isLockAddress(tmp)) *l=1; break;
             }
         } else {
@@ -256,11 +258,11 @@ static uintptr_t geted_32(dynarec_arm_t* dyn, uintptr_t addr, int ninst, uint8_t
         }
         if(nextop&0x80)
             i32 = F32S;
-        else 
+        else
             i32 = F8S;
         if(i32==0 || ((i32>=absmin) && (i32<=absmax)  && !(i32&mask)) || (unscaled && (i32>-256) && (i32<256))) {
             *fixaddress = i32;
-            if(unscaled && (i32>-256) && (i32<256))
+            if(unscaled && i32 && (i32>-256) && (i32<256))
                 *unscaled = 1;
             if((nextop&7)==4) {
                 if (sib_reg!=4) {
@@ -567,6 +569,7 @@ void jump_to_next(dynarec_arm_t* dyn, uintptr_t ip, int reg, int ninst)
     MAYUSE(dyn); MAYUSE(ninst);
     MESSAGE(LOG_DUMP, "Jump to next\n");
 
+    SMEND();
     if(reg) {
         if(reg!=xRIP) {
             MOVx_REG(xRIP, reg);
@@ -575,6 +578,10 @@ void jump_to_next(dynarec_arm_t* dyn, uintptr_t ip, int reg, int ninst)
         uintptr_t tbl = getJumpTable64();
         MAYUSE(tbl);
         TABLE64(x3, tbl);
+        #ifdef JMPTABL_SHIFT4
+        UBFXx(x2, xRIP, JMPTABL_START4, JMPTABL_SHIFT4);
+        LDRx_REG_LSL3(x3, x3, x2);
+        #endif
         UBFXx(x2, xRIP, JMPTABL_START3, JMPTABL_SHIFT3);
         LDRx_REG_LSL3(x3, x3, x2);
         UBFXx(x2, xRIP, JMPTABL_START2, JMPTABL_SHIFT2);
@@ -598,7 +605,6 @@ void jump_to_next(dynarec_arm_t* dyn, uintptr_t ip, int reg, int ninst)
     #ifdef HAVE_TRACE
     //MOVx(x3, 15);    no access to PC reg
     #endif
-    SMEND();
     BLR(x2); // save LR...
 }
 
@@ -612,16 +618,21 @@ void ret_to_epilog(dynarec_arm_t* dyn, int ninst, rex_t rex)
     if(box64_dynarec_callret) {
         // pop the actual return address for ARM stack
         LDPx_S7_offset(x2, x6, xSP, 0);
-        CBZx(x6, 5*4);
+        CBZx(x6, 6*4);
         ADDx_U12(xSP, xSP, 16);
         SUBx_REG(x6, x6, xRIP); // is it the right address?
         CBNZx(x6, 2*4);
         BLR(x2);
-        // not the correct return address, regular jump
+        // not the correct return address, regular jump, but purge the stack first, it's unsync now...
+        SUBx_U12(xSP, xSavedSP, 0);
     }
     uintptr_t tbl = getJumpTable64();
     NOTEST(x2);
     MOV64x(x2, tbl);
+    #ifdef JMPTABL_SHIFT4
+    UBFXx(x3, xRIP, JMPTABL_START4, JMPTABL_SHIFT4);
+    LDRx_REG_LSL3(x2, x2, x3);
+    #endif
     UBFXx(x3, xRIP, JMPTABL_START3, JMPTABL_SHIFT3);
     LDRx_REG_LSL3(x2, x2, x3);
     UBFXx(x3, xRIP, JMPTABL_START2, JMPTABL_SHIFT2);
@@ -650,16 +661,21 @@ void retn_to_epilog(dynarec_arm_t* dyn, int ninst, rex_t rex, int n)
     if(box64_dynarec_callret) {
         // pop the actual return address for ARM stack
         LDPx_S7_offset(x2, x6, xSP, 0);
-        CBZx(x6, 5*4);
+        CBZx(x6, 6*4);
         ADDx_U12(xSP, xSP, 16);
         SUBx_REG(x6, x6, xRIP); // is it the right address?
         CBNZx(x6, 2*4);
         BLR(x2);
         // not the correct return address, regular jump
+        SUBx_U12(xSP, xSavedSP, 0);
     }
     uintptr_t tbl = getJumpTable64();
     NOTEST(x2);
     MOV64x(x2, tbl);
+    #ifdef JMPTABL_SHIFT4
+    UBFXx(x3, xRIP, JMPTABL_START4, JMPTABL_SHIFT4);
+    LDRx_REG_LSL3(x2, x2, x3);
+    #endif
     UBFXx(x3, xRIP, JMPTABL_START3, JMPTABL_SHIFT3);
     LDRx_REG_LSL3(x2, x2, x3);
     UBFXx(x3, xRIP, JMPTABL_START2, JMPTABL_SHIFT2);
@@ -677,6 +693,7 @@ void iret_to_epilog(dynarec_arm_t* dyn, int ninst, int is64bits)
     //#warning TODO: is64bits
     MAYUSE(ninst);
     MESSAGE(LOG_DUMP, "IRet to epilog\n");
+    SMEND();
     // POP IP
     NOTEST(x2);
     if(is64bits) {
@@ -711,7 +728,6 @@ void iret_to_epilog(dynarec_arm_t* dyn, int ninst, int is64bits)
     MOVx_REG(xRSP, x3);
     // Ret....
     MOV64x(x2, (uintptr_t)arm64_epilog);  // epilog on purpose, CS might have changed!
-    SMEND();
     BR(x2);
     CLEARIP();
 }
@@ -852,11 +868,11 @@ static void x87_reset(dynarec_arm_t* dyn)
             dyn->n.neoncache[i].v = 0;
 }
 
-void x87_stackcount(dynarec_arm_t* dyn, int ninst, int scratch)
+int x87_stackcount(dynarec_arm_t* dyn, int ninst, int scratch)
 {
     MAYUSE(scratch);
     if(!dyn->n.x87stack)
-        return;
+        return 0;
     if(dyn->n.mmxcount)
         mmx_purgecache(dyn, ninst, 0, scratch);
     MESSAGE(LOG_DUMP, "\tSynch x87 Stackcount (%d)\n", dyn->n.x87stack);
@@ -879,10 +895,45 @@ void x87_stackcount(dynarec_arm_t* dyn, int ninst, int scratch)
     ANDw_mask(scratch, scratch, 0, 2);  //mask=7
     STRw_U12(scratch, xEmu, offsetof(x64emu_t, top));
     // reset x87stack, but not the stack count of neoncache
+    int ret = dyn->n.x87stack;
     dyn->n.x87stack = 0;
     dyn->n.stack_next -= dyn->n.stack;
     dyn->n.stack = 0;
     MESSAGE(LOG_DUMP, "\t------x87 Stackcount\n");
+    return ret;
+}
+
+void x87_unstackcount(dynarec_arm_t* dyn, int ninst, int scratch, int count)
+{
+    MAYUSE(scratch);
+    if(!count)
+        return;
+    if(dyn->n.mmxcount)
+        mmx_purgecache(dyn, ninst, 0, scratch);
+    MESSAGE(LOG_DUMP, "\tUnsynch x87 Stackcount (%d)\n", count);
+    int a = -count;
+    // Add x87stack to emu fpu_stack
+    LDRw_U12(scratch, xEmu, offsetof(x64emu_t, fpu_stack));
+    if(a>0) {
+        ADDw_U12(scratch, scratch, a);
+    } else {
+        SUBw_U12(scratch, scratch, -a);
+    }
+    STRw_U12(scratch, xEmu, offsetof(x64emu_t, fpu_stack));
+    // Sub x87stack to top, with and 7
+    LDRw_U12(scratch, xEmu, offsetof(x64emu_t, top));
+    if(a>0) {
+        SUBw_U12(scratch, scratch, a);
+    } else {
+        ADDw_U12(scratch, scratch, -a);
+    }
+    ANDw_mask(scratch, scratch, 0, 2);  //mask=7
+    STRw_U12(scratch, xEmu, offsetof(x64emu_t, top));
+    // reset x87stack, but not the stack count of neoncache
+    dyn->n.x87stack = count;
+    dyn->n.stack = count;
+    dyn->n.stack_next += dyn->n.stack;
+    MESSAGE(LOG_DUMP, "\t------x87 Unstackcount\n");
 }
 
 int neoncache_st_coherency(dynarec_arm_t* dyn, int ninst, int a, int b)
@@ -1098,7 +1149,11 @@ static void x87_reflectcache(dynarec_arm_t* dyn, int ninst, int s1, int s2, int
         if(dyn->n.x87cache[i]!=-1) {
             ADDw_U12(s3, s2, dyn->n.x87cache[i]);
             ANDw_mask(s3, s3, 0, 2); // mask=7   // (emu->top + i)&7
-            VSTR64_REG_LSL3(dyn->n.x87reg[i], s1, s3);
+            if(neoncache_get_st_f(dyn, ninst, dyn->n.x87cache[i])>=0) {
+                FCVT_D_S(SCRATCH0, dyn->n.x87reg[i]);
+                VSTR64_REG_LSL3(SCRATCH0, s1, s3);
+            } else
+                VSTR64_REG_LSL3(dyn->n.x87reg[i], s1, s3);
         }
 }
 
@@ -1135,6 +1190,8 @@ int x87_get_current_cache(dynarec_arm_t* dyn, int ninst, int st, int t)
             #if STEP == 1
             if(t==NEON_CACHE_ST_D && (dyn->n.neoncache[dyn->n.x87reg[i]].t==NEON_CACHE_ST_F || dyn->n.neoncache[dyn->n.x87reg[i]].t==NEON_CACHE_ST_I64))
                 neoncache_promote_double(dyn, ninst, st);
+            else if(t==NEON_CACHE_ST_I64 && (dyn->n.neoncache[dyn->n.x87reg[i]].t==NEON_CACHE_ST_F))
+                neoncache_promote_double(dyn, ninst, st);
             else if(t==NEON_CACHE_ST_F && (dyn->n.neoncache[dyn->n.x87reg[i]].t==NEON_CACHE_ST_I64))
                 neoncache_promote_double(dyn, ninst, st);
             #endif
@@ -1232,7 +1289,6 @@ void x87_refresh(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st)
 
 void x87_forget(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st)
 {
-    x87_stackcount(dyn, ninst, s1);
     int ret = -1;
     for (int i=0; (i<8) && (ret==-1); ++i)
         if(dyn->n.x87cache[i] == st)
@@ -1240,8 +1296,9 @@ void x87_forget(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st)
     if(ret==-1)    // nothing to do
         return;
     MESSAGE(LOG_DUMP, "\tForget x87 Cache for ST%d\n", st);
+    const int reg = dyn->n.x87reg[ret];
     #if STEP == 1
-    if(dyn->n.neoncache[dyn->n.x87reg[ret]].t==NEON_CACHE_ST_F || dyn->n.neoncache[dyn->n.x87reg[ret]].t==NEON_CACHE_ST_I64)
+    if(dyn->n.neoncache[reg].t==NEON_CACHE_ST_F || dyn->n.neoncache[reg].t==NEON_CACHE_ST_I64)
         neoncache_promote_double(dyn, ninst, st);
     #endif
     // prepare offset to fpu => s1
@@ -1249,15 +1306,28 @@ void x87_forget(dynarec_arm_t* dyn, int ninst, int s1, int s2, int st)
     // Get top
     LDRw_U12(s2, xEmu, offsetof(x64emu_t, top));
     // Update
-    if(st) {
-        ADDw_U12(s2, s2, st);
+    int ast = st - dyn->n.x87stack;
+    if(ast) {
+        if(ast>0) {
+            ADDw_U12(s2, s2, ast);
+        } else {
+            SUBw_U12(s2, s2, -ast);
+        }
         ANDw_mask(s2, s2, 0, 2); //mask=7    // (emu->top + i)&7
     }
-    VSTR64_REG_LSL3(dyn->n.x87reg[ret], s1, s2);
+    if(dyn->n.neoncache[reg].t==NEON_CACHE_ST_F) {
+        FCVT_D_S(31, reg);
+        VSTR64_REG_LSL3(31, s1, s2);
+    } else if(dyn->n.neoncache[reg].t==NEON_CACHE_ST_I64) {
+        SCVTFDD(31, reg);
+        VSTR64_REG_LSL3(31, s1, s2);
+    } else {
+        VSTR64_REG_LSL3(reg, s1, s2);
+    }
     MESSAGE(LOG_DUMP, "\t--------x87 Cache for ST%d\n", st);
     // and forget that cache
-    fpu_free_reg(dyn, dyn->n.x87reg[ret]);
-    dyn->n.neoncache[dyn->n.x87reg[ret]].v = 0;
+    fpu_free_reg(dyn, reg);
+    dyn->n.neoncache[reg].v = 0;
     dyn->n.x87cache[ret] = -1;
     dyn->n.x87reg[ret] = -1;
 }
@@ -1438,7 +1508,7 @@ static void mmx_reflectcache(dynarec_arm_t* dyn, int ninst, int s1)
 {
     for (int i=0; i<8; ++i)
         if(dyn->n.mmxcache[i]!=-1) {
-            VLDR64_U12(dyn->n.mmxcache[i], xEmu, offsetof(x64emu_t, mmx[i]));
+            VSTR64_U12(dyn->n.mmxcache[i], xEmu, offsetof(x64emu_t, mmx[i]));
         }
 }
 
@@ -1541,6 +1611,17 @@ static void sse_reflectcache(dynarec_arm_t* dyn, int ninst, int s1)
         }
 }
 
+void sse_reflect_reg(dynarec_arm_t* dyn, int ninst, int a)
+{
+    if(dyn->n.ssecache[a].v==-1)
+        return;
+    if(dyn->n.neoncache[dyn->n.ssecache[a].reg].t == NEON_CACHE_XMMW) {
+        VSTR128_U12(dyn->n.ssecache[a].reg, xEmu, offsetof(x64emu_t, xmm[a]));
+        /*dyn->n.neoncache[dyn->n.ssecache[a].reg].t = NEON_CACHE_XMMR;
+        dyn->n.ssecache[a].write = 0;*/
+    }
+}
+
 void fpu_pushcache(dynarec_arm_t* dyn, int ninst, int s1, int not07)
 {
     int start = not07?8:0;
@@ -1573,6 +1654,8 @@ void fpu_popcache(dynarec_arm_t* dyn, int ninst, int s1, int not07)
     for (int i=start; i<16; ++i)
         if(dyn->n.ssecache[i].v!=-1) {
             VLDR128_U12(dyn->n.ssecache[i].reg, xEmu, offsetof(x64emu_t, xmm[i]));
+            /*dyn->n.ssecache[i].write = 0;   // OPTIM: it's sync, so not write anymore
+            dyn->n.neoncache[dyn->n.ssecache[i].reg].t = NEON_CACHE_XMMR;*/
         }
     MESSAGE(LOG_DUMP, "\t------- Pop XMM Cache (%d)\n", n);
 }
@@ -1697,7 +1780,7 @@ static void loadCache(dynarec_arm_t* dyn, int ninst, int stack_cnt, int s1, int
             break;
         case NEON_CACHE_MM:
             MESSAGE(LOG_DUMP, "\t  - Loading %s\n", getCacheName(t, n));
-            VLDR64_U12(i, xEmu, offsetof(x64emu_t, mmx[i]));
+            VLDR64_U12(i, xEmu, offsetof(x64emu_t, mmx[n]));
             break;
         case NEON_CACHE_ST_D:
         case NEON_CACHE_ST_F:
@@ -1717,9 +1800,11 @@ static void loadCache(dynarec_arm_t* dyn, int ninst, int stack_cnt, int s1, int
                 ANDw_mask(s3, s3, 0, 2); //mask=7    // (emu->top + i)&7
             }
             *s3_top += a;
-            *s2_val = 0;
-            ADDx_REG_LSL(s2, xEmu, s3, 3);
-            VLDR64_U12(i, s2, offsetof(x64emu_t, x87));
+            if(*s2_val!=1) {
+                *s2_val = 1;
+                ADDx_U12(s2, xEmu, offsetof(x64emu_t, x87));
+            }
+            VLDR64_REG_LSL3(i, s2, s3);
             if(t==NEON_CACHE_ST_F) {
                 FCVT_S_D(i, i);
             }
@@ -1769,14 +1854,16 @@ static void unloadCache(dynarec_arm_t* dyn, int ninst, int stack_cnt, int s1, in
                 ANDw_mask(s3, s3, 0, 2); //mask=7    // (emu->top + i)&7
             }
             *s3_top += a;
-            ADDx_REG_LSL(s2, xEmu, s3, 3);
-            *s2_val = 0;
+            if(*s2_val!=1) {
+                *s2_val = 1;
+                ADDx_U12(s2, xEmu, offsetof(x64emu_t, x87));
+            }
             if(t==NEON_CACHE_ST_F) {
                 FCVT_D_S(i, i);
             } else if (t==NEON_CACHE_ST_I64) {
                 SCVTFDD(i, i);
             }
-            VSTR64_U12(i, s2, offsetof(x64emu_t, x87));
+            VSTR64_REG_LSL3(i, s2, s3);
             break;
         case NEON_CACHE_NONE:
         case NEON_CACHE_SCR:
@@ -1945,7 +2032,7 @@ static void flagsCacheTransform(dynarec_arm_t* dyn, int ninst, int s1)
     if(dyn->f.dfnone)  // flags are fully known, nothing we can do more
         return;
     MESSAGE(LOG_DUMP, "\tFlags fetch ---- ninst=%d -> %d\n", ninst, jmp);
-    int go = 0;
+    int go = (dyn->insts[jmp].f_entry.dfnone && !dyn->f.dfnone)?1:0;
     switch (dyn->insts[jmp].f_entry.pending) {
         case SF_UNKNOWN: break;
         case SF_SET:
@@ -1963,12 +2050,10 @@ static void flagsCacheTransform(dynarec_arm_t* dyn, int ninst, int s1)
             && dyn->f.pending!=SF_SET_PENDING
             && dyn->f.pending!=SF_PENDING)
                 go = 1;
-            else
-                go = (dyn->insts[jmp].f_entry.dfnone  == dyn->f.dfnone)?0:1;
+            else if (dyn->insts[jmp].f_entry.dfnone !=dyn->f.dfnone)
+                go = 1;
             break;
     }
-    if(dyn->insts[jmp].f_entry.dfnone && !dyn->f.dfnone)
-        go = 1;
     if(go) {
         if(dyn->f.pending!=SF_PENDING) {
             LDRw_U12(s1, xEmu, offsetof(x64emu_t, df));
@@ -2016,7 +2101,7 @@ void emit_pf(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4)
     // PF: (((emu->x64emu_parity_tab[(res) / 32] >> ((res) % 32)) & 1) == 0)
     ANDw_mask(s3, s1, 0b011011, 0b000010); // mask=0xE0
     LSRw(s3, s3, 5);
-    MOV64x(s4, (uintptr_t)GetParityTab());
+    TABLE64(s4, (uintptr_t)GetParityTab());
     LDRw_REG_LSL2(s4, s4, s3);
     ANDw_mask(s3, s1, 0, 0b000100); //0x1f
     LSRw_REG(s4, s4, s3);
@@ -2086,4 +2171,4 @@ void fpu_propagate_stack(dynarec_arm_t* dyn, int ninst)
     dyn->n.news = 0;
     dyn->n.stack_push = 0;
     dyn->n.swapped = 0;
-}
+}

src/dynarec/arm64/dynarec_arm64_helper.h

@@ -33,28 +33,31 @@
 #define PKip(a)   *(uint8_t*)(ip+a)
 
 // Strong mem emulation helpers
-// Sequence of Read will trigger a DMB on "first" read if strongmem is 2
-// Sequence of Write will trigger a DMB on "last" write if strongmem is 1
+#define SMREAD_MIN  2
+#define SMWRITE_MIN 1
+// Sequence of Read will trigger a DMB on "first" read if strongmem is >= SMREAD_MIN
+// Sequence of Write will trigger a DMB on "last" write if strongmem is >= 1
+// All Write operation that might use a lock all have a memory barrier if strongmem is >= SMWRITE_MIN
 // Opcode will read
-#define SMREAD()    if(!dyn->smread && box64_dynarec_strongmem>1) {SMDMB();}
+#define SMREAD()    if((dyn->smread==0) && (box64_dynarec_strongmem>SMREAD_MIN)) {SMDMB();} else dyn->smread=1
 // Opcode will read with option forced lock
-#define SMREADLOCK(lock)    if(lock || (!dyn->smread && box64_dynarec_strongmem>1)) {SMDMB();}
+#define SMREADLOCK(lock)    if((lock) || ((dyn->smread==0) && (box64_dynarec_strongmem>SMREAD_MIN))) {SMDMB();}
 // Opcode might read (depend on nextop)
 #define SMMIGHTREAD()   if(!MODREG) {SMREAD();}
 // Opcode has wrote
 #define SMWRITE()   dyn->smwrite=1
 // Opcode has wrote (strongmem>1 only)
-#define SMWRITE2()   if(box64_dynarec_strongmem>1) dyn->smwrite=1
+#define SMWRITE2()   if(box64_dynarec_strongmem>SMREAD_MIN) dyn->smwrite=1
 // Opcode has wrote with option forced lock
-#define SMWRITELOCK(lock)   if(lock) {SMDMB();} else dyn->smwrite=1
+#define SMWRITELOCK(lock)   if(lock || (box64_dynarec_strongmem>SMWRITE_MIN)) {SMDMB();} else dyn->smwrite=1
 // Opcode might have wrote (depend on nextop)
 #define SMMIGHTWRITE()   if(!MODREG) {SMWRITE();}
 // Start of sequence
 #define SMSTART()   SMEND()
 // End of sequence
-#define SMEND()     if(dyn->smwrite && box64_dynarec_strongmem) {DMB_ISH();} dyn->smwrite=0; dyn->smread=0;
+#define SMEND()     if(dyn->smwrite && box64_dynarec_strongmem) {if(box64_dynarec_strongmem){DSB_ISH();}else{DMB_ISH();}} dyn->smwrite=0; dyn->smread=0;
 // Force a Data memory barrier (for LOCK: prefix)
-#define SMDMB()     DMB_ISH(); dyn->smwrite=0; dyn->smread=1
+#define SMDMB()     if(box64_dynarec_strongmem){DSB_ISH();}else{DMB_ISH();} dyn->smwrite=0; dyn->smread=1
 
 //LOCK_* define
 #define LOCK_LOCK   (int*)1
@@ -66,7 +69,7 @@
                     ed = xRAX+(nextop&7)+(rex.b<<3);    \
                     wback = 0;                          \
                 } else {                                \
-                    SMREAD()                            \
+                    SMREAD();                           \
                     addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<(2+rex.w), (1<<(2+rex.w))-1, rex, NULL, 0, D); \
                     LDxw(x1, wback, fixedaddress);      \
                     ed = x1;                            \
@@ -85,7 +88,7 @@
                     wback = 0;                          \
                 } else {                                \
                     SMREAD();                           \
-                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, D); \
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, &fixedaddress, &unscaled, 0xfff<<(3-rex.is32bits), rex.is32bits?3:7, rex, NULL, 0, D); \
                     LDz(x1, wback, fixedaddress);       \
                     ed = x1;                            \
                 }
@@ -422,6 +425,9 @@
         VLD64(a, ed, fixedaddress);                                                                     \
     }
 
+// Get Ex as 64bits, not a quad (warning, x1 get used)
+#define GETEX64(a, w, D)    GETEXSD(a, w, D)
+
 // Get Ex as a single, not a quad (warning, x1 get used)
 #define GETEXSS(a, w, D)                                                                                \
     if(MODREG) {                                                                                        \
@@ -433,6 +439,20 @@
         VLD32(a, ed, fixedaddress);                                                                     \
     }
 
+// Get Ex as 32bits, not a quad (warning, x1 get used)
+#define GETEX32(a, w, D)    GETEXSS(a, w, D)
+
+// Get Ex as 16bits, not a quad (warning, x1 get used)
+#define GETEX16(a, w, D)                                                                                \
+    if(MODREG) {                                                                                        \
+        a = sse_get_reg(dyn, ninst, x1, (nextop&7)+(rex.b<<3), w);                                      \
+    } else {                                                                                            \
+        SMREAD();                                                                                       \
+        a = fpu_get_scratch(dyn);                                                                       \
+        addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<1, 1, rex, NULL, 0, D);   \
+        VLD16(a, ed, fixedaddress);                                                                     \
+    }
+
 // Get GM, might use x1, x2 and x3
 #define GETGM(a)                        \
     gd = ((nextop&0x38)>>3);            \
@@ -464,6 +484,9 @@
     TSTw_mask(xFlags, 0b010110, 0); \
     CNEGx(r, r, cNE)
 
+#define ALIGNED_ATOMICxw ((fixedaddress && !(fixedaddress&((1<<(2+rex.w)-1)))) || box64_dynarec_aligned_atomics)
+#define ALIGNED_ATOMICH ((fixedaddress && !(fixedaddress&1)) || box64_dynarec_aligned_atomics)
+
 // CALL will use x7 for the call address. Return value can be put in ret (unless ret is -1)
 // R0 will not be pushed/popd if ret is -2
 #define CALL(F, ret) call_c(dyn, ninst, F, x7, ret, 1, 0)
@@ -742,8 +765,39 @@
     LDP_REGS(R12, R13);     \
     LDP_REGS(R14, R15)
 
+#define X87_PUSH_OR_FAIL(var, dyn, ninst, scratch, t) \
+    if (dyn->n.stack == +8) {                         \
+        *ok = 0;                                      \
+        break;                                        \
+    }                                                 \
+    var = x87_do_push(dyn, ninst, scratch, t);
+
+#define X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, scratch) \
+    if (dyn->n.stack == +8) {                       \
+        *ok = 0;                                    \
+        break;                                      \
+    }                                               \
+    x87_do_push_empty(dyn, ninst, scratch);
+
+#define X87_POP_OR_FAIL(dyn, ninst, scratch) \
+    if (dyn->n.stack == -8) {                \
+        *ok = 0;                             \
+        break;                               \
+    }                                        \
+    x87_do_pop(dyn, ninst, scratch);
+
 #define SET_DFNONE(S)    if(!dyn->f.dfnone) {STRw_U12(wZR, 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_DF(S, N)        \
+    if((N)!=d_none) {       \
+        MOVZw(S, (N));      \
+        STRw_U12(S, xEmu, offsetof(x64emu_t, df)); \
+        if(dyn->f.pending==SF_PENDING && dyn->insts[ninst].x64.need_after && !(dyn->insts[ninst].x64.need_after&X_PEND)) {  \
+            CALL_(UpdateFlags, -1, 0);              \
+            dyn->f.pending = SF_SET;                \
+            SET_NODF();     \
+        }                   \
+        dyn->f.dfnone=0;    \
+    } else SET_DFNONE(S)
 #define SET_NODF()          dyn->f.dfnone = 0
 #define SET_DFOK()          dyn->f.dfnone = 1
 
@@ -784,7 +838,7 @@
     } else dyn->f.pending = SF_SET
 #endif
 #ifndef JUMP
-#define JUMP(A, C)
+#define JUMP(A, C) SMEND()
 #endif
 #ifndef BARRIER
 #define BARRIER(A)
@@ -792,6 +846,9 @@
 #ifndef BARRIER_NEXT
 #define BARRIER_NEXT(A)
 #endif
+#ifndef SET_HASCALLRET
+#define SET_HASCALLRET()
+#endif
 #define UFLAG_OP1(A) if(dyn->insts[ninst].x64.gen_flags) {STRxw_U12(A, xEmu, offsetof(x64emu_t, op1));}
 #define UFLAG_OP2(A) if(dyn->insts[ninst].x64.gen_flags) {STRxw_U12(A, xEmu, offsetof(x64emu_t, op2));}
 #define UFLAG_OP12(A1, A2) if(dyn->insts[ninst].x64.gen_flags) {STRxw_U12(A1, xEmu, offsetof(x64emu_t, op1));STRxw_U12(A2, xEmu, offsetof(x64emu_t, op2));}
@@ -874,6 +931,8 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 #define dynarec64_65       STEPNAME(dynarec64_65)
 #define dynarec64_66       STEPNAME(dynarec64_66)
 #define dynarec64_67       STEPNAME(dynarec64_67)
+#define dynarec64_67_32    STEPNAME(dynarec64_67_32)
+#define dynarec64_6764_32  STEPNAME(dynarec64_6764_32)
 #define dynarec64_D8       STEPNAME(dynarec64_D8)
 #define dynarec64_D9       STEPNAME(dynarec64_D9)
 #define dynarec64_DA       STEPNAME(dynarec64_DA)
@@ -884,6 +943,8 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 #define dynarec64_DF       STEPNAME(dynarec64_DF)
 #define dynarec64_F0       STEPNAME(dynarec64_F0)
 #define dynarec64_660F     STEPNAME(dynarec64_660F)
+#define dynarec64_66F20F   STEPNAME(dynarec64_66F20F)
+#define dynarec64_66F30F   STEPNAME(dynarec64_66F30F)
 #define dynarec64_6664     STEPNAME(dynarec64_6664)
 #define dynarec64_66F0     STEPNAME(dynarec64_66F0)
 #define dynarec64_F20F     STEPNAME(dynarec64_F20F)
@@ -965,16 +1026,38 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 #define emit_shr32      STEPNAME(emit_shr32)
 #define emit_shr32c     STEPNAME(emit_shr32c)
 #define emit_sar32c     STEPNAME(emit_sar32c)
+#define emit_shl8       STEPNAME(emit_shl8)
+#define emit_shl8c      STEPNAME(emit_shl8c)
+#define emit_shr8       STEPNAME(emit_shr8)
+#define emit_shr8c      STEPNAME(emit_shr8c)
+#define emit_sar8       STEPNAME(emit_sar8)
+#define emit_sar8c      STEPNAME(emit_sar8c)
+#define emit_shl16      STEPNAME(emit_shl16)
+#define emit_shl16c     STEPNAME(emit_shl16c)
+#define emit_shr16      STEPNAME(emit_shr16)
+#define emit_shr16c     STEPNAME(emit_shr16c)
+#define emit_sar16      STEPNAME(emit_sar16)
+#define emit_sar16c     STEPNAME(emit_sar16c)
 #define emit_rol32c     STEPNAME(emit_rol32c)
 #define emit_ror32c     STEPNAME(emit_ror32c)
+#define emit_rol8c      STEPNAME(emit_rol8c)
+#define emit_ror8c      STEPNAME(emit_ror8c)
+#define emit_rol16c     STEPNAME(emit_rol16c)
+#define emit_ror16c     STEPNAME(emit_ror16c)
 #define emit_shrd32c    STEPNAME(emit_shrd32c)
+#define emit_shrd32     STEPNAME(emit_shrd32)
 #define emit_shld32c    STEPNAME(emit_shld32c)
+#define emit_shld32     STEPNAME(emit_shld32)
+#define emit_shrd16c    STEPNAME(emit_shrd16c)
+#define emit_shrd16     STEPNAME(emit_shrd16)
+#define emit_shld16c    STEPNAME(emit_shld16c)
+#define emit_shld16     STEPNAME(emit_shld16)
 
 #define emit_pf         STEPNAME(emit_pf)
 
-#define x87_do_push     STEPNAME(x87_do_push)
-#define x87_do_push_empty STEPNAME(x87_do_push_empty)
-#define x87_do_pop      STEPNAME(x87_do_pop)
+#define x87_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)
@@ -984,6 +1067,7 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 #define x87_forget      STEPNAME(x87_forget)
 #define x87_reget_st    STEPNAME(x87_reget_st)
 #define x87_stackcount  STEPNAME(x87_stackcount)
+#define x87_unstackcount  STEPNAME(x87_unstackcount)
 #define x87_swapreg     STEPNAME(x87_swapreg)
 #define x87_setround    STEPNAME(x87_setround)
 #define x87_restoreround STEPNAME(x87_restoreround)
@@ -994,6 +1078,7 @@ void* arm64_next(x64emu_t* emu, uintptr_t addr);
 #define sse_get_reg_empty STEPNAME(sse_get_reg_empty)
 #define sse_forget_reg   STEPNAME(sse_forget_reg)
 #define sse_purge07cache STEPNAME(sse_purge07cache)
+#define sse_reflect_reg  STEPNAME(sse_reflect_reg)
 
 #define fpu_pushcache   STEPNAME(fpu_pushcache)
 #define fpu_popcache    STEPNAME(fpu_popcache)
@@ -1092,16 +1177,40 @@ void emit_shl32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, i
 void emit_shr32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s3, int s4);
 void emit_shr32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, int s3, int s4);
 void emit_sar32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, int s3, int s4);
+void emit_shl8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
+void emit_shl8c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_shr8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
+void emit_shr8c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_sar8(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
+void emit_sar8c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_shl16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
+void emit_shl16c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_shr16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
+void emit_shr16c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_sar16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s3, int s4);
+void emit_sar16c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
 void emit_rol32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, int s3, int s4);
 void emit_ror32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, uint32_t c, int s3, int s4);
+void emit_rol8c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_ror8c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_rol16c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
+void emit_ror16c(dynarec_arm_t* dyn, int ninst, int s1, uint32_t c, int s3, int s4);
 void emit_shrd32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, uint32_t c, int s3, int s4);
 void emit_shld32c(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, uint32_t c, int s3, int s4);
+void emit_shrd32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s5, int s3, int s4);
+void emit_shld32(dynarec_arm_t* dyn, int ninst, rex_t rex, int s1, int s2, int s5, int s3, int s4);
+void emit_shrd16c(dynarec_arm_t* dyn, int ninst, int s1, int s2, uint32_t c, int s3, int s4);
+void emit_shrd16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s5, int s3, int s4);
+void emit_shld16c(dynarec_arm_t* dyn, int ninst, int s1, int s2, uint32_t c, int s3, int s4);
+void emit_shld16(dynarec_arm_t* dyn, int ninst, int s1, int s2, int s5, int s3, int s4);
 
 void emit_pf(dynarec_arm_t* dyn, int ninst, int s1, int s3, int s4);
 
 // x87 helper
-// cache of the local stack counter, to avoid update at every call
-void x87_stackcount(dynarec_arm_t* dyn, int ninst, int scratch);
+// cache of the local stack counter, to avoid update at every call, return old internal stack counter
+int x87_stackcount(dynarec_arm_t* dyn, int ninst, int scratch);
+// revert local stack counter to previous version (return from x87_stackcount)
+void x87_unstackcount(dynarec_arm_t* dyn, int ninst, int scratch, int count);
 // fpu push. Return the Dd value to be used
 int x87_do_push(dynarec_arm_t* dyn, int ninst, int s1, int t);
 // fpu push. Do not allocate a cache register. Needs a scratch register to do x87stack synch (or 0 to not do it)
@@ -1154,8 +1263,8 @@ int neoncache_st_coherency(dynarec_arm_t* dyn, int ninst, int a, int b);
 #define ST_IS_F(A) (neoncache_get_current_st(dyn, ninst, A)==NEON_CACHE_ST_F)
 #define ST_IS_I64(A) (neoncache_get_current_st(dyn, ninst, A)==NEON_CACHE_ST_I64)
 #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)
+#define X87_ST0     neoncache_no_i64(dyn, ninst, 0, neoncache_get_current_st(dyn, ninst, 0))
+#define X87_ST(A)   neoncache_no_i64(dyn, ninst, A, neoncache_get_current_st(dyn, ninst, A))
 #else
 #define ST_IS_F(A) (neoncache_get_st(dyn, ninst, A)==NEON_CACHE_ST_F)
 #define ST_IS_I64(A) (neoncache_get_st(dyn, ninst, A)==NEON_CACHE_ST_I64)
@@ -1183,7 +1292,8 @@ int sse_get_reg_empty(dynarec_arm_t* dyn, int ninst, int s1, int a);
 void sse_forget_reg(dynarec_arm_t* dyn, int ninst, int a);
 // purge the XMM0..XMM7 cache (before function call)
 void sse_purge07cache(dynarec_arm_t* dyn, int ninst, int s1);
-
+// Push current value to the cache
+void sse_reflect_reg(dynarec_arm_t* dyn, int ninst, int a);
 // common coproc helpers
 // reset the cache
 void fpu_reset(dynarec_arm_t* dyn);
@@ -1208,16 +1318,20 @@ uintptr_t dynarec64_64(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
 //uintptr_t dynarec64_65(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep,int* ok, int* need_epilog);
 uintptr_t dynarec64_66(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
 uintptr_t dynarec64_67(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
+uintptr_t dynarec64_67_32(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
+uintptr_t dynarec64_6764_32(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int seg, int* ok, int* need_epilog);
 uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
 uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
-//uintptr_t dynarec64_DA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
+uintptr_t dynarec64_DA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
 uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
 uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
 uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
 uintptr_t dynarec64_DE(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
 uintptr_t dynarec64_DF(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
 uintptr_t dynarec64_F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
-uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
+uintptr_t dynarec64_66F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
+uintptr_t dynarec64_66F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
+uintptr_t dynarec64_660F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
 uintptr_t dynarec64_6664(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int seg, int* ok, int* need_epilog);
 uintptr_t dynarec64_66F0(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog);
 uintptr_t dynarec64_F20F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog);
@@ -1354,4 +1468,30 @@ uintptr_t dynarec64_F30F(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int n
             opcode = F8;                        \
         }
 
+#define COMP_ZFSF(s1, A)                        \
+    IFX(X_ZF|X_SF) {                            \
+        if(arm64_flagm) {                       \
+            SETF##A(s1);                        \
+            IFX(X_ZF) {                         \
+                CSETw(s3, cEQ);                 \
+                BFIw(xFlags, s3, F_ZF, 1);      \
+            }                                   \
+            IFX(X_SF) {                         \
+                CSETw(s3, cMI);                 \
+                BFIw(xFlags, s3, F_SF, 1);      \
+            }                                   \
+        } else {                                \
+            IFX(X_ZF) {                         \
+                ANDSw_mask(s1, s1, 0, (A)-1);   \
+                CSETw(s3, cEQ);                 \
+                BFIw(xFlags, s3, F_ZF, 1);      \
+            }                                   \
+            IFX(X_SF) {                         \
+                LSRw(s3, s1, (A)-1);            \
+                BFIw(xFlags, s3, F_SF, 1);      \
+            }                                   \
+        }                                       \
+    }
+
+
 #endif //__DYNAREC_ARM64_HELPER_H__

src/dynarec/arm64/dynarec_arm64_pass0.h

@@ -16,9 +16,10 @@
         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 JUMP(A, C)         add_jump(dyn, ninst); add_next(dyn, (uintptr_t)A); SMEND(); 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 BARRIER_NEXT(A) dyn->insts[ninst].x64.barrier_next = A
+#define SET_HASCALLRET()    dyn->insts[ninst].x64.has_callret = 1
 #define NEW_INST \
         ++dyn->size;                            \
         if(dyn->size+3>=dyn->cap) {             \

src/dynarec/arm64/dynarec_arm64_private.h

@@ -102,6 +102,9 @@ typedef struct dynarec_arm_s {
     uintptr_t*          next;       // variable array of "next" jump address
     int                 next_sz;
     int                 next_cap;
+    int*                jmps;       // variable array of jump instructions
+    int                 jmp_sz;
+    int                 jmp_cap;
     int*                predecessor;// single array of all predecessor
     dynablock_t*        dynablock;
     instsize_t*         instsize;
@@ -119,6 +122,8 @@ typedef struct dynarec_arm_s {
 
 void add_next(dynarec_arm_t *dyn, uintptr_t addr);
 uintptr_t get_closest_next(dynarec_arm_t *dyn, uintptr_t addr);
+void add_jump(dynarec_arm_t *dyn, int ninst);
+int get_first_jump(dynarec_arm_t *dyn, int next);
 int is_nops(dynarec_arm_t *dyn, uintptr_t addr, int n);
 int is_instructions(dynarec_arm_t *dyn, uintptr_t addr, int n);
 
@@ -126,7 +131,7 @@ int Table64(dynarec_arm_t *dyn, uint64_t val, int pass);  // add a value to tabl
 
 void CreateJmpNext(void* addr, void* next);
 
-#define GO_TRACE(A, B)      \
+#define GO_TRACE(A, B, s0)  \
     GETIP(addr);            \
     MOVx_REG(x1, xRIP);     \
     STORE_XEMU_CALL(xRIP);  \

src/dynarec/dynablock.c

@@ -41,10 +41,10 @@ dynablock_t* InvalidDynablock(dynablock_t* db, int need_lock)
         if(db->gone)
             return NULL; // already in the process of deletion!
         dynarec_log(LOG_DEBUG, "InvalidDynablock(%p), db->block=%p x64=%p:%p already gone=%d\n", db, db->block, db->x64_addr, db->x64_addr+db->x64_size-1, db->gone);
+        // remove jumptable without waiting
+        setJumpTableDefault64(db->x64_addr);
         if(need_lock)
             mutex_lock(&my_context->mutex_dyndump);
-        // remove jumptable
-        setJumpTableDefault64(db->x64_addr);
         db->done = 0;
         db->gone = 1;
         if(need_lock)
@@ -74,10 +74,10 @@ void FreeDynablock(dynablock_t* db, int need_lock)
         if(db->gone)
             return; // already in the process of deletion!
         dynarec_log(LOG_DEBUG, "FreeDynablock(%p), db->block=%p x64=%p:%p already gone=%d\n", db, db->block, db->x64_addr, db->x64_addr+db->x64_size-1, db->gone);
+        // remove jumptable without waiting
+        setJumpTableDefault64(db->x64_addr);
         if(need_lock)
             mutex_lock(&my_context->mutex_dyndump);
-        // remove jumptable
-        setJumpTableDefault64(db->x64_addr);
         dynarec_log(LOG_DEBUG, " -- FreeDyrecMap(%p, %d)\n", db->actual_block, db->size);
         db->done = 0;
         db->gone = 1;
@@ -171,11 +171,16 @@ dynablock_t *AddNewDynablock(uintptr_t addr)
 }
 
 //TODO: move this to dynrec_arm.c and track allocated structure to avoid memory leak
-static __thread struct __jmp_buf_tag dynarec_jmpbuf;
+static __thread JUMPBUFF dynarec_jmpbuf;
+#ifdef ANDROID
+#define DYN_JMPBUF dynarec_jmpbuf
+#else
+#define DYN_JMPBUF &dynarec_jmpbuf
+#endif
 
 void cancelFillBlock()
 {
-    longjmp(&dynarec_jmpbuf, 1);
+    longjmp(DYN_JMPBUF, 1);
 }
 
 /* 
@@ -208,7 +213,7 @@ static dynablock_t* internalDBGetBlock(x64emu_t* emu, uintptr_t addr, uintptr_t
 
     // fill the block
     block->x64_addr = (void*)addr;
-    if(sigsetjmp(&dynarec_jmpbuf, 1)) {
+    if(sigsetjmp(DYN_JMPBUF, 1)) {
         printf_log(LOG_INFO, "FillBlock at %p triggered a segfault, canceling\n", (void*)addr);
         FreeDynablock(block, 0);
         if(need_lock)

src/dynarec/dynarec.c

@@ -81,16 +81,38 @@ void DynaCall(x64emu_t* emu, uintptr_t addr)
     uint64_t old_rsi = R_RSI;
     uint64_t old_rbp = R_RBP;
     uint64_t old_rip = R_RIP;
+    // save defered flags
+    deferred_flags_t old_df = emu->df;
+    multiuint_t old_op1 = emu->op1;
+    multiuint_t old_op2 = emu->op2;
+    multiuint_t old_res = emu->res;
+    multiuint_t old_op1_sav= emu->op1_sav;
+    multiuint_t old_res_sav= emu->res_sav;
+    deferred_flags_t old_df_sav= emu->df_sav;
+    // uc_link
+    x64_ucontext_t* old_uc_link = emu->uc_link;
+    emu->uc_link = NULL;
+
     PushExit(emu);
     R_RIP = addr;
     emu->df = d_none;
     DynaRun(emu);
     emu->quit = 0;  // reset Quit flags...
     emu->df = d_none;
+    emu->uc_link = old_uc_link;
     if(emu->flags.quitonlongjmp && emu->flags.longjmp) {
         if(emu->flags.quitonlongjmp==1)
             emu->flags.longjmp = 0;   // don't change anything because of the longjmp
     } else {
+        // restore defered flags
+        emu->df = old_df;
+        emu->op1 = old_op1;
+        emu->op2 = old_op2;
+        emu->res = old_res;
+        emu->op1_sav = old_op1_sav;
+        emu->res_sav = old_res_sav;
+        emu->df_sav = old_df_sav;
+        // and the old registers
         R_RBX = old_rbx;
         R_RDI = old_rdi;
         R_RSI = old_rsi;
@@ -100,19 +122,25 @@ void DynaCall(x64emu_t* emu, uintptr_t addr)
     }
 }
 
+int my_setcontext(x64emu_t* emu, void* ucp);
 void DynaRun(x64emu_t* emu)
 {
     // prepare setjump for signal handling
-    struct __jmp_buf_tag jmpbuf[1] = {0};
+    JUMPBUFF jmpbuf[1] = {0};
     int skip = 0;
-    struct __jmp_buf_tag *old_jmpbuf = emu->jmpbuf;
+    JUMPBUFF *old_jmpbuf = emu->jmpbuf;
     emu->flags.jmpbuf_ready = 0;
 
     while(!(emu->quit)) {
         if(!emu->jmpbuf || (emu->flags.need_jmpbuf && emu->jmpbuf!=jmpbuf)) {
             emu->jmpbuf = jmpbuf;
             emu->flags.jmpbuf_ready = 1;
-            if((skip=sigsetjmp(emu->jmpbuf, 1))) {
+            #ifdef ANDROID
+            if((skip=sigsetjmp(*(JUMPBUFF*)emu->jmpbuf, 1))) 
+            #else
+            if((skip=sigsetjmp(emu->jmpbuf, 1))) 
+            #endif
+            {
                 printf_log(LOG_DEBUG, "Setjmp DynaRun, fs=0x%x\n", emu->segs[_FS]);
                 #ifdef DYNAREC
                 if(box64_dynarec_test) {
@@ -134,7 +162,7 @@ void DynaRun(x64emu_t* emu)
         else {
             int is32bits = (emu->segs[_CS]==0x23);
             dynablock_t* block = (skip)?NULL:DBGetBlock(emu, R_RIP, 1, is32bits);
-            if(!block || !block->block || !block->done) {
+            if(!block || !block->block || !block->done || ACCESS_FLAG(F_TF)) {
                 skip = 0;
                 // no block, of block doesn't have DynaRec content (yet, temp is not null)
                 // Use interpreter (should use single instruction step...)
@@ -153,6 +181,10 @@ void DynaRun(x64emu_t* emu)
                 emu->fork = 0;
                 emu = x64emu_fork(emu, forktype);
             }
+            if(emu->quit && emu->uc_link) {
+                emu->quit = 0;
+                my_setcontext(emu, emu->uc_link);
+            }
         }
 #endif
         if(emu->flags.need_jmpbuf)

src/dynarec/dynarec_native.c

@@ -94,6 +94,21 @@ uintptr_t get_closest_next(dynarec_native_t *dyn, uintptr_t addr) {
     }
     return best;
 }
+void add_jump(dynarec_native_t *dyn, int ninst) {
+    // add slots
+    if(dyn->jmp_sz == dyn->jmp_cap) {
+        dyn->jmp_cap += 64;
+        dyn->jmps = (int*)dynaRealloc(dyn->jmps, dyn->jmp_cap*sizeof(int));
+    }
+    dyn->jmps[dyn->jmp_sz++] = ninst;
+}
+int get_first_jump(dynarec_native_t *dyn, int next) {
+    for(int i=0; i<dyn->jmp_sz; ++i)
+        if(dyn->insts[dyn->jmps[i]].x64.jmp == next)
+            return i;
+    return -2;
+}
+
 #define PK(A) (*((uint8_t*)(addr+(A))))
 int is_nops(dynarec_native_t *dyn, uintptr_t addr, int n)
 {
@@ -356,11 +371,11 @@ static int updateNeed(dynarec_native_t* dyn, int ninst, uint8_t need) {
             dyn->insts[ninst].x64.gen_flags |= X_PEND;
         dyn->insts[ninst].x64.need_after = need;
         need = dyn->insts[ninst].x64.need_after&~dyn->insts[ninst].x64.gen_flags;
+
         if(dyn->insts[ninst].x64.may_set)
             need |= dyn->insts[ninst].x64.gen_flags;    // forward the flags
-        // Consume X_PEND if relevant
-        if((need&X_PEND) && (dyn->insts[ninst].x64.set_flags&SF_PENDING))
-            need &=~X_PEND;
+        else if((need&X_PEND) && (dyn->insts[ninst].x64.set_flags&SF_PENDING))
+            need &=~X_PEND;         // Consume X_PEND if relevant
         need |= dyn->insts[ninst].x64.use_flags;
         if(dyn->insts[ninst].x64.need_before == need)
             return ninst - 1;
@@ -392,6 +407,7 @@ void CancelBlock64(int need_lock)
     current_helper = NULL;
     if(helper) {
         dynaFree(helper->next);
+        dynaFree(helper->jmps);
         dynaFree(helper->insts);
         dynaFree(helper->predecessor);
         if(helper->table64 && (helper->table64!=(uint64_t*)helper->tablestart))
@@ -471,10 +487,6 @@ void* FillBlock64(dynablock_t* block, uintptr_t addr, int alternate, int is32bit
     helper.insts = (instruction_native_t*)dynaCalloc(helper.cap, sizeof(instruction_native_t));
     // pass 0, addresses, x64 jump addresses, overall size of the block
     uintptr_t end = native_pass0(&helper, addr, alternate, is32bits);
-    // no need for next anymore
-    dynaFree(helper.next);
-    helper.next_sz = helper.next_cap = 0;
-    helper.next = NULL;
     // basic checks
     if(!helper.size) {
         dynarec_log(LOG_INFO, "Warning, null-sized dynarec block (%p)\n", (void*)addr);
@@ -493,26 +505,57 @@ void* FillBlock64(dynablock_t* block, uintptr_t addr, int alternate, int is32bit
     // compute hash signature
     uint32_t hash = X31_hash_code((void*)addr, end-addr);
     // 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 || j==helper.insts[i].x64.addr) {
-                if(j==helper.insts[i].x64.addr) // if there is a loop on some opcode, make the block "always to tested"
-                    helper.always_test = 1;
-                helper.insts[i].x64.jmp_insts = -1;
-                helper.insts[i].x64.need_after |= X_PEND;
-            } else {
-                // find jump address instruction
-                int k=-1;
-                for(int i2=0; i2<helper.size && k==-1; ++i2) {
-                    if(helper.insts[i2].x64.addr==j)
-                        k=i2;
+    for(int ii=0; ii<helper.jmp_sz; ++ii) {
+        int i = helper.jmps[ii];
+        uintptr_t j = helper.insts[i].x64.jmp;
+        if(j<start || j>=end || j==helper.insts[i].x64.addr) {
+            if(j==helper.insts[i].x64.addr) // if there is a loop on some opcode, make the block "always to tested"
+                helper.always_test = 1;
+            helper.insts[i].x64.jmp_insts = -1;
+            helper.insts[i].x64.need_after |= X_PEND;
+        } else {
+            // find jump address instruction
+            int k=-1;
+            int search = ((j>=helper.insts[0].x64.addr) && j<helper.insts[0].x64.addr+helper.isize)?1:0;
+            int imin = 0;
+            int imax = helper.size-1;
+            int i2 = helper.size/2;
+            // dichotomy search
+            while(search) {
+                if(helper.insts[i2].x64.addr == j) {
+                    k = i2;
+                    search = 0;
+                } else if(helper.insts[i2].x64.addr>j) {
+                    imax = i2;
+                    i2 = (imax+imin)/2;
+                } else {
+                    imin = i2;
+                    i2 = (imax+imin)/2;
+                }
+                if(search && (imax-imin)<2) {
+                    search = 0;
+                    if(helper.insts[imin].x64.addr==j)
+                        k = imin;
+                    else if(helper.insts[imax].x64.addr==j)
+                        k = imax;
                 }
-                if(k!=-1 && !helper.insts[i].barrier_maybe)
-                    helper.insts[k].x64.barrier |= BARRIER_FULL;
-                helper.insts[i].x64.jmp_insts = k;
             }
+            /*for(int i2=0; i2<helper.size && k==-1; ++i2) {
+                if(helper.insts[i2].x64.addr==j)
+                    k=i2;
+            }*/
+            if(k!=-1 && !helper.insts[i].barrier_maybe)
+                helper.insts[k].x64.barrier |= BARRIER_FULL;
+            helper.insts[i].x64.jmp_insts = k;
         }
+    }
+    // no need for next and jmps anymore
+    dynaFree(helper.next);
+    helper.next_sz = helper.next_cap = 0;
+    helper.next = NULL;
+    dynaFree(helper.jmps);
+    helper.jmp_sz = helper.jmp_cap = 0;
+    helper.jmps = NULL;
     // fill predecessors with the jump address
     fillPredecessors(&helper);
 

src/dynarec/dynarec_native_functions.c

@@ -66,12 +66,24 @@ void native_fxtract(x64emu_t* emu)
 }
 void native_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_C1 = (tmp32s&1);
-    emu->sw.f.F87_C3 = ((tmp32s>>1)&1);
-    emu->sw.f.F87_C0 = ((tmp32s>>2)&1);
+    int e0, e1;
+    int64_t ll;
+    frexp(ST0.d, &e0);
+    frexp(ST1.d, &e1);
+    int32_t tmp32s = e0 - e1;
+    if(tmp32s<64)
+    {
+        ll = (int64_t)floor(ST0.d/ST1.d);
+        ST0.d = ST0.d - (ST1.d*ll);
+        emu->sw.f.F87_C2 = 0;
+        emu->sw.f.F87_C1 = (ll&1)?1:0;
+        emu->sw.f.F87_C3 = (ll&2)?1:0;
+        emu->sw.f.F87_C0 = (ll&4)?1:0;
+    } else {
+        ll = (int64_t)(floor((ST0.d/ST1.d))/exp2(tmp32s - 32));
+        ST0.d = ST0.d - ST1.d*ll*exp2(tmp32s - 32);
+        emu->sw.f.F87_C2 = 1;
+    }
 }
 void native_fyl2xp1(x64emu_t* emu)
 {
@@ -174,6 +186,17 @@ void native_singlestep(x64emu_t* emu)
     emit_signal(emu, SIGTRAP, (void*)R_RIP, 1);
 }
 
+void native_int3(x64emu_t* emu)
+{
+    emit_signal(emu, SIGTRAP, (void*)R_RIP, 128);
+}
+
+void native_div0(x64emu_t* emu)
+{
+    emu->test.test = 0;
+    emit_div0(emu,  (void*)R_RIP, 0);
+}
+
 void native_fsave(x64emu_t* emu, uint8_t* ed)
 {
     fpu_savenv(emu, (char*)ed, 0);
@@ -200,13 +223,24 @@ void native_frstor(x64emu_t* emu, uint8_t* ed)
 
 void native_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_C1 = (tmp32s&1);
-    emu->sw.f.F87_C3 = ((tmp32s>>1)&1);
-    emu->sw.f.F87_C0 = ((tmp32s>>2)&1);
+    int e0, e1;
+    int64_t ll;
+    frexp(ST0.d, &e0);
+    frexp(ST1.d, &e1);
+    int32_t tmp32s = e0 - e1;
+    if(tmp32s<64)
+    {
+        ll = (int64_t)round(ST0.d/ST1.d);
+        ST0.d = ST0.d - (ST1.d*ll);
+        emu->sw.f.F87_C2 = 0;
+        emu->sw.f.F87_C1 = (ll&1)?1:0;
+        emu->sw.f.F87_C3 = (ll&2)?1:0;
+        emu->sw.f.F87_C0 = (ll&4)?1:0;
+    } else {
+        ll = (int64_t)(trunc((ST0.d/ST1.d))/exp2(tmp32s - 32));
+        ST0.d = ST0.d - ST1.d*ll*exp2(tmp32s - 32);
+        emu->sw.f.F87_C2 = 1;
+    }
 }
 
 static uint8_t ff_mult(uint8_t a, uint8_t b)
@@ -371,9 +405,8 @@ static int flagsCacheNeedsTransform(dynarec_native_t* dyn, int ninst) {
         return 0;
     if(dyn->insts[ninst].f_exit.dfnone)  // flags are fully known, nothing we can do more
         return 0;
-/*    if((dyn->f.pending!=SF_SET)
-    && (dyn->f.pending!=SF_SET_PENDING)) {
-        if(dyn->f.pending!=SF_PENDING) {*/
+    if(dyn->insts[jmp].f_entry.dfnone && !dyn->insts[ninst].f_exit.dfnone)
+        return 1;
     switch (dyn->insts[jmp].f_entry.pending) {
         case SF_UNKNOWN: return 0;
         case SF_SET:
@@ -382,22 +415,14 @@ static int flagsCacheNeedsTransform(dynarec_native_t* dyn, int ninst) {
             else
                 return 0;
         case SF_SET_PENDING:
-            if(dyn->insts[ninst].f_exit.pending!=SF_SET
-            && dyn->insts[ninst].f_exit.pending!=SF_SET_PENDING
-            && dyn->insts[ninst].f_exit.pending!=SF_PENDING)
-                return 1;
-            else
+            if(dyn->insts[ninst].f_exit.pending==SF_SET_PENDING)
                 return 0;
+            return 1;
         case SF_PENDING:
-            if(dyn->insts[ninst].f_exit.pending!=SF_SET
-            && dyn->insts[ninst].f_exit.pending!=SF_SET_PENDING
-            && dyn->insts[ninst].f_exit.pending!=SF_PENDING)
-                return 1;
-            else
-                return (dyn->insts[jmp].f_entry.dfnone  == dyn->insts[ninst].f_exit.dfnone)?0:1;
+            if(dyn->insts[ninst].f_exit.pending==SF_PENDING || dyn->insts[ninst].f_exit.pending==SF_SET_PENDING)
+                return 0;
+            return (dyn->insts[jmp].f_entry.dfnone  == dyn->insts[ninst].f_exit.dfnone)?0:1;
     }
-    if(dyn->insts[jmp].f_entry.dfnone && !dyn->insts[ninst].f_exit.dfnone)
-        return 1;
     return 0;
 }
 
@@ -451,6 +476,12 @@ uintptr_t fakeed(dynarec_native_t* dyn, uintptr_t addr, int ninst, uint8_t nexto
     }
     return addr;
 }
+// return Ib on a mod/rm opcode without emiting anything
+uint8_t geted_ib(dynarec_native_t* dyn, uintptr_t addr, int ninst, uint8_t nextop)
+{
+    addr = fakeed(dyn, addr, ninst, nextop);
+    return F8;
+}
 #undef F8
 
 int isNativeCall(dynarec_native_t* dyn, uintptr_t addr, uintptr_t* calladdress, int* retn)

src/dynarec/dynarec_native_functions.h

@@ -46,7 +46,9 @@ void native_clflush(x64emu_t* emu, void* p);
 void native_ud(x64emu_t* emu);
 void native_priv(x64emu_t* emu);
 void native_singlestep(x64emu_t* emu);
+void native_int3(x64emu_t* emu);
 void native_int(x64emu_t* emu, int num);
+void native_div0(x64emu_t* emu);
 
 // Caches transformation (for loops) // Specific, need to be written par backend
 int CacheNeedsTransform(dynarec_native_t* dyn, int i1);
@@ -57,6 +59,8 @@ int getNominalPred(dynarec_native_t* dyn, int ninst);
 
 // Do the GETED, but don't emit anything...
 uintptr_t fakeed(dynarec_native_t* dyn, uintptr_t addr, int ninst, uint8_t nextop);
+// return Ib on a mod/rm opcode without emiting anything
+uint8_t geted_ib(dynarec_native_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_native_t* dyn, uintptr_t addr, uintptr_t* calladdress, int* retn);

src/dynarec/dynarec_native_pass.c

@@ -77,6 +77,11 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
             dyn->last_ip = 0;   // reset IP if some jump are coming here
         fpu_propagate_stack(dyn, ninst);
         NEW_INST;
+        #if STEP == 0
+        if(ninst && dyn->insts[ninst-1].x64.barrier_next) {
+            BARRIER(dyn->insts[ninst-1].x64.barrier_next);
+        }
+        #endif
         if(!ninst) {
             GOTEST(x1, x2);
         }
@@ -88,17 +93,17 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
         if(box64_dynarec_test) {
             MESSAGE(LOG_DUMP, "TEST INIT ----\n");
             fpu_reflectcache(dyn, ninst, x1, x2, x3);
-            GO_TRACE(x64test_init, 1);
+            GO_TRACE(x64test_init, 1, x5);
             fpu_unreflectcache(dyn, ninst, x1, x2, x3);
             MESSAGE(LOG_DUMP, "----------\n");
         }
 #ifdef HAVE_TRACE
         else if(my_context->dec && box64_dynarec_trace) {
-        if((trace_end == 0) 
+        if((trace_end == 0)
             || ((ip >= trace_start) && (ip < trace_end)))  {
                 MESSAGE(LOG_DUMP, "TRACE ----\n");
                 fpu_reflectcache(dyn, ninst, x1, x2, x3);
-                GO_TRACE(PrintTrace, 1);
+                GO_TRACE(PrintTrace, 1, x5);
                 fpu_unreflectcache(dyn, ninst, x1, x2, x3);
                 MESSAGE(LOG_DUMP, "----------\n");
             }
@@ -107,12 +112,13 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
 
         rep = 0;
         uint8_t pk = PK(0);
-        while((pk==0xF2) || (pk==0xF3)) {
-            rep = pk-0xF1;
-            ++addr;
-            pk = PK(0);
-        }
-        while(pk==0x3E || pk==0x26) {   //Branch Taken Hint ignored, same for ES: prefix
+        while((pk==0xF2) || (pk==0xF3) || (pk==0x3E) || (pk==0x26)) {
+            switch(pk) {
+                case 0xF2: rep = 1; break;
+                case 0xF3: rep = 2; break;
+                case 0x3E:
+                case 0x26: /* ignored */ break;
+            }
             ++addr;
             pk = PK(0);
         }
@@ -177,7 +183,9 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
         if(dyn->forward) {
             if(dyn->forward_to == addr && !need_epilog) {
                 // we made it!
-                if(box64_dynarec_dump) dynarec_log(LOG_NONE, "Forward extend block for %d bytes %p -> %p\n", dyn->forward_to-dyn->forward, (void*)dyn->forward, (void*)dyn->forward_to);
+                if(box64_dynarec_dump) dynarec_log(LOG_NONE, "Forward extend block for %d bytes %s%p -> %p\n", dyn->forward_to-dyn->forward, dyn->insts[dyn->forward_ninst].x64.has_callret?"(opt. call) ":"", (void*)dyn->forward, (void*)dyn->forward_to);
+                if(dyn->insts[dyn->forward_ninst].x64.has_callret && !dyn->insts[dyn->forward_ninst].x64.has_next)
+                    dyn->insts[dyn->forward_ninst].x64.has_next = 1;  // this block actually continue
                 dyn->forward = 0;
                 dyn->forward_to = 0;
                 dyn->forward_size = 0;
@@ -200,18 +208,18 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
             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)<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);
+                    if(dyn->insts[ninst].x64.has_callret && !dyn->insts[ninst].x64.has_next) {
+                        dyn->insts[ninst].x64.has_next = 1;  // this block actually continue
+                    } else {
+                        // need to find back that instruction to copy the caches, as previous version cannot be used anymore
+                        // and pred table is not ready yet
+                        reset_n = get_first_jump(dyn, next);
+                    }
+                    if(box64_dynarec_dump) dynarec_log(LOG_NONE, "Extend block %p, %s%p -> %p (ninst=%d, jump from %d)\n", dyn, dyn->insts[ninst].x64.has_callret?"(opt. call) ":"", (void*)addr, (void*)next, ninst, dyn->insts[ninst].x64.has_callret?ninst:reset_n);
                 } else if(next && (next-addr)<box64_dynarec_forward && (getProtection(next)&PROT_READ)/*box64_dynarec_bigblock>=stopblock*/) {
                     if(!((box64_dynarec_bigblock<stopblock) && !isJumpTableDefault64((void*)next))) {
                         if(dyn->forward) {
@@ -232,7 +240,7 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
             }
         #endif
         if(ok<0)  {
-            ok = 0; need_epilog=1; 
+            ok = 0; need_epilog=1;
             #if STEP == 0
             if(ninst) {
                 --ninst;
@@ -256,7 +264,7 @@ uintptr_t native_pass(dynarec_native_t* dyn, uintptr_t addr, int alternate, int
         }
         ++ninst;
         #if STEP == 0
-        if(ok && (((box64_dynarec_bigblock<stopblock) && !isJumpTableDefault64((void*)addr)) 
+        if(ok && (((box64_dynarec_bigblock<stopblock) && !isJumpTableDefault64((void*)addr))
             || (addr>=box64_nodynarec_start && addr<box64_nodynarec_end)))
         #else
         if(ok && (ninst==dyn->size))

src/dynarec/dynarec_private.h

@@ -33,6 +33,8 @@ typedef struct instruction_x64_s {
     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     barrier_next;   // next instruction needs a barrier
+    uint8_t     has_callret;    // this instruction have an optimised call setup
     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

src/dynarec/native_lock.h

@@ -28,6 +28,10 @@
 #define native_lock_incif0(A)               arm64_lock_incif0(A)
 #define native_lock_decifnot0(A)            arm64_lock_decifnot0(A)
 #define native_lock_store(A, B)             arm64_lock_store(A, B)
+#define native_lock_store_dd(A, B)          arm64_lock_store_dd(A, B)
+#define native_lock_get_b(A)                arm64_lock_get_b(A)
+#define native_lock_get_d(A)                arm64_lock_get_d(A)
+#define native_lock_get_dd(A)               arm64_lock_get_dd(A)
 
 #elif defined(RV64)
 #include "rv64/rv64_lock.h"
@@ -50,6 +54,7 @@
 #define native_lock_incif0(A)               rv64_lock_incif0(A)
 #define native_lock_decifnot0(A)            rv64_lock_decifnot0(A)
 #define native_lock_store(A, B)             rv64_lock_store(A, B)
+#define native_lock_store_dd(A, B)          rv64_lock_store_dd(A, B)
 #define native_lock_cas_d(A, B, C)          rv64_lock_cas_d(A, B, C)
 #define native_lock_cas_dd(A, B, C)         rv64_lock_cas_dd(A, B, C)
 
@@ -68,6 +73,9 @@
 // there is no atomic move on 16bytes, so faking it
 #define native_lock_read_dq(A, B, C)        *A=tmpcas=((uint64_t*)(C))[0]; *B=((uint64_t*)(C))[1];
 #define native_lock_write_dq(A, B, C)       rv64_lock_cas_dq(C, A, tmpcas, B);
+#define native_lock_get_b(A)                rv64_lock_get_b(A)
+#define native_lock_get_d(A)                rv64_lock_get_d(A)
+#define native_lock_get_dd(A)               rv64_lock_get_dd(A)
 
 #else
 #error Unsupported architecture

src/dynarec/rv64/dynarec_rv64_00_0.c

@@ -213,7 +213,7 @@ uintptr_t dynarec64_00_0(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             u8 = F8;
             ANDI(x1, xRAX, 0xff);
             emit_adc8c(dyn, ninst, x1, u8, x3, x4, x5, x6);
-            ANDI(xRAX, xRAX, -256);
+            ANDI(xRAX, xRAX, ~0xff);
             OR(xRAX, xRAX, x1);
             break;
         case 0x15:

src/dynarec/rv64/dynarec_rv64_00_1.c

@@ -172,7 +172,10 @@ uintptr_t dynarec64_00_1(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             addr = dynarec64_66(dyn, addr, ip, ninst, rex, rep, ok, need_epilog);
             break;
         case 0x67:
-            addr = dynarec64_67(dyn, addr, ip, ninst, rex, rep, ok, need_epilog);
+            if(rex.is32bits)
+                addr = dynarec64_67_32(dyn, addr, ip, ninst, rex, rep, ok, need_epilog);
+            else
+                addr = dynarec64_67(dyn, addr, ip, ninst, rex, rep, ok, need_epilog);
             break;
         case 0x68:
             INST_NAME("PUSH Id");
@@ -260,6 +263,31 @@ uintptr_t dynarec64_00_1(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             }
             break;
 
+        case 0x6C:
+        case 0x6D:
+            INST_NAME(opcode == 0x6C ? "INSB" : "INSD");
+            SETFLAGS(X_ALL, SF_SET); // Hack to set flags in "don't care" state
+            GETIP(ip);
+            STORE_XEMU_CALL(x3);
+            CALL(native_priv, -1);
+            LOAD_XEMU_CALL();
+            jump_to_epilog(dyn, 0, xRIP, ninst);
+            *need_epilog = 0;
+            *ok = 0;
+            break;
+        case 0x6E:
+        case 0x6F:
+            INST_NAME(opcode == 0x6C ? "OUTSB" : "OUTSD");
+            SETFLAGS(X_ALL, SF_SET); // Hack to set flags in "don't care" state
+            GETIP(ip);
+            STORE_XEMU_CALL(x3);
+            CALL(native_priv, -1);
+            LOAD_XEMU_CALL();
+            jump_to_epilog(dyn, 0, xRIP, ninst);
+            *need_epilog = 0;
+            *ok = 0;
+            break;
+
         #define GO(GETFLAGS, NO, YES, F)                                \
             READFLAGS(F);                                               \
             i8 = F8S;                                                   \

src/dynarec/rv64/dynarec_rv64_00_2.c

@@ -418,7 +418,7 @@ uintptr_t dynarec64_00_2(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
                 LHU(x1, ed, fixedaddress);
                 ed = x1;
             }
-            SW(ed, xEmu, offsetof(x64emu_t, segs[(nextop&0x38)>>3]));
+            SH(ed, xEmu, offsetof(x64emu_t, segs[(nextop&0x38)>>3]));
             SW(xZR, xEmu, offsetof(x64emu_t, segs_serial[(nextop&0x38)>>3]));
             break;
         case 0x8F:
@@ -499,15 +499,24 @@ uintptr_t dynarec64_00_2(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             if(box64_wine) {    // should this be done all the time?
                 ANDI(x1, xFlags, 1 << F_TF);
                 CBZ_NEXT(x1);
-                MOV64x(xRIP, addr);
-                STORE_XEMU_CALL();
-                CALL(native_singlestep, -1);
-                ANDI(xFlags, xFlags, ~(1 << F_TF));
+                // go to epilog, TF should trigger at end of next opcode, so using Interpretor only
+                jump_to_epilog(dyn, addr, 0, ninst);
             }
             break;
+        case 0x9E:
+            INST_NAME("SAHF");
+            SETFLAGS(X_CF | X_PF | X_AF | X_ZF | X_SF, SF_SUBSET);
+            ADDI(x1, xZR, ~0b11010101);
+            AND(xFlags, xFlags, x1);
+            NOT(x1, x1);
+            SRLI(x2, xRAX, 8);
+            AND(x1, x1, x2);
+            OR(xFlags, xFlags, x1);
+            SET_DFNONE();
+            break;
         case 0x9F:
             INST_NAME("LAHF");
-            READFLAGS(X_CF|X_PF|X_AF|X_ZF|X_SF);
+            READFLAGS(X_CF | X_PF | X_AF | X_ZF | X_SF);
             ANDI(x1, xFlags, 0xFF);
             SLLI(x1, x1, 8);
             MOV64x(x2, 0xffffffffffff00ffLL);
@@ -558,7 +567,7 @@ uintptr_t dynarec64_00_2(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
                 B_NEXT_nocond;
                 MARK2;  // Part with DF==1
                 LBU(x1, xRSI, 0);
-                LBU(x1, xRDI, 0);
+                SB(x1, xRDI, 0);
                 SUBI(xRSI, xRSI, 1);
                 SUBI(xRDI, xRDI, 1);
                 SUBI(xRCX, xRCX, 1);
@@ -646,6 +655,48 @@ uintptr_t dynarec64_00_2(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
                 break;
             }
             break;
+        case 0xA7:
+            switch (rep) {
+                case 1:
+                case 2:
+                    if (rep == 1) { INST_NAME("REPNZ CMPSD"); } else { INST_NAME("REPZ CMPSD"); }
+                    MAYSETFLAGS();
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    CBZ_NEXT(xRCX);
+                    ANDI(x1, xFlags, 1 << F_DF);
+                    BNEZ_MARK2(x1);
+                    MARK; // Part with DF==0
+                    LDxw(x1, xRSI, 0);
+                    ADDI(xRSI, xRSI, rex.w ? 8 : 4);
+                    LDxw(x2, xRDI, 0);
+                    ADDI(xRDI, xRDI, rex.w ? 8 : 4);
+                    SUBI(xRCX, xRCX, 1);
+                    if (rep == 1) { BEQ_MARK3(x1, x2); } else { BNE_MARK3(x1, x2); }
+                    BNEZ_MARK(xRCX);
+                    B_MARK3_nocond;
+                    MARK2; // Part with DF==1
+                    LDxw(x1, xRSI, 0);
+                    SUBI(xRSI, xRSI, rex.w ? 8 : 4);
+                    LDxw(x2, xRDI, 0);
+                    SUBI(xRDI, xRDI, rex.w ? 8 : 4);
+                    SUBI(xRCX, xRCX, 1);
+                    if (rep == 1) { BEQ_MARK3(x1, x2); } else { BNE_MARK3(x1, x2); }
+                    BNEZ_MARK2(xRCX);
+                    MARK3; // end
+                    emit_cmp32(dyn, ninst, rex, x1, x2, x3, x4, x5, x6);
+                    break;
+                default:
+                    INST_NAME("CMPSD");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    GETDIR(x3, x1, rex.w ? 8 : 4);
+                    LDxw(x1, xRSI, 0);
+                    LDxw(x2, xRDI, 0);
+                    ADD(xRSI, xRSI, x3);
+                    ADD(xRDI, xRDI, x3);
+                    emit_cmp32(dyn, ninst, rex, x1, x2, x3, x4, x5, x6);
+                    break;
+            }
+            break;
         case 0xA8:
             INST_NAME("TEST AL, Ib");
             SETFLAGS(X_ALL, SF_SET_PENDING);
@@ -711,6 +762,28 @@ uintptr_t dynarec64_00_2(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
                 ADD(xRDI, xRDI, x3);
             }
             break;
+        case 0xAC:
+            if (rep) {
+                DEFAULT;
+            } else {
+                INST_NAME("LODSB");
+                GETDIR(x1, x2, 1);
+                LBU(x2, xRSI, 0);
+                ADD(xRSI, xRSI, x1);
+                ANDI(xRAX, xRAX, ~0xff);
+                OR(xRAX, xRAX, x2);
+            }
+            break;
+        case 0xAD:
+            if (rep) {
+                DEFAULT;
+            } else {
+                INST_NAME("LODSD");
+                GETDIR(x1, x2, rex.w ? 8 : 4);
+                LDxw(xRAX, xRSI, 0);
+                ADD(xRSI, xRSI, x1);
+            }
+            break;
         case 0xAE:
             switch (rep) {
             case 1:
@@ -780,10 +853,9 @@ uintptr_t dynarec64_00_2(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
                 INST_NAME("SCASD");
                 SETFLAGS(X_ALL, SF_SET_PENDING);
                 GETDIR(x3, x1, rex.w?8:4);
-                AND(x1, xRAX, xMASK);
                 LDxw(x2, xRDI, 0);
                 ADD(xRDI, xRDI, x3);
-                emit_cmp32(dyn, ninst, rex, x1, x2, x3, x4, x5, x6);
+                emit_cmp32(dyn, ninst, rex, xRAX, x2, x3, x4, x5, x6);
                 break;
             }
             break;

src/dynarec/rv64/dynarec_rv64_00_3.c

@@ -319,7 +319,7 @@ uintptr_t dynarec64_00_3(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
                         addr+=8+8;
                     } else {
                         GETIP(ip+1); // read the 0xCC
-                        STORE_XEMU_CALL();
+                        STORE_XEMU_CALL(x3);
                         ADDI(x1, xEmu, (uint32_t)offsetof(x64emu_t, ip)); // setup addr as &emu->ip
                         CALL_S(x64Int3, -1);
                         LOAD_XEMU_CALL();
@@ -360,7 +360,7 @@ uintptr_t dynarec64_00_3(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
                 NOTEST(x1);
                 SMEND();
                 GETIP(addr);
-                STORE_XEMU_CALL();
+                STORE_XEMU_CALL(x3);
                 CALL_S(x86Syscall, -1);
                 LOAD_XEMU_CALL();
                 TABLE64(x3, addr); // expected return address
@@ -368,13 +368,13 @@ uintptr_t dynarec64_00_3(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
                 LW(x1, xEmu, offsetof(x64emu_t, quit));
                 BEQ_NEXT(x1, xZR);
                 MARK;
-                LOAD_XEMU_REM();
+                LOAD_XEMU_REM(x3);
                 jump_to_epilog(dyn, 0, xRIP, ninst);
             } else {
                 INST_NAME("INT n");
                 SETFLAGS(X_ALL, SF_SET); // Hack to set flags in "don't care" state
                 GETIP(ip);
-                STORE_XEMU_CALL();
+                STORE_XEMU_CALL(x3);
                 CALL(native_priv, -1);
                 LOAD_XEMU_CALL();
                 jump_to_epilog(dyn, 0, xRIP, ninst);
@@ -503,6 +503,16 @@ uintptr_t dynarec64_00_3(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
                     WBACK;
                     if(!wback && !rex.w) ZEROUP(ed);
                     break;
+                case 2:
+                    INST_NAME("RCL Ed, 1");
+                    MESSAGE("LOG_DUMP", "Need optimization\n");
+                    READFLAGS(X_CF);
+                    SETFLAGS(X_OF|X_CF, SF_SET);
+                    MOV32w(x2, 1);
+                    GETEDW(x4, x1, 0);
+                    CALL_(rex.w ? ((void*)rcl64) : ((void*)rcl32), ed, x4);
+                    WBACK;
+                    break;
                 case 3:
                     INST_NAME("RCR Ed, 1");
                     MESSAGE(LOG_DUMP, "Need Optimization\n");
@@ -604,7 +614,9 @@ uintptr_t dynarec64_00_3(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
         case 0xD9:
             addr = dynarec64_D9(dyn, addr, ip, ninst, rex, rep, ok, need_epilog);
             break;
-
+        case 0xDA:
+            addr = dynarec64_DA(dyn, addr, ip, ninst, rex, rep, ok, need_epilog);
+            break;
         case 0xDB:
             addr = dynarec64_DB(dyn, addr, ip, ninst, rex, rep, ok, need_epilog);
             break;
@@ -723,7 +735,7 @@ uintptr_t dynarec64_00_3(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
                         dyn->last_ip = addr;
                     } else {
                         GETIP_(dyn->insts[ninst].natcall); // read the 0xCC already
-                        STORE_XEMU_CALL();
+                        STORE_XEMU_CALL(x3);
                         ADDI(x1, xEmu, (uint32_t)offsetof(x64emu_t, ip)); // setup addr as &emu->ip
                         CALL_S(x64Int3, -1);
                         LOAD_XEMU_CALL();
@@ -755,7 +767,6 @@ uintptr_t dynarec64_00_3(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
                         SETFLAGS(X_ALL, SF_SET);    // Hack to set flags to "dont'care" state
                     }
                     // regular call
-                    //BARRIER_NEXT(1);
                     if(box64_dynarec_callret && box64_dynarec_bigblock>1) {
                         BARRIER(BARRIER_FULL);
                     } else {
@@ -1076,7 +1087,11 @@ uintptr_t dynarec64_00_3(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             break;
         case 0xFC:
             INST_NAME("CLD");
-            ANDI(xFlags, xFlags, ~(1<<F_CF));
+            ANDI(xFlags, xFlags, ~(1 << F_DF));
+            break;
+        case 0xFD:
+            INST_NAME("STD");
+            ORI(xFlags, xFlags, 1 << F_DF);
             break;
         case 0xFE:
             nextop = F8;
@@ -1168,12 +1183,12 @@ uintptr_t dynarec64_00_3(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
                         INST_NAME("JMP FAR Ed");
                         READFLAGS(X_PEND);
                         BARRIER(BARRIER_FLOAT);
-                        SMREAD()
+                        SMREAD();
                         addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, NULL, 0, 0);
                         LDxw(x1, wback, 0);
                         ed = x1;
                         LHU(x3, wback, rex.w?8:4);
-                        SW(x3, xEmu, offsetof(x64emu_t, segs[_CS]));
+                        SH(x3, xEmu, offsetof(x64emu_t, segs[_CS]));
                         SW(xZR, xEmu, offsetof(x64emu_t, segs_serial[_CS]));
                         jump_to_epilog(dyn, 0, ed, ninst);
                         *need_epilog = 0;

src/dynarec/rv64/dynarec_rv64_64.c

@@ -393,7 +393,44 @@ uintptr_t dynarec64_64(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                 LDxw(gd, x4, fixedaddress);
             }
             break;
-
+        case 0x8E:
+            INST_NAME("MOV Seg,Ew");
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            u8 = (nextop&0x38) >> 3;
+            if ((nextop&0xC0) == 0xC0) {
+                ed = xRAX+(nextop&7) + (rex.b<<3);
+            } else {
+                SMREAD();
+                addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, NULL, 0, 0);
+                ADD(x4, wback, x4);
+                LHU(x1, x4, 0);
+                ed = x1;
+            }
+            SH(ed, xEmu, offsetof(x64emu_t, segs[u8]));
+            SW(xZR, xEmu, offsetof(x64emu_t, segs_serial[u8]));
+            break;
+        case 0x8F:
+            INST_NAME("POP FS:Ed");
+            grab_segdata(dyn, addr, ninst, x4, seg);
+            nextop = F8;
+            if (MODREG) {
+                POP1z(xRAX + (nextop & 7) + (rex.b << 3));
+            } else {
+                POP1z(x3); // so this can handle POP [ESP] and maybe some variant too
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, NULL, 0, 0);
+                if (ed == xRSP) {
+                    ADD(x4, ed, x4);
+                    SDz(x3, x4, 0);
+                } else {
+                    // complicated to just allow a segfault that can be recovered correctly
+                    ADDIz(xRSP, xRSP, rex.is32bits ? -4 : -8);
+                    ADD(x4, ed, x4);
+                    SDz(x3, x4, 0);
+                    ADDIz(xRSP, xRSP, rex.is32bits ? 4 : 8);
+                }
+            }
+            break;
         case 0xA1:
             INST_NAME("MOV EAX,FS:Od");
             grab_segdata(dyn, addr, ninst, x4, seg);
@@ -489,6 +526,20 @@ uintptr_t dynarec64_64(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                 SMWRITE2();
             }
             break;
+
+        case 0xFF:
+            nextop = F8;
+            grab_segdata(dyn, addr, ninst, x6, seg);
+            switch((nextop>>3)&7) {
+                case 6: // Push Ed
+                    INST_NAME("PUSH Ed");
+                    GETEDOz(x6, 0, x3);
+                    PUSH1z(ed);
+                    break;
+                default:
+                    DEFAULT;
+            }
+            break;
         default:
             DEFAULT;
     }

src/dynarec/rv64/dynarec_rv64_66.c

@@ -15,6 +15,7 @@
 #include "emu/x64run_private.h"
 #include "x64trace.h"
 #include "dynarec_native.h"
+#include "custommem.h"
 
 #include "rv64_printer.h"
 #include "dynarec_rv64_private.h"
@@ -115,7 +116,11 @@ uintptr_t dynarec64_66(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             OR(xRAX, xRAX, x1);
             break;
         case 0x0F:
-            addr = dynarec64_660F(dyn, addr, ip, ninst, rex, ok, need_epilog);
+            switch(rep) {
+                case 0: addr = dynarec64_660F(dyn, addr, ip, ninst, rex, ok, need_epilog); break;
+                case 1: addr = dynarec64_66F20F(dyn, addr, ip, ninst, rex, ok, need_epilog); break;
+                case 2: addr = dynarec64_66F30F(dyn, addr, ip, ninst, rex, ok, need_epilog); break;
+            }
             break;
         case 0x11:
             INST_NAME("ADC Ew, Gw");
@@ -539,7 +544,7 @@ uintptr_t dynarec64_66(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                     OR(ed, ed, x2);
                 }
             } else {
-                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, &lock, 0, 0);
+                addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, &lock, 1, 0);
                 SH(gd, ed, fixedaddress);
                 SMWRITELOCK(lock);
             }
@@ -602,6 +607,29 @@ uintptr_t dynarec64_66(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             AND(x1, x1, x2);
             OR(xRAX, xRAX, x1);
             break;
+        case 0xA1:
+            INST_NAME("MOV EAX,Od");
+            if (rex.is32bits) u64 = F32; else u64 = F64;
+            MOV64z(x1, u64);
+            if (isLockAddress(u64)) lock = 1; else lock = 0;
+            SMREADLOCK(lock);
+            LHU(x2, x1, 0);
+            LUI(x3, 0xffff0);
+            AND(xRAX, xRAX, x3);
+            OR(xRAX, xRAX, x2);
+            break;
+        case 0xA3:
+            INST_NAME("MOV Od,EAX");
+            if(rex.is32bits)
+                u64 = F32;
+            else
+                u64 = F64;
+            MOV64z(x1, u64);
+            if(isLockAddress(u64)) lock=1; else lock = 0;
+            SH(xRAX, x1, 0);
+            SMWRITELOCK(lock);
+            break;
+
         case 0xA5:
             if(rep) {
                 INST_NAME("REP MOVSW");
@@ -633,6 +661,48 @@ uintptr_t dynarec64_66(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                 ADD(xRDI, xRDI, x3);
             }
             break;
+        case 0xA7:
+            switch (rep) {
+                case 1:
+                case 2:
+                    if (rep == 1) { INST_NAME("REPNZ CMPSW"); } else { INST_NAME("REPZ CMPSW"); }
+                    MAYSETFLAGS();
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    CBZ_NEXT(xRCX);
+                    ANDI(x1, xFlags, 1 << F_DF);
+                    BNEZ_MARK2(x1);
+                    MARK; // Part with DF==0
+                    LHU(x1, xRSI, 0);
+                    ADDI(xRSI, xRSI, 2);
+                    LHU(x2, xRDI, 0);
+                    ADDI(xRDI, xRDI, 2);
+                    SUBI(xRCX, xRCX, 1);
+                    if (rep == 1) { BEQ_MARK3(x1, x2); } else { BNE_MARK3(x1, x2); }
+                    BNEZ_MARK(xRCX);
+                    B_MARK3_nocond;
+                    MARK2; // Part with DF==1
+                    LHU(x1, xRSI, 0);
+                    SUBI(xRSI, xRSI, 2);
+                    LHU(x2, xRDI, 0);
+                    SUBI(xRDI, xRDI, 2);
+                    SUBI(xRCX, xRCX, 1);
+                    if (rep == 1) { BEQ_MARK3(x1, x2); } else { BNE_MARK3(x1, x2); }
+                    BNEZ_MARK2(xRCX);
+                    MARK3; // end
+                    emit_cmp16(dyn, ninst, x1, x2, x3, x4, x5, x6);
+                    break;
+                default:
+                    INST_NAME("CMPSW");
+                    SETFLAGS(X_ALL, SF_SET_PENDING);
+                    GETDIR(x3, x1, 2);
+                    LHU(x1, xRSI, 0);
+                    LHU(x2, xRDI, 0);
+                    ADD(xRSI, xRSI, x3);
+                    ADD(xRDI, xRDI, x3);
+                    emit_cmp16(dyn, ninst, x1, x2, x3, x4, x5, x6);
+                    break;
+            }
+            break;
         case 0xA9:
             INST_NAME("TEST AX,Iw");
             SETFLAGS(X_ALL, SF_SET_PENDING);
@@ -667,7 +737,19 @@ uintptr_t dynarec64_66(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                 ADD(xRDI, xRDI, x3);
             }
             break;
-
+        case 0xAD:
+            if (rep) {
+                DEFAULT;
+            } else {
+                INST_NAME("LODSW");
+                GETDIR(x1, x2, 2);
+                LHU(x2, xRSI, 0);
+                ADD(xRSI, xRSI, x1);
+                LUI(x1, 0xffff0);
+                AND(xRAX, xRAX, x1);
+                OR(xRAX, xRAX, x2);
+            }
+            break;
         case 0xAF:
             switch (rep) {
                 case 1:

src/dynarec/rv64/dynarec_rv64_66f20f.c

@@ -0,0 +1,58 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.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 "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynarec_native.h"
+
+#include "rv64_printer.h"
+#include "dynarec_rv64_private.h"
+#include "dynarec_rv64_functions.h"
+#include "dynarec_rv64_helper.h"
+
+uintptr_t dynarec64_66F20F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
+{
+    (void)ip;
+    (void)need_epilog;
+
+    uint8_t opcode = F8;
+    uint8_t nextop, u8, s8;
+    int32_t i32;
+    uint8_t gd, ed;
+    uint8_t wback, wb1, wb2, gback;
+    uint8_t eb1, eb2;
+    int64_t j64;
+    uint64_t tmp64u, tmp64u2;
+    int v0, v1;
+    int q0, q1;
+    int d0, d1;
+    int64_t fixedaddress, gdoffset;
+    int unscaled;
+
+    MAYUSE(d0);
+    MAYUSE(d1);
+    MAYUSE(q0);
+    MAYUSE(q1);
+    MAYUSE(eb1);
+    MAYUSE(eb2);
+    MAYUSE(j64);
+
+    static const int8_t round_round[] = { RD_RNE, RD_RDN, RD_RUP, RD_RTZ };
+
+    switch (opcode) {
+        default:
+            DEFAULT;
+    }
+    return addr;
+}

src/dynarec/rv64/dynarec_rv64_66f30f.c

@@ -0,0 +1,58 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.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 "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynarec_native.h"
+
+#include "rv64_printer.h"
+#include "dynarec_rv64_private.h"
+#include "dynarec_rv64_functions.h"
+#include "dynarec_rv64_helper.h"
+
+uintptr_t dynarec64_66F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
+{
+    (void)ip;
+    (void)need_epilog;
+
+    uint8_t opcode = F8;
+    uint8_t nextop, u8, s8;
+    int32_t i32;
+    uint8_t gd, ed;
+    uint8_t wback, wb1, wb2, gback;
+    uint8_t eb1, eb2;
+    int64_t j64;
+    uint64_t tmp64u, tmp64u2;
+    int v0, v1;
+    int q0, q1;
+    int d0, d1;
+    int64_t fixedaddress, gdoffset;
+    int unscaled;
+
+    MAYUSE(d0);
+    MAYUSE(d1);
+    MAYUSE(q0);
+    MAYUSE(q1);
+    MAYUSE(eb1);
+    MAYUSE(eb2);
+    MAYUSE(j64);
+
+    static const int8_t round_round[] = { RD_RNE, RD_RDN, RD_RUP, RD_RTZ };
+
+    switch (opcode) {
+        default:
+            DEFAULT;
+    }
+    return addr;
+}

src/dynarec/rv64/dynarec_rv64_67.c

@@ -180,6 +180,44 @@ uintptr_t dynarec64_67(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                             DEFAULT;
                     }
                     break;
+                case 0xB6:
+                    INST_NAME("MOVXZ Gd, Eb");
+                    nextop = F8;
+                    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
+                        }
+                        if (eb2) {
+                            SRLI(gd, eb1, 8);
+                            ANDI(gd, gd, 0xff);
+                        } else {
+                            ANDI(gd, eb1, 0xff);
+                        }
+                    } else {
+                        SMREAD();
+                        addr = geted32(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, &lock, 1, 0);
+                        LBU(gd, ed, fixedaddress);
+                    }
+                    break;
+                case 0xB7:
+                    INST_NAME("MOVZX Gd, Ew");
+                    nextop = F8;
+                    GETGD;
+                    if(MODREG) {
+                        ed = xRAX + (nextop & 7) + (rex.b << 3);
+                        ZEXTH(gd, ed);
+                    } else {
+                        SMREAD();
+                        addr = geted32(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, &lock, 1, 0);
+                        LHU(gd, ed, fixedaddress);
+                    }
+                    break;
                 default:
                     DEFAULT;
             }
@@ -410,6 +448,54 @@ uintptr_t dynarec64_67(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             } else
                 emit_cmp32_0(dyn, ninst, rex, xRAX, x3, x4);
             break;
+        case 0x63:
+            INST_NAME("MOVSXD Gd, Ed");
+            nextop = F8;
+            GETGD;
+            if (rex.w) {
+                if (MODREG) { // reg <= reg
+                    ADDIW(gd, xRAX + (nextop & 7) + (rex.b << 3), 0);
+                } else { // mem <= reg
+                    SMREAD();
+                    addr = geted32(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, &lock, 1, 0);
+                    LW(gd, ed, fixedaddress);
+                }
+            } else {
+                if (MODREG) { // reg <= reg
+                    AND(gd, xRAX + (nextop & 7) + (rex.b << 3), xMASK);
+                } else { // mem <= reg
+                    SMREAD();
+                    addr = geted32(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, &lock, 1, 0);
+                    LWU(gd, ed, fixedaddress);
+                }
+            }
+            break;
+        case 0x66:
+            opcode = F8;
+            switch (opcode) {
+                case 0x89:
+                    INST_NAME("MOV Ew, Gw");
+                    nextop = F8;
+                    GETGD; // don't need GETGW here
+                    if (MODREG) {
+                        ed = xRAX + (nextop & 7) + (rex.b << 3);
+                        if (ed != gd) {
+                            LUI(x1, 0xffff0);
+                            AND(ed, ed, x1);
+                            ZEXTH(x2, gd);
+                            OR(ed, ed, x2);
+                        }
+                    } else {
+                        addr = geted32(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, &lock, 1, 0);
+                        SH(gd, ed, fixedaddress);
+                        SMWRITELOCK(lock);
+                    }
+                    break;
+
+                default:
+                    DEFAULT;
+            }
+            break;
 
         case 0x81:
         case 0x83:
@@ -567,6 +653,23 @@ uintptr_t dynarec64_67(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                 }
             }
             break;
+
+        case 0xC7:
+            INST_NAME("MOV Ed, Id");
+            nextop = F8;
+            if(MODREG) { // reg <= i32
+                i64 = F32S;
+                ed = xRAX + (nextop & 7) + (rex.b << 3);
+                MOV64xw(ed, i64);
+            } else {     // mem <= i32
+                addr = geted32(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, &lock, 1, 0);
+                i64 = F32S;
+                MOV64xw(x3, i64);
+                SDxw(x3, ed, fixedaddress);
+                SMWRITELOCK(lock);
+            }
+            break;
+
         #define GO(Z)                                                   \
             BARRIER(BARRIER_MAYBE);                                     \
             JUMP(addr+i8, 1);                                           \

src/dynarec/rv64/dynarec_rv64_67_32.c

@@ -0,0 +1,59 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.h>
+#include <assert.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 "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "dynarec_native.h"
+
+#include "rv64_printer.h"
+#include "dynarec_rv64_private.h"
+#include "dynarec_rv64_helper.h"
+#include "dynarec_rv64_functions.h"
+
+uintptr_t dynarec64_67_32(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+{
+    (void)ip; (void)need_epilog;
+
+    uint8_t opcode = F8;
+    uint8_t nextop;
+    uint8_t gd, ed, wback, wb, wb1, wb2, gb1, gb2, eb1, eb2;
+    int64_t fixedaddress;
+    int unscaled;
+    int8_t  i8;
+    uint8_t u8;
+    int32_t i32;
+    int64_t j64, i64;
+    int cacheupd = 0;
+    int lock;
+    int v0, v1, s0;
+
+    if(!rex.is32bits) {
+        // should do a different file
+        DEFAULT;
+        return addr;
+    }
+
+    rep = 0;
+    while((opcode==0xF2) || (opcode==0xF3)) {
+        rep = opcode-0xF1;
+        opcode = F8;
+    }
+
+    switch(opcode) {
+        default:
+            DEFAULT;
+    }
+    return addr;
+}

src/dynarec/rv64/dynarec_rv64_d8.c

@@ -43,56 +43,45 @@ uintptr_t dynarec64_D8(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
 
     switch(nextop) {
         case 0xC0 ... 0xC7:
-        
+            INST_NAME("FADD ST0, STx");
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
+            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7, X87_COMBINE(0, nextop&7));
+            if(ST_IS_F(0)) {
+                FADDS(v1, v1, v2);
+            } else {
+                FADDD(v1, v1, v2);
+            }
+            break;
         case 0xC8 ... 0xCF:
-
+            INST_NAME("FMUL ST0, STx");
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop & 7));
+            v2 = x87_get_st(dyn, ninst, x1, x2, nextop & 7, X87_COMBINE(0, nextop & 7));
+            if (ST_IS_F(0)) {
+                FMULS(v1, v1, v2);
+            } else {
+                FMULD(v1, v1, v2);
+            }
+            break;
         case 0xD0 ... 0xD7:
-
+            INST_NAME("FCOM ST0, STx");
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop & 7));
+            v2 = x87_get_st(dyn, ninst, x1, x2, nextop & 7, X87_COMBINE(0, nextop & 7));
+            if (ST_IS_F(0)) {
+                FCOMS(v1, v2, x1, x2, x3, x4, x5);
+            } else {
+                FCOMD(v1, v2, x1, x2, x3, x4, x5);
+            }
+            break;
         case 0xD8 ... 0xDF:
             INST_NAME("FCOMP ST0, STx");
             v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7, X87_COMBINE(0, nextop&7));
-            LHU(x3, xEmu, offsetof(x64emu_t, sw));
-            MOV32w(x1, 0b1110100011111111); // mask off c0,c1,c2,c3
-            AND(x3, x3, x1);
+            v2 = x87_get_st(dyn, ninst, x1, x2, nextop & 7, X87_COMBINE(0, nextop & 7));
             if(ST_IS_F(0)) {
-                FEQS(x5, v1, v1);
-                FEQS(x4, v2, v2);
-                AND(x5, x5, x4);
-                BEQZ(x5, 24); // undefined/NaN
-                FEQS(x5, v1, v2);
-                BNEZ(x5, 28); // equal
-                FLTS(x3, v1, v2); // x3 = (v1<v2)?1:0
-                SLLI(x1, x3, 8);
-                J(20); // end
-                // undefined/NaN
-                LUI(x1, 1);
-                ADDI(x1, x1, 0b010100000000);
-                J(8); // end
-                // equal
-                LUI(x1, 1);
-                // end
+                FCOMS(v1, v2, x1, x2, x3, x4, x5);
             } else {
-                FEQD(x5, v1, v1);
-                FEQD(x4, v2, v2);
-                AND(x5, x5, x4);
-                BEQZ(x5, 24); // undefined/NaN
-                FEQD(x5, v1, v2);
-                BNEZ(x5, 28); // equal
-                FLTD(x3, v1, v2); // x3 = (v1<v2)?1:0
-                SLLI(x1, x3, 8);
-                J(20); // end
-                // undefined/NaN
-                LUI(x1, 1);
-                ADDI(x1, x1, 0b010100000000);
-                J(8); // end
-                // equal
-                LUI(x1, 1);
-                // end
+                FCOMD(v1, v2, x1, x2, x3, x4, x5);
             }
-            OR(x3, x3, x1);
-            SH(x3, xEmu, offsetof(x64emu_t, sw));
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xE0 ... 0xE7:
             INST_NAME("FSUB ST0, STx");
@@ -105,11 +94,19 @@ uintptr_t dynarec64_D8(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             }
             break;
         case 0xE8 ... 0xEF:
-
+            INST_NAME("FSUBR ST0, STx");
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop & 7));
+            v2 = x87_get_st(dyn, ninst, x1, x2, nextop & 7, X87_COMBINE(0, nextop & 7));
+            if (ST_IS_F(0)) {
+                FSUBS(v1, v2, v1);
+            } else {
+                FSUBD(v1, v2, v1);
+            }
+            break;
         case 0xF0 ... 0xF7:
             INST_NAME("FDIV ST0, STx");
-            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
-            v2 = x87_get_st(dyn, ninst, x1, x2, nextop&7, X87_COMBINE(0, nextop&7));
+            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 {
@@ -117,7 +114,14 @@ uintptr_t dynarec64_D8(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             }
             break;
         case 0xF8 ... 0xFF:
-            DEFAULT;
+            INST_NAME("FDIVR ST0, STx");
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop & 7));
+            v2 = x87_get_st(dyn, ninst, x1, x2, nextop & 7, X87_COMBINE(0, nextop & 7));
+            if (ST_IS_F(0)) {
+                FDIVS(v1, v2, v1);
+            } else {
+                FDIVD(v1, v2, v1);
+            }
             break;
 
         default:
@@ -148,6 +152,33 @@ uintptr_t dynarec64_D8(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                         FMULD(v1, v1, s0);
                     }
                     break;
+                case 2:
+                    INST_NAME("FCOM ST0, float[ED]");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+                    s0 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, NULL, 1, 0);
+                    FLW(s0, ed, fixedaddress);
+                    if (ST_IS_F(0)) {
+                        FCOMS(v1, s0, x1, x6, x3, x4, x5);
+                    } else {
+                        FCVTDS(s0, s0);
+                        FCOMD(v1, s0, x1, x6, x3, x4, x5);
+                    }
+                    break;
+                case 3:
+                    INST_NAME("FCOMP ST0, float[ED]");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+                    s0 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, NULL, 1, 0);
+                    FLW(s0, ed, fixedaddress);
+                    if (ST_IS_F(0)) {
+                        FCOMS(v1, s0, x1, x6, x3, x4, x5);
+                    } else {
+                        FCVTDS(s0, s0);
+                        FCOMD(v1, s0, x1, x6, x3, x4, x5);
+                    }
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
+                    break;
                 case 4:
                     INST_NAME("FSUB ST0, float[ED]");
                     v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
@@ -161,6 +192,19 @@ uintptr_t dynarec64_D8(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                         FSUBD(v1, v1, s0);
                     }
                     break;
+                case 5:
+                    INST_NAME("FSUBR ST0, float[ED]");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+                    s0 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, NULL, 1, 0);
+                    FLW(s0, ed, fixedaddress);
+                    if (ST_IS_F(0)) {
+                        FSUBS(v1, s0, v1);
+                    } else {
+                        FCVTDS(s0, s0);
+                        FSUBD(v1, s0, v1);
+                    }
+                    break;
                 case 6:
                     INST_NAME("FDIV ST0, float[ED]");
                     v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
@@ -174,8 +218,19 @@ uintptr_t dynarec64_D8(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                         FDIVD(v1, v1, s0);
                     }
                     break;
-                default:
-                    DEFAULT;
+                case 7:
+                    INST_NAME("FDIVR ST0, float[ED]");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
+                    s0 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, NULL, 1, 0);
+                    FLW(s0, ed, fixedaddress);
+                    if (ST_IS_F(0)) {
+                        FDIVS(v1, s0, v1);
+                    } else {
+                        FCVTDS(s0, s0);
+                        FDIVD(v1, s0, v1);
+                    }
+                    break;
             }
     }
     return addr;

src/dynarec/rv64/dynarec_rv64_d9.c

@@ -54,7 +54,7 @@ uintptr_t dynarec64_D9(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
         case 0xC6:
         case 0xC7:
             INST_NAME("FLD STx");
-            v2 = x87_do_push(dyn, ninst, x1, X87_ST(nextop&7));
+            X87_PUSH_OR_FAIL(v2, 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)) {
                 FMVS(v2, v1);
@@ -87,7 +87,7 @@ uintptr_t dynarec64_D9(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
 
         case 0xD8:
             INST_NAME("FSTPNCE ST0, ST0");
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xD9:
         case 0xDA:
@@ -101,7 +101,7 @@ uintptr_t dynarec64_D9(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             x87_get_st_empty(dyn, ninst, x1, x2, nextop&7, X87_ST(nextop&7));
             x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
             x87_swapreg(dyn, ninst, x1, x2, 0, nextop&7);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xE0:
             INST_NAME("FCHS");
@@ -135,7 +135,7 @@ uintptr_t dynarec64_D9(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
 
         case 0xE8:
             INST_NAME("FLD1");
-            v1 = x87_do_push(dyn, ninst, x1, EXT_CACHE_ST_F);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, EXT_CACHE_ST_F);
             if(ST_IS_F(0)) {
                 MOV32w(x1, 0x3f800000);
                 FMVWX(v1, x1);
@@ -146,32 +146,32 @@ uintptr_t dynarec64_D9(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             break;
         case 0xE9:
             INST_NAME("FLDL2T");
-            v1 = x87_do_push(dyn, ninst, x1, EXT_CACHE_ST_D);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, EXT_CACHE_ST_D);
             FTABLE64(v1, L2T);
             break;
-        case 0xEA:     
+        case 0xEA:
             INST_NAME("FLDL2E");
-            v1 = x87_do_push(dyn, ninst, x1, EXT_CACHE_ST_D);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, EXT_CACHE_ST_D);
             FTABLE64(v1, L2E);
             break;
         case 0xEB:
             INST_NAME("FLDPI");
-            v1 = x87_do_push(dyn, ninst, x1, EXT_CACHE_ST_D);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, EXT_CACHE_ST_D);
             FTABLE64(v1, PI);
             break;
         case 0xEC:
             INST_NAME("FLDLG2");
-            v1 = x87_do_push(dyn, ninst, x1, EXT_CACHE_ST_D);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, EXT_CACHE_ST_D);
             FTABLE64(v1, LG2);
             break;
         case 0xED:
             INST_NAME("FLDLN2");
-            v1 = x87_do_push(dyn, ninst, x1, EXT_CACHE_ST_D);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, EXT_CACHE_ST_D);
             FTABLE64(v1, LN2);
             break;
         case 0xEE:
             INST_NAME("FLDZ");
-            v1 = x87_do_push(dyn, ninst, x1, EXT_CACHE_ST_F);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, EXT_CACHE_ST_F);
             if(ST_IS_F(0)) {
                 FMVWX(v1, xZR);
             } else {
@@ -191,14 +191,14 @@ uintptr_t dynarec64_D9(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             x87_forget(dyn, ninst, x1, x2, 0);
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(native_fyl2x, -1);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xF2:
             INST_NAME("FPTAN");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
             x87_forget(dyn, ninst, x1, x2, 0);
             CALL(native_ftan, -1);
-            v1 = x87_do_push(dyn, ninst, x1, EXT_CACHE_ST_F);
+            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, EXT_CACHE_ST_F);
             if(ST_IS_F(0)) {
                 MOV32w(x1, 0x3f800000);
                 FMVWX(v1, x1);
@@ -213,12 +213,12 @@ uintptr_t dynarec64_D9(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             x87_forget(dyn, ninst, x1, x2, 0);
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(native_fpatan, -1);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xF4:
             INST_NAME("FXTRACT");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
-            x87_do_push_empty(dyn, ninst, 0);
+            X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, 0);
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(native_fxtract, -1);
             break;
@@ -258,7 +258,7 @@ uintptr_t dynarec64_D9(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             x87_forget(dyn, ninst, x1, x2, 0);
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(native_fyl2xp1, -1);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xFA:
             INST_NAME("FSQRT");
@@ -272,7 +272,7 @@ uintptr_t dynarec64_D9(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
         case 0xFB:
             INST_NAME("FSINCOS");
             MESSAGE(LOG_DUMP, "Need Optimization\n");
-            x87_do_push_empty(dyn, ninst, 0);
+            X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, 0);
             x87_forget(dyn, ninst, x1, x2, 1);
             CALL(native_fsincos, -1);
             break;
@@ -349,12 +349,12 @@ uintptr_t dynarec64_D9(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
         case 0xEF:
             DEFAULT;
             break;
-             
+
         default:
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("FLD ST0, float[ED]");
-                    v1 = x87_do_push(dyn, ninst, x1, box64_dynarec_x87double?EXT_CACHE_ST_D:EXT_CACHE_ST_F);
+                    X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, box64_dynarec_x87double?EXT_CACHE_ST_D:EXT_CACHE_ST_F);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, NULL, 1, 0);
                     FLW(v1, ed, fixedaddress);
                     if(!ST_IS_F(0)) {
@@ -381,7 +381,7 @@ uintptr_t dynarec64_D9(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                         FCVTSD(v1, v1);
                     }
                     FSW(v1, ed, fixedaddress);
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 case 4:
                     INST_NAME("FLDENV Ed");

src/dynarec/rv64/dynarec_rv64_da.c

@@ -0,0 +1,164 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <errno.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 "callback.h"
+#include "emu/x64run_private.h"
+#include "x64trace.h"
+#include "emu/x87emu_private.h"
+#include "dynarec_native.h"
+
+#include "rv64_printer.h"
+#include "dynarec_rv64_private.h"
+#include "dynarec_rv64_helper.h"
+#include "dynarec_rv64_functions.h"
+
+
+uintptr_t dynarec64_DA(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int rep, int* ok, int* need_epilog)
+{
+    uint8_t nextop = F8;
+    int64_t j64;
+    uint8_t ed;
+    uint8_t wback;
+    int v1, v2;
+    int d0;
+    int s0;
+    int64_t fixedaddress;
+    int unscaled;
+
+    MAYUSE(s0);
+    MAYUSE(d0);
+    MAYUSE(v2);
+    MAYUSE(v1);
+    MAYUSE(ed);
+    MAYUSE(j64);
+
+    switch (nextop) {
+        case 0xC0:
+        case 0xC1:
+        case 0xC2:
+        case 0xC3:
+        case 0xC4:
+        case 0xC5:
+        case 0xC6:
+        case 0xC7:
+            INST_NAME("FCMOVB ST0, STx");
+            READFLAGS(X_CF);
+            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));
+            ANDI(x1, xFlags, 1 << F_CF);
+            CBZ_NEXT(x1);
+            if (ST_IS_F(0))
+                FMVS(v1, v2);
+            else
+                FMVD(v1, v2);
+            break;
+        case 0xC8:
+        case 0xC9:
+        case 0xCA:
+        case 0xCB:
+        case 0xCC:
+        case 0xCD:
+        case 0xCE:
+        case 0xCF:
+            INST_NAME("FCMOVE ST0, STx");
+            READFLAGS(X_ZF);
+            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));
+            ANDI(x1, xFlags, 1 << F_ZF);
+            CBZ_NEXT(x1);
+            if (ST_IS_F(0))
+                FMVS(v1, v2);
+            else
+                FMVD(v1, v2);
+            break;
+        case 0xD0:
+        case 0xD1:
+        case 0xD2:
+        case 0xD3:
+        case 0xD4:
+        case 0xD5:
+        case 0xD6:
+        case 0xD7:
+            INST_NAME("FCMOVBE ST0, STx");
+            READFLAGS(X_CF | X_ZF);
+            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));
+            ANDI(x1, xFlags, (1 << F_CF) | (1 << F_ZF));
+            CBZ_NEXT(x1);
+            if (ST_IS_F(0))
+                FMVS(v1, v2);
+            else
+                FMVD(v1, v2);
+            break;
+        case 0xD8:
+        case 0xD9:
+        case 0xDA:
+        case 0xDB:
+        case 0xDC:
+        case 0xDD:
+        case 0xDE:
+        case 0xDF:
+            INST_NAME("FCMOVU ST0, STx");
+            READFLAGS(X_PF);
+            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));
+            ANDI(x1, xFlags, (1 << F_PF));
+            CBZ_NEXT(x1);
+            if (ST_IS_F(0))
+                FMVS(v1, v2);
+            else
+                FMVD(v1, v2);
+            break;
+        case 0xE9:
+            INST_NAME("FUCOMPP ST0, ST1");
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop & 7));
+            v2 = x87_get_st(dyn, ninst, x1, x2, 1, X87_COMBINE(0, nextop & 7));
+            if (ST_IS_F(0)) {
+                FCOMS(v1, v2, x1, x2, x3, x4, x5);
+            } else {
+                FCOMD(v1, v2, x1, x2, x3, x4, x5);
+            }
+            X87_POP_OR_FAIL(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
+            break;
+        case 0xE4:
+        case 0xF0:
+        case 0xF1:
+        case 0xF4:
+        case 0xF5:
+        case 0xF6:
+        case 0xF7:
+        case 0xF8:
+        case 0xF9:
+        case 0xFD:
+            DEFAULT;
+            break;
+
+        default:
+            switch ((nextop >> 3) & 7) {
+                case 6:
+                    INST_NAME("FIDIV ST0, Ed");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, EXT_CACHE_ST_D);
+                    v2 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, NULL, 1, 0);
+                    LW(x1, ed, fixedaddress);
+                    FCVTDW(v2, x1, RD_RNE);
+                    FDIVD(v1, v1, v2);
+                    break;
+                default:
+                    DEFAULT;
+                    break;
+            }
+    }
+    return addr;
+}

src/dynarec/rv64/dynarec_rv64_db.c

@@ -153,7 +153,7 @@ uintptr_t dynarec64_DB(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             SET_DFNONE();
             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));
-            IFX(F_ZF | F_PF | F_CF) {
+            IFX(X_ZF | X_PF | X_CF) {
                 if(ST_IS_F(0)) {
                     FEQS(x5, v1, v1);
                     FEQS(x4, v2, v2);
@@ -189,7 +189,7 @@ uintptr_t dynarec64_DB(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                 }
             }
             break;
-        case 0xF0:  
+        case 0xF0:
         case 0xF1:
         case 0xF2:
         case 0xF3:
@@ -213,7 +213,7 @@ uintptr_t dynarec64_DB(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("FILD ST0, Ed");
-                    v1 = x87_do_push(dyn, ninst, x1, EXT_CACHE_ST_D);
+                    X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, EXT_CACHE_ST_D);
                     addr = geted(dyn, addr, ninst, nextop, &ed, x2, x1, &fixedaddress, rex, NULL, 1, 0);
                     LW(x1, ed, fixedaddress);
                     FCVTDW(v1, x1, RD_RNE);    // i32 -> double
@@ -245,7 +245,7 @@ uintptr_t dynarec64_DB(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                     }
                     MARK2;
                     SW(x4, wback, fixedaddress);
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 case 5:
                     INST_NAME("FLD tbyte");
@@ -268,13 +268,13 @@ uintptr_t dynarec64_DB(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                         SH(x6, ed, fixedaddress+8);
                     } else {
                         if(box64_x87_no80bits) {
-                            v1 = x87_do_push(dyn, ninst, x1, EXT_CACHE_ST_D);
+                            X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, EXT_CACHE_ST_D);
                             FLD(v1, ed, fixedaddress);
                         } else {
                             if(ed!=x1) {
-                                MV(x1, ed);
+                                ADDI(x1, ed, fixedaddress);
                             }
-                            x87_do_push_empty(dyn, ninst, x3);
+                            X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, x3);
                             CALL(native_fld, -1);
                         }
                     }
@@ -293,7 +293,7 @@ uintptr_t dynarec64_DB(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                         }
                         CALL(native_fstp, -1);
                     }
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 default:
                     DEFAULT;

src/dynarec/rv64/dynarec_rv64_dc.c

@@ -39,69 +39,135 @@ uintptr_t dynarec64_DC(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
     switch(nextop) {
         case 0xC0 ... 0xC7:
             INST_NAME("FADD STx, ST0");
-            DEFAULT;
+            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 ... 0xCF:
             INST_NAME("FMUL STx, ST0");
-            DEFAULT;
+            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 ... 0xD7:
             INST_NAME("FCOM ST0, STx"); //yep
-            DEFAULT;
+            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)) {
+                FCOMS(v1, v2, x1, x2, x3, x4, x5);
+            } else {
+                FCOMD(v1, v2, x1, x2, x3, x4, x5);
+            }
             break;
         case 0xD8 ... 0xDF:
             INST_NAME("FCOMP ST0, STx");
-            DEFAULT;
+            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)) {
+                FCOMS(v1, v2, x1, x2, x3, x4, x5);
+            } else {
+                FCOMD(v1, v2, x1, x2, x3, x4, x5);
+            }
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xE0 ... 0xE7:
             INST_NAME("FSUBR STx, ST0");
-            DEFAULT;
-            break;
+            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 ... 0xEF:
             INST_NAME("FSUB STx, ST0");
-            DEFAULT;
+            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 ... 0xF7:
             INST_NAME("FDIVR STx, ST0");
-            DEFAULT;
+            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 ... 0xFF:
             INST_NAME("FDIV STx, ST0");
-            DEFAULT;
+            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, EXT_CACHE_ST_D);
+                    v2 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, NULL, 1, 0);
+                    FLD(v2, wback, fixedaddress);
+                    FADDD(v1, v1, v2);
+                    break;
+                case 1:
+                    INST_NAME("FMUL ST0, double[ED]");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, EXT_CACHE_ST_D);
+                    v2 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, NULL, 1, 0);
+                    FLD(v2, wback, fixedaddress);
+                    FMULD(v1, v1, v2);
+                    break;
+                case 2:
+                    INST_NAME("FCOM ST0, double[ED]");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, EXT_CACHE_ST_D);
+                    v2 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, NULL, 1, 0);
+                    FLD(v2, wback, fixedaddress);
+                    FCOMD(v1, v2, x1, x6, x3, x4, x5);
+                    break;
                 case 3:
                     INST_NAME("FCOMP ST0, double[ED]");
                     v1 = x87_get_st(dyn, ninst, x1, x2, 0, EXT_CACHE_ST_D);
                     v2 = fpu_get_scratch(dyn);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, NULL, 1, 0);
                     FLD(v2, wback, fixedaddress);
-
-                    LHU(x3, xEmu, offsetof(x64emu_t, sw));
-                    MOV32w(x1, 0b1110100011111111); // mask off c0,c1,c2,c3
-                    AND(x3, x3, x1);
-                    FEQD(x5, v1, v1);
-                    FEQD(x4, v2, v2);
-                    AND(x5, x5, x4);
-                    BEQZ(x5, 24); // undefined/NaN
-                    FEQD(x5, v1, v2);
-                    BNEZ(x5, 28); // equal
-                    FLTD(x3, v1, v2); // x3 = (v1<v2)?1:0
-                    SLLI(x1, x3, 8);
-                    J(20); // end
-                    // undefined/NaN
-                    LUI(x1, 1);
-                    ADDI(x1, x1, 0b010100000000);
-                    J(8); // end
-                    // equal
-                    LUI(x1, 1);
-                    // end
-                    OR(x3, x3, x1);
-                    SH(x3, xEmu, offsetof(x64emu_t, sw));
-
-                    x87_do_pop(dyn, ninst, x3);
+                    FCOMD(v1, v2, x1, x6, x3, x4, x5);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
+                    break;
+                case 4:
+                    INST_NAME("FSUB ST0, double[ED]");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, EXT_CACHE_ST_D);
+                    v2 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, NULL, 1, 0);
+                    FLD(v2, wback, fixedaddress);
+                    FSUBD(v1, v1, v2);
+                    break;
+                case 5:
+                    INST_NAME("FSUBR ST0, double[ED]");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, EXT_CACHE_ST_D);
+                    v2 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, NULL, 1, 0);
+                    FLD(v2, wback, fixedaddress);
+                    FSUBD(v1, v2, v1);
                     break;
                 case 6:
                     INST_NAME("FDIV ST0, double[ED]");
@@ -111,8 +177,14 @@ uintptr_t dynarec64_DC(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                     FLD(v2, wback, fixedaddress);
                     FDIVD(v1, v1, v2);
                     break;
-                default:
-                    DEFAULT;
+                case 7:
+                    INST_NAME("FDIVR ST0, double[ED]");
+                    v1 = x87_get_st(dyn, ninst, x1, x2, 0, EXT_CACHE_ST_D);
+                    v2 = fpu_get_scratch(dyn);
+                    addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, NULL, 1, 0);
+                    FLD(v2, wback, fixedaddress);
+                    FDIVD(v1, v2, v1);
+                    break;
             }
     }
     return addr;

src/dynarec/rv64/dynarec_rv64_dd.c

@@ -64,11 +64,17 @@ uintptr_t dynarec64_DD(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
         case 0xD6:
         case 0xD7:
             INST_NAME("FST ST0, STx");
-            DEFAULT;
+            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)) {
+                FMVS(v2, v1);
+            } else {
+                FMVD(v2, v1);
+            }
             break;
         case 0xD8:
             INST_NAME("FSTP ST0, ST0");
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xD9:
         case 0xDA:
@@ -82,7 +88,7 @@ uintptr_t dynarec64_DD(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             x87_get_st_empty(dyn, ninst, x1, x2, nextop&7, X87_ST(nextop&7));
             x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
             x87_swapreg(dyn, ninst, x1, x2, 0, nextop&7);
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xE0:
         case 0xE1:
@@ -93,7 +99,13 @@ uintptr_t dynarec64_DD(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
         case 0xE6:
         case 0xE7:
             INST_NAME("FUCOM ST0, STx");
-            DEFAULT;
+            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)) {
+                FCOMS(v1, v2, x1, x2, x3, x4, x5);
+            } else {
+                FCOMD(v1, v2, x1, x2, x3, x4, x5);
+            }
             break;
         case 0xE8:
         case 0xE9:
@@ -104,7 +116,14 @@ uintptr_t dynarec64_DD(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
         case 0xEE:
         case 0xEF:
             INST_NAME("FUCOMP ST0, STx");
-            DEFAULT;
+            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)) {
+                FCOMS(v1, v2, x1, x2, x3, x4, x5);
+            } else {
+                FCOMD(v1, v2, x1, x2, x3, x4, x5);
+            }
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xC8:
         case 0xC9:
@@ -137,7 +156,7 @@ uintptr_t dynarec64_DD(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("FLD double");
-                    v1 = x87_do_push(dyn, ninst, x1, EXT_CACHE_ST_D);
+                    X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, EXT_CACHE_ST_D);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, NULL, 1, 0);
                     FLD(v1, wback, fixedaddress);
                     break;
@@ -152,7 +171,7 @@ uintptr_t dynarec64_DD(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                     v1 = x87_get_st(dyn, ninst, x1, x2, 0, EXT_CACHE_ST_D);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, NULL, 1, 0);
                     FSD(v1, wback, fixedaddress);
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 case 7:
                     INST_NAME("FNSTSW m2byte");

src/dynarec/rv64/dynarec_rv64_de.c

@@ -52,7 +52,7 @@ uintptr_t dynarec64_DE(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             } else {
                 FADDD(v1, v1, v2);
             }
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xC8:
         case 0xC9:
@@ -70,7 +70,7 @@ uintptr_t dynarec64_DE(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             } else {
                 FMULD(v1, v1, v2);
             }
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xD0:
         case 0xD1:
@@ -80,12 +80,27 @@ uintptr_t dynarec64_DE(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
         case 0xD5:
         case 0xD6:
         case 0xD7:
-            // INST_NAME("FCOMP ST0, STx"); //yep
-            DEFAULT;
+            INST_NAME("FCOMP ST0, STx"); // yep
+            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)) {
+                FCOMS(v1, v2, x1, x2, x3, x4, x5);
+            } else {
+                FCOMD(v1, v2, x1, x2, x3, x4, x5);
+            }
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xD9:
-            // INST_NAME("FCOMPP ST0, STx");
-            DEFAULT;
+            INST_NAME("FCOMPP ST0, STx");
+            v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop & 7));
+            v2 = x87_get_st(dyn, ninst, x1, x2, nextop & 7, X87_COMBINE(0, nextop & 7));
+            if (ST_IS_F(0)) {
+                FCOMS(v1, v2, x1, x2, x3, x4, x5);
+            } else {
+                FCOMD(v1, v2, x1, x2, x3, x4, x5);
+            }
+            X87_POP_OR_FAIL(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xE0:
         case 0xE1:
@@ -103,7 +118,7 @@ uintptr_t dynarec64_DE(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             } else {
                 FSUBD(v1, v2, v1);
             }
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xE8:
         case 0xE9:
@@ -121,7 +136,7 @@ uintptr_t dynarec64_DE(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             } else {
                 FSUBD(v1, v1, v2);
             }
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xF0:
         case 0xF1:
@@ -139,8 +154,8 @@ uintptr_t dynarec64_DE(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             } else {
                 FDIVD(v1, v2, v1);
             }
-            x87_do_pop(dyn, ninst, x3);
-            break;       
+            X87_POP_OR_FAIL(dyn, ninst, x3);
+            break;
         case 0xF8:
         case 0xF9:
         case 0xFA:
@@ -157,7 +172,7 @@ uintptr_t dynarec64_DE(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             } else {
                 FDIVD(v1, v1, v2);
             }
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xD8:
         case 0xDA:
@@ -166,7 +181,8 @@ uintptr_t dynarec64_DE(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
         case 0xDD:
         case 0xDE:
         case 0xDF:
-            return 0;
+            DEFAULT;
+            break;
         default:
             switch((nextop>>3)&7) {
                 default:

src/dynarec/rv64/dynarec_rv64_df.c

@@ -46,7 +46,7 @@ uintptr_t dynarec64_DF(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
 
         case 0xE0:
             INST_NAME("FNSTSW AX");
-            LHU(x2, xEmu, offsetof(x64emu_t, top));
+            LWU(x2, xEmu, offsetof(x64emu_t, top));
             LHU(x1, xEmu, offsetof(x64emu_t, sw));
             MOV32w(x3, 0b1100011111111111); // mask
             AND(x1, x1, x3);
@@ -68,7 +68,7 @@ uintptr_t dynarec64_DF(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             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));
             CLEAR_FLAGS();
-            IFX(F_ZF | F_PF | F_CF) {
+            IFX(X_ZF | X_PF | X_CF) {
                 if(ST_IS_F(0)) {
                     FEQS(x5, v1, v1);
                     FEQS(x4, v2, v2);
@@ -103,7 +103,7 @@ uintptr_t dynarec64_DF(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                     // end
                 }
             }
-            x87_do_pop(dyn, ninst, x3);
+            X87_POP_OR_FAIL(dyn, ninst, x3);
             break;
         case 0xC8 ... 0xDF:
         case 0xE1 ... 0xE7:
@@ -115,7 +115,7 @@ uintptr_t dynarec64_DF(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             switch((nextop>>3)&7) {
                 case 0:
                     INST_NAME("FILD ST0, Ew");
-                    v1 = x87_do_push(dyn, ninst, x1, EXT_CACHE_ST_F);
+                    X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, EXT_CACHE_ST_F);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x3, x4, &fixedaddress, rex, NULL, 1, 0);
                     LH(x1, wback, fixedaddress);
                     if(ST_IS_F(0)) {
@@ -144,7 +144,7 @@ uintptr_t dynarec64_DF(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                     }
                     MARK2;
                     SH(x4, wback, fixedaddress);
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 case 3:
                     INST_NAME("FISTP Ew, ST0");
@@ -168,11 +168,11 @@ uintptr_t dynarec64_DF(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                     }
                     MARK2;
                     SH(x4, wback, fixedaddress);
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 case 5:
                     INST_NAME("FILD ST0, i64");
-                    v1 = x87_do_push(dyn, ninst, x1, EXT_CACHE_ST_D);
+                    X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, EXT_CACHE_ST_D);
                     addr = geted(dyn, addr, ninst, nextop, &wback, x2, x3, &fixedaddress, rex, NULL, 1, 0);
                     LD(x1, wback, fixedaddress);
                     if (rex.is32bits) {
@@ -193,6 +193,14 @@ uintptr_t dynarec64_DF(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                         FSD(v1, x5, 0);  // ref
                     }
                     break;
+                case 6:
+                    INST_NAME("FBSTP tbytes, ST0");
+                    x87_forget(dyn, ninst, x1, x2, 0);
+                    addr = geted(dyn, addr, ninst, nextop, &ed, x1, x2, &fixedaddress, rex, NULL, 0, 0);
+                    if (ed != x1) { MV(x1, ed); }
+                    CALL(fpu_fbst, -1);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
+                    break;
                 case 7:
                     INST_NAME("FISTP i64, ST0");
                     v1 = x87_get_st(dyn, ninst, x1, x2, 0, EXT_CACHE_ST_D);
@@ -233,7 +241,7 @@ uintptr_t dynarec64_DF(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                     SD(x4, wback, fixedaddress);
                     MARK3;
                     x87_restoreround(dyn, ninst, u8);
-                    x87_do_pop(dyn, ninst, x3);
+                    X87_POP_OR_FAIL(dyn, ninst, x3);
                     break;
                 default:
                     DEFAULT;

src/dynarec/rv64/dynarec_rv64_emit_math.c

@@ -63,7 +63,7 @@ void emit_add32(dynarec_rv64_t* dyn, int ninst, rex_t rex, int s1, int s2, int s
     }
     IFX(X_AF | X_OF) {
         if(rv64_zbb) {
-            ANDN(s3, s1, s3);   // s3 = ~res & (op1 | op2)
+            ANDN(s3, s3, s1);   // s3 = ~res & (op1 | op2)
         } else {
             NOT(s5, s1);   // s5 = ~res
             AND(s3, s5, s3);   // s3 = ~res & (op1 | op2)
@@ -160,7 +160,7 @@ void emit_add32c(dynarec_rv64_t* dyn, int ninst, rex_t rex, int s1, int64_t c, i
     }
     IFX(X_AF | X_OF) {
         if(rv64_zbb) {
-            ANDN(s3, s1, s3);   // s3 = ~res & (op1 | op2)
+            ANDN(s3, s3, s1);   // s3 = ~res & (op1 | op2)
         } else {
             NOT(s2, s1);   // s2 = ~res
             AND(s3, s2, s3);   // s3 = ~res & (op1 | op2)
@@ -218,7 +218,7 @@ void emit_add16(dynarec_rv64_t* dyn, int ninst, int s1, int s2, int s3, int s4,
     }
     IFX(X_AF | X_OF) {
         if(rv64_zbb) {
-            ANDN(s3, s1, s3);    // s3 = ~res & (op1 | op2)
+            ANDN(s3, s3, s1);    // s3 = ~res & (op1 | op2)
         } else {
             NOT(s5, s1);   // s5 = ~res
             AND(s3, s5, s3);   // s3 = ~res & (op1 | op2)
@@ -280,7 +280,7 @@ void emit_add8(dynarec_rv64_t* dyn, int ninst, int s1, int s2, int s3, int s4)
 
     IFX(X_AF|X_OF) {
         if(rv64_zbb) {
-            ANDN(s3, s1, s3);   // s3 = ~res & (op1 | op2)
+            ANDN(s3, s3, s1);   // s3 = ~res & (op1 | op2)
         } else {
             NOT(s4, s1);   // s4 = ~res
             AND(s3, s4, s3);   // s3 = ~res & (op1 | op2)
@@ -336,7 +336,7 @@ void emit_add8c(dynarec_rv64_t* dyn, int ninst, int s1, int c, int s2, int s3, i
         SET_DFNONE();
     }
     IFX(X_AF | X_OF) {
-        if(X_PEND) {} else {MOV32w(s4, c&0xff);}
+        IFX(X_PEND) {} else {MOV32w(s4, c&0xff);}
         OR(s3, s1, s4);       // s3 = op1 | op2
         AND(s4, s1, s4);       // s4 = op1 & op2
     }
@@ -344,7 +344,7 @@ void emit_add8c(dynarec_rv64_t* dyn, int ninst, int s1, int c, int s2, int s3, i
 
     IFX(X_AF|X_OF) {
         if(rv64_zbb) {
-            ANDN(s3, s1, s3);   // s3 = ~res & (op1 | op2)
+            ANDN(s3, s3, s1);   // s3 = ~res & (op1 | op2)
         } else {
             NOT(s2, s1);   // s2 = ~res
             AND(s3, s2, s3);   // s3 = ~res & (op1 | op2)
@@ -596,7 +596,7 @@ void emit_inc8(dynarec_rv64_t* dyn, int ninst, int s1, int s2, int s3, int s4)
     }
     IFX(X_AF | X_OF) {
         if(rv64_zbb) {
-            ANDN(s3, s1, s3);   // s3 = ~res & (op1 | op2)
+            ANDN(s3, s3, s1);   // s3 = ~res & (op1 | op2)
         } else {
             NOT(s2, s1);   // s2 = ~res
             AND(s3, s2, s3);   // s3 = ~res & (op1 | op2)
@@ -709,7 +709,7 @@ void emit_inc32(dynarec_rv64_t* dyn, int ninst, rex_t rex, int s1, int s2, int s
     }
     IFX(X_AF | X_OF) {
         if(rv64_zbb) {
-            ANDN(s3, s1, s3);    // s3 = ~res & (op1 | op2)
+            ANDN(s3, s3, s1);    // s3 = ~res & (op1 | op2)
         } else {
             NOT(s2, s1);   // s2 = ~res
             AND(s3, s2, s3);   // s3 = ~res & (op1 | op2)
@@ -825,7 +825,7 @@ void emit_inc16(dynarec_rv64_t* dyn, int ninst, int s1, int s2, int s3, int s4)
     }
     IFX(X_AF | X_OF) {
         if(rv64_zbb) {
-            ANDN(s3, s1, s3);   // s3 = ~res & (op1 | op2)
+            ANDN(s3, s3, s1);   // s3 = ~res & (op1 | op2)
         } else {
             NOT(s2, s1);   // s2 = ~res
             AND(s3, s2, s3);   // s3 = ~res & (op1 | op2)
@@ -968,8 +968,8 @@ void emit_adc8(dynarec_rv64_t* dyn, int ninst, int s1, int s2, int s3, int s4, i
         SET_DFNONE();
     }
     IFX(X_AF | X_OF) {
-        OR(s4, s1, s2);  // s3 = op1 | op2
-        AND(s5, s1, s2); // s4 = op1 & op2
+        OR(s4, s1, s2);  // s4 = op1 | op2
+        AND(s5, s1, s2); // s5 = op1 & op2
     }
 
     ADD(s1, s1, s2);
@@ -982,7 +982,7 @@ void emit_adc8(dynarec_rv64_t* dyn, int ninst, int s1, int s2, int s3, int s4, i
     }
     IFX(X_AF | X_OF) {
         if(rv64_zbb) {
-            ANDN(s3, s1, s4);   // s3 = ~res & (op1 | op2)
+            ANDN(s3, s4, s1);   // s3 = ~res & (op1 | op2)
         } else {
             NOT(s2, s1);     // s2 = ~res
             AND(s3, s2, s4); // s3 = ~res & (op1 | op2)
@@ -1304,8 +1304,8 @@ void emit_adc16(dynarec_rv64_t* dyn, int ninst, int s1, int s2, int s3, int s4,
         SET_DFNONE();
     }
     IFX(X_AF | X_OF) {
-        OR(s4, s1, s2);  // s3 = op1 | op2
-        AND(s5, s1, s2); // s4 = op1 & op2
+        OR(s4, s1, s2);  // s4 = op1 | op2
+        AND(s5, s1, s2); // s5 = op1 & op2
     }
 
     ADD(s1, s1, s2);
@@ -1318,7 +1318,7 @@ void emit_adc16(dynarec_rv64_t* dyn, int ninst, int s1, int s2, int s3, int s4,
     }
     IFX(X_AF | X_OF) {
         if(rv64_zbb) {
-            ANDN(s3, s1, s4);   // s3 = ~res & (op1 | op2)
+            ANDN(s3, s4, s1);   // s3 = ~res & (op1 | op2)
         } else {
             NOT(s2, s1);     // s2 = ~res
             AND(s3, s2, s4); // s3 = ~res & (op1 | op2)
@@ -1386,8 +1386,8 @@ void emit_adc32(dynarec_rv64_t* dyn, int ninst, rex_t rex, int s1, int s2, int s
         }
     }
     IFX(X_AF | X_OF) {
-        OR(s4, s1, s2);  // s3 = op1 | op2
-        AND(s5, s1, s2); // s4 = op1 & op2
+        OR(s4, s1, s2);  // s4 = op1 | op2
+        AND(s5, s1, s2); // s5 = op1 & op2
     }
 
     ADDxw(s1, s1, s2);
@@ -1404,7 +1404,7 @@ void emit_adc32(dynarec_rv64_t* dyn, int ninst, rex_t rex, int s1, int s2, int s
     }
     IFX(X_AF | X_OF) {
         if(rv64_zbb) {
-            ANDN(s3, s1, s4);   // s3 = ~res & (op1 | op2)
+            ANDN(s3, s4, s1);   // s3 = ~res & (op1 | op2)
         } else {
             NOT(s2, s1);     // s2 = ~res
             AND(s3, s2, s4); // s3 = ~res & (op1 | op2)

src/dynarec/rv64/dynarec_rv64_emit_shift.c

@@ -385,8 +385,8 @@ void emit_ror32(dynarec_rv64_t* dyn, int ninst, rex_t rex, int s1, int s2, int s
         SDxw(s1, xEmu, offsetof(x64emu_t, res));
     }
     IFX(X_CF) {
-        SRLI(s3, s1, rex.w?63:31);
-        AND(xFlags, xFlags, s3);
+        SRLIxw(s3, s1, rex.w?63:31);
+        OR(xFlags, xFlags, s3);
     }
     IFX(X_OF) {
         ADDI(s3, xZR, 1);
@@ -421,6 +421,8 @@ void emit_rol32c(dynarec_rv64_t* dyn, int ninst, rex_t rex, int s1, uint32_t c,
     }
     if(rv64_zbb) {
         RORIxw(s1, s1, (rex.w?64:32)-c);
+    } else if (rv64_xtheadbb) {
+        TH_SRRIxw(s1, s1, (rex.w?64:32)-c);
     } else {
         SLLIxw(s3, s1, c);
         SRLIxw(s1, s1, (rex.w?64:32)-c);
@@ -466,6 +468,8 @@ void emit_ror32c(dynarec_rv64_t* dyn, int ninst, rex_t rex, int s1, uint32_t c,
     }
     if(rv64_zbb) {
         RORIxw(s1, s1, c);
+    } else if (rv64_xtheadbb) {
+        TH_SRRIxw(s1, s1, c);
     } else {
         SRLIxw(s3, s1, c);
         SLLIxw(s1, s1, (rex.w?64:32)-c);
@@ -475,8 +479,8 @@ void emit_ror32c(dynarec_rv64_t* dyn, int ninst, rex_t rex, int s1, uint32_t c,
         SDxw(s1, xEmu, offsetof(x64emu_t, res));
     }
     IFX(X_CF) {
-        SRLI(s3, s1, rex.w?63:31);
-        AND(xFlags, xFlags, s3);
+        SRLIxw(s3, s1, rex.w?63:31);
+        OR(xFlags, xFlags, s3);
     }
     IFX(X_OF) {
         if(c==1) {

src/dynarec/rv64/dynarec_rv64_f0.c

@@ -49,8 +49,9 @@ uintptr_t dynarec64_F0(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
 
     GETREX();
 
-    // TODO: Take care of unligned memory access for all the LOCK ones.
-    // https://github.com/ptitSeb/box64/pull/604
+    // TODO: Add support for unligned memory access for all the LOCK ones.
+    // TODO: Add support for BOX4_DYNAREC_ALIGNED_ATOMICS.
+
     switch(opcode) {
         case 0x01:
             INST_NAME("LOCK ADD Ed, Gd");
@@ -214,13 +215,29 @@ uintptr_t dynarec64_F0(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                             } else {
                                 SMDMB();
                                 addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, LOCK_LOCK, 0, 0);
+                                ANDI(x1, wback, (1 << (rex.w + 2)) - 1);
+                                BNEZ_MARK3(x1);
+                                // Aligned
                                 MARKLOCK;
                                 LRxw(x1, wback, 1, 1);
                                 SUBxw(x3, x1, xRAX);
-                                BNE_MARK(x3, xZR);
+                                BNEZ_MARK(x3);
                                 // EAX == Ed
                                 SCxw(x4, gd, wback, 1, 1);
                                 BNEZ_MARKLOCK(x4);
+                                B_MARK_nocond;
+                                MARK3;
+                                // Unaligned
+                                ANDI(x5, wback, -(1 << (rex.w + 2)));
+                                MARK2; // Use MARK2 as a "MARKLOCK" since we're running out of marks.
+                                LDxw(x6, wback, 0);
+                                LRxw(x1, x5, 1, 1);
+                                SUBxw(x3, x6, xRAX);
+                                BNEZ_MARK(x3);
+                                // EAX == Ed
+                                SCxw(x4, x1, x5, 1, 1);
+                                BNEZ_MARK2(x4);
+                                SDxw(gd, wback, 0);
                                 MARK;
                                 UFLAG_IF {emit_cmp32(dyn, ninst, rex, xRAX, x1, x3, x4, x5, x6);}
                                 MVxw(xRAX, x1);
@@ -371,7 +388,7 @@ uintptr_t dynarec64_F0(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                 MARKLOCK;
                 LR_W(x5, wback, 1, 1);
                 ANDI(x4, x5, 0xff); // x4 = Ed.b[0]
-                ANDI(x5, x5, -256); // x5 = clear Ed.b[0]
+                ANDI(x5, x5, ~0xff); // x5 = clear Ed.b[0]
                 ADDW(x6, x4, x2);
                 ANDI(x9, xFlags, 1 << F_CF);
                 ADDW(x6, x6, x9);   // x6 = adc
@@ -533,20 +550,40 @@ uintptr_t dynarec64_F0(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
                         ed = xRAX+(nextop&7)+(rex.b<<3);
                         emit_add32c(dyn, ninst, rex, ed, i64, x3, x4, x5, x6);
                     } else {
+                        SMDMB();
                         addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, LOCK_LOCK, 0, (opcode==0x81)?4:1);
                         if(opcode==0x81) i64 = F32S; else i64 = F8S;
+                        if (i64 < -2048 || i64 >= 2048)
+                            MOV64xw(x4, i64);
+                        ANDI(x1, wback, (1 << (rex.w + 2)) - 1);
+                        BNEZ_MARK3(x1);
+                        // Aligned
                         MARKLOCK;
                         LRxw(x1, wback, 1, 1);
-                        if(i64>=-2048 && i64<2048)
+                        if (i64 >= -2048 && i64 < 2048)
                             ADDIxw(x4, x1, i64);
-                        else {
-                            MOV64xw(x4, i64);
+                        else
                             ADDxw(x4, x1, x4);
-                        }
                         SCxw(x3, x4, wback, 1, 1);
                         BNEZ_MARKLOCK(x3);
-                        IFX(X_ALL|X_PEND)
+                        B_MARK_nocond;
+                        MARK3;
+                        // Unaligned
+                        ANDI(x5, wback, -(1 << (rex.w + 2)));
+                        MARK2; // Use MARK2 as a "MARKLOCK" since we're running out of marks.
+                        LDxw(x6, wback, 0);
+                        LRxw(x1, x5, 1, 1);
+                        if (i64 >= -2048 && i64 < 2048)
+                            ADDIxw(x4, x6, i64);
+                        else
+                            ADDxw(x4, x6, x4);
+                        SCxw(x3, x1, x5, 1, 1);
+                        BNEZ_MARK2(x3);
+                        SDxw(x4, wback, 0);
+                        MARK;
+                        IFX (X_ALL | X_PEND)
                             emit_add32c(dyn, ninst, rex, x1, i64, x3, x4, x5, x6);
+                        SMDMB();
                     }
                     break;
                 case 1: // OR
@@ -640,6 +677,41 @@ uintptr_t dynarec64_F0(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ni
             }
             SMDMB();
             break;
+        case 0x87:
+            INST_NAME("LOCK XCHG Ed, Gd");
+            nextop = F8;
+            if (MODREG) {
+                GETGD;
+                GETED(0);
+                MV(x1, gd);
+                MV(gd, ed);
+                MV(ed, x1);
+            } else {
+                SMDMB();
+                GETGD;
+                addr = geted(dyn, addr, ninst, nextop, &wback, x2, x1, &fixedaddress, rex, LOCK_LOCK, 0, 0);
+                ANDI(x1, wback, (1 << (rex.w + 2)) - 1);
+                BNEZ_MARK3(x1);
+                // Aligned
+                MARKLOCK;
+                LRxw(x1, wback, 1, 1);
+                SCxw(x4, gd, wback, 1, 1);
+                BNEZ_MARKLOCK(x4);
+                B_MARK_nocond;
+                MARK3;
+                // Unaligned
+                ANDI(x5, wback, -(1 << (rex.w + 2)));
+                MARK2; // Use MARK2 as a "MARKLOCK" since we're running out of marks.
+                LDxw(x1, wback, 0);
+                LRxw(x3, x5, 1, 1);
+                SCxw(x4, x3, x5, 1, 1);
+                BNEZ_MARK2(x4);
+                SDxw(gd, wback, 0);
+                MARK;
+                MVxw(gd, x1);
+                SMDMB();
+            }
+            break;
         case 0xFF:
             nextop = F8;
             switch((nextop>>3)&7)

src/dynarec/rv64/dynarec_rv64_f20f.c

@@ -146,33 +146,6 @@ uintptr_t dynarec64_F20F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             opcode = F8;
             switch(opcode) {
 
-                case 0xF0:
-                    INST_NAME("(unsupported) CRC32 Gd, Eb)");
-                    nextop = F8;
-                    addr = fakeed(dyn, addr, ninst, nextop);
-                    SETFLAGS(X_ALL, SF_SET);    // Hack to set flags in "don't care" state
-                    GETIP(ip);
-                    STORE_XEMU_CALL();
-                    CALL(native_ud, -1);
-                    LOAD_XEMU_CALL();
-                    jump_to_epilog(dyn, 0, xRIP, ninst);
-                    *need_epilog = 0;
-                    *ok = 0;
-                    break;
-                case 0xF1:
-                    INST_NAME("(unsupported) CRC32 Gd, Ed)");
-                    nextop = F8;
-                    addr = fakeed(dyn, addr, ninst, nextop);
-                    SETFLAGS(X_ALL, SF_SET);    // Hack to set flags in "don't care" state
-                    GETIP(ip);
-                    STORE_XEMU_CALL();
-                    CALL(native_ud, -1);
-                    LOAD_XEMU_CALL();
-                    jump_to_epilog(dyn, 0, xRIP, ninst);
-                    *need_epilog = 0;
-                    *ok = 0;
-                    break;
-
                 default:
                     DEFAULT;
             }

src/dynarec/rv64/dynarec_rv64_f30f.c

@@ -24,7 +24,8 @@
 
 uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int ninst, rex_t rex, int* ok, int* need_epilog)
 {
-    (void)ip; (void)need_epilog;
+    (void)ip;
+    (void)need_epilog;
 
     uint8_t opcode = F8;
     uint8_t nextop, u8;
@@ -46,14 +47,14 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
     MAYUSE(v1);
     MAYUSE(j64);
 
-    switch(opcode) {
+    switch (opcode) {
         case 0x10:
             INST_NAME("MOVSS Gx, Ex");
             nextop = F8;
             GETG;
-            if(MODREG) {
+            if (MODREG) {
                 v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
-                q0 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3), 1);
+                q0 = sse_get_reg(dyn, ninst, x1, (nextop & 7) + (rex.b << 3), 1);
                 FMVS(v0, q0);
             } else {
                 v0 = sse_get_reg_empty(dyn, ninst, x1, gd, 1);
@@ -61,8 +62,8 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, x2, &fixedaddress, rex, NULL, 8, 0);
                 FLW(v0, ed, fixedaddress);
                 // reset upper part
-                SW(xZR, xEmu, offsetof(x64emu_t, xmm[gd])+4);
-                SD(xZR, xEmu, offsetof(x64emu_t, xmm[gd])+8);
+                SW(xZR, xEmu, offsetof(x64emu_t, xmm[gd]) + 4);
+                SD(xZR, xEmu, offsetof(x64emu_t, xmm[gd]) + 8);
             }
             break;
         case 0x11:
@@ -70,8 +71,8 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             nextop = F8;
             GETG;
             v0 = sse_get_reg(dyn, ninst, x1, gd, 1);
-            if(MODREG) {
-                q0 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3), 1);
+            if (MODREG) {
+                q0 = sse_get_reg(dyn, ninst, x1, (nextop & 7) + (rex.b << 3), 1);
                 FMVS(q0, v0);
             } else {
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, x2, &fixedaddress, rex, NULL, 1, 0);
@@ -88,12 +89,12 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
 
             // GX->ud[1] = GX->ud[0] = EX->ud[0];
             // GX->ud[3] = GX->ud[2] = EX->ud[2];
-            LD(x3, wback, fixedaddress+0);
-            SD(x3, gback, gdoffset+0);
-            SD(x3, gback, gdoffset+4);
-            LD(x3, wback, fixedaddress+8);
-            SD(x3, gback, gdoffset+8);
-            SD(x3, gback, gdoffset+12);
+            LD(x3, wback, fixedaddress + 0);
+            SD(x3, gback, gdoffset + 0);
+            SD(x3, gback, gdoffset + 4);
+            LD(x3, wback, fixedaddress + 8);
+            SD(x3, gback, gdoffset + 8);
+            SD(x3, gback, gdoffset + 12);
             break;
         case 0x1E:
             INST_NAME("NOP / ENDBR32 / ENDBR64");
@@ -106,7 +107,7 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             nextop = F8;
             GETGXSS(v0);
             GETED(0);
-            if(rex.w) {
+            if (rex.w) {
                 FCVTSL(v0, ed, RD_RNE);
             } else {
                 FCVTSW(v0, ed, RD_RNE);
@@ -118,17 +119,17 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             nextop = F8;
             GETGD;
             GETEXSS(d0, 0);
-            if(!box64_dynarec_fastround) {
-                FSFLAGSI(0);  // // reset all bits
+            if (!box64_dynarec_fastround) {
+                FSFLAGSI(0); // // reset all bits
             }
             FCVTSxw(gd, d0, RD_RTZ);
-            if(!rex.w)
+            if (!rex.w)
                 ZEROUP(gd);
-            if(!box64_dynarec_fastround) {
-                FRFLAGS(x5);   // get back FPSR to check the IOC bit
-                ANDI(x5, x5, (1<<FR_NV)|(1<<FR_OF));
+            if (!box64_dynarec_fastround) {
+                FRFLAGS(x5); // get back FPSR to check the IOC bit
+                ANDI(x5, x5, (1 << FR_NV) | (1 << FR_OF));
                 CBZ_NEXT(x5);
-                if(rex.w) {
+                if (rex.w) {
                     MOV64x(gd, 0x8000000000000000LL);
                 } else {
                     MOV32w(gd, 0x80000000);
@@ -140,19 +141,19 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             nextop = F8;
             GETGD;
             GETEXSS(d0, 0);
-            if(!box64_dynarec_fastround) {
-                FSFLAGSI(0);  // // reset all bits
+            if (!box64_dynarec_fastround) {
+                FSFLAGSI(0); // // reset all bits
             }
             u8 = sse_setround(dyn, ninst, x5, x6);
             FCVTSxw(gd, d0, RD_DYN);
             x87_restoreround(dyn, ninst, u8);
-            if(!rex.w)
+            if (!rex.w)
                 ZEROUP(gd);
-            if(!box64_dynarec_fastround) {
-                FRFLAGS(x5);   // get back FPSR to check the IOC bit
-                ANDI(x5, x5, (1<<FR_NV)|(1<<FR_OF));
+            if (!box64_dynarec_fastround) {
+                FRFLAGS(x5); // get back FPSR to check the IOC bit
+                ANDI(x5, x5, (1 << FR_NV) | (1 << FR_OF));
                 CBZ_NEXT(x5);
-                if(rex.w) {
+                if (rex.w) {
                     MOV64x(gd, 0x8000000000000000LL);
                 } else {
                     MOV32w(gd, 0x80000000);
@@ -257,31 +258,31 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             u8 = F8;
             int32_t idx;
 
-            idx = 4+((u8>>(0*2))&3);
-            LHU(x3, wback, fixedaddress+idx*2);
-            idx = 4+((u8>>(1*2))&3);
-            LHU(x4, wback, fixedaddress+idx*2);
-            idx = 4+((u8>>(2*2))&3);
-            LHU(x5, wback, fixedaddress+idx*2);
-            idx = 4+((u8>>(3*2))&3);
-            LHU(x6, wback, fixedaddress+idx*2);
+            idx = 4 + ((u8 >> (0 * 2)) & 3);
+            LHU(x3, wback, fixedaddress + idx * 2);
+            idx = 4 + ((u8 >> (1 * 2)) & 3);
+            LHU(x4, wback, fixedaddress + idx * 2);
+            idx = 4 + ((u8 >> (2 * 2)) & 3);
+            LHU(x5, wback, fixedaddress + idx * 2);
+            idx = 4 + ((u8 >> (3 * 2)) & 3);
+            LHU(x6, wback, fixedaddress + idx * 2);
 
-            SH(x3, gback, gdoffset+(4+0)*2);
-            SH(x4, gback, gdoffset+(4+1)*2);
-            SH(x5, gback, gdoffset+(4+2)*2);
-            SH(x6, gback, gdoffset+(4+3)*2);
+            SH(x3, gback, gdoffset + (4 + 0) * 2);
+            SH(x4, gback, gdoffset + (4 + 1) * 2);
+            SH(x5, gback, gdoffset + (4 + 2) * 2);
+            SH(x6, gback, gdoffset + (4 + 3) * 2);
 
-            if (!(MODREG && (gd==ed))) {
-                LD(x3, wback, fixedaddress+0);
-                SD(x3, gback, gdoffset+0);
+            if (!(MODREG && (gd == ed))) {
+                LD(x3, wback, fixedaddress + 0);
+                SD(x3, gback, gdoffset + 0);
             }
             break;
         case 0x7E:
             INST_NAME("MOVQ Gx, Ex");
             nextop = F8;
             // Will load Gx as SD. Is that a good choice?
-            if(MODREG) {
-                v1 = sse_get_reg(dyn, ninst, x1, (nextop&7) + (rex.b<<3), 0);
+            if (MODREG) {
+                v1 = sse_get_reg(dyn, ninst, x1, (nextop & 7) + (rex.b << 3), 0);
                 GETGXSD_empty(v0);
                 FMVD(v0, v1);
             } else {
@@ -290,7 +291,7 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
                 addr = geted(dyn, addr, ninst, nextop, &ed, x1, x2, &fixedaddress, rex, NULL, 1, 0);
                 FLD(v0, ed, fixedaddress);
             }
-            SD(xZR, xEmu, offsetof(x64emu_t, xmm[gd])+8);
+            SD(xZR, xEmu, offsetof(x64emu_t, xmm[gd]) + 8);
             break;
         case 0x7F:
             INST_NAME("MOVDQU Ex,Gx");
@@ -298,7 +299,7 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             GETGX();
             GETEX(x2, 0);
             SSE_LOOP_MV_Q2(x3);
-            if(!MODREG) SMWRITE2();
+            if (!MODREG) SMWRITE2();
             break;
 
         case 0x5B:
@@ -307,19 +308,19 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             GETGX();
             GETEX(x2, 0);
             v0 = fpu_get_scratch(dyn);
-            for(int i=0; i<4; ++i) {
-                if(!box64_dynarec_fastround) {
+            for (int i = 0; i < 4; ++i) {
+                if (!box64_dynarec_fastround) {
                     FSFLAGSI(0); // reset all bits
                 }
-                FLW(v0, wback, fixedaddress+i*4);
+                FLW(v0, wback, fixedaddress + i * 4);
                 FCVTWS(x3, v0, RD_RTZ);
-                if(!box64_dynarec_fastround) {
-                    FRFLAGS(x5);   // get back FPSR to check the IOC bit
-                    ANDI(x5, x5, (1<<FR_NV)|(1<<FR_OF));
+                if (!box64_dynarec_fastround) {
+                    FRFLAGS(x5); // get back FPSR to check the IOC bit
+                    ANDI(x5, x5, (1 << FR_NV) | (1 << FR_OF));
                     BEQZ(x5, 8);
                     MOV32w(x3, 0x80000000);
                 }
-                SW(x3, gback, gdoffset+i*4);
+                SW(x3, gback, gdoffset + i * 4);
             }
             break;
         case 0xB8:
@@ -329,17 +330,17 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             nextop = F8;
             GETED(0);
             GETGD;
-            if(!rex.w && MODREG) {
+            if (!rex.w && MODREG) {
                 AND(x4, ed, xMASK);
                 ed = x4;
             }
             CLEAR_FLAGS();
             BNE_MARK(ed, xZR);
-            ORI(xFlags, xFlags, 1<<F_ZF);
+            ORI(xFlags, xFlags, 1 << F_ZF);
             MOV32w(gd, 0);
             B_NEXT_nocond;
             MARK;
-            if(rv64_zbb) {
+            if (rv64_zbb) {
                 CPOPxw(gd, ed);
             } else {
                 TABLE64(x1, 0x5555555555555555uLL);
@@ -371,75 +372,75 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             nextop = F8;
             GETED(0);
             GETGD;
-            if(!rex.w && MODREG) {
+            if (!rex.w && MODREG) {
                 AND(x4, ed, xMASK);
                 ed = x4;
             }
-            ANDI(xFlags, xFlags, ~((1<<F_ZF) | (1<<F_CF)));
+            ANDI(xFlags, xFlags, ~((1 << F_ZF) | (1 << F_CF)));
             BNE_MARK(ed, xZR);
-            ORI(xFlags, xFlags, 1<<F_CF);
-            MOV32w(gd, rex.w?64:32);
+            ORI(xFlags, xFlags, 1 << F_CF);
+            MOV32w(gd, rex.w ? 64 : 32);
             B_NEXT_nocond;
             MARK;
-            if(rv64_zbb) {
+            if (rv64_zbb) {
                 CTZxw(gd, ed);
             } else {
                 NEG(x2, ed);
                 AND(x2, x2, ed);
                 TABLE64(x3, 0x03f79d71b4ca8b09ULL);
                 MUL(x2, x2, x3);
-                SRLI(x2, x2, 64-6);
+                SRLI(x2, x2, 64 - 6);
                 TABLE64(x1, (uintptr_t)&deBruijn64tab);
                 ADD(x1, x1, x2);
                 LBU(gd, x1, 0);
             }
-            BNE(gd, xZR, 4+4);
-            ORI(xFlags, xFlags, 1<<F_ZF);
+            BNE(gd, xZR, 4 + 4);
+            ORI(xFlags, xFlags, 1 << F_ZF);
             break;
         case 0xBD:
             INST_NAME("LZCNT Gd, Ed");
-            SETFLAGS(X_ZF|X_CF, SF_SUBSET);
+            SETFLAGS(X_ZF | X_CF, SF_SUBSET);
             SET_DFNONE();
             nextop = F8;
             GETED(0);
             GETGD;
-            if(!rex.w && MODREG) {
+            if (!rex.w && MODREG) {
                 AND(x4, ed, xMASK);
                 ed = x4;
             }
             BNE_MARK(ed, xZR);
-            MOV32w(gd, rex.w?64:32);
-            ANDI(xFlags, xFlags, ~(1<<F_ZF));
-            ORI(xFlags, xFlags, 1<<F_CF);
+            MOV32w(gd, rex.w ? 64 : 32);
+            ANDI(xFlags, xFlags, ~(1 << F_ZF));
+            ORI(xFlags, xFlags, 1 << F_CF);
             B_NEXT_nocond;
             MARK;
-            if(rv64_zbb) {
+            if (rv64_zbb) {
                 CLZxw(gd, ed);
             } else {
-                if(ed!=gd)
+                if (ed != gd)
                     u8 = gd;
                 else
                     u8 = x1;
-                ADDI(u8, xZR, rex.w?63:31);
-                if(rex.w) {
+                ADDI(u8, xZR, rex.w ? 63 : 31);
+                if (rex.w) {
                     MV(x2, ed);
                     SRLI(x3, x2, 32);
-                    BEQZ(x3, 4+2*4);
+                    BEQZ(x3, 4 + 2 * 4);
                     SUBI(u8, u8, 32);
                     MV(x2, x3);
                 } else {
                     AND(x2, ed, xMASK);
                 }
                 SRLI(x3, x2, 16);
-                BEQZ(x3, 4+2*4);
+                BEQZ(x3, 4 + 2 * 4);
                 SUBI(u8, u8, 16);
                 MV(x2, x3);
                 SRLI(x3, x2, 8);
-                BEQZ(x3, 4+2*4);
+                BEQZ(x3, 4 + 2 * 4);
                 SUBI(u8, u8, 8);
                 MV(x2, x3);
                 SRLI(x3, x2, 4);
-                BEQZ(x3, 4+2*4);
+                BEQZ(x3, 4 + 2 * 4);
                 SUBI(u8, u8, 4);
                 MV(x2, x3);
                 ANDI(x2, x2, 0b1111);
@@ -449,9 +450,9 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
                 SUB(gd, u8, x2);
                 MARK2;
             }
-            ANDI(xFlags, xFlags, ~((1<<F_ZF) | (1<<F_CF)));
-            BNE(gd, xZR, 4+4);
-            ORI(xFlags, xFlags, 1<<F_ZF);
+            ANDI(xFlags, xFlags, ~((1 << F_ZF) | (1 << F_CF)));
+            BNE(gd, xZR, 4 + 4);
+            ORI(xFlags, xFlags, 1 << F_ZF);
             break;
 
         case 0xC2:
@@ -460,9 +461,9 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             GETGXSS(d0);
             GETEXSS(d1, 1);
             u8 = F8;
-            if ((u8&7) == 0) {                                      // Equal
+            if ((u8 & 7) == 0) { // Equal
                 FEQS(x2, d0, d1);
-            } else if ((u8&7) == 4) {                               // Not Equal or unordered
+            } else if ((u8 & 7) == 4) { // Not Equal or unordered
                 FEQS(x2, d0, d1);
                 XORI(x2, x2, 1);
             } else {
@@ -471,27 +472,33 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
                 FEQS(x2, d1, d1);
                 AND(x2, x2, x3);
 
-                switch(u8&7) {
-                case 1: BEQ_MARK(x2, xZR); FLTS(x2, d0, d1); break; // Less than
-                case 2: BEQ_MARK(x2, xZR); FLES(x2, d0, d1); break; // Less or equal
-                case 3: XORI(x2, x2, 1); break;                     // NaN
-                case 5: {                                           // Greater or equal or unordered
-                    BEQ_MARK2(x2, xZR);
-                    FLES(x2, d1, d0);
-                    B_MARK_nocond;
-                    break;
-                }
-                case 6: {                                           // Greater or unordered, test inverted, N!=V so unordered or less than (inverted)
-                    BEQ_MARK2(x2, xZR);
-                    FLTS(x2, d1, d0);
-                    B_MARK_nocond;
-                    break;
-                }
-                case 7: break;                                      // Not NaN
+                switch (u8 & 7) {
+                    case 1:
+                        BEQ_MARK(x2, xZR);
+                        FLTS(x2, d0, d1);
+                        break; // Less than
+                    case 2:
+                        BEQ_MARK(x2, xZR);
+                        FLES(x2, d0, d1);
+                        break;                      // Less or equal
+                    case 3: XORI(x2, x2, 1); break; // NaN
+                    case 5: {                       // Greater or equal or unordered
+                        BEQ_MARK2(x2, xZR);
+                        FLES(x2, d1, d0);
+                        B_MARK_nocond;
+                        break;
+                    }
+                    case 6: { // Greater or unordered, test inverted, N!=V so unordered or less than (inverted)
+                        BEQ_MARK2(x2, xZR);
+                        FLTS(x2, d1, d0);
+                        B_MARK_nocond;
+                        break;
+                    }
+                    case 7: break; // Not NaN
                 }
 
                 MARK2;
-                if ((u8&7) == 5 || (u8&7) == 6) {
+                if ((u8 & 7) == 5 || (u8 & 7) == 6) {
                     MOV32w(x2, 1);
                 }
                 MARK;
@@ -507,12 +514,12 @@ uintptr_t dynarec64_F30F(dynarec_rv64_t* dyn, uintptr_t addr, uintptr_t ip, int
             GETEX(x2, 0);
             q0 = fpu_get_scratch(dyn);
             q1 = fpu_get_scratch(dyn);
-            LW(x3, wback, fixedaddress+0);
-            LW(x4, wback, fixedaddress+4);
+            LW(x3, wback, fixedaddress + 0);
+            LW(x4, wback, fixedaddress + 4);
             FCVTDW(q0, x3, RD_RTZ);
             FCVTDW(q1, x4, RD_RTZ);
-            FSD(q0, gback, gdoffset+0);
-            FSD(q1, gback, gdoffset+8);
+            FSD(q0, gback, gdoffset + 0);
+            FSD(q1, gback, gdoffset + 8);
             break;
 
         default:

src/dynarec/rv64/dynarec_rv64_functions.c

@@ -58,7 +58,7 @@ void fpu_free_reg(dynarec_rv64_t* dyn, int reg)
     int idx = EXTIDX(reg);
     // TODO: check upper limit?
     dyn->e.fpuused[idx] = 0;
-    if(dyn->e.extcache[reg].t!=EXT_CACHE_ST_F && dyn->e.extcache[reg].t!=EXT_CACHE_ST_D)
+    if(dyn->e.extcache[idx].t!=EXT_CACHE_ST_F && dyn->e.extcache[idx].t!=EXT_CACHE_ST_D)
         dyn->e.extcache[idx].v = 0;
 }
 // Get an MMX double reg
@@ -494,7 +494,7 @@ const char* getCacheName(int t, int n)
 void inst_name_pass3(dynarec_native_t* dyn, int ninst, const char* name, rex_t rex)
 {
     static const char* fnames[] = {
-        "ft0"," ft1", "ft2", "ft3", "ft4", "ft5", "ft6", "ft7",
+        "ft0", "ft1", "ft2", "ft3", "ft4", "ft5", "ft6", "ft7",
         "fs0", "fs1",
         "fa0", "fa1", "fa2", "fa3", "fa4", "fa5", "fa6", "fa7",
         "fs2", "fs3", "fs4", "fs5", "fs6", "fs7", "fs8", "fs9", "fs10", "fs11",
@@ -561,4 +561,4 @@ void print_newinst(dynarec_native_t* dyn, int ninst)
         ninst, dyn->block, dyn->native_size,
         (box64_dynarec_dump>1)?"\e[m":""
         );
-}
+}

src/dynarec/rv64/dynarec_rv64_helper.c

@@ -56,16 +56,7 @@ uintptr_t geted(dynarec_rv64_t* dyn, uintptr_t addr, int ninst, uint8_t nextop,
                 if (sib_reg!=4) {
                     if(tmp && ((tmp<-2048) || (tmp>maxval) || !i12)) {
                         MOV64x(scratch, tmp);
-                        if((sib>>6)) {
-                            if(rv64_zba) {
-                                SHxADD(ret, xRAX+sib_reg, sib>>6, scratch);
-                            } else {
-                                SLLI(ret, xRAX+sib_reg, (sib>>6));
-                                ADD(ret, ret, scratch);
-                            }
-                        } else {
-                            ADD(ret, xRAX+sib_reg, scratch);
-                        }
+                        ADDSL(ret, scratch, xRAX+sib_reg, sib>>6, ret);
                     } else {
                         if(sib>>6) {
                             SLLI(ret, xRAX+sib_reg, (sib>>6));
@@ -82,16 +73,7 @@ uintptr_t geted(dynarec_rv64_t* dyn, uintptr_t addr, int ninst, uint8_t nextop,
                 }
             } else {
                 if (sib_reg!=4) {
-                    if(sib>>6) {
-                        if(rv64_zba) {
-                            SHxADD(ret, xRAX+sib_reg, sib>>6, xRAX+sib_reg2);
-                        } else {
-                            SLLI(scratch, xRAX+sib_reg, (sib>>6));
-                            ADD(ret, xRAX+sib_reg2, scratch);
-                        }
-                    } else {
-                        ADD(ret, xRAX+sib_reg2, xRAX+sib_reg);
-                    }
+                    ADDSL(ret, xRAX+sib_reg2, xRAX+sib_reg, sib>>6, scratch);
                 } else {
                     ret = xRAX+sib_reg2;
                 }
@@ -146,16 +128,7 @@ uintptr_t geted(dynarec_rv64_t* dyn, uintptr_t addr, int ninst, uint8_t nextop,
             *fixaddress = i64;
             if((nextop&7)==4) {
                 if (sib_reg!=4) {
-                    if(sib>>6) {
-                        if(rv64_zba) {
-                            SHxADD(ret, xRAX+sib_reg, sib>>6, xRAX+sib_reg2);
-                        } else {
-                            SLLI(scratch, xRAX+sib_reg, (sib>>6));
-                            ADD(ret, xRAX+sib_reg2, scratch);
-                        }
-                    } else {
-                        ADD(ret, xRAX+sib_reg2, xRAX+sib_reg);
-                    }
+                    ADDSL(ret, xRAX+sib_reg2, xRAX+sib_reg, sib>>6, scratch);
                 } else {
                     ret = xRAX+sib_reg2;
                 }
@@ -165,16 +138,7 @@ uintptr_t geted(dynarec_rv64_t* dyn, uintptr_t addr, int ninst, uint8_t nextop,
             if(i64>=-2048 && i64<=2047) {
                 if((nextop&7)==4) {
                     if (sib_reg!=4) {
-                        if(sib>>6) {
-                            if(rv64_zba) {
-                                SHxADD(scratch, xRAX+sib_reg, sib>>6, xRAX+sib_reg2);
-                            } else {
-                                SLLI(scratch, xRAX+sib_reg, (sib>>6));
-                                ADD(scratch, xRAX+sib_reg2, scratch);
-                            }
-                        } else {
-                            ADD(scratch, xRAX+sib_reg2, xRAX+sib_reg);
-                        }
+                        ADDSL(scratch, xRAX+sib_reg2, xRAX+sib_reg, sib>>6, scratch);
                     } else {
                         scratch = xRAX+sib_reg2;
                     }
@@ -186,16 +150,7 @@ uintptr_t geted(dynarec_rv64_t* dyn, uintptr_t addr, int ninst, uint8_t nextop,
                 if((nextop&7)==4) {
                     if (sib_reg!=4) {
                         ADD(scratch, scratch, xRAX+sib_reg2);
-                        if(sib>>6) {
-                            if(rv64_zba) {
-                                SHxADD(ret, xRAX+sib_reg, sib>>6, scratch);
-                            } else {
-                                SLLI(ret, xRAX+sib_reg, (sib>>6));
-                                ADD(ret, scratch, ret);
-                            }
-                        } else {
-                            ADD(ret, scratch, xRAX+sib_reg);
-                        }
+                        ADDSL(ret, scratch, xRAX+sib_reg, sib>>6, ret);
                     } else {
                         PASS3(int tmp = xRAX+sib_reg2);
                         ADD(ret, tmp, scratch);
@@ -507,7 +462,9 @@ void jump_to_epilog_fast(dynarec_rv64_t* dyn, uintptr_t ip, int reg, int ninst)
     SMEND();
     BR(x2);
 }
-
+#ifdef JMPTABLE_SHIFT4
+#error TODO!
+#endif
 void jump_to_next(dynarec_rv64_t* dyn, uintptr_t ip, int reg, int ninst)
 {
     MAYUSE(dyn); MAYUSE(ninst);
@@ -1718,9 +1675,9 @@ static void swapCache(dynarec_rv64_t* dyn, int ninst, int i, int j, extcache_t *
     // SWAP
     ext_cache_t tmp;
     MESSAGE(LOG_DUMP, "\t  - Swapping %d <-> %d\n", i, j);
-    // There is no VSWP in Arm64 NEON to swap 2 register contents!
+    // There is no swap instruction in RV64 to swap 2 float registers!
     // so use a scratch...
-    #define SCRATCH 0
+    #define SCRATCH 2
     if(i_single)
         FMVS(SCRATCH, reg_i);
     else
@@ -1770,7 +1727,7 @@ static void loadCache(dynarec_rv64_t* dyn, int ninst, int stack_cnt, int s1, int
             break;
         case EXT_CACHE_MM:
             MESSAGE(LOG_DUMP, "\t  - Loading %s\n", getCacheName(t, n));
-            FLD(reg, xEmu, offsetof(x64emu_t, mmx[i]));
+            FLD(reg, xEmu, offsetof(x64emu_t, mmx[n]));
             break;
         case EXT_CACHE_ST_D:
         case EXT_CACHE_ST_F:
@@ -1926,7 +1883,9 @@ static void fpuCacheTransform(dynarec_rv64_t* dyn, int ninst, int s1, int s2, in
         s3_top = 0;
         stack_cnt = cache_i2.stack;
     }
-    extcache_t cache = dyn->e;
+
+    // FIXME: why we can't just use dyn->e here? debug that!
+    extcache_t cache = dyn->insts[ninst].e;
     int s1_val = 0;
     int s2_val = 0;
     // unload every uneeded cache

src/dynarec/rv64/dynarec_rv64_pass0.h

@@ -16,9 +16,10 @@
         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 JUMP(A, C)         add_jump(dyn, ninst); 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 BARRIER_NEXT(A) dyn->insts[ninst].x64.barrier_next = A
+#define SET_HASCALLRET()    dyn->insts[ninst].x64.has_callret = 1
 #define NEW_INST \
         ++dyn->size;                            \
         if(dyn->size+3>=dyn->cap) {             \

src/dynarec/rv64/dynarec_rv64_private.h

@@ -79,8 +79,8 @@ typedef struct instruction_rv64_s {
     int                 size2;      // size of the arm emitted instrucion after pass2
     int                 pred_sz;    // size of predecessor list
     int                 *pred;      // predecessor array
-    uintptr_t           mark, mark2, mark3;
-    uintptr_t           markf, markf2;
+    uintptr_t           mark[3];
+    uintptr_t           markf[2];
     uintptr_t           markseg;
     uintptr_t           marklock;
     int                 pass2choice;// value for choices that are fixed on pass2 for pass3
@@ -112,6 +112,9 @@ typedef struct dynarec_rv64_s {
     uintptr_t*          next;       // variable array of "next" jump address
     int                 next_sz;
     int                 next_cap;
+    int*                jmps;       // variable array of jump instructions
+    int                 jmp_sz;
+    int                 jmp_cap;
     int*                predecessor;// single array of all predecessor
     dynablock_t*        dynablock;
     instsize_t*         instsize;
@@ -132,6 +135,8 @@ typedef struct dynarec_rv64_s {
 
 void add_next(dynarec_rv64_t *dyn, uintptr_t addr);
 uintptr_t get_closest_next(dynarec_rv64_t *dyn, uintptr_t addr);
+void add_jump(dynarec_rv64_t *dyn, int ninst);
+int get_first_jump(dynarec_rv64_t *dyn, int next);
 int is_nops(dynarec_rv64_t *dyn, uintptr_t addr, int n);
 int is_instructions(dynarec_rv64_t *dyn, uintptr_t addr, int n);
 
@@ -139,12 +144,12 @@ int Table64(dynarec_rv64_t *dyn, uint64_t val, int pass);  // add a value to tab
 
 void CreateJmpNext(void* addr, void* next);
 
-#define GO_TRACE(A, B)      \
+#define GO_TRACE(A, B, s0)  \
     GETIP(addr);            \
     MV(A1, xRIP);           \
-    STORE_XEMU_CALL();      \
+    STORE_XEMU_CALL(s0);    \
     MOV64x(A2, B);          \
     CALL(A, -1);            \
     LOAD_XEMU_CALL()
 
-#endif //__DYNAREC_RV64_PRIVATE_H_
+#endif //__DYNAREC_RV64_PRIVATE_H_

src/dynarec/rv64/rv64_lock.S

@@ -17,9 +17,13 @@
 .global rv64_lock_incif0
 .global rv64_lock_decifnot0
 .global rv64_lock_store
+.global rv64_lock_store_dd
 .global rv64_lock_cas_d
 .global rv64_lock_cas_dd
 .global rv64_lock_cas_dq
+.global rv64_lock_get_b
+.global rv64_lock_get_d
+.global rv64_lock_get_dd
 
 rv64_lock_xchg_dd:
     // address is a0, value is a1, return old value in a0
@@ -202,6 +206,11 @@ rv64_lock_store:
     fence   rw, rw
     ret
 
+rv64_lock_store_dd:
+    sd      a1, 0(a0)
+    fence   rw, rw
+    ret
+
 rv64_lock_cas_d:
     lr.w t0, (a0)
     bne t0, a1, 1f
@@ -229,4 +238,19 @@ rv64_lock_cas_dq:
     ret
 1:
     li a0, 1
-    ret
+    ret
+
+rv64_lock_get_b:
+    fence   rw, rw
+    lb      a0, 0(a0)
+    ret
+
+rv64_lock_get_d:
+    fence   rw, rw
+    lw      a0, 0(a0)
+    ret
+
+rv64_lock_get_dd:
+    fence   rw, rw
+    ld      a0, 0(a0)
+    ret

src/dynarec/rv64/rv64_lock.h

@@ -44,9 +44,21 @@ extern int rv64_lock_decifnot0(void*p);
 // atomic store (with memory barrier)
 extern void rv64_lock_store(void*p, uint32_t v);
 
+// atomic store (with memory barrier)
+extern void rv64_lock_store_dd(void*p, uint64_t v);
+
 // (mostly) Atomically store val1 and val2 at [p] if old [p] is ref. Return 0 if OK, 1 is not. p needs to be aligned
 extern int rv64_lock_cas_dq(void* p, uint64_t ref, uint64_t val1, uint64_t val2);
 
+// atomic get (with memory barrier)
+extern uint32_t rv64_lock_get_b(void* p);
+
+// atomic get (with memory barrier)
+extern uint32_t rv64_lock_get_d(void* p);
+
+// atomic get (with memory barrier)
+extern void* rv64_lock_get_dd(void* p);
+
 // Not defined in assembler but in dynarec_rv64_functions
 uint8_t extract_byte(uint32_t val, void* address);
 uint32_t insert_byte(uint32_t val, uint8_t b, void* address);

src/elfs/elfload_dump.c

@@ -45,7 +45,9 @@ const char* DumpSection(Elf64_Shdr *s, char* SST) {
         GO(SHT_GNU_ATTRIBUTES);
         GO(SHT_GNU_HASH);
         GO(SHT_GNU_LIBLIST);
-        GO(SHT_CHECKSUM);
+        #ifndef TERMUX
+          GO(SHT_CHECKSUM);
+        #endif
         GO(SHT_LOSUNW);
         //GO(SHT_SUNW_move);
         GO(SHT_SUNW_COMDAT);

src/elfs/elfloader.c

@@ -38,6 +38,7 @@
 #endif
 #include "../emu/x64emu_private.h"
 #include "../emu/x64run_private.h"
+#include "../tools/bridge_private.h"
 #include "x64tls.h"
 
 void* my__IO_2_1_stderr_ = NULL;
@@ -69,6 +70,9 @@ elfheader_t* LoadAndCheckElfHeader(FILE* f, const char* name, int exec)
     h->globaldefver = NewDefaultVersion();
     h->weakdefver = NewDefaultVersion();
     h->refcnt = 1;
+
+    h->file = f;
+    h->fileno = fileno(f);
     
     return h;
 }
@@ -100,6 +104,8 @@ void FreeElfHeader(elfheader_t** head)
 
     box_free(h->name);
     box_free(h->path);
+    if(h->file)
+        fclose(h->file);
     box_free(h);
 
     *head = NULL;
@@ -169,201 +175,174 @@ const char* ElfPath(elfheader_t* head)
         return NULL;
     return head->path;
 }
-int AllocElfMemory(box64context_t* context, elfheader_t* head, int mainbin)
+#define ALIGN(p) (((p)+box64_pagesize-1)&~(box64_pagesize-1))
+int AllocLoadElfMemory(box64context_t* context, elfheader_t* head, int mainbin)
 {
     uintptr_t offs = 0;
+    loadProtectionFromMap();
+    int log_level = box64_load_addr?LOG_INFO:LOG_DEBUG;
+
+    head->multiblock_n = 0; // count PHEntrie with LOAD
+    uintptr_t max_align = (box64_pagesize-1);
+    for (size_t i=0; i<head->numPHEntries; ++i) 
+        if(head->PHEntries[i].p_type == PT_LOAD && head->PHEntries[i].p_flags) {
+            if(max_align < head->PHEntries[i].p_align-1)
+                max_align = head->PHEntries[i].p_align-1;
+            ++head->multiblock_n;
+        }
+
     if(!head->vaddr && box64_load_addr) {
-        offs = (uintptr_t)find47bitBlockNearHint((void*)box64_load_addr, head->memsz);
-        box64_load_addr += head->memsz;
+        offs = (uintptr_t)find47bitBlockNearHint((void*)((box64_load_addr+max_align)&~max_align), head->memsz, max_align);
+        box64_load_addr = offs + head->memsz;
         box64_load_addr = (box64_load_addr+0x10ffffffLL)&~0xffffffLL;
     }
-    int log_level = box64_load_addr?LOG_INFO:LOG_DEBUG;
-    if(!offs)
-        offs = head->vaddr;
-    if(head->vaddr) {
-        head->multiblock_n = 0; // count PHEntrie with LOAD
-        for (size_t i=0; i<head->numPHEntries; ++i) 
-            if(head->PHEntries[i].p_type == PT_LOAD && head->PHEntries[i].p_flags)
-                ++head->multiblock_n;
-        head->multiblock_size = (uint64_t*)box_calloc(head->multiblock_n, sizeof(uint64_t));
-        head->multiblock_offs = (uintptr_t*)box_calloc(head->multiblock_n, sizeof(uintptr_t));
-        head->multiblock = (void**)box_calloc(head->multiblock_n, sizeof(void*));
-        // and now, create all individual blocks
-        head->memory = (char*)0xffffffffffffffff;
-        int n = 0;
-        for (size_t i=0; i<head->numPHEntries; ++i) 
-            if(head->PHEntries[i].p_type == PT_LOAD && head->PHEntries[i].p_flags) {
-                Elf64_Phdr * e = &head->PHEntries[i];
-                uintptr_t bstart = e->p_vaddr;
-                uint64_t bsize = e->p_memsz;
-                uintptr_t balign = e->p_align;
-                if (balign) balign = balign-1; else balign = 1;
-                if(balign<4095) balign = 4095;
-                uintptr_t bend = (bstart + bsize + balign)&(~balign);
-                bstart &= ~balign;
-                int ok = 0;
-                for (int j=0; !ok && j<n; ++j) {
-                    uintptr_t start = head->multiblock_offs[j];
-                    uintptr_t end = head->multiblock_offs[j] + head->multiblock_size[j];
-                    start &= ~balign;
-                    if( (head->e_type == ET_DYN) ||
-                        (((bstart>=start) && (bstart<=end+0x1000)) || ((bend>=start) && (bend<=end)) || ((bstart<start) && (bend>end))))
-                    {
-                        // merge
-                        ok = 1;
-                        if(bstart<start)
-                            head->multiblock_offs[j] = bstart;
-                        head->multiblock_size[j] = ((bend>end)?bend:end) - head->multiblock_offs[j];
-                        --head->multiblock_n;
-                    }
-                }
-                if(!ok) {
-                    head->multiblock_offs[n] = bstart;
-                    head->multiblock_size[n] = bend - head->multiblock_offs[n];
-                    ++n;
-                }
-            }
-        head->multiblock_n = n; // might be less in fact
-        for (int i=0; i<head->multiblock_n; ++i) {
-            printf_log(log_level, "Allocating 0x%lx memory @%p for Elf \"%s\"\n", head->multiblock_size[i], (void*)head->multiblock_offs[i], head->name);
-            void* p = mmap((void*)head->multiblock_offs[i], head->multiblock_size[i]
-                , PROT_READ | PROT_WRITE | PROT_EXEC
-                , 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%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%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);
-                    // need to adjust vaddr!
-                    for (size_t j=0; j<head->numPHEntries; ++j) 
-                        if(head->PHEntries[j].p_type == PT_LOAD) {
-                            Elf64_Phdr * e = &head->PHEntries[j];
-                            if(e->p_vaddr>=head->multiblock_offs[j] && e->p_vaddr<(head->multiblock_offs[j]+head->multiblock_size[j])) {
-                                e->p_vaddr = e->p_vaddr - head->multiblock_offs[j] + (uintptr_t)p;
-                                if(!head->delta) head->delta = (intptr_t)p - (intptr_t)head->multiblock_offs[j];
-                            }
-                        }
-                }
-            }
-            setProtection((uintptr_t)p, head->multiblock_size[i], PROT_READ | PROT_WRITE | PROT_EXEC);
-            head->multiblock[i] = p;
-            if(p<(void*)head->memory)
-                head->memory = (char*)p;
-        }
-    } else {
-        // vaddr is 0, load everything has a One block
-        uintptr_t old_offs = offs;
-        if(!offs /*&& box64_wine*/)
-            offs = (uintptr_t)find47bitBlock(head->memsz); // limit to 47bits...
-        printf_log(log_level, "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(offs &&!old_offs && p!=MAP_FAILED && offs!=(uintptr_t)p) {
-            // try again
-            munmap(p, head->memsz);
-            loadProtectionFromMap();
-            offs = (uintptr_t)find47bitBlock(head->memsz);
-            printf_log(log_level, "New, try. Allocating 0x%zx memory @%p for Elf \"%s\"\n", head->memsz, (void*)offs, head->name);
-            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%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%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);
-        head->memory = p;
-        memset(p, 0, head->memsz);
-        head->delta = (intptr_t)p - (intptr_t)head->vaddr;
-        printf_log(log_level, "Got %p (delta=%p)\n", p, (void*)head->delta);
-
-        head->multiblock_n = 1;
-        head->multiblock_size = (uint64_t*)box_calloc(head->multiblock_n, sizeof(uint64_t));
-        head->multiblock_offs = (uintptr_t*)box_calloc(head->multiblock_n, sizeof(uintptr_t));
-        head->multiblock = (void**)box_calloc(head->multiblock_n, sizeof(void*));
-        head->multiblock_size[0] = head->memsz;
-        head->multiblock_offs[0] = (uintptr_t)p;
-        head->multiblock[0] = p;
+    if(!offs && !head->vaddr)
+        offs = (uintptr_t)find47bitBlockElf(head->memsz, mainbin, max_align); // limit to 47bits...
+    #if defined(PAGE8K) || defined(PAGE16K) || defined(PAGE64K)
+    // Will not try anything smart on pagesize != 4k....
+    void* image = mmap64((void*)(head->vaddr?head->vaddr:offs), head->memsz, PROT_READ|PROT_WRITE|PROT_EXEC, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
+    #else   // PAGE4K
+    // prereserve the whole elf image, without populating
+    void* image = mmap64((void*)(head->vaddr?head->vaddr:offs), head->memsz, 0, MAP_ANONYMOUS|MAP_PRIVATE|MAP_NORESERVE, -1, 0);
+    if(image!=MAP_FAILED && head->vaddr && image!=(void*)offs && (((uintptr_t)image)&max_align)) {
+        munmap(image, head->memsz);
+        image = mmap64(find47bitBlockElf(head->memsz, mainbin, max_align), head->memsz, 0, MAP_ANONYMOUS|MAP_PRIVATE|MAP_NORESERVE, -1, 0);
     }
+    #endif
+    if(image!=MAP_FAILED && !head->vaddr && image!=(void*)offs) {
+        printf_log(LOG_INFO, "Mmap64 for (@%p 0x%zx) for elf \"%s\" returned %p instead\n", (void*)(head->vaddr?head->vaddr:offs), head->memsz, head->name, image);
+        offs = (uintptr_t)image;
+    }
+    if(image==MAP_FAILED || image!=(void*)(head->vaddr?head->vaddr:offs)) {
+        printf_log(LOG_NONE, "%s cannot create memory map (@%p 0x%zx) for elf \"%s\"", (image==MAP_FAILED)?"Error:":"Warning:", (void*)(head->vaddr?head->vaddr:offs), head->memsz, head->name);
+        if(image==MAP_FAILED) {
+            printf_log(LOG_NONE, " error=%d/%s\n", errno, strerror(errno));
+        } else {
+            printf_log(LOG_NONE, " got %p\n", image);
+        }
+        if(image==MAP_FAILED)
+            return 1;
+        offs = (uintptr_t)image-head->vaddr;
+    }
+    printf_log(log_level, "Pre-allocated 0x%zx byte at %p for %s\n", head->memsz, image, head->name);
+    head->delta = offs;
+    printf_log(log_level, "Delta of %p (vaddr=%p) for Elf \"%s\"\n", (void*)offs, (void*)head->vaddr, head->name);
 
+    head->image = image;
+    #if defined(PAGE8K) || defined(PAGE16K) || defined(PAGE64K)
+    setProtection_elf((uintptr_t)image, head->memsz, PROT_READ|PROT_WRITE|PROT_EXEC);
+    #else
+    setProtection_elf((uintptr_t)image, head->memsz, 0);
+    #endif
+
+    head->multiblocks = (multiblock_t*)box_calloc(head->multiblock_n, sizeof(multiblock_t));
     head->tlsbase = AddTLSPartition(context, head->tlssize);
-
-    return 0;
-}
-
-void FreeElfMemory(elfheader_t* head)
-{
-    if(head->multiblock_n) {
-        for(int i=0; i<head->multiblock_n; ++i) {
-#ifdef DYNAREC
-            dynarec_log(LOG_INFO, "Free DynaBlocks for %s\n", head->path);
-            if(box64_dynarec) {
-                cleanDBFromAddressRange((uintptr_t)head->multiblock[i], head->multiblock_size[i], 1);
-                freeProtection((uintptr_t)head->multiblock[i], head->multiblock_size[i]);
-            }
-#endif
-            munmap(head->multiblock[i], head->multiblock_size[i]);
-        }
-        box_free(head->multiblock);
-        box_free(head->multiblock_size);
-        box_free(head->multiblock_offs);
-    }
-}
-
-int LoadElfMemory(FILE* f, box64context_t* context, elfheader_t* head)
-{
+    // and now, create all individual blocks
+    head->memory = (char*)0xffffffffffffffff;
+    int n = 0;
     for (size_t i=0; i<head->numPHEntries; ++i) {
-        if(head->PHEntries[i].p_type == PT_LOAD) {
+        if(head->PHEntries[i].p_type == PT_LOAD && head->PHEntries[i].p_flags) {
             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%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&(box64_pagesize-1))) {
-                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);
-                p = mmap(dest, e->p_filesz, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fileno(f), e->p_offset);
-            }
-            if(p!=dest) {
-                printf_log(LOG_DEBUG, "Loading block #%zu %p (0x%zx/0x%zx)\n",i, dest, e->p_filesz, e->p_memsz);
-                fseeko64(f, e->p_offset, SEEK_SET);
+
+            head->multiblocks[n].flags = e->p_flags;
+            head->multiblocks[n].offs = e->p_offset;
+            head->multiblocks[n].paddr = e->p_paddr + offs;
+            head->multiblocks[n].size = e->p_filesz;
+            head->multiblocks[n].align = e->p_align;
+            uint8_t prot = ((e->p_flags & PF_R)?PROT_READ:0)|((e->p_flags & PF_W)?PROT_WRITE:0)|((e->p_flags & PF_X)?PROT_EXEC:0);
+            #if defined(PAGE8K) || defined(PAGE16K) || defined(PAGE64K)
+                head->multiblocks[n].p = NULL;
                 if(e->p_filesz) {
-                    if(fread(dest, e->p_filesz, 1, f)!=1) {
-                        printf_log(LOG_NONE, "Fail to read PT_LOAD part #%zu (size=%zd)\n", i, e->p_filesz);
+                    fseeko64(head->file, head->multiblocks[n].offs, SEEK_SET);
+                    if(fread((void*)head->multiblocks[n].paddr, head->multiblocks[n].size, 1, head->file)!=1) {
+                        printf_log(LOG_NONE, "Cannot read elf block (@%p 0x%zx) for elf \"%s\"\n", (void*)head->multiblocks[n].offs, head->multiblocks[n].asize, head->name);
                         return 1;
                     }
                 }
+            #else //PAGE4K
+            // check if alignment is correct
+            uintptr_t balign = head->multiblocks[n].align-1;
+            if(balign<(box64_pagesize-1)) balign = (box64_pagesize-1);
+            head->multiblocks[n].asize = ALIGN(e->p_memsz+(e->p_paddr&balign));
+            int try_mmap = 1;
+            if(e->p_paddr&balign)
+                try_mmap = 0;
+            if(e->p_offset&(box64_pagesize-1))
+                try_mmap = 0;
+            if(e->p_memsz-e->p_filesz>(box64_pagesize-1))
+                try_mmap = 0;
+            if(!e->p_filesz)
+                try_mmap = 0;
+            if(try_mmap) {
+                printf_log(log_level, "Mmaping 0x%lx(0x%lx) bytes @%p for Elf \"%s\"\n", head->multiblocks[n].size, head->multiblocks[n].asize, (void*)head->multiblocks[n].paddr, head->name);
+                void* p = mmap64(
+                    (void*)head->multiblocks[n].paddr, 
+                    head->multiblocks[n].size, 
+                    prot,
+                    MAP_PRIVATE|MAP_FIXED, //((prot&PROT_WRITE)?MAP_SHARED:MAP_PRIVATE)|MAP_FIXED,
+                    head->fileno,
+                    e->p_offset
+                );
+                if(p==MAP_FAILED || p!=(void*)head->multiblocks[n].paddr) {
+                    try_mmap = 0;
+                    printf_log(log_level, "Mapping failed, using regular mmap+read");
+                } else {
+                    setProtection_elf((uintptr_t)p, head->multiblocks[n].asize, prot);
+                    head->multiblocks[n].p = p;
+
+                }
             }
+            if(!try_mmap) {
+                uintptr_t paddr = head->multiblocks[n].paddr&~balign;
+                size_t asize = head->multiblocks[n].asize;
+                printf_log(log_level, "Allocating 0x%zx (0x%zx) bytes @%p, will read 0x%zx @%p for Elf \"%s\"\n", asize, e->p_memsz, (void*)paddr, e->p_filesz, (void*)head->multiblocks[n].paddr, head->name);
+                void* p = mmap64(
+                    (void*)paddr,
+                    asize,
+                    prot|PROT_WRITE,
+                    MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
+                    -1,
+                    0
+                );
+                if(p==MAP_FAILED || p!=(void*)paddr) {
+                    printf_log(LOG_NONE, "Cannot create memory map (@%p 0x%zx/0x%zx) for elf \"%s\"", (void*)paddr, asize, balign, head->name);
+                    if(p==MAP_FAILED) {
+                        printf_log(LOG_NONE, " error=%d/%s\n", errno, strerror(errno));
+                    } else {
+                        printf_log(LOG_NONE, " got %p\n", p);
+                    }
+                    return 1;
+                }
+                setProtection_elf((uintptr_t)p, asize, prot);
+                head->multiblocks[n].p = p;
+                if(e->p_filesz) {
+                    fseeko64(head->file, head->multiblocks[n].offs, SEEK_SET);
+                    if(fread((void*)head->multiblocks[n].paddr, head->multiblocks[n].size, 1, head->file)!=1) {
+                        printf_log(LOG_NONE, "Cannot read elf block (@%p 0x%zx/0x%zx) for elf \"%s\"\n", (void*)head->multiblocks[n].offs, head->multiblocks[n].asize, balign, head->name);
+                        return 1;
+                    }
+                }
+                if(!(prot&PROT_WRITE))
+                    mprotect((void*)paddr, asize, prot);
+            }
+            #endif //PAGE4K
 #ifdef DYNAREC
             if(box64_dynarec && (e->p_flags & PF_X)) {
-                dynarec_log(LOG_DEBUG, "Add ELF eXecutable Memory %p:%p\n", dest, (void*)e->p_memsz);
-                addDBFromAddressRange((uintptr_t)dest, e->p_memsz);
+                dynarec_log(LOG_DEBUG, "Add ELF eXecutable Memory %p:%p\n", head->multiblocks[n].p, (void*)head->multiblocks[n].asize);
+                addDBFromAddressRange((uintptr_t)head->multiblocks[n].p, head->multiblocks[n].asize);
             }
 #endif
-
-            // zero'd difference between filesz and memsz
-            /*if(e->p_filesz != e->p_memsz)
-                memset(dest+e->p_filesz, 0, e->p_memsz - e->p_filesz);*/    //block is already 0'd at creation
+            if((uintptr_t)head->memory>(uintptr_t)head->multiblocks[n].p)
+                head->memory = (char*)head->multiblocks[n].p;
+            ++n;
         }
         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%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) {
+                fseeko64(head->file, e->p_offset, SEEK_SET);
+                if(fread(dest, e->p_filesz, 1, head->file)!=1) {
                     printf_log(LOG_NONE, "Fail to read PT_TLS part #%zu (size=%zd)\n", i, e->p_filesz);
                     return 1;
                 }
@@ -373,9 +352,34 @@ int LoadElfMemory(FILE* f, box64context_t* context, elfheader_t* head)
                 memset(dest+e->p_filesz, 0, e->p_memsz - e->p_filesz);
         }
     }
+
+    // can close the elf file now!
+    fclose(head->file);
+    head->file = NULL;
+    head->fileno = -1;
+
     return 0;
 }
 
+void FreeElfMemory(elfheader_t* head)
+{
+    if(head->multiblock_n) {
+#ifdef DYNAREC
+        for(int i=0; i<head->multiblock_n; ++i) {
+            dynarec_log(LOG_INFO, "Free DynaBlocks for %s\n", head->path);
+            if(box64_dynarec)
+                cleanDBFromAddressRange((uintptr_t)head->multiblocks[i].p, head->multiblocks[i].size, 1);
+            freeProtection((uintptr_t)head->multiblocks[i].p, head->multiblocks[i].asize);
+        }
+#endif
+        box_free(head->multiblocks);
+    }
+    // we only need to free the overall mmap, no need to free individual part as they are inside the big one
+    if(head->image && head->memsz)
+        munmap(head->image, head->memsz);
+    freeProtection((uintptr_t)head->image, head->memsz);
+}
+
 int isElfHasNeededVer(elfheader_t* head, const char* libname, elfheader_t* verneeded)
 {
     if(!verneeded || !head)
@@ -393,39 +397,6 @@ int isElfHasNeededVer(elfheader_t* head, const char* libname, elfheader_t* verne
     return 1;
 }
 
-int ReloadElfMemory(FILE* f, box64context_t* context, elfheader_t* head)
-{
-    (void)context;
-
-    for (size_t i=0; i<head->numPHEntries; ++i) {
-        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%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
-            cleanDBFromAddressRange((uintptr_t)dest, e->p_memsz, 0);
-            #endif
-            uint32_t page_offset = (uintptr_t)dest & (box64_pagesize - 1);
-            mprotect(dest - page_offset, e->p_memsz + page_offset, PROT_READ | PROT_WRITE | PROT_EXEC);
-            setProtection((uintptr_t)dest - page_offset, e->p_memsz + page_offset, PROT_READ | PROT_WRITE | PROT_EXEC);
-            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=%zd, return=%zd, feof=%d/ferror=%d/%s)\n", i, dest, e->p_filesz, r, feof(f), ferror(f), strerror(ferror(f)));
-                    return 1;
-                }
-            }
-            // zero'd difference between filesz and memsz
-            if(e->p_filesz != e->p_memsz)
-                memset(dest+e->p_filesz, 0, e->p_memsz - e->p_filesz);
-        }
-    }
-    // TLS data are just a copy, no need to re-load it
-    return 0;
-}
-
 int FindR64COPYRel(elfheader_t* h, const char* name, uintptr_t *offs, uint64_t** p, size_t size, int version, const char* vername)
 {
     if(!h)
@@ -471,7 +442,7 @@ int RelocateElfREL(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t*
         if(version!=-1) version &=0x7fff;
         const char* vername = GetSymbolVersion(head, version);
         const char* globdefver = (bind==STB_WEAK)?NULL:GetMaplibDefaultVersion(maplib, local_maplib, 0, symname);
-        const char* weakdefver = (bind==STB_WEAK)?GetMaplibDefaultVersion(maplib, local_maplib, 1, symname):NULL;
+        const char* weakdefver = (bind==STB_WEAK || !globdefver)?GetMaplibDefaultVersion(maplib, local_maplib, 1, symname):NULL;
         if(bind==STB_LOCAL) {
             if(!symname || !symname[0]) {
                 offs = sym->st_value + head->delta;
@@ -494,14 +465,14 @@ int RelocateElfREL(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t*
                 if(!offs && !end) {
                     if(!offs && !end && local_maplib)
                         GetGlobalWeakSymbolStartEnd(local_maplib, symname, &offs, &end, head, version, vername, globdefver, weakdefver);
-                    if(!offs && !end && local_maplib)
+                    if(!offs && !end)
                         GetGlobalWeakSymbolStartEnd(maplib, symname, &offs, &end, head, version, vername, globdefver, weakdefver);
                 }
             } else {
                 if(!offs && !end) {
                     if(!offs && !end && local_maplib)
                         GetGlobalSymbolStartEnd(local_maplib, symname, &offs, &end, head, version, vername, globdefver, weakdefver);
-                    if(!offs && !end && local_maplib)
+                    if(!offs && !end)
                         GetGlobalSymbolStartEnd(maplib, symname, &offs, &end, head, version, vername, globdefver, weakdefver);
                 }
             }
@@ -539,10 +510,10 @@ int RelocateElfREL(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t*
                     *p = globoffs;
                 } else {
                     // Look for same symbol already loaded but not in self (so no need for local_maplib here)
-                    if (GetGlobalNoWeakSymbolStartEnd(local_maplib?local_maplib:maplib, symname, &globoffs, &globend, version, vername, globdefver)) {
+                    /*if (GetGlobalNoWeakSymbolStartEnd(local_maplib?local_maplib:maplib, symname, &globoffs, &globend, version, vername, globdefver)) {
                         offs = globoffs;
                         end = globend;
-                    }
+                    }*/
                     if (!offs) {
                         if(strcmp(symname, "__gmon_start__") && strcmp(symname, "data_start") && strcmp(symname, "__data_start") && strcmp(symname, "collector_func_load"))
                             printf_log(LOG_NONE, "%s: Global Symbol %s (ver=%d/%s) not found, cannot apply R_X86_64_GLOB_DAT @%p (%p) in %s\n", (bind==STB_WEAK)?"Warning":"Error", symname, version, vername?vername:"(none)", p, *(void**)p, head->name);
@@ -676,7 +647,7 @@ int RelocateElfRELA(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t
             weakdefver = old_weakdefver;
         } else {
             old_globdefver = globdefver = (bind==STB_WEAK)?NULL:GetMaplibDefaultVersion(maplib, local_maplib, 0, symname);
-            old_weakdefver = weakdefver = (bind==STB_WEAK)?GetMaplibDefaultVersion(maplib, local_maplib, 1, symname):NULL;
+            old_weakdefver = weakdefver = (bind==STB_WEAK || !globdefver)?GetMaplibDefaultVersion(maplib, local_maplib, 1, symname):NULL;
         }
         if(bind==STB_LOCAL) {
             if(!symname || !symname[0]) {
@@ -715,9 +686,10 @@ int RelocateElfRELA(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t
                     offs = old_offs;
                     end = old_end;
                 } else {
-                    GetGlobalSymbolStartEnd(maplib, symname, &offs, &end, head, version, vername, globdefver, weakdefver);
                     if(!offs && !end && local_maplib)
                         GetGlobalSymbolStartEnd(local_maplib, symname, &offs, &end, head, version, vername, globdefver, weakdefver);
+                    if(!offs && !end)
+                        GetGlobalSymbolStartEnd(maplib, symname, &offs, &end, head, version, vername, globdefver, weakdefver);
                 }
             }
         }
@@ -787,10 +759,10 @@ int RelocateElfRELA(lib_t *maplib, lib_t *local_maplib, int bindnow, elfheader_t
                     *p = globoffs;
                 } else {
                     // Look for same symbol already loaded but not in self (so no need for local_maplib here)
-                    if (GetGlobalNoWeakSymbolStartEnd(local_maplib?local_maplib:maplib, symname, &globoffs, &globend, version, vername, globdefver)) {
+                    /*if (GetGlobalNoWeakSymbolStartEnd(local_maplib?local_maplib:maplib, symname, &globoffs, &globend, version, vername, globdefver)) {
                         offs = globoffs;
                         end = globend;
-                    }
+                    }*/
                     if (!offs) {
                         if(strcmp(symname, "__gmon_start__") && strcmp(symname, "data_start") && strcmp(symname, "__data_start") && strcmp(symname, "collector_func_load"))
                             printf_log((bind==STB_WEAK)?LOG_INFO: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);
@@ -1060,6 +1032,10 @@ uintptr_t GetLastByte(elfheader_t* h)
     return (uintptr_t)h->memory/* + h->delta*/ + h->memsz;
 }
 
+#ifndef STB_GNU_UNIQUE
+#define STB_GNU_UNIQUE	10
+#endif
+
 void checkHookedSymbols(elfheader_t* h); // in mallochook.c
 void AddSymbols(lib_t *maplib, kh_mapsymbols_t* mapsymbols, kh_mapsymbols_t* weaksymbols, kh_mapsymbols_t* localsymbols, elfheader_t* h)
 {
@@ -1273,8 +1249,8 @@ void RefreshElfTLS(elfheader_t* h)
         char* dest = (char*)(my_context->tlsdata+my_context->tlssize+h->tlsbase);
         printf_dump(LOG_DEBUG, "Refreshing main TLS block @%p from %p:0x%lx\n", dest, (void*)h->tlsaddr, h->tlsfilesize);
         memcpy(dest, (void*)(h->tlsaddr+h->delta), h->tlsfilesize);
-        tlsdatasize_t* ptr;
-        if ((ptr = (tlsdatasize_t*)pthread_getspecific(my_context->tlskey)) != NULL) {
+        if (pthread_getspecific(my_context->tlskey)) {
+            tlsdatasize_t* ptr = getTLSData(my_context);
             // refresh in tlsdata too
             dest = (char*)(ptr->data+h->tlsbase);
             printf_dump(LOG_DEBUG, "Refreshing active TLS block @%p from %p:0x%lx\n", dest, (void*)h->tlsaddr, h->tlssize-h->tlsfilesize);
@@ -1328,16 +1304,15 @@ void RunElfInit(elfheader_t* h, x64emu_t *emu)
     // reset Segs Cache
     memset(emu->segs_serial, 0, sizeof(emu->segs_serial));
     uintptr_t p = h->initentry + h->delta;
-    box64context_t* context = GetEmuContext(emu);
     // Refresh no-file part of TLS in case default value changed
     RefreshElfTLS(h);
     // check if in deferredInit
-    if(context->deferredInit) {
-        if(context->deferredInitSz==context->deferredInitCap) {
-            context->deferredInitCap += 4;
-            context->deferredInitList = (elfheader_t**)box_realloc(context->deferredInitList, context->deferredInitCap*sizeof(elfheader_t*));
+    if(my_context->deferredInit) {
+        if(my_context->deferredInitSz==my_context->deferredInitCap) {
+            my_context->deferredInitCap += 4;
+            my_context->deferredInitList = (elfheader_t**)box_realloc(my_context->deferredInitList, my_context->deferredInitCap*sizeof(elfheader_t*));
         }
-        context->deferredInitList[context->deferredInitSz++] = h;
+        my_context->deferredInitList[my_context->deferredInitSz++] = h;
         return;
     }
     h->init_done = 1;
@@ -1349,14 +1324,14 @@ void RunElfInit(elfheader_t* h, x64emu_t *emu)
     }
     printf_dump(LOG_DEBUG, "Calling Init for %s @%p\n", ElfName(h), (void*)p);
     if(h->initentry)
-        RunFunctionWithEmu(emu, 0, p, 3, context->argc, context->argv, context->envv);
+        RunFunctionWithEmu(emu, 0, p, 3, my_context->argc, my_context->argv, my_context->envv);
     printf_dump(LOG_DEBUG, "Done Init for %s\n", ElfName(h));
     // and check init array now
     Elf64_Addr *addr = (Elf64_Addr*)(h->initarray + h->delta);
     for (size_t i=0; i<h->initarray_sz; ++i) {
         if(addr[i]) {
             printf_dump(LOG_DEBUG, "Calling Init[%zu] for %s @%p\n", i, ElfName(h), (void*)addr[i]);
-            RunFunctionWithEmu(emu, 0, (uintptr_t)addr[i], 3, context->argc, context->argv, context->envv);
+            RunFunctionWithEmu(emu, 0, (uintptr_t)addr[i], 3, my_context->argc, my_context->argv, my_context->envv);
         }
     }
 
@@ -1391,6 +1366,10 @@ void RunElfFini(elfheader_t* h, x64emu_t *emu)
     if(!h || h->fini_done || !h->init_done)
         return;
     h->fini_done = 1;
+#ifdef ANDROID
+    // TODO: Fix .fini_array on Android
+    printf_log(LOG_DEBUG, "Android does not support Fini for %s\n", ElfName(h));
+#else
     // first check fini array
     Elf64_Addr *addr = (Elf64_Addr*)(h->finiarray + h->delta);
     for (int i=h->finiarray_sz-1; i>=0; --i) {
@@ -1405,6 +1384,7 @@ void RunElfFini(elfheader_t* h, x64emu_t *emu)
     }
     h->init_done = 0;   // can be re-inited again...
     return;
+#endif
 }
 
 uintptr_t GetElfInit(elfheader_t* h)
@@ -1418,7 +1398,7 @@ uintptr_t GetElfFini(elfheader_t* h)
 
 void* GetBaseAddress(elfheader_t* h)
 {
-    return h->memory;
+    return h->image;
 }
 
 void* GetElfDelta(elfheader_t* h)
@@ -1435,13 +1415,20 @@ int IsAddressInElfSpace(const elfheader_t* h, uintptr_t addr)
 {
     if(!h)
         return 0;
+    #if defined(PAGE8K) || defined(PAGE16K) || defined(PAGE64K)
+    uintptr_t base = (uintptr_t)h->image;
+    uintptr_t end = base+h->memsz;
+    if(base && addr>=base && addr<=end)
+        return 1;
+    #else //PAGE4K
     for(int i=0; i<h->multiblock_n; ++i) {
-        uintptr_t base = h->multiblock_offs[i];
-        uintptr_t end = h->multiblock_offs[i] + h->multiblock_size[i] - 1;
-        if(addr>=base && addr<=end)
+        uintptr_t base = (uintptr_t)h->multiblocks[i].p;
+        uintptr_t end = (uintptr_t)h->multiblocks[i].p + h->multiblocks[i].asize - 1;
+        if(base && addr>=base && addr<=end)
             return 1;
         
     }
+    #endif //PAGE4K
     return 0;
 }
 elfheader_t* FindElfAddress(box64context_t *context, uintptr_t addr)
@@ -1461,22 +1448,23 @@ const char* FindNearestSymbolName(elfheader_t* h, void* p, uintptr_t* start, uin
     const char* ret = NULL;
     uintptr_t s = 0;
     uint64_t size = 0;
-    #ifdef HAVE_TRACE
+    if((uintptr_t)p<0x10000)
+        return ret;
     if(!h) {
         if(getProtection((uintptr_t)p)&(PROT_READ)) {
-            if(*(uint8_t*)(p)==0xCC && *(uint8_t*)(p+1)=='S' && *(uint8_t*)(p+2)=='C') {
-                ret = getBridgeName(*(void**)(p+3+8));
+            uintptr_t adj_p = ((uintptr_t)p)&~(sizeof(onebridge_t));
+            if(*(uint8_t*)(adj_p)==0xCC && *(uint8_t*)(adj_p+1)=='S' && *(uint8_t*)(adj_p+2)=='C') {
+                ret = getBridgeName((void*)adj_p);
                 if(ret) {
                     if(start)
-                        *start = (uintptr_t)p;
+                        *start = (uintptr_t)adj_p;
                     if(sz)
-                        *sz = 32;
+                        *sz = sizeof(onebridge_t);
                 }
             }
         }
         return ret;
     }
-    #endif
     if(!h || h->fini_done)
         return ret;
 

src/elfs/elfloader_private.h

@@ -9,7 +9,17 @@ typedef struct kh_defaultversion_s kh_defaultversion_t;
 #include <elf.h>
 #include "elfloader.h"
 
-struct elfheader_s {
+typedef struct multiblock_s {
+    void*       p;
+    uintptr_t   offs;
+    uintptr_t   paddr;
+    uintptr_t   align;
+    uint64_t    size;
+    uint64_t    asize;
+    uint8_t     flags;
+} multiblock_t;
+
+typedef struct elfheader_s {
     char*       name;
     char*       path;   // Resolved path to file
     char*       soname; // soname of the elf
@@ -39,6 +49,7 @@ struct elfheader_s {
     uint32_t    flags;
 
     intptr_t    delta;  // should be 0
+    void*       image;
 
     uintptr_t   entrypoint;
     uintptr_t   initentry;
@@ -90,20 +101,21 @@ struct elfheader_s {
     int         malloc_hook_2;  // this elf hook malloc, hacking it
 
     char*       memory;     // char* and not void* to allow math on memory pointer
-    void**      multiblock;
-    uintptr_t*  multiblock_offs;
-    uint64_t*   multiblock_size;
+    multiblock_t*  multiblocks;
     int         multiblock_n;
 
     library_t   *lib;       // attached lib (exept on main elf)
     needed_libs_t* needed;
 
+    FILE*       file;
+    int         fileno;
+
     kh_mapsymbols_t   *mapsymbols;
     kh_mapsymbols_t   *weaksymbols;
     kh_mapsymbols_t   *localsymbols;
     kh_defaultversion_t *globaldefver;  // the global default version for symbols (the XXX@@vvvv of symbols)
     kh_defaultversion_t *weakdefver;    // the weak default version for symbols (the XXX@@vvvv of symbols)
-};
+} elfheader_t;
 
 #define R_X86_64_NONE           0       /* No reloc */
 #define R_X86_64_64             1       /* Direct 64 bit  */

src/elfs/elfparser.c

@@ -247,8 +247,10 @@ elfheader_t* ParseElfHeader(FILE* f, const char* name, int exec)
                     printf_dump(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);
+                    #ifndef ANDROID
+                        if(val)
+                            printf_log(LOG_NONE, "Warning, PreInit Array (size=%d) present and ignored!\n", val);
+                    #endif
                     break;
                 case DT_FINI: // Exit hook
                     h->finientry = ptr;
@@ -466,4 +468,19 @@ const char* GetNeededVersionString(elfheader_t* h, const char* libname, int idx)
         }
     }
     return NULL;
-}
+}
+
+int GetNeededVersionForLib(elfheader_t* h, const char* libname, const char* ver)
+{
+    if(!libname || !ver)
+        return 0;
+    int n = GetNeededVersionCnt(h, libname);
+    if(!n)
+        return 0;
+    for(int i=0; i<n; ++i) {
+        const char* vername = GetNeededVersionString(h, libname, i);
+        if(vername && !strcmp(ver, vername))
+            return 1;
+    }
+    return 0;
+}

src/emu/x64compstrings.c

@@ -0,0 +1,120 @@
+#include <stdint.h>
+
+#include "box64stack.h"
+#include "x64emu.h"
+#include "x64run_private.h"
+#include "x64emu_private.h"
+#include "x64compstrings.h"
+
+static int overrideIfDataInvalid(sse_regs_t* mem, int lmem, sse_regs_t* reg, int lreg, int j, int i, int imm8)
+{
+    int valid1 = (i<lreg);
+    int valid2 = (j<lmem);
+    if(!valid1 && !valid2)
+        switch((imm8>>2)&3) {
+            case 0b00:
+            case 0b01:  return 0;
+            case 0b10:
+            case 0b11:  return 1;
+        }
+    if(!valid1 && valid2)
+        switch((imm8>>2)&3) {
+            case 0b00:
+            case 0b01:
+            case 0b10:  return 0;
+            case 0b11:  return 1;
+        }
+    if(valid1 && !valid2)
+        return 0;
+    switch((imm8>>2)&3) {
+        case 0b01:  // range
+            switch (imm8&3) {
+                case 0b00:  // ub
+                    return (i&1)?((reg->ub[i]>=mem->ub[j])):((reg->ub[i]<=mem->ub[j]));
+                case 0b01:  // uw
+                    return (i&1)?((reg->uw[i]>=mem->uw[j])):((reg->uw[i]<=mem->uw[j]));
+                case 0b10:  // sb
+                    return (i&1)?((reg->sb[i]>=mem->sb[j])):((reg->sb[i]<=mem->sb[j]));
+                case 0b11:  // sw
+                    return (i&1)?((reg->sw[i]>=mem->sw[j])):((reg->sw[i]<=mem->sw[j]));
+            }
+            break;
+        default:    // the others
+            switch (imm8&1) {
+                case 0: // byte
+                    return (reg->ub[i] == mem->ub[j]);
+                case 1: // word
+                    return (reg->uw[i] == mem->uw[j]);
+            }
+    }
+}
+
+uint32_t sse42_compare_string_explicit_len(x64emu_t* emu, sse_regs_t* mem, int lmem, sse_regs_t* reg, int lreg, uint8_t imm8)
+{
+    // get number of packed byte/word
+    int n_packed = (imm8&1)?8:16;
+    if(lreg<0) lreg = -lreg;
+    if(lmem<0) lmem = -lmem;
+    if(lreg>n_packed) lreg = n_packed;
+    if(lmem>n_packed) lmem = n_packed;
+    // aggregate to intres1
+    uint32_t intres1 = 0;
+    switch((imm8>>2)&3) {
+        case 0b00:  //Equal any
+            for(int j=0; j<n_packed; ++j)
+                for(int i=0; i<n_packed; ++i) {
+                    intres1 |= overrideIfDataInvalid(mem, lmem, reg, lreg, j, i, imm8)<<j;
+                }
+            break;
+        case 0b01:  // Range
+            for(int j=0; j<n_packed; ++j)
+                for(int i=0; i<n_packed; i+=2) {
+                    intres1 |= (overrideIfDataInvalid(mem, lmem, reg, lreg, j, i, imm8) & overrideIfDataInvalid(mem, lmem, reg, lreg, j, i+1, imm8))<<j;
+                }
+            break;
+        case 0b10:  // Equal each
+            for(int i=0; i<n_packed; ++i) {
+                intres1 |= overrideIfDataInvalid(mem, lmem, reg, lreg, i, i, imm8)<<i;
+            }
+            break;
+        case 0b11:  // Equal ordered
+            intres1 = (1<<n_packed)-1;
+            for(int j=0; j<n_packed; ++j)
+                for(int i=0; i<n_packed-j; ++i) {
+                    int k = i+j;
+                    intres1 &= (((1<<n_packed)-1)^(1<<j)) | overrideIfDataInvalid(mem, lmem, reg, lreg, k, i, imm8)<<j;
+                }
+            break;
+    }
+    // build intres2
+    uint32_t intres2 = intres1;
+    switch((imm8>>4)&3) {
+        case 0b01: intres2 ^= ((1<<n_packed)-1); break;
+        case 0b11: intres2 ^= ((1<<lmem)-1); break;
+    }
+    // and now set the flags
+    RESET_FLAGS(emu);
+    CONDITIONAL_SET_FLAG(intres2, F_CF);
+    CONDITIONAL_SET_FLAG(lmem<n_packed, F_ZF);
+    CONDITIONAL_SET_FLAG(lreg<n_packed, F_SF);
+    CONDITIONAL_SET_FLAG(intres2&1, F_OF);
+    CLEAR_FLAG(F_AF);
+    CLEAR_FLAG(F_PF);
+    
+    return intres2;
+}
+
+uint32_t sse42_compare_string_implicit_len(x64emu_t* emu, sse_regs_t* mem, sse_regs_t* reg, uint8_t imm8)
+{
+    int lmem = 0;
+    int lreg = 0;
+    // get lmem and lreg
+    if(imm8&1) {
+        while(lmem<8 && mem->uw[lmem]) ++lmem;
+        while(lreg<8 && reg->uw[lreg]) ++lreg;
+    } else {
+        while(lmem<16 && mem->ub[lmem]) ++lmem;
+        while(lreg<16 && reg->ub[lreg]) ++lreg;
+    }
+    return sse42_compare_string_explicit_len(emu, mem, lmem, reg, lreg, imm8);
+}

src/emu/x64compstrings.h

@@ -0,0 +1,13 @@
+#ifndef __X64_CMPSTRINGS_H__
+#define __X64_CMPSTRINGS_H__
+
+#include <stdint.h>
+
+#include "regs.h"
+
+typedef struct x64emu_s x64emu_t;
+
+uint32_t sse42_compare_string_explicit_len(x64emu_t* emu, sse_regs_t* a, int la, sse_regs_t* b, int lb, uint8_t imm8);
+uint32_t sse42_compare_string_implicit_len(x64emu_t* emu, sse_regs_t* a, sse_regs_t* b, uint8_t imm8);
+
+#endif //__X64_CMPSTRINGS_H__